[PATCHES][RFC][CFR] mount-related stuff

[PATCHES][RFC][CFR] mount-related stuff

[Al Viro]

	The next pile of mount massage; it will grow - there will be
further modifications, as well as fixes and documentation, but this is
the subset I've got in more or less settled form right now.

	Review and testing would be very welcome.

	This series (-rc1-based) sits in
git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs.git #work.mount
individual patches in followups.

	Rough overview:

Part 1: trivial cleanups and helpers:

1) copy_tree(): don't set ->mnt_mountpoint on the root of copy
	Ancient bogosity, fortunately harmless, but confusing.
2) constify mnt_has_parent()
3) pnode: lift peers() into pnode.h
4) new predicate: mount_is_ancestor()
	Incidentally, I wonder if the "early bail out on move
of anon into the same anon" was not due to (now eliminated)
corner case in loop detection...  Christian?
5) constify is_local_mountpoint()
6) new predicate: anon_ns_root(mount)
7) dissolve_on_fput(): use anon_ns_root()
8) don't set MNT_LOCKED on parentless mounts
	Simplify the rules for MNT_LOCKED
9) clone_mnt(): simplify the propagation-related logics
	... making it somewhat easier to verify correctness wrt
propagation graph invariants.
10) do_umount(): simplify the "is it still mounted" checks
	it needs to check that mount is ours and it has gradually
grown an equivalent of such check, but it's badly obfuscated.

Part 2: (somewhat of a side story) restore the machinery for long-term
mounts from accumulated bitrot.

11) sanitize handling of long-term internal mounts

Part 3: propagate_umount() rewrite (posted last cycle)

12) Rewrite of propagate_umount()

Part 4: untangling do_move_mount()/attach_recursive_mnt().  This is one area that
will definitely grow - reliable avoidance of having multiple mounts with the same
parent/mountpoint pair will go in there.

13) attach_mnt(): expand in attach_recursive_mnt(), then lose the flag argument
14) do_move_mount(): take dropping the old mountpoint into attach_recursive_mnt()
15) get rid of mnt_set_mountpoint_beneath()
16) make commit_tree() usable in same-namespace move case
17) attach_recursive_mnt(): unify the mnt_change_mountpoint() logics
18) attach_recursive_mnt(): pass destination mount in all cases
19) attach_recursive_mnt(): get rid of flags entirely
20) do_move_mount(): get rid of 'attached' flag

Part 5: change locking for expiry lists.
21) attach_recursive_mnt(): remove from expiry list on move
22) take ->mnt_expire handling under mount_lock [read_seqlock_excl]

Part 6: struct mountpoint massage.
23) pivot_root(): reorder tree surgeries, collapse unhash_mnt() and put_mountpoint()
24) combine __put_mountpoint() with unhash_mnt()
25) get rid of mountpoint->m_count

Part 7: regularize mount refcounting a bit
26) don't have mounts pin their parents

[PATCH 01/26] copy_tree(): don't set ->mnt_mountpoint on the root of copy

[Al Viro]

It never made any sense - neither when copy_tree() had been introduced
(2.4.11-pre5), nor at any point afterwards.  Mountpoint is meaningless
without parent mount and the root of copied tree has no parent until we get
around to attaching it somewhere.  At that time we'll have mountpoint set;
before that we have no idea which dentry will be used as mountpoint.
IOW, copy_tree() should just leave the default value.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 1 -
 1 file changed, 1 deletion(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index e13d9ab4f564..5eeb17c39fcb 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -2259,7 +2259,6 @@ struct mount *copy_tree(struct mount *src_root, struct dentry *dentry,
 		return dst_mnt;
 
 	src_parent = src_root;
-	dst_mnt->mnt_mountpoint = src_root->mnt_mountpoint;
 
 	list_for_each_entry(src_root_child, &src_root->mnt_mounts, mnt_child) {
 		if (!is_subdir(src_root_child->mnt_mountpoint, dentry))
-- 
2.39.5

[PATCH 02/26] constify mnt_has_parent()

[Al Viro]

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/mount.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/fs/mount.h b/fs/mount.h
index ad7173037924..02e5d7b34d43 100644
--- a/fs/mount.h
+++ b/fs/mount.h
@@ -101,7 +101,7 @@ static inline struct mount *real_mount(struct vfsmount *mnt)
 	return container_of(mnt, struct mount, mnt);
 }
 
-static inline int mnt_has_parent(struct mount *mnt)
+static inline int mnt_has_parent(const struct mount *mnt)
 {
 	return mnt != mnt->mnt_parent;
 }
-- 
2.39.5

Re: [PATCH 02/26] constify mnt_has_parent()

[Christian Brauner]

On Tue, Jun 10, 2025 at 09:21:24AM +0100, Al Viro wrote:
> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
> ---

Reviewed-by: Christian Brauner <brauner@kernel.org>

[PATCH 03/26] pnode: lift peers() into pnode.h

[Al Viro]

it's going to be useful both in pnode.c and namespace.c

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/pnode.c | 5 -----
 fs/pnode.h | 5 +++++
 2 files changed, 5 insertions(+), 5 deletions(-)

diff --git a/fs/pnode.c b/fs/pnode.c
index ffd429b760d5..aa187144e389 100644
--- a/fs/pnode.c
+++ b/fs/pnode.c
@@ -216,11 +216,6 @@ static struct mount *next_group(struct mount *m, struct mount *origin)
 static struct mount *last_dest, *first_source, *last_source, *dest_master;
 static struct hlist_head *list;
 
-static inline bool peers(const struct mount *m1, const struct mount *m2)
-{
-	return m1->mnt_group_id == m2->mnt_group_id && m1->mnt_group_id;
-}
-
 static int propagate_one(struct mount *m, struct mountpoint *dest_mp)
 {
 	struct mount *child;
diff --git a/fs/pnode.h b/fs/pnode.h
index 34b6247af01d..b728b5de4adc 100644
--- a/fs/pnode.h
+++ b/fs/pnode.h
@@ -36,6 +36,11 @@ static inline void set_mnt_shared(struct mount *mnt)
 	mnt->mnt.mnt_flags |= MNT_SHARED;
 }
 
+static inline bool peers(const struct mount *m1, const struct mount *m2)
+{
+	return m1->mnt_group_id == m2->mnt_group_id && m1->mnt_group_id;
+}
+
 void change_mnt_propagation(struct mount *, int);
 int propagate_mnt(struct mount *, struct mountpoint *, struct mount *,
 		struct hlist_head *);
-- 
2.39.5

Re: [PATCH 03/26] pnode: lift peers() into pnode.h

[Christian Brauner]

On Tue, Jun 10, 2025 at 09:21:25AM +0100, Al Viro wrote:
> it's going to be useful both in pnode.c and namespace.c
> 
> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
> ---

Reviewed-by: Christian Brauner <brauner@kernel.org>

[PATCH 04/26] new predicate: mount_is_ancestor()

[Al Viro]

mount_is_ancestor(p1, p2) returns true iff there is a possibly
empty ancestry chain from p1 to p2.

Convert the open-coded checks.  Unlike those open-coded variants
it does not depend upon p1 not being root...

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 21 +++++++++++++++------
 1 file changed, 15 insertions(+), 6 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index 5eeb17c39fcb..b60cb35aa59c 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -3491,6 +3491,17 @@ static inline bool path_overmounted(const struct path *path)
 	return unlikely(!no_child);
 }
 
+/*
+ * Check if there is a possibly empty chain of descent from p1 to p2.
+ * Locks: namespace_sem (shared) or mount_lock (read_seqlock_excl).
+ */
+static bool mount_is_ancestor(const struct mount *p1, const struct mount *p2)
+{
+	while (p2 != p1 && mnt_has_parent(p2))
+		p2 = p2->mnt_parent;
+	return p2 == p1;
+}
+
 /**
  * can_move_mount_beneath - check that we can mount beneath the top mount
  * @from: mount to mount beneath
@@ -3542,9 +3553,8 @@ static int can_move_mount_beneath(const struct path *from,
 	if (parent_mnt_to == current->nsproxy->mnt_ns->root)
 		return -EINVAL;
 
-	for (struct mount *p = mnt_from; mnt_has_parent(p); p = p->mnt_parent)
-		if (p == mnt_to)
-			return -EINVAL;
+	if (mount_is_ancestor(mnt_to, mnt_from))
+		return -EINVAL;
 
 	/*
 	 * If the parent mount propagates to the child mount this would
@@ -3713,9 +3723,8 @@ static int do_move_mount(struct path *old_path,
 	err = -ELOOP;
 	if (!check_for_nsfs_mounts(old))
 		goto out;
-	for (; mnt_has_parent(p); p = p->mnt_parent)
-		if (p == old)
-			goto out;
+	if (mount_is_ancestor(old, p))
+		goto out;
 
 	err = attach_recursive_mnt(old, real_mount(new_path->mnt), mp, flags);
 	if (err)
-- 
2.39.5

Re: [PATCH 04/26] new predicate: mount_is_ancestor()

[Christian Brauner]

On Tue, Jun 10, 2025 at 09:21:26AM +0100, Al Viro wrote:
> mount_is_ancestor(p1, p2) returns true iff there is a possibly
> empty ancestry chain from p1 to p2.
> 
> Convert the open-coded checks.  Unlike those open-coded variants
> it does not depend upon p1 not being root...
> 
> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
> ---

Reviewed-by: Christian Brauner <brauner@kernel.org>

[PATCH 05/26] constify is_local_mountpoint()

[Al Viro]

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/mount.h     | 4 ++--
 fs/namespace.c | 2 +-
 2 files changed, 3 insertions(+), 3 deletions(-)

diff --git a/fs/mount.h b/fs/mount.h
index 02e5d7b34d43..9fe06e901cc8 100644
--- a/fs/mount.h
+++ b/fs/mount.h
@@ -146,8 +146,8 @@ struct proc_mounts {
 
 extern const struct seq_operations mounts_op;
 
-extern bool __is_local_mountpoint(struct dentry *dentry);
-static inline bool is_local_mountpoint(struct dentry *dentry)
+extern bool __is_local_mountpoint(const struct dentry *dentry);
+static inline bool is_local_mountpoint(const struct dentry *dentry)
 {
 	if (!d_mountpoint(dentry))
 		return false;
diff --git a/fs/namespace.c b/fs/namespace.c
index b60cb35aa59c..2fb5b9fcd2cd 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -894,7 +894,7 @@ struct vfsmount *lookup_mnt(const struct path *path)
  * namespace not just a mount that happens to have some specified
  * parent mount.
  */
-bool __is_local_mountpoint(struct dentry *dentry)
+bool __is_local_mountpoint(const struct dentry *dentry)
 {
 	struct mnt_namespace *ns = current->nsproxy->mnt_ns;
 	struct mount *mnt, *n;
-- 
2.39.5

Re: [PATCH 05/26] constify is_local_mountpoint()

[Christian Brauner]

On Tue, Jun 10, 2025 at 09:21:27AM +0100, Al Viro wrote:
> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
> ---

Reviewed-by: Christian Brauner <brauner@kernel.org>

[PATCH 06/26] new predicate: anon_ns_root(mount)

[Al Viro]

checks if mount is the root of an anonymouns namespace.
Switch open-coded equivalents to using it.

For mounts that belong to anon namespace !mnt_has_parent(mount)
is the same as mount == ns->root, and intent is more obvious in
the latter form.

NB: comment in do_mount_setattr() appears to be very confused...

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/mount.h     |  7 +++++++
 fs/namespace.c | 17 +++--------------
 2 files changed, 10 insertions(+), 14 deletions(-)

diff --git a/fs/mount.h b/fs/mount.h
index 9fe06e901cc8..18fa88ad752a 100644
--- a/fs/mount.h
+++ b/fs/mount.h
@@ -160,6 +160,13 @@ static inline bool is_anon_ns(struct mnt_namespace *ns)
 	return ns->seq == 0;
 }
 
+static inline bool anon_ns_root(const struct mount *m)
+{
+	struct mnt_namespace *ns = READ_ONCE(m->mnt_ns);
+
+	return !IS_ERR_OR_NULL(ns) && is_anon_ns(ns) && m == ns->root;
+}
+
 static inline bool mnt_ns_attached(const struct mount *mnt)
 {
 	return !RB_EMPTY_NODE(&mnt->mnt_node);
diff --git a/fs/namespace.c b/fs/namespace.c
index 2fb5b9fcd2cd..b229f74762de 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -2485,9 +2485,7 @@ struct vfsmount *clone_private_mount(const struct path *path)
 	 * loops get created.
 	 */
 	if (!check_mnt(old_mnt)) {
-		if (!is_mounted(&old_mnt->mnt) ||
-			!is_anon_ns(old_mnt->mnt_ns) ||
-			mnt_has_parent(old_mnt))
+		if (!anon_ns_root(old_mnt))
 			return ERR_PTR(-EINVAL);
 
 		if (!check_for_nsfs_mounts(old_mnt))
@@ -3657,9 +3655,6 @@ static int do_move_mount(struct path *old_path,
 	ns = old->mnt_ns;
 
 	err = -EINVAL;
-	/* The thing moved must be mounted... */
-	if (!is_mounted(&old->mnt))
-		goto out;
 
 	if (check_mnt(old)) {
 		/* if the source is in our namespace... */
@@ -3672,10 +3667,8 @@ static int do_move_mount(struct path *old_path,
 	} else {
 		/*
 		 * otherwise the source must be the root of some anon namespace.
-		 * AV: check for mount being root of an anon namespace is worth
-		 * an inlined predicate...
 		 */
-		if (!is_anon_ns(ns) || mnt_has_parent(old))
+		if (!anon_ns_root(old))
 			goto out;
 		/*
 		 * Bail out early if the target is within the same namespace -
@@ -5036,10 +5029,6 @@ static int do_mount_setattr(struct path *path, struct mount_kattr *kattr)
 	err = -EINVAL;
 	lock_mount_hash();
 
-	/* Ensure that this isn't anything purely vfs internal. */
-	if (!is_mounted(&mnt->mnt))
-		goto out;
-
 	/*
 	 * If this is an attached mount make sure it's located in the callers
 	 * mount namespace. If it's not don't let the caller interact with it.
@@ -5051,7 +5040,7 @@ static int do_mount_setattr(struct path *path, struct mount_kattr *kattr)
 	 * neither has a parent nor is it a detached mount so we cannot
 	 * unconditionally check for detached mounts.
 	 */
-	if ((mnt_has_parent(mnt) || !is_anon_ns(mnt->mnt_ns)) && !check_mnt(mnt))
+	if (!anon_ns_root(mnt) && !check_mnt(mnt))
 		goto out;
 
 	/*
-- 
2.39.5

Re: [PATCH 06/26] new predicate: anon_ns_root(mount)

[Christian Brauner]

On Tue, Jun 10, 2025 at 09:21:28AM +0100, Al Viro wrote:
> checks if mount is the root of an anonymouns namespace.
> Switch open-coded equivalents to using it.
> 
> For mounts that belong to anon namespace !mnt_has_parent(mount)
> is the same as mount == ns->root, and intent is more obvious in
> the latter form.
> 
> NB: comment in do_mount_setattr() appears to be very confused...

The comment just mentions a single case where we did regress userspace
some time ago because we didn't allowing changing mount properties on
the real rootfs (And we have this discussion on another thread.).

But I'm not sure why this belongs in the commit message in the first
place. Just remove the comment.

> 
> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
> ---

Reviewed-by: Christian Brauner <brauner@kernel.org>

>  fs/mount.h     |  7 +++++++
>  fs/namespace.c | 17 +++--------------
>  2 files changed, 10 insertions(+), 14 deletions(-)
> 
> diff --git a/fs/mount.h b/fs/mount.h
> index 9fe06e901cc8..18fa88ad752a 100644
> --- a/fs/mount.h
> +++ b/fs/mount.h
> @@ -160,6 +160,13 @@ static inline bool is_anon_ns(struct mnt_namespace *ns)
>  	return ns->seq == 0;
>  }
>  
> +static inline bool anon_ns_root(const struct mount *m)
> +{
> +	struct mnt_namespace *ns = READ_ONCE(m->mnt_ns);
> +
> +	return !IS_ERR_OR_NULL(ns) && is_anon_ns(ns) && m == ns->root;
> +}
> +
>  static inline bool mnt_ns_attached(const struct mount *mnt)
>  {
>  	return !RB_EMPTY_NODE(&mnt->mnt_node);
> diff --git a/fs/namespace.c b/fs/namespace.c
> index 2fb5b9fcd2cd..b229f74762de 100644
> --- a/fs/namespace.c
> +++ b/fs/namespace.c
> @@ -2485,9 +2485,7 @@ struct vfsmount *clone_private_mount(const struct path *path)
>  	 * loops get created.
>  	 */
>  	if (!check_mnt(old_mnt)) {
> -		if (!is_mounted(&old_mnt->mnt) ||
> -			!is_anon_ns(old_mnt->mnt_ns) ||
> -			mnt_has_parent(old_mnt))
> +		if (!anon_ns_root(old_mnt))
>  			return ERR_PTR(-EINVAL);
>  
>  		if (!check_for_nsfs_mounts(old_mnt))
> @@ -3657,9 +3655,6 @@ static int do_move_mount(struct path *old_path,
>  	ns = old->mnt_ns;
>  
>  	err = -EINVAL;
> -	/* The thing moved must be mounted... */
> -	if (!is_mounted(&old->mnt))
> -		goto out;
>  
>  	if (check_mnt(old)) {
>  		/* if the source is in our namespace... */
> @@ -3672,10 +3667,8 @@ static int do_move_mount(struct path *old_path,
>  	} else {
>  		/*
>  		 * otherwise the source must be the root of some anon namespace.
> -		 * AV: check for mount being root of an anon namespace is worth
> -		 * an inlined predicate...
>  		 */
> -		if (!is_anon_ns(ns) || mnt_has_parent(old))
> +		if (!anon_ns_root(old))
>  			goto out;
>  		/*
>  		 * Bail out early if the target is within the same namespace -
> @@ -5036,10 +5029,6 @@ static int do_mount_setattr(struct path *path, struct mount_kattr *kattr)
>  	err = -EINVAL;
>  	lock_mount_hash();
>  
> -	/* Ensure that this isn't anything purely vfs internal. */
> -	if (!is_mounted(&mnt->mnt))
> -		goto out;
> -
>  	/*
>  	 * If this is an attached mount make sure it's located in the callers
>  	 * mount namespace. If it's not don't let the caller interact with it.
> @@ -5051,7 +5040,7 @@ static int do_mount_setattr(struct path *path, struct mount_kattr *kattr)
>  	 * neither has a parent nor is it a detached mount so we cannot
>  	 * unconditionally check for detached mounts.
>  	 */
> -	if ((mnt_has_parent(mnt) || !is_anon_ns(mnt->mnt_ns)) && !check_mnt(mnt))
> +	if (!anon_ns_root(mnt) && !check_mnt(mnt))
>  		goto out;
>  
>  	/*
> -- 
> 2.39.5
> 

Re: [PATCH 06/26] new predicate: anon_ns_root(mount)

[Al Viro]

On Wed, Jun 11, 2025 at 12:39:28PM +0200, Christian Brauner wrote:
> On Tue, Jun 10, 2025 at 09:21:28AM +0100, Al Viro wrote:
> > checks if mount is the root of an anonymouns namespace.
> > Switch open-coded equivalents to using it.
> > 
> > For mounts that belong to anon namespace !mnt_has_parent(mount)
> > is the same as mount == ns->root, and intent is more obvious in
> > the latter form.
> > 
> > NB: comment in do_mount_setattr() appears to be very confused...
> 
> The comment just mentions a single case where we did regress userspace
> some time ago because we didn't allowing changing mount properties on
> the real rootfs (And we have this discussion on another thread.).
> 
> But I'm not sure why this belongs in the commit message in the first
> place. Just remove the comment.

Done

[PATCH 07/26] dissolve_on_fput(): use anon_ns_root()

[Al Viro]

that's the condition we are actually trying to check there...

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 62 +++++++++++---------------------------------------
 1 file changed, 13 insertions(+), 49 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index b229f74762de..e783eb801060 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -2329,67 +2329,31 @@ struct vfsmount *collect_mounts(const struct path *path)
 static void free_mnt_ns(struct mnt_namespace *);
 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *, bool);
 
-static inline bool must_dissolve(struct mnt_namespace *mnt_ns)
-{
-	/*
-        * This mount belonged to an anonymous mount namespace
-        * but was moved to a non-anonymous mount namespace and
-        * then unmounted.
-        */
-	if (unlikely(!mnt_ns))
-		return false;
-
-	/*
-        * This mount belongs to a non-anonymous mount namespace
-        * and we know that such a mount can never transition to
-        * an anonymous mount namespace again.
-        */
-	if (!is_anon_ns(mnt_ns)) {
-		/*
-		 * A detached mount either belongs to an anonymous mount
-		 * namespace or a non-anonymous mount namespace. It
-		 * should never belong to something purely internal.
-		 */
-		VFS_WARN_ON_ONCE(mnt_ns == MNT_NS_INTERNAL);
-		return false;
-	}
-
-	return true;
-}
-
 void dissolve_on_fput(struct vfsmount *mnt)
 {
 	struct mnt_namespace *ns;
 	struct mount *m = real_mount(mnt);
 
+	/*
+	 * m used to be the root of anon namespace; if it still is one,
+	 * we need to dissolve the mount tree and free that namespace.
+	 * Let's try to avoid taking namespace_sem if we can determine
+	 * that there's nothing to do without it - rcu_read_lock() is
+	 * enough to make anon_ns_root() memory-safe and once m has
+	 * left its namespace, it's no longer our concern, since it will
+	 * never become a root of anon ns again.
+	 */
+
 	scoped_guard(rcu) {
-		if (!must_dissolve(READ_ONCE(m->mnt_ns)))
+		if (!anon_ns_root(m))
 			return;
 	}
 
 	scoped_guard(namespace_lock, &namespace_sem) {
-		ns = m->mnt_ns;
-		if (!must_dissolve(ns))
-			return;
-
-		/*
-		 * After must_dissolve() we know that this is a detached
-		 * mount in an anonymous mount namespace.
-		 *
-		 * Now when mnt_has_parent() reports that this mount
-		 * tree has a parent, we know that this anonymous mount
-		 * tree has been moved to another anonymous mount
-		 * namespace.
-		 *
-		 * So when closing this file we cannot unmount the mount
-		 * tree. This will be done when the file referring to
-		 * the root of the anonymous mount namespace will be
-		 * closed (It could already be closed but it would sync
-		 * on @namespace_sem and wait for us to finish.).
-		 */
-		if (mnt_has_parent(m))
+		if (!anon_ns_root(m))
 			return;
 
+		ns = m->mnt_ns;
 		lock_mount_hash();
 		umount_tree(m, UMOUNT_CONNECTED);
 		unlock_mount_hash();
-- 
2.39.5

Re: [PATCH 07/26] dissolve_on_fput(): use anon_ns_root()

[Christian Brauner]

On Tue, Jun 10, 2025 at 09:21:29AM +0100, Al Viro wrote:
> that's the condition we are actually trying to check there...
> 
> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
> ---

Reviewed-by: Christian Brauner <brauner@kernel.org>

>  fs/namespace.c | 62 +++++++++++---------------------------------------
>  1 file changed, 13 insertions(+), 49 deletions(-)
> 
> diff --git a/fs/namespace.c b/fs/namespace.c
> index b229f74762de..e783eb801060 100644
> --- a/fs/namespace.c
> +++ b/fs/namespace.c
> @@ -2329,67 +2329,31 @@ struct vfsmount *collect_mounts(const struct path *path)
>  static void free_mnt_ns(struct mnt_namespace *);
>  static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *, bool);
>  
> -static inline bool must_dissolve(struct mnt_namespace *mnt_ns)
> -{
> -	/*
> -        * This mount belonged to an anonymous mount namespace
> -        * but was moved to a non-anonymous mount namespace and
> -        * then unmounted.
> -        */
> -	if (unlikely(!mnt_ns))
> -		return false;
> -
> -	/*
> -        * This mount belongs to a non-anonymous mount namespace
> -        * and we know that such a mount can never transition to
> -        * an anonymous mount namespace again.
> -        */
> -	if (!is_anon_ns(mnt_ns)) {
> -		/*
> -		 * A detached mount either belongs to an anonymous mount
> -		 * namespace or a non-anonymous mount namespace. It
> -		 * should never belong to something purely internal.
> -		 */
> -		VFS_WARN_ON_ONCE(mnt_ns == MNT_NS_INTERNAL);
> -		return false;
> -	}
> -
> -	return true;
> -}
> -
>  void dissolve_on_fput(struct vfsmount *mnt)
>  {
>  	struct mnt_namespace *ns;
>  	struct mount *m = real_mount(mnt);
>  
> +	/*
> +	 * m used to be the root of anon namespace; if it still is one,
> +	 * we need to dissolve the mount tree and free that namespace.
> +	 * Let's try to avoid taking namespace_sem if we can determine
> +	 * that there's nothing to do without it - rcu_read_lock() is
> +	 * enough to make anon_ns_root() memory-safe and once m has
> +	 * left its namespace, it's no longer our concern, since it will
> +	 * never become a root of anon ns again.
> +	 */
> +
>  	scoped_guard(rcu) {
> -		if (!must_dissolve(READ_ONCE(m->mnt_ns)))
> +		if (!anon_ns_root(m))
>  			return;
>  	}
>  
>  	scoped_guard(namespace_lock, &namespace_sem) {
> -		ns = m->mnt_ns;
> -		if (!must_dissolve(ns))
> -			return;
> -
> -		/*
> -		 * After must_dissolve() we know that this is a detached
> -		 * mount in an anonymous mount namespace.
> -		 *
> -		 * Now when mnt_has_parent() reports that this mount
> -		 * tree has a parent, we know that this anonymous mount
> -		 * tree has been moved to another anonymous mount
> -		 * namespace.
> -		 *
> -		 * So when closing this file we cannot unmount the mount
> -		 * tree. This will be done when the file referring to
> -		 * the root of the anonymous mount namespace will be
> -		 * closed (It could already be closed but it would sync
> -		 * on @namespace_sem and wait for us to finish.).
> -		 */
> -		if (mnt_has_parent(m))
> +		if (!anon_ns_root(m))
>  			return;
>  
> +		ns = m->mnt_ns;
>  		lock_mount_hash();
>  		umount_tree(m, UMOUNT_CONNECTED);
>  		unlock_mount_hash();
> -- 
> 2.39.5
> 

[PATCH 08/26] don't set MNT_LOCKED on parentless mounts

[Al Viro]

Originally MNT_LOCKED meant only one thing - "don't let this mount to
be peeled off its parent, we don't want to have its mountpoint exposed".
Accordingly, it had only been set on mounts that *do* have a parent.
Later it got overloaded with another use - setting it on the absolute
root had given free protection against umount(2) of absolute root
(was possible to trigger, oopsed).  Not a bad trick, but it ended
up costing more than it bought us.  Unfortunately, the cost included
both hard-to-reason-about logics and a subtle race between
mount -o remount,ro and mount --[r]bind - lockless &= ~MNT_LOCKED in
the end of __do_loopback() could race with sb_prepare_remount_readonly()
setting and clearing MNT_HOLD_WRITE (under mount_lock, as it should
be).  The race wouldn't be much of a problem (there are other ways to
deal with it), but the subtlety is.

Turns out that nobody except umount(2) had ever made use of having
MNT_LOCKED set on absolute root.  So let's give up on that trick,
clever as it had been, add an explicit check in do_umount() and
return to using MNT_LOCKED only for mounts that have a parent.

It means that
	* clone_mnt() no longer copies MNT_LOCKED
	* copy_tree() sets it on submounts if their counterparts had
been marked such, and does that right next to attach_mnt() in there,
in the same mount_lock scope.
	* __do_loopback() no longer needs to strip MNT_LOCKED off the
root of subtree it's about to return; no store, no race.
	* init_mount_tree() doesn't bother setting MNT_LOCKED on absolute
root.
	* lock_mnt_tree() does not set MNT_LOCKED on the subtree's root;
accordingly, its caller (loop in attach_recursive_mnt()) does not need to
bother stripping that MNT_LOCKED on root.  Note that lock_mnt_tree() setting
MNT_LOCKED on submounts happens in the same mount_lock scope as __attach_mnt()
(from commit_tree()) that makes them reachable.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 32 +++++++++++++++-----------------
 1 file changed, 15 insertions(+), 17 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index e783eb801060..d6c81eab6a11 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -1349,7 +1349,7 @@ static struct mount *clone_mnt(struct mount *old, struct dentry *root,
 	}
 
 	mnt->mnt.mnt_flags = old->mnt.mnt_flags;
-	mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL);
+	mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL|MNT_LOCKED);
 
 	atomic_inc(&sb->s_active);
 	mnt->mnt.mnt_idmap = mnt_idmap_get(mnt_idmap(&old->mnt));
@@ -2024,6 +2024,9 @@ static int do_umount(struct mount *mnt, int flags)
 	if (mnt->mnt.mnt_flags & MNT_LOCKED)
 		goto out;
 
+	if (!mnt_has_parent(mnt))
+		goto out;
+
 	event++;
 	if (flags & MNT_DETACH) {
 		if (mnt_ns_attached(mnt) || !list_empty(&mnt->mnt_list))
@@ -2293,6 +2296,8 @@ struct mount *copy_tree(struct mount *src_root, struct dentry *dentry,
 			if (IS_ERR(dst_mnt))
 				goto out;
 			lock_mount_hash();
+			if (src_mnt->mnt.mnt_flags & MNT_LOCKED)
+				dst_mnt->mnt.mnt_flags |= MNT_LOCKED;
 			list_add_tail(&dst_mnt->mnt_list, &res->mnt_list);
 			attach_mnt(dst_mnt, dst_parent, src_parent->mnt_mp, false);
 			unlock_mount_hash();
@@ -2508,7 +2513,7 @@ static void lock_mnt_tree(struct mount *mnt)
 		if (flags & MNT_NOEXEC)
 			flags |= MNT_LOCK_NOEXEC;
 		/* Don't allow unprivileged users to reveal what is under a mount */
-		if (list_empty(&p->mnt_expire))
+		if (list_empty(&p->mnt_expire) && p != mnt)
 			flags |= MNT_LOCKED;
 		p->mnt.mnt_flags = flags;
 	}
@@ -2719,7 +2724,6 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 		/* Notice when we are propagating across user namespaces */
 		if (child->mnt_parent->mnt_ns->user_ns != user_ns)
 			lock_mnt_tree(child);
-		child->mnt.mnt_flags &= ~MNT_LOCKED;
 		commit_tree(child);
 	}
 	put_mountpoint(smp);
@@ -2992,26 +2996,21 @@ static inline bool may_copy_tree(struct path *path)
 
 static struct mount *__do_loopback(struct path *old_path, int recurse)
 {
-	struct mount *mnt = ERR_PTR(-EINVAL), *old = real_mount(old_path->mnt);
+	struct mount *old = real_mount(old_path->mnt);
 
 	if (IS_MNT_UNBINDABLE(old))
-		return mnt;
+		return ERR_PTR(-EINVAL);
 
 	if (!may_copy_tree(old_path))
-		return mnt;
+		return ERR_PTR(-EINVAL);
 
 	if (!recurse && __has_locked_children(old, old_path->dentry))
-		return mnt;
+		return ERR_PTR(-EINVAL);
 
 	if (recurse)
-		mnt = copy_tree(old, old_path->dentry, CL_COPY_MNT_NS_FILE);
+		return copy_tree(old, old_path->dentry, CL_COPY_MNT_NS_FILE);
 	else
-		mnt = clone_mnt(old, old_path->dentry, 0);
-
-	if (!IS_ERR(mnt))
-		mnt->mnt.mnt_flags &= ~MNT_LOCKED;
-
-	return mnt;
+		return clone_mnt(old, old_path->dentry, 0);
 }
 
 /*
@@ -4756,11 +4755,11 @@ SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
 	if (!path_mounted(&root))
 		goto out4; /* not a mountpoint */
 	if (!mnt_has_parent(root_mnt))
-		goto out4; /* not attached */
+		goto out4; /* absolute root */
 	if (!path_mounted(&new))
 		goto out4; /* not a mountpoint */
 	if (!mnt_has_parent(new_mnt))
-		goto out4; /* not attached */
+		goto out4; /* absolute root */
 	/* make sure we can reach put_old from new_root */
 	if (!is_path_reachable(old_mnt, old.dentry, &new))
 		goto out4;
@@ -6176,7 +6175,6 @@ static void __init init_mount_tree(void)
 
 	root.mnt = mnt;
 	root.dentry = mnt->mnt_root;
-	mnt->mnt_flags |= MNT_LOCKED;
 
 	set_fs_pwd(current->fs, &root);
 	set_fs_root(current->fs, &root);
-- 
2.39.5

Re: [PATCH 08/26] don't set MNT_LOCKED on parentless mounts

[Christian Brauner]

On Tue, Jun 10, 2025 at 09:21:30AM +0100, Al Viro wrote:
> Originally MNT_LOCKED meant only one thing - "don't let this mount to
> be peeled off its parent, we don't want to have its mountpoint exposed".
> Accordingly, it had only been set on mounts that *do* have a parent.
> Later it got overloaded with another use - setting it on the absolute
> root had given free protection against umount(2) of absolute root
> (was possible to trigger, oopsed).  Not a bad trick, but it ended
> up costing more than it bought us.  Unfortunately, the cost included
> both hard-to-reason-about logics and a subtle race between
> mount -o remount,ro and mount --[r]bind - lockless &= ~MNT_LOCKED in
> the end of __do_loopback() could race with sb_prepare_remount_readonly()
> setting and clearing MNT_HOLD_WRITE (under mount_lock, as it should
> be).  The race wouldn't be much of a problem (there are other ways to
> deal with it), but the subtlety is.
> 
> Turns out that nobody except umount(2) had ever made use of having
> MNT_LOCKED set on absolute root.  So let's give up on that trick,
> clever as it had been, add an explicit check in do_umount() and
> return to using MNT_LOCKED only for mounts that have a parent.
> 
> It means that
> 	* clone_mnt() no longer copies MNT_LOCKED
> 	* copy_tree() sets it on submounts if their counterparts had
> been marked such, and does that right next to attach_mnt() in there,
> in the same mount_lock scope.
> 	* __do_loopback() no longer needs to strip MNT_LOCKED off the
> root of subtree it's about to return; no store, no race.
> 	* init_mount_tree() doesn't bother setting MNT_LOCKED on absolute
> root.
> 	* lock_mnt_tree() does not set MNT_LOCKED on the subtree's root;
> accordingly, its caller (loop in attach_recursive_mnt()) does not need to
> bother stripping that MNT_LOCKED on root.  Note that lock_mnt_tree() setting
> MNT_LOCKED on submounts happens in the same mount_lock scope as __attach_mnt()
> (from commit_tree()) that makes them reachable.
> 
> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
> ---

Reviewed-by: Christian Brauner <brauner@kernel.org>

>  fs/namespace.c | 32 +++++++++++++++-----------------
>  1 file changed, 15 insertions(+), 17 deletions(-)
> 
> diff --git a/fs/namespace.c b/fs/namespace.c
> index e783eb801060..d6c81eab6a11 100644
> --- a/fs/namespace.c
> +++ b/fs/namespace.c
> @@ -1349,7 +1349,7 @@ static struct mount *clone_mnt(struct mount *old, struct dentry *root,
>  	}
>  
>  	mnt->mnt.mnt_flags = old->mnt.mnt_flags;
> -	mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL);
> +	mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL|MNT_LOCKED);
>  
>  	atomic_inc(&sb->s_active);
>  	mnt->mnt.mnt_idmap = mnt_idmap_get(mnt_idmap(&old->mnt));
> @@ -2024,6 +2024,9 @@ static int do_umount(struct mount *mnt, int flags)
>  	if (mnt->mnt.mnt_flags & MNT_LOCKED)
>  		goto out;
>  

This deserves a comment imho.

> +	if (!mnt_has_parent(mnt))
> +		goto out;

[PATCH 09/26] clone_mnt(): simplify the propagation-related logics

[Al Viro]

The underlying rules are simple:
	* MNT_SHARED should be set iff ->mnt_group_id of new mount ends up
non-zero.
	* mounts should be on the same ->mnt_share cyclic list iff they have
the same non-zero ->mnt_group_id value.
	* CL_PRIVATE is mutually exclusive with MNT_SHARED, MNT_SLAVE,
MNT_SHARED_TO_SLAVE and MNT_EXPIRE; the whole point of that thing is to
get a clone of old mount that would *not* be on any namespace-related
lists.

The above allows to make the logics more straightforward; what's more,
it makes the proof that invariants are maintained much simpler.
The variant in mainline is safe (aside of a very narrow race with
unsafe modification of mnt_flags right after we had the mount exposed
in superblock's ->s_mounts; theoretically it can race with ro remount
of the original, but it's not easy to hit), but proof of its correctness
is really unpleasant.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 26 +++++++++++++-------------
 1 file changed, 13 insertions(+), 13 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index d6c81eab6a11..02e9f37f49b9 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -1337,6 +1337,9 @@ static struct mount *clone_mnt(struct mount *old, struct dentry *root,
 	if (!mnt)
 		return ERR_PTR(-ENOMEM);
 
+	mnt->mnt.mnt_flags = READ_ONCE(old->mnt.mnt_flags) &
+			     ~MNT_INTERNAL_FLAGS;
+
 	if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
 		mnt->mnt_group_id = 0; /* not a peer of original */
 	else
@@ -1348,8 +1351,8 @@ static struct mount *clone_mnt(struct mount *old, struct dentry *root,
 			goto out_free;
 	}
 
-	mnt->mnt.mnt_flags = old->mnt.mnt_flags;
-	mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL|MNT_LOCKED);
+	if (mnt->mnt_group_id)
+		set_mnt_shared(mnt);
 
 	atomic_inc(&sb->s_active);
 	mnt->mnt.mnt_idmap = mnt_idmap_get(mnt_idmap(&old->mnt));
@@ -1362,22 +1365,20 @@ static struct mount *clone_mnt(struct mount *old, struct dentry *root,
 	list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
 	unlock_mount_hash();
 
+	if (flag & CL_PRIVATE)	// we are done with it
+		return mnt;
+
+	if (peers(mnt, old))
+		list_add(&mnt->mnt_share, &old->mnt_share);
+
 	if ((flag & CL_SLAVE) ||
 	    ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
 		list_add(&mnt->mnt_slave, &old->mnt_slave_list);
 		mnt->mnt_master = old;
-		CLEAR_MNT_SHARED(mnt);
-	} else if (!(flag & CL_PRIVATE)) {
-		if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
-			list_add(&mnt->mnt_share, &old->mnt_share);
-		if (IS_MNT_SLAVE(old))
-			list_add(&mnt->mnt_slave, &old->mnt_slave);
+	} else if (IS_MNT_SLAVE(old)) {
+		list_add(&mnt->mnt_slave, &old->mnt_slave);
 		mnt->mnt_master = old->mnt_master;
-	} else {
-		CLEAR_MNT_SHARED(mnt);
 	}
-	if (flag & CL_MAKE_SHARED)
-		set_mnt_shared(mnt);
 
 	/* stick the duplicate mount on the same expiry list
 	 * as the original if that was on one */
@@ -1385,7 +1386,6 @@ static struct mount *clone_mnt(struct mount *old, struct dentry *root,
 		if (!list_empty(&old->mnt_expire))
 			list_add(&mnt->mnt_expire, &old->mnt_expire);
 	}
-
 	return mnt;
 
  out_free:
-- 
2.39.5

Re: [PATCH 09/26] clone_mnt(): simplify the propagation-related logics

[Christian Brauner]

On Tue, Jun 10, 2025 at 09:21:31AM +0100, Al Viro wrote:
> The underlying rules are simple:
> 	* MNT_SHARED should be set iff ->mnt_group_id of new mount ends up
> non-zero.
> 	* mounts should be on the same ->mnt_share cyclic list iff they have
> the same non-zero ->mnt_group_id value.
> 	* CL_PRIVATE is mutually exclusive with MNT_SHARED, MNT_SLAVE,
> MNT_SHARED_TO_SLAVE and MNT_EXPIRE; the whole point of that thing is to
> get a clone of old mount that would *not* be on any namespace-related
> lists.
> 
> The above allows to make the logics more straightforward; what's more,
> it makes the proof that invariants are maintained much simpler.
> The variant in mainline is safe (aside of a very narrow race with
> unsafe modification of mnt_flags right after we had the mount exposed
> in superblock's ->s_mounts; theoretically it can race with ro remount
> of the original, but it's not easy to hit), but proof of its correctness
> is really unpleasant.
> 
> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
> ---

Reviewed-by: Christian Brauner <brauner@kernel.org>

[PATCH 10/26] do_umount(): simplify the "is it still mounted" checks

[Al Viro]

Calls of do_umount() are always preceded by can_umount(), where we'd
done a racy check for mount belonging to our namespace; if it wasn't,
can_unmount() would've failed with -EINVAL and we wouldn't have
reached do_umount() at all.

That check needs to be redone once we have acquired namespace_sem
and in do_umount() we do that.  However, that's done in a very odd
way; we check that mount is still in rbtree of _some_ namespace or
its mnt_list is not empty.  It is equivalent to check_mnt(mnt) -
we know that earlier mnt was mounted in our namespace; if it has
stayed there, it's going to remain in rbtree of our namespace.
OTOH, if it ever had been removed from out namespace, it would be
removed from rbtree and it never would've re-added to a namespace
afterwards.  As for ->mnt_list, for something that had been mounted
in a namespace we'll never observe non-empty ->mnt_list while holding
namespace_sem - it does temporarily become non-empty during
umount_tree(), but that doesn't outlast the call of umount_tree(),
let alone dropping namespace_sem.

Things get much easier to follow if we replace that with (equivalent)
check_mnt(mnt) there.  What's more, currently we treat a failure of
that test as "quietly do nothing"; we might as well pretend that we'd
lost the race and fail on that the same way can_umount() would have.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 11 ++++++-----
 1 file changed, 6 insertions(+), 5 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index 02e9f37f49b9..5e82f1ef042a 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -2019,8 +2019,11 @@ static int do_umount(struct mount *mnt, int flags)
 	namespace_lock();
 	lock_mount_hash();
 
-	/* Recheck MNT_LOCKED with the locks held */
+	/* Repeat the earlier racy checks, now that we are holding the locks */
 	retval = -EINVAL;
+	if (!check_mnt(mnt))
+		goto out;
+
 	if (mnt->mnt.mnt_flags & MNT_LOCKED)
 		goto out;
 
@@ -2029,16 +2032,14 @@ static int do_umount(struct mount *mnt, int flags)
 
 	event++;
 	if (flags & MNT_DETACH) {
-		if (mnt_ns_attached(mnt) || !list_empty(&mnt->mnt_list))
-			umount_tree(mnt, UMOUNT_PROPAGATE);
+		umount_tree(mnt, UMOUNT_PROPAGATE);
 		retval = 0;
 	} else {
 		smp_mb(); // paired with __legitimize_mnt()
 		shrink_submounts(mnt);
 		retval = -EBUSY;
 		if (!propagate_mount_busy(mnt, 2)) {
-			if (mnt_ns_attached(mnt) || !list_empty(&mnt->mnt_list))
-				umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
+			umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
 			retval = 0;
 		}
 	}
-- 
2.39.5

Re: [PATCH 10/26] do_umount(): simplify the "is it still mounted" checks

[Christian Brauner]

On Tue, Jun 10, 2025 at 09:21:32AM +0100, Al Viro wrote:
> Calls of do_umount() are always preceded by can_umount(), where we'd
> done a racy check for mount belonging to our namespace; if it wasn't,
> can_unmount() would've failed with -EINVAL and we wouldn't have
> reached do_umount() at all.
> 
> That check needs to be redone once we have acquired namespace_sem
> and in do_umount() we do that.  However, that's done in a very odd
> way; we check that mount is still in rbtree of _some_ namespace or
> its mnt_list is not empty.  It is equivalent to check_mnt(mnt) -
> we know that earlier mnt was mounted in our namespace; if it has
> stayed there, it's going to remain in rbtree of our namespace.
> OTOH, if it ever had been removed from out namespace, it would be
> removed from rbtree and it never would've re-added to a namespace
> afterwards.  As for ->mnt_list, for something that had been mounted
> in a namespace we'll never observe non-empty ->mnt_list while holding
> namespace_sem - it does temporarily become non-empty during
> umount_tree(), but that doesn't outlast the call of umount_tree(),
> let alone dropping namespace_sem.
> 
> Things get much easier to follow if we replace that with (equivalent)
> check_mnt(mnt) there.  What's more, currently we treat a failure of
> that test as "quietly do nothing"; we might as well pretend that we'd
> lost the race and fail on that the same way can_umount() would have.
> 
> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
> ---

Reviewed-by: Christian Brauner <brauner@kernel.org>

[PATCH 11/26] sanitize handling of long-term internal mounts

[Al Viro]

Original rationale for those had been the reduced cost of mntput()
for the stuff that is mounted somewhere.  Mount refcount increments and
decrements are frequent; what's worse, they tend to concentrate on the
same instances and cacheline pingpong is quite noticable.

As the result, mount refcounts are per-cpu; that allows a very cheap
increment.  Plain decrement would be just as easy, but decrement-and-test
is anything but (we need to add the components up, with exclusion against
possible increment-from-zero, etc.).

Fortunately, there is a very common case where we can tell that decrement
won't be the final one - if the thing we are dropping is currently
mounted somewhere.  We have an RCU delay between the removal from mount
tree and dropping the reference that used to pin it there, so we can
just take rcu_read_lock() and check if the victim is mounted somewhere.
If it is, we can go ahead and decrement without and further checks -
the reference we are dropping is not the last one.  If it isn't, we
get all the fun with locking, carefully adding up components, etc.,
but the majority of refcount decrements end up taking the fast path.

There is a major exception, though - pipes and sockets.  Those live
on the internal filesystems that are not going to be mounted anywhere.
They are not going to be _un_mounted, of course, so having to take the
slow path every time a pipe or socket gets closed is really obnoxious.
Solution had been to mark them as long-lived ones - essentially faking
"they are mounted somewhere" indicator.

With minor modification that works even for ones that do eventually get
dropped - all it takes is making sure we have an RCU delay between
clearing the "mounted somewhere" indicator and dropping the reference.

There are some additional twists (if you want to drop a dozen of such
internal mounts, you'd be better off with clearing the indicator on
all of them, doing an RCU delay once, then dropping the references),
but in the basic form it had been
	* use kern_mount() if you want your internal mount to be
a long-term one.
	* use kern_unmount() to undo that.

Unfortunately, the things did rot a bit during the mount API reshuffling.
In several cases we have lost the "fake the indicator" part; kern_unmount()
on the unmount side remained (it doesn't warn if you use it on a mount
without the indicator), but all benefits regaring mntput() cost had been
lost.

To get rid of that bitrot, let's add a new helper that would work
with fs_context-based API: fc_mount_longterm().  It's a counterpart
of fc_mount() that does, on success, mark its result as long-term.
It must be paired with kern_unmount() or equivalents.

Converted:
	1) mqueue (it used to use kern_mount_data() and the umount side
is still as it used to be)
	2) hugetlbfs (used to use kern_mount_data(), internal mount is
never unmounted in this one)
	3) i915 gemfs (used to be kern_mount() + manual remount to set
options, still uses kern_unmount() on umount side)
	4) v3d gemfs (copied from i915)

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 drivers/gpu/drm/i915/gem/i915_gemfs.c | 21 ++++++++++++++++++---
 drivers/gpu/drm/v3d/v3d_gemfs.c       | 21 ++++++++++++++++++---
 fs/hugetlbfs/inode.c                  |  2 +-
 fs/namespace.c                        |  9 +++++++++
 include/linux/mount.h                 |  1 +
 ipc/mqueue.c                          |  2 +-
 6 files changed, 48 insertions(+), 8 deletions(-)

diff --git a/drivers/gpu/drm/i915/gem/i915_gemfs.c b/drivers/gpu/drm/i915/gem/i915_gemfs.c
index 65d84a93c525..a09e2eb47175 100644
--- a/drivers/gpu/drm/i915/gem/i915_gemfs.c
+++ b/drivers/gpu/drm/i915/gem/i915_gemfs.c
@@ -5,16 +5,23 @@
 
 #include <linux/fs.h>
 #include <linux/mount.h>
+#include <linux/fs_context.h>
 
 #include "i915_drv.h"
 #include "i915_gemfs.h"
 #include "i915_utils.h"
 
+static int add_param(struct fs_context *fc, const char *key, const char *val)
+{
+	return vfs_parse_fs_string(fc, key, val, strlen(val));
+}
+
 void i915_gemfs_init(struct drm_i915_private *i915)
 {
-	char huge_opt[] = "huge=within_size"; /* r/w */
 	struct file_system_type *type;
+	struct fs_context *fc;
 	struct vfsmount *gemfs;
+	int ret;
 
 	/*
 	 * By creating our own shmemfs mountpoint, we can pass in
@@ -38,8 +45,16 @@ void i915_gemfs_init(struct drm_i915_private *i915)
 	if (!type)
 		goto err;
 
-	gemfs = vfs_kern_mount(type, SB_KERNMOUNT, type->name, huge_opt);
-	if (IS_ERR(gemfs))
+	fc = fs_context_for_mount(type, SB_KERNMOUNT);
+	if (IS_ERR(fc))
+		goto err;
+	ret = add_param(fc, "source", "tmpfs");
+	if (!ret)
+		ret = add_param(fc, "huge", "within_size");
+	if (!ret)
+		gemfs = fc_mount_longterm(fc);
+	put_fs_context(fc);
+	if (ret)
 		goto err;
 
 	i915->mm.gemfs = gemfs;
diff --git a/drivers/gpu/drm/v3d/v3d_gemfs.c b/drivers/gpu/drm/v3d/v3d_gemfs.c
index 4c5e18590a5c..8ec6ed82b3d9 100644
--- a/drivers/gpu/drm/v3d/v3d_gemfs.c
+++ b/drivers/gpu/drm/v3d/v3d_gemfs.c
@@ -3,14 +3,21 @@
 
 #include <linux/fs.h>
 #include <linux/mount.h>
+#include <linux/fs_context.h>
 
 #include "v3d_drv.h"
 
+static int add_param(struct fs_context *fc, const char *key, const char *val)
+{
+	return vfs_parse_fs_string(fc, key, val, strlen(val));
+}
+
 void v3d_gemfs_init(struct v3d_dev *v3d)
 {
-	char huge_opt[] = "huge=within_size";
 	struct file_system_type *type;
+	struct fs_context *fc;
 	struct vfsmount *gemfs;
+	int ret;
 
 	/*
 	 * By creating our own shmemfs mountpoint, we can pass in
@@ -28,8 +35,16 @@ void v3d_gemfs_init(struct v3d_dev *v3d)
 	if (!type)
 		goto err;
 
-	gemfs = vfs_kern_mount(type, SB_KERNMOUNT, type->name, huge_opt);
-	if (IS_ERR(gemfs))
+	fc = fs_context_for_mount(type, SB_KERNMOUNT);
+	if (IS_ERR(fc))
+		goto err;
+	ret = add_param(fc, "source", "tmpfs");
+	if (!ret)
+		ret = add_param(fc, "huge", "within_size");
+	if (!ret)
+		gemfs = fc_mount_longterm(fc);
+	put_fs_context(fc);
+	if (ret)
 		goto err;
 
 	v3d->gemfs = gemfs;
diff --git a/fs/hugetlbfs/inode.c b/fs/hugetlbfs/inode.c
index e4de5425838d..4e0397775167 100644
--- a/fs/hugetlbfs/inode.c
+++ b/fs/hugetlbfs/inode.c
@@ -1587,7 +1587,7 @@ static struct vfsmount *__init mount_one_hugetlbfs(struct hstate *h)
 	} else {
 		struct hugetlbfs_fs_context *ctx = fc->fs_private;
 		ctx->hstate = h;
-		mnt = fc_mount(fc);
+		mnt = fc_mount_longterm(fc);
 		put_fs_context(fc);
 	}
 	if (IS_ERR(mnt))
diff --git a/fs/namespace.c b/fs/namespace.c
index 5e82f1ef042a..166d60a6f66b 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -1296,6 +1296,15 @@ struct vfsmount *fc_mount(struct fs_context *fc)
 }
 EXPORT_SYMBOL(fc_mount);
 
+struct vfsmount *fc_mount_longterm(struct fs_context *fc)
+{
+	struct vfsmount *mnt = fc_mount(fc);
+	if (!IS_ERR(mnt))
+		real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
+	return mnt;
+}
+EXPORT_SYMBOL(fc_mount_longterm);
+
 struct vfsmount *vfs_kern_mount(struct file_system_type *type,
 				int flags, const char *name,
 				void *data)
diff --git a/include/linux/mount.h b/include/linux/mount.h
index 4880f434c021..1d30041b37e1 100644
--- a/include/linux/mount.h
+++ b/include/linux/mount.h
@@ -98,6 +98,7 @@ int mnt_get_write_access(struct vfsmount *mnt);
 void mnt_put_write_access(struct vfsmount *mnt);
 
 extern struct vfsmount *fc_mount(struct fs_context *fc);
+extern struct vfsmount *fc_mount_longterm(struct fs_context *fc);
 extern struct vfsmount *vfs_create_mount(struct fs_context *fc);
 extern struct vfsmount *vfs_kern_mount(struct file_system_type *type,
 				      int flags, const char *name,
diff --git a/ipc/mqueue.c b/ipc/mqueue.c
index 82ed2d3c9846..de7432efbf4a 100644
--- a/ipc/mqueue.c
+++ b/ipc/mqueue.c
@@ -482,7 +482,7 @@ static struct vfsmount *mq_create_mount(struct ipc_namespace *ns)
 	put_user_ns(fc->user_ns);
 	fc->user_ns = get_user_ns(ctx->ipc_ns->user_ns);
 
-	mnt = fc_mount(fc);
+	mnt = fc_mount_longterm(fc);
 	put_fs_context(fc);
 	return mnt;
 }
-- 
2.39.5

Re: [PATCH 11/26] sanitize handling of long-term internal mounts

[Christian Brauner]

On Tue, Jun 10, 2025 at 09:21:33AM +0100, Al Viro wrote:
> Original rationale for those had been the reduced cost of mntput()
> for the stuff that is mounted somewhere.  Mount refcount increments and
> decrements are frequent; what's worse, they tend to concentrate on the
> same instances and cacheline pingpong is quite noticable.
> 
> As the result, mount refcounts are per-cpu; that allows a very cheap
> increment.  Plain decrement would be just as easy, but decrement-and-test
> is anything but (we need to add the components up, with exclusion against
> possible increment-from-zero, etc.).
> 
> Fortunately, there is a very common case where we can tell that decrement
> won't be the final one - if the thing we are dropping is currently
> mounted somewhere.  We have an RCU delay between the removal from mount
> tree and dropping the reference that used to pin it there, so we can
> just take rcu_read_lock() and check if the victim is mounted somewhere.
> If it is, we can go ahead and decrement without and further checks -
> the reference we are dropping is not the last one.  If it isn't, we
> get all the fun with locking, carefully adding up components, etc.,
> but the majority of refcount decrements end up taking the fast path.
> 
> There is a major exception, though - pipes and sockets.  Those live
> on the internal filesystems that are not going to be mounted anywhere.
> They are not going to be _un_mounted, of course, so having to take the
> slow path every time a pipe or socket gets closed is really obnoxious.
> Solution had been to mark them as long-lived ones - essentially faking
> "they are mounted somewhere" indicator.
> 
> With minor modification that works even for ones that do eventually get
> dropped - all it takes is making sure we have an RCU delay between
> clearing the "mounted somewhere" indicator and dropping the reference.
> 
> There are some additional twists (if you want to drop a dozen of such
> internal mounts, you'd be better off with clearing the indicator on
> all of them, doing an RCU delay once, then dropping the references),
> but in the basic form it had been
> 	* use kern_mount() if you want your internal mount to be
> a long-term one.
> 	* use kern_unmount() to undo that.
> 
> Unfortunately, the things did rot a bit during the mount API reshuffling.
> In several cases we have lost the "fake the indicator" part; kern_unmount()
> on the unmount side remained (it doesn't warn if you use it on a mount
> without the indicator), but all benefits regaring mntput() cost had been
> lost.
> 
> To get rid of that bitrot, let's add a new helper that would work
> with fs_context-based API: fc_mount_longterm().  It's a counterpart
> of fc_mount() that does, on success, mark its result as long-term.
> It must be paired with kern_unmount() or equivalents.
> 
> Converted:
> 	1) mqueue (it used to use kern_mount_data() and the umount side
> is still as it used to be)
> 	2) hugetlbfs (used to use kern_mount_data(), internal mount is
> never unmounted in this one)
> 	3) i915 gemfs (used to be kern_mount() + manual remount to set
> options, still uses kern_unmount() on umount side)
> 	4) v3d gemfs (copied from i915)
> 
> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
> ---

Reviewed-by: Christian Brauner <brauner@kernel.org>

[PATCH 12/26] Rewrite of propagate_umount()

[Al Viro]

The variant currently in the tree has problems; trying to prove
correctness has caught at least one class of bugs (reparenting
that ends up moving the visible location of reparented mount, due
to not excluding some of the counterparts on propagation that
should've been included).

I tried to prove that it's the only bug there; I'm still not sure
whether it is.  If anyone can reconstruct and write down an analysis
of the mainline implementation, I'll gladly review it; as it is,
I ended up doing a different implementation.  Candidate collection
phase is similar, but trimming the set down until it satisfies the
constraints turned out pretty different.

I hoped to do transformation as a massage series, but that turns out
to be too convoluted.  So it's a single patch replacing propagate_umount()
and friends in one go, with notes and analysis in D/f/propagate_umount.txt
(in addition to inline comments).

As far I can tell, it is provably correct and provably linear by the number
of mounts we need to look at in order to decide what should be unmounted.
It even builds and seems to survive testing...

Another nice thing that fell out of that is that ->mnt_umounting is no longer
needed.

Compared to the first version:
	* explicit MNT_UMOUNT_CANDIDATE flag for is_candidate()
	* trim_ancestors() only clears that flag, leaving the suckers on list
	* trim_one() and handle_locked() take the stuff with flag cleared off
the list.  That allows to iterate with list_for_each_entry_safe() when calling
trim_one() - it removes at most one element from the list now.
	* no globals - I didn't bother with any kind of context, not worth it.

	* Notes updated accordingly; I have not touch the terms yet.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 .../filesystems/propagate_umount.txt          | 484 ++++++++++++++++++
 fs/mount.h                                    |   1 -
 fs/namespace.c                                |   1 -
 fs/pnode.c                                    | 363 +++++++------
 fs/pnode.h                                    |   2 +-
 include/linux/mount.h                         |   3 +-
 6 files changed, 686 insertions(+), 168 deletions(-)
 create mode 100644 Documentation/filesystems/propagate_umount.txt

diff --git a/Documentation/filesystems/propagate_umount.txt b/Documentation/filesystems/propagate_umount.txt
new file mode 100644
index 000000000000..5b48540b4059
--- /dev/null
+++ b/Documentation/filesystems/propagate_umount.txt
@@ -0,0 +1,484 @@
+	Notes on propagate_umount()
+
+Umount propagation starts with a set of mounts we are already going to
+take out.  Ideally, we would like to add all downstream cognates to
+that set - anything with the same mountpoint as one of the removed
+mounts and with parent that would receive events from the parent of that
+mount.  However, there are some constraints the resulting set must
+satisfy.
+
+It is convenient to define several properties of sets of mounts:
+
+1) A set S of mounts is non-shifting if for any mount X belonging
+to S all subtrees mounted strictly inside of X (i.e. not overmounting
+the root of X) contain only elements of S.
+
+2) A set S is non-revealing if all locked mounts that belong to S have
+parents that also belong to S.
+
+3) A set S is closed if it contains all children of its elements.
+
+The set of mounts taken out by umount(2) must be non-shifting and
+non-revealing; the first constraint is what allows to reparent
+any remaining mounts and the second is what prevents the exposure
+of any concealed mountpoints.
+
+propagate_umount() takes the original set as an argument and tries to
+extend that set.  The original set is a full subtree and its root is
+unlocked; what matters is that it's closed and non-revealing.
+Resulting set may not be closed; there might still be mounts outside
+of that set, but only on top of stacks of root-overmounting elements
+of set.  They can be reparented to the place where the bottom of
+stack is attached to a mount that will survive.  NOTE: doing that
+will violate a constraint on having no more than one mount with
+the same parent/mountpoint pair; however, the caller (umount_tree())
+will immediately remedy that - it may keep unmounted element attached
+to parent, but only if the parent itself is unmounted.  Since all
+conflicts created by reparenting have common parent *not* in the
+set and one side of the conflict (bottom of the stack of overmounts)
+is in the set, it will be resolved.  However, we rely upon umount_tree()
+doing that pretty much immediately after the call of propagate_umount().
+
+Algorithm is based on two statements:
+	1) for any set S, there is a maximal non-shifting subset of S
+and it can be calculated in O(#S) time.
+	2) for any non-shifting set S, there is a maximal non-revealing
+subset of S.  That subset is also non-shifting and it can be calculated
+in O(#S) time.
+
+		Finding candidates.
+
+We are given a closed set U and we want to find all mounts that have
+the same mountpoint as some mount m in U *and* whose parent receives
+propagation from the parent of the same mount m.  Naive implementation
+would be
+	S = {}
+	for each m in U
+		add m to S
+		p = parent(m)
+		for each q in Propagation(p) - {p}
+			child = look_up(q, mountpoint(m))
+			if child
+				add child to S
+but that can lead to excessive work - there might be propagation among the
+subtrees of U, in which case we'd end up examining the same candidates
+many times.  Since propagation is transitive, the same will happen to
+everything downstream of that candidate and it's not hard to construct
+cases where the approach above leads to the time quadratic by the actual
+number of candidates.
+
+Note that if we run into a candidate we'd already seen, it must've been
+added on an earlier iteration of the outer loop - all additions made
+during one iteration of the outer loop have different parents.  So
+if we find a child already added to the set, we know that everything
+in Propagation(parent(child)) with the same mountpoint has been already
+added.
+	S = {}
+	for each m in U
+		if m in S
+			continue
+		add m to S
+		p = parent(m)
+		q = propagation_next(p, p)
+		while q
+			child = look_up(q, mountpoint(m))
+			if child
+				if child in S
+					q = skip_them(q, p)
+					continue;
+				add child to S
+			q = propagation_next(q, p)
+where
+skip_them(q, p)
+	keep walking Propagation(p) from q until we find something
+	not in Propagation(q)
+
+would get rid of that problem, but we need a sane implementation of
+skip_them().  That's not hard to do - split propagation_next() into
+"down into mnt_slave_list" and "forward-and-up" parts, with the
+skip_them() being "repeat the forward-and-up part until we get NULL
+or something that isn't a peer of the one we are skipping".
+
+Note that there can be no absolute roots among the extra candidates -
+they all come from mount lookups.  Absolute root among the original
+set is _currently_ impossible, but it might be worth protecting
+against.
+
+		Maximal non-shifting subsets.
+
+Let's call a mount m in a set S forbidden in that set if there is a
+subtree mounted strictly inside m and containing mounts that do not
+belong to S.
+
+The set is non-shifting when none of its elements are forbidden in it.
+
+If mount m is forbidden in a set S, it is forbidden in any subset S' it
+belongs to.  In other words, it can't belong to any of the non-shifting
+subsets of S.  If we had a way to find a forbidden mount or show that
+there's none, we could use it to find the maximal non-shifting subset
+simply by finding and removing them until none remain.
+
+Suppose mount m is forbidden in S; then any mounts forbidden in S - {m}
+must have been forbidden in S itself.  Indeed, since m has descendents
+that do not belong to S, any subtree that fits into S will fit into
+S - {m} as well.
+
+So in principle we could go through elements of S, checking if they
+are forbidden in S and removing the ones that are.  Removals will
+not invalidate the checks done for earlier mounts - if they were not
+forbidden at the time we checked, they won't become forbidden later.
+It's too costly to be practical, but there is a similar approach that
+is linear by size of S.
+
+Let's say that mount x in a set S is forbidden by mount y, if
+	* both x and y belong to S.
+	* there is a chain of mounts starting at x and leaving S
+	  immediately after passing through y, with the first
+	  mountpoint strictly inside x.
+Note 1: x may be equal to y - that's the case when something not
+belonging to S is mounted strictly inside x.
+Note 2: if y does not belong to S, it can't forbid anything in S.
+Note 3: if y has no children outside of S, it can't forbid anything in S.
+
+It's easy to show that mount x is forbidden in S if and only if x is
+forbidden in S by some mount y.  And it's easy to find all mounts in S
+forbidden by a given mount.
+
+Consider the following operation:
+	Trim(S, m) = S - {x : x is forbidden by m in S}
+
+Note that if m does not belong to S or has no children outside of S we
+are guaranteed that Trim(S, m) is equal to S.
+
+The following is true: if x is forbidden by y in Trim(S, m), it was
+already forbidden by y in S.
+
+Proof: Suppose x is forbidden by y in Trim(S, m).  Then there is a
+chain of mounts (x_0 = x, ..., x_k = y, x_{k+1} = r), such that x_{k+1}
+is the first element that doesn't belong to Trim(S, m) and the
+mountpoint of x_1 is strictly inside x.  If mount r belongs to S, it must
+have been removed by Trim(S, m), i.e. it was forbidden in S by m.
+Then there was a mount chain from r to some child of m that stayed in
+S all the way until m, but that's impossible since x belongs to Trim(S, m)
+and prepending (x_0, ..., x_k) to that chain demonstrates that x is also
+prohibited in S by m, and thus can't belong to Trim(S, m).
+Therefore r can not belong to S and our chain demonstrates that
+x is prohibited by y in S.  QED.
+
+Corollary: no mount is forbidden by m in Trim(S, m).  Indeed, any
+such mount would have been forbidden by m in S and thus would have been
+in the part of S removed in Trim(S, m).
+
+Corollary: no mount is forbidden by m in Trim(Trim(S, m), n).  Indeed,
+any such would have to have been forbidden by m in Trim(S, m), which
+is impossible.
+
+Corollary: after
+	S = Trim(S, x_1)
+	S = Trim(S, x_2)
+	...
+	S = Trim(S, x_k)
+no mount remaining in S will be forbidden by either of x_1,...,x_k.
+
+The following will reduce S to its maximal non-shifting subset:
+	visited = {}
+	while S contains elements not belonging to visited
+		let m be an arbitrary such element of S
+		S = Trim(S, m)
+		add m to visited
+
+S never grows, so the number of elements of S not belonging to visited
+decreases at least by one on each iteration.  When the loop terminates,
+all mounts remaining in S belong to visited.  It's easy to see that at
+the beginning of each iteration no mount remaining in S will be forbidden
+by any element of visited.  In other words, no mount remaining in S will
+be forbidden, i.e. final value of S will be non-shifting.  It will be
+the maximal non-shifting subset, since we were removing only forbidden
+elements.
+
+	There are two difficulties in implementing the above in linear
+time, both due to the fact that Trim() might need to remove more than one
+element.  Naive implementation of Trim() is vulnerable to running into a
+long chain of mounts, each mounted on top of parent's root.  Nothing in
+that chain is prohibited, so nothing gets removed from it.  We need to
+recognize such chains and avoid walking them again on subsequent calls of
+Trim(), otherwise we will end up with worst-case time being quadratic by
+the number of elements in S.  Another difficulty is in implementing the
+outer loop - we need to iterate through all elements of a shrinking set.
+That would be trivial if we never removed more than one element at a time
+(linked list, with list_for_each_entry_safe for iterator), but we may
+need to remove more than one entry, possibly including the ones we have
+already visited.
+
+	Let's start with naive algorithm for Trim():
+
+Trim_one(m)
+	found = false
+	for each n in children(m)
+		if n not in S
+			found = true
+			if (mountpoint(n) != root(m))
+				remove m from S
+				break
+	if found
+		Trim_ancestors(m)
+
+Trim_ancestors(m)
+	for (; parent(m) in S; m = parent(m)) {
+		if (mountpoint(m) != root(parent(m)))
+			remove parent(m) from S
+	}
+
+If m belongs to S, Trim_one(m) will replace S with Trim(S, m).
+Proof:
+	Consider the chains excluding elements from Trim(S, m).  The last
+two elements in such chain are m and some child of m that does not belong
+to S.  If m has no such children, Trim(S, m) is equal to S.
+	m itself is removed if and only if the chain has exactly two
+elements, i.e. when the last element does not overmount the root of m.
+In other words, that happens when m has a child not in S that does not
+overmount the root of m.
+	All other elements to remove will be ancestors of m, such that
+the entire descent chain from them to m is contained in S.  Let
+(x_0, x_1, ..., x_k = m) be the longest such chain.  x_i needs to be
+removed if and only if x_{i+1} does not overmount its root.  It's easy
+to see that Trim_ancestors(m) will iterate through that chain from
+x_k to x_1 and that it will remove exactly the elements that need to be
+removed.
+
+	Note that if the loop in Trim_ancestors() walks into an already
+visited element, we are guaranteed that remaining iterations will see
+only elements that had already been visited and remove none of them.
+That's the weakness that makes it vulnerable to long chains of full
+overmounts.
+
+	It's easy to deal with, if we can afford setting marks on
+elements of S; we would mark all elements already visited by
+Trim_ancestors() and have it bail out as soon as it sees an already
+marked element.
+
+	The problems with iterating through the set can be dealt with in
+several ways, depending upon the representation we choose for our set.
+One useful observation is that we are given a closed subset in S - the
+original set passed to propagate_umount().  Its elements can neither
+prohibit anything nor be prohibited by anything - all their descendents
+belong to S, so they can not occur anywhere in any excluding chain.
+In other words, the elements of that subset will remain in S until
+the end and Trim_one(S, m) is a no-op for all m from that subset.
+
+	That suggests keeping S as a disjoint union of a closed set U
+('will be unmounted, no matter what') and the set of all elements of
+S that do not belong to U.  That set ('candidates') is all we need
+to iterate through.  Let's represent it as a subset in a cyclic list,
+consisting of all list elements that are marked as candidates (initially -
+all of them).  Then we could have Trim_ancestors() only remove the mark,
+leaving the elements on the list.  Then Trim_one() would never remove
+anything other than its argument from the containing list, allowing to
+use list_for_each_entry_safe() as iterator.
+
+	Assuming that representation we get the following:
+
+	list_for_each_entry_safe(m, ..., Candidates, ...)
+		Trim_one(m)
+where
+Trim_one(m)
+	if (m is not marked as a candidate)
+		strip the "seen by Trim_ancestors" mark from m
+		remove m from the Candidates list
+		return
+		
+	remove_this = false
+	found = false
+	for each n in children(m)
+		if n not in S
+			found = true
+			if (mountpoint(n) != root(m))
+				remove_this = true
+				break
+	if found
+		Trim_ancestors(m)
+	if remove_this
+		strip the "seen by Trim_ancestors" mark from m
+		strip the "candidate" mark from m
+		remove m from the Candidate list
+
+Trim_ancestors(m)
+	for (p = parent(m); p is marked as candidate ; m = p, p = parent(p)) {
+		if m is marked as seen by Trim_ancestors
+			return
+		mark m as seen by Trim_ancestors
+		if (mountpoint(m) != root(p))
+			strip the "candidate" mark from p
+	}
+
+	Terminating condition in the loop in Trim_ancestors() is correct,
+since that that loop will never run into p belonging to U - p is always
+an ancestor of argument of Trim_one() and since U is closed, the argument
+of Trim_one() would also have to belong to U.  But Trim_one() is never
+called for elements of U.  In other words, p belongs to S if and only
+if it belongs to candidates.
+
+	Time complexity:
+* we get no more than O(#S) calls of Trim_one()
+* the loop over children in Trim_one() never looks at the same child
+twice through all the calls.
+* iterations of that loop for children in S are no more than O(#S)
+in the worst case
+* at most two children that are not elements of S are considered per
+call of Trim_one().
+* the loop in Trim_ancestors() sets its mark once per iteration and
+no element of S has is set more than once.
+
+	In the end we may have some elements excluded from S by
+Trim_ancestors() still stuck on the list.  We could do a separate
+loop removing them from the list (also no worse than O(#S) time),
+but it's easier to leave that until the next phase - there we will
+iterate through the candidates anyway.
+
+	The caller has already removed all elements of U from their parents'
+lists of children, which means that checking if child belongs to S is
+equivalent to checking if it's marked as a candidate; we'll never see
+the elements of U in the loop over children in Trim_one().
+
+	What's more, if we see that children(m) is empty and m is not
+locked, we can immediately move m into the committed subset (remove
+from the parent's list of children, etc.).  That's one fewer mount we'll
+have to look into when we check the list of children of its parent *and*
+when we get to building the non-revealing subset.
+
+		Maximal non-revealing subsets
+
+If S is not a non-revealing subset, there is a locked element x in S
+such that parent of x is not in S.
+
+Obviously, no non-revealing subset of S may contain x.  Removing such
+elements one by one will obviously end with the maximal non-revealing
+subset (possibly empty one).  Note that removal of an element will
+require removal of all its locked children, etc.
+
+If the set had been non-shifting, it will remain non-shifting after
+such removals.
+Proof: suppose S was non-shifting, x is a locked element of S, parent of x
+is not in S and S - {x} is not non-shifting.  Then there is an element m
+in S - {x} and a subtree mounted strictly inside m, such that m contains
+an element not in in S - {x}.  Since S is non-shifting, everything in
+that subtree must belong to S.  But that means that this subtree must
+contain x somewhere *and* that parent of x either belongs that subtree
+or is equal to m.  Either way it must belong to S.  Contradiction.
+
+// same representation as for finding maximal non-shifting subsets:
+// S is a disjoint union of a non-revealing set U (the ones we are committed
+// to unmount) and a set of candidates, represented as a subset of list
+// elements that have "is a candidate" mark on them.
+// Elements of U are removed from their parents' lists of children.
+// In the end candidates becomes empty and maximal non-revealing non-shifting
+// subset of S is now in U
+	while (Candidates list is non-empty)
+		handle_locked(first(Candidates))
+
+handle_locked(m)
+	if m is not marked as a candidate
+		strip the "seen by Trim_ancestors" mark from m
+		remove m from the list
+		return
+	cutoff = m
+	for (p = m; p in candidates; p = parent(p)) {
+		strip the "seen by Trim_ancestors" mark from p
+		strip the "candidate" mark from p
+		remove p from the Candidates list
+		if (!locked(p))
+			cutoff = parent(p)
+	}
+	if p in U
+		cutoff = p
+	while m != cutoff
+		remove m from children(parent(m))
+		add m to U
+		m = parent(m)
+
+Let (x_0, ..., x_n = m) be the maximal chain of descent of m within S.
+* If it contains some elements of U, let x_k be the last one of those.
+Then union of U with {x_{k+1}, ..., x_n} is obviously non-revealing.
+* otherwise if all its elements are locked, then none of {x_0, ..., x_n}
+may be elements of a non-revealing subset of S.
+* otherwise let x_k be the first unlocked element of the chain.  Then none
+of {x_0, ..., x_{k-1}} may be an element of a non-revealing subset of
+S and union of U and {x_k, ..., x_n} is non-revealing.
+
+handle_locked(m) finds which of these cases applies and adjusts Candidates
+and U accordingly.  U remains non-revealing, union of Candidates and
+U still contains any non-revealing subset of S and after the call of
+handle_locked(m) m is guaranteed to be not in Candidates list.  So having
+it called for each element of S would suffice to empty Candidates,
+leaving U the maximal non-revealing subset of S.
+
+However, handle_locked(m) is a no-op when m belongs to U, so it's enough
+to have it called for elements of Candidates list until none remain.
+
+Time complexity: number of calls of handle_locked() is limited by
+#Candidates, each iteration of the first loop in handle_locked() removes
+an element from the list, so their total number of executions is also
+limited by #Candidates; number of iterations in the second loop is no
+greater than the number of iterations of the first loop.
+
+
+		Reparenting
+
+After we'd calculated the final set, we still need to deal with
+reparenting - if an element of the final set has a child not in it,
+we need to reparent such child.
+
+Such children can only be root-overmounting (otherwise the set wouldn't
+be non-shifting) and their parents can not belong to the original set,
+since the original is guaranteed to be closed.
+
+
+		Putting all of that together
+
+The plan is to
+	* find all candidates
+	* trim down to maximal non-shifting subset
+	* trim down to maximal non-revealing subset
+	* reparent anything that needs to be reparented
+	* return the resulting set to the caller
+
+For the 2nd and 3rd steps we want to separate the set into growing
+non-revealing subset, initially containing the original set ("U" in
+terms of the pseudocode above) and everything we are still not sure about
+("candidates").  It means that for the output of the 1st step we'd like
+the extra candidates separated from the stuff already in the original set.
+For the 4th step we would like the additions to U separate from the
+original set.
+
+So let's go for
+	* original set ("set").  Linkage via mnt_list
+	* undecided candidates ("candidates").  Subset of a list,
+consisting of all its elements marked with a new flag (MNT_UMOUNT_CANDIDATE).
+Initially all elements of the list will be marked that way; in the
+end the list will become empty and no mounts will remain marked with
+that flag.
+	* Reuse MNT_MARKED for "has been already seen by trim_ancestors()".
+	* anything in U that hadn't been in the original set - elements of
+candidates will gradually be either discarded or moved there.  In other
+words, it's the candidates we have already decided to unmount.	Its role
+is reasonably close to the old "to_umount", so let's use that name.
+Linkage via mnt_list.
+
+For gather_candidates() we'll need to maintain both candidates (S -
+set) and intersection of S with set.  Use MNT_UMOUNT_CANDIDATE for
+all elements we encounter, putting the ones not already in the original
+set into the list of candidates.  When we are done, strip that flag from
+all elements of the original set.  That gives a cheap way to check
+if element belongs to S (in gather_candidates) and to candidates
+itself (at later stages).  Call that predicate is_candidate(); it would
+be m->mnt_flags & MNT_UMOUNT_CANDIDATE.
+
+All elements of the original set are marked with MNT_UMOUNT and we'll
+need the same for elements added when joining the contents of to_umount
+to set in the end.  Let's set MNT_UMOUNT at the time we add an element
+to to_umount; that's close to what the old 'umount_one' is doing, so
+let's keep that name.  It also gives us another predicate we need -
+"belongs to union of set and to_umount"; will_be_unmounted() for now.
+
+Removals from the candidates list should strip both MNT_MARKED and
+MNT_UMOUNT_CANDIDATE; call it remove_from_candidates_list().
diff --git a/fs/mount.h b/fs/mount.h
index 18fa88ad752a..9b3de2eef68a 100644
--- a/fs/mount.h
+++ b/fs/mount.h
@@ -79,7 +79,6 @@ struct mount {
 		struct hlist_node mnt_mp_list;	/* list mounts with the same mountpoint */
 		struct hlist_node mnt_umount;
 	};
-	struct list_head mnt_umounting; /* list entry for umount propagation */
 #ifdef CONFIG_FSNOTIFY
 	struct fsnotify_mark_connector __rcu *mnt_fsnotify_marks;
 	__u32 mnt_fsnotify_mask;
diff --git a/fs/namespace.c b/fs/namespace.c
index 166d60a6f66b..1ec7b1d63e94 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -383,7 +383,6 @@ static struct mount *alloc_vfsmnt(const char *name)
 		INIT_LIST_HEAD(&mnt->mnt_slave_list);
 		INIT_LIST_HEAD(&mnt->mnt_slave);
 		INIT_HLIST_NODE(&mnt->mnt_mp_list);
-		INIT_LIST_HEAD(&mnt->mnt_umounting);
 		INIT_HLIST_HEAD(&mnt->mnt_stuck_children);
 		RB_CLEAR_NODE(&mnt->mnt_node);
 		mnt->mnt.mnt_idmap = &nop_mnt_idmap;
diff --git a/fs/pnode.c b/fs/pnode.c
index aa187144e389..f1752dd499af 100644
--- a/fs/pnode.c
+++ b/fs/pnode.c
@@ -24,11 +24,6 @@ static inline struct mount *first_slave(struct mount *p)
 	return list_entry(p->mnt_slave_list.next, struct mount, mnt_slave);
 }
 
-static inline struct mount *last_slave(struct mount *p)
-{
-	return list_entry(p->mnt_slave_list.prev, struct mount, mnt_slave);
-}
-
 static inline struct mount *next_slave(struct mount *p)
 {
 	return list_entry(p->mnt_slave.next, struct mount, mnt_slave);
@@ -136,6 +131,23 @@ void change_mnt_propagation(struct mount *mnt, int type)
 	}
 }
 
+static struct mount *__propagation_next(struct mount *m,
+					 struct mount *origin)
+{
+	while (1) {
+		struct mount *master = m->mnt_master;
+
+		if (master == origin->mnt_master) {
+			struct mount *next = next_peer(m);
+			return (next == origin) ? NULL : next;
+		} else if (m->mnt_slave.next != &master->mnt_slave_list)
+			return next_slave(m);
+
+		/* back at master */
+		m = master;
+	}
+}
+
 /*
  * get the next mount in the propagation tree.
  * @m: the mount seen last
@@ -153,31 +165,21 @@ static struct mount *propagation_next(struct mount *m,
 	if (!IS_MNT_NEW(m) && !list_empty(&m->mnt_slave_list))
 		return first_slave(m);
 
-	while (1) {
-		struct mount *master = m->mnt_master;
-
-		if (master == origin->mnt_master) {
-			struct mount *next = next_peer(m);
-			return (next == origin) ? NULL : next;
-		} else if (m->mnt_slave.next != &master->mnt_slave_list)
-			return next_slave(m);
-
-		/* back at master */
-		m = master;
-	}
+	return __propagation_next(m, origin);
 }
 
 static struct mount *skip_propagation_subtree(struct mount *m,
 						struct mount *origin)
 {
 	/*
-	 * Advance m such that propagation_next will not return
-	 * the slaves of m.
+	 * Advance m past everything that gets propagation from it.
 	 */
-	if (!IS_MNT_NEW(m) && !list_empty(&m->mnt_slave_list))
-		m = last_slave(m);
+	struct mount *p = __propagation_next(m, origin);
+
+	while (p && peers(m, p))
+		p = __propagation_next(p, origin);
 
-	return m;
+	return p;
 }
 
 static struct mount *next_group(struct mount *m, struct mount *origin)
@@ -458,181 +460,214 @@ void propagate_mount_unlock(struct mount *mnt)
 	}
 }
 
-static void umount_one(struct mount *mnt, struct list_head *to_umount)
+static inline bool is_candidate(struct mount *m)
 {
-	CLEAR_MNT_MARK(mnt);
-	mnt->mnt.mnt_flags |= MNT_UMOUNT;
-	list_del_init(&mnt->mnt_child);
-	list_del_init(&mnt->mnt_umounting);
-	move_from_ns(mnt, to_umount);
+	return m->mnt.mnt_flags & MNT_UMOUNT_CANDIDATE;
 }
 
-/*
- * NOTE: unmounting 'mnt' naturally propagates to all other mounts its
- * parent propagates to.
- */
-static bool __propagate_umount(struct mount *mnt,
-			       struct list_head *to_umount,
-			       struct list_head *to_restore)
+static inline bool will_be_unmounted(struct mount *m)
 {
-	bool progress = false;
-	struct mount *child;
+	return m->mnt.mnt_flags & MNT_UMOUNT;
+}
 
-	/*
-	 * The state of the parent won't change if this mount is
-	 * already unmounted or marked as without children.
-	 */
-	if (mnt->mnt.mnt_flags & (MNT_UMOUNT | MNT_MARKED))
-		goto out;
+static void umount_one(struct mount *m, struct list_head *to_umount)
+{
+	m->mnt.mnt_flags |= MNT_UMOUNT;
+	list_del_init(&m->mnt_child);
+	move_from_ns(m, to_umount);
+}
 
-	/* Verify topper is the only grandchild that has not been
-	 * speculatively unmounted.
-	 */
-	list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
-		if (child->mnt_mountpoint == mnt->mnt.mnt_root)
-			continue;
-		if (!list_empty(&child->mnt_umounting) && IS_MNT_MARKED(child))
-			continue;
-		/* Found a mounted child */
-		goto children;
-	}
+static void remove_from_candidate_list(struct mount *m)
+{
+	m->mnt.mnt_flags &= ~(MNT_MARKED | MNT_UMOUNT_CANDIDATE);
+	list_del_init(&m->mnt_list);
+}
 
-	/* Mark mounts that can be unmounted if not locked */
-	SET_MNT_MARK(mnt);
-	progress = true;
+static void gather_candidates(struct list_head *set,
+			      struct list_head *candidates)
+{
+	struct mount *m, *p, *q;
 
-	/* If a mount is without children and not locked umount it. */
-	if (!IS_MNT_LOCKED(mnt)) {
-		umount_one(mnt, to_umount);
-	} else {
-children:
-		list_move_tail(&mnt->mnt_umounting, to_restore);
+	list_for_each_entry(m, set, mnt_list) {
+		if (is_candidate(m))
+			continue;
+		m->mnt.mnt_flags |= MNT_UMOUNT_CANDIDATE;
+		p = m->mnt_parent;
+		q = propagation_next(p, p);
+		while (q) {
+			struct mount *child = __lookup_mnt(&q->mnt,
+							   m->mnt_mountpoint);
+			if (child) {
+				/*
+				 * We might've already run into this one.  That
+				 * must've happened on earlier iteration of the
+				 * outer loop; in that case we can skip those
+				 * parents that get propagation from q - there
+				 * will be nothing new on those as well.
+				 */
+				if (is_candidate(child)) {
+					q = skip_propagation_subtree(q, p);
+					continue;
+				}
+				child->mnt.mnt_flags |= MNT_UMOUNT_CANDIDATE;
+				if (!will_be_unmounted(child))
+					list_add(&child->mnt_list, candidates);
+			}
+			q = propagation_next(q, p);
+		}
 	}
-out:
-	return progress;
+	list_for_each_entry(m, set, mnt_list)
+		m->mnt.mnt_flags &= ~MNT_UMOUNT_CANDIDATE;
 }
 
-static void umount_list(struct list_head *to_umount,
-			struct list_head *to_restore)
+/*
+ * We know that some child of @m can't be unmounted.  In all places where the
+ * chain of descent of @m has child not overmounting the root of parent,
+ * the parent can't be unmounted either.
+ */
+static void trim_ancestors(struct mount *m)
 {
-	struct mount *mnt, *child, *tmp;
-	list_for_each_entry(mnt, to_umount, mnt_list) {
-		list_for_each_entry_safe(child, tmp, &mnt->mnt_mounts, mnt_child) {
-			/* topper? */
-			if (child->mnt_mountpoint == mnt->mnt.mnt_root)
-				list_move_tail(&child->mnt_umounting, to_restore);
-			else
-				umount_one(child, to_umount);
-		}
+	struct mount *p;
+
+	for (p = m->mnt_parent; is_candidate(p); m = p, p = p->mnt_parent) {
+		if (IS_MNT_MARKED(m))	// all candidates beneath are overmounts
+			return;
+		SET_MNT_MARK(m);
+		if (m->mnt_mountpoint != p->mnt.mnt_root)
+			p->mnt.mnt_flags &= ~MNT_UMOUNT_CANDIDATE;
 	}
 }
 
-static void restore_mounts(struct list_head *to_restore)
+/*
+ * Find and exclude all umount candidates forbidden by @m
+ * (see Documentation/filesystems/propagate_umount.txt)
+ * If we can immediately tell that @m is OK to unmount (unlocked
+ * and all children are already committed to unmounting) commit
+ * to unmounting it.
+ * Only @m itself might be taken from the candidates list;
+ * anything found by trim_ancestors() is marked non-candidate
+ * and left on the list.
+ */
+static void trim_one(struct mount *m, struct list_head *to_umount)
 {
-	/* Restore mounts to a clean working state */
-	while (!list_empty(to_restore)) {
-		struct mount *mnt, *parent;
-		struct mountpoint *mp;
-
-		mnt = list_first_entry(to_restore, struct mount, mnt_umounting);
-		CLEAR_MNT_MARK(mnt);
-		list_del_init(&mnt->mnt_umounting);
-
-		/* Should this mount be reparented? */
-		mp = mnt->mnt_mp;
-		parent = mnt->mnt_parent;
-		while (parent->mnt.mnt_flags & MNT_UMOUNT) {
-			mp = parent->mnt_mp;
-			parent = parent->mnt_parent;
-		}
-		if (parent != mnt->mnt_parent) {
-			mnt_change_mountpoint(parent, mp, mnt);
-			mnt_notify_add(mnt);
+	bool remove_this = false, found = false, umount_this = false;
+	struct mount *n;
+
+	if (!is_candidate(m)) { // trim_ancestors() left it on list
+		remove_from_candidate_list(m);
+		return;
+	}
+
+	list_for_each_entry(n, &m->mnt_mounts, mnt_child) {
+		if (!is_candidate(n)) {
+			found = true;
+			if (n->mnt_mountpoint != m->mnt.mnt_root) {
+				remove_this = true;
+				break;
+			}
 		}
 	}
+	if (found) {
+		trim_ancestors(m);
+	} else if (!IS_MNT_LOCKED(m) && list_empty(&m->mnt_mounts)) {
+		remove_this = true;
+		umount_this = true;
+	}
+	if (remove_this) {
+		remove_from_candidate_list(m);
+		if (umount_this)
+			umount_one(m, to_umount);
+	}
 }
 
-static void cleanup_umount_visitations(struct list_head *visited)
+static void handle_locked(struct mount *m, struct list_head *to_umount)
 {
-	while (!list_empty(visited)) {
-		struct mount *mnt =
-			list_first_entry(visited, struct mount, mnt_umounting);
-		list_del_init(&mnt->mnt_umounting);
+	struct mount *cutoff = m, *p;
+
+	if (!is_candidate(m)) { // trim_ancestors() left it on list
+		remove_from_candidate_list(m);
+		return;
+	}
+	for (p = m; is_candidate(p); p = p->mnt_parent) {
+		remove_from_candidate_list(p);
+		if (!IS_MNT_LOCKED(p))
+			cutoff = p->mnt_parent;
+	}
+	if (will_be_unmounted(p))
+		cutoff = p;
+	while (m != cutoff) {
+		umount_one(m, to_umount);
+		m = m->mnt_parent;
 	}
 }
 
 /*
- * collect all mounts that receive propagation from the mount in @list,
- * and return these additional mounts in the same list.
- * @list: the list of mounts to be unmounted.
+ * @m is not to going away, and it overmounts the top of a stack of mounts
+ * that are going away.  We know that all of those are fully overmounted
+ * by the one above (@m being the topmost of the chain), so @m can be slid
+ * in place where the bottom of the stack is attached.
  *
- * vfsmount lock must be held for write
+ * NOTE: here we temporarily violate a constraint - two mounts end up with
+ * the same parent and mountpoint; that will be remedied as soon as we
+ * return from propagate_umount() - its caller (umount_tree()) will detach
+ * the stack from the parent it (and now @m) is attached to.  umount_tree()
+ * might choose to keep unmounted pieces stuck to each other, but it always
+ * detaches them from the mounts that remain in the tree.
  */
-int propagate_umount(struct list_head *list)
+static void reparent(struct mount *m)
 {
-	struct mount *mnt;
-	LIST_HEAD(to_restore);
-	LIST_HEAD(to_umount);
-	LIST_HEAD(visited);
-
-	/* Find candidates for unmounting */
-	list_for_each_entry_reverse(mnt, list, mnt_list) {
-		struct mount *parent = mnt->mnt_parent;
-		struct mount *m;
+	struct mount *p = m;
+	struct mountpoint *mp;
 
-		/*
-		 * If this mount has already been visited it is known that it's
-		 * entire peer group and all of their slaves in the propagation
-		 * tree for the mountpoint has already been visited and there is
-		 * no need to visit them again.
-		 */
-		if (!list_empty(&mnt->mnt_umounting))
-			continue;
+	do {
+		mp = p->mnt_mp;
+		p = p->mnt_parent;
+	} while (will_be_unmounted(p));
 
-		list_add_tail(&mnt->mnt_umounting, &visited);
-		for (m = propagation_next(parent, parent); m;
-		     m = propagation_next(m, parent)) {
-			struct mount *child = __lookup_mnt(&m->mnt,
-							   mnt->mnt_mountpoint);
-			if (!child)
-				continue;
+	mnt_change_mountpoint(p, mp, m);
+	mnt_notify_add(m);
+}
 
-			if (!list_empty(&child->mnt_umounting)) {
-				/*
-				 * If the child has already been visited it is
-				 * know that it's entire peer group and all of
-				 * their slaves in the propgation tree for the
-				 * mountpoint has already been visited and there
-				 * is no need to visit this subtree again.
-				 */
-				m = skip_propagation_subtree(m, parent);
-				continue;
-			} else if (child->mnt.mnt_flags & MNT_UMOUNT) {
-				/*
-				 * We have come across a partially unmounted
-				 * mount in a list that has not been visited
-				 * yet. Remember it has been visited and
-				 * continue about our merry way.
-				 */
-				list_add_tail(&child->mnt_umounting, &visited);
-				continue;
-			}
+/**
+ * propagate_umount - apply propagation rules to the set of mounts for umount()
+ * @set: the list of mounts to be unmounted.
+ *
+ * Collect all mounts that receive propagation from the mount in @set and have
+ * no obstacles to being unmounted.  Add these additional mounts to the set.
+ *
+ * See Documentation/filesystems/propagate_umount.txt if you do anything in
+ * this area.
+ *
+ * Locks held:
+ * mount_lock (write_seqlock), namespace_sem (exclusive).
+ */
+void propagate_umount(struct list_head *set)
+{
+	struct mount *m, *p;
+	LIST_HEAD(to_umount);	// committed to unmounting
+	LIST_HEAD(candidates);	// undecided umount candidates
 
-			/* Check the child and parents while progress is made */
-			while (__propagate_umount(child,
-						  &to_umount, &to_restore)) {
-				/* Is the parent a umount candidate? */
-				child = child->mnt_parent;
-				if (list_empty(&child->mnt_umounting))
-					break;
-			}
-		}
+	// collect all candidates
+	gather_candidates(set, &candidates);
+
+	// reduce the set until it's non-shifting
+	list_for_each_entry_safe(m, p, &candidates, mnt_list)
+		trim_one(m, &to_umount);
+
+	// ... and non-revealing
+	while (!list_empty(&candidates)) {
+		m = list_first_entry(&candidates,struct mount, mnt_list);
+		handle_locked(m, &to_umount);
 	}
 
-	umount_list(&to_umount, &to_restore);
-	restore_mounts(&to_restore);
-	cleanup_umount_visitations(&visited);
-	list_splice_tail(&to_umount, list);
+	// now to_umount consists of all acceptable candidates
+	// deal with reparenting of remaining overmounts on those
+	list_for_each_entry(m, &to_umount, mnt_list) {
+		while (!list_empty(&m->mnt_mounts)) // should be at most one
+			reparent(list_first_entry(&m->mnt_mounts,
+						  struct mount, mnt_child));
+	}
 
-	return 0;
+	// and fold them into the set
+	list_splice_tail_init(&to_umount, set);
 }
diff --git a/fs/pnode.h b/fs/pnode.h
index b728b5de4adc..bfc10c095cbf 100644
--- a/fs/pnode.h
+++ b/fs/pnode.h
@@ -44,7 +44,7 @@ static inline bool peers(const struct mount *m1, const struct mount *m2)
 void change_mnt_propagation(struct mount *, int);
 int propagate_mnt(struct mount *, struct mountpoint *, struct mount *,
 		struct hlist_head *);
-int propagate_umount(struct list_head *);
+void propagate_umount(struct list_head *);
 int propagate_mount_busy(struct mount *, int);
 void propagate_mount_unlock(struct mount *);
 void mnt_release_group_id(struct mount *);
diff --git a/include/linux/mount.h b/include/linux/mount.h
index 1d30041b37e1..cae7324650b6 100644
--- a/include/linux/mount.h
+++ b/include/linux/mount.h
@@ -40,6 +40,7 @@ enum mount_flags {
 
 	MNT_INTERNAL	= 0x4000,
 
+	MNT_UMOUNT_CANDIDATE	= 0x020000,
 	MNT_LOCK_ATIME		= 0x040000,
 	MNT_LOCK_NOEXEC		= 0x080000,
 	MNT_LOCK_NOSUID		= 0x100000,
@@ -66,7 +67,7 @@ enum mount_flags {
 
 	MNT_INTERNAL_FLAGS = MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL |
 			     MNT_DOOMED | MNT_SYNC_UMOUNT | MNT_MARKED |
-			     MNT_LOCKED,
+			     MNT_LOCKED | MNT_UMOUNT_CANDIDATE,
 };
 
 struct vfsmount {
-- 
2.39.5

Re: [PATCH 12/26] Rewrite of propagate_umount()

[Christian Brauner]

On Tue, Jun 10, 2025 at 09:21:34AM +0100, Al Viro wrote:
> The variant currently in the tree has problems; trying to prove
> correctness has caught at least one class of bugs (reparenting
> that ends up moving the visible location of reparented mount, due
> to not excluding some of the counterparts on propagation that
> should've been included).
> 
> I tried to prove that it's the only bug there; I'm still not sure
> whether it is.  If anyone can reconstruct and write down an analysis
> of the mainline implementation, I'll gladly review it; as it is,
> I ended up doing a different implementation.  Candidate collection
> phase is similar, but trimming the set down until it satisfies the
> constraints turned out pretty different.
> 
> I hoped to do transformation as a massage series, but that turns out
> to be too convoluted.  So it's a single patch replacing propagate_umount()
> and friends in one go, with notes and analysis in D/f/propagate_umount.txt
> (in addition to inline comments).
> 
> As far I can tell, it is provably correct and provably linear by the number
> of mounts we need to look at in order to decide what should be unmounted.
> It even builds and seems to survive testing...
> 
> Another nice thing that fell out of that is that ->mnt_umounting is no longer
> needed.
> 
> Compared to the first version:
> 	* explicit MNT_UMOUNT_CANDIDATE flag for is_candidate()
> 	* trim_ancestors() only clears that flag, leaving the suckers on list
> 	* trim_one() and handle_locked() take the stuff with flag cleared off
> the list.  That allows to iterate with list_for_each_entry_safe() when calling
> trim_one() - it removes at most one element from the list now.
> 	* no globals - I didn't bother with any kind of context, not worth it.
> 
> 	* Notes updated accordingly; I have not touch the terms yet.
> 
> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
> ---

Reviewed-by: Christian Brauner <brauner@kernel.org>

[PATCH 13/26] attach_mnt(): expand in attach_recursive_mnt(), then lose the flag argument

[Al Viro]

simpler that way - all but one caller pass false as 'beneath' argument,
and that one caller is actually happier with the call expanded - the
logics with choice of mountpoint is identical for 'moving' and 'attaching'
cases, and now that is no longer hidden.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 37 ++++++++++++-------------------------
 1 file changed, 12 insertions(+), 25 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index 1ec7b1d63e94..409ffbf35d7d 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -1115,16 +1115,10 @@ static void __attach_mnt(struct mount *mnt, struct mount *parent)
  * @parent:  the parent
  * @mnt:     the new mount
  * @mp:      the new mountpoint
- * @beneath: whether to mount @mnt beneath or on top of @parent
  *
- * If @beneath is false, mount @mnt at @mp on @parent. Then attach @mnt
+ * Mount @mnt at @mp on @parent. Then attach @mnt
  * to @parent's child mount list and to @mount_hashtable.
  *
- * If @beneath is true, remove @mnt from its current parent and
- * mountpoint and mount it on @mp on @parent, and mount @parent on the
- * old parent and old mountpoint of @mnt. Finally, attach @parent to
- * @mnt_hashtable and @parent->mnt_parent->mnt_mounts.
- *
  * Note, when __attach_mnt() is called @mnt->mnt_parent already points
  * to the correct parent.
  *
@@ -1132,18 +1126,9 @@ static void __attach_mnt(struct mount *mnt, struct mount *parent)
  *          to have been acquired in that order.
  */
 static void attach_mnt(struct mount *mnt, struct mount *parent,
-		       struct mountpoint *mp, bool beneath)
+		       struct mountpoint *mp)
 {
-	if (beneath)
-		mnt_set_mountpoint_beneath(mnt, parent, mp);
-	else
-		mnt_set_mountpoint(parent, mp, mnt);
-	/*
-	 * Note, @mnt->mnt_parent has to be used. If @mnt was mounted
-	 * beneath @parent then @mnt will need to be attached to
-	 * @parent's old parent, not @parent. IOW, @mnt->mnt_parent
-	 * isn't the same mount as @parent.
-	 */
+	mnt_set_mountpoint(parent, mp, mnt);
 	__attach_mnt(mnt, mnt->mnt_parent);
 }
 
@@ -1156,7 +1141,7 @@ void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct m
 	hlist_del_init(&mnt->mnt_mp_list);
 	hlist_del_init_rcu(&mnt->mnt_hash);
 
-	attach_mnt(mnt, parent, mp, false);
+	attach_mnt(mnt, parent, mp);
 
 	put_mountpoint(old_mp);
 	mnt_add_count(old_parent, -1);
@@ -2308,7 +2293,7 @@ struct mount *copy_tree(struct mount *src_root, struct dentry *dentry,
 			if (src_mnt->mnt.mnt_flags & MNT_LOCKED)
 				dst_mnt->mnt.mnt_flags |= MNT_LOCKED;
 			list_add_tail(&dst_mnt->mnt_list, &res->mnt_list);
-			attach_mnt(dst_mnt, dst_parent, src_parent->mnt_mp, false);
+			attach_mnt(dst_mnt, dst_parent, src_parent->mnt_mp);
 			unlock_mount_hash();
 		}
 	}
@@ -2701,10 +2686,12 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 	}
 
 	if (moving) {
-		if (beneath)
-			dest_mp = smp;
 		unhash_mnt(source_mnt);
-		attach_mnt(source_mnt, top_mnt, dest_mp, beneath);
+		if (beneath)
+			mnt_set_mountpoint_beneath(source_mnt, top_mnt, smp);
+		else
+			mnt_set_mountpoint(top_mnt, dest_mp, source_mnt);
+		__attach_mnt(source_mnt, source_mnt->mnt_parent);
 		mnt_notify_add(source_mnt);
 		touch_mnt_namespace(source_mnt->mnt_ns);
 	} else {
@@ -4783,9 +4770,9 @@ SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
 		root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
 	}
 	/* mount old root on put_old */
-	attach_mnt(root_mnt, old_mnt, old_mp, false);
+	attach_mnt(root_mnt, old_mnt, old_mp);
 	/* mount new_root on / */
-	attach_mnt(new_mnt, root_parent, root_mp, false);
+	attach_mnt(new_mnt, root_parent, root_mp);
 	mnt_add_count(root_parent, -1);
 	touch_mnt_namespace(current->nsproxy->mnt_ns);
 	/* A moved mount should not expire automatically */
-- 
2.39.5

Re: [PATCH 13/26] attach_mnt(): expand in attach_recursive_mnt(), then lose the flag argument

[Christian Brauner]

On Tue, Jun 10, 2025 at 09:21:35AM +0100, Al Viro wrote:
> simpler that way - all but one caller pass false as 'beneath' argument,
> and that one caller is actually happier with the call expanded - the
> logics with choice of mountpoint is identical for 'moving' and 'attaching'
> cases, and now that is no longer hidden.
> 
> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
> ---

Reviewed-by: Christian Brauner <brauner@kernel.org>

[PATCH 14/26] do_move_mount(): take dropping the old mountpoint into attach_recursive_mnt()

[Al Viro]

... and fold it with unhash_mnt() there - there's no need to retain a reference
to old_mp beyond that point, since by then all mountpoints we were going to add
are either explicitly pinned by get_mountpoint() or have stuff already added
to them.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 7 ++-----
 1 file changed, 2 insertions(+), 5 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index 409ffbf35d7d..b176075ad833 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -2686,7 +2686,7 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 	}
 
 	if (moving) {
-		unhash_mnt(source_mnt);
+		umount_mnt(source_mnt);
 		if (beneath)
 			mnt_set_mountpoint_beneath(source_mnt, top_mnt, smp);
 		else
@@ -3596,7 +3596,7 @@ static int do_move_mount(struct path *old_path,
 	struct mount *p;
 	struct mount *old;
 	struct mount *parent;
-	struct mountpoint *mp, *old_mp;
+	struct mountpoint *mp;
 	int err;
 	bool attached, beneath = flags & MNT_TREE_BENEATH;
 
@@ -3610,7 +3610,6 @@ static int do_move_mount(struct path *old_path,
 	attached = mnt_has_parent(old);
 	if (attached)
 		flags |= MNT_TREE_MOVE;
-	old_mp = old->mnt_mp;
 	ns = old->mnt_ns;
 
 	err = -EINVAL;
@@ -3685,8 +3684,6 @@ static int do_move_mount(struct path *old_path,
 	/* if the mount is moved, it should no longer be expire
 	 * automatically */
 	list_del_init(&old->mnt_expire);
-	if (attached)
-		put_mountpoint(old_mp);
 out:
 	unlock_mount(mp);
 	if (!err) {
-- 
2.39.5

Re: [PATCH 14/26] do_move_mount(): take dropping the old mountpoint into attach_recursive_mnt()

[Christian Brauner]

On Tue, Jun 10, 2025 at 09:21:36AM +0100, Al Viro wrote:
> ... and fold it with unhash_mnt() there - there's no need to retain a reference
> to old_mp beyond that point, since by then all mountpoints we were going to add
> are either explicitly pinned by get_mountpoint() or have stuff already added
> to them.
> 
> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
> ---

Reviewed-by: Christian Brauner <brauner@kernel.org>

[PATCH 15/26] get rid of mnt_set_mountpoint_beneath()

[Al Viro]

mnt_set_mountpoint_beneath() consists of attaching new mount side-by-side
with the one we want to mount beneath (by mnt_set_mountpoint()), followed
by mnt_change_mountpoint() shifting the the top mount onto the new one
(by mnt_change_mountpoint()).

Both callers of mnt_set_mountpoint_beneath (both in attach_recursive_mnt())
have the same form - in 'beneath' case we call mnt_set_mountpoint_beneath(),
otherwise - mnt_set_mountpoint().

The thing is, expressing that as unconditional mnt_set_mountpoint(),
followed, in 'beneath' case, by mnt_change_mountpoint() is just as easy.
And these mnt_change_mountpoint() callers are similar to the ones we
do when it comes to attaching propagated copies, which will allow more
cleanups in the next commits.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 37 ++++---------------------------------
 1 file changed, 4 insertions(+), 33 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index b176075ad833..22bbc30a2da1 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -1075,33 +1075,6 @@ void mnt_set_mountpoint(struct mount *mnt,
 	hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list);
 }
 
-/**
- * mnt_set_mountpoint_beneath - mount a mount beneath another one
- *
- * @new_parent: the source mount
- * @top_mnt:    the mount beneath which @new_parent is mounted
- * @new_mp:     the new mountpoint of @top_mnt on @new_parent
- *
- * Remove @top_mnt from its current mountpoint @top_mnt->mnt_mp and
- * parent @top_mnt->mnt_parent and mount it on top of @new_parent at
- * @new_mp. And mount @new_parent on the old parent and old
- * mountpoint of @top_mnt.
- *
- * Context: This function expects namespace_lock() and lock_mount_hash()
- *          to have been acquired in that order.
- */
-static void mnt_set_mountpoint_beneath(struct mount *new_parent,
-				       struct mount *top_mnt,
-				       struct mountpoint *new_mp)
-{
-	struct mount *old_top_parent = top_mnt->mnt_parent;
-	struct mountpoint *old_top_mp = top_mnt->mnt_mp;
-
-	mnt_set_mountpoint(old_top_parent, old_top_mp, new_parent);
-	mnt_change_mountpoint(new_parent, new_mp, top_mnt);
-}
-
-
 static void __attach_mnt(struct mount *mnt, struct mount *parent)
 {
 	hlist_add_head_rcu(&mnt->mnt_hash,
@@ -2687,10 +2660,9 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 
 	if (moving) {
 		umount_mnt(source_mnt);
+		mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
 		if (beneath)
-			mnt_set_mountpoint_beneath(source_mnt, top_mnt, smp);
-		else
-			mnt_set_mountpoint(top_mnt, dest_mp, source_mnt);
+			mnt_change_mountpoint(source_mnt, smp, top_mnt);
 		__attach_mnt(source_mnt, source_mnt->mnt_parent);
 		mnt_notify_add(source_mnt);
 		touch_mnt_namespace(source_mnt->mnt_ns);
@@ -2703,10 +2675,9 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 				move_from_ns(p, &head);
 			list_del_init(&head);
 		}
+		mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
 		if (beneath)
-			mnt_set_mountpoint_beneath(source_mnt, top_mnt, smp);
-		else
-			mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
+			mnt_change_mountpoint(source_mnt, smp, top_mnt);
 		commit_tree(source_mnt);
 	}
 
-- 
2.39.5

Re: [PATCH 15/26] get rid of mnt_set_mountpoint_beneath()

[Christian Brauner]

On Tue, Jun 10, 2025 at 09:21:37AM +0100, Al Viro wrote:
> mnt_set_mountpoint_beneath() consists of attaching new mount side-by-side
> with the one we want to mount beneath (by mnt_set_mountpoint()), followed
> by mnt_change_mountpoint() shifting the the top mount onto the new one
> (by mnt_change_mountpoint()).
> 
> Both callers of mnt_set_mountpoint_beneath (both in attach_recursive_mnt())
> have the same form - in 'beneath' case we call mnt_set_mountpoint_beneath(),
> otherwise - mnt_set_mountpoint().
> 
> The thing is, expressing that as unconditional mnt_set_mountpoint(),
> followed, in 'beneath' case, by mnt_change_mountpoint() is just as easy.
> And these mnt_change_mountpoint() callers are similar to the ones we
> do when it comes to attaching propagated copies, which will allow more
> cleanups in the next commits.
> 
> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
> ---

Reviewed-by: Christian Brauner <brauner@kernel.org>

[PATCH 16/26] make commit_tree() usable in same-namespace move case

[Al Viro]

Once attach_recursive_mnt() has created all copies of original subtree,
it needs to put them in place(s).

Steps needed for those are slightly different:
	1) in 'move' case, original copy doesn't need any rbtree
manipulations (everything's already in the same namespace where it will
be), but it needs to be detached from the current location
	2) in 'attach' case, original may be in anon namespace; if it is,
all those mounts need to removed from their current namespace before
insertion into the target one
	3) additional copies have a couple of extra twists - in case
of cross-userns propagation we need to lock everything other the root of
subtree and in case when we end up inserting under an existing mount,
that mount needs to be found (for original copy we have it explicitly
passed by the caller).

Quite a bit of that can be unified; as the first step, make commit_tree()
helper (inserting mounts into namespace, hashing the root of subtree
and marking the namespace as updated) usable in all cases; (2) and (3)
are already using it and for (1) we only need to make the insertion of
mounts into namespace conditional.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 30 ++++++++++++++----------------
 1 file changed, 14 insertions(+), 16 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index 22bbc30a2da1..50c46c084b13 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -1166,15 +1166,17 @@ static void commit_tree(struct mount *mnt)
 
 	BUG_ON(parent == mnt);
 
-	list_add_tail(&head, &mnt->mnt_list);
-	while (!list_empty(&head)) {
-		m = list_first_entry(&head, typeof(*m), mnt_list);
-		list_del(&m->mnt_list);
+	if (!mnt_ns_attached(mnt)) {
+		list_add_tail(&head, &mnt->mnt_list);
+		while (!list_empty(&head)) {
+			m = list_first_entry(&head, typeof(*m), mnt_list);
+			list_del(&m->mnt_list);
 
-		mnt_add_to_ns(n, m);
+			mnt_add_to_ns(n, m);
+		}
+		n->nr_mounts += n->pending_mounts;
+		n->pending_mounts = 0;
 	}
-	n->nr_mounts += n->pending_mounts;
-	n->pending_mounts = 0;
 
 	__attach_mnt(mnt, parent);
 	touch_mnt_namespace(n);
@@ -2660,12 +2662,7 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 
 	if (moving) {
 		umount_mnt(source_mnt);
-		mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
-		if (beneath)
-			mnt_change_mountpoint(source_mnt, smp, top_mnt);
-		__attach_mnt(source_mnt, source_mnt->mnt_parent);
 		mnt_notify_add(source_mnt);
-		touch_mnt_namespace(source_mnt->mnt_ns);
 	} else {
 		if (source_mnt->mnt_ns) {
 			LIST_HEAD(head);
@@ -2675,12 +2672,13 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 				move_from_ns(p, &head);
 			list_del_init(&head);
 		}
-		mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
-		if (beneath)
-			mnt_change_mountpoint(source_mnt, smp, top_mnt);
-		commit_tree(source_mnt);
 	}
 
+	mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
+	if (beneath)
+		mnt_change_mountpoint(source_mnt, smp, top_mnt);
+	commit_tree(source_mnt);
+
 	hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
 		struct mount *q;
 		hlist_del_init(&child->mnt_hash);
-- 
2.39.5

Re: [PATCH 16/26] make commit_tree() usable in same-namespace move case

[Christian Brauner]

On Tue, Jun 10, 2025 at 09:21:38AM +0100, Al Viro wrote:
> Once attach_recursive_mnt() has created all copies of original subtree,
> it needs to put them in place(s).
> 
> Steps needed for those are slightly different:
> 	1) in 'move' case, original copy doesn't need any rbtree
> manipulations (everything's already in the same namespace where it will
> be), but it needs to be detached from the current location
> 	2) in 'attach' case, original may be in anon namespace; if it is,
> all those mounts need to removed from their current namespace before
> insertion into the target one
> 	3) additional copies have a couple of extra twists - in case
> of cross-userns propagation we need to lock everything other the root of
> subtree and in case when we end up inserting under an existing mount,
> that mount needs to be found (for original copy we have it explicitly
> passed by the caller).
> 
> Quite a bit of that can be unified; as the first step, make commit_tree()
> helper (inserting mounts into namespace, hashing the root of subtree
> and marking the namespace as updated) usable in all cases; (2) and (3)
> are already using it and for (1) we only need to make the insertion of
> mounts into namespace conditional.
> 
> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
> ---

Reviewed-by: Christian Brauner <brauner@kernel.org>

[PATCH 17/26] attach_recursive_mnt(): unify the mnt_change_mountpoint() logics

[Al Viro]

The logics used for tucking under existing mount differs for original
and copies; copies do a mount hash lookup to see if mountpoint to be is
already overmounted, while the original is told explicitly.

But the same logics that is used for copies works for the original,
at which point the only place where we get very close to eliminating
the need of passing 'beneath' flag to attach_recursive_mnt().

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 4 +---
 1 file changed, 1 insertion(+), 3 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index 50c46c084b13..0e43301abb91 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -2675,9 +2675,7 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 	}
 
 	mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
-	if (beneath)
-		mnt_change_mountpoint(source_mnt, smp, top_mnt);
-	commit_tree(source_mnt);
+	hlist_add_head(&source_mnt->mnt_hash, &tree_list);
 
 	hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
 		struct mount *q;
-- 
2.39.5

Re: [PATCH 17/26] attach_recursive_mnt(): unify the mnt_change_mountpoint() logics

[Christian Brauner]

On Tue, Jun 10, 2025 at 09:21:39AM +0100, Al Viro wrote:
> The logics used for tucking under existing mount differs for original
> and copies; copies do a mount hash lookup to see if mountpoint to be is
> already overmounted, while the original is told explicitly.
> 
> But the same logics that is used for copies works for the original,
> at which point the only place where we get very close to eliminating
> the need of passing 'beneath' flag to attach_recursive_mnt().
> 
> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
> ---

Reviewed-by: Christian Brauner <brauner@kernel.org>

>  fs/namespace.c | 4 +---
>  1 file changed, 1 insertion(+), 3 deletions(-)
> 
> diff --git a/fs/namespace.c b/fs/namespace.c
> index 50c46c084b13..0e43301abb91 100644
> --- a/fs/namespace.c
> +++ b/fs/namespace.c
> @@ -2675,9 +2675,7 @@ static int attach_recursive_mnt(struct mount *source_mnt,
>  	}
>  
>  	mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
> -	if (beneath)
> -		mnt_change_mountpoint(source_mnt, smp, top_mnt);
> -	commit_tree(source_mnt);
> +	hlist_add_head(&source_mnt->mnt_hash, &tree_list);

Please add a comment here. Right now it's easy to understand even with
your mnt_change_mountpoint() changes. Afterwards the cases are folded
and imho that leaves readers wondering why that's correct.

>  
>  	hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
>  		struct mount *q;
> -- 
> 2.39.5
> 

Re: [PATCH 17/26] attach_recursive_mnt(): unify the mnt_change_mountpoint() logics

[Al Viro]

On Wed, Jun 11, 2025 at 01:05:52PM +0200, Christian Brauner wrote:
> > -	if (beneath)
> > -		mnt_change_mountpoint(source_mnt, smp, top_mnt);
> > -	commit_tree(source_mnt);
> > +	hlist_add_head(&source_mnt->mnt_hash, &tree_list);
> 
> Please add a comment here. Right now it's easy to understand even with
> your mnt_change_mountpoint() changes. Afterwards the cases are folded
> and imho that leaves readers wondering why that's correct.

Hmm...  Does the incremental below look sane for you?

diff --git a/fs/namespace.c b/fs/namespace.c
index d5a7d7da3932..15b7959b1771 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -2675,6 +2675,13 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 	}
 
 	mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
+	/*
+	 * Now the original copy is in the same state as the secondaries -
+	 * its root attached to mountpoint, but not hashed and all mounts
+	 * in it are either in our namespace or in no namespace at all.
+	 * Add the original to the list of copies and deal with the
+	 * rest of work for all of them uniformly.
+	 */
 	hlist_add_head(&source_mnt->mnt_hash, &tree_list);
 
 	hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {

Re: [PATCH 17/26] attach_recursive_mnt(): unify the mnt_change_mountpoint() logics

[Christian Brauner]

On Wed, Jun 11, 2025 at 07:12:23PM +0100, Al Viro wrote:
> On Wed, Jun 11, 2025 at 01:05:52PM +0200, Christian Brauner wrote:
> > > -	if (beneath)
> > > -		mnt_change_mountpoint(source_mnt, smp, top_mnt);
> > > -	commit_tree(source_mnt);
> > > +	hlist_add_head(&source_mnt->mnt_hash, &tree_list);
> > 
> > Please add a comment here. Right now it's easy to understand even with
> > your mnt_change_mountpoint() changes. Afterwards the cases are folded
> > and imho that leaves readers wondering why that's correct.
> 
> Hmm...  Does the incremental below look sane for you?

Yep, thanks!

> 
> diff --git a/fs/namespace.c b/fs/namespace.c
> index d5a7d7da3932..15b7959b1771 100644
> --- a/fs/namespace.c
> +++ b/fs/namespace.c
> @@ -2675,6 +2675,13 @@ static int attach_recursive_mnt(struct mount *source_mnt,
>  	}
>  
>  	mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
> +	/*
> +	 * Now the original copy is in the same state as the secondaries -
> +	 * its root attached to mountpoint, but not hashed and all mounts
> +	 * in it are either in our namespace or in no namespace at all.
> +	 * Add the original to the list of copies and deal with the
> +	 * rest of work for all of them uniformly.
> +	 */
>  	hlist_add_head(&source_mnt->mnt_hash, &tree_list);
>  
>  	hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {

[PATCH 18/26] attach_recursive_mnt(): pass destination mount in all cases

[Al Viro]

... and 'beneath' is no longer used there

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 17 ++++++-----------
 1 file changed, 6 insertions(+), 11 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index 0e43301abb91..571916df33fd 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -2547,7 +2547,7 @@ enum mnt_tree_flags_t {
 /**
  * attach_recursive_mnt - attach a source mount tree
  * @source_mnt: mount tree to be attached
- * @top_mnt:    mount that @source_mnt will be mounted on or mounted beneath
+ * @dest_mnt:   mount that @source_mnt will be mounted on
  * @dest_mp:    the mountpoint @source_mnt will be mounted at
  * @flags:      modify how @source_mnt is supposed to be attached
  *
@@ -2612,18 +2612,18 @@ enum mnt_tree_flags_t {
  *         Otherwise a negative error code is returned.
  */
 static int attach_recursive_mnt(struct mount *source_mnt,
-				struct mount *top_mnt,
+				struct mount *dest_mnt,
 				struct mountpoint *dest_mp,
 				enum mnt_tree_flags_t flags)
 {
 	struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
 	HLIST_HEAD(tree_list);
-	struct mnt_namespace *ns = top_mnt->mnt_ns;
+	struct mnt_namespace *ns = dest_mnt->mnt_ns;
 	struct mountpoint *smp;
-	struct mount *child, *dest_mnt, *p;
+	struct mount *child, *p;
 	struct hlist_node *n;
 	int err = 0;
-	bool moving = flags & MNT_TREE_MOVE, beneath = flags & MNT_TREE_BENEATH;
+	bool moving = flags & MNT_TREE_MOVE;
 
 	/*
 	 * Preallocate a mountpoint in case the new mounts need to be
@@ -2640,11 +2640,6 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 			goto out;
 	}
 
-	if (beneath)
-		dest_mnt = top_mnt->mnt_parent;
-	else
-		dest_mnt = top_mnt;
-
 	if (IS_MNT_SHARED(dest_mnt)) {
 		err = invent_group_ids(source_mnt, true);
 		if (err)
@@ -3644,7 +3639,7 @@ static int do_move_mount(struct path *old_path,
 	if (mount_is_ancestor(old, p))
 		goto out;
 
-	err = attach_recursive_mnt(old, real_mount(new_path->mnt), mp, flags);
+	err = attach_recursive_mnt(old, p, mp, flags);
 	if (err)
 		goto out;
 
-- 
2.39.5

Re: [PATCH 18/26] attach_recursive_mnt(): pass destination mount in all cases

[Christian Brauner]

On Tue, Jun 10, 2025 at 09:21:40AM +0100, Al Viro wrote:
> ... and 'beneath' is no longer used there
> 
> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
> ---

Reviewed-by: Christian Brauner <brauner@kernel.org>

[PATCH 19/26] attach_recursive_mnt(): get rid of flags entirely

[Al Viro]

move vs. attach is trivially detected as mnt_has_parent(source_mnt)...

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 18 ++++++------------
 1 file changed, 6 insertions(+), 12 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index 571916df33fd..5906ad173a28 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -2539,9 +2539,8 @@ int count_mounts(struct mnt_namespace *ns, struct mount *mnt)
 }
 
 enum mnt_tree_flags_t {
-	MNT_TREE_MOVE = BIT(0),
-	MNT_TREE_BENEATH = BIT(1),
-	MNT_TREE_PROPAGATION = BIT(2),
+	MNT_TREE_BENEATH = BIT(0),
+	MNT_TREE_PROPAGATION = BIT(1),
 };
 
 /**
@@ -2549,7 +2548,6 @@ enum mnt_tree_flags_t {
  * @source_mnt: mount tree to be attached
  * @dest_mnt:   mount that @source_mnt will be mounted on
  * @dest_mp:    the mountpoint @source_mnt will be mounted at
- * @flags:      modify how @source_mnt is supposed to be attached
  *
  *  NOTE: in the table below explains the semantics when a source mount
  *  of a given type is attached to a destination mount of a given type.
@@ -2613,8 +2611,7 @@ enum mnt_tree_flags_t {
  */
 static int attach_recursive_mnt(struct mount *source_mnt,
 				struct mount *dest_mnt,
-				struct mountpoint *dest_mp,
-				enum mnt_tree_flags_t flags)
+				struct mountpoint *dest_mp)
 {
 	struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
 	HLIST_HEAD(tree_list);
@@ -2623,7 +2620,7 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 	struct mount *child, *p;
 	struct hlist_node *n;
 	int err = 0;
-	bool moving = flags & MNT_TREE_MOVE;
+	bool moving = mnt_has_parent(source_mnt);
 
 	/*
 	 * Preallocate a mountpoint in case the new mounts need to be
@@ -2828,7 +2825,7 @@ static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
 	      d_is_dir(mnt->mnt.mnt_root))
 		return -ENOTDIR;
 
-	return attach_recursive_mnt(mnt, p, mp, 0);
+	return attach_recursive_mnt(mnt, p, mp);
 }
 
 /*
@@ -3570,8 +3567,6 @@ static int do_move_mount(struct path *old_path,
 	p = real_mount(new_path->mnt);
 	parent = old->mnt_parent;
 	attached = mnt_has_parent(old);
-	if (attached)
-		flags |= MNT_TREE_MOVE;
 	ns = old->mnt_ns;
 
 	err = -EINVAL;
@@ -3624,7 +3619,6 @@ static int do_move_mount(struct path *old_path,
 
 		err = -EINVAL;
 		p = p->mnt_parent;
-		flags |= MNT_TREE_BENEATH;
 	}
 
 	/*
@@ -3639,7 +3633,7 @@ static int do_move_mount(struct path *old_path,
 	if (mount_is_ancestor(old, p))
 		goto out;
 
-	err = attach_recursive_mnt(old, p, mp, flags);
+	err = attach_recursive_mnt(old, p, mp);
 	if (err)
 		goto out;
 
-- 
2.39.5

Re: [PATCH 19/26] attach_recursive_mnt(): get rid of flags entirely

[Christian Brauner]

On Tue, Jun 10, 2025 at 09:21:41AM +0100, Al Viro wrote:
> move vs. attach is trivially detected as mnt_has_parent(source_mnt)...
> 
> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
> ---

Reviewed-by: Christian Brauner <brauner@kernel.org>

[PATCH 20/26] do_move_mount(): get rid of 'attached' flag

[Al Viro]

'attached' serves as a proxy for "source is a subtree of our namespace
and not the entirety of anon namespace"; finish massaging it away.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 13 +++++--------
 1 file changed, 5 insertions(+), 8 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index 5906ad173a28..7dffe9f71896 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -3557,7 +3557,7 @@ static int do_move_mount(struct path *old_path,
 	struct mount *parent;
 	struct mountpoint *mp;
 	int err;
-	bool attached, beneath = flags & MNT_TREE_BENEATH;
+	bool beneath = flags & MNT_TREE_BENEATH;
 
 	mp = do_lock_mount(new_path, beneath);
 	if (IS_ERR(mp))
@@ -3566,7 +3566,6 @@ static int do_move_mount(struct path *old_path,
 	old = real_mount(old_path->mnt);
 	p = real_mount(new_path->mnt);
 	parent = old->mnt_parent;
-	attached = mnt_has_parent(old);
 	ns = old->mnt_ns;
 
 	err = -EINVAL;
@@ -3579,6 +3578,9 @@ static int do_move_mount(struct path *old_path,
 		/* ... and the target should be in our namespace */
 		if (!check_mnt(p))
 			goto out;
+		/* parent of the source should not be shared */
+		if (IS_MNT_SHARED(parent))
+			goto out;
 	} else {
 		/*
 		 * otherwise the source must be the root of some anon namespace.
@@ -3606,11 +3608,6 @@ static int do_move_mount(struct path *old_path,
 	if (d_is_dir(new_path->dentry) !=
 	    d_is_dir(old_path->dentry))
 		goto out;
-	/*
-	 * Don't move a mount residing in a shared parent.
-	 */
-	if (attached && IS_MNT_SHARED(parent))
-		goto out;
 
 	if (beneath) {
 		err = can_move_mount_beneath(old_path, new_path, mp);
@@ -3643,7 +3640,7 @@ static int do_move_mount(struct path *old_path,
 out:
 	unlock_mount(mp);
 	if (!err) {
-		if (attached) {
+		if (!is_anon_ns(ns)) {
 			mntput_no_expire(parent);
 		} else {
 			/* Make sure we notice when we leak mounts. */
-- 
2.39.5

Re: [PATCH 20/26] do_move_mount(): get rid of 'attached' flag

[Christian Brauner]

On Tue, Jun 10, 2025 at 09:21:42AM +0100, Al Viro wrote:
> 'attached' serves as a proxy for "source is a subtree of our namespace
> and not the entirety of anon namespace"; finish massaging it away.
> 
> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
> ---

Reviewed-by: Christian Brauner <brauner@kernel.org>

[PATCH 21/26] attach_recursive_mnt(): remove from expiry list on move

[Al Viro]

... rather than doing that in do_move_mount().  That's the main
obstacle to moving the protection of ->mnt_expire from namespace_sem
to mount_lock (spinlock-only), which would simplify several failure
exits.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 9 +++------
 1 file changed, 3 insertions(+), 6 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index 7dffe9f71896..d9ad214b3fec 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -2655,6 +2655,9 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 	if (moving) {
 		umount_mnt(source_mnt);
 		mnt_notify_add(source_mnt);
+		/* if the mount is moved, it should no longer be expired
+		 * automatically */
+		list_del_init(&source_mnt->mnt_expire);
 	} else {
 		if (source_mnt->mnt_ns) {
 			LIST_HEAD(head);
@@ -3631,12 +3634,6 @@ static int do_move_mount(struct path *old_path,
 		goto out;
 
 	err = attach_recursive_mnt(old, p, mp);
-	if (err)
-		goto out;
-
-	/* if the mount is moved, it should no longer be expire
-	 * automatically */
-	list_del_init(&old->mnt_expire);
 out:
 	unlock_mount(mp);
 	if (!err) {
-- 
2.39.5

Re: [PATCH 21/26] attach_recursive_mnt(): remove from expiry list on move

[Christian Brauner]

On Tue, Jun 10, 2025 at 09:21:43AM +0100, Al Viro wrote:
> ... rather than doing that in do_move_mount().  That's the main
> obstacle to moving the protection of ->mnt_expire from namespace_sem
> to mount_lock (spinlock-only), which would simplify several failure
> exits.
> 
> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
> ---

Reviewed-by: Christian Brauner <brauner@kernel.org>

[PATCH 22/26] take ->mnt_expire handling under mount_lock [read_seqlock_excl]

[Al Viro]

Doesn't take much massage, and we no longer need to make sure that
by the time of final mntput() the victim has been removed from the
list.  Makes life safer for ->d_automount() instances...

Rules:
	* all ->mnt_expire accesses are under mount_lock.
	* insertion into the list is done by mnt_set_expiry(), and
caller (->d_automount() instance) must hold a reference to mount
in question.  It shouldn't be done more than once for a mount.
	* if a mount on an expiry list is not yet mounted, it will
be ignored by anything that walks that list.
	* if the final mntput() finds its victim still on an expiry
list (in which case it must've never been mounted - umount_tree()
would've taken it out), it will remove the victim from the list.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 28 +++++++++++-----------------
 1 file changed, 11 insertions(+), 17 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index d9ad214b3fec..7df00ed26db5 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -1347,13 +1347,6 @@ static struct mount *clone_mnt(struct mount *old, struct dentry *root,
 		list_add(&mnt->mnt_slave, &old->mnt_slave);
 		mnt->mnt_master = old->mnt_master;
 	}
-
-	/* stick the duplicate mount on the same expiry list
-	 * as the original if that was on one */
-	if (flag & CL_EXPIRE) {
-		if (!list_empty(&old->mnt_expire))
-			list_add(&mnt->mnt_expire, &old->mnt_expire);
-	}
 	return mnt;
 
  out_free:
@@ -1446,6 +1439,8 @@ static void mntput_no_expire(struct mount *mnt)
 	rcu_read_unlock();
 
 	list_del(&mnt->mnt_instance);
+	if (unlikely(!list_empty(&mnt->mnt_expire)))
+		list_del(&mnt->mnt_expire);
 
 	if (unlikely(!list_empty(&mnt->mnt_mounts))) {
 		struct mount *p, *tmp;
@@ -2267,6 +2262,13 @@ struct mount *copy_tree(struct mount *src_root, struct dentry *dentry,
 			lock_mount_hash();
 			if (src_mnt->mnt.mnt_flags & MNT_LOCKED)
 				dst_mnt->mnt.mnt_flags |= MNT_LOCKED;
+			if (unlikely(flag & CL_EXPIRE)) {
+				/* stick the duplicate mount on the same expiry
+				 * list as the original if that was on one */
+				if (!list_empty(&src_mnt->mnt_expire))
+					list_add(&dst_mnt->mnt_expire,
+						 &src_mnt->mnt_expire);
+			}
 			list_add_tail(&dst_mnt->mnt_list, &res->mnt_list);
 			attach_mnt(dst_mnt, dst_parent, src_parent->mnt_mp);
 			unlock_mount_hash();
@@ -3848,12 +3850,6 @@ int finish_automount(struct vfsmount *m, const struct path *path)
 	namespace_unlock();
 	inode_unlock(dentry->d_inode);
 discard:
-	/* remove m from any expiration list it may be on */
-	if (!list_empty(&mnt->mnt_expire)) {
-		namespace_lock();
-		list_del_init(&mnt->mnt_expire);
-		namespace_unlock();
-	}
 	mntput(m);
 	return err;
 }
@@ -3865,11 +3861,9 @@ int finish_automount(struct vfsmount *m, const struct path *path)
  */
 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
 {
-	namespace_lock();
-
+	read_seqlock_excl(&mount_lock);
 	list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
-
-	namespace_unlock();
+	read_sequnlock_excl(&mount_lock);
 }
 EXPORT_SYMBOL(mnt_set_expiry);
 
-- 
2.39.5

Re: [PATCH 22/26] take ->mnt_expire handling under mount_lock [read_seqlock_excl]

[Christian Brauner]

On Tue, Jun 10, 2025 at 09:21:44AM +0100, Al Viro wrote:
> Doesn't take much massage, and we no longer need to make sure that
> by the time of final mntput() the victim has been removed from the
> list.  Makes life safer for ->d_automount() instances...
> 
> Rules:
> 	* all ->mnt_expire accesses are under mount_lock.
> 	* insertion into the list is done by mnt_set_expiry(), and
> caller (->d_automount() instance) must hold a reference to mount
> in question.  It shouldn't be done more than once for a mount.
> 	* if a mount on an expiry list is not yet mounted, it will
> be ignored by anything that walks that list.
> 	* if the final mntput() finds its victim still on an expiry
> list (in which case it must've never been mounted - umount_tree()
> would've taken it out), it will remove the victim from the list.
> 
> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
> ---

Reviewed-by: Christian Brauner <brauner@kernel.org>

[PATCH 23/26] pivot_root(): reorder tree surgeries, collapse unhash_mnt() and put_mountpoint()

[Al Viro]

attach new_mnt *before* detaching root_mnt; that way we don't need to keep hold
on the mountpoint and one more pair of unhash_mnt()/put_mountpoint() gets
folded together into umount_mnt().

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 11 +++++------
 1 file changed, 5 insertions(+), 6 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index 7df00ed26db5..60dcfe4aa976 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -4642,7 +4642,7 @@ SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
 {
 	struct path new, old, root;
 	struct mount *new_mnt, *root_mnt, *old_mnt, *root_parent, *ex_parent;
-	struct mountpoint *old_mp, *root_mp;
+	struct mountpoint *old_mp;
 	int error;
 
 	if (!may_mount())
@@ -4705,20 +4705,19 @@ SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
 		goto out4;
 	lock_mount_hash();
 	umount_mnt(new_mnt);
-	root_mp = unhash_mnt(root_mnt);  /* we'll need its mountpoint */
 	if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
 		new_mnt->mnt.mnt_flags |= MNT_LOCKED;
 		root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
 	}
-	/* mount old root on put_old */
-	attach_mnt(root_mnt, old_mnt, old_mp);
 	/* mount new_root on / */
-	attach_mnt(new_mnt, root_parent, root_mp);
+	attach_mnt(new_mnt, root_parent, root_mnt->mnt_mp);
+	umount_mnt(root_mnt);
 	mnt_add_count(root_parent, -1);
+	/* mount old root on put_old */
+	attach_mnt(root_mnt, old_mnt, old_mp);
 	touch_mnt_namespace(current->nsproxy->mnt_ns);
 	/* A moved mount should not expire automatically */
 	list_del_init(&new_mnt->mnt_expire);
-	put_mountpoint(root_mp);
 	unlock_mount_hash();
 	mnt_notify_add(root_mnt);
 	mnt_notify_add(new_mnt);
-- 
2.39.5

Re: [PATCH 23/26] pivot_root(): reorder tree surgeries, collapse unhash_mnt() and put_mountpoint()

[Christian Brauner]

On Tue, Jun 10, 2025 at 09:21:45AM +0100, Al Viro wrote:
> attach new_mnt *before* detaching root_mnt; that way we don't need to keep hold
> on the mountpoint and one more pair of unhash_mnt()/put_mountpoint() gets
> folded together into umount_mnt().
> 
> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
> ---

Reviewed-by: Christian Brauner <brauner@kernel.org>

[PATCH 24/26] combine __put_mountpoint() with unhash_mnt()

[Al Viro]

A call of unhash_mnt() is immediately followed by passing its return
value to __put_mountpoint(); the shrink list given to __put_mountpoint()
will be ex_mountpoints when called from umount_mnt() and list when called
from mntput_no_expire().

Replace with __umount_mnt(mount, shrink_list), moving the call of
__put_mountpoint() into it (and returning nothing), adjust the
callers.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index 60dcfe4aa976..777e4c3b2c12 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -1037,9 +1037,9 @@ static void __touch_mnt_namespace(struct mnt_namespace *ns)
 }
 
 /*
- * vfsmount lock must be held for write
+ * locks: mount_lock[write_seqlock]
  */
-static struct mountpoint *unhash_mnt(struct mount *mnt)
+static void __umount_mnt(struct mount *mnt, struct list_head *shrink_list)
 {
 	struct mountpoint *mp;
 	mnt->mnt_parent = mnt;
@@ -1049,15 +1049,15 @@ static struct mountpoint *unhash_mnt(struct mount *mnt)
 	hlist_del_init(&mnt->mnt_mp_list);
 	mp = mnt->mnt_mp;
 	mnt->mnt_mp = NULL;
-	return mp;
+	__put_mountpoint(mp, shrink_list);
 }
 
 /*
- * vfsmount lock must be held for write
+ * locks: mount_lock[write_seqlock], namespace_sem[excl] (for ex_mountpoints)
  */
 static void umount_mnt(struct mount *mnt)
 {
-	put_mountpoint(unhash_mnt(mnt));
+	__umount_mnt(mnt, &ex_mountpoints);
 }
 
 /*
@@ -1445,7 +1445,7 @@ static void mntput_no_expire(struct mount *mnt)
 	if (unlikely(!list_empty(&mnt->mnt_mounts))) {
 		struct mount *p, *tmp;
 		list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts,  mnt_child) {
-			__put_mountpoint(unhash_mnt(p), &list);
+			__umount_mnt(p, &list);
 			hlist_add_head(&p->mnt_umount, &mnt->mnt_stuck_children);
 		}
 	}
-- 
2.39.5

Re: [PATCH 24/26] combine __put_mountpoint() with unhash_mnt()

[Christian Brauner]

On Tue, Jun 10, 2025 at 09:21:46AM +0100, Al Viro wrote:
> A call of unhash_mnt() is immediately followed by passing its return
> value to __put_mountpoint(); the shrink list given to __put_mountpoint()
> will be ex_mountpoints when called from umount_mnt() and list when called
> from mntput_no_expire().
> 
> Replace with __umount_mnt(mount, shrink_list), moving the call of
> __put_mountpoint() into it (and returning nothing), adjust the
> callers.
> 
> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
> ---

Reviewed-by: Christian Brauner <brauner@kernel.org>

[PATCH 25/26] get rid of mountpoint->m_count

[Al Viro]

struct mountpoint has an odd kinda-sorta refcount in it.  It's always
either equal to or one above the number of mounts attached to that
mountpoint.

"One above" happens when a function takes a temporary reference to
mountpoint.  Things get simpler if we express that as inserting
a local object into ->m_list and removing it to drop the reference.

New calling conventions:

1) lock_mount(), do_lock_mount(), get_mountpoint() and lookup_mountpoint()
take an extra struct pinned_mountpoint * argument and returns 0/-E...
(or true/false in case of lookup_mountpoint()) instead of returning
struct mountpoint pointers.  In case of success, the struct mountpoint *
we used to get can be found as pinned_mountpoint.mp

2) unlock_mount() (always paired with lock_mount()/do_lock_mount()) takes
an address of struct pinned_mountpoint - the same that had been passed to
lock_mount()/do_lock_mount().

3) put_mountpoint() for a temporary reference (paired with get_mountpoint()
or lookup_mountpoint()) is replaced with unpin_mountpoint(), which takes
the address of pinned_mountpoint we passed to matching {get,lookup}_mountpoint().

4) all instances of pinned_mountpoint are local variables; they always live on
stack.  {} is used for initializer, after successful {get,lookup}_mountpoint()
we must make sure to call unpin_mountpoint() before leaving the scope and
after successful {do_,}lock_mount() we must make sure to call unlock_mount()
before leaving the scope.

5) all manipulations of ->m_count are gone, along with ->m_count itself.
struct mountpoint lives while its ->m_list is non-empty.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/mount.h     |   1 -
 fs/namespace.c | 186 ++++++++++++++++++++++++-------------------------
 2 files changed, 92 insertions(+), 95 deletions(-)

diff --git a/fs/mount.h b/fs/mount.h
index 9b3de2eef68a..684480087da2 100644
--- a/fs/mount.h
+++ b/fs/mount.h
@@ -44,7 +44,6 @@ struct mountpoint {
 	struct hlist_node m_hash;
 	struct dentry *m_dentry;
 	struct hlist_head m_list;
-	int m_count;
 };
 
 struct mount {
diff --git a/fs/namespace.c b/fs/namespace.c
index 777e4c3b2c12..1f1cf1d6a464 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -910,42 +910,48 @@ bool __is_local_mountpoint(const struct dentry *dentry)
 	return is_covered;
 }
 
-static struct mountpoint *lookup_mountpoint(struct dentry *dentry)
+struct pinned_mountpoint {
+	struct hlist_node node;
+	struct mountpoint *mp;
+};
+
+static bool lookup_mountpoint(struct dentry *dentry, struct pinned_mountpoint *m)
 {
 	struct hlist_head *chain = mp_hash(dentry);
 	struct mountpoint *mp;
 
 	hlist_for_each_entry(mp, chain, m_hash) {
 		if (mp->m_dentry == dentry) {
-			mp->m_count++;
-			return mp;
+			hlist_add_head(&m->node, &mp->m_list);
+			m->mp = mp;
+			return true;
 		}
 	}
-	return NULL;
+	return false;
 }
 
-static struct mountpoint *get_mountpoint(struct dentry *dentry)
+static int get_mountpoint(struct dentry *dentry, struct pinned_mountpoint *m)
 {
-	struct mountpoint *mp, *new = NULL;
+	struct mountpoint *mp __free(kfree) = NULL;
+	bool found;
 	int ret;
 
 	if (d_mountpoint(dentry)) {
 		/* might be worth a WARN_ON() */
 		if (d_unlinked(dentry))
-			return ERR_PTR(-ENOENT);
+			return -ENOENT;
 mountpoint:
 		read_seqlock_excl(&mount_lock);
-		mp = lookup_mountpoint(dentry);
+		found = lookup_mountpoint(dentry, m);
 		read_sequnlock_excl(&mount_lock);
-		if (mp)
-			goto done;
+		if (found)
+			return 0;
 	}
 
-	if (!new)
-		new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
-	if (!new)
-		return ERR_PTR(-ENOMEM);
-
+	if (!mp)
+		mp = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
+	if (!mp)
+		return -ENOMEM;
 
 	/* Exactly one processes may set d_mounted */
 	ret = d_set_mounted(dentry);
@@ -955,34 +961,28 @@ static struct mountpoint *get_mountpoint(struct dentry *dentry)
 		goto mountpoint;
 
 	/* The dentry is not available as a mountpoint? */
-	mp = ERR_PTR(ret);
 	if (ret)
-		goto done;
+		return ret;
 
 	/* Add the new mountpoint to the hash table */
 	read_seqlock_excl(&mount_lock);
-	new->m_dentry = dget(dentry);
-	new->m_count = 1;
-	hlist_add_head(&new->m_hash, mp_hash(dentry));
-	INIT_HLIST_HEAD(&new->m_list);
+	mp->m_dentry = dget(dentry);
+	hlist_add_head(&mp->m_hash, mp_hash(dentry));
+	INIT_HLIST_HEAD(&mp->m_list);
+	hlist_add_head(&m->node, &mp->m_list);
+	m->mp = no_free_ptr(mp);
 	read_sequnlock_excl(&mount_lock);
-
-	mp = new;
-	new = NULL;
-done:
-	kfree(new);
-	return mp;
+	return 0;
 }
 
 /*
  * vfsmount lock must be held.  Additionally, the caller is responsible
  * for serializing calls for given disposal list.
  */
-static void __put_mountpoint(struct mountpoint *mp, struct list_head *list)
+static void maybe_free_mountpoint(struct mountpoint *mp, struct list_head *list)
 {
-	if (!--mp->m_count) {
+	if (hlist_empty(&mp->m_list)) {
 		struct dentry *dentry = mp->m_dentry;
-		BUG_ON(!hlist_empty(&mp->m_list));
 		spin_lock(&dentry->d_lock);
 		dentry->d_flags &= ~DCACHE_MOUNTED;
 		spin_unlock(&dentry->d_lock);
@@ -992,10 +992,15 @@ static void __put_mountpoint(struct mountpoint *mp, struct list_head *list)
 	}
 }
 
-/* called with namespace_lock and vfsmount lock */
-static void put_mountpoint(struct mountpoint *mp)
+/*
+ * locks: mount_lock [read_seqlock_excl], namespace_sem [excl]
+ */
+static void unpin_mountpoint(struct pinned_mountpoint *m)
 {
-	__put_mountpoint(mp, &ex_mountpoints);
+	if (m->mp) {
+		hlist_del(&m->node);
+		maybe_free_mountpoint(m->mp, &ex_mountpoints);
+	}
 }
 
 static inline int check_mnt(struct mount *mnt)
@@ -1049,7 +1054,7 @@ static void __umount_mnt(struct mount *mnt, struct list_head *shrink_list)
 	hlist_del_init(&mnt->mnt_mp_list);
 	mp = mnt->mnt_mp;
 	mnt->mnt_mp = NULL;
-	__put_mountpoint(mp, shrink_list);
+	maybe_free_mountpoint(mp, shrink_list);
 }
 
 /*
@@ -1067,7 +1072,6 @@ void mnt_set_mountpoint(struct mount *mnt,
 			struct mountpoint *mp,
 			struct mount *child_mnt)
 {
-	mp->m_count++;
 	mnt_add_count(mnt, 1);	/* essentially, that's mntget */
 	child_mnt->mnt_mountpoint = mp->m_dentry;
 	child_mnt->mnt_parent = mnt;
@@ -1116,7 +1120,7 @@ void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct m
 
 	attach_mnt(mnt, parent, mp);
 
-	put_mountpoint(old_mp);
+	maybe_free_mountpoint(old_mp, &ex_mountpoints);
 	mnt_add_count(old_parent, -1);
 }
 
@@ -2024,25 +2028,24 @@ static int do_umount(struct mount *mnt, int flags)
  */
 void __detach_mounts(struct dentry *dentry)
 {
-	struct mountpoint *mp;
+	struct pinned_mountpoint mp = {};
 	struct mount *mnt;
 
 	namespace_lock();
 	lock_mount_hash();
-	mp = lookup_mountpoint(dentry);
-	if (!mp)
+	if (!lookup_mountpoint(dentry, &mp))
 		goto out_unlock;
 
 	event++;
-	while (!hlist_empty(&mp->m_list)) {
-		mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list);
+	while (mp.node.next) {
+		mnt = hlist_entry(mp.node.next, struct mount, mnt_mp_list);
 		if (mnt->mnt.mnt_flags & MNT_UMOUNT) {
 			umount_mnt(mnt);
 			hlist_add_head(&mnt->mnt_umount, &unmounted);
 		}
 		else umount_tree(mnt, UMOUNT_CONNECTED);
 	}
-	put_mountpoint(mp);
+	unpin_mountpoint(&mp);
 out_unlock:
 	unlock_mount_hash();
 	namespace_unlock();
@@ -2618,7 +2621,7 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 	struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
 	HLIST_HEAD(tree_list);
 	struct mnt_namespace *ns = dest_mnt->mnt_ns;
-	struct mountpoint *smp;
+	struct pinned_mountpoint root = {};
 	struct mount *child, *p;
 	struct hlist_node *n;
 	int err = 0;
@@ -2628,9 +2631,9 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 	 * Preallocate a mountpoint in case the new mounts need to be
 	 * mounted beneath mounts on the same mountpoint.
 	 */
-	smp = get_mountpoint(source_mnt->mnt.mnt_root);
-	if (IS_ERR(smp))
-		return PTR_ERR(smp);
+	err = get_mountpoint(source_mnt->mnt.mnt_root, &root);
+	if (err)
+		return err;
 
 	/* Is there space to add these mounts to the mount namespace? */
 	if (!moving) {
@@ -2680,13 +2683,13 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 		q = __lookup_mnt(&child->mnt_parent->mnt,
 				 child->mnt_mountpoint);
 		if (q)
-			mnt_change_mountpoint(child, smp, q);
+			mnt_change_mountpoint(child, root.mp, q);
 		/* Notice when we are propagating across user namespaces */
 		if (child->mnt_parent->mnt_ns->user_ns != user_ns)
 			lock_mnt_tree(child);
 		commit_tree(child);
 	}
-	put_mountpoint(smp);
+	unpin_mountpoint(&root);
 	unlock_mount_hash();
 
 	return 0;
@@ -2703,7 +2706,7 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 	ns->pending_mounts = 0;
 
 	read_seqlock_excl(&mount_lock);
-	put_mountpoint(smp);
+	unpin_mountpoint(&root);
 	read_sequnlock_excl(&mount_lock);
 
 	return err;
@@ -2743,12 +2746,12 @@ static int attach_recursive_mnt(struct mount *source_mnt,
  * Return: Either the target mountpoint on the top mount or the top
  *         mount's mountpoint.
  */
-static struct mountpoint *do_lock_mount(struct path *path, bool beneath)
+static int do_lock_mount(struct path *path, struct pinned_mountpoint *pinned, bool beneath)
 {
 	struct vfsmount *mnt = path->mnt;
 	struct dentry *dentry;
-	struct mountpoint *mp = ERR_PTR(-ENOENT);
 	struct path under = {};
+	int err = -ENOENT;
 
 	for (;;) {
 		struct mount *m = real_mount(mnt);
@@ -2786,8 +2789,8 @@ static struct mountpoint *do_lock_mount(struct path *path, bool beneath)
 			path->dentry = dget(mnt->mnt_root);
 			continue;	// got overmounted
 		}
-		mp = get_mountpoint(dentry);
-		if (IS_ERR(mp))
+		err = get_mountpoint(dentry, pinned);
+		if (err)
 			break;
 		if (beneath) {
 			/*
@@ -2798,25 +2801,25 @@ static struct mountpoint *do_lock_mount(struct path *path, bool beneath)
 			 */
 			path_put(&under);
 		}
-		return mp;
+		return 0;
 	}
 	namespace_unlock();
 	inode_unlock(dentry->d_inode);
 	if (beneath)
 		path_put(&under);
-	return mp;
+	return err;
 }
 
-static inline struct mountpoint *lock_mount(struct path *path)
+static inline int lock_mount(struct path *path, struct pinned_mountpoint *m)
 {
-	return do_lock_mount(path, false);
+	return do_lock_mount(path, m, false);
 }
 
-static void unlock_mount(struct mountpoint *where)
+static void unlock_mount(struct pinned_mountpoint *m)
 {
-	inode_unlock(where->m_dentry->d_inode);
+	inode_unlock(m->mp->m_dentry->d_inode);
 	read_seqlock_excl(&mount_lock);
-	put_mountpoint(where);
+	unpin_mountpoint(m);
 	read_sequnlock_excl(&mount_lock);
 	namespace_unlock();
 }
@@ -2981,7 +2984,7 @@ static int do_loopback(struct path *path, const char *old_name,
 {
 	struct path old_path;
 	struct mount *mnt = NULL, *parent;
-	struct mountpoint *mp;
+	struct pinned_mountpoint mp = {};
 	int err;
 	if (!old_name || !*old_name)
 		return -EINVAL;
@@ -2993,11 +2996,9 @@ static int do_loopback(struct path *path, const char *old_name,
 	if (mnt_ns_loop(old_path.dentry))
 		goto out;
 
-	mp = lock_mount(path);
-	if (IS_ERR(mp)) {
-		err = PTR_ERR(mp);
+	err = lock_mount(path, &mp);
+	if (err)
 		goto out;
-	}
 
 	parent = real_mount(path->mnt);
 	if (!check_mnt(parent))
@@ -3009,14 +3010,14 @@ static int do_loopback(struct path *path, const char *old_name,
 		goto out2;
 	}
 
-	err = graft_tree(mnt, parent, mp);
+	err = graft_tree(mnt, parent, mp.mp);
 	if (err) {
 		lock_mount_hash();
 		umount_tree(mnt, UMOUNT_SYNC);
 		unlock_mount_hash();
 	}
 out2:
-	unlock_mount(mp);
+	unlock_mount(&mp);
 out:
 	path_put(&old_path);
 	return err;
@@ -3560,13 +3561,13 @@ static int do_move_mount(struct path *old_path,
 	struct mount *p;
 	struct mount *old;
 	struct mount *parent;
-	struct mountpoint *mp;
+	struct pinned_mountpoint mp;
 	int err;
 	bool beneath = flags & MNT_TREE_BENEATH;
 
-	mp = do_lock_mount(new_path, beneath);
-	if (IS_ERR(mp))
-		return PTR_ERR(mp);
+	err = do_lock_mount(new_path, &mp, beneath);
+	if (err)
+		return err;
 
 	old = real_mount(old_path->mnt);
 	p = real_mount(new_path->mnt);
@@ -3615,7 +3616,7 @@ static int do_move_mount(struct path *old_path,
 		goto out;
 
 	if (beneath) {
-		err = can_move_mount_beneath(old_path, new_path, mp);
+		err = can_move_mount_beneath(old_path, new_path, mp.mp);
 		if (err)
 			goto out;
 
@@ -3635,9 +3636,9 @@ static int do_move_mount(struct path *old_path,
 	if (mount_is_ancestor(old, p))
 		goto out;
 
-	err = attach_recursive_mnt(old, p, mp);
+	err = attach_recursive_mnt(old, p, mp.mp);
 out:
-	unlock_mount(mp);
+	unlock_mount(&mp);
 	if (!err) {
 		if (!is_anon_ns(ns)) {
 			mntput_no_expire(parent);
@@ -3707,7 +3708,7 @@ static int do_new_mount_fc(struct fs_context *fc, struct path *mountpoint,
 			   unsigned int mnt_flags)
 {
 	struct vfsmount *mnt;
-	struct mountpoint *mp;
+	struct pinned_mountpoint mp = {};
 	struct super_block *sb = fc->root->d_sb;
 	int error;
 
@@ -3728,13 +3729,12 @@ static int do_new_mount_fc(struct fs_context *fc, struct path *mountpoint,
 
 	mnt_warn_timestamp_expiry(mountpoint, mnt);
 
-	mp = lock_mount(mountpoint);
-	if (IS_ERR(mp)) {
-		mntput(mnt);
-		return PTR_ERR(mp);
+	error = lock_mount(mountpoint, &mp);
+	if (!error) {
+		error = do_add_mount(real_mount(mnt), mp.mp,
+				     mountpoint, mnt_flags);
+		unlock_mount(&mp);
 	}
-	error = do_add_mount(real_mount(mnt), mp, mountpoint, mnt_flags);
-	unlock_mount(mp);
 	if (error < 0)
 		mntput(mnt);
 	return error;
@@ -3802,7 +3802,7 @@ static int do_new_mount(struct path *path, const char *fstype, int sb_flags,
 int finish_automount(struct vfsmount *m, const struct path *path)
 {
 	struct dentry *dentry = path->dentry;
-	struct mountpoint *mp;
+	struct pinned_mountpoint mp = {};
 	struct mount *mnt;
 	int err;
 
@@ -3834,14 +3834,13 @@ int finish_automount(struct vfsmount *m, const struct path *path)
 		err = 0;
 		goto discard_locked;
 	}
-	mp = get_mountpoint(dentry);
-	if (IS_ERR(mp)) {
-		err = PTR_ERR(mp);
+	err = get_mountpoint(dentry, &mp);
+	if (err)
 		goto discard_locked;
-	}
 
-	err = do_add_mount(mnt, mp, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
-	unlock_mount(mp);
+	err = do_add_mount(mnt, mp.mp, path,
+			   path->mnt->mnt_flags | MNT_SHRINKABLE);
+	unlock_mount(&mp);
 	if (unlikely(err))
 		goto discard;
 	return 0;
@@ -4642,7 +4641,7 @@ SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
 {
 	struct path new, old, root;
 	struct mount *new_mnt, *root_mnt, *old_mnt, *root_parent, *ex_parent;
-	struct mountpoint *old_mp;
+	struct pinned_mountpoint old_mp = {};
 	int error;
 
 	if (!may_mount())
@@ -4663,9 +4662,8 @@ SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
 		goto out2;
 
 	get_fs_root(current->fs, &root);
-	old_mp = lock_mount(&old);
-	error = PTR_ERR(old_mp);
-	if (IS_ERR(old_mp))
+	error = lock_mount(&old, &old_mp);
+	if (error)
 		goto out3;
 
 	error = -EINVAL;
@@ -4714,7 +4712,7 @@ SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
 	umount_mnt(root_mnt);
 	mnt_add_count(root_parent, -1);
 	/* mount old root on put_old */
-	attach_mnt(root_mnt, old_mnt, old_mp);
+	attach_mnt(root_mnt, old_mnt, old_mp.mp);
 	touch_mnt_namespace(current->nsproxy->mnt_ns);
 	/* A moved mount should not expire automatically */
 	list_del_init(&new_mnt->mnt_expire);
@@ -4724,7 +4722,7 @@ SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
 	chroot_fs_refs(&root, &new);
 	error = 0;
 out4:
-	unlock_mount(old_mp);
+	unlock_mount(&old_mp);
 	if (!error)
 		mntput_no_expire(ex_parent);
 out3:
-- 
2.39.5

Re: [PATCH 25/26] get rid of mountpoint->m_count

[Christian Brauner]

On Tue, Jun 10, 2025 at 09:21:47AM +0100, Al Viro wrote:
> struct mountpoint has an odd kinda-sorta refcount in it.  It's always
> either equal to or one above the number of mounts attached to that
> mountpoint.
> 
> "One above" happens when a function takes a temporary reference to
> mountpoint.  Things get simpler if we express that as inserting
> a local object into ->m_list and removing it to drop the reference.
> 
> New calling conventions:
> 
> 1) lock_mount(), do_lock_mount(), get_mountpoint() and lookup_mountpoint()
> take an extra struct pinned_mountpoint * argument and returns 0/-E...
> (or true/false in case of lookup_mountpoint()) instead of returning
> struct mountpoint pointers.  In case of success, the struct mountpoint *
> we used to get can be found as pinned_mountpoint.mp
> 
> 2) unlock_mount() (always paired with lock_mount()/do_lock_mount()) takes
> an address of struct pinned_mountpoint - the same that had been passed to
> lock_mount()/do_lock_mount().
> 
> 3) put_mountpoint() for a temporary reference (paired with get_mountpoint()
> or lookup_mountpoint()) is replaced with unpin_mountpoint(), which takes
> the address of pinned_mountpoint we passed to matching {get,lookup}_mountpoint().
> 
> 4) all instances of pinned_mountpoint are local variables; they always live on
> stack.  {} is used for initializer, after successful {get,lookup}_mountpoint()
> we must make sure to call unpin_mountpoint() before leaving the scope and

This feels well-suited for a DEFINE_FREE based annotation so that
unpin_mountpoint() is called when the scope ends.

> after successful {do_,}lock_mount() we must make sure to call unlock_mount()
> before leaving the scope.
> 
> 5) all manipulations of ->m_count are gone, along with ->m_count itself.
> struct mountpoint lives while its ->m_list is non-empty.
> 
> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
> ---

Debatable whether that's really an improvement but sure,
Reviewed-by: Christian Brauner <brauner@kernel.org>

>  fs/mount.h     |   1 -
>  fs/namespace.c | 186 ++++++++++++++++++++++++-------------------------
>  2 files changed, 92 insertions(+), 95 deletions(-)
> 
> diff --git a/fs/mount.h b/fs/mount.h
> index 9b3de2eef68a..684480087da2 100644
> --- a/fs/mount.h
> +++ b/fs/mount.h
> @@ -44,7 +44,6 @@ struct mountpoint {
>  	struct hlist_node m_hash;
>  	struct dentry *m_dentry;
>  	struct hlist_head m_list;
> -	int m_count;
>  };
>  
>  struct mount {
> diff --git a/fs/namespace.c b/fs/namespace.c
> index 777e4c3b2c12..1f1cf1d6a464 100644
> --- a/fs/namespace.c
> +++ b/fs/namespace.c
> @@ -910,42 +910,48 @@ bool __is_local_mountpoint(const struct dentry *dentry)
>  	return is_covered;
>  }
>  
> -static struct mountpoint *lookup_mountpoint(struct dentry *dentry)
> +struct pinned_mountpoint {
> +	struct hlist_node node;
> +	struct mountpoint *mp;
> +};
> +
> +static bool lookup_mountpoint(struct dentry *dentry, struct pinned_mountpoint *m)
>  {
>  	struct hlist_head *chain = mp_hash(dentry);
>  	struct mountpoint *mp;
>  
>  	hlist_for_each_entry(mp, chain, m_hash) {
>  		if (mp->m_dentry == dentry) {
> -			mp->m_count++;
> -			return mp;
> +			hlist_add_head(&m->node, &mp->m_list);
> +			m->mp = mp;
> +			return true;
>  		}
>  	}
> -	return NULL;
> +	return false;
>  }
>  
> -static struct mountpoint *get_mountpoint(struct dentry *dentry)
> +static int get_mountpoint(struct dentry *dentry, struct pinned_mountpoint *m)
>  {
> -	struct mountpoint *mp, *new = NULL;
> +	struct mountpoint *mp __free(kfree) = NULL;
> +	bool found;
>  	int ret;
>  
>  	if (d_mountpoint(dentry)) {
>  		/* might be worth a WARN_ON() */
>  		if (d_unlinked(dentry))
> -			return ERR_PTR(-ENOENT);
> +			return -ENOENT;
>  mountpoint:
>  		read_seqlock_excl(&mount_lock);
> -		mp = lookup_mountpoint(dentry);
> +		found = lookup_mountpoint(dentry, m);
>  		read_sequnlock_excl(&mount_lock);
> -		if (mp)
> -			goto done;
> +		if (found)
> +			return 0;
>  	}
>  
> -	if (!new)
> -		new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
> -	if (!new)
> -		return ERR_PTR(-ENOMEM);
> -
> +	if (!mp)
> +		mp = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
> +	if (!mp)
> +		return -ENOMEM;
>  
>  	/* Exactly one processes may set d_mounted */
>  	ret = d_set_mounted(dentry);
> @@ -955,34 +961,28 @@ static struct mountpoint *get_mountpoint(struct dentry *dentry)
>  		goto mountpoint;
>  
>  	/* The dentry is not available as a mountpoint? */
> -	mp = ERR_PTR(ret);
>  	if (ret)
> -		goto done;
> +		return ret;
>  
>  	/* Add the new mountpoint to the hash table */
>  	read_seqlock_excl(&mount_lock);
> -	new->m_dentry = dget(dentry);
> -	new->m_count = 1;
> -	hlist_add_head(&new->m_hash, mp_hash(dentry));
> -	INIT_HLIST_HEAD(&new->m_list);
> +	mp->m_dentry = dget(dentry);
> +	hlist_add_head(&mp->m_hash, mp_hash(dentry));
> +	INIT_HLIST_HEAD(&mp->m_list);
> +	hlist_add_head(&m->node, &mp->m_list);
> +	m->mp = no_free_ptr(mp);
>  	read_sequnlock_excl(&mount_lock);
> -
> -	mp = new;
> -	new = NULL;
> -done:
> -	kfree(new);
> -	return mp;
> +	return 0;
>  }
>  
>  /*
>   * vfsmount lock must be held.  Additionally, the caller is responsible
>   * for serializing calls for given disposal list.
>   */
> -static void __put_mountpoint(struct mountpoint *mp, struct list_head *list)
> +static void maybe_free_mountpoint(struct mountpoint *mp, struct list_head *list)
>  {
> -	if (!--mp->m_count) {
> +	if (hlist_empty(&mp->m_list)) {
>  		struct dentry *dentry = mp->m_dentry;
> -		BUG_ON(!hlist_empty(&mp->m_list));
>  		spin_lock(&dentry->d_lock);
>  		dentry->d_flags &= ~DCACHE_MOUNTED;
>  		spin_unlock(&dentry->d_lock);
> @@ -992,10 +992,15 @@ static void __put_mountpoint(struct mountpoint *mp, struct list_head *list)
>  	}
>  }
>  
> -/* called with namespace_lock and vfsmount lock */
> -static void put_mountpoint(struct mountpoint *mp)
> +/*
> + * locks: mount_lock [read_seqlock_excl], namespace_sem [excl]
> + */
> +static void unpin_mountpoint(struct pinned_mountpoint *m)
>  {
> -	__put_mountpoint(mp, &ex_mountpoints);
> +	if (m->mp) {
> +		hlist_del(&m->node);
> +		maybe_free_mountpoint(m->mp, &ex_mountpoints);
> +	}
>  }
>  
>  static inline int check_mnt(struct mount *mnt)
> @@ -1049,7 +1054,7 @@ static void __umount_mnt(struct mount *mnt, struct list_head *shrink_list)
>  	hlist_del_init(&mnt->mnt_mp_list);
>  	mp = mnt->mnt_mp;
>  	mnt->mnt_mp = NULL;
> -	__put_mountpoint(mp, shrink_list);
> +	maybe_free_mountpoint(mp, shrink_list);
>  }
>  
>  /*
> @@ -1067,7 +1072,6 @@ void mnt_set_mountpoint(struct mount *mnt,
>  			struct mountpoint *mp,
>  			struct mount *child_mnt)
>  {
> -	mp->m_count++;
>  	mnt_add_count(mnt, 1);	/* essentially, that's mntget */
>  	child_mnt->mnt_mountpoint = mp->m_dentry;
>  	child_mnt->mnt_parent = mnt;
> @@ -1116,7 +1120,7 @@ void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct m
>  
>  	attach_mnt(mnt, parent, mp);
>  
> -	put_mountpoint(old_mp);
> +	maybe_free_mountpoint(old_mp, &ex_mountpoints);
>  	mnt_add_count(old_parent, -1);
>  }
>  
> @@ -2024,25 +2028,24 @@ static int do_umount(struct mount *mnt, int flags)
>   */
>  void __detach_mounts(struct dentry *dentry)
>  {
> -	struct mountpoint *mp;
> +	struct pinned_mountpoint mp = {};
>  	struct mount *mnt;
>  
>  	namespace_lock();
>  	lock_mount_hash();
> -	mp = lookup_mountpoint(dentry);
> -	if (!mp)
> +	if (!lookup_mountpoint(dentry, &mp))
>  		goto out_unlock;
>  
>  	event++;
> -	while (!hlist_empty(&mp->m_list)) {
> -		mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list);
> +	while (mp.node.next) {
> +		mnt = hlist_entry(mp.node.next, struct mount, mnt_mp_list);
>  		if (mnt->mnt.mnt_flags & MNT_UMOUNT) {
>  			umount_mnt(mnt);
>  			hlist_add_head(&mnt->mnt_umount, &unmounted);
>  		}
>  		else umount_tree(mnt, UMOUNT_CONNECTED);
>  	}
> -	put_mountpoint(mp);
> +	unpin_mountpoint(&mp);
>  out_unlock:
>  	unlock_mount_hash();
>  	namespace_unlock();
> @@ -2618,7 +2621,7 @@ static int attach_recursive_mnt(struct mount *source_mnt,
>  	struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
>  	HLIST_HEAD(tree_list);
>  	struct mnt_namespace *ns = dest_mnt->mnt_ns;
> -	struct mountpoint *smp;
> +	struct pinned_mountpoint root = {};
>  	struct mount *child, *p;
>  	struct hlist_node *n;
>  	int err = 0;
> @@ -2628,9 +2631,9 @@ static int attach_recursive_mnt(struct mount *source_mnt,
>  	 * Preallocate a mountpoint in case the new mounts need to be
>  	 * mounted beneath mounts on the same mountpoint.
>  	 */
> -	smp = get_mountpoint(source_mnt->mnt.mnt_root);
> -	if (IS_ERR(smp))
> -		return PTR_ERR(smp);
> +	err = get_mountpoint(source_mnt->mnt.mnt_root, &root);
> +	if (err)
> +		return err;
>  
>  	/* Is there space to add these mounts to the mount namespace? */
>  	if (!moving) {
> @@ -2680,13 +2683,13 @@ static int attach_recursive_mnt(struct mount *source_mnt,
>  		q = __lookup_mnt(&child->mnt_parent->mnt,
>  				 child->mnt_mountpoint);
>  		if (q)
> -			mnt_change_mountpoint(child, smp, q);
> +			mnt_change_mountpoint(child, root.mp, q);
>  		/* Notice when we are propagating across user namespaces */
>  		if (child->mnt_parent->mnt_ns->user_ns != user_ns)
>  			lock_mnt_tree(child);
>  		commit_tree(child);
>  	}
> -	put_mountpoint(smp);
> +	unpin_mountpoint(&root);
>  	unlock_mount_hash();
>  
>  	return 0;
> @@ -2703,7 +2706,7 @@ static int attach_recursive_mnt(struct mount *source_mnt,
>  	ns->pending_mounts = 0;
>  
>  	read_seqlock_excl(&mount_lock);
> -	put_mountpoint(smp);
> +	unpin_mountpoint(&root);
>  	read_sequnlock_excl(&mount_lock);
>  
>  	return err;
> @@ -2743,12 +2746,12 @@ static int attach_recursive_mnt(struct mount *source_mnt,
>   * Return: Either the target mountpoint on the top mount or the top
>   *         mount's mountpoint.
>   */
> -static struct mountpoint *do_lock_mount(struct path *path, bool beneath)
> +static int do_lock_mount(struct path *path, struct pinned_mountpoint *pinned, bool beneath)
>  {
>  	struct vfsmount *mnt = path->mnt;
>  	struct dentry *dentry;
> -	struct mountpoint *mp = ERR_PTR(-ENOENT);
>  	struct path under = {};
> +	int err = -ENOENT;
>  
>  	for (;;) {
>  		struct mount *m = real_mount(mnt);
> @@ -2786,8 +2789,8 @@ static struct mountpoint *do_lock_mount(struct path *path, bool beneath)
>  			path->dentry = dget(mnt->mnt_root);
>  			continue;	// got overmounted
>  		}
> -		mp = get_mountpoint(dentry);
> -		if (IS_ERR(mp))
> +		err = get_mountpoint(dentry, pinned);
> +		if (err)
>  			break;
>  		if (beneath) {
>  			/*
> @@ -2798,25 +2801,25 @@ static struct mountpoint *do_lock_mount(struct path *path, bool beneath)
>  			 */
>  			path_put(&under);
>  		}
> -		return mp;
> +		return 0;
>  	}
>  	namespace_unlock();
>  	inode_unlock(dentry->d_inode);
>  	if (beneath)
>  		path_put(&under);
> -	return mp;
> +	return err;
>  }
>  
> -static inline struct mountpoint *lock_mount(struct path *path)
> +static inline int lock_mount(struct path *path, struct pinned_mountpoint *m)
>  {
> -	return do_lock_mount(path, false);
> +	return do_lock_mount(path, m, false);
>  }
>  
> -static void unlock_mount(struct mountpoint *where)
> +static void unlock_mount(struct pinned_mountpoint *m)
>  {
> -	inode_unlock(where->m_dentry->d_inode);
> +	inode_unlock(m->mp->m_dentry->d_inode);
>  	read_seqlock_excl(&mount_lock);
> -	put_mountpoint(where);
> +	unpin_mountpoint(m);
>  	read_sequnlock_excl(&mount_lock);
>  	namespace_unlock();
>  }
> @@ -2981,7 +2984,7 @@ static int do_loopback(struct path *path, const char *old_name,
>  {
>  	struct path old_path;
>  	struct mount *mnt = NULL, *parent;
> -	struct mountpoint *mp;
> +	struct pinned_mountpoint mp = {};
>  	int err;
>  	if (!old_name || !*old_name)
>  		return -EINVAL;
> @@ -2993,11 +2996,9 @@ static int do_loopback(struct path *path, const char *old_name,
>  	if (mnt_ns_loop(old_path.dentry))
>  		goto out;
>  
> -	mp = lock_mount(path);
> -	if (IS_ERR(mp)) {
> -		err = PTR_ERR(mp);
> +	err = lock_mount(path, &mp);
> +	if (err)
>  		goto out;
> -	}
>  
>  	parent = real_mount(path->mnt);
>  	if (!check_mnt(parent))
> @@ -3009,14 +3010,14 @@ static int do_loopback(struct path *path, const char *old_name,
>  		goto out2;
>  	}
>  
> -	err = graft_tree(mnt, parent, mp);
> +	err = graft_tree(mnt, parent, mp.mp);
>  	if (err) {
>  		lock_mount_hash();
>  		umount_tree(mnt, UMOUNT_SYNC);
>  		unlock_mount_hash();
>  	}
>  out2:
> -	unlock_mount(mp);
> +	unlock_mount(&mp);
>  out:
>  	path_put(&old_path);
>  	return err;
> @@ -3560,13 +3561,13 @@ static int do_move_mount(struct path *old_path,
>  	struct mount *p;
>  	struct mount *old;
>  	struct mount *parent;
> -	struct mountpoint *mp;
> +	struct pinned_mountpoint mp;
>  	int err;
>  	bool beneath = flags & MNT_TREE_BENEATH;
>  
> -	mp = do_lock_mount(new_path, beneath);
> -	if (IS_ERR(mp))
> -		return PTR_ERR(mp);
> +	err = do_lock_mount(new_path, &mp, beneath);
> +	if (err)
> +		return err;
>  
>  	old = real_mount(old_path->mnt);
>  	p = real_mount(new_path->mnt);
> @@ -3615,7 +3616,7 @@ static int do_move_mount(struct path *old_path,
>  		goto out;
>  
>  	if (beneath) {
> -		err = can_move_mount_beneath(old_path, new_path, mp);
> +		err = can_move_mount_beneath(old_path, new_path, mp.mp);
>  		if (err)
>  			goto out;
>  
> @@ -3635,9 +3636,9 @@ static int do_move_mount(struct path *old_path,
>  	if (mount_is_ancestor(old, p))
>  		goto out;
>  
> -	err = attach_recursive_mnt(old, p, mp);
> +	err = attach_recursive_mnt(old, p, mp.mp);
>  out:
> -	unlock_mount(mp);
> +	unlock_mount(&mp);
>  	if (!err) {
>  		if (!is_anon_ns(ns)) {
>  			mntput_no_expire(parent);
> @@ -3707,7 +3708,7 @@ static int do_new_mount_fc(struct fs_context *fc, struct path *mountpoint,
>  			   unsigned int mnt_flags)
>  {
>  	struct vfsmount *mnt;
> -	struct mountpoint *mp;
> +	struct pinned_mountpoint mp = {};
>  	struct super_block *sb = fc->root->d_sb;
>  	int error;
>  
> @@ -3728,13 +3729,12 @@ static int do_new_mount_fc(struct fs_context *fc, struct path *mountpoint,
>  
>  	mnt_warn_timestamp_expiry(mountpoint, mnt);
>  
> -	mp = lock_mount(mountpoint);
> -	if (IS_ERR(mp)) {
> -		mntput(mnt);
> -		return PTR_ERR(mp);
> +	error = lock_mount(mountpoint, &mp);
> +	if (!error) {
> +		error = do_add_mount(real_mount(mnt), mp.mp,
> +				     mountpoint, mnt_flags);
> +		unlock_mount(&mp);
>  	}
> -	error = do_add_mount(real_mount(mnt), mp, mountpoint, mnt_flags);
> -	unlock_mount(mp);
>  	if (error < 0)
>  		mntput(mnt);
>  	return error;
> @@ -3802,7 +3802,7 @@ static int do_new_mount(struct path *path, const char *fstype, int sb_flags,
>  int finish_automount(struct vfsmount *m, const struct path *path)
>  {
>  	struct dentry *dentry = path->dentry;
> -	struct mountpoint *mp;
> +	struct pinned_mountpoint mp = {};
>  	struct mount *mnt;
>  	int err;
>  
> @@ -3834,14 +3834,13 @@ int finish_automount(struct vfsmount *m, const struct path *path)
>  		err = 0;
>  		goto discard_locked;
>  	}
> -	mp = get_mountpoint(dentry);
> -	if (IS_ERR(mp)) {
> -		err = PTR_ERR(mp);
> +	err = get_mountpoint(dentry, &mp);
> +	if (err)
>  		goto discard_locked;
> -	}
>  
> -	err = do_add_mount(mnt, mp, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
> -	unlock_mount(mp);
> +	err = do_add_mount(mnt, mp.mp, path,
> +			   path->mnt->mnt_flags | MNT_SHRINKABLE);
> +	unlock_mount(&mp);
>  	if (unlikely(err))
>  		goto discard;
>  	return 0;
> @@ -4642,7 +4641,7 @@ SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
>  {
>  	struct path new, old, root;
>  	struct mount *new_mnt, *root_mnt, *old_mnt, *root_parent, *ex_parent;
> -	struct mountpoint *old_mp;
> +	struct pinned_mountpoint old_mp = {};
>  	int error;
>  
>  	if (!may_mount())
> @@ -4663,9 +4662,8 @@ SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
>  		goto out2;
>  
>  	get_fs_root(current->fs, &root);
> -	old_mp = lock_mount(&old);
> -	error = PTR_ERR(old_mp);
> -	if (IS_ERR(old_mp))
> +	error = lock_mount(&old, &old_mp);
> +	if (error)
>  		goto out3;
>  
>  	error = -EINVAL;
> @@ -4714,7 +4712,7 @@ SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
>  	umount_mnt(root_mnt);
>  	mnt_add_count(root_parent, -1);
>  	/* mount old root on put_old */
> -	attach_mnt(root_mnt, old_mnt, old_mp);
> +	attach_mnt(root_mnt, old_mnt, old_mp.mp);
>  	touch_mnt_namespace(current->nsproxy->mnt_ns);
>  	/* A moved mount should not expire automatically */
>  	list_del_init(&new_mnt->mnt_expire);
> @@ -4724,7 +4722,7 @@ SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
>  	chroot_fs_refs(&root, &new);
>  	error = 0;
>  out4:
> -	unlock_mount(old_mp);
> +	unlock_mount(&old_mp);
>  	if (!error)
>  		mntput_no_expire(ex_parent);
>  out3:
> -- 
> 2.39.5
> 

Re: [PATCH 25/26] get rid of mountpoint->m_count

[Al Viro]

On Wed, Jun 11, 2025 at 01:19:43PM +0200, Christian Brauner wrote:

> This feels well-suited for a DEFINE_FREE based annotation so that
> unpin_mountpoint() is called when the scope ends.

FWIW, I'd be more interested in having unlock_mount() treated that
way, but I'm not sure what syntax would make sense there.

scoped_cond_guard() is not a good fit, unfortunately...

Re: [PATCH 25/26] get rid of mountpoint->m_count

[Al Viro]

On Wed, Jun 11, 2025 at 07:47:00PM +0100, Al Viro wrote:
> On Wed, Jun 11, 2025 at 01:19:43PM +0200, Christian Brauner wrote:
> 
> > This feels well-suited for a DEFINE_FREE based annotation so that
> > unpin_mountpoint() is called when the scope ends.
> 
> FWIW, I'd be more interested in having unlock_mount() treated that
> way, but I'm not sure what syntax would make sense there.
> 
> scoped_cond_guard() is not a good fit, unfortunately...

Folks, how much would you hate the following trick:

lock_mount(path, &m) returning void, and indicating error by storing
ERR_PTR(-E...) into m.path; unlock_mount(&m) doing nothing if IS_ERR(m.mp);
users turned into

	scoped_guard(lock_mount, mp)(path) {
		if (IS_ERR(mp.mp))
			return ERR_PTR(mp.mp);
		....
	}

Re: [PATCH 25/26] get rid of mountpoint->m_count

[Linus Torvalds]

On Mon, 16 Jun 2025 at 13:38, Al Viro <viro@zeniv.linux.org.uk> wrote:
>
> Folks, how much would you hate the following trick:
>
> lock_mount(path, &m) returning void, and indicating error by storing
> ERR_PTR(-E...) into m.path; unlock_mount(&m) doing nothing if IS_ERR(m.mp);
> users turned into
>
>         scoped_guard(lock_mount, mp)(path) {
>                 if (IS_ERR(mp.mp))
>                         return ERR_PTR(mp.mp);
>                 ....
>         }

Eww.

It feels like using 'scoped_cond_guard()' is the *much* more
appropriate model, even if that interface is kind of disgusting.

The it would literally become

        scoped_cond_guard(lock_mount, return ERR_PTR(mp.mp), mp, path) {

except I do admit that our "cond" guard syntax is horribly nasty - see
the several threads people have had about trying to make them better.
You've been cc'd on at least some of them.

Maybe using lock_mount() could help figure out what the right syntax
for the conditional guards are.

             Linus

[PATCH 26/26] don't have mounts pin their parents

[Al Viro]

Simplify the rules for mount refcounts.  Current rules include:
	* being a namespace root => +1
	* being someone's child => +1
	* being someone's child => +1 to parent's refcount, unless you've
				   already been through umount_tree().

The last part is not needed at all.  It makes for more places where need
to decrement refcounts and it creates an asymmetry between the situations
for something that has never been a part of a namespace and something that
left one, both for no good reason.

If mount's refcount has additions from its children, we know that
	* it's either someone's child itself (and will remain so
until umount_tree(), at which point contributions from children
will disappear), or
	* or is the root of namespace (and will remain such until
it either becomes someone's child in another namespace or goes through
umount_tree()), or
	* it is the root of some tree copy, and is currently pinned
by the caller of copy_tree() (and remains such until it either gets
into namespace, or goes to umount_tree()).
In all cases we already have contribution(s) to refcount that will last
as long as the contribution from children remains.  In other words, the
lifetime is not affected by refcount contributions from children.

It might be useful for "is it busy" checks, but those are actually
no harder to express without it.

NB: propagate_mnt_busy() part is an equivalent transformation, ugly as it
is; the current logics is actually wrong and may give false negatives,
but fixing that is for a separate patch (probably earlier in the queue).

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 31 +++++++++--------------------
 fs/pnode.c     | 53 ++++++++++++++++++++------------------------------
 2 files changed, 30 insertions(+), 54 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index 1f1cf1d6a464..1bfc26098fe3 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -1072,7 +1072,6 @@ void mnt_set_mountpoint(struct mount *mnt,
 			struct mountpoint *mp,
 			struct mount *child_mnt)
 {
-	mnt_add_count(mnt, 1);	/* essentially, that's mntget */
 	child_mnt->mnt_mountpoint = mp->m_dentry;
 	child_mnt->mnt_parent = mnt;
 	child_mnt->mnt_mp = mp;
@@ -1112,7 +1111,6 @@ static void attach_mnt(struct mount *mnt, struct mount *parent,
 void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt)
 {
 	struct mountpoint *old_mp = mnt->mnt_mp;
-	struct mount *old_parent = mnt->mnt_parent;
 
 	list_del_init(&mnt->mnt_child);
 	hlist_del_init(&mnt->mnt_mp_list);
@@ -1121,7 +1119,6 @@ void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct m
 	attach_mnt(mnt, parent, mp);
 
 	maybe_free_mountpoint(old_mp, &ex_mountpoints);
-	mnt_add_count(old_parent, -1);
 }
 
 static inline struct mount *node_to_mount(struct rb_node *node)
@@ -1646,23 +1643,19 @@ const struct seq_operations mounts_op = {
 int may_umount_tree(struct vfsmount *m)
 {
 	struct mount *mnt = real_mount(m);
-	int actual_refs = 0;
-	int minimum_refs = 0;
-	struct mount *p;
-	BUG_ON(!m);
+	bool busy = false;
 
 	/* write lock needed for mnt_get_count */
 	lock_mount_hash();
-	for (p = mnt; p; p = next_mnt(p, mnt)) {
-		actual_refs += mnt_get_count(p);
-		minimum_refs += 2;
+	for (struct mount *p = mnt; p; p = next_mnt(p, mnt)) {
+		if (mnt_get_count(p) > (p == mnt ? 2 : 1)) {
+			busy = true;
+			break;
+		}
 	}
 	unlock_mount_hash();
 
-	if (actual_refs > minimum_refs)
-		return 0;
-
-	return 1;
+	return !busy;
 }
 
 EXPORT_SYMBOL(may_umount_tree);
@@ -1863,7 +1856,6 @@ static void umount_tree(struct mount *mnt, enum umount_tree_flags how)
 
 		disconnect = disconnect_mount(p, how);
 		if (mnt_has_parent(p)) {
-			mnt_add_count(p->mnt_parent, -1);
 			if (!disconnect) {
 				/* Don't forget about p */
 				list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts);
@@ -1940,7 +1932,7 @@ static int do_umount(struct mount *mnt, int flags)
 		 * all race cases, but it's a slowpath.
 		 */
 		lock_mount_hash();
-		if (mnt_get_count(mnt) != 2) {
+		if (!list_empty(&mnt->mnt_mounts) || mnt_get_count(mnt) != 2) {
 			unlock_mount_hash();
 			return -EBUSY;
 		}
@@ -3640,9 +3632,7 @@ static int do_move_mount(struct path *old_path,
 out:
 	unlock_mount(&mp);
 	if (!err) {
-		if (!is_anon_ns(ns)) {
-			mntput_no_expire(parent);
-		} else {
+		if (is_anon_ns(ns)) {
 			/* Make sure we notice when we leak mounts. */
 			VFS_WARN_ON_ONCE(!mnt_ns_empty(ns));
 			free_mnt_ns(ns);
@@ -4710,7 +4700,6 @@ SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
 	/* mount new_root on / */
 	attach_mnt(new_mnt, root_parent, root_mnt->mnt_mp);
 	umount_mnt(root_mnt);
-	mnt_add_count(root_parent, -1);
 	/* mount old root on put_old */
 	attach_mnt(root_mnt, old_mnt, old_mp.mp);
 	touch_mnt_namespace(current->nsproxy->mnt_ns);
@@ -4723,8 +4712,6 @@ SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
 	error = 0;
 out4:
 	unlock_mount(&old_mp);
-	if (!error)
-		mntput_no_expire(ex_parent);
 out3:
 	path_put(&root);
 out2:
diff --git a/fs/pnode.c b/fs/pnode.c
index f1752dd499af..efed6bb20c72 100644
--- a/fs/pnode.c
+++ b/fs/pnode.c
@@ -332,21 +332,6 @@ int propagate_mnt(struct mount *dest_mnt, struct mountpoint *dest_mp,
 	return ret;
 }
 
-static struct mount *find_topper(struct mount *mnt)
-{
-	/* If there is exactly one mount covering mnt completely return it. */
-	struct mount *child;
-
-	if (!list_is_singular(&mnt->mnt_mounts))
-		return NULL;
-
-	child = list_first_entry(&mnt->mnt_mounts, struct mount, mnt_child);
-	if (child->mnt_mountpoint != mnt->mnt.mnt_root)
-		return NULL;
-
-	return child;
-}
-
 /*
  * return true if the refcount is greater than count
  */
@@ -404,12 +389,8 @@ bool propagation_would_overmount(const struct mount *from,
  */
 int propagate_mount_busy(struct mount *mnt, int refcnt)
 {
-	struct mount *m, *child, *topper;
 	struct mount *parent = mnt->mnt_parent;
 
-	if (mnt == parent)
-		return do_refcount_check(mnt, refcnt);
-
 	/*
 	 * quickly check if the current mount can be unmounted.
 	 * If not, we don't have to go checking for all other
@@ -418,23 +399,31 @@ int propagate_mount_busy(struct mount *mnt, int refcnt)
 	if (!list_empty(&mnt->mnt_mounts) || do_refcount_check(mnt, refcnt))
 		return 1;
 
-	for (m = propagation_next(parent, parent); m;
+	if (mnt == parent)
+		return 0;
+
+	for (struct mount *m = propagation_next(parent, parent); m;
 	     		m = propagation_next(m, parent)) {
-		int count = 1;
-		child = __lookup_mnt(&m->mnt, mnt->mnt_mountpoint);
-		if (!child)
-			continue;
+		struct list_head *head;
+		struct mount *child = __lookup_mnt(&m->mnt, mnt->mnt_mountpoint);
 
-		/* Is there exactly one mount on the child that covers
-		 * it completely whose reference should be ignored?
-		 */
-		topper = find_topper(child);
-		if (topper)
-			count += 1;
-		else if (!list_empty(&child->mnt_mounts))
+		if (!child)
 			continue;
 
-		if (do_refcount_check(child, count))
+		head = &child->mnt_mounts;
+		if (!list_empty(head)) {
+			struct mount *p;
+			/*
+			 * a mount that covers child completely wouldn't prevent
+			 * it being pulled out; any other would.
+			 */
+			if (head->next != head->prev)
+				continue;
+			p = list_first_entry(head, struct mount, mnt_child);
+			if (p->mnt_mountpoint != p->mnt.mnt_root)
+				continue;
+		}
+		if (do_refcount_check(child, 1))
 			return 1;
 	}
 	return 0;
-- 
2.39.5

Re: [PATCH 26/26] don't have mounts pin their parents

[Christian Brauner]

On Tue, Jun 10, 2025 at 09:21:48AM +0100, Al Viro wrote:
> Simplify the rules for mount refcounts.  Current rules include:
> 	* being a namespace root => +1
> 	* being someone's child => +1
> 	* being someone's child => +1 to parent's refcount, unless you've
> 				   already been through umount_tree().
> 
> The last part is not needed at all.  It makes for more places where need
> to decrement refcounts and it creates an asymmetry between the situations
> for something that has never been a part of a namespace and something that
> left one, both for no good reason.
> 
> If mount's refcount has additions from its children, we know that
> 	* it's either someone's child itself (and will remain so
> until umount_tree(), at which point contributions from children
> will disappear), or
> 	* or is the root of namespace (and will remain such until
> it either becomes someone's child in another namespace or goes through
> umount_tree()), or
> 	* it is the root of some tree copy, and is currently pinned
> by the caller of copy_tree() (and remains such until it either gets
> into namespace, or goes to umount_tree()).
> In all cases we already have contribution(s) to refcount that will last
> as long as the contribution from children remains.  In other words, the
> lifetime is not affected by refcount contributions from children.
> 
> It might be useful for "is it busy" checks, but those are actually
> no harder to express without it.
> 
> NB: propagate_mnt_busy() part is an equivalent transformation, ugly as it
> is; the current logics is actually wrong and may give false negatives,
> but fixing that is for a separate patch (probably earlier in the queue).
> 
> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
> ---

Reviewed-by: Christian Brauner <brauner@kernel.org>

Re: [PATCH 26/26] don't have mounts pin their parents

[Ian Kent]

On 10/6/25 16:21, Al Viro wrote:
> Simplify the rules for mount refcounts.  Current rules include:
> 	* being a namespace root => +1
> 	* being someone's child => +1
> 	* being someone's child => +1 to parent's refcount, unless you've
> 				   already been through umount_tree().
>
> The last part is not needed at all.  It makes for more places where need
> to decrement refcounts and it creates an asymmetry between the situations
> for something that has never been a part of a namespace and something that
> left one, both for no good reason.
>
> If mount's refcount has additions from its children, we know that
> 	* it's either someone's child itself (and will remain so
> until umount_tree(), at which point contributions from children
> will disappear), or
> 	* or is the root of namespace (and will remain such until
> it either becomes someone's child in another namespace or goes through
> umount_tree()), or
> 	* it is the root of some tree copy, and is currently pinned
> by the caller of copy_tree() (and remains such until it either gets
> into namespace, or goes to umount_tree()).
> In all cases we already have contribution(s) to refcount that will last
> as long as the contribution from children remains.  In other words, the
> lifetime is not affected by refcount contributions from children.
>
> It might be useful for "is it busy" checks, but those are actually
> no harder to express without it.
>
> NB: propagate_mnt_busy() part is an equivalent transformation, ugly as it
> is; the current logics is actually wrong and may give false negatives,
> but fixing that is for a separate patch (probably earlier in the queue).

This looks a lot like the patch you sent over long ago, ;)


I did spend quite a bit of time on fixing may_umount_tree() but I was always

concerned about how the parent ref count change would be received and I had

trouble with one of my two patches which essentially tried to add an fairly

aggressive optimization. Early on I had it working but kept getting the 
feeling

it was too aggressive and later it stopped working and I couldn't work 
out how

to fix it so I dropped it.


In any case the first of my patches always worked and IMHO is adequate to be

used to fix the may_umount_tree() namespace awareness problem. I guess 
we may

soon see if all goes well with this series.


I'll run my two usual autofs tests against this (once I've setup an VM 
for it)

and report back.


>
> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
> ---
>   fs/namespace.c | 31 +++++++++--------------------
>   fs/pnode.c     | 53 ++++++++++++++++++++------------------------------
>   2 files changed, 30 insertions(+), 54 deletions(-)
>
> diff --git a/fs/namespace.c b/fs/namespace.c
> index 1f1cf1d6a464..1bfc26098fe3 100644
> --- a/fs/namespace.c
> +++ b/fs/namespace.c
> @@ -1072,7 +1072,6 @@ void mnt_set_mountpoint(struct mount *mnt,
>   			struct mountpoint *mp,
>   			struct mount *child_mnt)
>   {
> -	mnt_add_count(mnt, 1);	/* essentially, that's mntget */
>   	child_mnt->mnt_mountpoint = mp->m_dentry;
>   	child_mnt->mnt_parent = mnt;
>   	child_mnt->mnt_mp = mp;
> @@ -1112,7 +1111,6 @@ static void attach_mnt(struct mount *mnt, struct mount *parent,
>   void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt)
>   {
>   	struct mountpoint *old_mp = mnt->mnt_mp;
> -	struct mount *old_parent = mnt->mnt_parent;
>   
>   	list_del_init(&mnt->mnt_child);
>   	hlist_del_init(&mnt->mnt_mp_list);
> @@ -1121,7 +1119,6 @@ void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct m
>   	attach_mnt(mnt, parent, mp);
>   
>   	maybe_free_mountpoint(old_mp, &ex_mountpoints);
> -	mnt_add_count(old_parent, -1);
>   }
>   
>   static inline struct mount *node_to_mount(struct rb_node *node)
> @@ -1646,23 +1643,19 @@ const struct seq_operations mounts_op = {
>   int may_umount_tree(struct vfsmount *m)
>   {
>   	struct mount *mnt = real_mount(m);
> -	int actual_refs = 0;
> -	int minimum_refs = 0;
> -	struct mount *p;
> -	BUG_ON(!m);
> +	bool busy = false;
>   
>   	/* write lock needed for mnt_get_count */
>   	lock_mount_hash();
> -	for (p = mnt; p; p = next_mnt(p, mnt)) {
> -		actual_refs += mnt_get_count(p);
> -		minimum_refs += 2;
> +	for (struct mount *p = mnt; p; p = next_mnt(p, mnt)) {
> +		if (mnt_get_count(p) > (p == mnt ? 2 : 1)) {
> +			busy = true;
> +			break;
> +		}

This function is broken, it's not namespace aware (as you know).

But, surprisingly, it's adequate in many cases.

For example autofs uses this when expiring trees of mounts and v5 does 
this in subtrees

from the bottom up so the single level check works fine unless it's in 
use in another

namespace which (surprisingly) doesn't happen very often. In any case 
the consequence

is it fails to umount a subtree or two which must be handled by the 
daemon anyway.


Nevertheless the namespace awareness needs to be fixed.


Ian

>   	}
>   	unlock_mount_hash();
>   
> -	if (actual_refs > minimum_refs)
> -		return 0;
> -
> -	return 1;
> +	return !busy;
>   }
>   
>   EXPORT_SYMBOL(may_umount_tree);
> @@ -1863,7 +1856,6 @@ static void umount_tree(struct mount *mnt, enum umount_tree_flags how)
>   
>   		disconnect = disconnect_mount(p, how);
>   		if (mnt_has_parent(p)) {
> -			mnt_add_count(p->mnt_parent, -1);
>   			if (!disconnect) {
>   				/* Don't forget about p */
>   				list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts);
> @@ -1940,7 +1932,7 @@ static int do_umount(struct mount *mnt, int flags)
>   		 * all race cases, but it's a slowpath.
>   		 */
>   		lock_mount_hash();
> -		if (mnt_get_count(mnt) != 2) {
> +		if (!list_empty(&mnt->mnt_mounts) || mnt_get_count(mnt) != 2) {
>   			unlock_mount_hash();
>   			return -EBUSY;
>   		}
> @@ -3640,9 +3632,7 @@ static int do_move_mount(struct path *old_path,
>   out:
>   	unlock_mount(&mp);
>   	if (!err) {
> -		if (!is_anon_ns(ns)) {
> -			mntput_no_expire(parent);
> -		} else {
> +		if (is_anon_ns(ns)) {
>   			/* Make sure we notice when we leak mounts. */
>   			VFS_WARN_ON_ONCE(!mnt_ns_empty(ns));
>   			free_mnt_ns(ns);
> @@ -4710,7 +4700,6 @@ SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
>   	/* mount new_root on / */
>   	attach_mnt(new_mnt, root_parent, root_mnt->mnt_mp);
>   	umount_mnt(root_mnt);
> -	mnt_add_count(root_parent, -1);
>   	/* mount old root on put_old */
>   	attach_mnt(root_mnt, old_mnt, old_mp.mp);
>   	touch_mnt_namespace(current->nsproxy->mnt_ns);
> @@ -4723,8 +4712,6 @@ SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
>   	error = 0;
>   out4:
>   	unlock_mount(&old_mp);
> -	if (!error)
> -		mntput_no_expire(ex_parent);
>   out3:
>   	path_put(&root);
>   out2:
> diff --git a/fs/pnode.c b/fs/pnode.c
> index f1752dd499af..efed6bb20c72 100644
> --- a/fs/pnode.c
> +++ b/fs/pnode.c
> @@ -332,21 +332,6 @@ int propagate_mnt(struct mount *dest_mnt, struct mountpoint *dest_mp,
>   	return ret;
>   }
>   
> -static struct mount *find_topper(struct mount *mnt)
> -{
> -	/* If there is exactly one mount covering mnt completely return it. */
> -	struct mount *child;
> -
> -	if (!list_is_singular(&mnt->mnt_mounts))
> -		return NULL;
> -
> -	child = list_first_entry(&mnt->mnt_mounts, struct mount, mnt_child);
> -	if (child->mnt_mountpoint != mnt->mnt.mnt_root)
> -		return NULL;
> -
> -	return child;
> -}
> -
>   /*
>    * return true if the refcount is greater than count
>    */
> @@ -404,12 +389,8 @@ bool propagation_would_overmount(const struct mount *from,
>    */
>   int propagate_mount_busy(struct mount *mnt, int refcnt)
>   {
> -	struct mount *m, *child, *topper;
>   	struct mount *parent = mnt->mnt_parent;
>   
> -	if (mnt == parent)
> -		return do_refcount_check(mnt, refcnt);
> -
>   	/*
>   	 * quickly check if the current mount can be unmounted.
>   	 * If not, we don't have to go checking for all other
> @@ -418,23 +399,31 @@ int propagate_mount_busy(struct mount *mnt, int refcnt)
>   	if (!list_empty(&mnt->mnt_mounts) || do_refcount_check(mnt, refcnt))
>   		return 1;
>   
> -	for (m = propagation_next(parent, parent); m;
> +	if (mnt == parent)
> +		return 0;
> +
> +	for (struct mount *m = propagation_next(parent, parent); m;
>   	     		m = propagation_next(m, parent)) {
> -		int count = 1;
> -		child = __lookup_mnt(&m->mnt, mnt->mnt_mountpoint);
> -		if (!child)
> -			continue;
> +		struct list_head *head;
> +		struct mount *child = __lookup_mnt(&m->mnt, mnt->mnt_mountpoint);
>   
> -		/* Is there exactly one mount on the child that covers
> -		 * it completely whose reference should be ignored?
> -		 */
> -		topper = find_topper(child);
> -		if (topper)
> -			count += 1;
> -		else if (!list_empty(&child->mnt_mounts))
> +		if (!child)
>   			continue;
>   
> -		if (do_refcount_check(child, count))
> +		head = &child->mnt_mounts;
> +		if (!list_empty(head)) {
> +			struct mount *p;
> +			/*
> +			 * a mount that covers child completely wouldn't prevent
> +			 * it being pulled out; any other would.
> +			 */
> +			if (head->next != head->prev)
> +				continue;
> +			p = list_first_entry(head, struct mount, mnt_child);
> +			if (p->mnt_mountpoint != p->mnt.mnt_root)
> +				continue;
> +		}
> +		if (do_refcount_check(child, 1))
>   			return 1;
>   	}
>   	return 0;

Re: [PATCH 01/26] copy_tree(): don't set ->mnt_mountpoint on the root of copy

[Eric W. Biederman]

Al Viro <viro@zeniv.linux.org.uk> writes:

> It never made any sense - neither when copy_tree() had been introduced
> (2.4.11-pre5), nor at any point afterwards.  Mountpoint is meaningless
> without parent mount and the root of copied tree has no parent until we get
> around to attaching it somewhere.  At that time we'll have mountpoint set;
> before that we have no idea which dentry will be used as mountpoint.
> IOW, copy_tree() should just leave the default value.

I will just note that does not result in dst_mnt->mnt_mountpoint
being left as NULL.

Rather dst_mnt->mnt_mountpoint retains the value that clone_mnt
sets it to which is dst_mnt->mnt.mnt_root.

It would be nice to have a note that says something like leaving
dst_mnt->mnt_parent and dst_mnt->mnt_mountpoint alone indicates that the
mount is not mounted anywhere, and that the current situation of just
setting one of them completely confusing.

> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
> ---
>  fs/namespace.c | 1 -
>  1 file changed, 1 deletion(-)
>
> diff --git a/fs/namespace.c b/fs/namespace.c
> index e13d9ab4f564..5eeb17c39fcb 100644
> --- a/fs/namespace.c
> +++ b/fs/namespace.c
> @@ -2259,7 +2259,6 @@ struct mount *copy_tree(struct mount *src_root, struct dentry *dentry,
>  		return dst_mnt;
>  
>  	src_parent = src_root;
> -	dst_mnt->mnt_mountpoint = src_root->mnt_mountpoint;
>  
>  	list_for_each_entry(src_root_child, &src_root->mnt_mounts, mnt_child) {
>  		if (!is_subdir(src_root_child->mnt_mountpoint, dentry))

Re: [PATCH 01/26] copy_tree(): don't set ->mnt_mountpoint on the root of copy

[Al Viro]

On Tue, Jun 10, 2025 at 05:30:11PM -0500, Eric W. Biederman wrote:
> Al Viro <viro@zeniv.linux.org.uk> writes:
> 
> > It never made any sense - neither when copy_tree() had been introduced
> > (2.4.11-pre5), nor at any point afterwards.  Mountpoint is meaningless
> > without parent mount and the root of copied tree has no parent until we get
> > around to attaching it somewhere.  At that time we'll have mountpoint set;
> > before that we have no idea which dentry will be used as mountpoint.
> > IOW, copy_tree() should just leave the default value.
> 
> I will just note that does not result in dst_mnt->mnt_mountpoint
> being left as NULL.
> 
> Rather dst_mnt->mnt_mountpoint retains the value that clone_mnt
> sets it to which is dst_mnt->mnt.mnt_root.
> 
> It would be nice to have a note that says something like leaving
> dst_mnt->mnt_parent and dst_mnt->mnt_mountpoint alone indicates that the
> mount is not mounted anywhere, and that the current situation of just
> setting one of them completely confusing.

s/default value/& for a parentless mount/, perhaps?

<digs through the half-finished documentation>
----------------------------------------------------------------------------
                Rootwards linkage.

        Once a mount has been attached to a subtree of some filesystem,
it becomes a part of forest.  Past that stage each mount is either
parentless or has a parent mount and a mountpoint - some dentry on
the filesystem associated with the parent.

        The linkage is protected by mount_lock.

        Checking if mount is parentless is done by mnt_has_parent(mount);
it returns true for mounts that have a parent and false for parentless
ones.   

        Four fields of struct mount are involved in storing that linkage.
1) struct mount *mnt_parent
        Never NULL, points to self for parentless, to parent mount otherwise.
2) struct dentry *mnt_mountpoint
        Never NULL, points to root dentry of mount itself for parentless
and to mountpoint dentry otherwise.
3) struct mountpoint *mnt_mp.
        NULL for parentless, points to struct mountpoint associated with
mountpoint dentry otherwise.
4) struct hlist_node mnt_mp_list - linkage for the list all mounts sharing
the mountpoint.

        These fields are always updated together.  They make sense only
after mount has been attached to a filesystem - prior to that they happen
to contain NULL (and empty hlist_node), but they are visible only to whoever
had allocated the mount, so nobody else should care.[1]

        The values in these fields are not independent.  If mount m is not
parentless, m->mnt_parent->mnt.mnt_sb == m->mnt_mountpoint->d_sb,
m->mnt_mp->m_dentry == m->mnt_mountpoint and m->mnt_mp_list belongs to
the list anchored in m->mnt_mp->m_list.

        All accesses to ->mnt_mp_list and ->mnt_mp are under mount_lock.
        Access to ->mnt_parent and ->mnt_mountpoint under mount_lock is safe.
        Access to ->mnt_parent and ->mnt_mountpoint under rcu_read_lock() is
memory-safe; it needs to be validated with mount_lock seqcount component
afterwards.
        Access to ->mnt_parent and ->mnt_mountpoint under namespace_sem is
safe for anything crownwards of a pinned mount.  In particular, it is safe
for anything in a mount tree of any namespace, including its rbtree.  It
is also safe for anything reachable via the propagation graph. [XXX: probably
worth an explicit name for that state of a mount]

[1] it might be tempting to change the representation, so that parentless would
have NULL ->mnt_mountpoint; doing that would be a serious headache, though,
especially for RCU traversals towards parent mount.  We really depend upon never
seeing NULL in that field once mount has been attached to filesystem.
----------------------------------------------------------------------------

Re: [PATCHES][RFC][CFR] mount-related stuff

[Christian Brauner]

On Tue, Jun 10, 2025 at 09:17:58AM +0100, Al Viro wrote:
> 	The next pile of mount massage; it will grow - there will be
> further modifications, as well as fixes and documentation, but this is
> the subset I've got in more or less settled form right now.
> 
> 	Review and testing would be very welcome.
> 
> 	This series (-rc1-based) sits in
> git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs.git #work.mount
> individual patches in followups.
> 
> 	Rough overview:
> 
> Part 1: trivial cleanups and helpers:
> 
> 1) copy_tree(): don't set ->mnt_mountpoint on the root of copy
> 	Ancient bogosity, fortunately harmless, but confusing.
> 2) constify mnt_has_parent()
> 3) pnode: lift peers() into pnode.h
> 4) new predicate: mount_is_ancestor()
> 	Incidentally, I wonder if the "early bail out on move
> of anon into the same anon" was not due to (now eliminated)
> corner case in loop detection...  Christian?

No, that wasn't the reason. When moving mounts between anonymous mount
namespaces I wanted a very simple visual barrier that moving mounts into
the same anonymous mount namespace is not possible.

I even mentioned in the comment that this would be caught later but that
I like it being explicitly checked for.

Re: [PATCHES][RFC][CFR] mount-related stuff

[Al Viro]

On Wed, Jun 11, 2025 at 12:31:54PM +0200, Christian Brauner wrote:
> On Tue, Jun 10, 2025 at 09:17:58AM +0100, Al Viro wrote:
> > 	The next pile of mount massage; it will grow - there will be
> > further modifications, as well as fixes and documentation, but this is
> > the subset I've got in more or less settled form right now.
> > 
> > 	Review and testing would be very welcome.
> > 
> > 	This series (-rc1-based) sits in
> > git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs.git #work.mount
> > individual patches in followups.
> > 
> > 	Rough overview:
> > 
> > Part 1: trivial cleanups and helpers:
> > 
> > 1) copy_tree(): don't set ->mnt_mountpoint on the root of copy
> > 	Ancient bogosity, fortunately harmless, but confusing.
> > 2) constify mnt_has_parent()
> > 3) pnode: lift peers() into pnode.h
> > 4) new predicate: mount_is_ancestor()
> > 	Incidentally, I wonder if the "early bail out on move
> > of anon into the same anon" was not due to (now eliminated)
> > corner case in loop detection...  Christian?
> 
> No, that wasn't the reason. When moving mounts between anonymous mount
> namespaces I wanted a very simple visual barrier that moving mounts into
> the same anonymous mount namespace is not possible.
> 
> I even mentioned in the comment that this would be caught later but that
> I like it being explicitly checked for.

OK...  AFAICS, the way those tests were done it would not be caught later.
At the merge time loop detection had been the same as in mainline now:
        for (; mnt_has_parent(p); p = p->mnt_parent)
		if (p == old)
			goto out;
and that will never reach that goto out if mnt_has_parent(old) is false.
The early bailout avoided that problem, thus the question if that's where
it came from...

Re: [PATCHES][RFC][CFR] mount-related stuff

[Christian Brauner]

On Wed, Jun 11, 2025 at 06:51:36PM +0100, Al Viro wrote:
> On Wed, Jun 11, 2025 at 12:31:54PM +0200, Christian Brauner wrote:
> > On Tue, Jun 10, 2025 at 09:17:58AM +0100, Al Viro wrote:
> > > 	The next pile of mount massage; it will grow - there will be
> > > further modifications, as well as fixes and documentation, but this is
> > > the subset I've got in more or less settled form right now.
> > > 
> > > 	Review and testing would be very welcome.
> > > 
> > > 	This series (-rc1-based) sits in
> > > git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs.git #work.mount
> > > individual patches in followups.
> > > 
> > > 	Rough overview:
> > > 
> > > Part 1: trivial cleanups and helpers:
> > > 
> > > 1) copy_tree(): don't set ->mnt_mountpoint on the root of copy
> > > 	Ancient bogosity, fortunately harmless, but confusing.
> > > 2) constify mnt_has_parent()
> > > 3) pnode: lift peers() into pnode.h
> > > 4) new predicate: mount_is_ancestor()
> > > 	Incidentally, I wonder if the "early bail out on move
> > > of anon into the same anon" was not due to (now eliminated)
> > > corner case in loop detection...  Christian?
> > 
> > No, that wasn't the reason. When moving mounts between anonymous mount
> > namespaces I wanted a very simple visual barrier that moving mounts into
> > the same anonymous mount namespace is not possible.
> > 
> > I even mentioned in the comment that this would be caught later but that
> > I like it being explicitly checked for.
> 
> OK...  AFAICS, the way those tests were done it would not be caught later.
> At the merge time loop detection had been the same as in mainline now:
>         for (; mnt_has_parent(p); p = p->mnt_parent)
> 		if (p == old)
> 			goto out;
> and that will never reach that goto out if mnt_has_parent(old) is false.
> The early bailout avoided that problem, thus the question if that's where
> it came from...

Yeah, I mean doing it your way is obviously fine and correct.

[PATCHES v2][RFC][CFR] mount-related stuff

[Al Viro]

On Tue, Jun 10, 2025 at 09:17:58AM +0100, Al Viro wrote:
> 	The next pile of mount massage; it will grow - there will be
> further modifications, as well as fixes and documentation, but this is
> the subset I've got in more or less settled form right now.
> 
> 	Review and testing would be very welcome.
> 
> 	This series (-rc1-based) sits in
> git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs.git #work.mount
> individual patches in followups.

Updated variant force-pushed to
git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs.git #work.mount
individual patches in followups.  It seems to survive testing here, but
more testing and review would be very welcome.  That's still not all -
there's more stuff in local queue, but it needs more massage; this is
the reasonably settled-down subset at the moment.

Changes since v1:
Added fixes (-stable fodder, IMO):
	replace collect_mounts()/drop_collected_mounts() with safer variant
	attach_recursive_mnt(): do not lock the covering tree when sliding something under it

Another thing that is probably a -stable candidate:
	prevent mount hash conflicts
That's an old headache hopefully taken care of; what we get out of it
is the warranty that there won't be more than one hashed mount with
given parent/mountpoint at any given time.  I've pulled that very
early in the sequence, so that we had a chance to backport it.
That comes with two prereqs (reordered from the middle of v1 series),
both should be trivial to backport.

Added cleanups:
	__attach_mnt(): lose the second argument
	copy_tree(): don't link the mounts via mnt_list
All uses of ->mnt_list are transient now - basically, various sets
used during umount_tree().
	mount: separate the flags accessed only under namespace_sem
Makes for simpler locking; some of the flags are accessed only under
namespace_sem, and we already rely upon that in the readers; taking
them to a separate word avoids the need to grab mount_lock on the
write side.
	propagate_one(): get rid of dest_master
	propagate_mnt(): get rid of globals
Linus asked to get rid of fs/pnode.c globals; done.
	take freeing of emptied mnt_namespace to namespace_unlock()

A couple of commits made simpler by "prevent mount hash conflicts" -
	Rewrite of propagate_umount()
reparenting is guaranteed that there won't be more than one overmount now,
no loop needed.
	don't have mounts pin their parents
simpler logics for "is there something other than overmount?"

	Rough overview:

Part 1: fixes

1) replace collect_mounts()/drop_collected_mounts() with safer variant
2) attach_recursive_mnt(): do not lock the covering tree when sliding something under it

Part 2: getting rid of mount hash conflicts for good

3) attach_mnt(): expand in attach_recursive_mnt(), then lose the flag argument
4) get rid of mnt_set_mountpoint_beneath()
5) prevent mount hash conflicts

Part 3: trivial cleanups and helpers:

6) copy_tree(): don't set ->mnt_mountpoint on the root of copy
7) constify mnt_has_parent()
8) pnode: lift peers() into pnode.h
9) new predicate: mount_is_ancestor()
10) constify is_local_mountpoint()
11) new predicate: anon_ns_root(mount)
12) dissolve_on_fput(): use anon_ns_root()
13) __attach_mnt(): lose the second argument
	... and rename to make_visible()
14) don't set MNT_LOCKED on parentless mounts
15) clone_mnt(): simplify the propagation-related logics
16) do_umount(): simplify the "is it still mounted" checks

Part 4: (somewhat of a side story) restore the machinery for long-term
mounts from accumulated bitrot.

17) sanitize handling of long-term internal mounts

Part 5: propagate_umount() rewrite (posted last cycle)

18) Rewrite of propagate_umount()

Part 6: untangling do_move_mount()/attach_recursive_mnt().

19) make commit_tree() usable in same-namespace move case
20) attach_recursive_mnt(): unify the mnt_change_mountpoint() logics
21) attach_recursive_mnt(): pass destination mount in all cases
22) attach_recursive_mnt(): get rid of flags entirely
23) do_move_mount(): take dropping the old mountpoint into attach_recursive_mnt()
24) do_move_mount(): get rid of 'attached' flag

Part 7: change locking for expiry lists.
25) attach_recursive_mnt(): remove from expiry list on move
26) take ->mnt_expire handling under mount_lock [read_seqlock_excl]

Part 8: struct mountpoint massage.
27) pivot_root(): reorder tree surgeries, collapse unhash_mnt() and put_mountpoint()
28) combine __put_mountpoint() with unhash_mnt()
29) get rid of mountpoint->m_count

Part 9: regularize mount refcounting a bit
30) don't have mounts pin their parents

Part 10: misc stuff, will grow...
31) copy_tree(): don't link the mounts via mnt_list
32) mount: separate the flags accessed only under namespace_sem
33) propagate_one(): get rid of dest_master
34) propagate_mnt(): get rid of globals
35) take freeing of emptied mnt_namespace to namespace_unlock()

Diffstat:
 Documentation/filesystems/porting.rst          |   9 +
 Documentation/filesystems/propagate_umount.txt | 484 +++++++++++++++
 drivers/gpu/drm/i915/gem/i915_gemfs.c          |  21 +-
 drivers/gpu/drm/v3d/v3d_gemfs.c                |  21 +-
 fs/hugetlbfs/inode.c                           |   2 +-
 fs/mount.h                                     |  36 +-
 fs/namespace.c                                 | 783 +++++++++++--------------
 fs/pnode.c                                     | 499 ++++++++--------
 fs/pnode.h                                     |  28 +-
 include/linux/mount.h                          |  24 +-
 ipc/mqueue.c                                   |   2 +-
 kernel/audit_tree.c                            |  63 +-
 12 files changed, 1214 insertions(+), 758 deletions(-)
 create mode 100644 Documentation/filesystems/propagate_umount.txt

[PATCH v2 01/35] replace collect_mounts()/drop_collected_mounts() with a safer variant

[Al Viro]

collect_mounts() has several problems - one can't iterate over the results
directly, so it has to be done with callback passed to iterate_mounts();
it has an oopsable race with d_invalidate(); it creates temporary clones
of mounts invisibly for sync umount (IOW, you can have non-lazy umount
succeed leaving filesystem not mounted anywhere and yet still busy).

A saner approach is to give caller an array of struct path that would pin
every mount in a subtree, without cloning any mounts.

        * collect_mounts()/drop_collected_mounts()/iterate_mounts() is gone
        * collect_paths(where, preallocated, size) gives either ERR_PTR(-E...) or
a pointer to array of struct path, one for each chunk of tree visible under
'where' (i.e. the first element is a copy of where, followed by (mount,root)
for everything mounted under it - the same set collect_mounts() would give).
Unlike collect_mounts(), the mounts are *not* cloned - we just get pinning
references to the roots of subtrees in the caller's namespace.
        Array is terminated by {NULL, NULL} struct path.  If it fits into
preallocated array (on-stack, normally), that's where it goes; otherwise
it's allocated by kmalloc_array().  Passing 0 as size means that 'preallocated'
is ignored (and expected to be NULL).
        * drop_collected_paths(paths, preallocated) is given the array returned
by an earlier call of collect_paths() and the preallocated array passed to that
call.  All mount/dentry references are dropped and array is kfree'd if it's not
equal to 'preallocated'.
        * instead of iterate_mounts(), users should just iterate over array
of struct path - nothing exotic is needed for that.  Existing users (all in
audit_tree.c) are converted.

Fixes: 80b5dce8c59b0 ("vfs: Add a function to lazily unmount all mounts from any dentry")
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 Documentation/filesystems/porting.rst |  9 +++
 fs/namespace.c                        | 97 ++++++++++++++++-----------
 fs/pnode.h                            |  2 -
 include/linux/mount.h                 |  6 +-
 kernel/audit_tree.c                   | 63 +++++++++--------
 5 files changed, 104 insertions(+), 73 deletions(-)

diff --git a/Documentation/filesystems/porting.rst b/Documentation/filesystems/porting.rst
index 3616d7161dab..a5734bdd1cc7 100644
--- a/Documentation/filesystems/porting.rst
+++ b/Documentation/filesystems/porting.rst
@@ -1249,3 +1249,12 @@ Using try_lookup_noperm() will require linux/namei.h to be included.
 
 Calling conventions for ->d_automount() have changed; we should *not* grab
 an extra reference to new mount - it should be returned with refcount 1.
+
+---
+
+collect_mounts()/drop_collected_mounts()/iterate_mounts() are gone now.
+Replacement is collect_paths()/drop_collected_path(), with no special
+iterator needed.  Instead of a cloned mount tree, the new interface returns
+an array of struct path, one for each mount collect_mounts() would've
+created.  These struct path point to locations in the caller's namespace
+that would be roots of the cloned mounts.
diff --git a/fs/namespace.c b/fs/namespace.c
index e13d9ab4f564..14601ec4c2c5 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -2310,21 +2310,62 @@ struct mount *copy_tree(struct mount *src_root, struct dentry *dentry,
 	return dst_mnt;
 }
 
-/* Caller should check returned pointer for errors */
+static inline bool extend_array(struct path **res, struct path **to_free,
+				unsigned n, unsigned *count, unsigned new_count)
+{
+	struct path *p;
+
+	if (likely(n < *count))
+		return true;
+	p = kmalloc_array(new_count, sizeof(struct path), GFP_KERNEL);
+	if (p && *count)
+		memcpy(p, *res, *count * sizeof(struct path));
+	*count = new_count;
+	kfree(*to_free);
+	*to_free = *res = p;
+	return p;
+}
 
-struct vfsmount *collect_mounts(const struct path *path)
+struct path *collect_paths(const struct path *path,
+			      struct path *prealloc, unsigned count)
 {
-	struct mount *tree;
-	namespace_lock();
-	if (!check_mnt(real_mount(path->mnt)))
-		tree = ERR_PTR(-EINVAL);
-	else
-		tree = copy_tree(real_mount(path->mnt), path->dentry,
-				 CL_COPY_ALL | CL_PRIVATE);
-	namespace_unlock();
-	if (IS_ERR(tree))
-		return ERR_CAST(tree);
-	return &tree->mnt;
+	struct mount *root = real_mount(path->mnt);
+	struct mount *child;
+	struct path *res = prealloc, *to_free = NULL;
+	unsigned n = 0;
+
+	guard(rwsem_read)(&namespace_sem);
+
+	if (!check_mnt(root))
+		return ERR_PTR(-EINVAL);
+	if (!extend_array(&res, &to_free, 0, &count, 32))
+		return ERR_PTR(-ENOMEM);
+	res[n++] = *path;
+	list_for_each_entry(child, &root->mnt_mounts, mnt_child) {
+		if (!is_subdir(child->mnt_mountpoint, path->dentry))
+			continue;
+		for (struct mount *m = child; m; m = next_mnt(m, child)) {
+			if (!extend_array(&res, &to_free, n, &count, 2 * count))
+				return ERR_PTR(-ENOMEM);
+			res[n].mnt = &m->mnt;
+			res[n].dentry = m->mnt.mnt_root;
+			n++;
+		}
+	}
+	if (!extend_array(&res, &to_free, n, &count, count + 1))
+		return ERR_PTR(-ENOMEM);
+	memset(res + n, 0, (count - n) * sizeof(struct path));
+	for (struct path *p = res; p->mnt; p++)
+		path_get(p);
+	return res;
+}
+
+void drop_collected_paths(struct path *paths, struct path *prealloc)
+{
+	for (struct path *p = paths; p->mnt; p++)
+		path_put(p);
+	if (paths != prealloc)
+		kfree(paths);
 }
 
 static void free_mnt_ns(struct mnt_namespace *);
@@ -2401,15 +2442,6 @@ void dissolve_on_fput(struct vfsmount *mnt)
 	free_mnt_ns(ns);
 }
 
-void drop_collected_mounts(struct vfsmount *mnt)
-{
-	namespace_lock();
-	lock_mount_hash();
-	umount_tree(real_mount(mnt), 0);
-	unlock_mount_hash();
-	namespace_unlock();
-}
-
 static bool __has_locked_children(struct mount *mnt, struct dentry *dentry)
 {
 	struct mount *child;
@@ -2511,21 +2543,6 @@ struct vfsmount *clone_private_mount(const struct path *path)
 }
 EXPORT_SYMBOL_GPL(clone_private_mount);
 
-int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
-		   struct vfsmount *root)
-{
-	struct mount *mnt;
-	int res = f(root, arg);
-	if (res)
-		return res;
-	list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
-		res = f(&mnt->mnt, arg);
-		if (res)
-			return res;
-	}
-	return 0;
-}
-
 static void lock_mnt_tree(struct mount *mnt)
 {
 	struct mount *p;
@@ -6262,7 +6279,11 @@ void put_mnt_ns(struct mnt_namespace *ns)
 {
 	if (!refcount_dec_and_test(&ns->ns.count))
 		return;
-	drop_collected_mounts(&ns->root->mnt);
+	namespace_lock();
+	lock_mount_hash();
+	umount_tree(ns->root, 0);
+	unlock_mount_hash();
+	namespace_unlock();
 	free_mnt_ns(ns);
 }
 
diff --git a/fs/pnode.h b/fs/pnode.h
index 34b6247af01d..2d026fb98b18 100644
--- a/fs/pnode.h
+++ b/fs/pnode.h
@@ -28,8 +28,6 @@
 #define CL_SHARED_TO_SLAVE	0x20
 #define CL_COPY_MNT_NS_FILE	0x40
 
-#define CL_COPY_ALL		(CL_COPY_UNBINDABLE | CL_COPY_MNT_NS_FILE)
-
 static inline void set_mnt_shared(struct mount *mnt)
 {
 	mnt->mnt.mnt_flags &= ~MNT_SHARED_MASK;
diff --git a/include/linux/mount.h b/include/linux/mount.h
index 4880f434c021..1a508beba446 100644
--- a/include/linux/mount.h
+++ b/include/linux/mount.h
@@ -116,10 +116,8 @@ extern int may_umount_tree(struct vfsmount *);
 extern int may_umount(struct vfsmount *);
 int do_mount(const char *, const char __user *,
 		     const char *, unsigned long, void *);
-extern struct vfsmount *collect_mounts(const struct path *);
-extern void drop_collected_mounts(struct vfsmount *);
-extern int iterate_mounts(int (*)(struct vfsmount *, void *), void *,
-			  struct vfsmount *);
+extern struct path *collect_paths(const struct path *, struct path *, unsigned);
+extern void drop_collected_paths(struct path *, struct path *);
 extern void kern_unmount_array(struct vfsmount *mnt[], unsigned int num);
 
 extern int cifs_root_data(char **dev, char **opts);
diff --git a/kernel/audit_tree.c b/kernel/audit_tree.c
index f2f38903b2fe..68e042ae93c7 100644
--- a/kernel/audit_tree.c
+++ b/kernel/audit_tree.c
@@ -668,12 +668,6 @@ int audit_remove_tree_rule(struct audit_krule *rule)
 	return 0;
 }
 
-static int compare_root(struct vfsmount *mnt, void *arg)
-{
-	return inode_to_key(d_backing_inode(mnt->mnt_root)) ==
-	       (unsigned long)arg;
-}
-
 void audit_trim_trees(void)
 {
 	struct list_head cursor;
@@ -683,8 +677,9 @@ void audit_trim_trees(void)
 	while (cursor.next != &tree_list) {
 		struct audit_tree *tree;
 		struct path path;
-		struct vfsmount *root_mnt;
 		struct audit_node *node;
+		struct path *paths;
+		struct path array[16];
 		int err;
 
 		tree = container_of(cursor.next, struct audit_tree, list);
@@ -696,9 +691,9 @@ void audit_trim_trees(void)
 		if (err)
 			goto skip_it;
 
-		root_mnt = collect_mounts(&path);
+		paths = collect_paths(&path, array, 16);
 		path_put(&path);
-		if (IS_ERR(root_mnt))
+		if (IS_ERR(paths))
 			goto skip_it;
 
 		spin_lock(&hash_lock);
@@ -706,14 +701,17 @@ void audit_trim_trees(void)
 			struct audit_chunk *chunk = find_chunk(node);
 			/* this could be NULL if the watch is dying else where... */
 			node->index |= 1U<<31;
-			if (iterate_mounts(compare_root,
-					   (void *)(chunk->key),
-					   root_mnt))
-				node->index &= ~(1U<<31);
+			for (struct path *p = paths; p->dentry; p++) {
+				struct inode *inode = p->dentry->d_inode;
+				if (inode_to_key(inode) == chunk->key) {
+					node->index &= ~(1U<<31);
+					break;
+				}
+			}
 		}
 		spin_unlock(&hash_lock);
 		trim_marked(tree);
-		drop_collected_mounts(root_mnt);
+		drop_collected_paths(paths, array);
 skip_it:
 		put_tree(tree);
 		mutex_lock(&audit_filter_mutex);
@@ -742,9 +740,14 @@ void audit_put_tree(struct audit_tree *tree)
 	put_tree(tree);
 }
 
-static int tag_mount(struct vfsmount *mnt, void *arg)
+static int tag_mounts(struct path *paths, struct audit_tree *tree)
 {
-	return tag_chunk(d_backing_inode(mnt->mnt_root), arg);
+	for (struct path *p = paths; p->dentry; p++) {
+		int err = tag_chunk(p->dentry->d_inode, tree);
+		if (err)
+			return err;
+	}
+	return 0;
 }
 
 /*
@@ -801,7 +804,8 @@ int audit_add_tree_rule(struct audit_krule *rule)
 {
 	struct audit_tree *seed = rule->tree, *tree;
 	struct path path;
-	struct vfsmount *mnt;
+	struct path array[16];
+	struct path *paths;
 	int err;
 
 	rule->tree = NULL;
@@ -828,16 +832,16 @@ int audit_add_tree_rule(struct audit_krule *rule)
 	err = kern_path(tree->pathname, 0, &path);
 	if (err)
 		goto Err;
-	mnt = collect_mounts(&path);
+	paths = collect_paths(paths, array, 16);
 	path_put(&path);
-	if (IS_ERR(mnt)) {
-		err = PTR_ERR(mnt);
+	if (IS_ERR(paths)) {
+		err = PTR_ERR(paths);
 		goto Err;
 	}
 
 	get_tree(tree);
-	err = iterate_mounts(tag_mount, tree, mnt);
-	drop_collected_mounts(mnt);
+	err = tag_mounts(paths, tree);
+	drop_collected_paths(paths, array);
 
 	if (!err) {
 		struct audit_node *node;
@@ -872,20 +876,21 @@ int audit_tag_tree(char *old, char *new)
 	struct list_head cursor, barrier;
 	int failed = 0;
 	struct path path1, path2;
-	struct vfsmount *tagged;
+	struct path array[16];
+	struct path *paths;
 	int err;
 
 	err = kern_path(new, 0, &path2);
 	if (err)
 		return err;
-	tagged = collect_mounts(&path2);
+	paths = collect_paths(&path2, array, 16);
 	path_put(&path2);
-	if (IS_ERR(tagged))
-		return PTR_ERR(tagged);
+	if (IS_ERR(paths))
+		return PTR_ERR(paths);
 
 	err = kern_path(old, 0, &path1);
 	if (err) {
-		drop_collected_mounts(tagged);
+		drop_collected_paths(paths, array);
 		return err;
 	}
 
@@ -914,7 +919,7 @@ int audit_tag_tree(char *old, char *new)
 			continue;
 		}
 
-		failed = iterate_mounts(tag_mount, tree, tagged);
+		failed = tag_mounts(paths, tree);
 		if (failed) {
 			put_tree(tree);
 			mutex_lock(&audit_filter_mutex);
@@ -955,7 +960,7 @@ int audit_tag_tree(char *old, char *new)
 	list_del(&cursor);
 	mutex_unlock(&audit_filter_mutex);
 	path_put(&path1);
-	drop_collected_mounts(tagged);
+	drop_collected_paths(paths, array);
 	return failed;
 }
 
-- 
2.39.5

[PATCH v2 02/35] attach_recursive_mnt(): do not lock the covering tree when sliding something under it

[Al Viro]

If we are propagating across the userns boundary, we need to lock the
mounts added there.  However, in case when something has already
been mounted there and we end up sliding a new tree under that,
the stuff that had been there before should not get locked.

IOW, lock_mnt_tree() should be called before we reparent the
preexisting tree on top of what we are adding.

Fixes: 3bd045cc9c4b ("separate copying and locking mount tree on cross-userns copies")
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index 14601ec4c2c5..eed83254492f 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -2768,14 +2768,14 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 	hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
 		struct mount *q;
 		hlist_del_init(&child->mnt_hash);
-		q = __lookup_mnt(&child->mnt_parent->mnt,
-				 child->mnt_mountpoint);
-		if (q)
-			mnt_change_mountpoint(child, smp, q);
 		/* Notice when we are propagating across user namespaces */
 		if (child->mnt_parent->mnt_ns->user_ns != user_ns)
 			lock_mnt_tree(child);
 		child->mnt.mnt_flags &= ~MNT_LOCKED;
+		q = __lookup_mnt(&child->mnt_parent->mnt,
+				 child->mnt_mountpoint);
+		if (q)
+			mnt_change_mountpoint(child, smp, q);
 		commit_tree(child);
 	}
 	put_mountpoint(smp);
-- 
2.39.5

[PATCH v2 03/35] attach_mnt(): expand in attach_recursive_mnt(), then lose the flag argument

[Al Viro]

simpler that way - all but one caller pass false as 'beneath' argument,
and that one caller is actually happier with the call expanded - the
logics with choice of mountpoint is identical for 'moving' and 'attaching'
cases, and now that is no longer hidden.

Reviewed-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 37 ++++++++++++-------------------------
 1 file changed, 12 insertions(+), 25 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index eed83254492f..c6cac3603661 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -1116,16 +1116,10 @@ static void __attach_mnt(struct mount *mnt, struct mount *parent)
  * @parent:  the parent
  * @mnt:     the new mount
  * @mp:      the new mountpoint
- * @beneath: whether to mount @mnt beneath or on top of @parent
  *
- * If @beneath is false, mount @mnt at @mp on @parent. Then attach @mnt
+ * Mount @mnt at @mp on @parent. Then attach @mnt
  * to @parent's child mount list and to @mount_hashtable.
  *
- * If @beneath is true, remove @mnt from its current parent and
- * mountpoint and mount it on @mp on @parent, and mount @parent on the
- * old parent and old mountpoint of @mnt. Finally, attach @parent to
- * @mnt_hashtable and @parent->mnt_parent->mnt_mounts.
- *
  * Note, when __attach_mnt() is called @mnt->mnt_parent already points
  * to the correct parent.
  *
@@ -1133,18 +1127,9 @@ static void __attach_mnt(struct mount *mnt, struct mount *parent)
  *          to have been acquired in that order.
  */
 static void attach_mnt(struct mount *mnt, struct mount *parent,
-		       struct mountpoint *mp, bool beneath)
+		       struct mountpoint *mp)
 {
-	if (beneath)
-		mnt_set_mountpoint_beneath(mnt, parent, mp);
-	else
-		mnt_set_mountpoint(parent, mp, mnt);
-	/*
-	 * Note, @mnt->mnt_parent has to be used. If @mnt was mounted
-	 * beneath @parent then @mnt will need to be attached to
-	 * @parent's old parent, not @parent. IOW, @mnt->mnt_parent
-	 * isn't the same mount as @parent.
-	 */
+	mnt_set_mountpoint(parent, mp, mnt);
 	__attach_mnt(mnt, mnt->mnt_parent);
 }
 
@@ -1157,7 +1142,7 @@ void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct m
 	hlist_del_init(&mnt->mnt_mp_list);
 	hlist_del_init_rcu(&mnt->mnt_hash);
 
-	attach_mnt(mnt, parent, mp, false);
+	attach_mnt(mnt, parent, mp);
 
 	put_mountpoint(old_mp);
 	mnt_add_count(old_parent, -1);
@@ -2295,7 +2280,7 @@ struct mount *copy_tree(struct mount *src_root, struct dentry *dentry,
 				goto out;
 			lock_mount_hash();
 			list_add_tail(&dst_mnt->mnt_list, &res->mnt_list);
-			attach_mnt(dst_mnt, dst_parent, src_parent->mnt_mp, false);
+			attach_mnt(dst_mnt, dst_parent, src_parent->mnt_mp);
 			unlock_mount_hash();
 		}
 	}
@@ -2743,10 +2728,12 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 	}
 
 	if (moving) {
-		if (beneath)
-			dest_mp = smp;
 		unhash_mnt(source_mnt);
-		attach_mnt(source_mnt, top_mnt, dest_mp, beneath);
+		if (beneath)
+			mnt_set_mountpoint_beneath(source_mnt, top_mnt, smp);
+		else
+			mnt_set_mountpoint(top_mnt, dest_mp, source_mnt);
+		__attach_mnt(source_mnt, source_mnt->mnt_parent);
 		mnt_notify_add(source_mnt);
 		touch_mnt_namespace(source_mnt->mnt_ns);
 	} else {
@@ -4827,9 +4814,9 @@ SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
 		root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
 	}
 	/* mount old root on put_old */
-	attach_mnt(root_mnt, old_mnt, old_mp, false);
+	attach_mnt(root_mnt, old_mnt, old_mp);
 	/* mount new_root on / */
-	attach_mnt(new_mnt, root_parent, root_mp, false);
+	attach_mnt(new_mnt, root_parent, root_mp);
 	mnt_add_count(root_parent, -1);
 	touch_mnt_namespace(current->nsproxy->mnt_ns);
 	/* A moved mount should not expire automatically */
-- 
2.39.5

[PATCH v2 04/35] get rid of mnt_set_mountpoint_beneath()

[Al Viro]

mnt_set_mountpoint_beneath() consists of attaching new mount side-by-side
with the one we want to mount beneath (by mnt_set_mountpoint()), followed
by mnt_change_mountpoint() shifting the the top mount onto the new one
(by mnt_change_mountpoint()).

Both callers of mnt_set_mountpoint_beneath (both in attach_recursive_mnt())
have the same form - in 'beneath' case we call mnt_set_mountpoint_beneath(),
otherwise - mnt_set_mountpoint().

The thing is, expressing that as unconditional mnt_set_mountpoint(),
followed, in 'beneath' case, by mnt_change_mountpoint() is just as easy.
And these mnt_change_mountpoint() callers are similar to the ones we
do when it comes to attaching propagated copies, which will allow more
cleanups in the next commits.

Reviewed-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 37 ++++---------------------------------
 1 file changed, 4 insertions(+), 33 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index c6cac3603661..18ab7241749a 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -1076,33 +1076,6 @@ void mnt_set_mountpoint(struct mount *mnt,
 	hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list);
 }
 
-/**
- * mnt_set_mountpoint_beneath - mount a mount beneath another one
- *
- * @new_parent: the source mount
- * @top_mnt:    the mount beneath which @new_parent is mounted
- * @new_mp:     the new mountpoint of @top_mnt on @new_parent
- *
- * Remove @top_mnt from its current mountpoint @top_mnt->mnt_mp and
- * parent @top_mnt->mnt_parent and mount it on top of @new_parent at
- * @new_mp. And mount @new_parent on the old parent and old
- * mountpoint of @top_mnt.
- *
- * Context: This function expects namespace_lock() and lock_mount_hash()
- *          to have been acquired in that order.
- */
-static void mnt_set_mountpoint_beneath(struct mount *new_parent,
-				       struct mount *top_mnt,
-				       struct mountpoint *new_mp)
-{
-	struct mount *old_top_parent = top_mnt->mnt_parent;
-	struct mountpoint *old_top_mp = top_mnt->mnt_mp;
-
-	mnt_set_mountpoint(old_top_parent, old_top_mp, new_parent);
-	mnt_change_mountpoint(new_parent, new_mp, top_mnt);
-}
-
-
 static void __attach_mnt(struct mount *mnt, struct mount *parent)
 {
 	hlist_add_head_rcu(&mnt->mnt_hash,
@@ -2729,10 +2702,9 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 
 	if (moving) {
 		unhash_mnt(source_mnt);
+		mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
 		if (beneath)
-			mnt_set_mountpoint_beneath(source_mnt, top_mnt, smp);
-		else
-			mnt_set_mountpoint(top_mnt, dest_mp, source_mnt);
+			mnt_change_mountpoint(source_mnt, smp, top_mnt);
 		__attach_mnt(source_mnt, source_mnt->mnt_parent);
 		mnt_notify_add(source_mnt);
 		touch_mnt_namespace(source_mnt->mnt_ns);
@@ -2745,10 +2717,9 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 				move_from_ns(p, &head);
 			list_del_init(&head);
 		}
+		mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
 		if (beneath)
-			mnt_set_mountpoint_beneath(source_mnt, top_mnt, smp);
-		else
-			mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
+			mnt_change_mountpoint(source_mnt, smp, top_mnt);
 		commit_tree(source_mnt);
 	}
 
-- 
2.39.5

[PATCH v2 05/35] prevent mount hash conflicts

[Al Viro]

Currently it's still possible to run into a pathological situation when
two hashed mounts share both parent and mountpoint.  That does not work
well, for obvious reasons.

We are not far from getting rid of that; the only remaining gap is
attach_recursive_mnt() not being careful enough when sliding a tree
under existing mount (for propagated copies or in 'beneath' case for
the original one).

To deal with that cleanly we need to be able to find overmounts
(i.e. mounts on top of parent's root); we could do hash lookups or scan
the list of children but either would be costly.  Since one of the results
we get from that will be prevention of multiple parallel overmounts, let's
just bite the bullet and store a (non-counting) reference to overmount
in struct mount.

With that done, closing the hole in attach_recursive_mnt() becomes easy
- we just need to follow the chain of overmounts before we change the
mountpoint of the mount we are sliding things under.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/mount.h     |  1 +
 fs/namespace.c | 27 ++++++++++++++++++++++-----
 2 files changed, 23 insertions(+), 5 deletions(-)

diff --git a/fs/mount.h b/fs/mount.h
index ad7173037924..b8beafdd6d24 100644
--- a/fs/mount.h
+++ b/fs/mount.h
@@ -92,6 +92,7 @@ struct mount {
 	int mnt_expiry_mark;		/* true if marked for expiry */
 	struct hlist_head mnt_pins;
 	struct hlist_head mnt_stuck_children;
+	struct mount *overmount;	/* mounted on ->mnt_root */
 } __randomize_layout;
 
 #define MNT_NS_INTERNAL ERR_PTR(-EINVAL) /* distinct from any mnt_namespace */
diff --git a/fs/namespace.c b/fs/namespace.c
index 18ab7241749a..ea10e32ca80f 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -1043,6 +1043,9 @@ static void __touch_mnt_namespace(struct mnt_namespace *ns)
 static struct mountpoint *unhash_mnt(struct mount *mnt)
 {
 	struct mountpoint *mp;
+	struct mount *parent = mnt->mnt_parent;
+	if (unlikely(parent->overmount == mnt))
+		parent->overmount = NULL;
 	mnt->mnt_parent = mnt;
 	mnt->mnt_mountpoint = mnt->mnt.mnt_root;
 	list_del_init(&mnt->mnt_child);
@@ -1078,6 +1081,8 @@ void mnt_set_mountpoint(struct mount *mnt,
 
 static void __attach_mnt(struct mount *mnt, struct mount *parent)
 {
+	if (unlikely(mnt->mnt_mountpoint == parent->mnt.mnt_root))
+		parent->overmount = mnt;
 	hlist_add_head_rcu(&mnt->mnt_hash,
 			   m_hash(&parent->mnt, mnt->mnt_mountpoint));
 	list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
@@ -2660,7 +2665,9 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 	HLIST_HEAD(tree_list);
 	struct mnt_namespace *ns = top_mnt->mnt_ns;
 	struct mountpoint *smp;
+	struct mountpoint *secondary = NULL;
 	struct mount *child, *dest_mnt, *p;
+	struct mount *top;
 	struct hlist_node *n;
 	int err = 0;
 	bool moving = flags & MNT_TREE_MOVE, beneath = flags & MNT_TREE_BENEATH;
@@ -2669,9 +2676,15 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 	 * Preallocate a mountpoint in case the new mounts need to be
 	 * mounted beneath mounts on the same mountpoint.
 	 */
-	smp = get_mountpoint(source_mnt->mnt.mnt_root);
+	for (top = source_mnt; unlikely(top->overmount); top = top->overmount) {
+		if (!secondary && is_mnt_ns_file(top->mnt.mnt_root))
+			secondary = top->mnt_mp;
+	}
+	smp = get_mountpoint(top->mnt.mnt_root);
 	if (IS_ERR(smp))
 		return PTR_ERR(smp);
+	if (!secondary)
+		secondary = smp;
 
 	/* Is there space to add these mounts to the mount namespace? */
 	if (!moving) {
@@ -2704,7 +2717,7 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 		unhash_mnt(source_mnt);
 		mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
 		if (beneath)
-			mnt_change_mountpoint(source_mnt, smp, top_mnt);
+			mnt_change_mountpoint(top, smp, top_mnt);
 		__attach_mnt(source_mnt, source_mnt->mnt_parent);
 		mnt_notify_add(source_mnt);
 		touch_mnt_namespace(source_mnt->mnt_ns);
@@ -2719,7 +2732,7 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 		}
 		mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
 		if (beneath)
-			mnt_change_mountpoint(source_mnt, smp, top_mnt);
+			mnt_change_mountpoint(top, smp, top_mnt);
 		commit_tree(source_mnt);
 	}
 
@@ -2732,8 +2745,12 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 		child->mnt.mnt_flags &= ~MNT_LOCKED;
 		q = __lookup_mnt(&child->mnt_parent->mnt,
 				 child->mnt_mountpoint);
-		if (q)
-			mnt_change_mountpoint(child, smp, q);
+		if (q) {
+			struct mount *r = child;
+			while (unlikely(r->overmount))
+				r = r->overmount;
+			mnt_change_mountpoint(r, secondary, q);
+		}
 		commit_tree(child);
 	}
 	put_mountpoint(smp);
-- 
2.39.5

[PATCH v2 06/35] copy_tree(): don't set ->mnt_mountpoint on the root of copy

[Al Viro]

It never made any sense - neither when copy_tree() had been introduced
(2.4.11-pre5), nor at any point afterwards.  Mountpoint is meaningless
without parent mount and the root of copied tree has no parent until we get
around to attaching it somewhere.  At that time we'll have mountpoint set;
before that we have no idea which dentry will be used as mountpoint.
IOW, copy_tree() should just leave the default value.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 1 -
 1 file changed, 1 deletion(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index ea10e32ca80f..e5b4ea106b94 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -2222,7 +2222,6 @@ struct mount *copy_tree(struct mount *src_root, struct dentry *dentry,
 		return dst_mnt;
 
 	src_parent = src_root;
-	dst_mnt->mnt_mountpoint = src_root->mnt_mountpoint;
 
 	list_for_each_entry(src_root_child, &src_root->mnt_mounts, mnt_child) {
 		if (!is_subdir(src_root_child->mnt_mountpoint, dentry))
-- 
2.39.5

[PATCH v2 07/35] constify mnt_has_parent()

[Al Viro]

Reviewed-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/mount.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/fs/mount.h b/fs/mount.h
index b8beafdd6d24..c4d417cd7953 100644
--- a/fs/mount.h
+++ b/fs/mount.h
@@ -102,7 +102,7 @@ static inline struct mount *real_mount(struct vfsmount *mnt)
 	return container_of(mnt, struct mount, mnt);
 }
 
-static inline int mnt_has_parent(struct mount *mnt)
+static inline int mnt_has_parent(const struct mount *mnt)
 {
 	return mnt != mnt->mnt_parent;
 }
-- 
2.39.5

[PATCH v2 08/35] pnode: lift peers() into pnode.h

[Al Viro]

it's going to be useful both in pnode.c and namespace.c

Reviewed-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/pnode.c | 5 -----
 fs/pnode.h | 5 +++++
 2 files changed, 5 insertions(+), 5 deletions(-)

diff --git a/fs/pnode.c b/fs/pnode.c
index ffd429b760d5..aa187144e389 100644
--- a/fs/pnode.c
+++ b/fs/pnode.c
@@ -216,11 +216,6 @@ static struct mount *next_group(struct mount *m, struct mount *origin)
 static struct mount *last_dest, *first_source, *last_source, *dest_master;
 static struct hlist_head *list;
 
-static inline bool peers(const struct mount *m1, const struct mount *m2)
-{
-	return m1->mnt_group_id == m2->mnt_group_id && m1->mnt_group_id;
-}
-
 static int propagate_one(struct mount *m, struct mountpoint *dest_mp)
 {
 	struct mount *child;
diff --git a/fs/pnode.h b/fs/pnode.h
index 2d026fb98b18..93fa9311bd07 100644
--- a/fs/pnode.h
+++ b/fs/pnode.h
@@ -34,6 +34,11 @@ static inline void set_mnt_shared(struct mount *mnt)
 	mnt->mnt.mnt_flags |= MNT_SHARED;
 }
 
+static inline bool peers(const struct mount *m1, const struct mount *m2)
+{
+	return m1->mnt_group_id == m2->mnt_group_id && m1->mnt_group_id;
+}
+
 void change_mnt_propagation(struct mount *, int);
 int propagate_mnt(struct mount *, struct mountpoint *, struct mount *,
 		struct hlist_head *);
-- 
2.39.5

[PATCH v2 09/35] new predicate: mount_is_ancestor()

[Al Viro]

mount_is_ancestor(p1, p2) returns true iff there is a possibly
empty ancestry chain from p1 to p2.

Convert the open-coded checks.  Unlike those open-coded variants
it does not depend upon p1 not being root...

Reviewed-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 21 +++++++++++++++------
 1 file changed, 15 insertions(+), 6 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index e5b4ea106b94..7454f9efaa27 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -3483,6 +3483,17 @@ static inline bool path_overmounted(const struct path *path)
 	return unlikely(!no_child);
 }
 
+/*
+ * Check if there is a possibly empty chain of descent from p1 to p2.
+ * Locks: namespace_sem (shared) or mount_lock (read_seqlock_excl).
+ */
+static bool mount_is_ancestor(const struct mount *p1, const struct mount *p2)
+{
+	while (p2 != p1 && mnt_has_parent(p2))
+		p2 = p2->mnt_parent;
+	return p2 == p1;
+}
+
 /**
  * can_move_mount_beneath - check that we can mount beneath the top mount
  * @from: mount to mount beneath
@@ -3534,9 +3545,8 @@ static int can_move_mount_beneath(const struct path *from,
 	if (parent_mnt_to == current->nsproxy->mnt_ns->root)
 		return -EINVAL;
 
-	for (struct mount *p = mnt_from; mnt_has_parent(p); p = p->mnt_parent)
-		if (p == mnt_to)
-			return -EINVAL;
+	if (mount_is_ancestor(mnt_to, mnt_from))
+		return -EINVAL;
 
 	/*
 	 * If the parent mount propagates to the child mount this would
@@ -3705,9 +3715,8 @@ static int do_move_mount(struct path *old_path,
 	err = -ELOOP;
 	if (!check_for_nsfs_mounts(old))
 		goto out;
-	for (; mnt_has_parent(p); p = p->mnt_parent)
-		if (p == old)
-			goto out;
+	if (mount_is_ancestor(old, p))
+		goto out;
 
 	err = attach_recursive_mnt(old, real_mount(new_path->mnt), mp, flags);
 	if (err)
-- 
2.39.5

[PATCH v2 10/35] constify is_local_mountpoint()

[Al Viro]

Reviewed-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/mount.h     | 4 ++--
 fs/namespace.c | 2 +-
 2 files changed, 3 insertions(+), 3 deletions(-)

diff --git a/fs/mount.h b/fs/mount.h
index c4d417cd7953..f10776003643 100644
--- a/fs/mount.h
+++ b/fs/mount.h
@@ -147,8 +147,8 @@ struct proc_mounts {
 
 extern const struct seq_operations mounts_op;
 
-extern bool __is_local_mountpoint(struct dentry *dentry);
-static inline bool is_local_mountpoint(struct dentry *dentry)
+extern bool __is_local_mountpoint(const struct dentry *dentry);
+static inline bool is_local_mountpoint(const struct dentry *dentry)
 {
 	if (!d_mountpoint(dentry))
 		return false;
diff --git a/fs/namespace.c b/fs/namespace.c
index 7454f9efaa27..1d68bfc3dc35 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -894,7 +894,7 @@ struct vfsmount *lookup_mnt(const struct path *path)
  * namespace not just a mount that happens to have some specified
  * parent mount.
  */
-bool __is_local_mountpoint(struct dentry *dentry)
+bool __is_local_mountpoint(const struct dentry *dentry)
 {
 	struct mnt_namespace *ns = current->nsproxy->mnt_ns;
 	struct mount *mnt, *n;
-- 
2.39.5

[PATCH v2 11/35] new predicate: anon_ns_root(mount)

[Al Viro]

checks if mount is the root of an anonymouns namespace.
Switch open-coded equivalents to using it.

For mounts that belong to anon namespace !mnt_has_parent(mount)
is the same as mount == ns->root, and intent is more obvious in
the latter form.

Reviewed-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/mount.h     |  7 +++++++
 fs/namespace.c | 28 +++-------------------------
 2 files changed, 10 insertions(+), 25 deletions(-)

diff --git a/fs/mount.h b/fs/mount.h
index f10776003643..f20e6ed845fe 100644
--- a/fs/mount.h
+++ b/fs/mount.h
@@ -161,6 +161,13 @@ static inline bool is_anon_ns(struct mnt_namespace *ns)
 	return ns->seq == 0;
 }
 
+static inline bool anon_ns_root(const struct mount *m)
+{
+	struct mnt_namespace *ns = READ_ONCE(m->mnt_ns);
+
+	return !IS_ERR_OR_NULL(ns) && is_anon_ns(ns) && m == ns->root;
+}
+
 static inline bool mnt_ns_attached(const struct mount *mnt)
 {
 	return !RB_EMPTY_NODE(&mnt->mnt_node);
diff --git a/fs/namespace.c b/fs/namespace.c
index 1d68bfc3dc35..82791f636442 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -2480,9 +2480,7 @@ struct vfsmount *clone_private_mount(const struct path *path)
 	 * loops get created.
 	 */
 	if (!check_mnt(old_mnt)) {
-		if (!is_mounted(&old_mnt->mnt) ||
-			!is_anon_ns(old_mnt->mnt_ns) ||
-			mnt_has_parent(old_mnt))
+		if (!anon_ns_root(old_mnt))
 			return ERR_PTR(-EINVAL);
 
 		if (!check_for_nsfs_mounts(old_mnt))
@@ -3649,9 +3647,6 @@ static int do_move_mount(struct path *old_path,
 	ns = old->mnt_ns;
 
 	err = -EINVAL;
-	/* The thing moved must be mounted... */
-	if (!is_mounted(&old->mnt))
-		goto out;
 
 	if (check_mnt(old)) {
 		/* if the source is in our namespace... */
@@ -3664,10 +3659,8 @@ static int do_move_mount(struct path *old_path,
 	} else {
 		/*
 		 * otherwise the source must be the root of some anon namespace.
-		 * AV: check for mount being root of an anon namespace is worth
-		 * an inlined predicate...
 		 */
-		if (!is_anon_ns(ns) || mnt_has_parent(old))
+		if (!anon_ns_root(old))
 			goto out;
 		/*
 		 * Bail out early if the target is within the same namespace -
@@ -5028,22 +5021,7 @@ static int do_mount_setattr(struct path *path, struct mount_kattr *kattr)
 	err = -EINVAL;
 	lock_mount_hash();
 
-	/* Ensure that this isn't anything purely vfs internal. */
-	if (!is_mounted(&mnt->mnt))
-		goto out;
-
-	/*
-	 * If this is an attached mount make sure it's located in the callers
-	 * mount namespace. If it's not don't let the caller interact with it.
-	 *
-	 * If this mount doesn't have a parent it's most often simply a
-	 * detached mount with an anonymous mount namespace. IOW, something
-	 * that's simply not attached yet. But there are apparently also users
-	 * that do change mount properties on the rootfs itself. That obviously
-	 * neither has a parent nor is it a detached mount so we cannot
-	 * unconditionally check for detached mounts.
-	 */
-	if ((mnt_has_parent(mnt) || !is_anon_ns(mnt->mnt_ns)) && !check_mnt(mnt))
+	if (!anon_ns_root(mnt) && !check_mnt(mnt))
 		goto out;
 
 	/*
-- 
2.39.5

[PATCH v2 12/35] dissolve_on_fput(): use anon_ns_root()

[Al Viro]

that's the condition we are actually trying to check there...

Reviewed-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 62 +++++++++++---------------------------------------
 1 file changed, 13 insertions(+), 49 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index 82791f636442..fb15bd04333a 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -2333,67 +2333,31 @@ void drop_collected_paths(struct path *paths, struct path *prealloc)
 static void free_mnt_ns(struct mnt_namespace *);
 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *, bool);
 
-static inline bool must_dissolve(struct mnt_namespace *mnt_ns)
-{
-	/*
-        * This mount belonged to an anonymous mount namespace
-        * but was moved to a non-anonymous mount namespace and
-        * then unmounted.
-        */
-	if (unlikely(!mnt_ns))
-		return false;
-
-	/*
-        * This mount belongs to a non-anonymous mount namespace
-        * and we know that such a mount can never transition to
-        * an anonymous mount namespace again.
-        */
-	if (!is_anon_ns(mnt_ns)) {
-		/*
-		 * A detached mount either belongs to an anonymous mount
-		 * namespace or a non-anonymous mount namespace. It
-		 * should never belong to something purely internal.
-		 */
-		VFS_WARN_ON_ONCE(mnt_ns == MNT_NS_INTERNAL);
-		return false;
-	}
-
-	return true;
-}
-
 void dissolve_on_fput(struct vfsmount *mnt)
 {
 	struct mnt_namespace *ns;
 	struct mount *m = real_mount(mnt);
 
+	/*
+	 * m used to be the root of anon namespace; if it still is one,
+	 * we need to dissolve the mount tree and free that namespace.
+	 * Let's try to avoid taking namespace_sem if we can determine
+	 * that there's nothing to do without it - rcu_read_lock() is
+	 * enough to make anon_ns_root() memory-safe and once m has
+	 * left its namespace, it's no longer our concern, since it will
+	 * never become a root of anon ns again.
+	 */
+
 	scoped_guard(rcu) {
-		if (!must_dissolve(READ_ONCE(m->mnt_ns)))
+		if (!anon_ns_root(m))
 			return;
 	}
 
 	scoped_guard(namespace_lock, &namespace_sem) {
-		ns = m->mnt_ns;
-		if (!must_dissolve(ns))
-			return;
-
-		/*
-		 * After must_dissolve() we know that this is a detached
-		 * mount in an anonymous mount namespace.
-		 *
-		 * Now when mnt_has_parent() reports that this mount
-		 * tree has a parent, we know that this anonymous mount
-		 * tree has been moved to another anonymous mount
-		 * namespace.
-		 *
-		 * So when closing this file we cannot unmount the mount
-		 * tree. This will be done when the file referring to
-		 * the root of the anonymous mount namespace will be
-		 * closed (It could already be closed but it would sync
-		 * on @namespace_sem and wait for us to finish.).
-		 */
-		if (mnt_has_parent(m))
+		if (!anon_ns_root(m))
 			return;
 
+		ns = m->mnt_ns;
 		lock_mount_hash();
 		umount_tree(m, UMOUNT_CONNECTED);
 		unlock_mount_hash();
-- 
2.39.5

[PATCH v2 13/35] __attach_mnt(): lose the second argument

[Al Viro]

It's always ->mnt_parent of the first one.  What the function does is
making a mount (with already set parent and mountpoint) visible - in
mount hash and in the parent's list of children.

IOW, it takes the existing rootwards linkage and sets the matching
crownwards linkage.

Renamed to make_visible(), while we are at it.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 11 ++++++-----
 1 file changed, 6 insertions(+), 5 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index fb15bd04333a..9ac9a82d2aee 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -1079,8 +1079,9 @@ void mnt_set_mountpoint(struct mount *mnt,
 	hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list);
 }
 
-static void __attach_mnt(struct mount *mnt, struct mount *parent)
+static void make_visible(struct mount *mnt)
 {
+	struct mount *parent = mnt->mnt_parent;
 	if (unlikely(mnt->mnt_mountpoint == parent->mnt.mnt_root))
 		parent->overmount = mnt;
 	hlist_add_head_rcu(&mnt->mnt_hash,
@@ -1098,7 +1099,7 @@ static void __attach_mnt(struct mount *mnt, struct mount *parent)
  * Mount @mnt at @mp on @parent. Then attach @mnt
  * to @parent's child mount list and to @mount_hashtable.
  *
- * Note, when __attach_mnt() is called @mnt->mnt_parent already points
+ * Note, when make_visible() is called @mnt->mnt_parent already points
  * to the correct parent.
  *
  * Context: This function expects namespace_lock() and lock_mount_hash()
@@ -1108,7 +1109,7 @@ static void attach_mnt(struct mount *mnt, struct mount *parent,
 		       struct mountpoint *mp)
 {
 	mnt_set_mountpoint(parent, mp, mnt);
-	__attach_mnt(mnt, mnt->mnt_parent);
+	make_visible(mnt);
 }
 
 void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt)
@@ -1182,7 +1183,7 @@ static void commit_tree(struct mount *mnt)
 	n->nr_mounts += n->pending_mounts;
 	n->pending_mounts = 0;
 
-	__attach_mnt(mnt, parent);
+	make_visible(mnt);
 	touch_mnt_namespace(n);
 }
 
@@ -2679,7 +2680,7 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 		mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
 		if (beneath)
 			mnt_change_mountpoint(top, smp, top_mnt);
-		__attach_mnt(source_mnt, source_mnt->mnt_parent);
+		make_visible(source_mnt);
 		mnt_notify_add(source_mnt);
 		touch_mnt_namespace(source_mnt->mnt_ns);
 	} else {
-- 
2.39.5

[PATCH v2 14/35] don't set MNT_LOCKED on parentless mounts

[Al Viro]

Originally MNT_LOCKED meant only one thing - "don't let this mount to
be peeled off its parent, we don't want to have its mountpoint exposed".
Accordingly, it had only been set on mounts that *do* have a parent.
Later it got overloaded with another use - setting it on the absolute
root had given free protection against umount(2) of absolute root
(was possible to trigger, oopsed).  Not a bad trick, but it ended
up costing more than it bought us.  Unfortunately, the cost included
both hard-to-reason-about logics and a subtle race between
mount -o remount,ro and mount --[r]bind - lockless &= ~MNT_LOCKED in
the end of __do_loopback() could race with sb_prepare_remount_readonly()
setting and clearing MNT_HOLD_WRITE (under mount_lock, as it should
be).  The race wouldn't be much of a problem (there are other ways to
deal with it), but the subtlety is.

Turns out that nobody except umount(2) had ever made use of having
MNT_LOCKED set on absolute root.  So let's give up on that trick,
clever as it had been, add an explicit check in do_umount() and
return to using MNT_LOCKED only for mounts that have a parent.

It means that
	* clone_mnt() no longer copies MNT_LOCKED
	* copy_tree() sets it on submounts if their counterparts had
been marked such, and does that right next to attach_mnt() in there,
in the same mount_lock scope.
	* __do_loopback() no longer needs to strip MNT_LOCKED off the
root of subtree it's about to return; no store, no race.
	* init_mount_tree() doesn't bother setting MNT_LOCKED on absolute
root.
	* lock_mnt_tree() does not set MNT_LOCKED on the subtree's root;
accordingly, its caller (loop in attach_recursive_mnt()) does not need to
bother stripping that MNT_LOCKED on root.  Note that lock_mnt_tree() setting
MNT_LOCKED on submounts happens in the same mount_lock scope as __attach_mnt()
(from commit_tree()) that makes them reachable.

Reviewed-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 32 +++++++++++++++-----------------
 1 file changed, 15 insertions(+), 17 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index 9ac9a82d2aee..bfc85d50e8cb 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -1313,7 +1313,7 @@ static struct mount *clone_mnt(struct mount *old, struct dentry *root,
 	}
 
 	mnt->mnt.mnt_flags = old->mnt.mnt_flags;
-	mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL);
+	mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL|MNT_LOCKED);
 
 	atomic_inc(&sb->s_active);
 	mnt->mnt.mnt_idmap = mnt_idmap_get(mnt_idmap(&old->mnt));
@@ -1988,6 +1988,9 @@ static int do_umount(struct mount *mnt, int flags)
 	if (mnt->mnt.mnt_flags & MNT_LOCKED)
 		goto out;
 
+	if (!mnt_has_parent(mnt)) /* not the absolute root */
+		goto out;
+
 	event++;
 	if (flags & MNT_DETACH) {
 		if (mnt_ns_attached(mnt) || !list_empty(&mnt->mnt_list))
@@ -2257,6 +2260,8 @@ struct mount *copy_tree(struct mount *src_root, struct dentry *dentry,
 			if (IS_ERR(dst_mnt))
 				goto out;
 			lock_mount_hash();
+			if (src_mnt->mnt.mnt_flags & MNT_LOCKED)
+				dst_mnt->mnt.mnt_flags |= MNT_LOCKED;
 			list_add_tail(&dst_mnt->mnt_list, &res->mnt_list);
 			attach_mnt(dst_mnt, dst_parent, src_parent->mnt_mp);
 			unlock_mount_hash();
@@ -2489,7 +2494,7 @@ static void lock_mnt_tree(struct mount *mnt)
 		if (flags & MNT_NOEXEC)
 			flags |= MNT_LOCK_NOEXEC;
 		/* Don't allow unprivileged users to reveal what is under a mount */
-		if (list_empty(&p->mnt_expire))
+		if (list_empty(&p->mnt_expire) && p != mnt)
 			flags |= MNT_LOCKED;
 		p->mnt.mnt_flags = flags;
 	}
@@ -2704,7 +2709,6 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 		/* Notice when we are propagating across user namespaces */
 		if (child->mnt_parent->mnt_ns->user_ns != user_ns)
 			lock_mnt_tree(child);
-		child->mnt.mnt_flags &= ~MNT_LOCKED;
 		q = __lookup_mnt(&child->mnt_parent->mnt,
 				 child->mnt_mountpoint);
 		if (q) {
@@ -2985,26 +2989,21 @@ static inline bool may_copy_tree(struct path *path)
 
 static struct mount *__do_loopback(struct path *old_path, int recurse)
 {
-	struct mount *mnt = ERR_PTR(-EINVAL), *old = real_mount(old_path->mnt);
+	struct mount *old = real_mount(old_path->mnt);
 
 	if (IS_MNT_UNBINDABLE(old))
-		return mnt;
+		return ERR_PTR(-EINVAL);
 
 	if (!may_copy_tree(old_path))
-		return mnt;
+		return ERR_PTR(-EINVAL);
 
 	if (!recurse && __has_locked_children(old, old_path->dentry))
-		return mnt;
+		return ERR_PTR(-EINVAL);
 
 	if (recurse)
-		mnt = copy_tree(old, old_path->dentry, CL_COPY_MNT_NS_FILE);
+		return copy_tree(old, old_path->dentry, CL_COPY_MNT_NS_FILE);
 	else
-		mnt = clone_mnt(old, old_path->dentry, 0);
-
-	if (!IS_ERR(mnt))
-		mnt->mnt.mnt_flags &= ~MNT_LOCKED;
-
-	return mnt;
+		return clone_mnt(old, old_path->dentry, 0);
 }
 
 /*
@@ -4749,11 +4748,11 @@ SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
 	if (!path_mounted(&root))
 		goto out4; /* not a mountpoint */
 	if (!mnt_has_parent(root_mnt))
-		goto out4; /* not attached */
+		goto out4; /* absolute root */
 	if (!path_mounted(&new))
 		goto out4; /* not a mountpoint */
 	if (!mnt_has_parent(new_mnt))
-		goto out4; /* not attached */
+		goto out4; /* absolute root */
 	/* make sure we can reach put_old from new_root */
 	if (!is_path_reachable(old_mnt, old.dentry, &new))
 		goto out4;
@@ -6158,7 +6157,6 @@ static void __init init_mount_tree(void)
 
 	root.mnt = mnt;
 	root.dentry = mnt->mnt_root;
-	mnt->mnt_flags |= MNT_LOCKED;
 
 	set_fs_pwd(current->fs, &root);
 	set_fs_root(current->fs, &root);
-- 
2.39.5

[PATCH v2 15/35] clone_mnt(): simplify the propagation-related logics

[Al Viro]

The underlying rules are simple:
	* MNT_SHARED should be set iff ->mnt_group_id of new mount ends up
non-zero.
	* mounts should be on the same ->mnt_share cyclic list iff they have
the same non-zero ->mnt_group_id value.
	* CL_PRIVATE is mutually exclusive with MNT_SHARED, MNT_SLAVE,
MNT_SHARED_TO_SLAVE and MNT_EXPIRE; the whole point of that thing is to
get a clone of old mount that would *not* be on any namespace-related
lists.

The above allows to make the logics more straightforward; what's more,
it makes the proof that invariants are maintained much simpler.
The variant in mainline is safe (aside of a very narrow race with
unsafe modification of mnt_flags right after we had the mount exposed
in superblock's ->s_mounts; theoretically it can race with ro remount
of the original, but it's not easy to hit), but proof of its correctness
is really unpleasant.

Reviewed-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 26 +++++++++++++-------------
 1 file changed, 13 insertions(+), 13 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index bfc85d50e8cb..fd453848c2c7 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -1301,6 +1301,9 @@ static struct mount *clone_mnt(struct mount *old, struct dentry *root,
 	if (!mnt)
 		return ERR_PTR(-ENOMEM);
 
+	mnt->mnt.mnt_flags = READ_ONCE(old->mnt.mnt_flags) &
+			     ~MNT_INTERNAL_FLAGS;
+
 	if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
 		mnt->mnt_group_id = 0; /* not a peer of original */
 	else
@@ -1312,8 +1315,8 @@ static struct mount *clone_mnt(struct mount *old, struct dentry *root,
 			goto out_free;
 	}
 
-	mnt->mnt.mnt_flags = old->mnt.mnt_flags;
-	mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL|MNT_LOCKED);
+	if (mnt->mnt_group_id)
+		set_mnt_shared(mnt);
 
 	atomic_inc(&sb->s_active);
 	mnt->mnt.mnt_idmap = mnt_idmap_get(mnt_idmap(&old->mnt));
@@ -1326,22 +1329,20 @@ static struct mount *clone_mnt(struct mount *old, struct dentry *root,
 	list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
 	unlock_mount_hash();
 
+	if (flag & CL_PRIVATE)	// we are done with it
+		return mnt;
+
+	if (peers(mnt, old))
+		list_add(&mnt->mnt_share, &old->mnt_share);
+
 	if ((flag & CL_SLAVE) ||
 	    ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
 		list_add(&mnt->mnt_slave, &old->mnt_slave_list);
 		mnt->mnt_master = old;
-		CLEAR_MNT_SHARED(mnt);
-	} else if (!(flag & CL_PRIVATE)) {
-		if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
-			list_add(&mnt->mnt_share, &old->mnt_share);
-		if (IS_MNT_SLAVE(old))
-			list_add(&mnt->mnt_slave, &old->mnt_slave);
+	} else if (IS_MNT_SLAVE(old)) {
+		list_add(&mnt->mnt_slave, &old->mnt_slave);
 		mnt->mnt_master = old->mnt_master;
-	} else {
-		CLEAR_MNT_SHARED(mnt);
 	}
-	if (flag & CL_MAKE_SHARED)
-		set_mnt_shared(mnt);
 
 	/* stick the duplicate mount on the same expiry list
 	 * as the original if that was on one */
@@ -1349,7 +1350,6 @@ static struct mount *clone_mnt(struct mount *old, struct dentry *root,
 		if (!list_empty(&old->mnt_expire))
 			list_add(&mnt->mnt_expire, &old->mnt_expire);
 	}
-
 	return mnt;
 
  out_free:
-- 
2.39.5

[PATCH v2 16/35] do_umount(): simplify the "is it still mounted" checks

[Al Viro]

Calls of do_umount() are always preceded by can_umount(), where we'd
done a racy check for mount belonging to our namespace; if it wasn't,
can_unmount() would've failed with -EINVAL and we wouldn't have
reached do_umount() at all.

That check needs to be redone once we have acquired namespace_sem
and in do_umount() we do that.  However, that's done in a very odd
way; we check that mount is still in rbtree of _some_ namespace or
its mnt_list is not empty.  It is equivalent to check_mnt(mnt) -
we know that earlier mnt was mounted in our namespace; if it has
stayed there, it's going to remain in rbtree of our namespace.
OTOH, if it ever had been removed from out namespace, it would be
removed from rbtree and it never would've re-added to a namespace
afterwards.  As for ->mnt_list, for something that had been mounted
in a namespace we'll never observe non-empty ->mnt_list while holding
namespace_sem - it does temporarily become non-empty during
umount_tree(), but that doesn't outlast the call of umount_tree(),
let alone dropping namespace_sem.

Things get much easier to follow if we replace that with (equivalent)
check_mnt(mnt) there.  What's more, currently we treat a failure of
that test as "quietly do nothing"; we might as well pretend that we'd
lost the race and fail on that the same way can_umount() would have.

Reviewed-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 11 ++++++-----
 1 file changed, 6 insertions(+), 5 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index fd453848c2c7..a7bf07d88da4 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -1983,8 +1983,11 @@ static int do_umount(struct mount *mnt, int flags)
 	namespace_lock();
 	lock_mount_hash();
 
-	/* Recheck MNT_LOCKED with the locks held */
+	/* Repeat the earlier racy checks, now that we are holding the locks */
 	retval = -EINVAL;
+	if (!check_mnt(mnt))
+		goto out;
+
 	if (mnt->mnt.mnt_flags & MNT_LOCKED)
 		goto out;
 
@@ -1993,16 +1996,14 @@ static int do_umount(struct mount *mnt, int flags)
 
 	event++;
 	if (flags & MNT_DETACH) {
-		if (mnt_ns_attached(mnt) || !list_empty(&mnt->mnt_list))
-			umount_tree(mnt, UMOUNT_PROPAGATE);
+		umount_tree(mnt, UMOUNT_PROPAGATE);
 		retval = 0;
 	} else {
 		smp_mb(); // paired with __legitimize_mnt()
 		shrink_submounts(mnt);
 		retval = -EBUSY;
 		if (!propagate_mount_busy(mnt, 2)) {
-			if (mnt_ns_attached(mnt) || !list_empty(&mnt->mnt_list))
-				umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
+			umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
 			retval = 0;
 		}
 	}
-- 
2.39.5

[PATCH v2 17/35] sanitize handling of long-term internal mounts

[Al Viro]

Original rationale for those had been the reduced cost of mntput()
for the stuff that is mounted somewhere.  Mount refcount increments and
decrements are frequent; what's worse, they tend to concentrate on the
same instances and cacheline pingpong is quite noticable.

As the result, mount refcounts are per-cpu; that allows a very cheap
increment.  Plain decrement would be just as easy, but decrement-and-test
is anything but (we need to add the components up, with exclusion against
possible increment-from-zero, etc.).

Fortunately, there is a very common case where we can tell that decrement
won't be the final one - if the thing we are dropping is currently
mounted somewhere.  We have an RCU delay between the removal from mount
tree and dropping the reference that used to pin it there, so we can
just take rcu_read_lock() and check if the victim is mounted somewhere.
If it is, we can go ahead and decrement without and further checks -
the reference we are dropping is not the last one.  If it isn't, we
get all the fun with locking, carefully adding up components, etc.,
but the majority of refcount decrements end up taking the fast path.

There is a major exception, though - pipes and sockets.  Those live
on the internal filesystems that are not going to be mounted anywhere.
They are not going to be _un_mounted, of course, so having to take the
slow path every time a pipe or socket gets closed is really obnoxious.
Solution had been to mark them as long-lived ones - essentially faking
"they are mounted somewhere" indicator.

With minor modification that works even for ones that do eventually get
dropped - all it takes is making sure we have an RCU delay between
clearing the "mounted somewhere" indicator and dropping the reference.

There are some additional twists (if you want to drop a dozen of such
internal mounts, you'd be better off with clearing the indicator on
all of them, doing an RCU delay once, then dropping the references),
but in the basic form it had been
	* use kern_mount() if you want your internal mount to be
a long-term one.
	* use kern_unmount() to undo that.

Unfortunately, the things did rot a bit during the mount API reshuffling.
In several cases we have lost the "fake the indicator" part; kern_unmount()
on the unmount side remained (it doesn't warn if you use it on a mount
without the indicator), but all benefits regaring mntput() cost had been
lost.

To get rid of that bitrot, let's add a new helper that would work
with fs_context-based API: fc_mount_longterm().  It's a counterpart
of fc_mount() that does, on success, mark its result as long-term.
It must be paired with kern_unmount() or equivalents.

Converted:
	1) mqueue (it used to use kern_mount_data() and the umount side
is still as it used to be)
	2) hugetlbfs (used to use kern_mount_data(), internal mount is
never unmounted in this one)
	3) i915 gemfs (used to be kern_mount() + manual remount to set
options, still uses kern_unmount() on umount side)
	4) v3d gemfs (copied from i915)

Reviewed-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 drivers/gpu/drm/i915/gem/i915_gemfs.c | 21 ++++++++++++++++++---
 drivers/gpu/drm/v3d/v3d_gemfs.c       | 21 ++++++++++++++++++---
 fs/hugetlbfs/inode.c                  |  2 +-
 fs/namespace.c                        |  9 +++++++++
 include/linux/mount.h                 |  1 +
 ipc/mqueue.c                          |  2 +-
 6 files changed, 48 insertions(+), 8 deletions(-)

diff --git a/drivers/gpu/drm/i915/gem/i915_gemfs.c b/drivers/gpu/drm/i915/gem/i915_gemfs.c
index 65d84a93c525..a09e2eb47175 100644
--- a/drivers/gpu/drm/i915/gem/i915_gemfs.c
+++ b/drivers/gpu/drm/i915/gem/i915_gemfs.c
@@ -5,16 +5,23 @@
 
 #include <linux/fs.h>
 #include <linux/mount.h>
+#include <linux/fs_context.h>
 
 #include "i915_drv.h"
 #include "i915_gemfs.h"
 #include "i915_utils.h"
 
+static int add_param(struct fs_context *fc, const char *key, const char *val)
+{
+	return vfs_parse_fs_string(fc, key, val, strlen(val));
+}
+
 void i915_gemfs_init(struct drm_i915_private *i915)
 {
-	char huge_opt[] = "huge=within_size"; /* r/w */
 	struct file_system_type *type;
+	struct fs_context *fc;
 	struct vfsmount *gemfs;
+	int ret;
 
 	/*
 	 * By creating our own shmemfs mountpoint, we can pass in
@@ -38,8 +45,16 @@ void i915_gemfs_init(struct drm_i915_private *i915)
 	if (!type)
 		goto err;
 
-	gemfs = vfs_kern_mount(type, SB_KERNMOUNT, type->name, huge_opt);
-	if (IS_ERR(gemfs))
+	fc = fs_context_for_mount(type, SB_KERNMOUNT);
+	if (IS_ERR(fc))
+		goto err;
+	ret = add_param(fc, "source", "tmpfs");
+	if (!ret)
+		ret = add_param(fc, "huge", "within_size");
+	if (!ret)
+		gemfs = fc_mount_longterm(fc);
+	put_fs_context(fc);
+	if (ret)
 		goto err;
 
 	i915->mm.gemfs = gemfs;
diff --git a/drivers/gpu/drm/v3d/v3d_gemfs.c b/drivers/gpu/drm/v3d/v3d_gemfs.c
index 4c5e18590a5c..8ec6ed82b3d9 100644
--- a/drivers/gpu/drm/v3d/v3d_gemfs.c
+++ b/drivers/gpu/drm/v3d/v3d_gemfs.c
@@ -3,14 +3,21 @@
 
 #include <linux/fs.h>
 #include <linux/mount.h>
+#include <linux/fs_context.h>
 
 #include "v3d_drv.h"
 
+static int add_param(struct fs_context *fc, const char *key, const char *val)
+{
+	return vfs_parse_fs_string(fc, key, val, strlen(val));
+}
+
 void v3d_gemfs_init(struct v3d_dev *v3d)
 {
-	char huge_opt[] = "huge=within_size";
 	struct file_system_type *type;
+	struct fs_context *fc;
 	struct vfsmount *gemfs;
+	int ret;
 
 	/*
 	 * By creating our own shmemfs mountpoint, we can pass in
@@ -28,8 +35,16 @@ void v3d_gemfs_init(struct v3d_dev *v3d)
 	if (!type)
 		goto err;
 
-	gemfs = vfs_kern_mount(type, SB_KERNMOUNT, type->name, huge_opt);
-	if (IS_ERR(gemfs))
+	fc = fs_context_for_mount(type, SB_KERNMOUNT);
+	if (IS_ERR(fc))
+		goto err;
+	ret = add_param(fc, "source", "tmpfs");
+	if (!ret)
+		ret = add_param(fc, "huge", "within_size");
+	if (!ret)
+		gemfs = fc_mount_longterm(fc);
+	put_fs_context(fc);
+	if (ret)
 		goto err;
 
 	v3d->gemfs = gemfs;
diff --git a/fs/hugetlbfs/inode.c b/fs/hugetlbfs/inode.c
index e4de5425838d..4e0397775167 100644
--- a/fs/hugetlbfs/inode.c
+++ b/fs/hugetlbfs/inode.c
@@ -1587,7 +1587,7 @@ static struct vfsmount *__init mount_one_hugetlbfs(struct hstate *h)
 	} else {
 		struct hugetlbfs_fs_context *ctx = fc->fs_private;
 		ctx->hstate = h;
-		mnt = fc_mount(fc);
+		mnt = fc_mount_longterm(fc);
 		put_fs_context(fc);
 	}
 	if (IS_ERR(mnt))
diff --git a/fs/namespace.c b/fs/namespace.c
index a7bf07d88da4..96bc3b804974 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -1260,6 +1260,15 @@ struct vfsmount *fc_mount(struct fs_context *fc)
 }
 EXPORT_SYMBOL(fc_mount);
 
+struct vfsmount *fc_mount_longterm(struct fs_context *fc)
+{
+	struct vfsmount *mnt = fc_mount(fc);
+	if (!IS_ERR(mnt))
+		real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
+	return mnt;
+}
+EXPORT_SYMBOL(fc_mount_longterm);
+
 struct vfsmount *vfs_kern_mount(struct file_system_type *type,
 				int flags, const char *name,
 				void *data)
diff --git a/include/linux/mount.h b/include/linux/mount.h
index 1a508beba446..c145820fcbbf 100644
--- a/include/linux/mount.h
+++ b/include/linux/mount.h
@@ -98,6 +98,7 @@ int mnt_get_write_access(struct vfsmount *mnt);
 void mnt_put_write_access(struct vfsmount *mnt);
 
 extern struct vfsmount *fc_mount(struct fs_context *fc);
+extern struct vfsmount *fc_mount_longterm(struct fs_context *fc);
 extern struct vfsmount *vfs_create_mount(struct fs_context *fc);
 extern struct vfsmount *vfs_kern_mount(struct file_system_type *type,
 				      int flags, const char *name,
diff --git a/ipc/mqueue.c b/ipc/mqueue.c
index 82ed2d3c9846..de7432efbf4a 100644
--- a/ipc/mqueue.c
+++ b/ipc/mqueue.c
@@ -482,7 +482,7 @@ static struct vfsmount *mq_create_mount(struct ipc_namespace *ns)
 	put_user_ns(fc->user_ns);
 	fc->user_ns = get_user_ns(ctx->ipc_ns->user_ns);
 
-	mnt = fc_mount(fc);
+	mnt = fc_mount_longterm(fc);
 	put_fs_context(fc);
 	return mnt;
 }
-- 
2.39.5

Re: [PATCH v2 17/35] sanitize handling of long-term internal mounts

[Linus Torvalds]

So from a quick scan through the patches, they all looked really nice
(but note the "quick scan" - I didn't apply this, much less test
anything).

I did react to this one, though - not very complicated, but this patch
struck me as kind of ugly compared to most of the others.

On Sun, 22 Jun 2025 at 21:54, Al Viro <viro@zeniv.linux.org.uk> wrote:
>
> -       gemfs = vfs_kern_mount(type, SB_KERNMOUNT, type->name, huge_opt);
> -       if (IS_ERR(gemfs))
> +       fc = fs_context_for_mount(type, SB_KERNMOUNT);
> +       if (IS_ERR(fc))
> +               goto err;
> +       ret = add_param(fc, "source", "tmpfs");
> +       if (!ret)
> +               ret = add_param(fc, "huge", "within_size");
> +       if (!ret)
> +               gemfs = fc_mount_longterm(fc);
> +       put_fs_context(fc);
> +       if (ret)
>                 goto err;

So this "fs_context_for_mount() + N * add_param()" pattern ends up
showing up twice, with that 'add_param()' helper done twice too.

And that's ignoring the _existing_ users of "fs_context_for_mount() +
N * vfs_parse_fs_string()", which are really the same except they
don't wrap it with that 'add_param()' helper.

I'm not objecting to the patch, and I don't really even have a
solution: many of the existing cases actually do need the more
complicated vfs_parse_fs_string() interface because they don't want
that simple 'strlen()' for size.

I just feel that at a minimum you shouldn't implement add_param()
twice, because some other users *would* want to do that.

So I wish you had made that a real helper - which would obviously then
also force a naming change ("fs_context_add_param()".

Or maybe even go further and some helper to doi that
"fs_context_for_mount()" _with_ a list of param's to be added?

I do think that could be done later (separately), but wanted to just
mention this because I reacted to this patch.

              Linus

Re: [PATCH v2 17/35] sanitize handling of long-term internal mounts

[Al Viro]

On Mon, Jun 23, 2025 at 09:18:33AM -0700, Linus Torvalds wrote:

> I'm not objecting to the patch, and I don't really even have a
> solution: many of the existing cases actually do need the more
> complicated vfs_parse_fs_string() interface because they don't want
> that simple 'strlen()' for size.

I don't know...  7 callers with explicit strlen():
drivers/gpu/drm/i915/gem/i915_gemfs.c:16:       return vfs_parse_fs_string(fc, key, val, strlen(val));
drivers/gpu/drm/v3d/v3d_gemfs.c:12:     return vfs_parse_fs_string(fc, key, val, strlen(val));
fs/namespace.c:1284:            ret = vfs_parse_fs_string(fc, "source",
fs/namespace.c:3799:            err = vfs_parse_fs_string(fc, "subtype",
fs/namespace.c:3802:            err = vfs_parse_fs_string(fc, "source", name, strlen(name));
fs/nfs/fs_context.c:1230:                       ret = vfs_parse_fs_string(fc, "context",
kernel/trace/trace.c:10280:     ret = vfs_parse_fs_string(fc, "source",

3 callers that could as well use strlen(), except that some of them need
to cope with NULL (using 0 for length in that case):
fs/fs_context.c:230:                    ret = vfs_parse_fs_string(fc, key, value, v_len);
fs/nfs/namespace.c:293:         ret = vfs_parse_fs_string(fc, "source", p, buffer + 4096 - p);
fs/smb/client/fs_context.c:785:         ret = vfs_parse_fs_string(fc, key, value, len);

1 caller that really does need len < strlen(s):
fs/afs/mntpt.c:140:                     ret = vfs_parse_fs_string(fc, "source", content, size - 1);

> I just feel that at a minimum you shouldn't implement add_param()
> twice, because some other users *would* want to do that.
> 
> So I wish you had made that a real helper - which would obviously then
> also force a naming change ("fs_context_add_param()".

May the bikeshedding commence ;-)

> Or maybe even go further and some helper to doi that
> "fs_context_for_mount()" _with_ a list of param's to be added?

Vararg, presumably?

> I do think that could be done later (separately), but wanted to just
> mention this because I reacted to this patch.

Re: [PATCH v2 17/35] sanitize handling of long-term internal mounts

[Linus Torvalds]

On Mon, 23 Jun 2025 at 10:03, Al Viro <viro@zeniv.linux.org.uk> wrote:
>
> I don't know...  7 callers with explicit strlen():

Yeah. Most of them clearly just want the string length.

But there were clearly a couple that really didn't have a
NUL-termination thing. And one - that nfs namespace thing - that do
have a string, but get the string length differently.

> > Or maybe even go further and some helper to doi that
> > "fs_context_for_mount()" _with_ a list of param's to be added?
>
> Vararg, presumably?

Or just pass in a descriptor struct / array that can be NULL?

But this really is not a huge deal. I was just looking at your series,
and most things cleaned things up, and this one just grated on me a
bit.

           Linus

[RFC] vfs_parse_fs_string() calling conventions change (was Re: [PATCH v2 17/35] sanitize handling of long-term internal mounts)

[Al Viro]

On Mon, Jun 23, 2025 at 06:03:15PM +0100, Al Viro wrote:
> I don't know...  7 callers with explicit strlen():
> drivers/gpu/drm/i915/gem/i915_gemfs.c:16:       return vfs_parse_fs_string(fc, key, val, strlen(val));
> drivers/gpu/drm/v3d/v3d_gemfs.c:12:     return vfs_parse_fs_string(fc, key, val, strlen(val));
> fs/namespace.c:1284:            ret = vfs_parse_fs_string(fc, "source",
> fs/namespace.c:3799:            err = vfs_parse_fs_string(fc, "subtype",
> fs/namespace.c:3802:            err = vfs_parse_fs_string(fc, "source", name, strlen(name));
> fs/nfs/fs_context.c:1230:                       ret = vfs_parse_fs_string(fc, "context",
> kernel/trace/trace.c:10280:     ret = vfs_parse_fs_string(fc, "source",
> 
> 3 callers that could as well use strlen(), except that some of them need
> to cope with NULL (using 0 for length in that case):
> fs/fs_context.c:230:                    ret = vfs_parse_fs_string(fc, key, value, v_len);
> fs/nfs/namespace.c:293:         ret = vfs_parse_fs_string(fc, "source", p, buffer + 4096 - p);
> fs/smb/client/fs_context.c:785:         ret = vfs_parse_fs_string(fc, key, value, len);
> 
> 1 caller that really does need len < strlen(s):
> fs/afs/mntpt.c:140:                     ret = vfs_parse_fs_string(fc, "source", content, size - 1);
> 
> > I just feel that at a minimum you shouldn't implement add_param()
> > twice, because some other users *would* want to do that.
> > 
> > So I wish you had made that a real helper - which would obviously then
> > also force a naming change ("fs_context_add_param()".

Frankly, looking at that stuff...  I wonder if we should add
vfs_parse_fs_qstr() for "comes with length" variant and lose the length
argument of vfs_parse_fs_string().

Yes, it's a flagday change.  Compiler will immediately catch any place
that needs to be converted, and D/f/porting.rst part should be clear
enough.

How about something like the following (completely untested), on top of -rc3?
Objections, anyone?

[PATCH] change the calling conventions for vfs_parse_fs_string()

Absolute majority of callers are passing the 4th argument equal to
strlen() of the 3rd one.

Drop the v_size argument, add vfs_parse_fs_qstr() for the cases that
want independent length.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
diff --git a/Documentation/filesystems/mount_api.rst b/Documentation/filesystems/mount_api.rst
index e149b89118c8..c99ab1f7fea4 100644
--- a/Documentation/filesystems/mount_api.rst
+++ b/Documentation/filesystems/mount_api.rst
@@ -504,10 +504,18 @@ returned.
      clear the pointer, but then becomes responsible for disposing of the
      object.
 
+   * ::
+
+       int vfs_parse_fs_qstr(struct fs_context *fc, const char *key,
+			       const struct qstr *value);
+
+     A wrapper around vfs_parse_fs_param() that copies the value string it is
+     passed.
+
    * ::
 
        int vfs_parse_fs_string(struct fs_context *fc, const char *key,
-			       const char *value, size_t v_size);
+			       const char *value);
 
      A wrapper around vfs_parse_fs_param() that copies the value string it is
      passed.
diff --git a/Documentation/filesystems/porting.rst b/Documentation/filesystems/porting.rst
index 3616d7161dab..cbbabdf130a6 100644
--- a/Documentation/filesystems/porting.rst
+++ b/Documentation/filesystems/porting.rst
@@ -1249,3 +1249,13 @@ Using try_lookup_noperm() will require linux/namei.h to be included.
 
 Calling conventions for ->d_automount() have changed; we should *not* grab
 an extra reference to new mount - it should be returned with refcount 1.
+
+---
+
+**mandatory**
+
+Calling conventions for vfs_parse_fs_string() have changed; it does *not*
+take length anymore (value ? strlen(value) : 0 is used).  If you want
+a different length, use
+	vfs_parse_fs_qstr(fc, key, &QSTR_INIT(value, len))
+instead.
diff --git a/fs/afs/mntpt.c b/fs/afs/mntpt.c
index 9434a5399f2b..33d1bb2b75bf 100644
--- a/fs/afs/mntpt.c
+++ b/fs/afs/mntpt.c
@@ -137,7 +137,8 @@ static int afs_mntpt_set_params(struct fs_context *fc, struct dentry *mntpt)
 
 		ret = -EINVAL;
 		if (content[size - 1] == '.')
-			ret = vfs_parse_fs_string(fc, "source", content, size - 1);
+			ret = vfs_parse_fs_qstr(fc, "source",
+						&QSTR_INIT(content, size - 1));
 		do_delayed_call(&cleanup);
 		if (ret < 0)
 			return ret;
diff --git a/fs/fs_context.c b/fs/fs_context.c
index 666e61753aed..93b7ebf8d927 100644
--- a/fs/fs_context.c
+++ b/fs/fs_context.c
@@ -161,25 +161,24 @@ int vfs_parse_fs_param(struct fs_context *fc, struct fs_parameter *param)
 EXPORT_SYMBOL(vfs_parse_fs_param);
 
 /**
- * vfs_parse_fs_string - Convenience function to just parse a string.
+ * vfs_parse_fs_qstr - Convenience function to just parse a string.
  * @fc: Filesystem context.
  * @key: Parameter name.
  * @value: Default value.
- * @v_size: Maximum number of bytes in the value.
  */
-int vfs_parse_fs_string(struct fs_context *fc, const char *key,
-			const char *value, size_t v_size)
+int vfs_parse_fs_qstr(struct fs_context *fc, const char *key,
+			const struct qstr *value)
 {
 	int ret;
 
 	struct fs_parameter param = {
 		.key	= key,
 		.type	= fs_value_is_flag,
-		.size	= v_size,
+		.size	= value ? value->len : 0,
 	};
 
 	if (value) {
-		param.string = kmemdup_nul(value, v_size, GFP_KERNEL);
+		param.string = kmemdup_nul(value->name, value->len, GFP_KERNEL);
 		if (!param.string)
 			return -ENOMEM;
 		param.type = fs_value_is_string;
@@ -189,7 +188,7 @@ int vfs_parse_fs_string(struct fs_context *fc, const char *key,
 	kfree(param.string);
 	return ret;
 }
-EXPORT_SYMBOL(vfs_parse_fs_string);
+EXPORT_SYMBOL(vfs_parse_fs_qstr);
 
 /**
  * vfs_parse_monolithic_sep - Parse key[=val][,key[=val]]* mount data
@@ -218,16 +217,14 @@ int vfs_parse_monolithic_sep(struct fs_context *fc, void *data,
 
 	while ((key = sep(&options)) != NULL) {
 		if (*key) {
-			size_t v_len = 0;
 			char *value = strchr(key, '=');
 
 			if (value) {
 				if (unlikely(value == key))
 					continue;
 				*value++ = 0;
-				v_len = strlen(value);
 			}
-			ret = vfs_parse_fs_string(fc, key, value, v_len);
+			ret = vfs_parse_fs_string(fc, key, value);
 			if (ret < 0)
 				break;
 		}
diff --git a/fs/namespace.c b/fs/namespace.c
index e13d9ab4f564..c5cc8406d24c 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -1312,8 +1312,7 @@ struct vfsmount *vfs_kern_mount(struct file_system_type *type,
 		return ERR_CAST(fc);
 
 	if (name)
-		ret = vfs_parse_fs_string(fc, "source",
-					  name, strlen(name));
+		ret = vfs_parse_fs_string(fc, "source", name);
 	if (!ret)
 		ret = parse_monolithic_mount_data(fc, data);
 	if (!ret)
@@ -3873,10 +3872,9 @@ static int do_new_mount(struct path *path, const char *fstype, int sb_flags,
 	fc->oldapi = true;
 
 	if (subtype)
-		err = vfs_parse_fs_string(fc, "subtype",
-					  subtype, strlen(subtype));
+		err = vfs_parse_fs_string(fc, "subtype", subtype);
 	if (!err && name)
-		err = vfs_parse_fs_string(fc, "source", name, strlen(name));
+		err = vfs_parse_fs_string(fc, "source", name);
 	if (!err)
 		err = parse_monolithic_mount_data(fc, data);
 	if (!err && !mount_capable(fc))
diff --git a/fs/nfs/fs_context.c b/fs/nfs/fs_context.c
index 13f71ca8c974..de1ecb10831b 100644
--- a/fs/nfs/fs_context.c
+++ b/fs/nfs/fs_context.c
@@ -1227,8 +1227,7 @@ static int nfs23_parse_monolithic(struct fs_context *fc,
 			int ret;
 
 			data->context[NFS_MAX_CONTEXT_LEN] = '\0';
-			ret = vfs_parse_fs_string(fc, "context",
-						  data->context, strlen(data->context));
+			ret = vfs_parse_fs_string(fc, "context", data->context);
 			if (ret < 0)
 				return ret;
 #else
diff --git a/fs/nfs/namespace.c b/fs/nfs/namespace.c
index 7f1ec9c67ff2..2fe97ab57575 100644
--- a/fs/nfs/namespace.c
+++ b/fs/nfs/namespace.c
@@ -290,7 +290,8 @@ int nfs_do_submount(struct fs_context *fc)
 		nfs_errorf(fc, "NFS: Couldn't determine submount pathname");
 		ret = PTR_ERR(p);
 	} else {
-		ret = vfs_parse_fs_string(fc, "source", p, buffer + 4096 - p);
+		ret = vfs_parse_fs_qstr(fc, "source",
+					&QSTR_INIT(p, buffer + 4096 - p));
 		if (!ret)
 			ret = vfs_get_tree(fc);
 	}
diff --git a/fs/smb/client/fs_context.c b/fs/smb/client/fs_context.c
index a634a34d4086..e750906c9c49 100644
--- a/fs/smb/client/fs_context.c
+++ b/fs/smb/client/fs_context.c
@@ -773,16 +773,14 @@ static int smb3_fs_context_parse_monolithic(struct fs_context *fc,
 		}
 
 
-		len = 0;
 		value = strchr(key, '=');
 		if (value) {
 			if (value == key)
 				continue;
 			*value++ = 0;
-			len = strlen(value);
 		}
 
-		ret = vfs_parse_fs_string(fc, key, value, len);
+		ret = vfs_parse_fs_string(fc, key, value);
 		if (ret < 0)
 			break;
 	}
diff --git a/include/linux/fs_context.h b/include/linux/fs_context.h
index a19e4bd32e4d..a735ed23dd9f 100644
--- a/include/linux/fs_context.h
+++ b/include/linux/fs_context.h
@@ -134,8 +134,13 @@ extern struct fs_context *fs_context_for_submount(struct file_system_type *fs_ty
 
 extern struct fs_context *vfs_dup_fs_context(struct fs_context *fc);
 extern int vfs_parse_fs_param(struct fs_context *fc, struct fs_parameter *param);
-extern int vfs_parse_fs_string(struct fs_context *fc, const char *key,
-			       const char *value, size_t v_size);
+extern int vfs_parse_fs_qstr(struct fs_context *fc, const char *key,
+				const struct qstr *value);
+static inline int vfs_parse_fs_string(struct fs_context *fc, const char *key,
+			       const char *value)
+{
+	return vfs_parse_fs_qstr(fc, key, value ? &QSTR(value) : NULL);
+}
 int vfs_parse_monolithic_sep(struct fs_context *fc, void *data,
 			     char *(*sep)(char **));
 extern int generic_parse_monolithic(struct fs_context *fc, void *data);
diff --git a/kernel/trace/trace.c b/kernel/trace/trace.c
index 95ae7c4e5835..77ff2608118b 100644
--- a/kernel/trace/trace.c
+++ b/kernel/trace/trace.c
@@ -10277,8 +10277,7 @@ static struct vfsmount *trace_automount(struct dentry *mntpt, void *ingore)
 	if (IS_ERR(fc))
 		return ERR_CAST(fc);
 
-	ret = vfs_parse_fs_string(fc, "source",
-				  "tracefs", strlen("tracefs"));
+	ret = vfs_parse_fs_string(fc, "source", "tracefs");
 	if (!ret)
 		mnt = fc_mount(fc);
 	else

Re: [RFC] vfs_parse_fs_string() calling conventions change (was Re: [PATCH v2 17/35] sanitize handling of long-term internal mounts)

[Al Viro]

On Sat, Jun 28, 2025 at 08:58:49AM +0100, Al Viro wrote:
> Yes, it's a flagday change.  Compiler will immediately catch any place
> that needs to be converted, and D/f/porting.rst part should be clear
> enough.
> 
> How about something like the following (completely untested), on top of -rc3?
> Objections, anyone?
 
After fixing a braino (s/QSTR_INIT/QSTR_LEN/) it even builds and seems to work...

[PATCH] change the calling conventions for vfs_parse_fs_string()

Absolute majority of callers are passing the 4th argument equal to
strlen() of the 3rd one.

Drop the v_size argument, add vfs_parse_fs_qstr() for the cases that
want independent length.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
diff --git a/Documentation/filesystems/mount_api.rst b/Documentation/filesystems/mount_api.rst
index e149b89118c8..c99ab1f7fea4 100644
--- a/Documentation/filesystems/mount_api.rst
+++ b/Documentation/filesystems/mount_api.rst
@@ -504,10 +504,18 @@ returned.
      clear the pointer, but then becomes responsible for disposing of the
      object.
 
+   * ::
+
+       int vfs_parse_fs_qstr(struct fs_context *fc, const char *key,
+			       const struct qstr *value);
+
+     A wrapper around vfs_parse_fs_param() that copies the value string it is
+     passed.
+
    * ::
 
        int vfs_parse_fs_string(struct fs_context *fc, const char *key,
-			       const char *value, size_t v_size);
+			       const char *value);
 
      A wrapper around vfs_parse_fs_param() that copies the value string it is
      passed.
diff --git a/Documentation/filesystems/porting.rst b/Documentation/filesystems/porting.rst
index 3616d7161dab..6ed66ed90eb6 100644
--- a/Documentation/filesystems/porting.rst
+++ b/Documentation/filesystems/porting.rst
@@ -1249,3 +1249,13 @@ Using try_lookup_noperm() will require linux/namei.h to be included.
 
 Calling conventions for ->d_automount() have changed; we should *not* grab
 an extra reference to new mount - it should be returned with refcount 1.
+
+---
+
+**mandatory**
+
+Calling conventions for vfs_parse_fs_string() have changed; it does *not*
+take length anymore (value ? strlen(value) : 0 is used).  If you want
+a different length, use
+	vfs_parse_fs_qstr(fc, key, &QSTR_LEN(value, len))
+instead.
diff --git a/fs/afs/mntpt.c b/fs/afs/mntpt.c
index 9434a5399f2b..1ad048e6e164 100644
--- a/fs/afs/mntpt.c
+++ b/fs/afs/mntpt.c
@@ -137,7 +137,8 @@ static int afs_mntpt_set_params(struct fs_context *fc, struct dentry *mntpt)
 
 		ret = -EINVAL;
 		if (content[size - 1] == '.')
-			ret = vfs_parse_fs_string(fc, "source", content, size - 1);
+			ret = vfs_parse_fs_qstr(fc, "source",
+						&QSTR_LEN(content, size - 1));
 		do_delayed_call(&cleanup);
 		if (ret < 0)
 			return ret;
diff --git a/fs/fs_context.c b/fs/fs_context.c
index 666e61753aed..93b7ebf8d927 100644
--- a/fs/fs_context.c
+++ b/fs/fs_context.c
@@ -161,25 +161,24 @@ int vfs_parse_fs_param(struct fs_context *fc, struct fs_parameter *param)
 EXPORT_SYMBOL(vfs_parse_fs_param);
 
 /**
- * vfs_parse_fs_string - Convenience function to just parse a string.
+ * vfs_parse_fs_qstr - Convenience function to just parse a string.
  * @fc: Filesystem context.
  * @key: Parameter name.
  * @value: Default value.
- * @v_size: Maximum number of bytes in the value.
  */
-int vfs_parse_fs_string(struct fs_context *fc, const char *key,
-			const char *value, size_t v_size)
+int vfs_parse_fs_qstr(struct fs_context *fc, const char *key,
+			const struct qstr *value)
 {
 	int ret;
 
 	struct fs_parameter param = {
 		.key	= key,
 		.type	= fs_value_is_flag,
-		.size	= v_size,
+		.size	= value ? value->len : 0,
 	};
 
 	if (value) {
-		param.string = kmemdup_nul(value, v_size, GFP_KERNEL);
+		param.string = kmemdup_nul(value->name, value->len, GFP_KERNEL);
 		if (!param.string)
 			return -ENOMEM;
 		param.type = fs_value_is_string;
@@ -189,7 +188,7 @@ int vfs_parse_fs_string(struct fs_context *fc, const char *key,
 	kfree(param.string);
 	return ret;
 }
-EXPORT_SYMBOL(vfs_parse_fs_string);
+EXPORT_SYMBOL(vfs_parse_fs_qstr);
 
 /**
  * vfs_parse_monolithic_sep - Parse key[=val][,key[=val]]* mount data
@@ -218,16 +217,14 @@ int vfs_parse_monolithic_sep(struct fs_context *fc, void *data,
 
 	while ((key = sep(&options)) != NULL) {
 		if (*key) {
-			size_t v_len = 0;
 			char *value = strchr(key, '=');
 
 			if (value) {
 				if (unlikely(value == key))
 					continue;
 				*value++ = 0;
-				v_len = strlen(value);
 			}
-			ret = vfs_parse_fs_string(fc, key, value, v_len);
+			ret = vfs_parse_fs_string(fc, key, value);
 			if (ret < 0)
 				break;
 		}
diff --git a/fs/namespace.c b/fs/namespace.c
index e13d9ab4f564..c5cc8406d24c 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -1312,8 +1312,7 @@ struct vfsmount *vfs_kern_mount(struct file_system_type *type,
 		return ERR_CAST(fc);
 
 	if (name)
-		ret = vfs_parse_fs_string(fc, "source",
-					  name, strlen(name));
+		ret = vfs_parse_fs_string(fc, "source", name);
 	if (!ret)
 		ret = parse_monolithic_mount_data(fc, data);
 	if (!ret)
@@ -3873,10 +3872,9 @@ static int do_new_mount(struct path *path, const char *fstype, int sb_flags,
 	fc->oldapi = true;
 
 	if (subtype)
-		err = vfs_parse_fs_string(fc, "subtype",
-					  subtype, strlen(subtype));
+		err = vfs_parse_fs_string(fc, "subtype", subtype);
 	if (!err && name)
-		err = vfs_parse_fs_string(fc, "source", name, strlen(name));
+		err = vfs_parse_fs_string(fc, "source", name);
 	if (!err)
 		err = parse_monolithic_mount_data(fc, data);
 	if (!err && !mount_capable(fc))
diff --git a/fs/nfs/fs_context.c b/fs/nfs/fs_context.c
index 13f71ca8c974..de1ecb10831b 100644
--- a/fs/nfs/fs_context.c
+++ b/fs/nfs/fs_context.c
@@ -1227,8 +1227,7 @@ static int nfs23_parse_monolithic(struct fs_context *fc,
 			int ret;
 
 			data->context[NFS_MAX_CONTEXT_LEN] = '\0';
-			ret = vfs_parse_fs_string(fc, "context",
-						  data->context, strlen(data->context));
+			ret = vfs_parse_fs_string(fc, "context", data->context);
 			if (ret < 0)
 				return ret;
 #else
diff --git a/fs/nfs/namespace.c b/fs/nfs/namespace.c
index 7f1ec9c67ff2..5735c0448b4c 100644
--- a/fs/nfs/namespace.c
+++ b/fs/nfs/namespace.c
@@ -290,7 +290,8 @@ int nfs_do_submount(struct fs_context *fc)
 		nfs_errorf(fc, "NFS: Couldn't determine submount pathname");
 		ret = PTR_ERR(p);
 	} else {
-		ret = vfs_parse_fs_string(fc, "source", p, buffer + 4096 - p);
+		ret = vfs_parse_fs_qstr(fc, "source",
+					&QSTR_LEN(p, buffer + 4096 - p));
 		if (!ret)
 			ret = vfs_get_tree(fc);
 	}
diff --git a/fs/smb/client/fs_context.c b/fs/smb/client/fs_context.c
index a634a34d4086..e750906c9c49 100644
--- a/fs/smb/client/fs_context.c
+++ b/fs/smb/client/fs_context.c
@@ -773,16 +773,14 @@ static int smb3_fs_context_parse_monolithic(struct fs_context *fc,
 		}
 
 
-		len = 0;
 		value = strchr(key, '=');
 		if (value) {
 			if (value == key)
 				continue;
 			*value++ = 0;
-			len = strlen(value);
 		}
 
-		ret = vfs_parse_fs_string(fc, key, value, len);
+		ret = vfs_parse_fs_string(fc, key, value);
 		if (ret < 0)
 			break;
 	}
diff --git a/include/linux/fs_context.h b/include/linux/fs_context.h
index a19e4bd32e4d..a735ed23dd9f 100644
--- a/include/linux/fs_context.h
+++ b/include/linux/fs_context.h
@@ -134,8 +134,13 @@ extern struct fs_context *fs_context_for_submount(struct file_system_type *fs_ty
 
 extern struct fs_context *vfs_dup_fs_context(struct fs_context *fc);
 extern int vfs_parse_fs_param(struct fs_context *fc, struct fs_parameter *param);
-extern int vfs_parse_fs_string(struct fs_context *fc, const char *key,
-			       const char *value, size_t v_size);
+extern int vfs_parse_fs_qstr(struct fs_context *fc, const char *key,
+				const struct qstr *value);
+static inline int vfs_parse_fs_string(struct fs_context *fc, const char *key,
+			       const char *value)
+{
+	return vfs_parse_fs_qstr(fc, key, value ? &QSTR(value) : NULL);
+}
 int vfs_parse_monolithic_sep(struct fs_context *fc, void *data,
 			     char *(*sep)(char **));
 extern int generic_parse_monolithic(struct fs_context *fc, void *data);
diff --git a/kernel/trace/trace.c b/kernel/trace/trace.c
index 95ae7c4e5835..77ff2608118b 100644
--- a/kernel/trace/trace.c
+++ b/kernel/trace/trace.c
@@ -10277,8 +10277,7 @@ static struct vfsmount *trace_automount(struct dentry *mntpt, void *ingore)
 	if (IS_ERR(fc))
 		return ERR_CAST(fc);
 
-	ret = vfs_parse_fs_string(fc, "source",
-				  "tracefs", strlen("tracefs"));
+	ret = vfs_parse_fs_string(fc, "source", "tracefs");
 	if (!ret)
 		mnt = fc_mount(fc);
 	else

Re: [RFC] vfs_parse_fs_string() calling conventions change (was Re: [PATCH v2 17/35] sanitize handling of long-term internal mounts)

[Al Viro]

On Sat, Jun 28, 2025 at 05:28:25PM +0100, Al Viro wrote:
> On Sat, Jun 28, 2025 at 08:58:49AM +0100, Al Viro wrote:
> > Yes, it's a flagday change.  Compiler will immediately catch any place
> > that needs to be converted, and D/f/porting.rst part should be clear
> > enough.
> > 
> > How about something like the following (completely untested), on top of -rc3?
> > Objections, anyone?
>  
> After fixing a braino (s/QSTR_INIT/QSTR_LEN/) it even builds and seems to work...

While we are at it, there are at least some open-coded instances.  IMO
that part of do_nfs4_mount() is better this way, and I wonder if we should
add vfs_parse_fs_printf(fc, key, fmt, ...) as well...

diff --git a/fs/nfs/nfs4super.c b/fs/nfs/nfs4super.c
index b29a26923ce0..92ac12cee26e 100644
--- a/fs/nfs/nfs4super.c
+++ b/fs/nfs/nfs4super.c
@@ -158,12 +158,6 @@ static int do_nfs4_mount(struct nfs_server *server,
 		.dirfd	= -1,
 	};
 
-	struct fs_parameter param_fsc = {
-		.key	= "fsc",
-		.type	= fs_value_is_string,
-		.dirfd	= -1,
-	};
-
 	if (IS_ERR(server))
 		return PTR_ERR(server);
 
@@ -181,15 +175,7 @@ static int do_nfs4_mount(struct nfs_server *server,
 	root_ctx->server = server;
 
 	if (ctx->fscache_uniq) {
-		len = strlen(ctx->fscache_uniq);
-		param_fsc.size = len;
-		param_fsc.string = kmemdup_nul(ctx->fscache_uniq, len, GFP_KERNEL);
-		if (param_fsc.string == NULL) {
-			put_fs_context(root_fc);
-			return -ENOMEM;
-		}
-		ret = vfs_parse_fs_param(root_fc, &param_fsc);
-		kfree(param_fsc.string);
+		ret = vfs_parse_fs_string(root_fc, "fsc", ctx->fscache_uniq);
 		if (ret < 0) {
 			put_fs_context(root_fc);
 			return ret;

Re: [RFC] vfs_parse_fs_string() calling conventions change (was Re: [PATCH v2 17/35] sanitize handling of long-term internal mounts)

[Linus Torvalds]

On Sat, 28 Jun 2025 at 00:58, Al Viro <viro@zeniv.linux.org.uk> wrote:
>
> Frankly, looking at that stuff...  I wonder if we should add
> vfs_parse_fs_qstr() for "comes with length" variant and lose the length
> argument of vfs_parse_fs_string().

Yeah, I had had the same reaction that the whole "str,len" thing could
be a qstr.

So no objections.

> Yes, it's a flagday change.

Yes, not optimal, but there aren't that many call-sites, and as you
say, anything that gets messed up (due to stable backports or
whatever) will be caught by the compiler anyway.

            Linus

Re: [RFC] vfs_parse_fs_string() calling conventions change (was Re: [PATCH v2 17/35] sanitize handling of long-term internal mounts)

[David Howells]

Al Viro <viro@zeniv.linux.org.uk> wrote:

> Frankly, looking at that stuff...  I wonder if we should add
> vfs_parse_fs_qstr() for "comes with length" variant and lose the length
> argument of vfs_parse_fs_string().

Um - why use a qstr?  Does using a qstr actually gain anything much?  Why not
just, say:

extern int vfs_parse_fs_str_len(struct fs_context *fc, const char *key,
				const char *value, size_t v_size);
static inline int vfs_parse_fs_string(struct fs_context *fc, const char *key,
			       const char *value)
{
	return vfs_parse_fs_str_len(fc, key, value, value ? strlen(value) : 0);
}

Is any arch other than i386 limited to fewer than four register arguments?

David

Re: [RFC] vfs_parse_fs_string() calling conventions change (was Re: [PATCH v2 17/35] sanitize handling of long-term internal mounts)

[Al Viro]

On Mon, Jun 30, 2025 at 04:19:53PM +0100, David Howells wrote:
> Al Viro <viro@zeniv.linux.org.uk> wrote:
> 
> > Frankly, looking at that stuff...  I wonder if we should add
> > vfs_parse_fs_qstr() for "comes with length" variant and lose the length
> > argument of vfs_parse_fs_string().
> 
> Um - why use a qstr?  Does using a qstr actually gain anything much?  Why not
> just, say:
> 
> extern int vfs_parse_fs_str_len(struct fs_context *fc, const char *key,
> 				const char *value, size_t v_size);
> static inline int vfs_parse_fs_string(struct fs_context *fc, const char *key,
> 			       const char *value)
> {
> 	return vfs_parse_fs_str_len(fc, key, value, value ? strlen(value) : 0);
> }
> 
> Is any arch other than i386 limited to fewer than four register arguments?

amd64, for one, but what does that have to do with anything?  If we are parsing
mount parameters at point where one or two words on stack are critical for
stack overflow, we are fucked anyway.

It's just that qstr is the usual type for string-with-length...

Re: [RFC] vfs_parse_fs_string() calling conventions change (was Re: [PATCH v2 17/35] sanitize handling of long-term internal mounts)

[Linus Torvalds]

On Mon, 30 Jun 2025 at 09:55, Al Viro <viro@zeniv.linux.org.uk> wrote:
> > Is any arch other than i386 limited to fewer than four register arguments?
>
> amd64, for one

No, x86-64 has 6 argument registers.

Anyway, the use of 'qstr' doesn't actually change any of that, since
it will use exactly the same number of argument registers as having
separate 'char *, len' arguments.

But the thing that makes qstr nice is that it's not only a good
abstraction, we also have useful helper macros for exactly this use.

IOW, instead of passing in "name, strlen(name)", you can use
"QSTR(name)" which does exactly that.

             Linus

[PATCH v2 18/35] Rewrite of propagate_umount()

[Al Viro]

The variant currently in the tree has problems; trying to prove
correctness has caught at least one class of bugs (reparenting
that ends up moving the visible location of reparented mount, due
to not excluding some of the counterparts on propagation that
should've been included).

I tried to prove that it's the only bug there; I'm still not sure
whether it is.  If anyone can reconstruct and write down an analysis
of the mainline implementation, I'll gladly review it; as it is,
I ended up doing a different implementation.  Candidate collection
phase is similar, but trimming the set down until it satisfies the
constraints turned out pretty different.

I hoped to do transformation as a massage series, but that turns out
to be too convoluted.  So it's a single patch replacing propagate_umount()
and friends in one go, with notes and analysis in D/f/propagate_umount.txt
(in addition to inline comments).

As far I can tell, it is provably correct and provably linear by the number
of mounts we need to look at in order to decide what should be unmounted.
It even builds and seems to survive testing...

Another nice thing that fell out of that is that ->mnt_umounting is no longer
needed.

Compared to the first version:
	* explicit MNT_UMOUNT_CANDIDATE flag for is_candidate()
	* trim_ancestors() only clears that flag, leaving the suckers on list
	* trim_one() and handle_locked() take the stuff with flag cleared off
the list.  That allows to iterate with list_for_each_entry_safe() when calling
trim_one() - it removes at most one element from the list now.
	* no globals - I didn't bother with any kind of context, not worth it.

	* Notes updated accordingly; I have not touch the terms yet.

Reviewed-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 .../filesystems/propagate_umount.txt          | 484 ++++++++++++++++++
 fs/mount.h                                    |   1 -
 fs/namespace.c                                |   1 -
 fs/pnode.c                                    | 362 +++++++------
 fs/pnode.h                                    |   2 +-
 include/linux/mount.h                         |   3 +-
 6 files changed, 685 insertions(+), 168 deletions(-)
 create mode 100644 Documentation/filesystems/propagate_umount.txt

diff --git a/Documentation/filesystems/propagate_umount.txt b/Documentation/filesystems/propagate_umount.txt
new file mode 100644
index 000000000000..5b48540b4059
--- /dev/null
+++ b/Documentation/filesystems/propagate_umount.txt
@@ -0,0 +1,484 @@
+	Notes on propagate_umount()
+
+Umount propagation starts with a set of mounts we are already going to
+take out.  Ideally, we would like to add all downstream cognates to
+that set - anything with the same mountpoint as one of the removed
+mounts and with parent that would receive events from the parent of that
+mount.  However, there are some constraints the resulting set must
+satisfy.
+
+It is convenient to define several properties of sets of mounts:
+
+1) A set S of mounts is non-shifting if for any mount X belonging
+to S all subtrees mounted strictly inside of X (i.e. not overmounting
+the root of X) contain only elements of S.
+
+2) A set S is non-revealing if all locked mounts that belong to S have
+parents that also belong to S.
+
+3) A set S is closed if it contains all children of its elements.
+
+The set of mounts taken out by umount(2) must be non-shifting and
+non-revealing; the first constraint is what allows to reparent
+any remaining mounts and the second is what prevents the exposure
+of any concealed mountpoints.
+
+propagate_umount() takes the original set as an argument and tries to
+extend that set.  The original set is a full subtree and its root is
+unlocked; what matters is that it's closed and non-revealing.
+Resulting set may not be closed; there might still be mounts outside
+of that set, but only on top of stacks of root-overmounting elements
+of set.  They can be reparented to the place where the bottom of
+stack is attached to a mount that will survive.  NOTE: doing that
+will violate a constraint on having no more than one mount with
+the same parent/mountpoint pair; however, the caller (umount_tree())
+will immediately remedy that - it may keep unmounted element attached
+to parent, but only if the parent itself is unmounted.  Since all
+conflicts created by reparenting have common parent *not* in the
+set and one side of the conflict (bottom of the stack of overmounts)
+is in the set, it will be resolved.  However, we rely upon umount_tree()
+doing that pretty much immediately after the call of propagate_umount().
+
+Algorithm is based on two statements:
+	1) for any set S, there is a maximal non-shifting subset of S
+and it can be calculated in O(#S) time.
+	2) for any non-shifting set S, there is a maximal non-revealing
+subset of S.  That subset is also non-shifting and it can be calculated
+in O(#S) time.
+
+		Finding candidates.
+
+We are given a closed set U and we want to find all mounts that have
+the same mountpoint as some mount m in U *and* whose parent receives
+propagation from the parent of the same mount m.  Naive implementation
+would be
+	S = {}
+	for each m in U
+		add m to S
+		p = parent(m)
+		for each q in Propagation(p) - {p}
+			child = look_up(q, mountpoint(m))
+			if child
+				add child to S
+but that can lead to excessive work - there might be propagation among the
+subtrees of U, in which case we'd end up examining the same candidates
+many times.  Since propagation is transitive, the same will happen to
+everything downstream of that candidate and it's not hard to construct
+cases where the approach above leads to the time quadratic by the actual
+number of candidates.
+
+Note that if we run into a candidate we'd already seen, it must've been
+added on an earlier iteration of the outer loop - all additions made
+during one iteration of the outer loop have different parents.  So
+if we find a child already added to the set, we know that everything
+in Propagation(parent(child)) with the same mountpoint has been already
+added.
+	S = {}
+	for each m in U
+		if m in S
+			continue
+		add m to S
+		p = parent(m)
+		q = propagation_next(p, p)
+		while q
+			child = look_up(q, mountpoint(m))
+			if child
+				if child in S
+					q = skip_them(q, p)
+					continue;
+				add child to S
+			q = propagation_next(q, p)
+where
+skip_them(q, p)
+	keep walking Propagation(p) from q until we find something
+	not in Propagation(q)
+
+would get rid of that problem, but we need a sane implementation of
+skip_them().  That's not hard to do - split propagation_next() into
+"down into mnt_slave_list" and "forward-and-up" parts, with the
+skip_them() being "repeat the forward-and-up part until we get NULL
+or something that isn't a peer of the one we are skipping".
+
+Note that there can be no absolute roots among the extra candidates -
+they all come from mount lookups.  Absolute root among the original
+set is _currently_ impossible, but it might be worth protecting
+against.
+
+		Maximal non-shifting subsets.
+
+Let's call a mount m in a set S forbidden in that set if there is a
+subtree mounted strictly inside m and containing mounts that do not
+belong to S.
+
+The set is non-shifting when none of its elements are forbidden in it.
+
+If mount m is forbidden in a set S, it is forbidden in any subset S' it
+belongs to.  In other words, it can't belong to any of the non-shifting
+subsets of S.  If we had a way to find a forbidden mount or show that
+there's none, we could use it to find the maximal non-shifting subset
+simply by finding and removing them until none remain.
+
+Suppose mount m is forbidden in S; then any mounts forbidden in S - {m}
+must have been forbidden in S itself.  Indeed, since m has descendents
+that do not belong to S, any subtree that fits into S will fit into
+S - {m} as well.
+
+So in principle we could go through elements of S, checking if they
+are forbidden in S and removing the ones that are.  Removals will
+not invalidate the checks done for earlier mounts - if they were not
+forbidden at the time we checked, they won't become forbidden later.
+It's too costly to be practical, but there is a similar approach that
+is linear by size of S.
+
+Let's say that mount x in a set S is forbidden by mount y, if
+	* both x and y belong to S.
+	* there is a chain of mounts starting at x and leaving S
+	  immediately after passing through y, with the first
+	  mountpoint strictly inside x.
+Note 1: x may be equal to y - that's the case when something not
+belonging to S is mounted strictly inside x.
+Note 2: if y does not belong to S, it can't forbid anything in S.
+Note 3: if y has no children outside of S, it can't forbid anything in S.
+
+It's easy to show that mount x is forbidden in S if and only if x is
+forbidden in S by some mount y.  And it's easy to find all mounts in S
+forbidden by a given mount.
+
+Consider the following operation:
+	Trim(S, m) = S - {x : x is forbidden by m in S}
+
+Note that if m does not belong to S or has no children outside of S we
+are guaranteed that Trim(S, m) is equal to S.
+
+The following is true: if x is forbidden by y in Trim(S, m), it was
+already forbidden by y in S.
+
+Proof: Suppose x is forbidden by y in Trim(S, m).  Then there is a
+chain of mounts (x_0 = x, ..., x_k = y, x_{k+1} = r), such that x_{k+1}
+is the first element that doesn't belong to Trim(S, m) and the
+mountpoint of x_1 is strictly inside x.  If mount r belongs to S, it must
+have been removed by Trim(S, m), i.e. it was forbidden in S by m.
+Then there was a mount chain from r to some child of m that stayed in
+S all the way until m, but that's impossible since x belongs to Trim(S, m)
+and prepending (x_0, ..., x_k) to that chain demonstrates that x is also
+prohibited in S by m, and thus can't belong to Trim(S, m).
+Therefore r can not belong to S and our chain demonstrates that
+x is prohibited by y in S.  QED.
+
+Corollary: no mount is forbidden by m in Trim(S, m).  Indeed, any
+such mount would have been forbidden by m in S and thus would have been
+in the part of S removed in Trim(S, m).
+
+Corollary: no mount is forbidden by m in Trim(Trim(S, m), n).  Indeed,
+any such would have to have been forbidden by m in Trim(S, m), which
+is impossible.
+
+Corollary: after
+	S = Trim(S, x_1)
+	S = Trim(S, x_2)
+	...
+	S = Trim(S, x_k)
+no mount remaining in S will be forbidden by either of x_1,...,x_k.
+
+The following will reduce S to its maximal non-shifting subset:
+	visited = {}
+	while S contains elements not belonging to visited
+		let m be an arbitrary such element of S
+		S = Trim(S, m)
+		add m to visited
+
+S never grows, so the number of elements of S not belonging to visited
+decreases at least by one on each iteration.  When the loop terminates,
+all mounts remaining in S belong to visited.  It's easy to see that at
+the beginning of each iteration no mount remaining in S will be forbidden
+by any element of visited.  In other words, no mount remaining in S will
+be forbidden, i.e. final value of S will be non-shifting.  It will be
+the maximal non-shifting subset, since we were removing only forbidden
+elements.
+
+	There are two difficulties in implementing the above in linear
+time, both due to the fact that Trim() might need to remove more than one
+element.  Naive implementation of Trim() is vulnerable to running into a
+long chain of mounts, each mounted on top of parent's root.  Nothing in
+that chain is prohibited, so nothing gets removed from it.  We need to
+recognize such chains and avoid walking them again on subsequent calls of
+Trim(), otherwise we will end up with worst-case time being quadratic by
+the number of elements in S.  Another difficulty is in implementing the
+outer loop - we need to iterate through all elements of a shrinking set.
+That would be trivial if we never removed more than one element at a time
+(linked list, with list_for_each_entry_safe for iterator), but we may
+need to remove more than one entry, possibly including the ones we have
+already visited.
+
+	Let's start with naive algorithm for Trim():
+
+Trim_one(m)
+	found = false
+	for each n in children(m)
+		if n not in S
+			found = true
+			if (mountpoint(n) != root(m))
+				remove m from S
+				break
+	if found
+		Trim_ancestors(m)
+
+Trim_ancestors(m)
+	for (; parent(m) in S; m = parent(m)) {
+		if (mountpoint(m) != root(parent(m)))
+			remove parent(m) from S
+	}
+
+If m belongs to S, Trim_one(m) will replace S with Trim(S, m).
+Proof:
+	Consider the chains excluding elements from Trim(S, m).  The last
+two elements in such chain are m and some child of m that does not belong
+to S.  If m has no such children, Trim(S, m) is equal to S.
+	m itself is removed if and only if the chain has exactly two
+elements, i.e. when the last element does not overmount the root of m.
+In other words, that happens when m has a child not in S that does not
+overmount the root of m.
+	All other elements to remove will be ancestors of m, such that
+the entire descent chain from them to m is contained in S.  Let
+(x_0, x_1, ..., x_k = m) be the longest such chain.  x_i needs to be
+removed if and only if x_{i+1} does not overmount its root.  It's easy
+to see that Trim_ancestors(m) will iterate through that chain from
+x_k to x_1 and that it will remove exactly the elements that need to be
+removed.
+
+	Note that if the loop in Trim_ancestors() walks into an already
+visited element, we are guaranteed that remaining iterations will see
+only elements that had already been visited and remove none of them.
+That's the weakness that makes it vulnerable to long chains of full
+overmounts.
+
+	It's easy to deal with, if we can afford setting marks on
+elements of S; we would mark all elements already visited by
+Trim_ancestors() and have it bail out as soon as it sees an already
+marked element.
+
+	The problems with iterating through the set can be dealt with in
+several ways, depending upon the representation we choose for our set.
+One useful observation is that we are given a closed subset in S - the
+original set passed to propagate_umount().  Its elements can neither
+prohibit anything nor be prohibited by anything - all their descendents
+belong to S, so they can not occur anywhere in any excluding chain.
+In other words, the elements of that subset will remain in S until
+the end and Trim_one(S, m) is a no-op for all m from that subset.
+
+	That suggests keeping S as a disjoint union of a closed set U
+('will be unmounted, no matter what') and the set of all elements of
+S that do not belong to U.  That set ('candidates') is all we need
+to iterate through.  Let's represent it as a subset in a cyclic list,
+consisting of all list elements that are marked as candidates (initially -
+all of them).  Then we could have Trim_ancestors() only remove the mark,
+leaving the elements on the list.  Then Trim_one() would never remove
+anything other than its argument from the containing list, allowing to
+use list_for_each_entry_safe() as iterator.
+
+	Assuming that representation we get the following:
+
+	list_for_each_entry_safe(m, ..., Candidates, ...)
+		Trim_one(m)
+where
+Trim_one(m)
+	if (m is not marked as a candidate)
+		strip the "seen by Trim_ancestors" mark from m
+		remove m from the Candidates list
+		return
+		
+	remove_this = false
+	found = false
+	for each n in children(m)
+		if n not in S
+			found = true
+			if (mountpoint(n) != root(m))
+				remove_this = true
+				break
+	if found
+		Trim_ancestors(m)
+	if remove_this
+		strip the "seen by Trim_ancestors" mark from m
+		strip the "candidate" mark from m
+		remove m from the Candidate list
+
+Trim_ancestors(m)
+	for (p = parent(m); p is marked as candidate ; m = p, p = parent(p)) {
+		if m is marked as seen by Trim_ancestors
+			return
+		mark m as seen by Trim_ancestors
+		if (mountpoint(m) != root(p))
+			strip the "candidate" mark from p
+	}
+
+	Terminating condition in the loop in Trim_ancestors() is correct,
+since that that loop will never run into p belonging to U - p is always
+an ancestor of argument of Trim_one() and since U is closed, the argument
+of Trim_one() would also have to belong to U.  But Trim_one() is never
+called for elements of U.  In other words, p belongs to S if and only
+if it belongs to candidates.
+
+	Time complexity:
+* we get no more than O(#S) calls of Trim_one()
+* the loop over children in Trim_one() never looks at the same child
+twice through all the calls.
+* iterations of that loop for children in S are no more than O(#S)
+in the worst case
+* at most two children that are not elements of S are considered per
+call of Trim_one().
+* the loop in Trim_ancestors() sets its mark once per iteration and
+no element of S has is set more than once.
+
+	In the end we may have some elements excluded from S by
+Trim_ancestors() still stuck on the list.  We could do a separate
+loop removing them from the list (also no worse than O(#S) time),
+but it's easier to leave that until the next phase - there we will
+iterate through the candidates anyway.
+
+	The caller has already removed all elements of U from their parents'
+lists of children, which means that checking if child belongs to S is
+equivalent to checking if it's marked as a candidate; we'll never see
+the elements of U in the loop over children in Trim_one().
+
+	What's more, if we see that children(m) is empty and m is not
+locked, we can immediately move m into the committed subset (remove
+from the parent's list of children, etc.).  That's one fewer mount we'll
+have to look into when we check the list of children of its parent *and*
+when we get to building the non-revealing subset.
+
+		Maximal non-revealing subsets
+
+If S is not a non-revealing subset, there is a locked element x in S
+such that parent of x is not in S.
+
+Obviously, no non-revealing subset of S may contain x.  Removing such
+elements one by one will obviously end with the maximal non-revealing
+subset (possibly empty one).  Note that removal of an element will
+require removal of all its locked children, etc.
+
+If the set had been non-shifting, it will remain non-shifting after
+such removals.
+Proof: suppose S was non-shifting, x is a locked element of S, parent of x
+is not in S and S - {x} is not non-shifting.  Then there is an element m
+in S - {x} and a subtree mounted strictly inside m, such that m contains
+an element not in in S - {x}.  Since S is non-shifting, everything in
+that subtree must belong to S.  But that means that this subtree must
+contain x somewhere *and* that parent of x either belongs that subtree
+or is equal to m.  Either way it must belong to S.  Contradiction.
+
+// same representation as for finding maximal non-shifting subsets:
+// S is a disjoint union of a non-revealing set U (the ones we are committed
+// to unmount) and a set of candidates, represented as a subset of list
+// elements that have "is a candidate" mark on them.
+// Elements of U are removed from their parents' lists of children.
+// In the end candidates becomes empty and maximal non-revealing non-shifting
+// subset of S is now in U
+	while (Candidates list is non-empty)
+		handle_locked(first(Candidates))
+
+handle_locked(m)
+	if m is not marked as a candidate
+		strip the "seen by Trim_ancestors" mark from m
+		remove m from the list
+		return
+	cutoff = m
+	for (p = m; p in candidates; p = parent(p)) {
+		strip the "seen by Trim_ancestors" mark from p
+		strip the "candidate" mark from p
+		remove p from the Candidates list
+		if (!locked(p))
+			cutoff = parent(p)
+	}
+	if p in U
+		cutoff = p
+	while m != cutoff
+		remove m from children(parent(m))
+		add m to U
+		m = parent(m)
+
+Let (x_0, ..., x_n = m) be the maximal chain of descent of m within S.
+* If it contains some elements of U, let x_k be the last one of those.
+Then union of U with {x_{k+1}, ..., x_n} is obviously non-revealing.
+* otherwise if all its elements are locked, then none of {x_0, ..., x_n}
+may be elements of a non-revealing subset of S.
+* otherwise let x_k be the first unlocked element of the chain.  Then none
+of {x_0, ..., x_{k-1}} may be an element of a non-revealing subset of
+S and union of U and {x_k, ..., x_n} is non-revealing.
+
+handle_locked(m) finds which of these cases applies and adjusts Candidates
+and U accordingly.  U remains non-revealing, union of Candidates and
+U still contains any non-revealing subset of S and after the call of
+handle_locked(m) m is guaranteed to be not in Candidates list.  So having
+it called for each element of S would suffice to empty Candidates,
+leaving U the maximal non-revealing subset of S.
+
+However, handle_locked(m) is a no-op when m belongs to U, so it's enough
+to have it called for elements of Candidates list until none remain.
+
+Time complexity: number of calls of handle_locked() is limited by
+#Candidates, each iteration of the first loop in handle_locked() removes
+an element from the list, so their total number of executions is also
+limited by #Candidates; number of iterations in the second loop is no
+greater than the number of iterations of the first loop.
+
+
+		Reparenting
+
+After we'd calculated the final set, we still need to deal with
+reparenting - if an element of the final set has a child not in it,
+we need to reparent such child.
+
+Such children can only be root-overmounting (otherwise the set wouldn't
+be non-shifting) and their parents can not belong to the original set,
+since the original is guaranteed to be closed.
+
+
+		Putting all of that together
+
+The plan is to
+	* find all candidates
+	* trim down to maximal non-shifting subset
+	* trim down to maximal non-revealing subset
+	* reparent anything that needs to be reparented
+	* return the resulting set to the caller
+
+For the 2nd and 3rd steps we want to separate the set into growing
+non-revealing subset, initially containing the original set ("U" in
+terms of the pseudocode above) and everything we are still not sure about
+("candidates").  It means that for the output of the 1st step we'd like
+the extra candidates separated from the stuff already in the original set.
+For the 4th step we would like the additions to U separate from the
+original set.
+
+So let's go for
+	* original set ("set").  Linkage via mnt_list
+	* undecided candidates ("candidates").  Subset of a list,
+consisting of all its elements marked with a new flag (MNT_UMOUNT_CANDIDATE).
+Initially all elements of the list will be marked that way; in the
+end the list will become empty and no mounts will remain marked with
+that flag.
+	* Reuse MNT_MARKED for "has been already seen by trim_ancestors()".
+	* anything in U that hadn't been in the original set - elements of
+candidates will gradually be either discarded or moved there.  In other
+words, it's the candidates we have already decided to unmount.	Its role
+is reasonably close to the old "to_umount", so let's use that name.
+Linkage via mnt_list.
+
+For gather_candidates() we'll need to maintain both candidates (S -
+set) and intersection of S with set.  Use MNT_UMOUNT_CANDIDATE for
+all elements we encounter, putting the ones not already in the original
+set into the list of candidates.  When we are done, strip that flag from
+all elements of the original set.  That gives a cheap way to check
+if element belongs to S (in gather_candidates) and to candidates
+itself (at later stages).  Call that predicate is_candidate(); it would
+be m->mnt_flags & MNT_UMOUNT_CANDIDATE.
+
+All elements of the original set are marked with MNT_UMOUNT and we'll
+need the same for elements added when joining the contents of to_umount
+to set in the end.  Let's set MNT_UMOUNT at the time we add an element
+to to_umount; that's close to what the old 'umount_one' is doing, so
+let's keep that name.  It also gives us another predicate we need -
+"belongs to union of set and to_umount"; will_be_unmounted() for now.
+
+Removals from the candidates list should strip both MNT_MARKED and
+MNT_UMOUNT_CANDIDATE; call it remove_from_candidates_list().
diff --git a/fs/mount.h b/fs/mount.h
index f20e6ed845fe..fb93d3e16724 100644
--- a/fs/mount.h
+++ b/fs/mount.h
@@ -79,7 +79,6 @@ struct mount {
 		struct hlist_node mnt_mp_list;	/* list mounts with the same mountpoint */
 		struct hlist_node mnt_umount;
 	};
-	struct list_head mnt_umounting; /* list entry for umount propagation */
 #ifdef CONFIG_FSNOTIFY
 	struct fsnotify_mark_connector __rcu *mnt_fsnotify_marks;
 	__u32 mnt_fsnotify_mask;
diff --git a/fs/namespace.c b/fs/namespace.c
index 96bc3b804974..d99c05f7031f 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -383,7 +383,6 @@ static struct mount *alloc_vfsmnt(const char *name)
 		INIT_LIST_HEAD(&mnt->mnt_slave_list);
 		INIT_LIST_HEAD(&mnt->mnt_slave);
 		INIT_HLIST_NODE(&mnt->mnt_mp_list);
-		INIT_LIST_HEAD(&mnt->mnt_umounting);
 		INIT_HLIST_HEAD(&mnt->mnt_stuck_children);
 		RB_CLEAR_NODE(&mnt->mnt_node);
 		mnt->mnt.mnt_idmap = &nop_mnt_idmap;
diff --git a/fs/pnode.c b/fs/pnode.c
index aa187144e389..9aa892263a38 100644
--- a/fs/pnode.c
+++ b/fs/pnode.c
@@ -24,11 +24,6 @@ static inline struct mount *first_slave(struct mount *p)
 	return list_entry(p->mnt_slave_list.next, struct mount, mnt_slave);
 }
 
-static inline struct mount *last_slave(struct mount *p)
-{
-	return list_entry(p->mnt_slave_list.prev, struct mount, mnt_slave);
-}
-
 static inline struct mount *next_slave(struct mount *p)
 {
 	return list_entry(p->mnt_slave.next, struct mount, mnt_slave);
@@ -136,6 +131,23 @@ void change_mnt_propagation(struct mount *mnt, int type)
 	}
 }
 
+static struct mount *__propagation_next(struct mount *m,
+					 struct mount *origin)
+{
+	while (1) {
+		struct mount *master = m->mnt_master;
+
+		if (master == origin->mnt_master) {
+			struct mount *next = next_peer(m);
+			return (next == origin) ? NULL : next;
+		} else if (m->mnt_slave.next != &master->mnt_slave_list)
+			return next_slave(m);
+
+		/* back at master */
+		m = master;
+	}
+}
+
 /*
  * get the next mount in the propagation tree.
  * @m: the mount seen last
@@ -153,31 +165,21 @@ static struct mount *propagation_next(struct mount *m,
 	if (!IS_MNT_NEW(m) && !list_empty(&m->mnt_slave_list))
 		return first_slave(m);
 
-	while (1) {
-		struct mount *master = m->mnt_master;
-
-		if (master == origin->mnt_master) {
-			struct mount *next = next_peer(m);
-			return (next == origin) ? NULL : next;
-		} else if (m->mnt_slave.next != &master->mnt_slave_list)
-			return next_slave(m);
-
-		/* back at master */
-		m = master;
-	}
+	return __propagation_next(m, origin);
 }
 
 static struct mount *skip_propagation_subtree(struct mount *m,
 						struct mount *origin)
 {
 	/*
-	 * Advance m such that propagation_next will not return
-	 * the slaves of m.
+	 * Advance m past everything that gets propagation from it.
 	 */
-	if (!IS_MNT_NEW(m) && !list_empty(&m->mnt_slave_list))
-		m = last_slave(m);
+	struct mount *p = __propagation_next(m, origin);
+
+	while (p && peers(m, p))
+		p = __propagation_next(p, origin);
 
-	return m;
+	return p;
 }
 
 static struct mount *next_group(struct mount *m, struct mount *origin)
@@ -458,181 +460,213 @@ void propagate_mount_unlock(struct mount *mnt)
 	}
 }
 
-static void umount_one(struct mount *mnt, struct list_head *to_umount)
+static inline bool is_candidate(struct mount *m)
 {
-	CLEAR_MNT_MARK(mnt);
-	mnt->mnt.mnt_flags |= MNT_UMOUNT;
-	list_del_init(&mnt->mnt_child);
-	list_del_init(&mnt->mnt_umounting);
-	move_from_ns(mnt, to_umount);
+	return m->mnt.mnt_flags & MNT_UMOUNT_CANDIDATE;
 }
 
-/*
- * NOTE: unmounting 'mnt' naturally propagates to all other mounts its
- * parent propagates to.
- */
-static bool __propagate_umount(struct mount *mnt,
-			       struct list_head *to_umount,
-			       struct list_head *to_restore)
+static inline bool will_be_unmounted(struct mount *m)
 {
-	bool progress = false;
-	struct mount *child;
+	return m->mnt.mnt_flags & MNT_UMOUNT;
+}
 
-	/*
-	 * The state of the parent won't change if this mount is
-	 * already unmounted or marked as without children.
-	 */
-	if (mnt->mnt.mnt_flags & (MNT_UMOUNT | MNT_MARKED))
-		goto out;
+static void umount_one(struct mount *m, struct list_head *to_umount)
+{
+	m->mnt.mnt_flags |= MNT_UMOUNT;
+	list_del_init(&m->mnt_child);
+	move_from_ns(m, to_umount);
+}
 
-	/* Verify topper is the only grandchild that has not been
-	 * speculatively unmounted.
-	 */
-	list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
-		if (child->mnt_mountpoint == mnt->mnt.mnt_root)
-			continue;
-		if (!list_empty(&child->mnt_umounting) && IS_MNT_MARKED(child))
-			continue;
-		/* Found a mounted child */
-		goto children;
-	}
+static void remove_from_candidate_list(struct mount *m)
+{
+	m->mnt.mnt_flags &= ~(MNT_MARKED | MNT_UMOUNT_CANDIDATE);
+	list_del_init(&m->mnt_list);
+}
 
-	/* Mark mounts that can be unmounted if not locked */
-	SET_MNT_MARK(mnt);
-	progress = true;
+static void gather_candidates(struct list_head *set,
+			      struct list_head *candidates)
+{
+	struct mount *m, *p, *q;
 
-	/* If a mount is without children and not locked umount it. */
-	if (!IS_MNT_LOCKED(mnt)) {
-		umount_one(mnt, to_umount);
-	} else {
-children:
-		list_move_tail(&mnt->mnt_umounting, to_restore);
+	list_for_each_entry(m, set, mnt_list) {
+		if (is_candidate(m))
+			continue;
+		m->mnt.mnt_flags |= MNT_UMOUNT_CANDIDATE;
+		p = m->mnt_parent;
+		q = propagation_next(p, p);
+		while (q) {
+			struct mount *child = __lookup_mnt(&q->mnt,
+							   m->mnt_mountpoint);
+			if (child) {
+				/*
+				 * We might've already run into this one.  That
+				 * must've happened on earlier iteration of the
+				 * outer loop; in that case we can skip those
+				 * parents that get propagation from q - there
+				 * will be nothing new on those as well.
+				 */
+				if (is_candidate(child)) {
+					q = skip_propagation_subtree(q, p);
+					continue;
+				}
+				child->mnt.mnt_flags |= MNT_UMOUNT_CANDIDATE;
+				if (!will_be_unmounted(child))
+					list_add(&child->mnt_list, candidates);
+			}
+			q = propagation_next(q, p);
+		}
 	}
-out:
-	return progress;
+	list_for_each_entry(m, set, mnt_list)
+		m->mnt.mnt_flags &= ~MNT_UMOUNT_CANDIDATE;
 }
 
-static void umount_list(struct list_head *to_umount,
-			struct list_head *to_restore)
+/*
+ * We know that some child of @m can't be unmounted.  In all places where the
+ * chain of descent of @m has child not overmounting the root of parent,
+ * the parent can't be unmounted either.
+ */
+static void trim_ancestors(struct mount *m)
 {
-	struct mount *mnt, *child, *tmp;
-	list_for_each_entry(mnt, to_umount, mnt_list) {
-		list_for_each_entry_safe(child, tmp, &mnt->mnt_mounts, mnt_child) {
-			/* topper? */
-			if (child->mnt_mountpoint == mnt->mnt.mnt_root)
-				list_move_tail(&child->mnt_umounting, to_restore);
-			else
-				umount_one(child, to_umount);
-		}
+	struct mount *p;
+
+	for (p = m->mnt_parent; is_candidate(p); m = p, p = p->mnt_parent) {
+		if (IS_MNT_MARKED(m))	// all candidates beneath are overmounts
+			return;
+		SET_MNT_MARK(m);
+		if (m->mnt_mountpoint != p->mnt.mnt_root)
+			p->mnt.mnt_flags &= ~MNT_UMOUNT_CANDIDATE;
 	}
 }
 
-static void restore_mounts(struct list_head *to_restore)
+/*
+ * Find and exclude all umount candidates forbidden by @m
+ * (see Documentation/filesystems/propagate_umount.txt)
+ * If we can immediately tell that @m is OK to unmount (unlocked
+ * and all children are already committed to unmounting) commit
+ * to unmounting it.
+ * Only @m itself might be taken from the candidates list;
+ * anything found by trim_ancestors() is marked non-candidate
+ * and left on the list.
+ */
+static void trim_one(struct mount *m, struct list_head *to_umount)
 {
-	/* Restore mounts to a clean working state */
-	while (!list_empty(to_restore)) {
-		struct mount *mnt, *parent;
-		struct mountpoint *mp;
-
-		mnt = list_first_entry(to_restore, struct mount, mnt_umounting);
-		CLEAR_MNT_MARK(mnt);
-		list_del_init(&mnt->mnt_umounting);
-
-		/* Should this mount be reparented? */
-		mp = mnt->mnt_mp;
-		parent = mnt->mnt_parent;
-		while (parent->mnt.mnt_flags & MNT_UMOUNT) {
-			mp = parent->mnt_mp;
-			parent = parent->mnt_parent;
-		}
-		if (parent != mnt->mnt_parent) {
-			mnt_change_mountpoint(parent, mp, mnt);
-			mnt_notify_add(mnt);
+	bool remove_this = false, found = false, umount_this = false;
+	struct mount *n;
+
+	if (!is_candidate(m)) { // trim_ancestors() left it on list
+		remove_from_candidate_list(m);
+		return;
+	}
+
+	list_for_each_entry(n, &m->mnt_mounts, mnt_child) {
+		if (!is_candidate(n)) {
+			found = true;
+			if (n->mnt_mountpoint != m->mnt.mnt_root) {
+				remove_this = true;
+				break;
+			}
 		}
 	}
+	if (found) {
+		trim_ancestors(m);
+	} else if (!IS_MNT_LOCKED(m) && list_empty(&m->mnt_mounts)) {
+		remove_this = true;
+		umount_this = true;
+	}
+	if (remove_this) {
+		remove_from_candidate_list(m);
+		if (umount_this)
+			umount_one(m, to_umount);
+	}
 }
 
-static void cleanup_umount_visitations(struct list_head *visited)
+static void handle_locked(struct mount *m, struct list_head *to_umount)
 {
-	while (!list_empty(visited)) {
-		struct mount *mnt =
-			list_first_entry(visited, struct mount, mnt_umounting);
-		list_del_init(&mnt->mnt_umounting);
+	struct mount *cutoff = m, *p;
+
+	if (!is_candidate(m)) { // trim_ancestors() left it on list
+		remove_from_candidate_list(m);
+		return;
+	}
+	for (p = m; is_candidate(p); p = p->mnt_parent) {
+		remove_from_candidate_list(p);
+		if (!IS_MNT_LOCKED(p))
+			cutoff = p->mnt_parent;
+	}
+	if (will_be_unmounted(p))
+		cutoff = p;
+	while (m != cutoff) {
+		umount_one(m, to_umount);
+		m = m->mnt_parent;
 	}
 }
 
 /*
- * collect all mounts that receive propagation from the mount in @list,
- * and return these additional mounts in the same list.
- * @list: the list of mounts to be unmounted.
+ * @m is not to going away, and it overmounts the top of a stack of mounts
+ * that are going away.  We know that all of those are fully overmounted
+ * by the one above (@m being the topmost of the chain), so @m can be slid
+ * in place where the bottom of the stack is attached.
  *
- * vfsmount lock must be held for write
+ * NOTE: here we temporarily violate a constraint - two mounts end up with
+ * the same parent and mountpoint; that will be remedied as soon as we
+ * return from propagate_umount() - its caller (umount_tree()) will detach
+ * the stack from the parent it (and now @m) is attached to.  umount_tree()
+ * might choose to keep unmounted pieces stuck to each other, but it always
+ * detaches them from the mounts that remain in the tree.
  */
-int propagate_umount(struct list_head *list)
+static void reparent(struct mount *m)
 {
-	struct mount *mnt;
-	LIST_HEAD(to_restore);
-	LIST_HEAD(to_umount);
-	LIST_HEAD(visited);
-
-	/* Find candidates for unmounting */
-	list_for_each_entry_reverse(mnt, list, mnt_list) {
-		struct mount *parent = mnt->mnt_parent;
-		struct mount *m;
+	struct mount *p = m;
+	struct mountpoint *mp;
 
-		/*
-		 * If this mount has already been visited it is known that it's
-		 * entire peer group and all of their slaves in the propagation
-		 * tree for the mountpoint has already been visited and there is
-		 * no need to visit them again.
-		 */
-		if (!list_empty(&mnt->mnt_umounting))
-			continue;
+	do {
+		mp = p->mnt_mp;
+		p = p->mnt_parent;
+	} while (will_be_unmounted(p));
 
-		list_add_tail(&mnt->mnt_umounting, &visited);
-		for (m = propagation_next(parent, parent); m;
-		     m = propagation_next(m, parent)) {
-			struct mount *child = __lookup_mnt(&m->mnt,
-							   mnt->mnt_mountpoint);
-			if (!child)
-				continue;
+	mnt_change_mountpoint(p, mp, m);
+	mnt_notify_add(m);
+}
 
-			if (!list_empty(&child->mnt_umounting)) {
-				/*
-				 * If the child has already been visited it is
-				 * know that it's entire peer group and all of
-				 * their slaves in the propgation tree for the
-				 * mountpoint has already been visited and there
-				 * is no need to visit this subtree again.
-				 */
-				m = skip_propagation_subtree(m, parent);
-				continue;
-			} else if (child->mnt.mnt_flags & MNT_UMOUNT) {
-				/*
-				 * We have come across a partially unmounted
-				 * mount in a list that has not been visited
-				 * yet. Remember it has been visited and
-				 * continue about our merry way.
-				 */
-				list_add_tail(&child->mnt_umounting, &visited);
-				continue;
-			}
+/**
+ * propagate_umount - apply propagation rules to the set of mounts for umount()
+ * @set: the list of mounts to be unmounted.
+ *
+ * Collect all mounts that receive propagation from the mount in @set and have
+ * no obstacles to being unmounted.  Add these additional mounts to the set.
+ *
+ * See Documentation/filesystems/propagate_umount.txt if you do anything in
+ * this area.
+ *
+ * Locks held:
+ * mount_lock (write_seqlock), namespace_sem (exclusive).
+ */
+void propagate_umount(struct list_head *set)
+{
+	struct mount *m, *p;
+	LIST_HEAD(to_umount);	// committed to unmounting
+	LIST_HEAD(candidates);	// undecided umount candidates
 
-			/* Check the child and parents while progress is made */
-			while (__propagate_umount(child,
-						  &to_umount, &to_restore)) {
-				/* Is the parent a umount candidate? */
-				child = child->mnt_parent;
-				if (list_empty(&child->mnt_umounting))
-					break;
-			}
-		}
+	// collect all candidates
+	gather_candidates(set, &candidates);
+
+	// reduce the set until it's non-shifting
+	list_for_each_entry_safe(m, p, &candidates, mnt_list)
+		trim_one(m, &to_umount);
+
+	// ... and non-revealing
+	while (!list_empty(&candidates)) {
+		m = list_first_entry(&candidates,struct mount, mnt_list);
+		handle_locked(m, &to_umount);
 	}
 
-	umount_list(&to_umount, &to_restore);
-	restore_mounts(&to_restore);
-	cleanup_umount_visitations(&visited);
-	list_splice_tail(&to_umount, list);
+	// now to_umount consists of all acceptable candidates
+	// deal with reparenting of remaining overmounts on those
+	list_for_each_entry(m, &to_umount, mnt_list) {
+		if (m->overmount)
+			reparent(m->overmount);
+	}
 
-	return 0;
+	// and fold them into the set
+	list_splice_tail_init(&to_umount, set);
 }
diff --git a/fs/pnode.h b/fs/pnode.h
index 93fa9311bd07..04f1ac53aa49 100644
--- a/fs/pnode.h
+++ b/fs/pnode.h
@@ -42,7 +42,7 @@ static inline bool peers(const struct mount *m1, const struct mount *m2)
 void change_mnt_propagation(struct mount *, int);
 int propagate_mnt(struct mount *, struct mountpoint *, struct mount *,
 		struct hlist_head *);
-int propagate_umount(struct list_head *);
+void propagate_umount(struct list_head *);
 int propagate_mount_busy(struct mount *, int);
 void propagate_mount_unlock(struct mount *);
 void mnt_release_group_id(struct mount *);
diff --git a/include/linux/mount.h b/include/linux/mount.h
index c145820fcbbf..65fa8442c00a 100644
--- a/include/linux/mount.h
+++ b/include/linux/mount.h
@@ -40,6 +40,7 @@ enum mount_flags {
 
 	MNT_INTERNAL	= 0x4000,
 
+	MNT_UMOUNT_CANDIDATE	= 0x020000,
 	MNT_LOCK_ATIME		= 0x040000,
 	MNT_LOCK_NOEXEC		= 0x080000,
 	MNT_LOCK_NOSUID		= 0x100000,
@@ -66,7 +67,7 @@ enum mount_flags {
 
 	MNT_INTERNAL_FLAGS = MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL |
 			     MNT_DOOMED | MNT_SYNC_UMOUNT | MNT_MARKED |
-			     MNT_LOCKED,
+			     MNT_LOCKED | MNT_UMOUNT_CANDIDATE,
 };
 
 struct vfsmount {
-- 
2.39.5

[PATCH v2 19/35] make commit_tree() usable in same-namespace move case

[Al Viro]

Once attach_recursive_mnt() has created all copies of original subtree,
it needs to put them in place(s).

Steps needed for those are slightly different:
	1) in 'move' case, original copy doesn't need any rbtree
manipulations (everything's already in the same namespace where it will
be), but it needs to be detached from the current location
	2) in 'attach' case, original may be in anon namespace; if it is,
all those mounts need to removed from their current namespace before
insertion into the target one
	3) additional copies have a couple of extra twists - in case
of cross-userns propagation we need to lock everything other the root of
subtree and in case when we end up inserting under an existing mount,
that mount needs to be found (for original copy we have it explicitly
passed by the caller).

Quite a bit of that can be unified; as the first step, make commit_tree()
helper (inserting mounts into namespace, hashing the root of subtree
and marking the namespace as updated) usable in all cases; (2) and (3)
are already using it and for (1) we only need to make the insertion of
mounts into namespace conditional.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 30 ++++++++++++++----------------
 1 file changed, 14 insertions(+), 16 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index d99c05f7031f..a73f16926830 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -1172,15 +1172,17 @@ static void commit_tree(struct mount *mnt)
 
 	BUG_ON(parent == mnt);
 
-	list_add_tail(&head, &mnt->mnt_list);
-	while (!list_empty(&head)) {
-		m = list_first_entry(&head, typeof(*m), mnt_list);
-		list_del(&m->mnt_list);
+	if (!mnt_ns_attached(mnt)) {
+		list_add_tail(&head, &mnt->mnt_list);
+		while (!list_empty(&head)) {
+			m = list_first_entry(&head, typeof(*m), mnt_list);
+			list_del(&m->mnt_list);
 
-		mnt_add_to_ns(n, m);
+			mnt_add_to_ns(n, m);
+		}
+		n->nr_mounts += n->pending_mounts;
+		n->pending_mounts = 0;
 	}
-	n->nr_mounts += n->pending_mounts;
-	n->pending_mounts = 0;
 
 	make_visible(mnt);
 	touch_mnt_namespace(n);
@@ -2691,12 +2693,7 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 
 	if (moving) {
 		unhash_mnt(source_mnt);
-		mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
-		if (beneath)
-			mnt_change_mountpoint(top, smp, top_mnt);
-		make_visible(source_mnt);
 		mnt_notify_add(source_mnt);
-		touch_mnt_namespace(source_mnt->mnt_ns);
 	} else {
 		if (source_mnt->mnt_ns) {
 			LIST_HEAD(head);
@@ -2706,12 +2703,13 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 				move_from_ns(p, &head);
 			list_del_init(&head);
 		}
-		mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
-		if (beneath)
-			mnt_change_mountpoint(top, smp, top_mnt);
-		commit_tree(source_mnt);
 	}
 
+	mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
+	if (beneath)
+		mnt_change_mountpoint(top, smp, top_mnt);
+	commit_tree(source_mnt);
+
 	hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
 		struct mount *q;
 		hlist_del_init(&child->mnt_hash);
-- 
2.39.5

[PATCH v2 20/35] attach_recursive_mnt(): unify the mnt_change_mountpoint() logics

[Al Viro]

The logics used for tucking under existing mount differs for original
and copies; copies do a mount hash lookup to see if mountpoint to be is
already overmounted, while the original is told explicitly.

But the same logics that is used for copies works for the original,
at which point the only place where we get very close to eliminating
the need of passing 'beneath' flag to attach_recursive_mnt().

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 24 +++++++++++++++---------
 1 file changed, 15 insertions(+), 9 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index a73f16926830..6092aaf604a7 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -2643,7 +2643,7 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 	HLIST_HEAD(tree_list);
 	struct mnt_namespace *ns = top_mnt->mnt_ns;
 	struct mountpoint *smp;
-	struct mountpoint *secondary = NULL;
+	struct mountpoint *shorter = NULL;
 	struct mount *child, *dest_mnt, *p;
 	struct mount *top;
 	struct hlist_node *n;
@@ -2655,14 +2655,12 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 	 * mounted beneath mounts on the same mountpoint.
 	 */
 	for (top = source_mnt; unlikely(top->overmount); top = top->overmount) {
-		if (!secondary && is_mnt_ns_file(top->mnt.mnt_root))
-			secondary = top->mnt_mp;
+		if (!shorter && is_mnt_ns_file(top->mnt.mnt_root))
+			shorter = top->mnt_mp;
 	}
 	smp = get_mountpoint(top->mnt.mnt_root);
 	if (IS_ERR(smp))
 		return PTR_ERR(smp);
-	if (!secondary)
-		secondary = smp;
 
 	/* Is there space to add these mounts to the mount namespace? */
 	if (!moving) {
@@ -2706,9 +2704,14 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 	}
 
 	mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
-	if (beneath)
-		mnt_change_mountpoint(top, smp, top_mnt);
-	commit_tree(source_mnt);
+	/*
+	 * Now the original copy is in the same state as the secondaries -
+	 * its root attached to mountpoint, but not hashed and all mounts
+	 * in it are either in our namespace or in no namespace at all.
+	 * Add the original to the list of copies and deal with the
+	 * rest of work for all of them uniformly.
+	 */
+	hlist_add_head(&source_mnt->mnt_hash, &tree_list);
 
 	hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
 		struct mount *q;
@@ -2719,10 +2722,13 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 		q = __lookup_mnt(&child->mnt_parent->mnt,
 				 child->mnt_mountpoint);
 		if (q) {
+			struct mountpoint *mp = smp;
 			struct mount *r = child;
 			while (unlikely(r->overmount))
 				r = r->overmount;
-			mnt_change_mountpoint(r, secondary, q);
+			if (unlikely(shorter) && child != source_mnt)
+				mp = shorter;
+			mnt_change_mountpoint(r, mp, q);
 		}
 		commit_tree(child);
 	}
-- 
2.39.5

[PATCH v2 21/35] attach_recursive_mnt(): pass destination mount in all cases

[Al Viro]

... and 'beneath' is no longer used there

Reviewed-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 17 ++++++-----------
 1 file changed, 6 insertions(+), 11 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index 6092aaf604a7..be3bfd99dc46 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -2570,7 +2570,7 @@ enum mnt_tree_flags_t {
 /**
  * attach_recursive_mnt - attach a source mount tree
  * @source_mnt: mount tree to be attached
- * @top_mnt:    mount that @source_mnt will be mounted on or mounted beneath
+ * @dest_mnt:   mount that @source_mnt will be mounted on
  * @dest_mp:    the mountpoint @source_mnt will be mounted at
  * @flags:      modify how @source_mnt is supposed to be attached
  *
@@ -2635,20 +2635,20 @@ enum mnt_tree_flags_t {
  *         Otherwise a negative error code is returned.
  */
 static int attach_recursive_mnt(struct mount *source_mnt,
-				struct mount *top_mnt,
+				struct mount *dest_mnt,
 				struct mountpoint *dest_mp,
 				enum mnt_tree_flags_t flags)
 {
 	struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
 	HLIST_HEAD(tree_list);
-	struct mnt_namespace *ns = top_mnt->mnt_ns;
+	struct mnt_namespace *ns = dest_mnt->mnt_ns;
 	struct mountpoint *smp;
 	struct mountpoint *shorter = NULL;
-	struct mount *child, *dest_mnt, *p;
+	struct mount *child, *p;
 	struct mount *top;
 	struct hlist_node *n;
 	int err = 0;
-	bool moving = flags & MNT_TREE_MOVE, beneath = flags & MNT_TREE_BENEATH;
+	bool moving = flags & MNT_TREE_MOVE;
 
 	/*
 	 * Preallocate a mountpoint in case the new mounts need to be
@@ -2669,11 +2669,6 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 			goto out;
 	}
 
-	if (beneath)
-		dest_mnt = top_mnt->mnt_parent;
-	else
-		dest_mnt = top_mnt;
-
 	if (IS_MNT_SHARED(dest_mnt)) {
 		err = invent_group_ids(source_mnt, true);
 		if (err)
@@ -3688,7 +3683,7 @@ static int do_move_mount(struct path *old_path,
 	if (mount_is_ancestor(old, p))
 		goto out;
 
-	err = attach_recursive_mnt(old, real_mount(new_path->mnt), mp, flags);
+	err = attach_recursive_mnt(old, p, mp, flags);
 	if (err)
 		goto out;
 
-- 
2.39.5

[PATCH v2 22/35] attach_recursive_mnt(): get rid of flags entirely

[Al Viro]

move vs. attach is trivially detected as mnt_has_parent(source_mnt)...

Reviewed-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 18 ++++++------------
 1 file changed, 6 insertions(+), 12 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index be3bfd99dc46..f9b320975cac 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -2562,9 +2562,8 @@ int count_mounts(struct mnt_namespace *ns, struct mount *mnt)
 }
 
 enum mnt_tree_flags_t {
-	MNT_TREE_MOVE = BIT(0),
-	MNT_TREE_BENEATH = BIT(1),
-	MNT_TREE_PROPAGATION = BIT(2),
+	MNT_TREE_BENEATH = BIT(0),
+	MNT_TREE_PROPAGATION = BIT(1),
 };
 
 /**
@@ -2572,7 +2571,6 @@ enum mnt_tree_flags_t {
  * @source_mnt: mount tree to be attached
  * @dest_mnt:   mount that @source_mnt will be mounted on
  * @dest_mp:    the mountpoint @source_mnt will be mounted at
- * @flags:      modify how @source_mnt is supposed to be attached
  *
  *  NOTE: in the table below explains the semantics when a source mount
  *  of a given type is attached to a destination mount of a given type.
@@ -2636,8 +2634,7 @@ enum mnt_tree_flags_t {
  */
 static int attach_recursive_mnt(struct mount *source_mnt,
 				struct mount *dest_mnt,
-				struct mountpoint *dest_mp,
-				enum mnt_tree_flags_t flags)
+				struct mountpoint *dest_mp)
 {
 	struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
 	HLIST_HEAD(tree_list);
@@ -2648,7 +2645,7 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 	struct mount *top;
 	struct hlist_node *n;
 	int err = 0;
-	bool moving = flags & MNT_TREE_MOVE;
+	bool moving = mnt_has_parent(source_mnt);
 
 	/*
 	 * Preallocate a mountpoint in case the new mounts need to be
@@ -2871,7 +2868,7 @@ static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
 	      d_is_dir(mnt->mnt.mnt_root))
 		return -ENOTDIR;
 
-	return attach_recursive_mnt(mnt, p, mp, 0);
+	return attach_recursive_mnt(mnt, p, mp);
 }
 
 /*
@@ -3613,8 +3610,6 @@ static int do_move_mount(struct path *old_path,
 	p = real_mount(new_path->mnt);
 	parent = old->mnt_parent;
 	attached = mnt_has_parent(old);
-	if (attached)
-		flags |= MNT_TREE_MOVE;
 	old_mp = old->mnt_mp;
 	ns = old->mnt_ns;
 
@@ -3668,7 +3663,6 @@ static int do_move_mount(struct path *old_path,
 
 		err = -EINVAL;
 		p = p->mnt_parent;
-		flags |= MNT_TREE_BENEATH;
 	}
 
 	/*
@@ -3683,7 +3677,7 @@ static int do_move_mount(struct path *old_path,
 	if (mount_is_ancestor(old, p))
 		goto out;
 
-	err = attach_recursive_mnt(old, p, mp, flags);
+	err = attach_recursive_mnt(old, p, mp);
 	if (err)
 		goto out;
 
-- 
2.39.5

[PATCH v2 23/35] do_move_mount(): take dropping the old mountpoint into attach_recursive_mnt()

[Al Viro]

... and fold it with unhash_mnt() there - there's no need to retain a reference
to old_mp beyond that point, since by then all mountpoints we were going to add
are either explicitly pinned by get_mountpoint() or have stuff already added
to them.

Reviewed-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 7 ++-----
 1 file changed, 2 insertions(+), 5 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index f9b320975cac..912c64d1dd79 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -2682,7 +2682,7 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 	}
 
 	if (moving) {
-		unhash_mnt(source_mnt);
+		umount_mnt(source_mnt);
 		mnt_notify_add(source_mnt);
 	} else {
 		if (source_mnt->mnt_ns) {
@@ -3598,7 +3598,7 @@ static int do_move_mount(struct path *old_path,
 	struct mount *p;
 	struct mount *old;
 	struct mount *parent;
-	struct mountpoint *mp, *old_mp;
+	struct mountpoint *mp;
 	int err;
 	bool attached, beneath = flags & MNT_TREE_BENEATH;
 
@@ -3610,7 +3610,6 @@ static int do_move_mount(struct path *old_path,
 	p = real_mount(new_path->mnt);
 	parent = old->mnt_parent;
 	attached = mnt_has_parent(old);
-	old_mp = old->mnt_mp;
 	ns = old->mnt_ns;
 
 	err = -EINVAL;
@@ -3684,8 +3683,6 @@ static int do_move_mount(struct path *old_path,
 	/* if the mount is moved, it should no longer be expire
 	 * automatically */
 	list_del_init(&old->mnt_expire);
-	if (attached)
-		put_mountpoint(old_mp);
 out:
 	unlock_mount(mp);
 	if (!err) {
-- 
2.39.5

[PATCH v2 24/35] do_move_mount(): get rid of 'attached' flag

[Al Viro]

'attached' serves as a proxy for "source is a subtree of our namespace
and not the entirety of anon namespace"; finish massaging it away.

Reviewed-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 13 +++++--------
 1 file changed, 5 insertions(+), 8 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index 912c64d1dd79..b6f2172aa5e1 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -3600,7 +3600,7 @@ static int do_move_mount(struct path *old_path,
 	struct mount *parent;
 	struct mountpoint *mp;
 	int err;
-	bool attached, beneath = flags & MNT_TREE_BENEATH;
+	bool beneath = flags & MNT_TREE_BENEATH;
 
 	mp = do_lock_mount(new_path, beneath);
 	if (IS_ERR(mp))
@@ -3609,7 +3609,6 @@ static int do_move_mount(struct path *old_path,
 	old = real_mount(old_path->mnt);
 	p = real_mount(new_path->mnt);
 	parent = old->mnt_parent;
-	attached = mnt_has_parent(old);
 	ns = old->mnt_ns;
 
 	err = -EINVAL;
@@ -3622,6 +3621,9 @@ static int do_move_mount(struct path *old_path,
 		/* ... and the target should be in our namespace */
 		if (!check_mnt(p))
 			goto out;
+		/* parent of the source should not be shared */
+		if (IS_MNT_SHARED(parent))
+			goto out;
 	} else {
 		/*
 		 * otherwise the source must be the root of some anon namespace.
@@ -3649,11 +3651,6 @@ static int do_move_mount(struct path *old_path,
 	if (d_is_dir(new_path->dentry) !=
 	    d_is_dir(old_path->dentry))
 		goto out;
-	/*
-	 * Don't move a mount residing in a shared parent.
-	 */
-	if (attached && IS_MNT_SHARED(parent))
-		goto out;
 
 	if (beneath) {
 		err = can_move_mount_beneath(old_path, new_path, mp);
@@ -3686,7 +3683,7 @@ static int do_move_mount(struct path *old_path,
 out:
 	unlock_mount(mp);
 	if (!err) {
-		if (attached) {
+		if (!is_anon_ns(ns)) {
 			mntput_no_expire(parent);
 		} else {
 			/* Make sure we notice when we leak mounts. */
-- 
2.39.5

[PATCH v2 25/35] attach_recursive_mnt(): remove from expiry list on move

[Al Viro]

... rather than doing that in do_move_mount().  That's the main
obstacle to moving the protection of ->mnt_expire from namespace_sem
to mount_lock (spinlock-only), which would simplify several failure
exits.

Reviewed-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 9 +++------
 1 file changed, 3 insertions(+), 6 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index b6f2172aa5e1..c4aba4e096ae 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -2684,6 +2684,9 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 	if (moving) {
 		umount_mnt(source_mnt);
 		mnt_notify_add(source_mnt);
+		/* if the mount is moved, it should no longer be expired
+		 * automatically */
+		list_del_init(&source_mnt->mnt_expire);
 	} else {
 		if (source_mnt->mnt_ns) {
 			LIST_HEAD(head);
@@ -3674,12 +3677,6 @@ static int do_move_mount(struct path *old_path,
 		goto out;
 
 	err = attach_recursive_mnt(old, p, mp);
-	if (err)
-		goto out;
-
-	/* if the mount is moved, it should no longer be expire
-	 * automatically */
-	list_del_init(&old->mnt_expire);
 out:
 	unlock_mount(mp);
 	if (!err) {
-- 
2.39.5

[PATCH v2 26/35] take ->mnt_expire handling under mount_lock [read_seqlock_excl]

[Al Viro]

Doesn't take much massage, and we no longer need to make sure that
by the time of final mntput() the victim has been removed from the
list.  Makes life safer for ->d_automount() instances...

Rules:
	* all ->mnt_expire accesses are under mount_lock.
	* insertion into the list is done by mnt_set_expiry(), and
caller (->d_automount() instance) must hold a reference to mount
in question.  It shouldn't be done more than once for a mount.
	* if a mount on an expiry list is not yet mounted, it will
be ignored by anything that walks that list.
	* if the final mntput() finds its victim still on an expiry
list (in which case it must've never been mounted - umount_tree()
would've taken it out), it will remove the victim from the list.

Reviewed-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 28 +++++++++++-----------------
 1 file changed, 11 insertions(+), 17 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index c4aba4e096ae..164b80108cc4 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -1353,13 +1353,6 @@ static struct mount *clone_mnt(struct mount *old, struct dentry *root,
 		list_add(&mnt->mnt_slave, &old->mnt_slave);
 		mnt->mnt_master = old->mnt_master;
 	}
-
-	/* stick the duplicate mount on the same expiry list
-	 * as the original if that was on one */
-	if (flag & CL_EXPIRE) {
-		if (!list_empty(&old->mnt_expire))
-			list_add(&mnt->mnt_expire, &old->mnt_expire);
-	}
 	return mnt;
 
  out_free:
@@ -1452,6 +1445,8 @@ static void mntput_no_expire(struct mount *mnt)
 	rcu_read_unlock();
 
 	list_del(&mnt->mnt_instance);
+	if (unlikely(!list_empty(&mnt->mnt_expire)))
+		list_del(&mnt->mnt_expire);
 
 	if (unlikely(!list_empty(&mnt->mnt_mounts))) {
 		struct mount *p, *tmp;
@@ -2273,6 +2268,13 @@ struct mount *copy_tree(struct mount *src_root, struct dentry *dentry,
 			lock_mount_hash();
 			if (src_mnt->mnt.mnt_flags & MNT_LOCKED)
 				dst_mnt->mnt.mnt_flags |= MNT_LOCKED;
+			if (unlikely(flag & CL_EXPIRE)) {
+				/* stick the duplicate mount on the same expiry
+				 * list as the original if that was on one */
+				if (!list_empty(&src_mnt->mnt_expire))
+					list_add(&dst_mnt->mnt_expire,
+						 &src_mnt->mnt_expire);
+			}
 			list_add_tail(&dst_mnt->mnt_list, &res->mnt_list);
 			attach_mnt(dst_mnt, dst_parent, src_parent->mnt_mp);
 			unlock_mount_hash();
@@ -3891,12 +3893,6 @@ int finish_automount(struct vfsmount *m, const struct path *path)
 	namespace_unlock();
 	inode_unlock(dentry->d_inode);
 discard:
-	/* remove m from any expiration list it may be on */
-	if (!list_empty(&mnt->mnt_expire)) {
-		namespace_lock();
-		list_del_init(&mnt->mnt_expire);
-		namespace_unlock();
-	}
 	mntput(m);
 	return err;
 }
@@ -3908,11 +3904,9 @@ int finish_automount(struct vfsmount *m, const struct path *path)
  */
 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
 {
-	namespace_lock();
-
+	read_seqlock_excl(&mount_lock);
 	list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
-
-	namespace_unlock();
+	read_sequnlock_excl(&mount_lock);
 }
 EXPORT_SYMBOL(mnt_set_expiry);
 
-- 
2.39.5

[PATCH v2 27/35] pivot_root(): reorder tree surgeries, collapse unhash_mnt() and put_mountpoint()

[Al Viro]

attach new_mnt *before* detaching root_mnt; that way we don't need to keep hold
on the mountpoint and one more pair of unhash_mnt()/put_mountpoint() gets
folded together into umount_mnt().

Reviewed-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 11 +++++------
 1 file changed, 5 insertions(+), 6 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index 164b80108cc4..834eed9d4493 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -4685,7 +4685,7 @@ SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
 {
 	struct path new, old, root;
 	struct mount *new_mnt, *root_mnt, *old_mnt, *root_parent, *ex_parent;
-	struct mountpoint *old_mp, *root_mp;
+	struct mountpoint *old_mp;
 	int error;
 
 	if (!may_mount())
@@ -4748,20 +4748,19 @@ SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
 		goto out4;
 	lock_mount_hash();
 	umount_mnt(new_mnt);
-	root_mp = unhash_mnt(root_mnt);  /* we'll need its mountpoint */
 	if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
 		new_mnt->mnt.mnt_flags |= MNT_LOCKED;
 		root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
 	}
-	/* mount old root on put_old */
-	attach_mnt(root_mnt, old_mnt, old_mp);
 	/* mount new_root on / */
-	attach_mnt(new_mnt, root_parent, root_mp);
+	attach_mnt(new_mnt, root_parent, root_mnt->mnt_mp);
+	umount_mnt(root_mnt);
 	mnt_add_count(root_parent, -1);
+	/* mount old root on put_old */
+	attach_mnt(root_mnt, old_mnt, old_mp);
 	touch_mnt_namespace(current->nsproxy->mnt_ns);
 	/* A moved mount should not expire automatically */
 	list_del_init(&new_mnt->mnt_expire);
-	put_mountpoint(root_mp);
 	unlock_mount_hash();
 	mnt_notify_add(root_mnt);
 	mnt_notify_add(new_mnt);
-- 
2.39.5

[PATCH v2 28/35] combine __put_mountpoint() with unhash_mnt()

[Al Viro]

A call of unhash_mnt() is immediately followed by passing its return
value to __put_mountpoint(); the shrink list given to __put_mountpoint()
will be ex_mountpoints when called from umount_mnt() and list when called
from mntput_no_expire().

Replace with __umount_mnt(mount, shrink_list), moving the call of
__put_mountpoint() into it (and returning nothing), adjust the
callers.

Reviewed-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index 834eed9d4493..5a18ba6e7df2 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -1037,9 +1037,9 @@ static void __touch_mnt_namespace(struct mnt_namespace *ns)
 }
 
 /*
- * vfsmount lock must be held for write
+ * locks: mount_lock[write_seqlock]
  */
-static struct mountpoint *unhash_mnt(struct mount *mnt)
+static void __umount_mnt(struct mount *mnt, struct list_head *shrink_list)
 {
 	struct mountpoint *mp;
 	struct mount *parent = mnt->mnt_parent;
@@ -1052,15 +1052,15 @@ static struct mountpoint *unhash_mnt(struct mount *mnt)
 	hlist_del_init(&mnt->mnt_mp_list);
 	mp = mnt->mnt_mp;
 	mnt->mnt_mp = NULL;
-	return mp;
+	__put_mountpoint(mp, shrink_list);
 }
 
 /*
- * vfsmount lock must be held for write
+ * locks: mount_lock[write_seqlock], namespace_sem[excl] (for ex_mountpoints)
  */
 static void umount_mnt(struct mount *mnt)
 {
-	put_mountpoint(unhash_mnt(mnt));
+	__umount_mnt(mnt, &ex_mountpoints);
 }
 
 /*
@@ -1451,7 +1451,7 @@ static void mntput_no_expire(struct mount *mnt)
 	if (unlikely(!list_empty(&mnt->mnt_mounts))) {
 		struct mount *p, *tmp;
 		list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts,  mnt_child) {
-			__put_mountpoint(unhash_mnt(p), &list);
+			__umount_mnt(p, &list);
 			hlist_add_head(&p->mnt_umount, &mnt->mnt_stuck_children);
 		}
 	}
-- 
2.39.5

[PATCH v2 29/35] get rid of mountpoint->m_count

[Al Viro]

struct mountpoint has an odd kinda-sorta refcount in it.  It's always
either equal to or one above the number of mounts attached to that
mountpoint.

"One above" happens when a function takes a temporary reference to
mountpoint.  Things get simpler if we express that as inserting
a local object into ->m_list and removing it to drop the reference.

New calling conventions:

1) lock_mount(), do_lock_mount(), get_mountpoint() and lookup_mountpoint()
take an extra struct pinned_mountpoint * argument and returns 0/-E...
(or true/false in case of lookup_mountpoint()) instead of returning
struct mountpoint pointers.  In case of success, the struct mountpoint *
we used to get can be found as pinned_mountpoint.mp

2) unlock_mount() (always paired with lock_mount()/do_lock_mount()) takes
an address of struct pinned_mountpoint - the same that had been passed to
lock_mount()/do_lock_mount().

3) put_mountpoint() for a temporary reference (paired with get_mountpoint()
or lookup_mountpoint()) is replaced with unpin_mountpoint(), which takes
the address of pinned_mountpoint we passed to matching {get,lookup}_mountpoint().

4) all instances of pinned_mountpoint are local variables; they always live on
stack.  {} is used for initializer, after successful {get,lookup}_mountpoint()
we must make sure to call unpin_mountpoint() before leaving the scope and
after successful {do_,}lock_mount() we must make sure to call unlock_mount()
before leaving the scope.

5) all manipulations of ->m_count are gone, along with ->m_count itself.
struct mountpoint lives while its ->m_list is non-empty.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/mount.h     |   1 -
 fs/namespace.c | 186 ++++++++++++++++++++++++-------------------------
 2 files changed, 92 insertions(+), 95 deletions(-)

diff --git a/fs/mount.h b/fs/mount.h
index fb93d3e16724..4355c482a841 100644
--- a/fs/mount.h
+++ b/fs/mount.h
@@ -44,7 +44,6 @@ struct mountpoint {
 	struct hlist_node m_hash;
 	struct dentry *m_dentry;
 	struct hlist_head m_list;
-	int m_count;
 };
 
 struct mount {
diff --git a/fs/namespace.c b/fs/namespace.c
index 5a18ba6e7df2..debc43282b26 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -910,42 +910,48 @@ bool __is_local_mountpoint(const struct dentry *dentry)
 	return is_covered;
 }
 
-static struct mountpoint *lookup_mountpoint(struct dentry *dentry)
+struct pinned_mountpoint {
+	struct hlist_node node;
+	struct mountpoint *mp;
+};
+
+static bool lookup_mountpoint(struct dentry *dentry, struct pinned_mountpoint *m)
 {
 	struct hlist_head *chain = mp_hash(dentry);
 	struct mountpoint *mp;
 
 	hlist_for_each_entry(mp, chain, m_hash) {
 		if (mp->m_dentry == dentry) {
-			mp->m_count++;
-			return mp;
+			hlist_add_head(&m->node, &mp->m_list);
+			m->mp = mp;
+			return true;
 		}
 	}
-	return NULL;
+	return false;
 }
 
-static struct mountpoint *get_mountpoint(struct dentry *dentry)
+static int get_mountpoint(struct dentry *dentry, struct pinned_mountpoint *m)
 {
-	struct mountpoint *mp, *new = NULL;
+	struct mountpoint *mp __free(kfree) = NULL;
+	bool found;
 	int ret;
 
 	if (d_mountpoint(dentry)) {
 		/* might be worth a WARN_ON() */
 		if (d_unlinked(dentry))
-			return ERR_PTR(-ENOENT);
+			return -ENOENT;
 mountpoint:
 		read_seqlock_excl(&mount_lock);
-		mp = lookup_mountpoint(dentry);
+		found = lookup_mountpoint(dentry, m);
 		read_sequnlock_excl(&mount_lock);
-		if (mp)
-			goto done;
+		if (found)
+			return 0;
 	}
 
-	if (!new)
-		new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
-	if (!new)
-		return ERR_PTR(-ENOMEM);
-
+	if (!mp)
+		mp = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
+	if (!mp)
+		return -ENOMEM;
 
 	/* Exactly one processes may set d_mounted */
 	ret = d_set_mounted(dentry);
@@ -955,34 +961,28 @@ static struct mountpoint *get_mountpoint(struct dentry *dentry)
 		goto mountpoint;
 
 	/* The dentry is not available as a mountpoint? */
-	mp = ERR_PTR(ret);
 	if (ret)
-		goto done;
+		return ret;
 
 	/* Add the new mountpoint to the hash table */
 	read_seqlock_excl(&mount_lock);
-	new->m_dentry = dget(dentry);
-	new->m_count = 1;
-	hlist_add_head(&new->m_hash, mp_hash(dentry));
-	INIT_HLIST_HEAD(&new->m_list);
+	mp->m_dentry = dget(dentry);
+	hlist_add_head(&mp->m_hash, mp_hash(dentry));
+	INIT_HLIST_HEAD(&mp->m_list);
+	hlist_add_head(&m->node, &mp->m_list);
+	m->mp = no_free_ptr(mp);
 	read_sequnlock_excl(&mount_lock);
-
-	mp = new;
-	new = NULL;
-done:
-	kfree(new);
-	return mp;
+	return 0;
 }
 
 /*
  * vfsmount lock must be held.  Additionally, the caller is responsible
  * for serializing calls for given disposal list.
  */
-static void __put_mountpoint(struct mountpoint *mp, struct list_head *list)
+static void maybe_free_mountpoint(struct mountpoint *mp, struct list_head *list)
 {
-	if (!--mp->m_count) {
+	if (hlist_empty(&mp->m_list)) {
 		struct dentry *dentry = mp->m_dentry;
-		BUG_ON(!hlist_empty(&mp->m_list));
 		spin_lock(&dentry->d_lock);
 		dentry->d_flags &= ~DCACHE_MOUNTED;
 		spin_unlock(&dentry->d_lock);
@@ -992,10 +992,15 @@ static void __put_mountpoint(struct mountpoint *mp, struct list_head *list)
 	}
 }
 
-/* called with namespace_lock and vfsmount lock */
-static void put_mountpoint(struct mountpoint *mp)
+/*
+ * locks: mount_lock [read_seqlock_excl], namespace_sem [excl]
+ */
+static void unpin_mountpoint(struct pinned_mountpoint *m)
 {
-	__put_mountpoint(mp, &ex_mountpoints);
+	if (m->mp) {
+		hlist_del(&m->node);
+		maybe_free_mountpoint(m->mp, &ex_mountpoints);
+	}
 }
 
 static inline int check_mnt(struct mount *mnt)
@@ -1052,7 +1057,7 @@ static void __umount_mnt(struct mount *mnt, struct list_head *shrink_list)
 	hlist_del_init(&mnt->mnt_mp_list);
 	mp = mnt->mnt_mp;
 	mnt->mnt_mp = NULL;
-	__put_mountpoint(mp, shrink_list);
+	maybe_free_mountpoint(mp, shrink_list);
 }
 
 /*
@@ -1070,7 +1075,6 @@ void mnt_set_mountpoint(struct mount *mnt,
 			struct mountpoint *mp,
 			struct mount *child_mnt)
 {
-	mp->m_count++;
 	mnt_add_count(mnt, 1);	/* essentially, that's mntget */
 	child_mnt->mnt_mountpoint = mp->m_dentry;
 	child_mnt->mnt_parent = mnt;
@@ -1122,7 +1126,7 @@ void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct m
 
 	attach_mnt(mnt, parent, mp);
 
-	put_mountpoint(old_mp);
+	maybe_free_mountpoint(old_mp, &ex_mountpoints);
 	mnt_add_count(old_parent, -1);
 }
 
@@ -2030,25 +2034,24 @@ static int do_umount(struct mount *mnt, int flags)
  */
 void __detach_mounts(struct dentry *dentry)
 {
-	struct mountpoint *mp;
+	struct pinned_mountpoint mp = {};
 	struct mount *mnt;
 
 	namespace_lock();
 	lock_mount_hash();
-	mp = lookup_mountpoint(dentry);
-	if (!mp)
+	if (!lookup_mountpoint(dentry, &mp))
 		goto out_unlock;
 
 	event++;
-	while (!hlist_empty(&mp->m_list)) {
-		mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list);
+	while (mp.node.next) {
+		mnt = hlist_entry(mp.node.next, struct mount, mnt_mp_list);
 		if (mnt->mnt.mnt_flags & MNT_UMOUNT) {
 			umount_mnt(mnt);
 			hlist_add_head(&mnt->mnt_umount, &unmounted);
 		}
 		else umount_tree(mnt, UMOUNT_CONNECTED);
 	}
-	put_mountpoint(mp);
+	unpin_mountpoint(&mp);
 out_unlock:
 	unlock_mount_hash();
 	namespace_unlock();
@@ -2641,7 +2644,7 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 	struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
 	HLIST_HEAD(tree_list);
 	struct mnt_namespace *ns = dest_mnt->mnt_ns;
-	struct mountpoint *smp;
+	struct pinned_mountpoint root = {};
 	struct mountpoint *shorter = NULL;
 	struct mount *child, *p;
 	struct mount *top;
@@ -2657,9 +2660,9 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 		if (!shorter && is_mnt_ns_file(top->mnt.mnt_root))
 			shorter = top->mnt_mp;
 	}
-	smp = get_mountpoint(top->mnt.mnt_root);
-	if (IS_ERR(smp))
-		return PTR_ERR(smp);
+	err = get_mountpoint(top->mnt.mnt_root, &root);
+	if (err)
+		return err;
 
 	/* Is there space to add these mounts to the mount namespace? */
 	if (!moving) {
@@ -2719,7 +2722,7 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 		q = __lookup_mnt(&child->mnt_parent->mnt,
 				 child->mnt_mountpoint);
 		if (q) {
-			struct mountpoint *mp = smp;
+			struct mountpoint *mp = root.mp;
 			struct mount *r = child;
 			while (unlikely(r->overmount))
 				r = r->overmount;
@@ -2729,7 +2732,7 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 		}
 		commit_tree(child);
 	}
-	put_mountpoint(smp);
+	unpin_mountpoint(&root);
 	unlock_mount_hash();
 
 	return 0;
@@ -2746,7 +2749,7 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 	ns->pending_mounts = 0;
 
 	read_seqlock_excl(&mount_lock);
-	put_mountpoint(smp);
+	unpin_mountpoint(&root);
 	read_sequnlock_excl(&mount_lock);
 
 	return err;
@@ -2786,12 +2789,12 @@ static int attach_recursive_mnt(struct mount *source_mnt,
  * Return: Either the target mountpoint on the top mount or the top
  *         mount's mountpoint.
  */
-static struct mountpoint *do_lock_mount(struct path *path, bool beneath)
+static int do_lock_mount(struct path *path, struct pinned_mountpoint *pinned, bool beneath)
 {
 	struct vfsmount *mnt = path->mnt;
 	struct dentry *dentry;
-	struct mountpoint *mp = ERR_PTR(-ENOENT);
 	struct path under = {};
+	int err = -ENOENT;
 
 	for (;;) {
 		struct mount *m = real_mount(mnt);
@@ -2829,8 +2832,8 @@ static struct mountpoint *do_lock_mount(struct path *path, bool beneath)
 			path->dentry = dget(mnt->mnt_root);
 			continue;	// got overmounted
 		}
-		mp = get_mountpoint(dentry);
-		if (IS_ERR(mp))
+		err = get_mountpoint(dentry, pinned);
+		if (err)
 			break;
 		if (beneath) {
 			/*
@@ -2841,25 +2844,25 @@ static struct mountpoint *do_lock_mount(struct path *path, bool beneath)
 			 */
 			path_put(&under);
 		}
-		return mp;
+		return 0;
 	}
 	namespace_unlock();
 	inode_unlock(dentry->d_inode);
 	if (beneath)
 		path_put(&under);
-	return mp;
+	return err;
 }
 
-static inline struct mountpoint *lock_mount(struct path *path)
+static inline int lock_mount(struct path *path, struct pinned_mountpoint *m)
 {
-	return do_lock_mount(path, false);
+	return do_lock_mount(path, m, false);
 }
 
-static void unlock_mount(struct mountpoint *where)
+static void unlock_mount(struct pinned_mountpoint *m)
 {
-	inode_unlock(where->m_dentry->d_inode);
+	inode_unlock(m->mp->m_dentry->d_inode);
 	read_seqlock_excl(&mount_lock);
-	put_mountpoint(where);
+	unpin_mountpoint(m);
 	read_sequnlock_excl(&mount_lock);
 	namespace_unlock();
 }
@@ -3024,7 +3027,7 @@ static int do_loopback(struct path *path, const char *old_name,
 {
 	struct path old_path;
 	struct mount *mnt = NULL, *parent;
-	struct mountpoint *mp;
+	struct pinned_mountpoint mp = {};
 	int err;
 	if (!old_name || !*old_name)
 		return -EINVAL;
@@ -3036,11 +3039,9 @@ static int do_loopback(struct path *path, const char *old_name,
 	if (mnt_ns_loop(old_path.dentry))
 		goto out;
 
-	mp = lock_mount(path);
-	if (IS_ERR(mp)) {
-		err = PTR_ERR(mp);
+	err = lock_mount(path, &mp);
+	if (err)
 		goto out;
-	}
 
 	parent = real_mount(path->mnt);
 	if (!check_mnt(parent))
@@ -3052,14 +3053,14 @@ static int do_loopback(struct path *path, const char *old_name,
 		goto out2;
 	}
 
-	err = graft_tree(mnt, parent, mp);
+	err = graft_tree(mnt, parent, mp.mp);
 	if (err) {
 		lock_mount_hash();
 		umount_tree(mnt, UMOUNT_SYNC);
 		unlock_mount_hash();
 	}
 out2:
-	unlock_mount(mp);
+	unlock_mount(&mp);
 out:
 	path_put(&old_path);
 	return err;
@@ -3603,13 +3604,13 @@ static int do_move_mount(struct path *old_path,
 	struct mount *p;
 	struct mount *old;
 	struct mount *parent;
-	struct mountpoint *mp;
+	struct pinned_mountpoint mp;
 	int err;
 	bool beneath = flags & MNT_TREE_BENEATH;
 
-	mp = do_lock_mount(new_path, beneath);
-	if (IS_ERR(mp))
-		return PTR_ERR(mp);
+	err = do_lock_mount(new_path, &mp, beneath);
+	if (err)
+		return err;
 
 	old = real_mount(old_path->mnt);
 	p = real_mount(new_path->mnt);
@@ -3658,7 +3659,7 @@ static int do_move_mount(struct path *old_path,
 		goto out;
 
 	if (beneath) {
-		err = can_move_mount_beneath(old_path, new_path, mp);
+		err = can_move_mount_beneath(old_path, new_path, mp.mp);
 		if (err)
 			goto out;
 
@@ -3678,9 +3679,9 @@ static int do_move_mount(struct path *old_path,
 	if (mount_is_ancestor(old, p))
 		goto out;
 
-	err = attach_recursive_mnt(old, p, mp);
+	err = attach_recursive_mnt(old, p, mp.mp);
 out:
-	unlock_mount(mp);
+	unlock_mount(&mp);
 	if (!err) {
 		if (!is_anon_ns(ns)) {
 			mntput_no_expire(parent);
@@ -3750,7 +3751,7 @@ static int do_new_mount_fc(struct fs_context *fc, struct path *mountpoint,
 			   unsigned int mnt_flags)
 {
 	struct vfsmount *mnt;
-	struct mountpoint *mp;
+	struct pinned_mountpoint mp = {};
 	struct super_block *sb = fc->root->d_sb;
 	int error;
 
@@ -3771,13 +3772,12 @@ static int do_new_mount_fc(struct fs_context *fc, struct path *mountpoint,
 
 	mnt_warn_timestamp_expiry(mountpoint, mnt);
 
-	mp = lock_mount(mountpoint);
-	if (IS_ERR(mp)) {
-		mntput(mnt);
-		return PTR_ERR(mp);
+	error = lock_mount(mountpoint, &mp);
+	if (!error) {
+		error = do_add_mount(real_mount(mnt), mp.mp,
+				     mountpoint, mnt_flags);
+		unlock_mount(&mp);
 	}
-	error = do_add_mount(real_mount(mnt), mp, mountpoint, mnt_flags);
-	unlock_mount(mp);
 	if (error < 0)
 		mntput(mnt);
 	return error;
@@ -3845,7 +3845,7 @@ static int do_new_mount(struct path *path, const char *fstype, int sb_flags,
 int finish_automount(struct vfsmount *m, const struct path *path)
 {
 	struct dentry *dentry = path->dentry;
-	struct mountpoint *mp;
+	struct pinned_mountpoint mp = {};
 	struct mount *mnt;
 	int err;
 
@@ -3877,14 +3877,13 @@ int finish_automount(struct vfsmount *m, const struct path *path)
 		err = 0;
 		goto discard_locked;
 	}
-	mp = get_mountpoint(dentry);
-	if (IS_ERR(mp)) {
-		err = PTR_ERR(mp);
+	err = get_mountpoint(dentry, &mp);
+	if (err)
 		goto discard_locked;
-	}
 
-	err = do_add_mount(mnt, mp, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
-	unlock_mount(mp);
+	err = do_add_mount(mnt, mp.mp, path,
+			   path->mnt->mnt_flags | MNT_SHRINKABLE);
+	unlock_mount(&mp);
 	if (unlikely(err))
 		goto discard;
 	return 0;
@@ -4685,7 +4684,7 @@ SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
 {
 	struct path new, old, root;
 	struct mount *new_mnt, *root_mnt, *old_mnt, *root_parent, *ex_parent;
-	struct mountpoint *old_mp;
+	struct pinned_mountpoint old_mp = {};
 	int error;
 
 	if (!may_mount())
@@ -4706,9 +4705,8 @@ SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
 		goto out2;
 
 	get_fs_root(current->fs, &root);
-	old_mp = lock_mount(&old);
-	error = PTR_ERR(old_mp);
-	if (IS_ERR(old_mp))
+	error = lock_mount(&old, &old_mp);
+	if (error)
 		goto out3;
 
 	error = -EINVAL;
@@ -4757,7 +4755,7 @@ SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
 	umount_mnt(root_mnt);
 	mnt_add_count(root_parent, -1);
 	/* mount old root on put_old */
-	attach_mnt(root_mnt, old_mnt, old_mp);
+	attach_mnt(root_mnt, old_mnt, old_mp.mp);
 	touch_mnt_namespace(current->nsproxy->mnt_ns);
 	/* A moved mount should not expire automatically */
 	list_del_init(&new_mnt->mnt_expire);
@@ -4767,7 +4765,7 @@ SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
 	chroot_fs_refs(&root, &new);
 	error = 0;
 out4:
-	unlock_mount(old_mp);
+	unlock_mount(&old_mp);
 	if (!error)
 		mntput_no_expire(ex_parent);
 out3:
-- 
2.39.5

[PATCH v2 30/35] don't have mounts pin their parents

[Al Viro]

Simplify the rules for mount refcounts.  Current rules include:
	* being a namespace root => +1
	* being someone's child => +1
	* being someone's child => +1 to parent's refcount, unless you've
				   already been through umount_tree().

The last part is not needed at all.  It makes for more places where need
to decrement refcounts and it creates an asymmetry between the situations
for something that has never been a part of a namespace and something that
left one, both for no good reason.

If mount's refcount has additions from its children, we know that
	* it's either someone's child itself (and will remain so
until umount_tree(), at which point contributions from children
will disappear), or
	* or is the root of namespace (and will remain such until
it either becomes someone's child in another namespace or goes through
umount_tree()), or
	* it is the root of some tree copy, and is currently pinned
by the caller of copy_tree() (and remains such until it either gets
into namespace, or goes to umount_tree()).
In all cases we already have contribution(s) to refcount that will last
as long as the contribution from children remains.  In other words, the
lifetime is not affected by refcount contributions from children.

It might be useful for "is it busy" checks, but those are actually
no harder to express without it.

NB: propagate_mnt_busy() part is an equivalent transformation, ugly as it
is; the current logics is actually wrong and may give false negatives,
but fixing that is for a separate patch (probably earlier in the queue).

Reviewed-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 31 +++++++++----------------------
 fs/pnode.c     | 49 +++++++++++++++++--------------------------------
 2 files changed, 26 insertions(+), 54 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index debc43282b26..4b123e2384ca 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -1075,7 +1075,6 @@ void mnt_set_mountpoint(struct mount *mnt,
 			struct mountpoint *mp,
 			struct mount *child_mnt)
 {
-	mnt_add_count(mnt, 1);	/* essentially, that's mntget */
 	child_mnt->mnt_mountpoint = mp->m_dentry;
 	child_mnt->mnt_parent = mnt;
 	child_mnt->mnt_mp = mp;
@@ -1118,7 +1117,6 @@ static void attach_mnt(struct mount *mnt, struct mount *parent,
 void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt)
 {
 	struct mountpoint *old_mp = mnt->mnt_mp;
-	struct mount *old_parent = mnt->mnt_parent;
 
 	list_del_init(&mnt->mnt_child);
 	hlist_del_init(&mnt->mnt_mp_list);
@@ -1127,7 +1125,6 @@ void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct m
 	attach_mnt(mnt, parent, mp);
 
 	maybe_free_mountpoint(old_mp, &ex_mountpoints);
-	mnt_add_count(old_parent, -1);
 }
 
 static inline struct mount *node_to_mount(struct rb_node *node)
@@ -1652,23 +1649,19 @@ const struct seq_operations mounts_op = {
 int may_umount_tree(struct vfsmount *m)
 {
 	struct mount *mnt = real_mount(m);
-	int actual_refs = 0;
-	int minimum_refs = 0;
-	struct mount *p;
-	BUG_ON(!m);
+	bool busy = false;
 
 	/* write lock needed for mnt_get_count */
 	lock_mount_hash();
-	for (p = mnt; p; p = next_mnt(p, mnt)) {
-		actual_refs += mnt_get_count(p);
-		minimum_refs += 2;
+	for (struct mount *p = mnt; p; p = next_mnt(p, mnt)) {
+		if (mnt_get_count(p) > (p == mnt ? 2 : 1)) {
+			busy = true;
+			break;
+		}
 	}
 	unlock_mount_hash();
 
-	if (actual_refs > minimum_refs)
-		return 0;
-
-	return 1;
+	return !busy;
 }
 
 EXPORT_SYMBOL(may_umount_tree);
@@ -1869,7 +1862,6 @@ static void umount_tree(struct mount *mnt, enum umount_tree_flags how)
 
 		disconnect = disconnect_mount(p, how);
 		if (mnt_has_parent(p)) {
-			mnt_add_count(p->mnt_parent, -1);
 			if (!disconnect) {
 				/* Don't forget about p */
 				list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts);
@@ -1946,7 +1938,7 @@ static int do_umount(struct mount *mnt, int flags)
 		 * all race cases, but it's a slowpath.
 		 */
 		lock_mount_hash();
-		if (mnt_get_count(mnt) != 2) {
+		if (!list_empty(&mnt->mnt_mounts) || mnt_get_count(mnt) != 2) {
 			unlock_mount_hash();
 			return -EBUSY;
 		}
@@ -3683,9 +3675,7 @@ static int do_move_mount(struct path *old_path,
 out:
 	unlock_mount(&mp);
 	if (!err) {
-		if (!is_anon_ns(ns)) {
-			mntput_no_expire(parent);
-		} else {
+		if (is_anon_ns(ns)) {
 			/* Make sure we notice when we leak mounts. */
 			VFS_WARN_ON_ONCE(!mnt_ns_empty(ns));
 			free_mnt_ns(ns);
@@ -4753,7 +4743,6 @@ SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
 	/* mount new_root on / */
 	attach_mnt(new_mnt, root_parent, root_mnt->mnt_mp);
 	umount_mnt(root_mnt);
-	mnt_add_count(root_parent, -1);
 	/* mount old root on put_old */
 	attach_mnt(root_mnt, old_mnt, old_mp.mp);
 	touch_mnt_namespace(current->nsproxy->mnt_ns);
@@ -4766,8 +4755,6 @@ SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
 	error = 0;
 out4:
 	unlock_mount(&old_mp);
-	if (!error)
-		mntput_no_expire(ex_parent);
 out3:
 	path_put(&root);
 out2:
diff --git a/fs/pnode.c b/fs/pnode.c
index 9aa892263a38..73a64c55deb3 100644
--- a/fs/pnode.c
+++ b/fs/pnode.c
@@ -332,21 +332,6 @@ int propagate_mnt(struct mount *dest_mnt, struct mountpoint *dest_mp,
 	return ret;
 }
 
-static struct mount *find_topper(struct mount *mnt)
-{
-	/* If there is exactly one mount covering mnt completely return it. */
-	struct mount *child;
-
-	if (!list_is_singular(&mnt->mnt_mounts))
-		return NULL;
-
-	child = list_first_entry(&mnt->mnt_mounts, struct mount, mnt_child);
-	if (child->mnt_mountpoint != mnt->mnt.mnt_root)
-		return NULL;
-
-	return child;
-}
-
 /*
  * return true if the refcount is greater than count
  */
@@ -404,12 +389,8 @@ bool propagation_would_overmount(const struct mount *from,
  */
 int propagate_mount_busy(struct mount *mnt, int refcnt)
 {
-	struct mount *m, *child, *topper;
 	struct mount *parent = mnt->mnt_parent;
 
-	if (mnt == parent)
-		return do_refcount_check(mnt, refcnt);
-
 	/*
 	 * quickly check if the current mount can be unmounted.
 	 * If not, we don't have to go checking for all other
@@ -418,23 +399,27 @@ int propagate_mount_busy(struct mount *mnt, int refcnt)
 	if (!list_empty(&mnt->mnt_mounts) || do_refcount_check(mnt, refcnt))
 		return 1;
 
-	for (m = propagation_next(parent, parent); m;
+	if (mnt == parent)
+		return 0;
+
+	for (struct mount *m = propagation_next(parent, parent); m;
 	     		m = propagation_next(m, parent)) {
-		int count = 1;
-		child = __lookup_mnt(&m->mnt, mnt->mnt_mountpoint);
-		if (!child)
-			continue;
+		struct list_head *head;
+		struct mount *child = __lookup_mnt(&m->mnt, mnt->mnt_mountpoint);
 
-		/* Is there exactly one mount on the child that covers
-		 * it completely whose reference should be ignored?
-		 */
-		topper = find_topper(child);
-		if (topper)
-			count += 1;
-		else if (!list_empty(&child->mnt_mounts))
+		if (!child)
 			continue;
 
-		if (do_refcount_check(child, count))
+		head = &child->mnt_mounts;
+		if (!list_empty(head)) {
+			/*
+			 * a mount that covers child completely wouldn't prevent
+			 * it being pulled out; any other would.
+			 */
+			if (!list_is_singular(head) || !child->overmount)
+				continue;
+		}
+		if (do_refcount_check(child, 1))
 			return 1;
 	}
 	return 0;
-- 
2.39.5

[PATCH v2 31/35] copy_tree(): don't link the mounts via mnt_list

[Al Viro]

The only place that really needs to be adjusted is commit_tree() -
there we need to iterate through the copy and we might as well
use next_mnt() for that.  However, in case when our tree has been
slid under something already mounted (propagation to a mountpoint
that already has something mounted on it or a 'beneath' move_mount)
we need to take care not to walk into the overmounting tree.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/mount.h     |  3 +--
 fs/namespace.c | 60 ++++++++++++++++++++------------------------------
 fs/pnode.c     |  3 ++-
 3 files changed, 27 insertions(+), 39 deletions(-)

diff --git a/fs/mount.h b/fs/mount.h
index 4355c482a841..c5b170b6cb3c 100644
--- a/fs/mount.h
+++ b/fs/mount.h
@@ -176,7 +176,7 @@ static inline bool mnt_ns_empty(const struct mnt_namespace *ns)
 	return RB_EMPTY_ROOT(&ns->mounts);
 }
 
-static inline void move_from_ns(struct mount *mnt, struct list_head *dt_list)
+static inline void move_from_ns(struct mount *mnt)
 {
 	struct mnt_namespace *ns = mnt->mnt_ns;
 	WARN_ON(!mnt_ns_attached(mnt));
@@ -186,7 +186,6 @@ static inline void move_from_ns(struct mount *mnt, struct list_head *dt_list)
 		ns->mnt_first_node = rb_next(&mnt->mnt_node);
 	rb_erase(&mnt->mnt_node, &ns->mounts);
 	RB_CLEAR_NODE(&mnt->mnt_node);
-	list_add_tail(&mnt->mnt_list, dt_list);
 }
 
 bool has_locked_children(struct mount *mnt, struct dentry *dentry);
diff --git a/fs/namespace.c b/fs/namespace.c
index 4b123e2384ca..5556c5edbae9 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -1161,34 +1161,6 @@ static void mnt_add_to_ns(struct mnt_namespace *ns, struct mount *mnt)
 	mnt_notify_add(mnt);
 }
 
-/*
- * vfsmount lock must be held for write
- */
-static void commit_tree(struct mount *mnt)
-{
-	struct mount *parent = mnt->mnt_parent;
-	struct mount *m;
-	LIST_HEAD(head);
-	struct mnt_namespace *n = parent->mnt_ns;
-
-	BUG_ON(parent == mnt);
-
-	if (!mnt_ns_attached(mnt)) {
-		list_add_tail(&head, &mnt->mnt_list);
-		while (!list_empty(&head)) {
-			m = list_first_entry(&head, typeof(*m), mnt_list);
-			list_del(&m->mnt_list);
-
-			mnt_add_to_ns(n, m);
-		}
-		n->nr_mounts += n->pending_mounts;
-		n->pending_mounts = 0;
-	}
-
-	make_visible(mnt);
-	touch_mnt_namespace(n);
-}
-
 static struct mount *next_mnt(struct mount *p, struct mount *root)
 {
 	struct list_head *next = p->mnt_mounts.next;
@@ -1215,6 +1187,27 @@ static struct mount *skip_mnt_tree(struct mount *p)
 	return p;
 }
 
+/*
+ * vfsmount lock must be held for write
+ */
+static void commit_tree(struct mount *mnt)
+{
+	struct mnt_namespace *n = mnt->mnt_parent->mnt_ns;
+
+	if (!mnt_ns_attached(mnt)) {
+		for (struct mount *m = mnt; m; m = next_mnt(m, mnt))
+			if (unlikely(mnt_ns_attached(m)))
+				m = skip_mnt_tree(m);
+			else
+				mnt_add_to_ns(n, m);
+		n->nr_mounts += n->pending_mounts;
+		n->pending_mounts = 0;
+	}
+
+	make_visible(mnt);
+	touch_mnt_namespace(n);
+}
+
 /**
  * vfs_create_mount - Create a mount for a configured superblock
  * @fc: The configuration context with the superblock attached
@@ -1831,9 +1824,8 @@ static void umount_tree(struct mount *mnt, enum umount_tree_flags how)
 	for (p = mnt; p; p = next_mnt(p, mnt)) {
 		p->mnt.mnt_flags |= MNT_UMOUNT;
 		if (mnt_ns_attached(p))
-			move_from_ns(p, &tmp_list);
-		else
-			list_move(&p->mnt_list, &tmp_list);
+			move_from_ns(p);
+		list_add_tail(&p->mnt_list, &tmp_list);
 	}
 
 	/* Hide the mounts from mnt_mounts */
@@ -2270,7 +2262,6 @@ struct mount *copy_tree(struct mount *src_root, struct dentry *dentry,
 					list_add(&dst_mnt->mnt_expire,
 						 &src_mnt->mnt_expire);
 			}
-			list_add_tail(&dst_mnt->mnt_list, &res->mnt_list);
 			attach_mnt(dst_mnt, dst_parent, src_parent->mnt_mp);
 			unlock_mount_hash();
 		}
@@ -2686,12 +2677,9 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 		list_del_init(&source_mnt->mnt_expire);
 	} else {
 		if (source_mnt->mnt_ns) {
-			LIST_HEAD(head);
-
 			/* move from anon - the caller will destroy */
 			for (p = source_mnt; p; p = next_mnt(p, source_mnt))
-				move_from_ns(p, &head);
-			list_del_init(&head);
+				move_from_ns(p);
 		}
 	}
 
diff --git a/fs/pnode.c b/fs/pnode.c
index 73a64c55deb3..f897a501bee7 100644
--- a/fs/pnode.c
+++ b/fs/pnode.c
@@ -459,7 +459,8 @@ static void umount_one(struct mount *m, struct list_head *to_umount)
 {
 	m->mnt.mnt_flags |= MNT_UMOUNT;
 	list_del_init(&m->mnt_child);
-	move_from_ns(m, to_umount);
+	move_from_ns(m);
+	list_add_tail(&m->mnt_list, to_umount);
 }
 
 static void remove_from_candidate_list(struct mount *m)
-- 
2.39.5

[PATCH v2 32/35] mount: separate the flags accessed only under namespace_sem

[Al Viro]

Several flags are updated and checked only under namespace_sem; we are
already making use of that when we are checking them without mount_lock,
but we have to hold mount_lock for all updates, which makes things
clumsier than they have to be.

Take MNT_SHARED, MNT_UNBINDABLE, MNT_MARKED and MNT_UMOUNT_CANDIDATE
into a separate field (->mnt_t_flags), renaming them to T_SHARED,
etc. to avoid confusion.  All accesses must be under namespace_sem.

That changes locking requirements for mnt_change_propagation() and
set_mnt_shared() - only namespace_sem is needed now.  The same goes
for SET_MNT_MARKED et.al.

There might be more flags moved from ->mnt_flags to that field;
this is just the initial set.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 .../filesystems/propagate_umount.txt          | 12 +++++-----
 fs/mount.h                                    | 17 ++++++++++++++
 fs/namespace.c                                |  4 ----
 fs/pnode.c                                    | 22 +++++++++----------
 fs/pnode.h                                    | 19 +++++++++-------
 include/linux/mount.h                         | 18 ++-------------
 6 files changed, 46 insertions(+), 46 deletions(-)

diff --git a/Documentation/filesystems/propagate_umount.txt b/Documentation/filesystems/propagate_umount.txt
index 5b48540b4059..d665144e9e6e 100644
--- a/Documentation/filesystems/propagate_umount.txt
+++ b/Documentation/filesystems/propagate_umount.txt
@@ -453,11 +453,11 @@ original set.
 So let's go for
 	* original set ("set").  Linkage via mnt_list
 	* undecided candidates ("candidates").  Subset of a list,
-consisting of all its elements marked with a new flag (MNT_UMOUNT_CANDIDATE).
+consisting of all its elements marked with a new flag (T_UMOUNT_CANDIDATE).
 Initially all elements of the list will be marked that way; in the
 end the list will become empty and no mounts will remain marked with
 that flag.
-	* Reuse MNT_MARKED for "has been already seen by trim_ancestors()".
+	* Reuse T_MARKED for "has been already seen by trim_ancestors()".
 	* anything in U that hadn't been in the original set - elements of
 candidates will gradually be either discarded or moved there.  In other
 words, it's the candidates we have already decided to unmount.	Its role
@@ -465,13 +465,13 @@ is reasonably close to the old "to_umount", so let's use that name.
 Linkage via mnt_list.
 
 For gather_candidates() we'll need to maintain both candidates (S -
-set) and intersection of S with set.  Use MNT_UMOUNT_CANDIDATE for
+set) and intersection of S with set.  Use T_UMOUNT_CANDIDATE for
 all elements we encounter, putting the ones not already in the original
 set into the list of candidates.  When we are done, strip that flag from
 all elements of the original set.  That gives a cheap way to check
 if element belongs to S (in gather_candidates) and to candidates
 itself (at later stages).  Call that predicate is_candidate(); it would
-be m->mnt_flags & MNT_UMOUNT_CANDIDATE.
+be m->mnt_t_flags & T_UMOUNT_CANDIDATE.
 
 All elements of the original set are marked with MNT_UMOUNT and we'll
 need the same for elements added when joining the contents of to_umount
@@ -480,5 +480,5 @@ to to_umount; that's close to what the old 'umount_one' is doing, so
 let's keep that name.  It also gives us another predicate we need -
 "belongs to union of set and to_umount"; will_be_unmounted() for now.
 
-Removals from the candidates list should strip both MNT_MARKED and
-MNT_UMOUNT_CANDIDATE; call it remove_from_candidates_list().
+Removals from the candidates list should strip both T_MARKED and
+T_UMOUNT_CANDIDATE; call it remove_from_candidates_list().
diff --git a/fs/mount.h b/fs/mount.h
index c5b170b6cb3c..250e84c0f2ce 100644
--- a/fs/mount.h
+++ b/fs/mount.h
@@ -84,6 +84,7 @@ struct mount {
 	struct list_head to_notify;	/* need to queue notification */
 	struct mnt_namespace *prev_ns;	/* previous namespace (NULL if none) */
 #endif
+	int mnt_t_flags;		/* namespace_sem-protected flags */
 	int mnt_id;			/* mount identifier, reused */
 	u64 mnt_id_unique;		/* mount ID unique until reboot */
 	int mnt_group_id;		/* peer group identifier */
@@ -93,6 +94,22 @@ struct mount {
 	struct mount *overmount;	/* mounted on ->mnt_root */
 } __randomize_layout;
 
+enum {
+	T_SHARED		= 1, /* mount is shared */
+	T_UNBINDABLE		= 2, /* mount is unbindable */
+	T_MARKED		= 4, /* internal mark for propagate_... */
+	T_UMOUNT_CANDIDATE	= 8, /* for propagate_umount */
+
+	/*
+	 * T_SHARED_MASK is the set of flags that should be cleared when a
+	 * mount becomes shared.  Currently, this is only the flag that says a
+	 * mount cannot be bind mounted, since this is how we create a mount
+	 * that shares events with another mount.  If you add a new T_*
+	 * flag, consider how it interacts with shared mounts.
+	 */
+	T_SHARED_MASK	= T_UNBINDABLE,
+};
+
 #define MNT_NS_INTERNAL ERR_PTR(-EINVAL) /* distinct from any mnt_namespace */
 
 static inline struct mount *real_mount(struct vfsmount *mnt)
diff --git a/fs/namespace.c b/fs/namespace.c
index 5556c5edbae9..4ea72ecd2621 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -2905,10 +2905,8 @@ static int do_change_type(struct path *path, int ms_flags)
 			goto out_unlock;
 	}
 
-	lock_mount_hash();
 	for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
 		change_mnt_propagation(m, type);
-	unlock_mount_hash();
 
  out_unlock:
 	namespace_unlock();
@@ -3397,9 +3395,7 @@ static int do_set_group(struct path *from_path, struct path *to_path)
 	if (IS_MNT_SHARED(from)) {
 		to->mnt_group_id = from->mnt_group_id;
 		list_add(&to->mnt_share, &from->mnt_share);
-		lock_mount_hash();
 		set_mnt_shared(to);
-		unlock_mount_hash();
 	}
 
 	err = 0;
diff --git a/fs/pnode.c b/fs/pnode.c
index f897a501bee7..0ae14f7f754f 100644
--- a/fs/pnode.c
+++ b/fs/pnode.c
@@ -112,7 +112,7 @@ static int do_make_slave(struct mount *mnt)
 }
 
 /*
- * vfsmount lock must be held for write
+ * EXCL[namespace_sem]
  */
 void change_mnt_propagation(struct mount *mnt, int type)
 {
@@ -125,9 +125,9 @@ void change_mnt_propagation(struct mount *mnt, int type)
 		list_del_init(&mnt->mnt_slave);
 		mnt->mnt_master = NULL;
 		if (type == MS_UNBINDABLE)
-			mnt->mnt.mnt_flags |= MNT_UNBINDABLE;
+			mnt->mnt_t_flags |= T_UNBINDABLE;
 		else
-			mnt->mnt.mnt_flags &= ~MNT_UNBINDABLE;
+			mnt->mnt_t_flags &= ~T_UNBINDABLE;
 	}
 }
 
@@ -263,9 +263,9 @@ static int propagate_one(struct mount *m, struct mountpoint *dest_mp)
 		return PTR_ERR(child);
 	read_seqlock_excl(&mount_lock);
 	mnt_set_mountpoint(m, dest_mp, child);
+	read_sequnlock_excl(&mount_lock);
 	if (m->mnt_master != dest_master)
 		SET_MNT_MARK(m->mnt_master);
-	read_sequnlock_excl(&mount_lock);
 	last_dest = m;
 	last_source = child;
 	hlist_add_head(&child->mnt_hash, list);
@@ -322,13 +322,11 @@ int propagate_mnt(struct mount *dest_mnt, struct mountpoint *dest_mp,
 		} while (n != m);
 	}
 out:
-	read_seqlock_excl(&mount_lock);
 	hlist_for_each_entry(n, tree_list, mnt_hash) {
 		m = n->mnt_parent;
 		if (m->mnt_master != dest_mnt->mnt_master)
 			CLEAR_MNT_MARK(m->mnt_master);
 	}
-	read_sequnlock_excl(&mount_lock);
 	return ret;
 }
 
@@ -447,7 +445,7 @@ void propagate_mount_unlock(struct mount *mnt)
 
 static inline bool is_candidate(struct mount *m)
 {
-	return m->mnt.mnt_flags & MNT_UMOUNT_CANDIDATE;
+	return m->mnt_t_flags & T_UMOUNT_CANDIDATE;
 }
 
 static inline bool will_be_unmounted(struct mount *m)
@@ -465,7 +463,7 @@ static void umount_one(struct mount *m, struct list_head *to_umount)
 
 static void remove_from_candidate_list(struct mount *m)
 {
-	m->mnt.mnt_flags &= ~(MNT_MARKED | MNT_UMOUNT_CANDIDATE);
+	m->mnt_t_flags &= ~(T_MARKED | T_UMOUNT_CANDIDATE);
 	list_del_init(&m->mnt_list);
 }
 
@@ -477,7 +475,7 @@ static void gather_candidates(struct list_head *set,
 	list_for_each_entry(m, set, mnt_list) {
 		if (is_candidate(m))
 			continue;
-		m->mnt.mnt_flags |= MNT_UMOUNT_CANDIDATE;
+		m->mnt_t_flags |= T_UMOUNT_CANDIDATE;
 		p = m->mnt_parent;
 		q = propagation_next(p, p);
 		while (q) {
@@ -495,7 +493,7 @@ static void gather_candidates(struct list_head *set,
 					q = skip_propagation_subtree(q, p);
 					continue;
 				}
-				child->mnt.mnt_flags |= MNT_UMOUNT_CANDIDATE;
+				child->mnt_t_flags |= T_UMOUNT_CANDIDATE;
 				if (!will_be_unmounted(child))
 					list_add(&child->mnt_list, candidates);
 			}
@@ -503,7 +501,7 @@ static void gather_candidates(struct list_head *set,
 		}
 	}
 	list_for_each_entry(m, set, mnt_list)
-		m->mnt.mnt_flags &= ~MNT_UMOUNT_CANDIDATE;
+		m->mnt_t_flags &= ~T_UMOUNT_CANDIDATE;
 }
 
 /*
@@ -520,7 +518,7 @@ static void trim_ancestors(struct mount *m)
 			return;
 		SET_MNT_MARK(m);
 		if (m->mnt_mountpoint != p->mnt.mnt_root)
-			p->mnt.mnt_flags &= ~MNT_UMOUNT_CANDIDATE;
+			p->mnt_t_flags &= ~T_UMOUNT_CANDIDATE;
 	}
 }
 
diff --git a/fs/pnode.h b/fs/pnode.h
index 04f1ac53aa49..507e30e7a420 100644
--- a/fs/pnode.h
+++ b/fs/pnode.h
@@ -10,14 +10,14 @@
 #include <linux/list.h>
 #include "mount.h"
 
-#define IS_MNT_SHARED(m) ((m)->mnt.mnt_flags & MNT_SHARED)
+#define IS_MNT_SHARED(m) ((m)->mnt_t_flags & T_SHARED)
 #define IS_MNT_SLAVE(m) ((m)->mnt_master)
 #define IS_MNT_NEW(m) (!(m)->mnt_ns)
-#define CLEAR_MNT_SHARED(m) ((m)->mnt.mnt_flags &= ~MNT_SHARED)
-#define IS_MNT_UNBINDABLE(m) ((m)->mnt.mnt_flags & MNT_UNBINDABLE)
-#define IS_MNT_MARKED(m) ((m)->mnt.mnt_flags & MNT_MARKED)
-#define SET_MNT_MARK(m) ((m)->mnt.mnt_flags |= MNT_MARKED)
-#define CLEAR_MNT_MARK(m) ((m)->mnt.mnt_flags &= ~MNT_MARKED)
+#define CLEAR_MNT_SHARED(m) ((m)->mnt_t_flags &= ~T_SHARED)
+#define IS_MNT_UNBINDABLE(m) ((m)->mnt_t_flags & T_UNBINDABLE)
+#define IS_MNT_MARKED(m) ((m)->mnt_t_flags & T_MARKED)
+#define SET_MNT_MARK(m) ((m)->mnt_t_flags |= T_MARKED)
+#define CLEAR_MNT_MARK(m) ((m)->mnt_t_flags &= ~T_MARKED)
 #define IS_MNT_LOCKED(m) ((m)->mnt.mnt_flags & MNT_LOCKED)
 
 #define CL_EXPIRE    		0x01
@@ -28,10 +28,13 @@
 #define CL_SHARED_TO_SLAVE	0x20
 #define CL_COPY_MNT_NS_FILE	0x40
 
+/*
+ * EXCL[namespace_sem]
+ */
 static inline void set_mnt_shared(struct mount *mnt)
 {
-	mnt->mnt.mnt_flags &= ~MNT_SHARED_MASK;
-	mnt->mnt.mnt_flags |= MNT_SHARED;
+	mnt->mnt_t_flags &= ~T_SHARED_MASK;
+	mnt->mnt_t_flags |= T_SHARED;
 }
 
 static inline bool peers(const struct mount *m1, const struct mount *m2)
diff --git a/include/linux/mount.h b/include/linux/mount.h
index 65fa8442c00a..5f9c053b0897 100644
--- a/include/linux/mount.h
+++ b/include/linux/mount.h
@@ -35,12 +35,8 @@ enum mount_flags {
 	MNT_SHRINKABLE	= 0x100,
 	MNT_WRITE_HOLD	= 0x200,
 
-	MNT_SHARED	= 0x1000, /* if the vfsmount is a shared mount */
-	MNT_UNBINDABLE	= 0x2000, /* if the vfsmount is a unbindable mount */
-
 	MNT_INTERNAL	= 0x4000,
 
-	MNT_UMOUNT_CANDIDATE	= 0x020000,
 	MNT_LOCK_ATIME		= 0x040000,
 	MNT_LOCK_NOEXEC		= 0x080000,
 	MNT_LOCK_NOSUID		= 0x100000,
@@ -49,25 +45,15 @@ enum mount_flags {
 	MNT_LOCKED		= 0x800000,
 	MNT_DOOMED		= 0x1000000,
 	MNT_SYNC_UMOUNT		= 0x2000000,
-	MNT_MARKED		= 0x4000000,
 	MNT_UMOUNT		= 0x8000000,
 
-	/*
-	 * MNT_SHARED_MASK is the set of flags that should be cleared when a
-	 * mount becomes shared.  Currently, this is only the flag that says a
-	 * mount cannot be bind mounted, since this is how we create a mount
-	 * that shares events with another mount.  If you add a new MNT_*
-	 * flag, consider how it interacts with shared mounts.
-	 */
-	MNT_SHARED_MASK	= MNT_UNBINDABLE,
 	MNT_USER_SETTABLE_MASK  = MNT_NOSUID | MNT_NODEV | MNT_NOEXEC
 				  | MNT_NOATIME | MNT_NODIRATIME | MNT_RELATIME
 				  | MNT_READONLY | MNT_NOSYMFOLLOW,
 	MNT_ATIME_MASK = MNT_NOATIME | MNT_NODIRATIME | MNT_RELATIME,
 
-	MNT_INTERNAL_FLAGS = MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL |
-			     MNT_DOOMED | MNT_SYNC_UMOUNT | MNT_MARKED |
-			     MNT_LOCKED | MNT_UMOUNT_CANDIDATE,
+	MNT_INTERNAL_FLAGS = MNT_WRITE_HOLD | MNT_INTERNAL | MNT_DOOMED |
+			     MNT_SYNC_UMOUNT | MNT_LOCKED
 };
 
 struct vfsmount {
-- 
2.39.5

[PATCH v2 33/35] propagate_one(): get rid of dest_master

[Al Viro]

propagate_mnt() takes the subtree we are about to attach and creates
its copies, setting the propagation between those.  Each copy is cloned
either from the original or from one of the already created copies.
The tricky part is choosing the right copy to serve as a master when we
are starting a new peer group.

The algorithm for doing that selection puts temporary marks on the masters
of mountpoints that already got a copy created for them; since the initial
peer group might have no master at all, we need to special-case that when
looking for the mark.  Currently we do that by memorizing the master of
original peer group.  It works, but we get yet another piece of data to
pass from propagate_mnt() to propagate_one().

Alternative is to mark the master of original peer group if not NULL,
turning the check into "master is not NULL or it is marked".  Less data
to pass around and memory safety is more obvious that way...

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/pnode.c | 13 ++++++++-----
 1 file changed, 8 insertions(+), 5 deletions(-)

diff --git a/fs/pnode.c b/fs/pnode.c
index 0ae14f7f754f..b54f7ca8cff5 100644
--- a/fs/pnode.c
+++ b/fs/pnode.c
@@ -215,7 +215,7 @@ static struct mount *next_group(struct mount *m, struct mount *origin)
 }
 
 /* all accesses are serialized by namespace_sem */
-static struct mount *last_dest, *first_source, *last_source, *dest_master;
+static struct mount *last_dest, *first_source, *last_source;
 static struct hlist_head *list;
 
 static int propagate_one(struct mount *m, struct mountpoint *dest_mp)
@@ -239,7 +239,7 @@ static int propagate_one(struct mount *m, struct mountpoint *dest_mp)
 		bool done;
 		for (n = m; ; n = p) {
 			p = n->mnt_master;
-			if (p == dest_master || IS_MNT_MARKED(p))
+			if (!p || IS_MNT_MARKED(p))
 				break;
 		}
 		do {
@@ -264,7 +264,7 @@ static int propagate_one(struct mount *m, struct mountpoint *dest_mp)
 	read_seqlock_excl(&mount_lock);
 	mnt_set_mountpoint(m, dest_mp, child);
 	read_sequnlock_excl(&mount_lock);
-	if (m->mnt_master != dest_master)
+	if (m->mnt_master)
 		SET_MNT_MARK(m->mnt_master);
 	last_dest = m;
 	last_source = child;
@@ -300,7 +300,8 @@ int propagate_mnt(struct mount *dest_mnt, struct mountpoint *dest_mp,
 	first_source = source_mnt;
 	last_source = source_mnt;
 	list = tree_list;
-	dest_master = dest_mnt->mnt_master;
+	if (dest_mnt->mnt_master)
+		SET_MNT_MARK(dest_mnt->mnt_master);
 
 	/* all peers of dest_mnt, except dest_mnt itself */
 	for (n = next_peer(dest_mnt); n != dest_mnt; n = next_peer(n)) {
@@ -324,9 +325,11 @@ int propagate_mnt(struct mount *dest_mnt, struct mountpoint *dest_mp,
 out:
 	hlist_for_each_entry(n, tree_list, mnt_hash) {
 		m = n->mnt_parent;
-		if (m->mnt_master != dest_mnt->mnt_master)
+		if (m->mnt_master)
 			CLEAR_MNT_MARK(m->mnt_master);
 	}
+	if (dest_mnt->mnt_master)
+		CLEAR_MNT_MARK(dest_mnt->mnt_master);
 	return ret;
 }
 
-- 
2.39.5

[PATCH v2 34/35] propagate_mnt(): get rid of globals

[Al Viro]

TBH, I'm not sure it makes the things better, but...

Fixed a stale comment, while we are at it - propagate_mnt() does *not*
use ->mnt_list for linkage and it does not attach the secondaries to
the original - they go into the caller-supplied list.  It had gone
stale before the entire thing got merged into the tree...

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/pnode.c | 61 +++++++++++++++++++++++++++---------------------------
 1 file changed, 31 insertions(+), 30 deletions(-)

diff --git a/fs/pnode.c b/fs/pnode.c
index b54f7ca8cff5..dacb7f515eed 100644
--- a/fs/pnode.c
+++ b/fs/pnode.c
@@ -214,25 +214,28 @@ static struct mount *next_group(struct mount *m, struct mount *origin)
 	}
 }
 
-/* all accesses are serialized by namespace_sem */
-static struct mount *last_dest, *first_source, *last_source;
-static struct hlist_head *list;
+struct propagate_mnt_context {
+	struct mountpoint *dest_mp;
+	struct hlist_head *list;
+	struct mount *last_dest, *source, *last_source;
+};
 
-static int propagate_one(struct mount *m, struct mountpoint *dest_mp)
+static int propagate_one(struct mount *m, struct propagate_mnt_context *ctx)
 {
-	struct mount *child;
+	struct mount *last_source = ctx->last_source;
+	struct mount *copy;
 	int type;
 	/* skip ones added by this propagate_mnt() */
 	if (IS_MNT_NEW(m))
 		return 0;
 	/* skip if mountpoint isn't visible in m */
-	if (!is_subdir(dest_mp->m_dentry, m->mnt.mnt_root))
+	if (!is_subdir(ctx->dest_mp->m_dentry, m->mnt.mnt_root))
 		return 0;
 	/* skip if m is in the anon_ns */
 	if (is_anon_ns(m->mnt_ns))
 		return 0;
 
-	if (peers(m, last_dest)) {
+	if (peers(m, ctx->last_dest)) {
 		type = CL_MAKE_SHARED;
 	} else {
 		struct mount *n, *p;
@@ -244,7 +247,7 @@ static int propagate_one(struct mount *m, struct mountpoint *dest_mp)
 		}
 		do {
 			struct mount *parent = last_source->mnt_parent;
-			if (peers(last_source, first_source))
+			if (peers(last_source, ctx->source))
 				break;
 			done = parent->mnt_master == p;
 			if (done && peers(n, parent))
@@ -258,18 +261,18 @@ static int propagate_one(struct mount *m, struct mountpoint *dest_mp)
 			type |= CL_MAKE_SHARED;
 	}
 		
-	child = copy_tree(last_source, last_source->mnt.mnt_root, type);
-	if (IS_ERR(child))
-		return PTR_ERR(child);
+	copy = copy_tree(last_source, last_source->mnt.mnt_root, type);
+	if (IS_ERR(copy))
+		return PTR_ERR(copy);
 	read_seqlock_excl(&mount_lock);
-	mnt_set_mountpoint(m, dest_mp, child);
+	mnt_set_mountpoint(m, ctx->dest_mp, copy);
 	read_sequnlock_excl(&mount_lock);
 	if (m->mnt_master)
 		SET_MNT_MARK(m->mnt_master);
-	last_dest = m;
-	last_source = child;
-	hlist_add_head(&child->mnt_hash, list);
-	return count_mounts(m->mnt_ns, child);
+	ctx->last_dest = m;
+	ctx->last_source = copy;
+	hlist_add_head(&copy->mnt_hash, ctx->list);
+	return count_mounts(m->mnt_ns, copy);
 }
 
 /*
@@ -277,35 +280,33 @@ static int propagate_one(struct mount *m, struct mountpoint *dest_mp)
  * dentry 'dest_dentry'. And propagate that mount to
  * all the peer and slave mounts of 'dest_mnt'.
  * Link all the new mounts into a propagation tree headed at
- * source_mnt. Also link all the new mounts using ->mnt_list
- * headed at source_mnt's ->mnt_list
+ * source_mnt.  Roots of all copies placed into 'tree_list',
+ * linked by ->mnt_hash.
  *
  * @dest_mnt: destination mount.
  * @dest_dentry: destination dentry.
  * @source_mnt: source mount.
- * @tree_list : list of heads of trees to be attached.
+ * @tree_list : list of trees to be attached.
  */
 int propagate_mnt(struct mount *dest_mnt, struct mountpoint *dest_mp,
 		    struct mount *source_mnt, struct hlist_head *tree_list)
 {
 	struct mount *m, *n;
 	int ret = 0;
+	struct propagate_mnt_context ctx = {
+		.source = source_mnt,
+		.dest_mp = dest_mp,
+		.list = tree_list,
+		.last_source = source_mnt,
+		.last_dest = dest_mnt,
+	};
 
-	/*
-	 * we don't want to bother passing tons of arguments to
-	 * propagate_one(); everything is serialized by namespace_sem,
-	 * so globals will do just fine.
-	 */
-	last_dest = dest_mnt;
-	first_source = source_mnt;
-	last_source = source_mnt;
-	list = tree_list;
 	if (dest_mnt->mnt_master)
 		SET_MNT_MARK(dest_mnt->mnt_master);
 
 	/* all peers of dest_mnt, except dest_mnt itself */
 	for (n = next_peer(dest_mnt); n != dest_mnt; n = next_peer(n)) {
-		ret = propagate_one(n, dest_mp);
+		ret = propagate_one(n, &ctx);
 		if (ret)
 			goto out;
 	}
@@ -316,7 +317,7 @@ int propagate_mnt(struct mount *dest_mnt, struct mountpoint *dest_mp,
 		/* everything in that slave group */
 		n = m;
 		do {
-			ret = propagate_one(n, dest_mp);
+			ret = propagate_one(n, &ctx);
 			if (ret)
 				goto out;
 			n = next_peer(n);
-- 
2.39.5

[PATCH v2 35/35] take freeing of emptied mnt_namespace to namespace_unlock()

[Al Viro]

Freeing of a namespace must be delayed until after we'd dealt with mount
notifications (in namespace_unlock()).  The reasons are not immediately
obvious (they are buried in ->prev_ns handling in mnt_notify()), and
having that free_mnt_ns() explicitly called after namespace_unlock()
is asking for trouble - it does feel like they should be OK to free
as soon as they've been emptied.

Make the things more explicit by setting 'emptied_ns' under namespace_sem
and having namespace_unlock() free the sucker as soon as it's safe to free.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 27 ++++++++++++---------------
 1 file changed, 12 insertions(+), 15 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index 4ea72ecd2621..b8abac07b794 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -79,6 +79,7 @@ static struct kmem_cache *mnt_cache __ro_after_init;
 static DECLARE_RWSEM(namespace_sem);
 static HLIST_HEAD(unmounted);	/* protected by namespace_sem */
 static LIST_HEAD(ex_mountpoints); /* protected by namespace_sem */
+static struct mnt_namespace *emptied_ns; /* protected by namespace_sem */
 static DEFINE_SEQLOCK(mnt_ns_tree_lock);
 
 #ifdef CONFIG_FSNOTIFY
@@ -1730,15 +1731,18 @@ static bool need_notify_mnt_list(void)
 }
 #endif
 
+static void free_mnt_ns(struct mnt_namespace *);
 static void namespace_unlock(void)
 {
 	struct hlist_head head;
 	struct hlist_node *p;
 	struct mount *m;
+	struct mnt_namespace *ns = emptied_ns;
 	LIST_HEAD(list);
 
 	hlist_move_list(&unmounted, &head);
 	list_splice_init(&ex_mountpoints, &list);
+	emptied_ns = NULL;
 
 	if (need_notify_mnt_list()) {
 		/*
@@ -1752,6 +1756,11 @@ static void namespace_unlock(void)
 	} else {
 		up_write(&namespace_sem);
 	}
+	if (unlikely(ns)) {
+		/* Make sure we notice when we leak mounts. */
+		VFS_WARN_ON_ONCE(!mnt_ns_empty(ns));
+		free_mnt_ns(ns);
+	}
 
 	shrink_dentry_list(&list);
 
@@ -2335,12 +2344,10 @@ void drop_collected_paths(struct path *paths, struct path *prealloc)
 		kfree(paths);
 }
 
-static void free_mnt_ns(struct mnt_namespace *);
 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *, bool);
 
 void dissolve_on_fput(struct vfsmount *mnt)
 {
-	struct mnt_namespace *ns;
 	struct mount *m = real_mount(mnt);
 
 	/*
@@ -2362,15 +2369,11 @@ void dissolve_on_fput(struct vfsmount *mnt)
 		if (!anon_ns_root(m))
 			return;
 
-		ns = m->mnt_ns;
+		emptied_ns = m->mnt_ns;
 		lock_mount_hash();
 		umount_tree(m, UMOUNT_CONNECTED);
 		unlock_mount_hash();
 	}
-
-	/* Make sure we notice when we leak mounts. */
-	VFS_WARN_ON_ONCE(!mnt_ns_empty(ns));
-	free_mnt_ns(ns);
 }
 
 static bool __has_locked_children(struct mount *mnt, struct dentry *dentry)
@@ -2678,6 +2681,7 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 	} else {
 		if (source_mnt->mnt_ns) {
 			/* move from anon - the caller will destroy */
+			emptied_ns = source_mnt->mnt_ns;
 			for (p = source_mnt; p; p = next_mnt(p, source_mnt))
 				move_from_ns(p);
 		}
@@ -3658,13 +3662,6 @@ static int do_move_mount(struct path *old_path,
 	err = attach_recursive_mnt(old, p, mp.mp);
 out:
 	unlock_mount(&mp);
-	if (!err) {
-		if (is_anon_ns(ns)) {
-			/* Make sure we notice when we leak mounts. */
-			VFS_WARN_ON_ONCE(!mnt_ns_empty(ns));
-			free_mnt_ns(ns);
-		}
-	}
 	return err;
 }
 
@@ -6159,11 +6156,11 @@ void put_mnt_ns(struct mnt_namespace *ns)
 	if (!refcount_dec_and_test(&ns->ns.count))
 		return;
 	namespace_lock();
+	emptied_ns = ns;
 	lock_mount_hash();
 	umount_tree(ns->root, 0);
 	unlock_mount_hash();
 	namespace_unlock();
-	free_mnt_ns(ns);
 }
 
 struct vfsmount *kern_mount(struct file_system_type *type)
-- 
2.39.5

Re: [PATCH v2 01/35] replace collect_mounts()/drop_collected_mounts() with a safer variant

[Al Viro]

On Mon, Jun 23, 2025 at 05:53:54AM +0100, Al Viro wrote:
> @@ -828,16 +832,16 @@ int audit_add_tree_rule(struct audit_krule *rule)
>  	err = kern_path(tree->pathname, 0, &path);
>  	if (err)
>  		goto Err;
> -	mnt = collect_mounts(&path);
> +	paths = collect_paths(paths, array, 16);

that would be
	paths = collect_paths(&path, array, 16);
of course.  Kudos to venkat88@linux.ibm.com for spotting the breakage.

Al, off to find out what's wrong with his local audit-related tests...

Re: [PATCHES v2][RFC][CFR] mount-related stuff

[Ian Kent]

On 23/6/25 12:49, Al Viro wrote:
> On Tue, Jun 10, 2025 at 09:17:58AM +0100, Al Viro wrote:
>> 	The next pile of mount massage; it will grow - there will be
>> further modifications, as well as fixes and documentation, but this is
>> the subset I've got in more or less settled form right now.
>>
>> 	Review and testing would be very welcome.
>>
>> 	This series (-rc1-based) sits in
>> git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs.git #work.mount
>> individual patches in followups.
> Updated variant force-pushed to
> git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs.git #work.mount
> individual patches in followups.  It seems to survive testing here, but
> more testing and review would be very welcome.  That's still not all -
> there's more stuff in local queue, but it needs more massage; this is
> the reasonably settled-down subset at the moment.

I'll have a look through these too and run my tests against them.

Btw, I did run my tests against v1 without any noticeable problem.


I also have revived my patch to make may_umount_tree() namespace aware

and it still seems to work fine. I'm not going to spend time on my second

patch that was meant to add some optimization because it seemed too

aggressive somehow and stopped working a while after it was done and I

couldn't see why. So for now I've dropped it.


Ian

>
> Changes since v1:
> Added fixes (-stable fodder, IMO):
> 	replace collect_mounts()/drop_collected_mounts() with safer variant
> 	attach_recursive_mnt(): do not lock the covering tree when sliding something under it
>
> Another thing that is probably a -stable candidate:
> 	prevent mount hash conflicts
> That's an old headache hopefully taken care of; what we get out of it
> is the warranty that there won't be more than one hashed mount with
> given parent/mountpoint at any given time.  I've pulled that very
> early in the sequence, so that we had a chance to backport it.
> That comes with two prereqs (reordered from the middle of v1 series),
> both should be trivial to backport.
>
> Added cleanups:
> 	__attach_mnt(): lose the second argument
> 	copy_tree(): don't link the mounts via mnt_list
> All uses of ->mnt_list are transient now - basically, various sets
> used during umount_tree().
> 	mount: separate the flags accessed only under namespace_sem
> Makes for simpler locking; some of the flags are accessed only under
> namespace_sem, and we already rely upon that in the readers; taking
> them to a separate word avoids the need to grab mount_lock on the
> write side.
> 	propagate_one(): get rid of dest_master
> 	propagate_mnt(): get rid of globals
> Linus asked to get rid of fs/pnode.c globals; done.
> 	take freeing of emptied mnt_namespace to namespace_unlock()
>
> A couple of commits made simpler by "prevent mount hash conflicts" -
> 	Rewrite of propagate_umount()
> reparenting is guaranteed that there won't be more than one overmount now,
> no loop needed.
> 	don't have mounts pin their parents
> simpler logics for "is there something other than overmount?"
>
> 	Rough overview:
>
> Part 1: fixes
>
> 1) replace collect_mounts()/drop_collected_mounts() with safer variant
> 2) attach_recursive_mnt(): do not lock the covering tree when sliding something under it
>
> Part 2: getting rid of mount hash conflicts for good
>
> 3) attach_mnt(): expand in attach_recursive_mnt(), then lose the flag argument
> 4) get rid of mnt_set_mountpoint_beneath()
> 5) prevent mount hash conflicts
>
> Part 3: trivial cleanups and helpers:
>
> 6) copy_tree(): don't set ->mnt_mountpoint on the root of copy
> 7) constify mnt_has_parent()
> 8) pnode: lift peers() into pnode.h
> 9) new predicate: mount_is_ancestor()
> 10) constify is_local_mountpoint()
> 11) new predicate: anon_ns_root(mount)
> 12) dissolve_on_fput(): use anon_ns_root()
> 13) __attach_mnt(): lose the second argument
> 	... and rename to make_visible()
> 14) don't set MNT_LOCKED on parentless mounts
> 15) clone_mnt(): simplify the propagation-related logics
> 16) do_umount(): simplify the "is it still mounted" checks
>
> Part 4: (somewhat of a side story) restore the machinery for long-term
> mounts from accumulated bitrot.
>
> 17) sanitize handling of long-term internal mounts
>
> Part 5: propagate_umount() rewrite (posted last cycle)
>
> 18) Rewrite of propagate_umount()
>
> Part 6: untangling do_move_mount()/attach_recursive_mnt().
>
> 19) make commit_tree() usable in same-namespace move case
> 20) attach_recursive_mnt(): unify the mnt_change_mountpoint() logics
> 21) attach_recursive_mnt(): pass destination mount in all cases
> 22) attach_recursive_mnt(): get rid of flags entirely
> 23) do_move_mount(): take dropping the old mountpoint into attach_recursive_mnt()
> 24) do_move_mount(): get rid of 'attached' flag
>
> Part 7: change locking for expiry lists.
> 25) attach_recursive_mnt(): remove from expiry list on move
> 26) take ->mnt_expire handling under mount_lock [read_seqlock_excl]
>
> Part 8: struct mountpoint massage.
> 27) pivot_root(): reorder tree surgeries, collapse unhash_mnt() and put_mountpoint()
> 28) combine __put_mountpoint() with unhash_mnt()
> 29) get rid of mountpoint->m_count
>
> Part 9: regularize mount refcounting a bit
> 30) don't have mounts pin their parents
>
> Part 10: misc stuff, will grow...
> 31) copy_tree(): don't link the mounts via mnt_list
> 32) mount: separate the flags accessed only under namespace_sem
> 33) propagate_one(): get rid of dest_master
> 34) propagate_mnt(): get rid of globals
> 35) take freeing of emptied mnt_namespace to namespace_unlock()
>
> Diffstat:
>   Documentation/filesystems/porting.rst          |   9 +
>   Documentation/filesystems/propagate_umount.txt | 484 +++++++++++++++
>   drivers/gpu/drm/i915/gem/i915_gemfs.c          |  21 +-
>   drivers/gpu/drm/v3d/v3d_gemfs.c                |  21 +-
>   fs/hugetlbfs/inode.c                           |   2 +-
>   fs/mount.h                                     |  36 +-
>   fs/namespace.c                                 | 783 +++++++++++--------------
>   fs/pnode.c                                     | 499 ++++++++--------
>   fs/pnode.h                                     |  28 +-
>   include/linux/mount.h                          |  24 +-
>   ipc/mqueue.c                                   |   2 +-
>   kernel/audit_tree.c                            |  63 +-
>   12 files changed, 1214 insertions(+), 758 deletions(-)
>   create mode 100644 Documentation/filesystems/propagate_umount.txt
>

Re: [PATCHES v2][RFC][CFR] mount-related stuff

[Al Viro]

On Mon, Jun 23, 2025 at 05:06:52PM +0800, Ian Kent wrote:

> I also have revived my patch to make may_umount_tree() namespace aware
> 
> and it still seems to work fine.

Could you post it?

Re: [PATCHES v2][RFC][CFR] mount-related stuff

[Ian Kent]

On 24/6/25 02:55, Al Viro wrote:
> On Mon, Jun 23, 2025 at 05:06:52PM +0800, Ian Kent wrote:
>
>> I also have revived my patch to make may_umount_tree() namespace aware
>>
>> and it still seems to work fine.
> Could you post it?


Not sure the formatting will be ok since my email setup is a mess.

It's against the v1 of your series.


Your advice (and Christian and others) would be much appreciated.


vfs: make may_umount_tree() mount namespace aware

From: Ian Kent <ikent@redhat.com>

Change may_umount_tree() to also check if propagated mounts are busy
during autofs expire runs.

Also alter may_umount_tree() to take a flag to indicate a reference to
the passed in mount is held.

This avoids unnecessary umount requests being sent to the automount
daemon if a mount in another mount namespace is in use when the expire
check is done.

Signed-off-by: Ian Kent <raven@themaw.net>
---
  fs/autofs/expire.c    |    4 ++--
  fs/namespace.c        |   36 ++++++++++++++++++++++++++++++------
  fs/pnode.c            |   32 ++++++++++++++++++++++++++++++++
  fs/pnode.h            |    1 +
  include/linux/mount.h |    5 ++++-
  5 files changed, 69 insertions(+), 9 deletions(-)

diff --git a/fs/autofs/expire.c b/fs/autofs/expire.c
index 5c2d459e1e48..c303d11f4c12 100644
--- a/fs/autofs/expire.c
+++ b/fs/autofs/expire.c
@@ -55,7 +55,7 @@ static int autofs_mount_busy(struct vfsmount *mnt,
      }

      /* Update the expiry counter if fs is busy */
-    if (!may_umount_tree(path.mnt)) {
+    if (!may_umount_tree(path.mnt, TREE_BUSY_REFERENCED)) {
          struct autofs_info *ino;

          ino = autofs_dentry_ino(top);
@@ -156,7 +156,7 @@ static int autofs_direct_busy(struct vfsmount *mnt,
          return 0;

      /* If it's busy update the expiry counters */
-    if (!may_umount_tree(mnt)) {
+    if (!may_umount_tree(mnt, TREE_BUSY_REFERENCED)) {
          struct autofs_info *ino;

          ino = autofs_dentry_ino(top);
diff --git a/fs/namespace.c b/fs/namespace.c
index bb95e5102916..3cb90bb46b94 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -1635,25 +1635,49 @@ const struct seq_operations mounts_op = {
  /**
   * may_umount_tree - check if a mount tree is busy
   * @m: root of mount tree
+ * @flags: behaviour modifier flags:
+ *     TREE_BUSY_REFERENCED caller holds additional reference
+ *     to @m.
   *
   * This is called to check if a tree of mounts has any
   * open files, pwds, chroots or sub mounts that are
   * busy.
   */
-int may_umount_tree(struct vfsmount *m)
+bool may_umount_tree(struct vfsmount *m, unsigned int flags)
  {
      struct mount *mnt = real_mount(m);
+    struct mount *p, *q;
      bool busy = false;

-    /* write lock needed for mnt_get_count */
+    down_read(&namespace_sem);
      lock_mount_hash();
-    for (struct mount *p = mnt; p; p = next_mnt(p, mnt)) {
-        if (mnt_get_count(p) > (p == mnt ? 2 : 1)) {
-            busy = true;
-            break;
+    for (p = mnt; p; p = next_mnt(p, mnt)) {
+        unsigned int f = 0;
+
+        if (p->mnt_mountpoint != mnt->mnt.mnt_root) {
+            if (p == mnt)
+                f = flags;
+            if (propagate_mount_tree_busy(p, f)) {
+                busy = true;
+                break;
+            }
+            continue;
+        }
+
+        /* p is a covering mnt, need to check if p or any of its
+         * children are in use. A reference to p is not held so
+         * don't pass TREE_BUSY_REFERENCED to the propagation
+         * helper.
+         */
+        for (q = p; q; q = next_mnt(q, p)) {
+            if (propagate_mount_tree_busy(q, f)) {
+                busy = true;
+                break;
+            }
          }
      }
      unlock_mount_hash();
+    up_read(&namespace_sem);

      return !busy;
  }
diff --git a/fs/pnode.c b/fs/pnode.c
index efed6bb20c72..e4222a008039 100644
--- a/fs/pnode.c
+++ b/fs/pnode.c
@@ -429,6 +429,38 @@ int propagate_mount_busy(struct mount *mnt, int refcnt)
      return 0;
  }

+/*
+ * Check if the mount tree at 'mnt' is in use or any of its
+ * propogated mounts are in use.
+ * @mnt: the mount to be checked
+ * @flags: see may_umount_tree() for modifier descriptions.
+ *
+ * Check if mnt or any of its propogated mounts have a reference
+ * count greater than the minimum reference count (ie. are in use).
+ */
+int propagate_mount_tree_busy(struct mount *mnt, unsigned int flags)
+{
+    struct mount *m;
+    struct mount *parent = mnt->mnt_parent;
+    int refcnt = flags & TREE_BUSY_REFERENCED ? 2 : 1;
+
+    if (do_refcount_check(mnt, refcnt))
+        return 1;
+
+    for (m = propagation_next(parent, parent); m;
+            m = propagation_next(m, parent)) {
+        struct mount *child;
+
+        child = __lookup_mnt(&m->mnt, mnt->mnt_mountpoint);
+        if (!child)
+            continue;
+
+        if (do_refcount_check(child, 1))
+            return 1;
+    }
+    return 0;
+}
+
  /*
   * Clear MNT_LOCKED when it can be shown to be safe.
   *
diff --git a/fs/pnode.h b/fs/pnode.h
index bfc10c095cbf..a0d2974e57d7 100644
--- a/fs/pnode.h
+++ b/fs/pnode.h
@@ -46,6 +46,7 @@ int propagate_mnt(struct mount *, struct mountpoint *, 
struct mount *,
          struct hlist_head *);
  void propagate_umount(struct list_head *);
  int propagate_mount_busy(struct mount *, int);
+int propagate_mount_tree_busy(struct mount *, unsigned int);
  void propagate_mount_unlock(struct mount *);
  void mnt_release_group_id(struct mount *);
  int get_dominating_id(struct mount *mnt, const struct path *root);
diff --git a/include/linux/mount.h b/include/linux/mount.h
index cae7324650b6..d66555cc8e96 100644
--- a/include/linux/mount.h
+++ b/include/linux/mount.h
@@ -114,7 +114,10 @@ extern bool our_mnt(struct vfsmount *mnt);

  extern struct vfsmount *kern_mount(struct file_system_type *);
  extern void kern_unmount(struct vfsmount *mnt);
-extern int may_umount_tree(struct vfsmount *);
+
+#define TREE_BUSY_REFERENCED        0x01
+
+extern bool may_umount_tree(struct vfsmount *, unsigned int);
  extern int may_umount(struct vfsmount *);
  int do_mount(const char *, const char __user *,
               const char *, unsigned long, void *);

Re: [PATCHES v2][RFC][CFR] mount-related stuff

[Al Viro]

On Tue, Jun 24, 2025 at 02:48:53PM +0800, Ian Kent wrote:

> Also alter may_umount_tree() to take a flag to indicate a reference to
> the passed in mount is held.
> 
> This avoids unnecessary umount requests being sent to the automount
> daemon if a mount in another mount namespace is in use when the expire
> check is done.

Huh?  I'm probably missing something, but all callers of may_umount_tree()
seem to be passing that flag...  propagate_mount_tree_busy() - sure, but why
does may_umount_tree() get that?

I'm half-asleep at the moment (3am here), so maybe it will make more sense
in the morning...

Re: [PATCHES v2][RFC][CFR] mount-related stuff

[Ian Kent]

On 24/6/25 15:05, Al Viro wrote:
> On Tue, Jun 24, 2025 at 02:48:53PM +0800, Ian Kent wrote:
>
>> Also alter may_umount_tree() to take a flag to indicate a reference to
>> the passed in mount is held.
>>
>> This avoids unnecessary umount requests being sent to the automount
>> daemon if a mount in another mount namespace is in use when the expire
>> check is done.
> Huh?  I'm probably missing something, but all callers of may_umount_tree()
> seem to be passing that flag...  propagate_mount_tree_busy() - sure, but why
> does may_umount_tree() get that?
>
> I'm half-asleep at the moment (3am here), so maybe it will make more sense
> in the morning...

Yes, fair point, maybe it should be propagate_mount_tree_busy() only 
that grows

a flag as it's the function that gets called without knowing whether the 
caller

holds an additional reference to the mount or not.


Ian

Re: [PATCHES v2][RFC][CFR] mount-related stuff

[Al Viro]

On Tue, Jun 24, 2025 at 02:48:53PM +0800, Ian Kent wrote:

> +    for (p = mnt; p; p = next_mnt(p, mnt)) {
> +        unsigned int f = 0;
> +
> +        if (p->mnt_mountpoint != mnt->mnt.mnt_root) {

???  The loop goes over everything mounted on mnt, no matter how
deep it is.  Do you mean "p is mounted on the root of its parent",
or is it "p is mounted on some mount of the same fs, with mountpoint
that just happens to be equal to root dentry of mnt (which may be
not the mount p is mounted on)"?

> +        /* p is a covering mnt, need to check if p or any of its
> +         * children are in use. A reference to p is not held so
> +         * don't pass TREE_BUSY_REFERENCED to the propagation
> +         * helper.
> +         */

... so for these you keep walking through the subtree on them (nevermind
that outer loop will walk it as well)...

> +        for (q = p; q; q = next_mnt(q, p)) {
> +            if (propagate_mount_tree_busy(q, f)) {
> +                busy = true;
> +                break;
> +            }

... and yet you still keep going in the outer loop?  Confused...
>      }
>      unlock_mount_hash();
> +    up_read(&namespace_sem);

> + * count greater than the minimum reference count (ie. are in use).
> + */
> +int propagate_mount_tree_busy(struct mount *mnt, unsigned int flags)
> +{
> +    struct mount *m;
> +    struct mount *parent = mnt->mnt_parent;
> +    int refcnt = flags & TREE_BUSY_REFERENCED ? 2 : 1;
> +
> +    if (do_refcount_check(mnt, refcnt))
> +        return 1;
> +
> +    for (m = propagation_next(parent, parent); m;
> +            m = propagation_next(m, parent)) {
> +        struct mount *child;
> +
> +        child = __lookup_mnt(&m->mnt, mnt->mnt_mountpoint);
> +        if (!child)
> +            continue;
> +
> +        if (do_refcount_check(child, 1))
> +            return 1;
> +    }
> +    return 0;
> +}

What is the daemon expected to do with your subtree?  Take it apart with
a series of sync (== non-lazy) umount(2)?  I presume it is normal for
it to run into -EBUSY halfway through - i.e. get rid of some, but not
all of the subtree, right?

Re: [PATCHES v2][RFC][CFR] mount-related stuff

[Ian Kent]

On 25/6/25 15:57, Al Viro wrote:
> On Tue, Jun 24, 2025 at 02:48:53PM +0800, Ian Kent wrote:
>
>> +    for (p = mnt; p; p = next_mnt(p, mnt)) {
>> +        unsigned int f = 0;
>> +
>> +        if (p->mnt_mountpoint != mnt->mnt.mnt_root) {
> ???  The loop goes over everything mounted on mnt, no matter how
> deep it is.  Do you mean "p is mounted on the root of its parent",
> or is it "p is mounted on some mount of the same fs, with mountpoint
> that just happens to be equal to root dentry of mnt (which may be
> not the mount p is mounted on)"?

I was trying to check if p is not covered but that's not what it does.


>
>> +        /* p is a covering mnt, need to check if p or any of its
>> +         * children are in use. A reference to p is not held so
>> +         * don't pass TREE_BUSY_REFERENCED to the propagation
>> +         * helper.
>> +         */
> ... so for these you keep walking through the subtree on them (nevermind
> that outer loop will walk it as well)...
>
>> +        for (q = p; q; q = next_mnt(q, p)) {
>> +            if (propagate_mount_tree_busy(q, f)) {
>> +                busy = true;
>> +                break;
>> +            }
> ... and yet you still keep going in the outer loop?  Confused...

Yes, I've not got this right at all.


>>       }
>>       unlock_mount_hash();
>> +    up_read(&namespace_sem);
>> + * count greater than the minimum reference count (ie. are in use).
>> + */
>> +int propagate_mount_tree_busy(struct mount *mnt, unsigned int flags)
>> +{
>> +    struct mount *m;
>> +    struct mount *parent = mnt->mnt_parent;
>> +    int refcnt = flags & TREE_BUSY_REFERENCED ? 2 : 1;
>> +
>> +    if (do_refcount_check(mnt, refcnt))
>> +        return 1;
>> +
>> +    for (m = propagation_next(parent, parent); m;
>> +            m = propagation_next(m, parent)) {
>> +        struct mount *child;
>> +
>> +        child = __lookup_mnt(&m->mnt, mnt->mnt_mountpoint);
>> +        if (!child)
>> +            continue;
>> +
>> +        if (do_refcount_check(child, 1))
>> +            return 1;
>> +    }
>> +    return 0;
>> +}
> What is the daemon expected to do with your subtree?  Take it apart with
> a series of sync (== non-lazy) umount(2)?  I presume it is normal for
> it to run into -EBUSY halfway through - i.e. get rid of some, but not
> all of the subtree, right?

All I need is to check if a mount and its children are in use, essentially

check if a mount subtree is in use in some way, working directory or open

file(s).


I think what I should be doing is very similar to what may_umount() does but

for the passed in mount and it's children with the adjustment for the 
ref held

on the passed in mount.


I rather like your implementation of may_umount() and 
propagate_mount_busy().

I think the only difference to what I need is that the passed in mount 
won't be

the global root and will usually have children (so I'd invert that 
list_empty()

check or leave it out for this check).


I'll have a go at that refactor and post it, we'll see if I can come up with

something acceptable.


The other difficulty is that automount can be run in a separate mount 
namespace

(eg. a container) which introduces a special case environment. I'm not 
sure how

the propagation will behave in this case. Atm. I'm thinking it might be 
sufficient

to just use what may_umount_tree() is now so it deliberately only checks 
the mount

tree in it's own namespace.


Ian

Re: [PATCHES v2][RFC][CFR] mount-related stuff

[Ian Kent]

On 25/6/25 18:58, Ian Kent wrote:
> On 25/6/25 15:57, Al Viro wrote:
>> On Tue, Jun 24, 2025 at 02:48:53PM +0800, Ian Kent wrote:
>>
>>> +    for (p = mnt; p; p = next_mnt(p, mnt)) {
>>> +        unsigned int f = 0;
>>> +
>>> +        if (p->mnt_mountpoint != mnt->mnt.mnt_root) {
>> ???  The loop goes over everything mounted on mnt, no matter how
>> deep it is.  Do you mean "p is mounted on the root of its parent",
>> or is it "p is mounted on some mount of the same fs, with mountpoint
>> that just happens to be equal to root dentry of mnt (which may be
>> not the mount p is mounted on)"?
>
> I was trying to check if p is not covered but that's not what it does.
>
>
>>
>>> +        /* p is a covering mnt, need to check if p or any of its
>>> +         * children are in use. A reference to p is not held so
>>> +         * don't pass TREE_BUSY_REFERENCED to the propagation
>>> +         * helper.
>>> +         */
>> ... so for these you keep walking through the subtree on them (nevermind
>> that outer loop will walk it as well)...
>>
>>> +        for (q = p; q; q = next_mnt(q, p)) {
>>> +            if (propagate_mount_tree_busy(q, f)) {
>>> +                busy = true;
>>> +                break;
>>> +            }
>> ... and yet you still keep going in the outer loop?  Confused...
>
> Yes, I've not got this right at all.
>
>
>>>       }
>>>       unlock_mount_hash();
>>> +    up_read(&namespace_sem);
>>> + * count greater than the minimum reference count (ie. are in use).
>>> + */
>>> +int propagate_mount_tree_busy(struct mount *mnt, unsigned int flags)
>>> +{
>>> +    struct mount *m;
>>> +    struct mount *parent = mnt->mnt_parent;
>>> +    int refcnt = flags & TREE_BUSY_REFERENCED ? 2 : 1;
>>> +
>>> +    if (do_refcount_check(mnt, refcnt))
>>> +        return 1;
>>> +
>>> +    for (m = propagation_next(parent, parent); m;
>>> +            m = propagation_next(m, parent)) {
>>> +        struct mount *child;
>>> +
>>> +        child = __lookup_mnt(&m->mnt, mnt->mnt_mountpoint);
>>> +        if (!child)
>>> +            continue;
>>> +
>>> +        if (do_refcount_check(child, 1))
>>> +            return 1;
>>> +    }
>>> +    return 0;
>>> +}
>> What is the daemon expected to do with your subtree?  Take it apart with
>> a series of sync (== non-lazy) umount(2)?  I presume it is normal for
>> it to run into -EBUSY halfway through - i.e. get rid of some, but not
>> all of the subtree, right?
>
> All I need is to check if a mount and its children are in use, 
> essentially
>
> check if a mount subtree is in use in some way, working directory or open
>
> file(s).
>
>
> I think what I should be doing is very similar to what may_umount() 
> does but
>
> for the passed in mount and it's children with the adjustment for the 
> ref held
>
> on the passed in mount.
>
>
> I rather like your implementation of may_umount() and 
> propagate_mount_busy().
>
> I think the only difference to what I need is that the passed in mount 
> won't be
>
> the global root and will usually have children (so I'd invert that 
> list_empty()
>
> check or leave it out for this check).
>
>
> I'll have a go at that refactor and post it, we'll see if I can come 
> up with
>
> something acceptable.
>
>
> The other difficulty is that automount can be run in a separate mount 
> namespace
>
> (eg. a container) which introduces a special case environment. I'm not 
> sure how
>
> the propagation will behave in this case. Atm. I'm thinking it might 
> be sufficient
>
> to just use what may_umount_tree() is now so it deliberately only 
> checks the mount
>
> tree in it's own namespace.


So I clearly got this wrong so I've started over.

Unfortunately it looks like I have found a bug so I can get to the 
checking the propagation

slave behaviors I want to work on.


The "unshare -m" command will create a shell with a new mount namespace 
that is propagation

private.


If I mount something in the origin mount namespace and then run the 
unshare command it is

included in the new namespace.


But, with the series here, it looks like if I set my woring directory to 
this mount in the

new namespace and then umount the above mount in the origin namespace 
the umount gets propagated

to the created propagation private namespace and without regard for the 
process working directory.


The behavior in earlier kernels is the mount remains requiring it be 
umounted in the namespace

or be dissolved on namespace destruction.


Ian

[PATCHES v3][RFC][CFR] mount-related stuff

[Al Viro]

Updated variant (-rc4-based) force-pushed to
git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs.git #work.mount
individual patches in followups.  It seems to survive testing here, but
more testing and review would be very welcome.  Again, that is not all -
there's more stuff coming...

Folks, please review - if nobody objects, it goes into #for-next in
a day or two.

Changes since v2:
Fixes went into mainline.

Added change_mnt_propagation() stuff: cleanups and getting rid of
potentially O(N^2) work in umount() - when a long slave list gets moved
from one doomed mount to another, with O(list length) work on each move.

In the same area, mnt_slave_list/mnt_slave turned into hlist.

Added propagate_mnt() series - refactoring instead of brute-force
"pass a structure around instead of playing with globals".

Added a few ->mnt_group_id-related cleanups.

New: ##32--44,46--48
Slight changes in #16 (Rewrite of propagate_umount()) and #30
(mount: separate the flags accessed only under namespace_sem).

	Rough overview:

Part 1: getting rid of mount hash conflicts for good
1) attach_mnt(): expand in attach_recursive_mnt(), then lose the flag argument
2) get rid of mnt_set_mountpoint_beneath()
3) prevent mount hash conflicts

Part 2: trivial cleanups and helpers:
4) copy_tree(): don't set ->mnt_mountpoint on the root of copy
5) constify mnt_has_parent()
6) pnode: lift peers() into pnode.h
7) new predicate: mount_is_ancestor()
8) constify is_local_mountpoint()
9) new predicate: anon_ns_root(mount)
10) dissolve_on_fput(): use anon_ns_root()
11) __attach_mnt(): lose the second argument
12) don't set MNT_LOCKED on parentless mounts
13) clone_mnt(): simplify the propagation-related logics
14) do_umount(): simplify the "is it still mounted" checks

Part 3: (somewhat of a side story) restore the machinery for long-term
mounts from accumulated bitrot.
15) sanitize handling of long-term internal mounts
	Still unchanged; might end up moved on top of #work.fs_context
with its change of vfs_fs_parse_string() calling conventions.

Part 4: propagate_umount() rewrite (posted last cycle)
16) Rewrite of propagate_umount()

Part 5: untangling do_move_mount()/attach_recursive_mnt().
17) make commit_tree() usable in same-namespace move case
18) attach_recursive_mnt(): unify the mnt_change_mountpoint() logics
19) attach_recursive_mnt(): pass destination mount in all cases
20) attach_recursive_mnt(): get rid of flags entirely
21) do_move_mount(): take dropping the old mountpoint into attach_recursive_mnt()
22) do_move_mount(): get rid of 'attached' flag

Part 6: change locking for expiry lists.
23) attach_recursive_mnt(): remove from expiry list on move
24) take ->mnt_expire handling under mount_lock [read_seqlock_excl]

Part 7: struct mountpoint massage.
25) pivot_root(): reorder tree surgeries, collapse unhash_mnt() and put_mountpoint()
26) combine __put_mountpoint() with unhash_mnt()
27) get rid of mountpoint->m_count

Part 8: regularize mount refcounting a bit
28) don't have mounts pin their parents

Part 9: propagate_mnt() massage
29) mount: separate the flags accessed only under namespace_sem
30) propagate_one(): get rid of dest_master
31) propagate_mnt(): handle all peer groups in the same loop
32) propagate_one(): separate the "do we need secondary here?" logics
33) propagate_one(): separate the "what should be the master for this copy" part
34) propagate_one(): fold into the sole caller
35) fs/pnode.c: get rid of globals
36) propagate_mnt(): get rid of last_dest
37) propagate_mnt(): fix comment and convert to kernel-doc, while we are at it

Part 10: change_mnt_propagation() massage
38) change_mnt_propagation() cleanups, step 1
39) change_mnt_propagation(): do_make_slave() is a no-op unless IS_MNT_SHARED()
	these two are preliminary massage, getting do_make_slave() into shape for
40) do_make_slave(): choose new master sanely
	... getting rid of excessive work on umount().  The thing is, when
mount stops propagating events (e.g. when it gets taken out), we need to
transfer its slave list to its peer (if exists) or to its master.  If there's
neither, we need to dissolve that slave list.
	Each member of slave list needs at least to have ->mnt_master switched
to new value.  Unfortunately, if the chosen new master is itself getting taken
out on the same umount(2), the entire thing needs to be repeated there, etc.
and it doesn't take much to construct a situation when we have 2N mounts and
umount(2) taking out half of them will end up moving the slave list (consisting
of the other half) through all of those, resulting in N^2 reassignments of
->mnt_master alone.  Not hard to avoid, we just need to figure out where the
thing will settle and transfer it there from the very beginning.
41) turn do_make_slave() into transfer_propagation()
	cleanup, getting the things into convenient shape for...
42) mnt_slave_list/mnt_slave: turn into hlist_head/hlist_node
	what it says on the can.
43) change_mnt_propagation(): move ->mnt_master assignment into MS_SLAVE case
	finishing touches on the cleanups series.

Part 11: misc stuff, will grow...
44) copy_tree(): don't link the mounts via mnt_list
45) take freeing of emptied mnt_namespace to namespace_unlock()
46) get rid of CL_SHARE_TO_SLAVE
47) invent_group_ids(): zero ->mnt_group_id always implies !IS_MNT_SHARED()
48) statmount_mnt_basic(): simplify the logics for group id

Diffstat:
 Documentation/filesystems/propagate_umount.txt | 484 +++++++++++++++++
 drivers/gpu/drm/i915/gem/i915_gemfs.c          |  21 +-
 drivers/gpu/drm/v3d/v3d_gemfs.c                |  21 +-
 fs/hugetlbfs/inode.c                           |   2 +-
 fs/mount.h                                     |  40 +-
 fs/namespace.c                                 | 711 ++++++++++---------------
 fs/pnode.c                                     | 697 ++++++++++++------------
 fs/pnode.h                                     |  27 +-
 include/linux/mount.h                          |  18 +-
 ipc/mqueue.c                                   |   2 +-
 10 files changed, 1216 insertions(+), 807 deletions(-)
 create mode 100644 Documentation/filesystems/propagate_umount.txt

[PATCH v3 01/48] attach_mnt(): expand in attach_recursive_mnt(), then lose the flag argument

[Al Viro]

simpler that way - all but one caller pass false as 'beneath' argument,
and that one caller is actually happier with the call expanded - the
logics with choice of mountpoint is identical for 'moving' and 'attaching'
cases, and now that is no longer hidden.

Reviewed-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 37 ++++++++++++-------------------------
 1 file changed, 12 insertions(+), 25 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index 54c59e091919..1761d2c2fdae 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -1116,16 +1116,10 @@ static void __attach_mnt(struct mount *mnt, struct mount *parent)
  * @parent:  the parent
  * @mnt:     the new mount
  * @mp:      the new mountpoint
- * @beneath: whether to mount @mnt beneath or on top of @parent
  *
- * If @beneath is false, mount @mnt at @mp on @parent. Then attach @mnt
+ * Mount @mnt at @mp on @parent. Then attach @mnt
  * to @parent's child mount list and to @mount_hashtable.
  *
- * If @beneath is true, remove @mnt from its current parent and
- * mountpoint and mount it on @mp on @parent, and mount @parent on the
- * old parent and old mountpoint of @mnt. Finally, attach @parent to
- * @mnt_hashtable and @parent->mnt_parent->mnt_mounts.
- *
  * Note, when __attach_mnt() is called @mnt->mnt_parent already points
  * to the correct parent.
  *
@@ -1133,18 +1127,9 @@ static void __attach_mnt(struct mount *mnt, struct mount *parent)
  *          to have been acquired in that order.
  */
 static void attach_mnt(struct mount *mnt, struct mount *parent,
-		       struct mountpoint *mp, bool beneath)
+		       struct mountpoint *mp)
 {
-	if (beneath)
-		mnt_set_mountpoint_beneath(mnt, parent, mp);
-	else
-		mnt_set_mountpoint(parent, mp, mnt);
-	/*
-	 * Note, @mnt->mnt_parent has to be used. If @mnt was mounted
-	 * beneath @parent then @mnt will need to be attached to
-	 * @parent's old parent, not @parent. IOW, @mnt->mnt_parent
-	 * isn't the same mount as @parent.
-	 */
+	mnt_set_mountpoint(parent, mp, mnt);
 	__attach_mnt(mnt, mnt->mnt_parent);
 }
 
@@ -1157,7 +1142,7 @@ void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct m
 	hlist_del_init(&mnt->mnt_mp_list);
 	hlist_del_init_rcu(&mnt->mnt_hash);
 
-	attach_mnt(mnt, parent, mp, false);
+	attach_mnt(mnt, parent, mp);
 
 	put_mountpoint(old_mp);
 	mnt_add_count(old_parent, -1);
@@ -2295,7 +2280,7 @@ struct mount *copy_tree(struct mount *src_root, struct dentry *dentry,
 				goto out;
 			lock_mount_hash();
 			list_add_tail(&dst_mnt->mnt_list, &res->mnt_list);
-			attach_mnt(dst_mnt, dst_parent, src_parent->mnt_mp, false);
+			attach_mnt(dst_mnt, dst_parent, src_parent->mnt_mp);
 			unlock_mount_hash();
 		}
 	}
@@ -2743,10 +2728,12 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 	}
 
 	if (moving) {
-		if (beneath)
-			dest_mp = smp;
 		unhash_mnt(source_mnt);
-		attach_mnt(source_mnt, top_mnt, dest_mp, beneath);
+		if (beneath)
+			mnt_set_mountpoint_beneath(source_mnt, top_mnt, smp);
+		else
+			mnt_set_mountpoint(top_mnt, dest_mp, source_mnt);
+		__attach_mnt(source_mnt, source_mnt->mnt_parent);
 		mnt_notify_add(source_mnt);
 		touch_mnt_namespace(source_mnt->mnt_ns);
 	} else {
@@ -4827,9 +4814,9 @@ SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
 		root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
 	}
 	/* mount old root on put_old */
-	attach_mnt(root_mnt, old_mnt, old_mp, false);
+	attach_mnt(root_mnt, old_mnt, old_mp);
 	/* mount new_root on / */
-	attach_mnt(new_mnt, root_parent, root_mp, false);
+	attach_mnt(new_mnt, root_parent, root_mp);
 	mnt_add_count(root_parent, -1);
 	touch_mnt_namespace(current->nsproxy->mnt_ns);
 	/* A moved mount should not expire automatically */
-- 
2.39.5

[PATCH v3 02/48] get rid of mnt_set_mountpoint_beneath()

[Al Viro]

mnt_set_mountpoint_beneath() consists of attaching new mount side-by-side
with the one we want to mount beneath (by mnt_set_mountpoint()), followed
by mnt_change_mountpoint() shifting the the top mount onto the new one
(by mnt_change_mountpoint()).

Both callers of mnt_set_mountpoint_beneath (both in attach_recursive_mnt())
have the same form - in 'beneath' case we call mnt_set_mountpoint_beneath(),
otherwise - mnt_set_mountpoint().

The thing is, expressing that as unconditional mnt_set_mountpoint(),
followed, in 'beneath' case, by mnt_change_mountpoint() is just as easy.
And these mnt_change_mountpoint() callers are similar to the ones we
do when it comes to attaching propagated copies, which will allow more
cleanups in the next commits.

Reviewed-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 37 ++++---------------------------------
 1 file changed, 4 insertions(+), 33 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index 1761d2c2fdae..888816289154 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -1076,33 +1076,6 @@ void mnt_set_mountpoint(struct mount *mnt,
 	hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list);
 }
 
-/**
- * mnt_set_mountpoint_beneath - mount a mount beneath another one
- *
- * @new_parent: the source mount
- * @top_mnt:    the mount beneath which @new_parent is mounted
- * @new_mp:     the new mountpoint of @top_mnt on @new_parent
- *
- * Remove @top_mnt from its current mountpoint @top_mnt->mnt_mp and
- * parent @top_mnt->mnt_parent and mount it on top of @new_parent at
- * @new_mp. And mount @new_parent on the old parent and old
- * mountpoint of @top_mnt.
- *
- * Context: This function expects namespace_lock() and lock_mount_hash()
- *          to have been acquired in that order.
- */
-static void mnt_set_mountpoint_beneath(struct mount *new_parent,
-				       struct mount *top_mnt,
-				       struct mountpoint *new_mp)
-{
-	struct mount *old_top_parent = top_mnt->mnt_parent;
-	struct mountpoint *old_top_mp = top_mnt->mnt_mp;
-
-	mnt_set_mountpoint(old_top_parent, old_top_mp, new_parent);
-	mnt_change_mountpoint(new_parent, new_mp, top_mnt);
-}
-
-
 static void __attach_mnt(struct mount *mnt, struct mount *parent)
 {
 	hlist_add_head_rcu(&mnt->mnt_hash,
@@ -2729,10 +2702,9 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 
 	if (moving) {
 		unhash_mnt(source_mnt);
+		mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
 		if (beneath)
-			mnt_set_mountpoint_beneath(source_mnt, top_mnt, smp);
-		else
-			mnt_set_mountpoint(top_mnt, dest_mp, source_mnt);
+			mnt_change_mountpoint(source_mnt, smp, top_mnt);
 		__attach_mnt(source_mnt, source_mnt->mnt_parent);
 		mnt_notify_add(source_mnt);
 		touch_mnt_namespace(source_mnt->mnt_ns);
@@ -2745,10 +2717,9 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 				move_from_ns(p, &head);
 			list_del_init(&head);
 		}
+		mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
 		if (beneath)
-			mnt_set_mountpoint_beneath(source_mnt, top_mnt, smp);
-		else
-			mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
+			mnt_change_mountpoint(source_mnt, smp, top_mnt);
 		commit_tree(source_mnt);
 	}
 
-- 
2.39.5

[PATCH v3 03/48] prevent mount hash conflicts

[Al Viro]

Currently it's still possible to run into a pathological situation when
two hashed mounts share both parent and mountpoint.  That does not work
well, for obvious reasons.

We are not far from getting rid of that; the only remaining gap is
attach_recursive_mnt() not being careful enough when sliding a tree
under existing mount (for propagated copies or in 'beneath' case for
the original one).

To deal with that cleanly we need to be able to find overmounts
(i.e. mounts on top of parent's root); we could do hash lookups or scan
the list of children but either would be costly.  Since one of the results
we get from that will be prevention of multiple parallel overmounts, let's
just bite the bullet and store a (non-counting) reference to overmount
in struct mount.

With that done, closing the hole in attach_recursive_mnt() becomes easy
- we just need to follow the chain of overmounts before we change the
mountpoint of the mount we are sliding things under.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/mount.h     |  1 +
 fs/namespace.c | 27 ++++++++++++++++++++++-----
 2 files changed, 23 insertions(+), 5 deletions(-)

diff --git a/fs/mount.h b/fs/mount.h
index ad7173037924..b8beafdd6d24 100644
--- a/fs/mount.h
+++ b/fs/mount.h
@@ -92,6 +92,7 @@ struct mount {
 	int mnt_expiry_mark;		/* true if marked for expiry */
 	struct hlist_head mnt_pins;
 	struct hlist_head mnt_stuck_children;
+	struct mount *overmount;	/* mounted on ->mnt_root */
 } __randomize_layout;
 
 #define MNT_NS_INTERNAL ERR_PTR(-EINVAL) /* distinct from any mnt_namespace */
diff --git a/fs/namespace.c b/fs/namespace.c
index 888816289154..9b732d74c2cc 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -1043,6 +1043,9 @@ static void __touch_mnt_namespace(struct mnt_namespace *ns)
 static struct mountpoint *unhash_mnt(struct mount *mnt)
 {
 	struct mountpoint *mp;
+	struct mount *parent = mnt->mnt_parent;
+	if (unlikely(parent->overmount == mnt))
+		parent->overmount = NULL;
 	mnt->mnt_parent = mnt;
 	mnt->mnt_mountpoint = mnt->mnt.mnt_root;
 	list_del_init(&mnt->mnt_child);
@@ -1078,6 +1081,8 @@ void mnt_set_mountpoint(struct mount *mnt,
 
 static void __attach_mnt(struct mount *mnt, struct mount *parent)
 {
+	if (unlikely(mnt->mnt_mountpoint == parent->mnt.mnt_root))
+		parent->overmount = mnt;
 	hlist_add_head_rcu(&mnt->mnt_hash,
 			   m_hash(&parent->mnt, mnt->mnt_mountpoint));
 	list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
@@ -2660,7 +2665,9 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 	HLIST_HEAD(tree_list);
 	struct mnt_namespace *ns = top_mnt->mnt_ns;
 	struct mountpoint *smp;
+	struct mountpoint *secondary = NULL;
 	struct mount *child, *dest_mnt, *p;
+	struct mount *top;
 	struct hlist_node *n;
 	int err = 0;
 	bool moving = flags & MNT_TREE_MOVE, beneath = flags & MNT_TREE_BENEATH;
@@ -2669,9 +2676,15 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 	 * Preallocate a mountpoint in case the new mounts need to be
 	 * mounted beneath mounts on the same mountpoint.
 	 */
-	smp = get_mountpoint(source_mnt->mnt.mnt_root);
+	for (top = source_mnt; unlikely(top->overmount); top = top->overmount) {
+		if (!secondary && is_mnt_ns_file(top->mnt.mnt_root))
+			secondary = top->mnt_mp;
+	}
+	smp = get_mountpoint(top->mnt.mnt_root);
 	if (IS_ERR(smp))
 		return PTR_ERR(smp);
+	if (!secondary)
+		secondary = smp;
 
 	/* Is there space to add these mounts to the mount namespace? */
 	if (!moving) {
@@ -2704,7 +2717,7 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 		unhash_mnt(source_mnt);
 		mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
 		if (beneath)
-			mnt_change_mountpoint(source_mnt, smp, top_mnt);
+			mnt_change_mountpoint(top, smp, top_mnt);
 		__attach_mnt(source_mnt, source_mnt->mnt_parent);
 		mnt_notify_add(source_mnt);
 		touch_mnt_namespace(source_mnt->mnt_ns);
@@ -2719,7 +2732,7 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 		}
 		mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
 		if (beneath)
-			mnt_change_mountpoint(source_mnt, smp, top_mnt);
+			mnt_change_mountpoint(top, smp, top_mnt);
 		commit_tree(source_mnt);
 	}
 
@@ -2732,8 +2745,12 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 		child->mnt.mnt_flags &= ~MNT_LOCKED;
 		q = __lookup_mnt(&child->mnt_parent->mnt,
 				 child->mnt_mountpoint);
-		if (q)
-			mnt_change_mountpoint(child, smp, q);
+		if (q) {
+			struct mount *r = child;
+			while (unlikely(r->overmount))
+				r = r->overmount;
+			mnt_change_mountpoint(r, secondary, q);
+		}
 		commit_tree(child);
 	}
 	put_mountpoint(smp);
-- 
2.39.5

[PATCH v3 04/48] copy_tree(): don't set ->mnt_mountpoint on the root of copy

[Al Viro]

It never made any sense - neither when copy_tree() had been introduced
(2.4.11-pre5), nor at any point afterwards.  Mountpoint is meaningless
without parent mount and the root of copied tree has no parent until we get
around to attaching it somewhere.  At that time we'll have mountpoint set;
before that we have no idea which dentry will be used as mountpoint.
IOW, copy_tree() should just leave the default value.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 1 -
 1 file changed, 1 deletion(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index 9b732d74c2cc..f0a56dbceff9 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -2222,7 +2222,6 @@ struct mount *copy_tree(struct mount *src_root, struct dentry *dentry,
 		return dst_mnt;
 
 	src_parent = src_root;
-	dst_mnt->mnt_mountpoint = src_root->mnt_mountpoint;
 
 	list_for_each_entry(src_root_child, &src_root->mnt_mounts, mnt_child) {
 		if (!is_subdir(src_root_child->mnt_mountpoint, dentry))
-- 
2.39.5

[PATCH v3 05/48] constify mnt_has_parent()

[Al Viro]

Reviewed-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/mount.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/fs/mount.h b/fs/mount.h
index b8beafdd6d24..c4d417cd7953 100644
--- a/fs/mount.h
+++ b/fs/mount.h
@@ -102,7 +102,7 @@ static inline struct mount *real_mount(struct vfsmount *mnt)
 	return container_of(mnt, struct mount, mnt);
 }
 
-static inline int mnt_has_parent(struct mount *mnt)
+static inline int mnt_has_parent(const struct mount *mnt)
 {
 	return mnt != mnt->mnt_parent;
 }
-- 
2.39.5

[PATCH v3 06/48] pnode: lift peers() into pnode.h

[Al Viro]

it's going to be useful both in pnode.c and namespace.c

Reviewed-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/pnode.c | 5 -----
 fs/pnode.h | 5 +++++
 2 files changed, 5 insertions(+), 5 deletions(-)

diff --git a/fs/pnode.c b/fs/pnode.c
index ffd429b760d5..aa187144e389 100644
--- a/fs/pnode.c
+++ b/fs/pnode.c
@@ -216,11 +216,6 @@ static struct mount *next_group(struct mount *m, struct mount *origin)
 static struct mount *last_dest, *first_source, *last_source, *dest_master;
 static struct hlist_head *list;
 
-static inline bool peers(const struct mount *m1, const struct mount *m2)
-{
-	return m1->mnt_group_id == m2->mnt_group_id && m1->mnt_group_id;
-}
-
 static int propagate_one(struct mount *m, struct mountpoint *dest_mp)
 {
 	struct mount *child;
diff --git a/fs/pnode.h b/fs/pnode.h
index 2d026fb98b18..93fa9311bd07 100644
--- a/fs/pnode.h
+++ b/fs/pnode.h
@@ -34,6 +34,11 @@ static inline void set_mnt_shared(struct mount *mnt)
 	mnt->mnt.mnt_flags |= MNT_SHARED;
 }
 
+static inline bool peers(const struct mount *m1, const struct mount *m2)
+{
+	return m1->mnt_group_id == m2->mnt_group_id && m1->mnt_group_id;
+}
+
 void change_mnt_propagation(struct mount *, int);
 int propagate_mnt(struct mount *, struct mountpoint *, struct mount *,
 		struct hlist_head *);
-- 
2.39.5

[PATCH v3 07/48] new predicate: mount_is_ancestor()

[Al Viro]

mount_is_ancestor(p1, p2) returns true iff there is a possibly
empty ancestry chain from p1 to p2.

Convert the open-coded checks.  Unlike those open-coded variants
it does not depend upon p1 not being root...

Reviewed-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 21 +++++++++++++++------
 1 file changed, 15 insertions(+), 6 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index f0a56dbceff9..aa93e1a48b5d 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -3483,6 +3483,17 @@ static inline bool path_overmounted(const struct path *path)
 	return unlikely(!no_child);
 }
 
+/*
+ * Check if there is a possibly empty chain of descent from p1 to p2.
+ * Locks: namespace_sem (shared) or mount_lock (read_seqlock_excl).
+ */
+static bool mount_is_ancestor(const struct mount *p1, const struct mount *p2)
+{
+	while (p2 != p1 && mnt_has_parent(p2))
+		p2 = p2->mnt_parent;
+	return p2 == p1;
+}
+
 /**
  * can_move_mount_beneath - check that we can mount beneath the top mount
  * @from: mount to mount beneath
@@ -3534,9 +3545,8 @@ static int can_move_mount_beneath(const struct path *from,
 	if (parent_mnt_to == current->nsproxy->mnt_ns->root)
 		return -EINVAL;
 
-	for (struct mount *p = mnt_from; mnt_has_parent(p); p = p->mnt_parent)
-		if (p == mnt_to)
-			return -EINVAL;
+	if (mount_is_ancestor(mnt_to, mnt_from))
+		return -EINVAL;
 
 	/*
 	 * If the parent mount propagates to the child mount this would
@@ -3705,9 +3715,8 @@ static int do_move_mount(struct path *old_path,
 	err = -ELOOP;
 	if (!check_for_nsfs_mounts(old))
 		goto out;
-	for (; mnt_has_parent(p); p = p->mnt_parent)
-		if (p == old)
-			goto out;
+	if (mount_is_ancestor(old, p))
+		goto out;
 
 	err = attach_recursive_mnt(old, real_mount(new_path->mnt), mp, flags);
 	if (err)
-- 
2.39.5

[PATCH v3 08/48] constify is_local_mountpoint()

[Al Viro]

Reviewed-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/mount.h     | 4 ++--
 fs/namespace.c | 2 +-
 2 files changed, 3 insertions(+), 3 deletions(-)

diff --git a/fs/mount.h b/fs/mount.h
index c4d417cd7953..f10776003643 100644
--- a/fs/mount.h
+++ b/fs/mount.h
@@ -147,8 +147,8 @@ struct proc_mounts {
 
 extern const struct seq_operations mounts_op;
 
-extern bool __is_local_mountpoint(struct dentry *dentry);
-static inline bool is_local_mountpoint(struct dentry *dentry)
+extern bool __is_local_mountpoint(const struct dentry *dentry);
+static inline bool is_local_mountpoint(const struct dentry *dentry)
 {
 	if (!d_mountpoint(dentry))
 		return false;
diff --git a/fs/namespace.c b/fs/namespace.c
index aa93e1a48b5d..c4feb8315978 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -894,7 +894,7 @@ struct vfsmount *lookup_mnt(const struct path *path)
  * namespace not just a mount that happens to have some specified
  * parent mount.
  */
-bool __is_local_mountpoint(struct dentry *dentry)
+bool __is_local_mountpoint(const struct dentry *dentry)
 {
 	struct mnt_namespace *ns = current->nsproxy->mnt_ns;
 	struct mount *mnt, *n;
-- 
2.39.5

[PATCH v3 09/48] new predicate: anon_ns_root(mount)

[Al Viro]

checks if mount is the root of an anonymouns namespace.
Switch open-coded equivalents to using it.

For mounts that belong to anon namespace !mnt_has_parent(mount)
is the same as mount == ns->root, and intent is more obvious in
the latter form.

Reviewed-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/mount.h     |  7 +++++++
 fs/namespace.c | 28 +++-------------------------
 2 files changed, 10 insertions(+), 25 deletions(-)

diff --git a/fs/mount.h b/fs/mount.h
index f10776003643..f20e6ed845fe 100644
--- a/fs/mount.h
+++ b/fs/mount.h
@@ -161,6 +161,13 @@ static inline bool is_anon_ns(struct mnt_namespace *ns)
 	return ns->seq == 0;
 }
 
+static inline bool anon_ns_root(const struct mount *m)
+{
+	struct mnt_namespace *ns = READ_ONCE(m->mnt_ns);
+
+	return !IS_ERR_OR_NULL(ns) && is_anon_ns(ns) && m == ns->root;
+}
+
 static inline bool mnt_ns_attached(const struct mount *mnt)
 {
 	return !RB_EMPTY_NODE(&mnt->mnt_node);
diff --git a/fs/namespace.c b/fs/namespace.c
index c4feb8315978..ea01fea2ac93 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -2480,9 +2480,7 @@ struct vfsmount *clone_private_mount(const struct path *path)
 	 * loops get created.
 	 */
 	if (!check_mnt(old_mnt)) {
-		if (!is_mounted(&old_mnt->mnt) ||
-			!is_anon_ns(old_mnt->mnt_ns) ||
-			mnt_has_parent(old_mnt))
+		if (!anon_ns_root(old_mnt))
 			return ERR_PTR(-EINVAL);
 
 		if (!check_for_nsfs_mounts(old_mnt))
@@ -3649,9 +3647,6 @@ static int do_move_mount(struct path *old_path,
 	ns = old->mnt_ns;
 
 	err = -EINVAL;
-	/* The thing moved must be mounted... */
-	if (!is_mounted(&old->mnt))
-		goto out;
 
 	if (check_mnt(old)) {
 		/* if the source is in our namespace... */
@@ -3664,10 +3659,8 @@ static int do_move_mount(struct path *old_path,
 	} else {
 		/*
 		 * otherwise the source must be the root of some anon namespace.
-		 * AV: check for mount being root of an anon namespace is worth
-		 * an inlined predicate...
 		 */
-		if (!is_anon_ns(ns) || mnt_has_parent(old))
+		if (!anon_ns_root(old))
 			goto out;
 		/*
 		 * Bail out early if the target is within the same namespace -
@@ -5028,22 +5021,7 @@ static int do_mount_setattr(struct path *path, struct mount_kattr *kattr)
 	err = -EINVAL;
 	lock_mount_hash();
 
-	/* Ensure that this isn't anything purely vfs internal. */
-	if (!is_mounted(&mnt->mnt))
-		goto out;
-
-	/*
-	 * If this is an attached mount make sure it's located in the callers
-	 * mount namespace. If it's not don't let the caller interact with it.
-	 *
-	 * If this mount doesn't have a parent it's most often simply a
-	 * detached mount with an anonymous mount namespace. IOW, something
-	 * that's simply not attached yet. But there are apparently also users
-	 * that do change mount properties on the rootfs itself. That obviously
-	 * neither has a parent nor is it a detached mount so we cannot
-	 * unconditionally check for detached mounts.
-	 */
-	if ((mnt_has_parent(mnt) || !is_anon_ns(mnt->mnt_ns)) && !check_mnt(mnt))
+	if (!anon_ns_root(mnt) && !check_mnt(mnt))
 		goto out;
 
 	/*
-- 
2.39.5

[PATCH v3 10/48] dissolve_on_fput(): use anon_ns_root()

[Al Viro]

that's the condition we are actually trying to check there...

Reviewed-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 62 +++++++++++---------------------------------------
 1 file changed, 13 insertions(+), 49 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index ea01fea2ac93..151d5f3360b9 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -2333,67 +2333,31 @@ void drop_collected_paths(struct path *paths, struct path *prealloc)
 static void free_mnt_ns(struct mnt_namespace *);
 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *, bool);
 
-static inline bool must_dissolve(struct mnt_namespace *mnt_ns)
-{
-	/*
-        * This mount belonged to an anonymous mount namespace
-        * but was moved to a non-anonymous mount namespace and
-        * then unmounted.
-        */
-	if (unlikely(!mnt_ns))
-		return false;
-
-	/*
-        * This mount belongs to a non-anonymous mount namespace
-        * and we know that such a mount can never transition to
-        * an anonymous mount namespace again.
-        */
-	if (!is_anon_ns(mnt_ns)) {
-		/*
-		 * A detached mount either belongs to an anonymous mount
-		 * namespace or a non-anonymous mount namespace. It
-		 * should never belong to something purely internal.
-		 */
-		VFS_WARN_ON_ONCE(mnt_ns == MNT_NS_INTERNAL);
-		return false;
-	}
-
-	return true;
-}
-
 void dissolve_on_fput(struct vfsmount *mnt)
 {
 	struct mnt_namespace *ns;
 	struct mount *m = real_mount(mnt);
 
+	/*
+	 * m used to be the root of anon namespace; if it still is one,
+	 * we need to dissolve the mount tree and free that namespace.
+	 * Let's try to avoid taking namespace_sem if we can determine
+	 * that there's nothing to do without it - rcu_read_lock() is
+	 * enough to make anon_ns_root() memory-safe and once m has
+	 * left its namespace, it's no longer our concern, since it will
+	 * never become a root of anon ns again.
+	 */
+
 	scoped_guard(rcu) {
-		if (!must_dissolve(READ_ONCE(m->mnt_ns)))
+		if (!anon_ns_root(m))
 			return;
 	}
 
 	scoped_guard(namespace_lock, &namespace_sem) {
-		ns = m->mnt_ns;
-		if (!must_dissolve(ns))
-			return;
-
-		/*
-		 * After must_dissolve() we know that this is a detached
-		 * mount in an anonymous mount namespace.
-		 *
-		 * Now when mnt_has_parent() reports that this mount
-		 * tree has a parent, we know that this anonymous mount
-		 * tree has been moved to another anonymous mount
-		 * namespace.
-		 *
-		 * So when closing this file we cannot unmount the mount
-		 * tree. This will be done when the file referring to
-		 * the root of the anonymous mount namespace will be
-		 * closed (It could already be closed but it would sync
-		 * on @namespace_sem and wait for us to finish.).
-		 */
-		if (mnt_has_parent(m))
+		if (!anon_ns_root(m))
 			return;
 
+		ns = m->mnt_ns;
 		lock_mount_hash();
 		umount_tree(m, UMOUNT_CONNECTED);
 		unlock_mount_hash();
-- 
2.39.5

[PATCH v3 11/48] __attach_mnt(): lose the second argument

[Al Viro]

It's always ->mnt_parent of the first one.  What the function does is
making a mount (with already set parent and mountpoint) visible - in
mount hash and in the parent's list of children.

IOW, it takes the existing rootwards linkage and sets the matching
crownwards linkage.

Renamed to make_visible(), while we are at it.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 11 ++++++-----
 1 file changed, 6 insertions(+), 5 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index 151d5f3360b9..75d45d0b615c 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -1079,8 +1079,9 @@ void mnt_set_mountpoint(struct mount *mnt,
 	hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list);
 }
 
-static void __attach_mnt(struct mount *mnt, struct mount *parent)
+static void make_visible(struct mount *mnt)
 {
+	struct mount *parent = mnt->mnt_parent;
 	if (unlikely(mnt->mnt_mountpoint == parent->mnt.mnt_root))
 		parent->overmount = mnt;
 	hlist_add_head_rcu(&mnt->mnt_hash,
@@ -1098,7 +1099,7 @@ static void __attach_mnt(struct mount *mnt, struct mount *parent)
  * Mount @mnt at @mp on @parent. Then attach @mnt
  * to @parent's child mount list and to @mount_hashtable.
  *
- * Note, when __attach_mnt() is called @mnt->mnt_parent already points
+ * Note, when make_visible() is called @mnt->mnt_parent already points
  * to the correct parent.
  *
  * Context: This function expects namespace_lock() and lock_mount_hash()
@@ -1108,7 +1109,7 @@ static void attach_mnt(struct mount *mnt, struct mount *parent,
 		       struct mountpoint *mp)
 {
 	mnt_set_mountpoint(parent, mp, mnt);
-	__attach_mnt(mnt, mnt->mnt_parent);
+	make_visible(mnt);
 }
 
 void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt)
@@ -1182,7 +1183,7 @@ static void commit_tree(struct mount *mnt)
 	n->nr_mounts += n->pending_mounts;
 	n->pending_mounts = 0;
 
-	__attach_mnt(mnt, parent);
+	make_visible(mnt);
 	touch_mnt_namespace(n);
 }
 
@@ -2679,7 +2680,7 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 		mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
 		if (beneath)
 			mnt_change_mountpoint(top, smp, top_mnt);
-		__attach_mnt(source_mnt, source_mnt->mnt_parent);
+		make_visible(source_mnt);
 		mnt_notify_add(source_mnt);
 		touch_mnt_namespace(source_mnt->mnt_ns);
 	} else {
-- 
2.39.5

[PATCH v3 12/48] don't set MNT_LOCKED on parentless mounts

[Al Viro]

Originally MNT_LOCKED meant only one thing - "don't let this mount to
be peeled off its parent, we don't want to have its mountpoint exposed".
Accordingly, it had only been set on mounts that *do* have a parent.
Later it got overloaded with another use - setting it on the absolute
root had given free protection against umount(2) of absolute root
(was possible to trigger, oopsed).  Not a bad trick, but it ended
up costing more than it bought us.  Unfortunately, the cost included
both hard-to-reason-about logics and a subtle race between
mount -o remount,ro and mount --[r]bind - lockless &= ~MNT_LOCKED in
the end of __do_loopback() could race with sb_prepare_remount_readonly()
setting and clearing MNT_HOLD_WRITE (under mount_lock, as it should
be).  The race wouldn't be much of a problem (there are other ways to
deal with it), but the subtlety is.

Turns out that nobody except umount(2) had ever made use of having
MNT_LOCKED set on absolute root.  So let's give up on that trick,
clever as it had been, add an explicit check in do_umount() and
return to using MNT_LOCKED only for mounts that have a parent.

It means that
	* clone_mnt() no longer copies MNT_LOCKED
	* copy_tree() sets it on submounts if their counterparts had
been marked such, and does that right next to attach_mnt() in there,
in the same mount_lock scope.
	* __do_loopback() no longer needs to strip MNT_LOCKED off the
root of subtree it's about to return; no store, no race.
	* init_mount_tree() doesn't bother setting MNT_LOCKED on absolute
root.
	* lock_mnt_tree() does not set MNT_LOCKED on the subtree's root;
accordingly, its caller (loop in attach_recursive_mnt()) does not need to
bother stripping that MNT_LOCKED on root.  Note that lock_mnt_tree() setting
MNT_LOCKED on submounts happens in the same mount_lock scope as __attach_mnt()
(from commit_tree()) that makes them reachable.

Reviewed-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 32 +++++++++++++++-----------------
 1 file changed, 15 insertions(+), 17 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index 75d45d0b615c..791904128f1e 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -1313,7 +1313,7 @@ static struct mount *clone_mnt(struct mount *old, struct dentry *root,
 	}
 
 	mnt->mnt.mnt_flags = old->mnt.mnt_flags;
-	mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL);
+	mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL|MNT_LOCKED);
 
 	atomic_inc(&sb->s_active);
 	mnt->mnt.mnt_idmap = mnt_idmap_get(mnt_idmap(&old->mnt));
@@ -1988,6 +1988,9 @@ static int do_umount(struct mount *mnt, int flags)
 	if (mnt->mnt.mnt_flags & MNT_LOCKED)
 		goto out;
 
+	if (!mnt_has_parent(mnt)) /* not the absolute root */
+		goto out;
+
 	event++;
 	if (flags & MNT_DETACH) {
 		if (mnt_ns_attached(mnt) || !list_empty(&mnt->mnt_list))
@@ -2257,6 +2260,8 @@ struct mount *copy_tree(struct mount *src_root, struct dentry *dentry,
 			if (IS_ERR(dst_mnt))
 				goto out;
 			lock_mount_hash();
+			if (src_mnt->mnt.mnt_flags & MNT_LOCKED)
+				dst_mnt->mnt.mnt_flags |= MNT_LOCKED;
 			list_add_tail(&dst_mnt->mnt_list, &res->mnt_list);
 			attach_mnt(dst_mnt, dst_parent, src_parent->mnt_mp);
 			unlock_mount_hash();
@@ -2489,7 +2494,7 @@ static void lock_mnt_tree(struct mount *mnt)
 		if (flags & MNT_NOEXEC)
 			flags |= MNT_LOCK_NOEXEC;
 		/* Don't allow unprivileged users to reveal what is under a mount */
-		if (list_empty(&p->mnt_expire))
+		if (list_empty(&p->mnt_expire) && p != mnt)
 			flags |= MNT_LOCKED;
 		p->mnt.mnt_flags = flags;
 	}
@@ -2704,7 +2709,6 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 		/* Notice when we are propagating across user namespaces */
 		if (child->mnt_parent->mnt_ns->user_ns != user_ns)
 			lock_mnt_tree(child);
-		child->mnt.mnt_flags &= ~MNT_LOCKED;
 		q = __lookup_mnt(&child->mnt_parent->mnt,
 				 child->mnt_mountpoint);
 		if (q) {
@@ -2985,26 +2989,21 @@ static inline bool may_copy_tree(struct path *path)
 
 static struct mount *__do_loopback(struct path *old_path, int recurse)
 {
-	struct mount *mnt = ERR_PTR(-EINVAL), *old = real_mount(old_path->mnt);
+	struct mount *old = real_mount(old_path->mnt);
 
 	if (IS_MNT_UNBINDABLE(old))
-		return mnt;
+		return ERR_PTR(-EINVAL);
 
 	if (!may_copy_tree(old_path))
-		return mnt;
+		return ERR_PTR(-EINVAL);
 
 	if (!recurse && __has_locked_children(old, old_path->dentry))
-		return mnt;
+		return ERR_PTR(-EINVAL);
 
 	if (recurse)
-		mnt = copy_tree(old, old_path->dentry, CL_COPY_MNT_NS_FILE);
+		return copy_tree(old, old_path->dentry, CL_COPY_MNT_NS_FILE);
 	else
-		mnt = clone_mnt(old, old_path->dentry, 0);
-
-	if (!IS_ERR(mnt))
-		mnt->mnt.mnt_flags &= ~MNT_LOCKED;
-
-	return mnt;
+		return clone_mnt(old, old_path->dentry, 0);
 }
 
 /*
@@ -4749,11 +4748,11 @@ SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
 	if (!path_mounted(&root))
 		goto out4; /* not a mountpoint */
 	if (!mnt_has_parent(root_mnt))
-		goto out4; /* not attached */
+		goto out4; /* absolute root */
 	if (!path_mounted(&new))
 		goto out4; /* not a mountpoint */
 	if (!mnt_has_parent(new_mnt))
-		goto out4; /* not attached */
+		goto out4; /* absolute root */
 	/* make sure we can reach put_old from new_root */
 	if (!is_path_reachable(old_mnt, old.dentry, &new))
 		goto out4;
@@ -6154,7 +6153,6 @@ static void __init init_mount_tree(void)
 
 	root.mnt = mnt;
 	root.dentry = mnt->mnt_root;
-	mnt->mnt_flags |= MNT_LOCKED;
 
 	set_fs_pwd(current->fs, &root);
 	set_fs_root(current->fs, &root);
-- 
2.39.5

[PATCH v3 13/48] clone_mnt(): simplify the propagation-related logics

[Al Viro]

The underlying rules are simple:
	* MNT_SHARED should be set iff ->mnt_group_id of new mount ends up
non-zero.
	* mounts should be on the same ->mnt_share cyclic list iff they have
the same non-zero ->mnt_group_id value.
	* CL_PRIVATE is mutually exclusive with MNT_SHARED, MNT_SLAVE,
MNT_SHARED_TO_SLAVE and MNT_EXPIRE; the whole point of that thing is to
get a clone of old mount that would *not* be on any namespace-related
lists.

The above allows to make the logics more straightforward; what's more,
it makes the proof that invariants are maintained much simpler.
The variant in mainline is safe (aside of a very narrow race with
unsafe modification of mnt_flags right after we had the mount exposed
in superblock's ->s_mounts; theoretically it can race with ro remount
of the original, but it's not easy to hit), but proof of its correctness
is really unpleasant.

Reviewed-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 26 +++++++++++++-------------
 1 file changed, 13 insertions(+), 13 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index 791904128f1e..12cf69da4320 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -1301,6 +1301,9 @@ static struct mount *clone_mnt(struct mount *old, struct dentry *root,
 	if (!mnt)
 		return ERR_PTR(-ENOMEM);
 
+	mnt->mnt.mnt_flags = READ_ONCE(old->mnt.mnt_flags) &
+			     ~MNT_INTERNAL_FLAGS;
+
 	if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
 		mnt->mnt_group_id = 0; /* not a peer of original */
 	else
@@ -1312,8 +1315,8 @@ static struct mount *clone_mnt(struct mount *old, struct dentry *root,
 			goto out_free;
 	}
 
-	mnt->mnt.mnt_flags = old->mnt.mnt_flags;
-	mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL|MNT_LOCKED);
+	if (mnt->mnt_group_id)
+		set_mnt_shared(mnt);
 
 	atomic_inc(&sb->s_active);
 	mnt->mnt.mnt_idmap = mnt_idmap_get(mnt_idmap(&old->mnt));
@@ -1326,22 +1329,20 @@ static struct mount *clone_mnt(struct mount *old, struct dentry *root,
 	list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
 	unlock_mount_hash();
 
+	if (flag & CL_PRIVATE)	// we are done with it
+		return mnt;
+
+	if (peers(mnt, old))
+		list_add(&mnt->mnt_share, &old->mnt_share);
+
 	if ((flag & CL_SLAVE) ||
 	    ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
 		list_add(&mnt->mnt_slave, &old->mnt_slave_list);
 		mnt->mnt_master = old;
-		CLEAR_MNT_SHARED(mnt);
-	} else if (!(flag & CL_PRIVATE)) {
-		if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
-			list_add(&mnt->mnt_share, &old->mnt_share);
-		if (IS_MNT_SLAVE(old))
-			list_add(&mnt->mnt_slave, &old->mnt_slave);
+	} else if (IS_MNT_SLAVE(old)) {
+		list_add(&mnt->mnt_slave, &old->mnt_slave);
 		mnt->mnt_master = old->mnt_master;
-	} else {
-		CLEAR_MNT_SHARED(mnt);
 	}
-	if (flag & CL_MAKE_SHARED)
-		set_mnt_shared(mnt);
 
 	/* stick the duplicate mount on the same expiry list
 	 * as the original if that was on one */
@@ -1349,7 +1350,6 @@ static struct mount *clone_mnt(struct mount *old, struct dentry *root,
 		if (!list_empty(&old->mnt_expire))
 			list_add(&mnt->mnt_expire, &old->mnt_expire);
 	}
-
 	return mnt;
 
  out_free:
-- 
2.39.5

[PATCH v3 14/48] do_umount(): simplify the "is it still mounted" checks

[Al Viro]

Calls of do_umount() are always preceded by can_umount(), where we'd
done a racy check for mount belonging to our namespace; if it wasn't,
can_unmount() would've failed with -EINVAL and we wouldn't have
reached do_umount() at all.

That check needs to be redone once we have acquired namespace_sem
and in do_umount() we do that.  However, that's done in a very odd
way; we check that mount is still in rbtree of _some_ namespace or
its mnt_list is not empty.  It is equivalent to check_mnt(mnt) -
we know that earlier mnt was mounted in our namespace; if it has
stayed there, it's going to remain in rbtree of our namespace.
OTOH, if it ever had been removed from out namespace, it would be
removed from rbtree and it never would've re-added to a namespace
afterwards.  As for ->mnt_list, for something that had been mounted
in a namespace we'll never observe non-empty ->mnt_list while holding
namespace_sem - it does temporarily become non-empty during
umount_tree(), but that doesn't outlast the call of umount_tree(),
let alone dropping namespace_sem.

Things get much easier to follow if we replace that with (equivalent)
check_mnt(mnt) there.  What's more, currently we treat a failure of
that test as "quietly do nothing"; we might as well pretend that we'd
lost the race and fail on that the same way can_umount() would have.

Reviewed-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 11 ++++++-----
 1 file changed, 6 insertions(+), 5 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index 12cf69da4320..57b0974a5d1e 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -1983,8 +1983,11 @@ static int do_umount(struct mount *mnt, int flags)
 	namespace_lock();
 	lock_mount_hash();
 
-	/* Recheck MNT_LOCKED with the locks held */
+	/* Repeat the earlier racy checks, now that we are holding the locks */
 	retval = -EINVAL;
+	if (!check_mnt(mnt))
+		goto out;
+
 	if (mnt->mnt.mnt_flags & MNT_LOCKED)
 		goto out;
 
@@ -1993,16 +1996,14 @@ static int do_umount(struct mount *mnt, int flags)
 
 	event++;
 	if (flags & MNT_DETACH) {
-		if (mnt_ns_attached(mnt) || !list_empty(&mnt->mnt_list))
-			umount_tree(mnt, UMOUNT_PROPAGATE);
+		umount_tree(mnt, UMOUNT_PROPAGATE);
 		retval = 0;
 	} else {
 		smp_mb(); // paired with __legitimize_mnt()
 		shrink_submounts(mnt);
 		retval = -EBUSY;
 		if (!propagate_mount_busy(mnt, 2)) {
-			if (mnt_ns_attached(mnt) || !list_empty(&mnt->mnt_list))
-				umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
+			umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
 			retval = 0;
 		}
 	}
-- 
2.39.5

[PATCH v3 15/48] sanitize handling of long-term internal mounts

[Al Viro]

Original rationale for those had been the reduced cost of mntput()
for the stuff that is mounted somewhere.  Mount refcount increments and
decrements are frequent; what's worse, they tend to concentrate on the
same instances and cacheline pingpong is quite noticable.

As the result, mount refcounts are per-cpu; that allows a very cheap
increment.  Plain decrement would be just as easy, but decrement-and-test
is anything but (we need to add the components up, with exclusion against
possible increment-from-zero, etc.).

Fortunately, there is a very common case where we can tell that decrement
won't be the final one - if the thing we are dropping is currently
mounted somewhere.  We have an RCU delay between the removal from mount
tree and dropping the reference that used to pin it there, so we can
just take rcu_read_lock() and check if the victim is mounted somewhere.
If it is, we can go ahead and decrement without and further checks -
the reference we are dropping is not the last one.  If it isn't, we
get all the fun with locking, carefully adding up components, etc.,
but the majority of refcount decrements end up taking the fast path.

There is a major exception, though - pipes and sockets.  Those live
on the internal filesystems that are not going to be mounted anywhere.
They are not going to be _un_mounted, of course, so having to take the
slow path every time a pipe or socket gets closed is really obnoxious.
Solution had been to mark them as long-lived ones - essentially faking
"they are mounted somewhere" indicator.

With minor modification that works even for ones that do eventually get
dropped - all it takes is making sure we have an RCU delay between
clearing the "mounted somewhere" indicator and dropping the reference.

There are some additional twists (if you want to drop a dozen of such
internal mounts, you'd be better off with clearing the indicator on
all of them, doing an RCU delay once, then dropping the references),
but in the basic form it had been
	* use kern_mount() if you want your internal mount to be
a long-term one.
	* use kern_unmount() to undo that.

Unfortunately, the things did rot a bit during the mount API reshuffling.
In several cases we have lost the "fake the indicator" part; kern_unmount()
on the unmount side remained (it doesn't warn if you use it on a mount
without the indicator), but all benefits regaring mntput() cost had been
lost.

To get rid of that bitrot, let's add a new helper that would work
with fs_context-based API: fc_mount_longterm().  It's a counterpart
of fc_mount() that does, on success, mark its result as long-term.
It must be paired with kern_unmount() or equivalents.

Converted:
	1) mqueue (it used to use kern_mount_data() and the umount side
is still as it used to be)
	2) hugetlbfs (used to use kern_mount_data(), internal mount is
never unmounted in this one)
	3) i915 gemfs (used to be kern_mount() + manual remount to set
options, still uses kern_unmount() on umount side)
	4) v3d gemfs (copied from i915)

Reviewed-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 drivers/gpu/drm/i915/gem/i915_gemfs.c | 21 ++++++++++++++++++---
 drivers/gpu/drm/v3d/v3d_gemfs.c       | 21 ++++++++++++++++++---
 fs/hugetlbfs/inode.c                  |  2 +-
 fs/namespace.c                        |  9 +++++++++
 include/linux/mount.h                 |  1 +
 ipc/mqueue.c                          |  2 +-
 6 files changed, 48 insertions(+), 8 deletions(-)

diff --git a/drivers/gpu/drm/i915/gem/i915_gemfs.c b/drivers/gpu/drm/i915/gem/i915_gemfs.c
index 65d84a93c525..a09e2eb47175 100644
--- a/drivers/gpu/drm/i915/gem/i915_gemfs.c
+++ b/drivers/gpu/drm/i915/gem/i915_gemfs.c
@@ -5,16 +5,23 @@
 
 #include <linux/fs.h>
 #include <linux/mount.h>
+#include <linux/fs_context.h>
 
 #include "i915_drv.h"
 #include "i915_gemfs.h"
 #include "i915_utils.h"
 
+static int add_param(struct fs_context *fc, const char *key, const char *val)
+{
+	return vfs_parse_fs_string(fc, key, val, strlen(val));
+}
+
 void i915_gemfs_init(struct drm_i915_private *i915)
 {
-	char huge_opt[] = "huge=within_size"; /* r/w */
 	struct file_system_type *type;
+	struct fs_context *fc;
 	struct vfsmount *gemfs;
+	int ret;
 
 	/*
 	 * By creating our own shmemfs mountpoint, we can pass in
@@ -38,8 +45,16 @@ void i915_gemfs_init(struct drm_i915_private *i915)
 	if (!type)
 		goto err;
 
-	gemfs = vfs_kern_mount(type, SB_KERNMOUNT, type->name, huge_opt);
-	if (IS_ERR(gemfs))
+	fc = fs_context_for_mount(type, SB_KERNMOUNT);
+	if (IS_ERR(fc))
+		goto err;
+	ret = add_param(fc, "source", "tmpfs");
+	if (!ret)
+		ret = add_param(fc, "huge", "within_size");
+	if (!ret)
+		gemfs = fc_mount_longterm(fc);
+	put_fs_context(fc);
+	if (ret)
 		goto err;
 
 	i915->mm.gemfs = gemfs;
diff --git a/drivers/gpu/drm/v3d/v3d_gemfs.c b/drivers/gpu/drm/v3d/v3d_gemfs.c
index 4c5e18590a5c..8ec6ed82b3d9 100644
--- a/drivers/gpu/drm/v3d/v3d_gemfs.c
+++ b/drivers/gpu/drm/v3d/v3d_gemfs.c
@@ -3,14 +3,21 @@
 
 #include <linux/fs.h>
 #include <linux/mount.h>
+#include <linux/fs_context.h>
 
 #include "v3d_drv.h"
 
+static int add_param(struct fs_context *fc, const char *key, const char *val)
+{
+	return vfs_parse_fs_string(fc, key, val, strlen(val));
+}
+
 void v3d_gemfs_init(struct v3d_dev *v3d)
 {
-	char huge_opt[] = "huge=within_size";
 	struct file_system_type *type;
+	struct fs_context *fc;
 	struct vfsmount *gemfs;
+	int ret;
 
 	/*
 	 * By creating our own shmemfs mountpoint, we can pass in
@@ -28,8 +35,16 @@ void v3d_gemfs_init(struct v3d_dev *v3d)
 	if (!type)
 		goto err;
 
-	gemfs = vfs_kern_mount(type, SB_KERNMOUNT, type->name, huge_opt);
-	if (IS_ERR(gemfs))
+	fc = fs_context_for_mount(type, SB_KERNMOUNT);
+	if (IS_ERR(fc))
+		goto err;
+	ret = add_param(fc, "source", "tmpfs");
+	if (!ret)
+		ret = add_param(fc, "huge", "within_size");
+	if (!ret)
+		gemfs = fc_mount_longterm(fc);
+	put_fs_context(fc);
+	if (ret)
 		goto err;
 
 	v3d->gemfs = gemfs;
diff --git a/fs/hugetlbfs/inode.c b/fs/hugetlbfs/inode.c
index e4de5425838d..4e0397775167 100644
--- a/fs/hugetlbfs/inode.c
+++ b/fs/hugetlbfs/inode.c
@@ -1587,7 +1587,7 @@ static struct vfsmount *__init mount_one_hugetlbfs(struct hstate *h)
 	} else {
 		struct hugetlbfs_fs_context *ctx = fc->fs_private;
 		ctx->hstate = h;
-		mnt = fc_mount(fc);
+		mnt = fc_mount_longterm(fc);
 		put_fs_context(fc);
 	}
 	if (IS_ERR(mnt))
diff --git a/fs/namespace.c b/fs/namespace.c
index 57b0974a5d1e..6a0697eeda74 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -1260,6 +1260,15 @@ struct vfsmount *fc_mount(struct fs_context *fc)
 }
 EXPORT_SYMBOL(fc_mount);
 
+struct vfsmount *fc_mount_longterm(struct fs_context *fc)
+{
+	struct vfsmount *mnt = fc_mount(fc);
+	if (!IS_ERR(mnt))
+		real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
+	return mnt;
+}
+EXPORT_SYMBOL(fc_mount_longterm);
+
 struct vfsmount *vfs_kern_mount(struct file_system_type *type,
 				int flags, const char *name,
 				void *data)
diff --git a/include/linux/mount.h b/include/linux/mount.h
index 1a508beba446..c145820fcbbf 100644
--- a/include/linux/mount.h
+++ b/include/linux/mount.h
@@ -98,6 +98,7 @@ int mnt_get_write_access(struct vfsmount *mnt);
 void mnt_put_write_access(struct vfsmount *mnt);
 
 extern struct vfsmount *fc_mount(struct fs_context *fc);
+extern struct vfsmount *fc_mount_longterm(struct fs_context *fc);
 extern struct vfsmount *vfs_create_mount(struct fs_context *fc);
 extern struct vfsmount *vfs_kern_mount(struct file_system_type *type,
 				      int flags, const char *name,
diff --git a/ipc/mqueue.c b/ipc/mqueue.c
index 82ed2d3c9846..de7432efbf4a 100644
--- a/ipc/mqueue.c
+++ b/ipc/mqueue.c
@@ -482,7 +482,7 @@ static struct vfsmount *mq_create_mount(struct ipc_namespace *ns)
 	put_user_ns(fc->user_ns);
 	fc->user_ns = get_user_ns(ctx->ipc_ns->user_ns);
 
-	mnt = fc_mount(fc);
+	mnt = fc_mount_longterm(fc);
 	put_fs_context(fc);
 	return mnt;
 }
-- 
2.39.5

[PATCH v3 16/48] Rewrite of propagate_umount()

[Al Viro]

The variant currently in the tree has problems; trying to prove
correctness has caught at least one class of bugs (reparenting
that ends up moving the visible location of reparented mount, due
to not excluding some of the counterparts on propagation that
should've been included).

I tried to prove that it's the only bug there; I'm still not sure
whether it is.  If anyone can reconstruct and write down an analysis
of the mainline implementation, I'll gladly review it; as it is,
I ended up doing a different implementation.  Candidate collection
phase is similar, but trimming the set down until it satisfies the
constraints turned out pretty different.

I hoped to do transformation as a massage series, but that turns out
to be too convoluted.  So it's a single patch replacing propagate_umount()
and friends in one go, with notes and analysis in D/f/propagate_umount.txt
(in addition to inline comments).

As far I can tell, it is provably correct and provably linear by the number
of mounts we need to look at in order to decide what should be unmounted.
It even builds and seems to survive testing...

Another nice thing that fell out of that is that ->mnt_umounting is no longer
needed.

Compared to the first version:
	* explicit MNT_UMOUNT_CANDIDATE flag for is_candidate()
	* trim_ancestors() only clears that flag, leaving the suckers on list
	* trim_one() and handle_locked() take the stuff with flag cleared off
the list.  That allows to iterate with list_for_each_entry_safe() when calling
trim_one() - it removes at most one element from the list now.
	* no globals - I didn't bother with any kind of context, not worth it.

	* Notes updated accordingly; I have not touch the terms yet.

Reviewed-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 .../filesystems/propagate_umount.txt          | 484 ++++++++++++++++++
 fs/mount.h                                    |   1 -
 fs/namespace.c                                |   1 -
 fs/pnode.c                                    | 362 +++++++------
 fs/pnode.h                                    |   2 +-
 include/linux/mount.h                         |   3 +-
 6 files changed, 685 insertions(+), 168 deletions(-)
 create mode 100644 Documentation/filesystems/propagate_umount.txt

diff --git a/Documentation/filesystems/propagate_umount.txt b/Documentation/filesystems/propagate_umount.txt
new file mode 100644
index 000000000000..6906903a8aa2
--- /dev/null
+++ b/Documentation/filesystems/propagate_umount.txt
@@ -0,0 +1,484 @@
+	Notes on propagate_umount()
+
+Umount propagation starts with a set of mounts we are already going to
+take out.  Ideally, we would like to add all downstream cognates to
+that set - anything with the same mountpoint as one of the removed
+mounts and with parent that would receive events from the parent of that
+mount.  However, there are some constraints the resulting set must
+satisfy.
+
+It is convenient to define several properties of sets of mounts:
+
+1) A set S of mounts is non-shifting if for any mount X belonging
+to S all subtrees mounted strictly inside of X (i.e. not overmounting
+the root of X) contain only elements of S.
+
+2) A set S is non-revealing if all locked mounts that belong to S have
+parents that also belong to S.
+
+3) A set S is closed if it contains all children of its elements.
+
+The set of mounts taken out by umount(2) must be non-shifting and
+non-revealing; the first constraint is what allows to reparent
+any remaining mounts and the second is what prevents the exposure
+of any concealed mountpoints.
+
+propagate_umount() takes the original set as an argument and tries to
+extend that set.  The original set is a full subtree and its root is
+unlocked; what matters is that it's closed and non-revealing.
+Resulting set may not be closed; there might still be mounts outside
+of that set, but only on top of stacks of root-overmounting elements
+of set.  They can be reparented to the place where the bottom of
+stack is attached to a mount that will survive.  NOTE: doing that
+will violate a constraint on having no more than one mount with
+the same parent/mountpoint pair; however, the caller (umount_tree())
+will immediately remedy that - it may keep unmounted element attached
+to parent, but only if the parent itself is unmounted.  Since all
+conflicts created by reparenting have common parent *not* in the
+set and one side of the conflict (bottom of the stack of overmounts)
+is in the set, it will be resolved.  However, we rely upon umount_tree()
+doing that pretty much immediately after the call of propagate_umount().
+
+Algorithm is based on two statements:
+	1) for any set S, there is a maximal non-shifting subset of S
+and it can be calculated in O(#S) time.
+	2) for any non-shifting set S, there is a maximal non-revealing
+subset of S.  That subset is also non-shifting and it can be calculated
+in O(#S) time.
+
+		Finding candidates.
+
+We are given a closed set U and we want to find all mounts that have
+the same mountpoint as some mount m in U *and* whose parent receives
+propagation from the parent of the same mount m.  Naive implementation
+would be
+	S = {}
+	for each m in U
+		add m to S
+		p = parent(m)
+		for each q in Propagation(p) - {p}
+			child = look_up(q, mountpoint(m))
+			if child
+				add child to S
+but that can lead to excessive work - there might be propagation among the
+subtrees of U, in which case we'd end up examining the same candidates
+many times.  Since propagation is transitive, the same will happen to
+everything downstream of that candidate and it's not hard to construct
+cases where the approach above leads to the time quadratic by the actual
+number of candidates.
+
+Note that if we run into a candidate we'd already seen, it must've been
+added on an earlier iteration of the outer loop - all additions made
+during one iteration of the outer loop have different parents.  So
+if we find a child already added to the set, we know that everything
+in Propagation(parent(child)) with the same mountpoint has been already
+added.
+	S = {}
+	for each m in U
+		if m in S
+			continue
+		add m to S
+		p = parent(m)
+		q = propagation_next(p, p)
+		while q
+			child = look_up(q, mountpoint(m))
+			if child
+				if child in S
+					q = skip_them(q, p)
+					continue;
+				add child to S
+			q = propagation_next(q, p)
+where
+skip_them(q, p)
+	keep walking Propagation(p) from q until we find something
+	not in Propagation(q)
+
+would get rid of that problem, but we need a sane implementation of
+skip_them().  That's not hard to do - split propagation_next() into
+"down into mnt_slave_list" and "forward-and-up" parts, with the
+skip_them() being "repeat the forward-and-up part until we get NULL
+or something that isn't a peer of the one we are skipping".
+
+Note that there can be no absolute roots among the extra candidates -
+they all come from mount lookups.  Absolute root among the original
+set is _currently_ impossible, but it might be worth protecting
+against.
+
+		Maximal non-shifting subsets.
+
+Let's call a mount m in a set S forbidden in that set if there is a
+subtree mounted strictly inside m and containing mounts that do not
+belong to S.
+
+The set is non-shifting when none of its elements are forbidden in it.
+
+If mount m is forbidden in a set S, it is forbidden in any subset S' it
+belongs to.  In other words, it can't belong to any of the non-shifting
+subsets of S.  If we had a way to find a forbidden mount or show that
+there's none, we could use it to find the maximal non-shifting subset
+simply by finding and removing them until none remain.
+
+Suppose mount m is forbidden in S; then any mounts forbidden in S - {m}
+must have been forbidden in S itself.  Indeed, since m has descendents
+that do not belong to S, any subtree that fits into S will fit into
+S - {m} as well.
+
+So in principle we could go through elements of S, checking if they
+are forbidden in S and removing the ones that are.  Removals will
+not invalidate the checks done for earlier mounts - if they were not
+forbidden at the time we checked, they won't become forbidden later.
+It's too costly to be practical, but there is a similar approach that
+is linear by size of S.
+
+Let's say that mount x in a set S is forbidden by mount y, if
+	* both x and y belong to S.
+	* there is a chain of mounts starting at x and leaving S
+	  immediately after passing through y, with the first
+	  mountpoint strictly inside x.
+Note 1: x may be equal to y - that's the case when something not
+belonging to S is mounted strictly inside x.
+Note 2: if y does not belong to S, it can't forbid anything in S.
+Note 3: if y has no children outside of S, it can't forbid anything in S.
+
+It's easy to show that mount x is forbidden in S if and only if x is
+forbidden in S by some mount y.  And it's easy to find all mounts in S
+forbidden by a given mount.
+
+Consider the following operation:
+	Trim(S, m) = S - {x : x is forbidden by m in S}
+
+Note that if m does not belong to S or has no children outside of S we
+are guaranteed that Trim(S, m) is equal to S.
+
+The following is true: if x is forbidden by y in Trim(S, m), it was
+already forbidden by y in S.
+
+Proof: Suppose x is forbidden by y in Trim(S, m).  Then there is a
+chain of mounts (x_0 = x, ..., x_k = y, x_{k+1} = r), such that x_{k+1}
+is the first element that doesn't belong to Trim(S, m) and the
+mountpoint of x_1 is strictly inside x.  If mount r belongs to S, it must
+have been removed by Trim(S, m), i.e. it was forbidden in S by m.
+Then there was a mount chain from r to some child of m that stayed in
+S all the way until m, but that's impossible since x belongs to Trim(S, m)
+and prepending (x_0, ..., x_k) to that chain demonstrates that x is also
+forbidden in S by m, and thus can't belong to Trim(S, m).
+Therefore r can not belong to S and our chain demonstrates that
+x is forbidden by y in S.  QED.
+
+Corollary: no mount is forbidden by m in Trim(S, m).  Indeed, any
+such mount would have been forbidden by m in S and thus would have been
+in the part of S removed in Trim(S, m).
+
+Corollary: no mount is forbidden by m in Trim(Trim(S, m), n).  Indeed,
+any such would have to have been forbidden by m in Trim(S, m), which
+is impossible.
+
+Corollary: after
+	S = Trim(S, x_1)
+	S = Trim(S, x_2)
+	...
+	S = Trim(S, x_k)
+no mount remaining in S will be forbidden by either of x_1,...,x_k.
+
+The following will reduce S to its maximal non-shifting subset:
+	visited = {}
+	while S contains elements not belonging to visited
+		let m be an arbitrary such element of S
+		S = Trim(S, m)
+		add m to visited
+
+S never grows, so the number of elements of S not belonging to visited
+decreases at least by one on each iteration.  When the loop terminates,
+all mounts remaining in S belong to visited.  It's easy to see that at
+the beginning of each iteration no mount remaining in S will be forbidden
+by any element of visited.  In other words, no mount remaining in S will
+be forbidden, i.e. final value of S will be non-shifting.  It will be
+the maximal non-shifting subset, since we were removing only forbidden
+elements.
+
+	There are two difficulties in implementing the above in linear
+time, both due to the fact that Trim() might need to remove more than one
+element.  Naive implementation of Trim() is vulnerable to running into a
+long chain of mounts, each mounted on top of parent's root.  Nothing in
+that chain is forbidden, so nothing gets removed from it.  We need to
+recognize such chains and avoid walking them again on subsequent calls of
+Trim(), otherwise we will end up with worst-case time being quadratic by
+the number of elements in S.  Another difficulty is in implementing the
+outer loop - we need to iterate through all elements of a shrinking set.
+That would be trivial if we never removed more than one element at a time
+(linked list, with list_for_each_entry_safe for iterator), but we may
+need to remove more than one entry, possibly including the ones we have
+already visited.
+
+	Let's start with naive algorithm for Trim():
+
+Trim_one(m)
+	found = false
+	for each n in children(m)
+		if n not in S
+			found = true
+			if (mountpoint(n) != root(m))
+				remove m from S
+				break
+	if found
+		Trim_ancestors(m)
+
+Trim_ancestors(m)
+	for (; parent(m) in S; m = parent(m)) {
+		if (mountpoint(m) != root(parent(m)))
+			remove parent(m) from S
+	}
+
+If m belongs to S, Trim_one(m) will replace S with Trim(S, m).
+Proof:
+	Consider the chains excluding elements from Trim(S, m).  The last
+two elements in such chain are m and some child of m that does not belong
+to S.  If m has no such children, Trim(S, m) is equal to S.
+	m itself is removed if and only if the chain has exactly two
+elements, i.e. when the last element does not overmount the root of m.
+In other words, that happens when m has a child not in S that does not
+overmount the root of m.
+	All other elements to remove will be ancestors of m, such that
+the entire descent chain from them to m is contained in S.  Let
+(x_0, x_1, ..., x_k = m) be the longest such chain.  x_i needs to be
+removed if and only if x_{i+1} does not overmount its root.  It's easy
+to see that Trim_ancestors(m) will iterate through that chain from
+x_k to x_1 and that it will remove exactly the elements that need to be
+removed.
+
+	Note that if the loop in Trim_ancestors() walks into an already
+visited element, we are guaranteed that remaining iterations will see
+only elements that had already been visited and remove none of them.
+That's the weakness that makes it vulnerable to long chains of full
+overmounts.
+
+	It's easy to deal with, if we can afford setting marks on
+elements of S; we would mark all elements already visited by
+Trim_ancestors() and have it bail out as soon as it sees an already
+marked element.
+
+	The problems with iterating through the set can be dealt with in
+several ways, depending upon the representation we choose for our set.
+One useful observation is that we are given a closed subset in S - the
+original set passed to propagate_umount().  Its elements can neither
+forbid anything nor be forbidden by anything - all their descendents
+belong to S, so they can not occur anywhere in any excluding chain.
+In other words, the elements of that subset will remain in S until
+the end and Trim_one(S, m) is a no-op for all m from that subset.
+
+	That suggests keeping S as a disjoint union of a closed set U
+('will be unmounted, no matter what') and the set of all elements of
+S that do not belong to U.  That set ('candidates') is all we need
+to iterate through.  Let's represent it as a subset in a cyclic list,
+consisting of all list elements that are marked as candidates (initially -
+all of them).  Then we could have Trim_ancestors() only remove the mark,
+leaving the elements on the list.  Then Trim_one() would never remove
+anything other than its argument from the containing list, allowing to
+use list_for_each_entry_safe() as iterator.
+
+	Assuming that representation we get the following:
+
+	list_for_each_entry_safe(m, ..., Candidates, ...)
+		Trim_one(m)
+where
+Trim_one(m)
+	if (m is not marked as a candidate)
+		strip the "seen by Trim_ancestors" mark from m
+		remove m from the Candidates list
+		return
+		
+	remove_this = false
+	found = false
+	for each n in children(m)
+		if n not in S
+			found = true
+			if (mountpoint(n) != root(m))
+				remove_this = true
+				break
+	if found
+		Trim_ancestors(m)
+	if remove_this
+		strip the "seen by Trim_ancestors" mark from m
+		strip the "candidate" mark from m
+		remove m from the Candidate list
+
+Trim_ancestors(m)
+	for (p = parent(m); p is marked as candidate ; m = p, p = parent(p)) {
+		if m is marked as seen by Trim_ancestors
+			return
+		mark m as seen by Trim_ancestors
+		if (mountpoint(m) != root(p))
+			strip the "candidate" mark from p
+	}
+
+	Terminating condition in the loop in Trim_ancestors() is correct,
+since that that loop will never run into p belonging to U - p is always
+an ancestor of argument of Trim_one() and since U is closed, the argument
+of Trim_one() would also have to belong to U.  But Trim_one() is never
+called for elements of U.  In other words, p belongs to S if and only
+if it belongs to candidates.
+
+	Time complexity:
+* we get no more than O(#S) calls of Trim_one()
+* the loop over children in Trim_one() never looks at the same child
+twice through all the calls.
+* iterations of that loop for children in S are no more than O(#S)
+in the worst case
+* at most two children that are not elements of S are considered per
+call of Trim_one().
+* the loop in Trim_ancestors() sets its mark once per iteration and
+no element of S has is set more than once.
+
+	In the end we may have some elements excluded from S by
+Trim_ancestors() still stuck on the list.  We could do a separate
+loop removing them from the list (also no worse than O(#S) time),
+but it's easier to leave that until the next phase - there we will
+iterate through the candidates anyway.
+
+	The caller has already removed all elements of U from their parents'
+lists of children, which means that checking if child belongs to S is
+equivalent to checking if it's marked as a candidate; we'll never see
+the elements of U in the loop over children in Trim_one().
+
+	What's more, if we see that children(m) is empty and m is not
+locked, we can immediately move m into the committed subset (remove
+from the parent's list of children, etc.).  That's one fewer mount we'll
+have to look into when we check the list of children of its parent *and*
+when we get to building the non-revealing subset.
+
+		Maximal non-revealing subsets
+
+If S is not a non-revealing subset, there is a locked element x in S
+such that parent of x is not in S.
+
+Obviously, no non-revealing subset of S may contain x.  Removing such
+elements one by one will obviously end with the maximal non-revealing
+subset (possibly empty one).  Note that removal of an element will
+require removal of all its locked children, etc.
+
+If the set had been non-shifting, it will remain non-shifting after
+such removals.
+Proof: suppose S was non-shifting, x is a locked element of S, parent of x
+is not in S and S - {x} is not non-shifting.  Then there is an element m
+in S - {x} and a subtree mounted strictly inside m, such that m contains
+an element not in in S - {x}.  Since S is non-shifting, everything in
+that subtree must belong to S.  But that means that this subtree must
+contain x somewhere *and* that parent of x either belongs that subtree
+or is equal to m.  Either way it must belong to S.  Contradiction.
+
+// same representation as for finding maximal non-shifting subsets:
+// S is a disjoint union of a non-revealing set U (the ones we are committed
+// to unmount) and a set of candidates, represented as a subset of list
+// elements that have "is a candidate" mark on them.
+// Elements of U are removed from their parents' lists of children.
+// In the end candidates becomes empty and maximal non-revealing non-shifting
+// subset of S is now in U
+	while (Candidates list is non-empty)
+		handle_locked(first(Candidates))
+
+handle_locked(m)
+	if m is not marked as a candidate
+		strip the "seen by Trim_ancestors" mark from m
+		remove m from the list
+		return
+	cutoff = m
+	for (p = m; p in candidates; p = parent(p)) {
+		strip the "seen by Trim_ancestors" mark from p
+		strip the "candidate" mark from p
+		remove p from the Candidates list
+		if (!locked(p))
+			cutoff = parent(p)
+	}
+	if p in U
+		cutoff = p
+	while m != cutoff
+		remove m from children(parent(m))
+		add m to U
+		m = parent(m)
+
+Let (x_0, ..., x_n = m) be the maximal chain of descent of m within S.
+* If it contains some elements of U, let x_k be the last one of those.
+Then union of U with {x_{k+1}, ..., x_n} is obviously non-revealing.
+* otherwise if all its elements are locked, then none of {x_0, ..., x_n}
+may be elements of a non-revealing subset of S.
+* otherwise let x_k be the first unlocked element of the chain.  Then none
+of {x_0, ..., x_{k-1}} may be an element of a non-revealing subset of
+S and union of U and {x_k, ..., x_n} is non-revealing.
+
+handle_locked(m) finds which of these cases applies and adjusts Candidates
+and U accordingly.  U remains non-revealing, union of Candidates and
+U still contains any non-revealing subset of S and after the call of
+handle_locked(m) m is guaranteed to be not in Candidates list.  So having
+it called for each element of S would suffice to empty Candidates,
+leaving U the maximal non-revealing subset of S.
+
+However, handle_locked(m) is a no-op when m belongs to U, so it's enough
+to have it called for elements of Candidates list until none remain.
+
+Time complexity: number of calls of handle_locked() is limited by
+#Candidates, each iteration of the first loop in handle_locked() removes
+an element from the list, so their total number of executions is also
+limited by #Candidates; number of iterations in the second loop is no
+greater than the number of iterations of the first loop.
+
+
+		Reparenting
+
+After we'd calculated the final set, we still need to deal with
+reparenting - if an element of the final set has a child not in it,
+we need to reparent such child.
+
+Such children can only be root-overmounting (otherwise the set wouldn't
+be non-shifting) and their parents can not belong to the original set,
+since the original is guaranteed to be closed.
+
+
+		Putting all of that together
+
+The plan is to
+	* find all candidates
+	* trim down to maximal non-shifting subset
+	* trim down to maximal non-revealing subset
+	* reparent anything that needs to be reparented
+	* return the resulting set to the caller
+
+For the 2nd and 3rd steps we want to separate the set into growing
+non-revealing subset, initially containing the original set ("U" in
+terms of the pseudocode above) and everything we are still not sure about
+("candidates").  It means that for the output of the 1st step we'd like
+the extra candidates separated from the stuff already in the original set.
+For the 4th step we would like the additions to U separate from the
+original set.
+
+So let's go for
+	* original set ("set").  Linkage via mnt_list
+	* undecided candidates ("candidates").  Subset of a list,
+consisting of all its elements marked with a new flag (MNT_UMOUNT_CANDIDATE).
+Initially all elements of the list will be marked that way; in the
+end the list will become empty and no mounts will remain marked with
+that flag.
+	* Reuse MNT_MARKED for "has been already seen by trim_ancestors()".
+	* anything in U that hadn't been in the original set - elements of
+candidates will gradually be either discarded or moved there.  In other
+words, it's the candidates we have already decided to unmount.	Its role
+is reasonably close to the old "to_umount", so let's use that name.
+Linkage via mnt_list.
+
+For gather_candidates() we'll need to maintain both candidates (S -
+set) and intersection of S with set.  Use MNT_UMOUNT_CANDIDATE for
+all elements we encounter, putting the ones not already in the original
+set into the list of candidates.  When we are done, strip that flag from
+all elements of the original set.  That gives a cheap way to check
+if element belongs to S (in gather_candidates) and to candidates
+itself (at later stages).  Call that predicate is_candidate(); it would
+be m->mnt_flags & MNT_UMOUNT_CANDIDATE.
+
+All elements of the original set are marked with MNT_UMOUNT and we'll
+need the same for elements added when joining the contents of to_umount
+to set in the end.  Let's set MNT_UMOUNT at the time we add an element
+to to_umount; that's close to what the old 'umount_one' is doing, so
+let's keep that name.  It also gives us another predicate we need -
+"belongs to union of set and to_umount"; will_be_unmounted() for now.
+
+Removals from the candidates list should strip both MNT_MARKED and
+MNT_UMOUNT_CANDIDATE; call it remove_from_candidates_list().
diff --git a/fs/mount.h b/fs/mount.h
index f20e6ed845fe..fb93d3e16724 100644
--- a/fs/mount.h
+++ b/fs/mount.h
@@ -79,7 +79,6 @@ struct mount {
 		struct hlist_node mnt_mp_list;	/* list mounts with the same mountpoint */
 		struct hlist_node mnt_umount;
 	};
-	struct list_head mnt_umounting; /* list entry for umount propagation */
 #ifdef CONFIG_FSNOTIFY
 	struct fsnotify_mark_connector __rcu *mnt_fsnotify_marks;
 	__u32 mnt_fsnotify_mask;
diff --git a/fs/namespace.c b/fs/namespace.c
index 6a0697eeda74..f64895d47d70 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -383,7 +383,6 @@ static struct mount *alloc_vfsmnt(const char *name)
 		INIT_LIST_HEAD(&mnt->mnt_slave_list);
 		INIT_LIST_HEAD(&mnt->mnt_slave);
 		INIT_HLIST_NODE(&mnt->mnt_mp_list);
-		INIT_LIST_HEAD(&mnt->mnt_umounting);
 		INIT_HLIST_HEAD(&mnt->mnt_stuck_children);
 		RB_CLEAR_NODE(&mnt->mnt_node);
 		mnt->mnt.mnt_idmap = &nop_mnt_idmap;
diff --git a/fs/pnode.c b/fs/pnode.c
index aa187144e389..901d40946d34 100644
--- a/fs/pnode.c
+++ b/fs/pnode.c
@@ -24,11 +24,6 @@ static inline struct mount *first_slave(struct mount *p)
 	return list_entry(p->mnt_slave_list.next, struct mount, mnt_slave);
 }
 
-static inline struct mount *last_slave(struct mount *p)
-{
-	return list_entry(p->mnt_slave_list.prev, struct mount, mnt_slave);
-}
-
 static inline struct mount *next_slave(struct mount *p)
 {
 	return list_entry(p->mnt_slave.next, struct mount, mnt_slave);
@@ -136,6 +131,23 @@ void change_mnt_propagation(struct mount *mnt, int type)
 	}
 }
 
+static struct mount *__propagation_next(struct mount *m,
+					 struct mount *origin)
+{
+	while (1) {
+		struct mount *master = m->mnt_master;
+
+		if (master == origin->mnt_master) {
+			struct mount *next = next_peer(m);
+			return (next == origin) ? NULL : next;
+		} else if (m->mnt_slave.next != &master->mnt_slave_list)
+			return next_slave(m);
+
+		/* back at master */
+		m = master;
+	}
+}
+
 /*
  * get the next mount in the propagation tree.
  * @m: the mount seen last
@@ -153,31 +165,21 @@ static struct mount *propagation_next(struct mount *m,
 	if (!IS_MNT_NEW(m) && !list_empty(&m->mnt_slave_list))
 		return first_slave(m);
 
-	while (1) {
-		struct mount *master = m->mnt_master;
-
-		if (master == origin->mnt_master) {
-			struct mount *next = next_peer(m);
-			return (next == origin) ? NULL : next;
-		} else if (m->mnt_slave.next != &master->mnt_slave_list)
-			return next_slave(m);
-
-		/* back at master */
-		m = master;
-	}
+	return __propagation_next(m, origin);
 }
 
 static struct mount *skip_propagation_subtree(struct mount *m,
 						struct mount *origin)
 {
 	/*
-	 * Advance m such that propagation_next will not return
-	 * the slaves of m.
+	 * Advance m past everything that gets propagation from it.
 	 */
-	if (!IS_MNT_NEW(m) && !list_empty(&m->mnt_slave_list))
-		m = last_slave(m);
+	struct mount *p = __propagation_next(m, origin);
+
+	while (p && peers(m, p))
+		p = __propagation_next(p, origin);
 
-	return m;
+	return p;
 }
 
 static struct mount *next_group(struct mount *m, struct mount *origin)
@@ -458,181 +460,213 @@ void propagate_mount_unlock(struct mount *mnt)
 	}
 }
 
-static void umount_one(struct mount *mnt, struct list_head *to_umount)
+static inline bool is_candidate(struct mount *m)
 {
-	CLEAR_MNT_MARK(mnt);
-	mnt->mnt.mnt_flags |= MNT_UMOUNT;
-	list_del_init(&mnt->mnt_child);
-	list_del_init(&mnt->mnt_umounting);
-	move_from_ns(mnt, to_umount);
+	return m->mnt.mnt_flags & MNT_UMOUNT_CANDIDATE;
 }
 
-/*
- * NOTE: unmounting 'mnt' naturally propagates to all other mounts its
- * parent propagates to.
- */
-static bool __propagate_umount(struct mount *mnt,
-			       struct list_head *to_umount,
-			       struct list_head *to_restore)
+static inline bool will_be_unmounted(struct mount *m)
 {
-	bool progress = false;
-	struct mount *child;
+	return m->mnt.mnt_flags & MNT_UMOUNT;
+}
 
-	/*
-	 * The state of the parent won't change if this mount is
-	 * already unmounted or marked as without children.
-	 */
-	if (mnt->mnt.mnt_flags & (MNT_UMOUNT | MNT_MARKED))
-		goto out;
+static void umount_one(struct mount *m, struct list_head *to_umount)
+{
+	m->mnt.mnt_flags |= MNT_UMOUNT;
+	list_del_init(&m->mnt_child);
+	move_from_ns(m, to_umount);
+}
 
-	/* Verify topper is the only grandchild that has not been
-	 * speculatively unmounted.
-	 */
-	list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
-		if (child->mnt_mountpoint == mnt->mnt.mnt_root)
-			continue;
-		if (!list_empty(&child->mnt_umounting) && IS_MNT_MARKED(child))
-			continue;
-		/* Found a mounted child */
-		goto children;
-	}
+static void remove_from_candidate_list(struct mount *m)
+{
+	m->mnt.mnt_flags &= ~(MNT_MARKED | MNT_UMOUNT_CANDIDATE);
+	list_del_init(&m->mnt_list);
+}
 
-	/* Mark mounts that can be unmounted if not locked */
-	SET_MNT_MARK(mnt);
-	progress = true;
+static void gather_candidates(struct list_head *set,
+			      struct list_head *candidates)
+{
+	struct mount *m, *p, *q;
 
-	/* If a mount is without children and not locked umount it. */
-	if (!IS_MNT_LOCKED(mnt)) {
-		umount_one(mnt, to_umount);
-	} else {
-children:
-		list_move_tail(&mnt->mnt_umounting, to_restore);
+	list_for_each_entry(m, set, mnt_list) {
+		if (is_candidate(m))
+			continue;
+		m->mnt.mnt_flags |= MNT_UMOUNT_CANDIDATE;
+		p = m->mnt_parent;
+		q = propagation_next(p, p);
+		while (q) {
+			struct mount *child = __lookup_mnt(&q->mnt,
+							   m->mnt_mountpoint);
+			if (child) {
+				/*
+				 * We might've already run into this one.  That
+				 * must've happened on earlier iteration of the
+				 * outer loop; in that case we can skip those
+				 * parents that get propagation from q - there
+				 * will be nothing new on those as well.
+				 */
+				if (is_candidate(child)) {
+					q = skip_propagation_subtree(q, p);
+					continue;
+				}
+				child->mnt.mnt_flags |= MNT_UMOUNT_CANDIDATE;
+				if (!will_be_unmounted(child))
+					list_add(&child->mnt_list, candidates);
+			}
+			q = propagation_next(q, p);
+		}
 	}
-out:
-	return progress;
+	list_for_each_entry(m, set, mnt_list)
+		m->mnt.mnt_flags &= ~MNT_UMOUNT_CANDIDATE;
 }
 
-static void umount_list(struct list_head *to_umount,
-			struct list_head *to_restore)
+/*
+ * We know that some child of @m can't be unmounted.  In all places where the
+ * chain of descent of @m has child not overmounting the root of parent,
+ * the parent can't be unmounted either.
+ */
+static void trim_ancestors(struct mount *m)
 {
-	struct mount *mnt, *child, *tmp;
-	list_for_each_entry(mnt, to_umount, mnt_list) {
-		list_for_each_entry_safe(child, tmp, &mnt->mnt_mounts, mnt_child) {
-			/* topper? */
-			if (child->mnt_mountpoint == mnt->mnt.mnt_root)
-				list_move_tail(&child->mnt_umounting, to_restore);
-			else
-				umount_one(child, to_umount);
-		}
+	struct mount *p;
+
+	for (p = m->mnt_parent; is_candidate(p); m = p, p = p->mnt_parent) {
+		if (IS_MNT_MARKED(m))	// all candidates beneath are overmounts
+			return;
+		SET_MNT_MARK(m);
+		if (m != p->overmount)
+			p->mnt.mnt_flags &= ~MNT_UMOUNT_CANDIDATE;
 	}
 }
 
-static void restore_mounts(struct list_head *to_restore)
+/*
+ * Find and exclude all umount candidates forbidden by @m
+ * (see Documentation/filesystems/propagate_umount.txt)
+ * If we can immediately tell that @m is OK to unmount (unlocked
+ * and all children are already committed to unmounting) commit
+ * to unmounting it.
+ * Only @m itself might be taken from the candidates list;
+ * anything found by trim_ancestors() is marked non-candidate
+ * and left on the list.
+ */
+static void trim_one(struct mount *m, struct list_head *to_umount)
 {
-	/* Restore mounts to a clean working state */
-	while (!list_empty(to_restore)) {
-		struct mount *mnt, *parent;
-		struct mountpoint *mp;
-
-		mnt = list_first_entry(to_restore, struct mount, mnt_umounting);
-		CLEAR_MNT_MARK(mnt);
-		list_del_init(&mnt->mnt_umounting);
-
-		/* Should this mount be reparented? */
-		mp = mnt->mnt_mp;
-		parent = mnt->mnt_parent;
-		while (parent->mnt.mnt_flags & MNT_UMOUNT) {
-			mp = parent->mnt_mp;
-			parent = parent->mnt_parent;
-		}
-		if (parent != mnt->mnt_parent) {
-			mnt_change_mountpoint(parent, mp, mnt);
-			mnt_notify_add(mnt);
+	bool remove_this = false, found = false, umount_this = false;
+	struct mount *n;
+
+	if (!is_candidate(m)) { // trim_ancestors() left it on list
+		remove_from_candidate_list(m);
+		return;
+	}
+
+	list_for_each_entry(n, &m->mnt_mounts, mnt_child) {
+		if (!is_candidate(n)) {
+			found = true;
+			if (n != m->overmount) {
+				remove_this = true;
+				break;
+			}
 		}
 	}
+	if (found) {
+		trim_ancestors(m);
+	} else if (!IS_MNT_LOCKED(m) && list_empty(&m->mnt_mounts)) {
+		remove_this = true;
+		umount_this = true;
+	}
+	if (remove_this) {
+		remove_from_candidate_list(m);
+		if (umount_this)
+			umount_one(m, to_umount);
+	}
 }
 
-static void cleanup_umount_visitations(struct list_head *visited)
+static void handle_locked(struct mount *m, struct list_head *to_umount)
 {
-	while (!list_empty(visited)) {
-		struct mount *mnt =
-			list_first_entry(visited, struct mount, mnt_umounting);
-		list_del_init(&mnt->mnt_umounting);
+	struct mount *cutoff = m, *p;
+
+	if (!is_candidate(m)) { // trim_ancestors() left it on list
+		remove_from_candidate_list(m);
+		return;
+	}
+	for (p = m; is_candidate(p); p = p->mnt_parent) {
+		remove_from_candidate_list(p);
+		if (!IS_MNT_LOCKED(p))
+			cutoff = p->mnt_parent;
+	}
+	if (will_be_unmounted(p))
+		cutoff = p;
+	while (m != cutoff) {
+		umount_one(m, to_umount);
+		m = m->mnt_parent;
 	}
 }
 
 /*
- * collect all mounts that receive propagation from the mount in @list,
- * and return these additional mounts in the same list.
- * @list: the list of mounts to be unmounted.
+ * @m is not to going away, and it overmounts the top of a stack of mounts
+ * that are going away.  We know that all of those are fully overmounted
+ * by the one above (@m being the topmost of the chain), so @m can be slid
+ * in place where the bottom of the stack is attached.
  *
- * vfsmount lock must be held for write
+ * NOTE: here we temporarily violate a constraint - two mounts end up with
+ * the same parent and mountpoint; that will be remedied as soon as we
+ * return from propagate_umount() - its caller (umount_tree()) will detach
+ * the stack from the parent it (and now @m) is attached to.  umount_tree()
+ * might choose to keep unmounted pieces stuck to each other, but it always
+ * detaches them from the mounts that remain in the tree.
  */
-int propagate_umount(struct list_head *list)
+static void reparent(struct mount *m)
 {
-	struct mount *mnt;
-	LIST_HEAD(to_restore);
-	LIST_HEAD(to_umount);
-	LIST_HEAD(visited);
-
-	/* Find candidates for unmounting */
-	list_for_each_entry_reverse(mnt, list, mnt_list) {
-		struct mount *parent = mnt->mnt_parent;
-		struct mount *m;
+	struct mount *p = m;
+	struct mountpoint *mp;
 
-		/*
-		 * If this mount has already been visited it is known that it's
-		 * entire peer group and all of their slaves in the propagation
-		 * tree for the mountpoint has already been visited and there is
-		 * no need to visit them again.
-		 */
-		if (!list_empty(&mnt->mnt_umounting))
-			continue;
+	do {
+		mp = p->mnt_mp;
+		p = p->mnt_parent;
+	} while (will_be_unmounted(p));
 
-		list_add_tail(&mnt->mnt_umounting, &visited);
-		for (m = propagation_next(parent, parent); m;
-		     m = propagation_next(m, parent)) {
-			struct mount *child = __lookup_mnt(&m->mnt,
-							   mnt->mnt_mountpoint);
-			if (!child)
-				continue;
+	mnt_change_mountpoint(p, mp, m);
+	mnt_notify_add(m);
+}
 
-			if (!list_empty(&child->mnt_umounting)) {
-				/*
-				 * If the child has already been visited it is
-				 * know that it's entire peer group and all of
-				 * their slaves in the propgation tree for the
-				 * mountpoint has already been visited and there
-				 * is no need to visit this subtree again.
-				 */
-				m = skip_propagation_subtree(m, parent);
-				continue;
-			} else if (child->mnt.mnt_flags & MNT_UMOUNT) {
-				/*
-				 * We have come across a partially unmounted
-				 * mount in a list that has not been visited
-				 * yet. Remember it has been visited and
-				 * continue about our merry way.
-				 */
-				list_add_tail(&child->mnt_umounting, &visited);
-				continue;
-			}
+/**
+ * propagate_umount - apply propagation rules to the set of mounts for umount()
+ * @set: the list of mounts to be unmounted.
+ *
+ * Collect all mounts that receive propagation from the mount in @set and have
+ * no obstacles to being unmounted.  Add these additional mounts to the set.
+ *
+ * See Documentation/filesystems/propagate_umount.txt if you do anything in
+ * this area.
+ *
+ * Locks held:
+ * mount_lock (write_seqlock), namespace_sem (exclusive).
+ */
+void propagate_umount(struct list_head *set)
+{
+	struct mount *m, *p;
+	LIST_HEAD(to_umount);	// committed to unmounting
+	LIST_HEAD(candidates);	// undecided umount candidates
 
-			/* Check the child and parents while progress is made */
-			while (__propagate_umount(child,
-						  &to_umount, &to_restore)) {
-				/* Is the parent a umount candidate? */
-				child = child->mnt_parent;
-				if (list_empty(&child->mnt_umounting))
-					break;
-			}
-		}
+	// collect all candidates
+	gather_candidates(set, &candidates);
+
+	// reduce the set until it's non-shifting
+	list_for_each_entry_safe(m, p, &candidates, mnt_list)
+		trim_one(m, &to_umount);
+
+	// ... and non-revealing
+	while (!list_empty(&candidates)) {
+		m = list_first_entry(&candidates,struct mount, mnt_list);
+		handle_locked(m, &to_umount);
 	}
 
-	umount_list(&to_umount, &to_restore);
-	restore_mounts(&to_restore);
-	cleanup_umount_visitations(&visited);
-	list_splice_tail(&to_umount, list);
+	// now to_umount consists of all acceptable candidates
+	// deal with reparenting of remaining overmounts on those
+	list_for_each_entry(m, &to_umount, mnt_list) {
+		if (m->overmount)
+			reparent(m->overmount);
+	}
 
-	return 0;
+	// and fold them into the set
+	list_splice_tail_init(&to_umount, set);
 }
diff --git a/fs/pnode.h b/fs/pnode.h
index 93fa9311bd07..04f1ac53aa49 100644
--- a/fs/pnode.h
+++ b/fs/pnode.h
@@ -42,7 +42,7 @@ static inline bool peers(const struct mount *m1, const struct mount *m2)
 void change_mnt_propagation(struct mount *, int);
 int propagate_mnt(struct mount *, struct mountpoint *, struct mount *,
 		struct hlist_head *);
-int propagate_umount(struct list_head *);
+void propagate_umount(struct list_head *);
 int propagate_mount_busy(struct mount *, int);
 void propagate_mount_unlock(struct mount *);
 void mnt_release_group_id(struct mount *);
diff --git a/include/linux/mount.h b/include/linux/mount.h
index c145820fcbbf..65fa8442c00a 100644
--- a/include/linux/mount.h
+++ b/include/linux/mount.h
@@ -40,6 +40,7 @@ enum mount_flags {
 
 	MNT_INTERNAL	= 0x4000,
 
+	MNT_UMOUNT_CANDIDATE	= 0x020000,
 	MNT_LOCK_ATIME		= 0x040000,
 	MNT_LOCK_NOEXEC		= 0x080000,
 	MNT_LOCK_NOSUID		= 0x100000,
@@ -66,7 +67,7 @@ enum mount_flags {
 
 	MNT_INTERNAL_FLAGS = MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL |
 			     MNT_DOOMED | MNT_SYNC_UMOUNT | MNT_MARKED |
-			     MNT_LOCKED,
+			     MNT_LOCKED | MNT_UMOUNT_CANDIDATE,
 };
 
 struct vfsmount {
-- 
2.39.5

[PATCH v3 17/48] make commit_tree() usable in same-namespace move case

[Al Viro]

Once attach_recursive_mnt() has created all copies of original subtree,
it needs to put them in place(s).

Steps needed for those are slightly different:
	1) in 'move' case, original copy doesn't need any rbtree
manipulations (everything's already in the same namespace where it will
be), but it needs to be detached from the current location
	2) in 'attach' case, original may be in anon namespace; if it is,
all those mounts need to removed from their current namespace before
insertion into the target one
	3) additional copies have a couple of extra twists - in case
of cross-userns propagation we need to lock everything other the root of
subtree and in case when we end up inserting under an existing mount,
that mount needs to be found (for original copy we have it explicitly
passed by the caller).

Quite a bit of that can be unified; as the first step, make commit_tree()
helper (inserting mounts into namespace, hashing the root of subtree
and marking the namespace as updated) usable in all cases; (2) and (3)
are already using it and for (1) we only need to make the insertion of
mounts into namespace conditional.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 30 ++++++++++++++----------------
 1 file changed, 14 insertions(+), 16 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index f64895d47d70..937c2a1825f2 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -1172,15 +1172,17 @@ static void commit_tree(struct mount *mnt)
 
 	BUG_ON(parent == mnt);
 
-	list_add_tail(&head, &mnt->mnt_list);
-	while (!list_empty(&head)) {
-		m = list_first_entry(&head, typeof(*m), mnt_list);
-		list_del(&m->mnt_list);
+	if (!mnt_ns_attached(mnt)) {
+		list_add_tail(&head, &mnt->mnt_list);
+		while (!list_empty(&head)) {
+			m = list_first_entry(&head, typeof(*m), mnt_list);
+			list_del(&m->mnt_list);
 
-		mnt_add_to_ns(n, m);
+			mnt_add_to_ns(n, m);
+		}
+		n->nr_mounts += n->pending_mounts;
+		n->pending_mounts = 0;
 	}
-	n->nr_mounts += n->pending_mounts;
-	n->pending_mounts = 0;
 
 	make_visible(mnt);
 	touch_mnt_namespace(n);
@@ -2691,12 +2693,7 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 
 	if (moving) {
 		unhash_mnt(source_mnt);
-		mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
-		if (beneath)
-			mnt_change_mountpoint(top, smp, top_mnt);
-		make_visible(source_mnt);
 		mnt_notify_add(source_mnt);
-		touch_mnt_namespace(source_mnt->mnt_ns);
 	} else {
 		if (source_mnt->mnt_ns) {
 			LIST_HEAD(head);
@@ -2706,12 +2703,13 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 				move_from_ns(p, &head);
 			list_del_init(&head);
 		}
-		mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
-		if (beneath)
-			mnt_change_mountpoint(top, smp, top_mnt);
-		commit_tree(source_mnt);
 	}
 
+	mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
+	if (beneath)
+		mnt_change_mountpoint(top, smp, top_mnt);
+	commit_tree(source_mnt);
+
 	hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
 		struct mount *q;
 		hlist_del_init(&child->mnt_hash);
-- 
2.39.5

[PATCH v3 18/48] attach_recursive_mnt(): unify the mnt_change_mountpoint() logics

[Al Viro]

The logics used for tucking under existing mount differs for original
and copies; copies do a mount hash lookup to see if mountpoint to be is
already overmounted, while the original is told explicitly.

But the same logics that is used for copies works for the original,
at which point the only place where we get very close to eliminating
the need of passing 'beneath' flag to attach_recursive_mnt().

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 24 +++++++++++++++---------
 1 file changed, 15 insertions(+), 9 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index 937c2a1825f2..9b8d07df4aa5 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -2643,7 +2643,7 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 	HLIST_HEAD(tree_list);
 	struct mnt_namespace *ns = top_mnt->mnt_ns;
 	struct mountpoint *smp;
-	struct mountpoint *secondary = NULL;
+	struct mountpoint *shorter = NULL;
 	struct mount *child, *dest_mnt, *p;
 	struct mount *top;
 	struct hlist_node *n;
@@ -2655,14 +2655,12 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 	 * mounted beneath mounts on the same mountpoint.
 	 */
 	for (top = source_mnt; unlikely(top->overmount); top = top->overmount) {
-		if (!secondary && is_mnt_ns_file(top->mnt.mnt_root))
-			secondary = top->mnt_mp;
+		if (!shorter && is_mnt_ns_file(top->mnt.mnt_root))
+			shorter = top->mnt_mp;
 	}
 	smp = get_mountpoint(top->mnt.mnt_root);
 	if (IS_ERR(smp))
 		return PTR_ERR(smp);
-	if (!secondary)
-		secondary = smp;
 
 	/* Is there space to add these mounts to the mount namespace? */
 	if (!moving) {
@@ -2706,9 +2704,14 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 	}
 
 	mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
-	if (beneath)
-		mnt_change_mountpoint(top, smp, top_mnt);
-	commit_tree(source_mnt);
+	/*
+	 * Now the original copy is in the same state as the secondaries -
+	 * its root attached to mountpoint, but not hashed and all mounts
+	 * in it are either in our namespace or in no namespace at all.
+	 * Add the original to the list of copies and deal with the
+	 * rest of work for all of them uniformly.
+	 */
+	hlist_add_head(&source_mnt->mnt_hash, &tree_list);
 
 	hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
 		struct mount *q;
@@ -2719,10 +2722,13 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 		q = __lookup_mnt(&child->mnt_parent->mnt,
 				 child->mnt_mountpoint);
 		if (q) {
+			struct mountpoint *mp = smp;
 			struct mount *r = child;
 			while (unlikely(r->overmount))
 				r = r->overmount;
-			mnt_change_mountpoint(r, secondary, q);
+			if (unlikely(shorter) && child != source_mnt)
+				mp = shorter;
+			mnt_change_mountpoint(r, mp, q);
 		}
 		commit_tree(child);
 	}
-- 
2.39.5

[PATCH v3 19/48] attach_recursive_mnt(): pass destination mount in all cases

[Al Viro]

... and 'beneath' is no longer used there

Reviewed-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 17 ++++++-----------
 1 file changed, 6 insertions(+), 11 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index 9b8d07df4aa5..449e66436b4f 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -2570,7 +2570,7 @@ enum mnt_tree_flags_t {
 /**
  * attach_recursive_mnt - attach a source mount tree
  * @source_mnt: mount tree to be attached
- * @top_mnt:    mount that @source_mnt will be mounted on or mounted beneath
+ * @dest_mnt:   mount that @source_mnt will be mounted on
  * @dest_mp:    the mountpoint @source_mnt will be mounted at
  * @flags:      modify how @source_mnt is supposed to be attached
  *
@@ -2635,20 +2635,20 @@ enum mnt_tree_flags_t {
  *         Otherwise a negative error code is returned.
  */
 static int attach_recursive_mnt(struct mount *source_mnt,
-				struct mount *top_mnt,
+				struct mount *dest_mnt,
 				struct mountpoint *dest_mp,
 				enum mnt_tree_flags_t flags)
 {
 	struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
 	HLIST_HEAD(tree_list);
-	struct mnt_namespace *ns = top_mnt->mnt_ns;
+	struct mnt_namespace *ns = dest_mnt->mnt_ns;
 	struct mountpoint *smp;
 	struct mountpoint *shorter = NULL;
-	struct mount *child, *dest_mnt, *p;
+	struct mount *child, *p;
 	struct mount *top;
 	struct hlist_node *n;
 	int err = 0;
-	bool moving = flags & MNT_TREE_MOVE, beneath = flags & MNT_TREE_BENEATH;
+	bool moving = flags & MNT_TREE_MOVE;
 
 	/*
 	 * Preallocate a mountpoint in case the new mounts need to be
@@ -2669,11 +2669,6 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 			goto out;
 	}
 
-	if (beneath)
-		dest_mnt = top_mnt->mnt_parent;
-	else
-		dest_mnt = top_mnt;
-
 	if (IS_MNT_SHARED(dest_mnt)) {
 		err = invent_group_ids(source_mnt, true);
 		if (err)
@@ -3688,7 +3683,7 @@ static int do_move_mount(struct path *old_path,
 	if (mount_is_ancestor(old, p))
 		goto out;
 
-	err = attach_recursive_mnt(old, real_mount(new_path->mnt), mp, flags);
+	err = attach_recursive_mnt(old, p, mp, flags);
 	if (err)
 		goto out;
 
-- 
2.39.5

[PATCH v3 20/48] attach_recursive_mnt(): get rid of flags entirely

[Al Viro]

move vs. attach is trivially detected as mnt_has_parent(source_mnt)...

Reviewed-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 18 ++++++------------
 1 file changed, 6 insertions(+), 12 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index 449e66436b4f..adb37f06ba68 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -2562,9 +2562,8 @@ int count_mounts(struct mnt_namespace *ns, struct mount *mnt)
 }
 
 enum mnt_tree_flags_t {
-	MNT_TREE_MOVE = BIT(0),
-	MNT_TREE_BENEATH = BIT(1),
-	MNT_TREE_PROPAGATION = BIT(2),
+	MNT_TREE_BENEATH = BIT(0),
+	MNT_TREE_PROPAGATION = BIT(1),
 };
 
 /**
@@ -2572,7 +2571,6 @@ enum mnt_tree_flags_t {
  * @source_mnt: mount tree to be attached
  * @dest_mnt:   mount that @source_mnt will be mounted on
  * @dest_mp:    the mountpoint @source_mnt will be mounted at
- * @flags:      modify how @source_mnt is supposed to be attached
  *
  *  NOTE: in the table below explains the semantics when a source mount
  *  of a given type is attached to a destination mount of a given type.
@@ -2636,8 +2634,7 @@ enum mnt_tree_flags_t {
  */
 static int attach_recursive_mnt(struct mount *source_mnt,
 				struct mount *dest_mnt,
-				struct mountpoint *dest_mp,
-				enum mnt_tree_flags_t flags)
+				struct mountpoint *dest_mp)
 {
 	struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
 	HLIST_HEAD(tree_list);
@@ -2648,7 +2645,7 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 	struct mount *top;
 	struct hlist_node *n;
 	int err = 0;
-	bool moving = flags & MNT_TREE_MOVE;
+	bool moving = mnt_has_parent(source_mnt);
 
 	/*
 	 * Preallocate a mountpoint in case the new mounts need to be
@@ -2871,7 +2868,7 @@ static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
 	      d_is_dir(mnt->mnt.mnt_root))
 		return -ENOTDIR;
 
-	return attach_recursive_mnt(mnt, p, mp, 0);
+	return attach_recursive_mnt(mnt, p, mp);
 }
 
 /*
@@ -3613,8 +3610,6 @@ static int do_move_mount(struct path *old_path,
 	p = real_mount(new_path->mnt);
 	parent = old->mnt_parent;
 	attached = mnt_has_parent(old);
-	if (attached)
-		flags |= MNT_TREE_MOVE;
 	old_mp = old->mnt_mp;
 	ns = old->mnt_ns;
 
@@ -3668,7 +3663,6 @@ static int do_move_mount(struct path *old_path,
 
 		err = -EINVAL;
 		p = p->mnt_parent;
-		flags |= MNT_TREE_BENEATH;
 	}
 
 	/*
@@ -3683,7 +3677,7 @@ static int do_move_mount(struct path *old_path,
 	if (mount_is_ancestor(old, p))
 		goto out;
 
-	err = attach_recursive_mnt(old, p, mp, flags);
+	err = attach_recursive_mnt(old, p, mp);
 	if (err)
 		goto out;
 
-- 
2.39.5

[PATCH v3 21/48] do_move_mount(): take dropping the old mountpoint into attach_recursive_mnt()

[Al Viro]

... and fold it with unhash_mnt() there - there's no need to retain a reference
to old_mp beyond that point, since by then all mountpoints we were going to add
are either explicitly pinned by get_mountpoint() or have stuff already added
to them.

Reviewed-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 7 ++-----
 1 file changed, 2 insertions(+), 5 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index adb37f06ba68..e5f8fde57c99 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -2682,7 +2682,7 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 	}
 
 	if (moving) {
-		unhash_mnt(source_mnt);
+		umount_mnt(source_mnt);
 		mnt_notify_add(source_mnt);
 	} else {
 		if (source_mnt->mnt_ns) {
@@ -3598,7 +3598,7 @@ static int do_move_mount(struct path *old_path,
 	struct mount *p;
 	struct mount *old;
 	struct mount *parent;
-	struct mountpoint *mp, *old_mp;
+	struct mountpoint *mp;
 	int err;
 	bool attached, beneath = flags & MNT_TREE_BENEATH;
 
@@ -3610,7 +3610,6 @@ static int do_move_mount(struct path *old_path,
 	p = real_mount(new_path->mnt);
 	parent = old->mnt_parent;
 	attached = mnt_has_parent(old);
-	old_mp = old->mnt_mp;
 	ns = old->mnt_ns;
 
 	err = -EINVAL;
@@ -3684,8 +3683,6 @@ static int do_move_mount(struct path *old_path,
 	/* if the mount is moved, it should no longer be expire
 	 * automatically */
 	list_del_init(&old->mnt_expire);
-	if (attached)
-		put_mountpoint(old_mp);
 out:
 	unlock_mount(mp);
 	if (!err) {
-- 
2.39.5

[PATCH v3 22/48] do_move_mount(): get rid of 'attached' flag

[Al Viro]

'attached' serves as a proxy for "source is a subtree of our namespace
and not the entirety of anon namespace"; finish massaging it away.

Reviewed-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 13 +++++--------
 1 file changed, 5 insertions(+), 8 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index e5f8fde57c99..7c7cc14da1ee 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -3600,7 +3600,7 @@ static int do_move_mount(struct path *old_path,
 	struct mount *parent;
 	struct mountpoint *mp;
 	int err;
-	bool attached, beneath = flags & MNT_TREE_BENEATH;
+	bool beneath = flags & MNT_TREE_BENEATH;
 
 	mp = do_lock_mount(new_path, beneath);
 	if (IS_ERR(mp))
@@ -3609,7 +3609,6 @@ static int do_move_mount(struct path *old_path,
 	old = real_mount(old_path->mnt);
 	p = real_mount(new_path->mnt);
 	parent = old->mnt_parent;
-	attached = mnt_has_parent(old);
 	ns = old->mnt_ns;
 
 	err = -EINVAL;
@@ -3622,6 +3621,9 @@ static int do_move_mount(struct path *old_path,
 		/* ... and the target should be in our namespace */
 		if (!check_mnt(p))
 			goto out;
+		/* parent of the source should not be shared */
+		if (IS_MNT_SHARED(parent))
+			goto out;
 	} else {
 		/*
 		 * otherwise the source must be the root of some anon namespace.
@@ -3649,11 +3651,6 @@ static int do_move_mount(struct path *old_path,
 	if (d_is_dir(new_path->dentry) !=
 	    d_is_dir(old_path->dentry))
 		goto out;
-	/*
-	 * Don't move a mount residing in a shared parent.
-	 */
-	if (attached && IS_MNT_SHARED(parent))
-		goto out;
 
 	if (beneath) {
 		err = can_move_mount_beneath(old_path, new_path, mp);
@@ -3686,7 +3683,7 @@ static int do_move_mount(struct path *old_path,
 out:
 	unlock_mount(mp);
 	if (!err) {
-		if (attached) {
+		if (!is_anon_ns(ns)) {
 			mntput_no_expire(parent);
 		} else {
 			/* Make sure we notice when we leak mounts. */
-- 
2.39.5

[PATCH v3 23/48] attach_recursive_mnt(): remove from expiry list on move

[Al Viro]

... rather than doing that in do_move_mount().  That's the main
obstacle to moving the protection of ->mnt_expire from namespace_sem
to mount_lock (spinlock-only), which would simplify several failure
exits.

Reviewed-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 9 +++------
 1 file changed, 3 insertions(+), 6 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index 7c7cc14da1ee..e8dc8af87548 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -2684,6 +2684,9 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 	if (moving) {
 		umount_mnt(source_mnt);
 		mnt_notify_add(source_mnt);
+		/* if the mount is moved, it should no longer be expired
+		 * automatically */
+		list_del_init(&source_mnt->mnt_expire);
 	} else {
 		if (source_mnt->mnt_ns) {
 			LIST_HEAD(head);
@@ -3674,12 +3677,6 @@ static int do_move_mount(struct path *old_path,
 		goto out;
 
 	err = attach_recursive_mnt(old, p, mp);
-	if (err)
-		goto out;
-
-	/* if the mount is moved, it should no longer be expire
-	 * automatically */
-	list_del_init(&old->mnt_expire);
 out:
 	unlock_mount(mp);
 	if (!err) {
-- 
2.39.5

[PATCH v3 24/48] take ->mnt_expire handling under mount_lock [read_seqlock_excl]

[Al Viro]

Doesn't take much massage, and we no longer need to make sure that
by the time of final mntput() the victim has been removed from the
list.  Makes life safer for ->d_automount() instances...

Rules:
	* all ->mnt_expire accesses are under mount_lock.
	* insertion into the list is done by mnt_set_expiry(), and
caller (->d_automount() instance) must hold a reference to mount
in question.  It shouldn't be done more than once for a mount.
	* if a mount on an expiry list is not yet mounted, it will
be ignored by anything that walks that list.
	* if the final mntput() finds its victim still on an expiry
list (in which case it must've never been mounted - umount_tree()
would've taken it out), it will remove the victim from the list.

Reviewed-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 28 +++++++++++-----------------
 1 file changed, 11 insertions(+), 17 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index e8dc8af87548..ff2281f780dc 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -1353,13 +1353,6 @@ static struct mount *clone_mnt(struct mount *old, struct dentry *root,
 		list_add(&mnt->mnt_slave, &old->mnt_slave);
 		mnt->mnt_master = old->mnt_master;
 	}
-
-	/* stick the duplicate mount on the same expiry list
-	 * as the original if that was on one */
-	if (flag & CL_EXPIRE) {
-		if (!list_empty(&old->mnt_expire))
-			list_add(&mnt->mnt_expire, &old->mnt_expire);
-	}
 	return mnt;
 
  out_free:
@@ -1452,6 +1445,8 @@ static void mntput_no_expire(struct mount *mnt)
 	rcu_read_unlock();
 
 	list_del(&mnt->mnt_instance);
+	if (unlikely(!list_empty(&mnt->mnt_expire)))
+		list_del(&mnt->mnt_expire);
 
 	if (unlikely(!list_empty(&mnt->mnt_mounts))) {
 		struct mount *p, *tmp;
@@ -2273,6 +2268,13 @@ struct mount *copy_tree(struct mount *src_root, struct dentry *dentry,
 			lock_mount_hash();
 			if (src_mnt->mnt.mnt_flags & MNT_LOCKED)
 				dst_mnt->mnt.mnt_flags |= MNT_LOCKED;
+			if (unlikely(flag & CL_EXPIRE)) {
+				/* stick the duplicate mount on the same expiry
+				 * list as the original if that was on one */
+				if (!list_empty(&src_mnt->mnt_expire))
+					list_add(&dst_mnt->mnt_expire,
+						 &src_mnt->mnt_expire);
+			}
 			list_add_tail(&dst_mnt->mnt_list, &res->mnt_list);
 			attach_mnt(dst_mnt, dst_parent, src_parent->mnt_mp);
 			unlock_mount_hash();
@@ -3891,12 +3893,6 @@ int finish_automount(struct vfsmount *m, const struct path *path)
 	namespace_unlock();
 	inode_unlock(dentry->d_inode);
 discard:
-	/* remove m from any expiration list it may be on */
-	if (!list_empty(&mnt->mnt_expire)) {
-		namespace_lock();
-		list_del_init(&mnt->mnt_expire);
-		namespace_unlock();
-	}
 	mntput(m);
 	return err;
 }
@@ -3908,11 +3904,9 @@ int finish_automount(struct vfsmount *m, const struct path *path)
  */
 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
 {
-	namespace_lock();
-
+	read_seqlock_excl(&mount_lock);
 	list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
-
-	namespace_unlock();
+	read_sequnlock_excl(&mount_lock);
 }
 EXPORT_SYMBOL(mnt_set_expiry);
 
-- 
2.39.5

[PATCH v3 25/48] pivot_root(): reorder tree surgeries, collapse unhash_mnt() and put_mountpoint()

[Al Viro]

attach new_mnt *before* detaching root_mnt; that way we don't need to keep hold
on the mountpoint and one more pair of unhash_mnt()/put_mountpoint() gets
folded together into umount_mnt().

Reviewed-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 11 +++++------
 1 file changed, 5 insertions(+), 6 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index ff2281f780dc..eee73e945a54 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -4685,7 +4685,7 @@ SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
 {
 	struct path new, old, root;
 	struct mount *new_mnt, *root_mnt, *old_mnt, *root_parent, *ex_parent;
-	struct mountpoint *old_mp, *root_mp;
+	struct mountpoint *old_mp;
 	int error;
 
 	if (!may_mount())
@@ -4748,20 +4748,19 @@ SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
 		goto out4;
 	lock_mount_hash();
 	umount_mnt(new_mnt);
-	root_mp = unhash_mnt(root_mnt);  /* we'll need its mountpoint */
 	if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
 		new_mnt->mnt.mnt_flags |= MNT_LOCKED;
 		root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
 	}
-	/* mount old root on put_old */
-	attach_mnt(root_mnt, old_mnt, old_mp);
 	/* mount new_root on / */
-	attach_mnt(new_mnt, root_parent, root_mp);
+	attach_mnt(new_mnt, root_parent, root_mnt->mnt_mp);
+	umount_mnt(root_mnt);
 	mnt_add_count(root_parent, -1);
+	/* mount old root on put_old */
+	attach_mnt(root_mnt, old_mnt, old_mp);
 	touch_mnt_namespace(current->nsproxy->mnt_ns);
 	/* A moved mount should not expire automatically */
 	list_del_init(&new_mnt->mnt_expire);
-	put_mountpoint(root_mp);
 	unlock_mount_hash();
 	mnt_notify_add(root_mnt);
 	mnt_notify_add(new_mnt);
-- 
2.39.5

[PATCH v3 26/48] combine __put_mountpoint() with unhash_mnt()

[Al Viro]

A call of unhash_mnt() is immediately followed by passing its return
value to __put_mountpoint(); the shrink list given to __put_mountpoint()
will be ex_mountpoints when called from umount_mnt() and list when called
from mntput_no_expire().

Replace with __umount_mnt(mount, shrink_list), moving the call of
__put_mountpoint() into it (and returning nothing), adjust the
callers.

Reviewed-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index eee73e945a54..521ffa52c906 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -1037,9 +1037,9 @@ static void __touch_mnt_namespace(struct mnt_namespace *ns)
 }
 
 /*
- * vfsmount lock must be held for write
+ * locks: mount_lock[write_seqlock]
  */
-static struct mountpoint *unhash_mnt(struct mount *mnt)
+static void __umount_mnt(struct mount *mnt, struct list_head *shrink_list)
 {
 	struct mountpoint *mp;
 	struct mount *parent = mnt->mnt_parent;
@@ -1052,15 +1052,15 @@ static struct mountpoint *unhash_mnt(struct mount *mnt)
 	hlist_del_init(&mnt->mnt_mp_list);
 	mp = mnt->mnt_mp;
 	mnt->mnt_mp = NULL;
-	return mp;
+	__put_mountpoint(mp, shrink_list);
 }
 
 /*
- * vfsmount lock must be held for write
+ * locks: mount_lock[write_seqlock], namespace_sem[excl] (for ex_mountpoints)
  */
 static void umount_mnt(struct mount *mnt)
 {
-	put_mountpoint(unhash_mnt(mnt));
+	__umount_mnt(mnt, &ex_mountpoints);
 }
 
 /*
@@ -1451,7 +1451,7 @@ static void mntput_no_expire(struct mount *mnt)
 	if (unlikely(!list_empty(&mnt->mnt_mounts))) {
 		struct mount *p, *tmp;
 		list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts,  mnt_child) {
-			__put_mountpoint(unhash_mnt(p), &list);
+			__umount_mnt(p, &list);
 			hlist_add_head(&p->mnt_umount, &mnt->mnt_stuck_children);
 		}
 	}
-- 
2.39.5

[PATCH v3 27/48] get rid of mountpoint->m_count

[Al Viro]

struct mountpoint has an odd kinda-sorta refcount in it.  It's always
either equal to or one above the number of mounts attached to that
mountpoint.

"One above" happens when a function takes a temporary reference to
mountpoint.  Things get simpler if we express that as inserting
a local object into ->m_list and removing it to drop the reference.

New calling conventions:

1) lock_mount(), do_lock_mount(), get_mountpoint() and lookup_mountpoint()
take an extra struct pinned_mountpoint * argument and returns 0/-E...
(or true/false in case of lookup_mountpoint()) instead of returning
struct mountpoint pointers.  In case of success, the struct mountpoint *
we used to get can be found as pinned_mountpoint.mp

2) unlock_mount() (always paired with lock_mount()/do_lock_mount()) takes
an address of struct pinned_mountpoint - the same that had been passed to
lock_mount()/do_lock_mount().

3) put_mountpoint() for a temporary reference (paired with get_mountpoint()
or lookup_mountpoint()) is replaced with unpin_mountpoint(), which takes
the address of pinned_mountpoint we passed to matching {get,lookup}_mountpoint().

4) all instances of pinned_mountpoint are local variables; they always live on
stack.  {} is used for initializer, after successful {get,lookup}_mountpoint()
we must make sure to call unpin_mountpoint() before leaving the scope and
after successful {do_,}lock_mount() we must make sure to call unlock_mount()
before leaving the scope.

5) all manipulations of ->m_count are gone, along with ->m_count itself.
struct mountpoint lives while its ->m_list is non-empty.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/mount.h     |   1 -
 fs/namespace.c | 186 ++++++++++++++++++++++++-------------------------
 2 files changed, 92 insertions(+), 95 deletions(-)

diff --git a/fs/mount.h b/fs/mount.h
index fb93d3e16724..4355c482a841 100644
--- a/fs/mount.h
+++ b/fs/mount.h
@@ -44,7 +44,6 @@ struct mountpoint {
 	struct hlist_node m_hash;
 	struct dentry *m_dentry;
 	struct hlist_head m_list;
-	int m_count;
 };
 
 struct mount {
diff --git a/fs/namespace.c b/fs/namespace.c
index 521ffa52c906..6df0436bfcb9 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -910,42 +910,48 @@ bool __is_local_mountpoint(const struct dentry *dentry)
 	return is_covered;
 }
 
-static struct mountpoint *lookup_mountpoint(struct dentry *dentry)
+struct pinned_mountpoint {
+	struct hlist_node node;
+	struct mountpoint *mp;
+};
+
+static bool lookup_mountpoint(struct dentry *dentry, struct pinned_mountpoint *m)
 {
 	struct hlist_head *chain = mp_hash(dentry);
 	struct mountpoint *mp;
 
 	hlist_for_each_entry(mp, chain, m_hash) {
 		if (mp->m_dentry == dentry) {
-			mp->m_count++;
-			return mp;
+			hlist_add_head(&m->node, &mp->m_list);
+			m->mp = mp;
+			return true;
 		}
 	}
-	return NULL;
+	return false;
 }
 
-static struct mountpoint *get_mountpoint(struct dentry *dentry)
+static int get_mountpoint(struct dentry *dentry, struct pinned_mountpoint *m)
 {
-	struct mountpoint *mp, *new = NULL;
+	struct mountpoint *mp __free(kfree) = NULL;
+	bool found;
 	int ret;
 
 	if (d_mountpoint(dentry)) {
 		/* might be worth a WARN_ON() */
 		if (d_unlinked(dentry))
-			return ERR_PTR(-ENOENT);
+			return -ENOENT;
 mountpoint:
 		read_seqlock_excl(&mount_lock);
-		mp = lookup_mountpoint(dentry);
+		found = lookup_mountpoint(dentry, m);
 		read_sequnlock_excl(&mount_lock);
-		if (mp)
-			goto done;
+		if (found)
+			return 0;
 	}
 
-	if (!new)
-		new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
-	if (!new)
-		return ERR_PTR(-ENOMEM);
-
+	if (!mp)
+		mp = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
+	if (!mp)
+		return -ENOMEM;
 
 	/* Exactly one processes may set d_mounted */
 	ret = d_set_mounted(dentry);
@@ -955,34 +961,28 @@ static struct mountpoint *get_mountpoint(struct dentry *dentry)
 		goto mountpoint;
 
 	/* The dentry is not available as a mountpoint? */
-	mp = ERR_PTR(ret);
 	if (ret)
-		goto done;
+		return ret;
 
 	/* Add the new mountpoint to the hash table */
 	read_seqlock_excl(&mount_lock);
-	new->m_dentry = dget(dentry);
-	new->m_count = 1;
-	hlist_add_head(&new->m_hash, mp_hash(dentry));
-	INIT_HLIST_HEAD(&new->m_list);
+	mp->m_dentry = dget(dentry);
+	hlist_add_head(&mp->m_hash, mp_hash(dentry));
+	INIT_HLIST_HEAD(&mp->m_list);
+	hlist_add_head(&m->node, &mp->m_list);
+	m->mp = no_free_ptr(mp);
 	read_sequnlock_excl(&mount_lock);
-
-	mp = new;
-	new = NULL;
-done:
-	kfree(new);
-	return mp;
+	return 0;
 }
 
 /*
  * vfsmount lock must be held.  Additionally, the caller is responsible
  * for serializing calls for given disposal list.
  */
-static void __put_mountpoint(struct mountpoint *mp, struct list_head *list)
+static void maybe_free_mountpoint(struct mountpoint *mp, struct list_head *list)
 {
-	if (!--mp->m_count) {
+	if (hlist_empty(&mp->m_list)) {
 		struct dentry *dentry = mp->m_dentry;
-		BUG_ON(!hlist_empty(&mp->m_list));
 		spin_lock(&dentry->d_lock);
 		dentry->d_flags &= ~DCACHE_MOUNTED;
 		spin_unlock(&dentry->d_lock);
@@ -992,10 +992,15 @@ static void __put_mountpoint(struct mountpoint *mp, struct list_head *list)
 	}
 }
 
-/* called with namespace_lock and vfsmount lock */
-static void put_mountpoint(struct mountpoint *mp)
+/*
+ * locks: mount_lock [read_seqlock_excl], namespace_sem [excl]
+ */
+static void unpin_mountpoint(struct pinned_mountpoint *m)
 {
-	__put_mountpoint(mp, &ex_mountpoints);
+	if (m->mp) {
+		hlist_del(&m->node);
+		maybe_free_mountpoint(m->mp, &ex_mountpoints);
+	}
 }
 
 static inline int check_mnt(struct mount *mnt)
@@ -1052,7 +1057,7 @@ static void __umount_mnt(struct mount *mnt, struct list_head *shrink_list)
 	hlist_del_init(&mnt->mnt_mp_list);
 	mp = mnt->mnt_mp;
 	mnt->mnt_mp = NULL;
-	__put_mountpoint(mp, shrink_list);
+	maybe_free_mountpoint(mp, shrink_list);
 }
 
 /*
@@ -1070,7 +1075,6 @@ void mnt_set_mountpoint(struct mount *mnt,
 			struct mountpoint *mp,
 			struct mount *child_mnt)
 {
-	mp->m_count++;
 	mnt_add_count(mnt, 1);	/* essentially, that's mntget */
 	child_mnt->mnt_mountpoint = mp->m_dentry;
 	child_mnt->mnt_parent = mnt;
@@ -1122,7 +1126,7 @@ void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct m
 
 	attach_mnt(mnt, parent, mp);
 
-	put_mountpoint(old_mp);
+	maybe_free_mountpoint(old_mp, &ex_mountpoints);
 	mnt_add_count(old_parent, -1);
 }
 
@@ -2030,25 +2034,24 @@ static int do_umount(struct mount *mnt, int flags)
  */
 void __detach_mounts(struct dentry *dentry)
 {
-	struct mountpoint *mp;
+	struct pinned_mountpoint mp = {};
 	struct mount *mnt;
 
 	namespace_lock();
 	lock_mount_hash();
-	mp = lookup_mountpoint(dentry);
-	if (!mp)
+	if (!lookup_mountpoint(dentry, &mp))
 		goto out_unlock;
 
 	event++;
-	while (!hlist_empty(&mp->m_list)) {
-		mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list);
+	while (mp.node.next) {
+		mnt = hlist_entry(mp.node.next, struct mount, mnt_mp_list);
 		if (mnt->mnt.mnt_flags & MNT_UMOUNT) {
 			umount_mnt(mnt);
 			hlist_add_head(&mnt->mnt_umount, &unmounted);
 		}
 		else umount_tree(mnt, UMOUNT_CONNECTED);
 	}
-	put_mountpoint(mp);
+	unpin_mountpoint(&mp);
 out_unlock:
 	unlock_mount_hash();
 	namespace_unlock();
@@ -2641,7 +2644,7 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 	struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
 	HLIST_HEAD(tree_list);
 	struct mnt_namespace *ns = dest_mnt->mnt_ns;
-	struct mountpoint *smp;
+	struct pinned_mountpoint root = {};
 	struct mountpoint *shorter = NULL;
 	struct mount *child, *p;
 	struct mount *top;
@@ -2657,9 +2660,9 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 		if (!shorter && is_mnt_ns_file(top->mnt.mnt_root))
 			shorter = top->mnt_mp;
 	}
-	smp = get_mountpoint(top->mnt.mnt_root);
-	if (IS_ERR(smp))
-		return PTR_ERR(smp);
+	err = get_mountpoint(top->mnt.mnt_root, &root);
+	if (err)
+		return err;
 
 	/* Is there space to add these mounts to the mount namespace? */
 	if (!moving) {
@@ -2719,7 +2722,7 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 		q = __lookup_mnt(&child->mnt_parent->mnt,
 				 child->mnt_mountpoint);
 		if (q) {
-			struct mountpoint *mp = smp;
+			struct mountpoint *mp = root.mp;
 			struct mount *r = child;
 			while (unlikely(r->overmount))
 				r = r->overmount;
@@ -2729,7 +2732,7 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 		}
 		commit_tree(child);
 	}
-	put_mountpoint(smp);
+	unpin_mountpoint(&root);
 	unlock_mount_hash();
 
 	return 0;
@@ -2746,7 +2749,7 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 	ns->pending_mounts = 0;
 
 	read_seqlock_excl(&mount_lock);
-	put_mountpoint(smp);
+	unpin_mountpoint(&root);
 	read_sequnlock_excl(&mount_lock);
 
 	return err;
@@ -2786,12 +2789,12 @@ static int attach_recursive_mnt(struct mount *source_mnt,
  * Return: Either the target mountpoint on the top mount or the top
  *         mount's mountpoint.
  */
-static struct mountpoint *do_lock_mount(struct path *path, bool beneath)
+static int do_lock_mount(struct path *path, struct pinned_mountpoint *pinned, bool beneath)
 {
 	struct vfsmount *mnt = path->mnt;
 	struct dentry *dentry;
-	struct mountpoint *mp = ERR_PTR(-ENOENT);
 	struct path under = {};
+	int err = -ENOENT;
 
 	for (;;) {
 		struct mount *m = real_mount(mnt);
@@ -2829,8 +2832,8 @@ static struct mountpoint *do_lock_mount(struct path *path, bool beneath)
 			path->dentry = dget(mnt->mnt_root);
 			continue;	// got overmounted
 		}
-		mp = get_mountpoint(dentry);
-		if (IS_ERR(mp))
+		err = get_mountpoint(dentry, pinned);
+		if (err)
 			break;
 		if (beneath) {
 			/*
@@ -2841,25 +2844,25 @@ static struct mountpoint *do_lock_mount(struct path *path, bool beneath)
 			 */
 			path_put(&under);
 		}
-		return mp;
+		return 0;
 	}
 	namespace_unlock();
 	inode_unlock(dentry->d_inode);
 	if (beneath)
 		path_put(&under);
-	return mp;
+	return err;
 }
 
-static inline struct mountpoint *lock_mount(struct path *path)
+static inline int lock_mount(struct path *path, struct pinned_mountpoint *m)
 {
-	return do_lock_mount(path, false);
+	return do_lock_mount(path, m, false);
 }
 
-static void unlock_mount(struct mountpoint *where)
+static void unlock_mount(struct pinned_mountpoint *m)
 {
-	inode_unlock(where->m_dentry->d_inode);
+	inode_unlock(m->mp->m_dentry->d_inode);
 	read_seqlock_excl(&mount_lock);
-	put_mountpoint(where);
+	unpin_mountpoint(m);
 	read_sequnlock_excl(&mount_lock);
 	namespace_unlock();
 }
@@ -3024,7 +3027,7 @@ static int do_loopback(struct path *path, const char *old_name,
 {
 	struct path old_path;
 	struct mount *mnt = NULL, *parent;
-	struct mountpoint *mp;
+	struct pinned_mountpoint mp = {};
 	int err;
 	if (!old_name || !*old_name)
 		return -EINVAL;
@@ -3036,11 +3039,9 @@ static int do_loopback(struct path *path, const char *old_name,
 	if (mnt_ns_loop(old_path.dentry))
 		goto out;
 
-	mp = lock_mount(path);
-	if (IS_ERR(mp)) {
-		err = PTR_ERR(mp);
+	err = lock_mount(path, &mp);
+	if (err)
 		goto out;
-	}
 
 	parent = real_mount(path->mnt);
 	if (!check_mnt(parent))
@@ -3052,14 +3053,14 @@ static int do_loopback(struct path *path, const char *old_name,
 		goto out2;
 	}
 
-	err = graft_tree(mnt, parent, mp);
+	err = graft_tree(mnt, parent, mp.mp);
 	if (err) {
 		lock_mount_hash();
 		umount_tree(mnt, UMOUNT_SYNC);
 		unlock_mount_hash();
 	}
 out2:
-	unlock_mount(mp);
+	unlock_mount(&mp);
 out:
 	path_put(&old_path);
 	return err;
@@ -3603,13 +3604,13 @@ static int do_move_mount(struct path *old_path,
 	struct mount *p;
 	struct mount *old;
 	struct mount *parent;
-	struct mountpoint *mp;
+	struct pinned_mountpoint mp;
 	int err;
 	bool beneath = flags & MNT_TREE_BENEATH;
 
-	mp = do_lock_mount(new_path, beneath);
-	if (IS_ERR(mp))
-		return PTR_ERR(mp);
+	err = do_lock_mount(new_path, &mp, beneath);
+	if (err)
+		return err;
 
 	old = real_mount(old_path->mnt);
 	p = real_mount(new_path->mnt);
@@ -3658,7 +3659,7 @@ static int do_move_mount(struct path *old_path,
 		goto out;
 
 	if (beneath) {
-		err = can_move_mount_beneath(old_path, new_path, mp);
+		err = can_move_mount_beneath(old_path, new_path, mp.mp);
 		if (err)
 			goto out;
 
@@ -3678,9 +3679,9 @@ static int do_move_mount(struct path *old_path,
 	if (mount_is_ancestor(old, p))
 		goto out;
 
-	err = attach_recursive_mnt(old, p, mp);
+	err = attach_recursive_mnt(old, p, mp.mp);
 out:
-	unlock_mount(mp);
+	unlock_mount(&mp);
 	if (!err) {
 		if (!is_anon_ns(ns)) {
 			mntput_no_expire(parent);
@@ -3750,7 +3751,7 @@ static int do_new_mount_fc(struct fs_context *fc, struct path *mountpoint,
 			   unsigned int mnt_flags)
 {
 	struct vfsmount *mnt;
-	struct mountpoint *mp;
+	struct pinned_mountpoint mp = {};
 	struct super_block *sb = fc->root->d_sb;
 	int error;
 
@@ -3771,13 +3772,12 @@ static int do_new_mount_fc(struct fs_context *fc, struct path *mountpoint,
 
 	mnt_warn_timestamp_expiry(mountpoint, mnt);
 
-	mp = lock_mount(mountpoint);
-	if (IS_ERR(mp)) {
-		mntput(mnt);
-		return PTR_ERR(mp);
+	error = lock_mount(mountpoint, &mp);
+	if (!error) {
+		error = do_add_mount(real_mount(mnt), mp.mp,
+				     mountpoint, mnt_flags);
+		unlock_mount(&mp);
 	}
-	error = do_add_mount(real_mount(mnt), mp, mountpoint, mnt_flags);
-	unlock_mount(mp);
 	if (error < 0)
 		mntput(mnt);
 	return error;
@@ -3845,7 +3845,7 @@ static int do_new_mount(struct path *path, const char *fstype, int sb_flags,
 int finish_automount(struct vfsmount *m, const struct path *path)
 {
 	struct dentry *dentry = path->dentry;
-	struct mountpoint *mp;
+	struct pinned_mountpoint mp = {};
 	struct mount *mnt;
 	int err;
 
@@ -3877,14 +3877,13 @@ int finish_automount(struct vfsmount *m, const struct path *path)
 		err = 0;
 		goto discard_locked;
 	}
-	mp = get_mountpoint(dentry);
-	if (IS_ERR(mp)) {
-		err = PTR_ERR(mp);
+	err = get_mountpoint(dentry, &mp);
+	if (err)
 		goto discard_locked;
-	}
 
-	err = do_add_mount(mnt, mp, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
-	unlock_mount(mp);
+	err = do_add_mount(mnt, mp.mp, path,
+			   path->mnt->mnt_flags | MNT_SHRINKABLE);
+	unlock_mount(&mp);
 	if (unlikely(err))
 		goto discard;
 	return 0;
@@ -4685,7 +4684,7 @@ SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
 {
 	struct path new, old, root;
 	struct mount *new_mnt, *root_mnt, *old_mnt, *root_parent, *ex_parent;
-	struct mountpoint *old_mp;
+	struct pinned_mountpoint old_mp = {};
 	int error;
 
 	if (!may_mount())
@@ -4706,9 +4705,8 @@ SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
 		goto out2;
 
 	get_fs_root(current->fs, &root);
-	old_mp = lock_mount(&old);
-	error = PTR_ERR(old_mp);
-	if (IS_ERR(old_mp))
+	error = lock_mount(&old, &old_mp);
+	if (error)
 		goto out3;
 
 	error = -EINVAL;
@@ -4757,7 +4755,7 @@ SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
 	umount_mnt(root_mnt);
 	mnt_add_count(root_parent, -1);
 	/* mount old root on put_old */
-	attach_mnt(root_mnt, old_mnt, old_mp);
+	attach_mnt(root_mnt, old_mnt, old_mp.mp);
 	touch_mnt_namespace(current->nsproxy->mnt_ns);
 	/* A moved mount should not expire automatically */
 	list_del_init(&new_mnt->mnt_expire);
@@ -4767,7 +4765,7 @@ SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
 	chroot_fs_refs(&root, &new);
 	error = 0;
 out4:
-	unlock_mount(old_mp);
+	unlock_mount(&old_mp);
 	if (!error)
 		mntput_no_expire(ex_parent);
 out3:
-- 
2.39.5

[PATCH v3 28/48] don't have mounts pin their parents

[Al Viro]

Simplify the rules for mount refcounts.  Current rules include:
	* being a namespace root => +1
	* being someone's child => +1
	* being someone's child => +1 to parent's refcount, unless you've
				   already been through umount_tree().

The last part is not needed at all.  It makes for more places where need
to decrement refcounts and it creates an asymmetry between the situations
for something that has never been a part of a namespace and something that
left one, both for no good reason.

If mount's refcount has additions from its children, we know that
	* it's either someone's child itself (and will remain so
until umount_tree(), at which point contributions from children
will disappear), or
	* or is the root of namespace (and will remain such until
it either becomes someone's child in another namespace or goes through
umount_tree()), or
	* it is the root of some tree copy, and is currently pinned
by the caller of copy_tree() (and remains such until it either gets
into namespace, or goes to umount_tree()).
In all cases we already have contribution(s) to refcount that will last
as long as the contribution from children remains.  In other words, the
lifetime is not affected by refcount contributions from children.

It might be useful for "is it busy" checks, but those are actually
no harder to express without it.

NB: propagate_mnt_busy() part is an equivalent transformation, ugly as it
is; the current logics is actually wrong and may give false negatives,
but fixing that is for a separate patch (probably earlier in the queue).

Reviewed-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 31 +++++++++----------------------
 fs/pnode.c     | 49 +++++++++++++++++--------------------------------
 2 files changed, 26 insertions(+), 54 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index 6df0436bfcb9..4bdf6a6e75ca 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -1075,7 +1075,6 @@ void mnt_set_mountpoint(struct mount *mnt,
 			struct mountpoint *mp,
 			struct mount *child_mnt)
 {
-	mnt_add_count(mnt, 1);	/* essentially, that's mntget */
 	child_mnt->mnt_mountpoint = mp->m_dentry;
 	child_mnt->mnt_parent = mnt;
 	child_mnt->mnt_mp = mp;
@@ -1118,7 +1117,6 @@ static void attach_mnt(struct mount *mnt, struct mount *parent,
 void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt)
 {
 	struct mountpoint *old_mp = mnt->mnt_mp;
-	struct mount *old_parent = mnt->mnt_parent;
 
 	list_del_init(&mnt->mnt_child);
 	hlist_del_init(&mnt->mnt_mp_list);
@@ -1127,7 +1125,6 @@ void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct m
 	attach_mnt(mnt, parent, mp);
 
 	maybe_free_mountpoint(old_mp, &ex_mountpoints);
-	mnt_add_count(old_parent, -1);
 }
 
 static inline struct mount *node_to_mount(struct rb_node *node)
@@ -1652,23 +1649,19 @@ const struct seq_operations mounts_op = {
 int may_umount_tree(struct vfsmount *m)
 {
 	struct mount *mnt = real_mount(m);
-	int actual_refs = 0;
-	int minimum_refs = 0;
-	struct mount *p;
-	BUG_ON(!m);
+	bool busy = false;
 
 	/* write lock needed for mnt_get_count */
 	lock_mount_hash();
-	for (p = mnt; p; p = next_mnt(p, mnt)) {
-		actual_refs += mnt_get_count(p);
-		minimum_refs += 2;
+	for (struct mount *p = mnt; p; p = next_mnt(p, mnt)) {
+		if (mnt_get_count(p) > (p == mnt ? 2 : 1)) {
+			busy = true;
+			break;
+		}
 	}
 	unlock_mount_hash();
 
-	if (actual_refs > minimum_refs)
-		return 0;
-
-	return 1;
+	return !busy;
 }
 
 EXPORT_SYMBOL(may_umount_tree);
@@ -1869,7 +1862,6 @@ static void umount_tree(struct mount *mnt, enum umount_tree_flags how)
 
 		disconnect = disconnect_mount(p, how);
 		if (mnt_has_parent(p)) {
-			mnt_add_count(p->mnt_parent, -1);
 			if (!disconnect) {
 				/* Don't forget about p */
 				list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts);
@@ -1946,7 +1938,7 @@ static int do_umount(struct mount *mnt, int flags)
 		 * all race cases, but it's a slowpath.
 		 */
 		lock_mount_hash();
-		if (mnt_get_count(mnt) != 2) {
+		if (!list_empty(&mnt->mnt_mounts) || mnt_get_count(mnt) != 2) {
 			unlock_mount_hash();
 			return -EBUSY;
 		}
@@ -3683,9 +3675,7 @@ static int do_move_mount(struct path *old_path,
 out:
 	unlock_mount(&mp);
 	if (!err) {
-		if (!is_anon_ns(ns)) {
-			mntput_no_expire(parent);
-		} else {
+		if (is_anon_ns(ns)) {
 			/* Make sure we notice when we leak mounts. */
 			VFS_WARN_ON_ONCE(!mnt_ns_empty(ns));
 			free_mnt_ns(ns);
@@ -4753,7 +4743,6 @@ SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
 	/* mount new_root on / */
 	attach_mnt(new_mnt, root_parent, root_mnt->mnt_mp);
 	umount_mnt(root_mnt);
-	mnt_add_count(root_parent, -1);
 	/* mount old root on put_old */
 	attach_mnt(root_mnt, old_mnt, old_mp.mp);
 	touch_mnt_namespace(current->nsproxy->mnt_ns);
@@ -4766,8 +4755,6 @@ SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
 	error = 0;
 out4:
 	unlock_mount(&old_mp);
-	if (!error)
-		mntput_no_expire(ex_parent);
 out3:
 	path_put(&root);
 out2:
diff --git a/fs/pnode.c b/fs/pnode.c
index 901d40946d34..827d71736ac5 100644
--- a/fs/pnode.c
+++ b/fs/pnode.c
@@ -332,21 +332,6 @@ int propagate_mnt(struct mount *dest_mnt, struct mountpoint *dest_mp,
 	return ret;
 }
 
-static struct mount *find_topper(struct mount *mnt)
-{
-	/* If there is exactly one mount covering mnt completely return it. */
-	struct mount *child;
-
-	if (!list_is_singular(&mnt->mnt_mounts))
-		return NULL;
-
-	child = list_first_entry(&mnt->mnt_mounts, struct mount, mnt_child);
-	if (child->mnt_mountpoint != mnt->mnt.mnt_root)
-		return NULL;
-
-	return child;
-}
-
 /*
  * return true if the refcount is greater than count
  */
@@ -404,12 +389,8 @@ bool propagation_would_overmount(const struct mount *from,
  */
 int propagate_mount_busy(struct mount *mnt, int refcnt)
 {
-	struct mount *m, *child, *topper;
 	struct mount *parent = mnt->mnt_parent;
 
-	if (mnt == parent)
-		return do_refcount_check(mnt, refcnt);
-
 	/*
 	 * quickly check if the current mount can be unmounted.
 	 * If not, we don't have to go checking for all other
@@ -418,23 +399,27 @@ int propagate_mount_busy(struct mount *mnt, int refcnt)
 	if (!list_empty(&mnt->mnt_mounts) || do_refcount_check(mnt, refcnt))
 		return 1;
 
-	for (m = propagation_next(parent, parent); m;
+	if (mnt == parent)
+		return 0;
+
+	for (struct mount *m = propagation_next(parent, parent); m;
 	     		m = propagation_next(m, parent)) {
-		int count = 1;
-		child = __lookup_mnt(&m->mnt, mnt->mnt_mountpoint);
-		if (!child)
-			continue;
+		struct list_head *head;
+		struct mount *child = __lookup_mnt(&m->mnt, mnt->mnt_mountpoint);
 
-		/* Is there exactly one mount on the child that covers
-		 * it completely whose reference should be ignored?
-		 */
-		topper = find_topper(child);
-		if (topper)
-			count += 1;
-		else if (!list_empty(&child->mnt_mounts))
+		if (!child)
 			continue;
 
-		if (do_refcount_check(child, count))
+		head = &child->mnt_mounts;
+		if (!list_empty(head)) {
+			/*
+			 * a mount that covers child completely wouldn't prevent
+			 * it being pulled out; any other would.
+			 */
+			if (!list_is_singular(head) || !child->overmount)
+				continue;
+		}
+		if (do_refcount_check(child, 1))
 			return 1;
 	}
 	return 0;
-- 
2.39.5

[PATCH v3 29/48] mount: separate the flags accessed only under namespace_sem

[Al Viro]

Several flags are updated and checked only under namespace_sem; we are
already making use of that when we are checking them without mount_lock,
but we have to hold mount_lock for all updates, which makes things
clumsier than they have to be.

Take MNT_SHARED, MNT_UNBINDABLE, MNT_MARKED and MNT_UMOUNT_CANDIDATE
into a separate field (->mnt_t_flags), renaming them to T_SHARED,
etc. to avoid confusion.  All accesses must be under namespace_sem.

That changes locking requirements for mnt_change_propagation() and
set_mnt_shared() - only namespace_sem is needed now.  The same goes
for SET_MNT_MARKED et.al.

There might be more flags moved from ->mnt_flags to that field;
this is just the initial set.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 .../filesystems/propagate_umount.txt          | 12 +++++-----
 fs/mount.h                                    | 17 ++++++++++++++
 fs/namespace.c                                |  4 ----
 fs/pnode.c                                    | 22 +++++++++----------
 fs/pnode.h                                    | 19 +++++++++-------
 include/linux/mount.h                         | 18 ++-------------
 6 files changed, 46 insertions(+), 46 deletions(-)

diff --git a/Documentation/filesystems/propagate_umount.txt b/Documentation/filesystems/propagate_umount.txt
index 6906903a8aa2..c90349e5b889 100644
--- a/Documentation/filesystems/propagate_umount.txt
+++ b/Documentation/filesystems/propagate_umount.txt
@@ -453,11 +453,11 @@ original set.
 So let's go for
 	* original set ("set").  Linkage via mnt_list
 	* undecided candidates ("candidates").  Subset of a list,
-consisting of all its elements marked with a new flag (MNT_UMOUNT_CANDIDATE).
+consisting of all its elements marked with a new flag (T_UMOUNT_CANDIDATE).
 Initially all elements of the list will be marked that way; in the
 end the list will become empty and no mounts will remain marked with
 that flag.
-	* Reuse MNT_MARKED for "has been already seen by trim_ancestors()".
+	* Reuse T_MARKED for "has been already seen by trim_ancestors()".
 	* anything in U that hadn't been in the original set - elements of
 candidates will gradually be either discarded or moved there.  In other
 words, it's the candidates we have already decided to unmount.	Its role
@@ -465,13 +465,13 @@ is reasonably close to the old "to_umount", so let's use that name.
 Linkage via mnt_list.
 
 For gather_candidates() we'll need to maintain both candidates (S -
-set) and intersection of S with set.  Use MNT_UMOUNT_CANDIDATE for
+set) and intersection of S with set.  Use T_UMOUNT_CANDIDATE for
 all elements we encounter, putting the ones not already in the original
 set into the list of candidates.  When we are done, strip that flag from
 all elements of the original set.  That gives a cheap way to check
 if element belongs to S (in gather_candidates) and to candidates
 itself (at later stages).  Call that predicate is_candidate(); it would
-be m->mnt_flags & MNT_UMOUNT_CANDIDATE.
+be m->mnt_t_flags & T_UMOUNT_CANDIDATE.
 
 All elements of the original set are marked with MNT_UMOUNT and we'll
 need the same for elements added when joining the contents of to_umount
@@ -480,5 +480,5 @@ to to_umount; that's close to what the old 'umount_one' is doing, so
 let's keep that name.  It also gives us another predicate we need -
 "belongs to union of set and to_umount"; will_be_unmounted() for now.
 
-Removals from the candidates list should strip both MNT_MARKED and
-MNT_UMOUNT_CANDIDATE; call it remove_from_candidates_list().
+Removals from the candidates list should strip both T_MARKED and
+T_UMOUNT_CANDIDATE; call it remove_from_candidates_list().
diff --git a/fs/mount.h b/fs/mount.h
index 4355c482a841..f299dc85446d 100644
--- a/fs/mount.h
+++ b/fs/mount.h
@@ -84,6 +84,7 @@ struct mount {
 	struct list_head to_notify;	/* need to queue notification */
 	struct mnt_namespace *prev_ns;	/* previous namespace (NULL if none) */
 #endif
+	int mnt_t_flags;		/* namespace_sem-protected flags */
 	int mnt_id;			/* mount identifier, reused */
 	u64 mnt_id_unique;		/* mount ID unique until reboot */
 	int mnt_group_id;		/* peer group identifier */
@@ -93,6 +94,22 @@ struct mount {
 	struct mount *overmount;	/* mounted on ->mnt_root */
 } __randomize_layout;
 
+enum {
+	T_SHARED		= 1, /* mount is shared */
+	T_UNBINDABLE		= 2, /* mount is unbindable */
+	T_MARKED		= 4, /* internal mark for propagate_... */
+	T_UMOUNT_CANDIDATE	= 8, /* for propagate_umount */
+
+	/*
+	 * T_SHARED_MASK is the set of flags that should be cleared when a
+	 * mount becomes shared.  Currently, this is only the flag that says a
+	 * mount cannot be bind mounted, since this is how we create a mount
+	 * that shares events with another mount.  If you add a new T_*
+	 * flag, consider how it interacts with shared mounts.
+	 */
+	T_SHARED_MASK	= T_UNBINDABLE,
+};
+
 #define MNT_NS_INTERNAL ERR_PTR(-EINVAL) /* distinct from any mnt_namespace */
 
 static inline struct mount *real_mount(struct vfsmount *mnt)
diff --git a/fs/namespace.c b/fs/namespace.c
index 4bdf6a6e75ca..da27365418a5 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -2917,10 +2917,8 @@ static int do_change_type(struct path *path, int ms_flags)
 			goto out_unlock;
 	}
 
-	lock_mount_hash();
 	for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
 		change_mnt_propagation(m, type);
-	unlock_mount_hash();
 
  out_unlock:
 	namespace_unlock();
@@ -3409,9 +3407,7 @@ static int do_set_group(struct path *from_path, struct path *to_path)
 	if (IS_MNT_SHARED(from)) {
 		to->mnt_group_id = from->mnt_group_id;
 		list_add(&to->mnt_share, &from->mnt_share);
-		lock_mount_hash();
 		set_mnt_shared(to);
-		unlock_mount_hash();
 	}
 
 	err = 0;
diff --git a/fs/pnode.c b/fs/pnode.c
index 827d71736ac5..b997663de6d0 100644
--- a/fs/pnode.c
+++ b/fs/pnode.c
@@ -112,7 +112,7 @@ static int do_make_slave(struct mount *mnt)
 }
 
 /*
- * vfsmount lock must be held for write
+ * EXCL[namespace_sem]
  */
 void change_mnt_propagation(struct mount *mnt, int type)
 {
@@ -125,9 +125,9 @@ void change_mnt_propagation(struct mount *mnt, int type)
 		list_del_init(&mnt->mnt_slave);
 		mnt->mnt_master = NULL;
 		if (type == MS_UNBINDABLE)
-			mnt->mnt.mnt_flags |= MNT_UNBINDABLE;
+			mnt->mnt_t_flags |= T_UNBINDABLE;
 		else
-			mnt->mnt.mnt_flags &= ~MNT_UNBINDABLE;
+			mnt->mnt_t_flags &= ~T_UNBINDABLE;
 	}
 }
 
@@ -263,9 +263,9 @@ static int propagate_one(struct mount *m, struct mountpoint *dest_mp)
 		return PTR_ERR(child);
 	read_seqlock_excl(&mount_lock);
 	mnt_set_mountpoint(m, dest_mp, child);
+	read_sequnlock_excl(&mount_lock);
 	if (m->mnt_master != dest_master)
 		SET_MNT_MARK(m->mnt_master);
-	read_sequnlock_excl(&mount_lock);
 	last_dest = m;
 	last_source = child;
 	hlist_add_head(&child->mnt_hash, list);
@@ -322,13 +322,11 @@ int propagate_mnt(struct mount *dest_mnt, struct mountpoint *dest_mp,
 		} while (n != m);
 	}
 out:
-	read_seqlock_excl(&mount_lock);
 	hlist_for_each_entry(n, tree_list, mnt_hash) {
 		m = n->mnt_parent;
 		if (m->mnt_master != dest_mnt->mnt_master)
 			CLEAR_MNT_MARK(m->mnt_master);
 	}
-	read_sequnlock_excl(&mount_lock);
 	return ret;
 }
 
@@ -447,7 +445,7 @@ void propagate_mount_unlock(struct mount *mnt)
 
 static inline bool is_candidate(struct mount *m)
 {
-	return m->mnt.mnt_flags & MNT_UMOUNT_CANDIDATE;
+	return m->mnt_t_flags & T_UMOUNT_CANDIDATE;
 }
 
 static inline bool will_be_unmounted(struct mount *m)
@@ -464,7 +462,7 @@ static void umount_one(struct mount *m, struct list_head *to_umount)
 
 static void remove_from_candidate_list(struct mount *m)
 {
-	m->mnt.mnt_flags &= ~(MNT_MARKED | MNT_UMOUNT_CANDIDATE);
+	m->mnt_t_flags &= ~(T_MARKED | T_UMOUNT_CANDIDATE);
 	list_del_init(&m->mnt_list);
 }
 
@@ -476,7 +474,7 @@ static void gather_candidates(struct list_head *set,
 	list_for_each_entry(m, set, mnt_list) {
 		if (is_candidate(m))
 			continue;
-		m->mnt.mnt_flags |= MNT_UMOUNT_CANDIDATE;
+		m->mnt_t_flags |= T_UMOUNT_CANDIDATE;
 		p = m->mnt_parent;
 		q = propagation_next(p, p);
 		while (q) {
@@ -494,7 +492,7 @@ static void gather_candidates(struct list_head *set,
 					q = skip_propagation_subtree(q, p);
 					continue;
 				}
-				child->mnt.mnt_flags |= MNT_UMOUNT_CANDIDATE;
+				child->mnt_t_flags |= T_UMOUNT_CANDIDATE;
 				if (!will_be_unmounted(child))
 					list_add(&child->mnt_list, candidates);
 			}
@@ -502,7 +500,7 @@ static void gather_candidates(struct list_head *set,
 		}
 	}
 	list_for_each_entry(m, set, mnt_list)
-		m->mnt.mnt_flags &= ~MNT_UMOUNT_CANDIDATE;
+		m->mnt_t_flags &= ~T_UMOUNT_CANDIDATE;
 }
 
 /*
@@ -519,7 +517,7 @@ static void trim_ancestors(struct mount *m)
 			return;
 		SET_MNT_MARK(m);
 		if (m != p->overmount)
-			p->mnt.mnt_flags &= ~MNT_UMOUNT_CANDIDATE;
+			p->mnt_t_flags &= ~T_UMOUNT_CANDIDATE;
 	}
 }
 
diff --git a/fs/pnode.h b/fs/pnode.h
index 04f1ac53aa49..507e30e7a420 100644
--- a/fs/pnode.h
+++ b/fs/pnode.h
@@ -10,14 +10,14 @@
 #include <linux/list.h>
 #include "mount.h"
 
-#define IS_MNT_SHARED(m) ((m)->mnt.mnt_flags & MNT_SHARED)
+#define IS_MNT_SHARED(m) ((m)->mnt_t_flags & T_SHARED)
 #define IS_MNT_SLAVE(m) ((m)->mnt_master)
 #define IS_MNT_NEW(m) (!(m)->mnt_ns)
-#define CLEAR_MNT_SHARED(m) ((m)->mnt.mnt_flags &= ~MNT_SHARED)
-#define IS_MNT_UNBINDABLE(m) ((m)->mnt.mnt_flags & MNT_UNBINDABLE)
-#define IS_MNT_MARKED(m) ((m)->mnt.mnt_flags & MNT_MARKED)
-#define SET_MNT_MARK(m) ((m)->mnt.mnt_flags |= MNT_MARKED)
-#define CLEAR_MNT_MARK(m) ((m)->mnt.mnt_flags &= ~MNT_MARKED)
+#define CLEAR_MNT_SHARED(m) ((m)->mnt_t_flags &= ~T_SHARED)
+#define IS_MNT_UNBINDABLE(m) ((m)->mnt_t_flags & T_UNBINDABLE)
+#define IS_MNT_MARKED(m) ((m)->mnt_t_flags & T_MARKED)
+#define SET_MNT_MARK(m) ((m)->mnt_t_flags |= T_MARKED)
+#define CLEAR_MNT_MARK(m) ((m)->mnt_t_flags &= ~T_MARKED)
 #define IS_MNT_LOCKED(m) ((m)->mnt.mnt_flags & MNT_LOCKED)
 
 #define CL_EXPIRE    		0x01
@@ -28,10 +28,13 @@
 #define CL_SHARED_TO_SLAVE	0x20
 #define CL_COPY_MNT_NS_FILE	0x40
 
+/*
+ * EXCL[namespace_sem]
+ */
 static inline void set_mnt_shared(struct mount *mnt)
 {
-	mnt->mnt.mnt_flags &= ~MNT_SHARED_MASK;
-	mnt->mnt.mnt_flags |= MNT_SHARED;
+	mnt->mnt_t_flags &= ~T_SHARED_MASK;
+	mnt->mnt_t_flags |= T_SHARED;
 }
 
 static inline bool peers(const struct mount *m1, const struct mount *m2)
diff --git a/include/linux/mount.h b/include/linux/mount.h
index 65fa8442c00a..5f9c053b0897 100644
--- a/include/linux/mount.h
+++ b/include/linux/mount.h
@@ -35,12 +35,8 @@ enum mount_flags {
 	MNT_SHRINKABLE	= 0x100,
 	MNT_WRITE_HOLD	= 0x200,
 
-	MNT_SHARED	= 0x1000, /* if the vfsmount is a shared mount */
-	MNT_UNBINDABLE	= 0x2000, /* if the vfsmount is a unbindable mount */
-
 	MNT_INTERNAL	= 0x4000,
 
-	MNT_UMOUNT_CANDIDATE	= 0x020000,
 	MNT_LOCK_ATIME		= 0x040000,
 	MNT_LOCK_NOEXEC		= 0x080000,
 	MNT_LOCK_NOSUID		= 0x100000,
@@ -49,25 +45,15 @@ enum mount_flags {
 	MNT_LOCKED		= 0x800000,
 	MNT_DOOMED		= 0x1000000,
 	MNT_SYNC_UMOUNT		= 0x2000000,
-	MNT_MARKED		= 0x4000000,
 	MNT_UMOUNT		= 0x8000000,
 
-	/*
-	 * MNT_SHARED_MASK is the set of flags that should be cleared when a
-	 * mount becomes shared.  Currently, this is only the flag that says a
-	 * mount cannot be bind mounted, since this is how we create a mount
-	 * that shares events with another mount.  If you add a new MNT_*
-	 * flag, consider how it interacts with shared mounts.
-	 */
-	MNT_SHARED_MASK	= MNT_UNBINDABLE,
 	MNT_USER_SETTABLE_MASK  = MNT_NOSUID | MNT_NODEV | MNT_NOEXEC
 				  | MNT_NOATIME | MNT_NODIRATIME | MNT_RELATIME
 				  | MNT_READONLY | MNT_NOSYMFOLLOW,
 	MNT_ATIME_MASK = MNT_NOATIME | MNT_NODIRATIME | MNT_RELATIME,
 
-	MNT_INTERNAL_FLAGS = MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL |
-			     MNT_DOOMED | MNT_SYNC_UMOUNT | MNT_MARKED |
-			     MNT_LOCKED | MNT_UMOUNT_CANDIDATE,
+	MNT_INTERNAL_FLAGS = MNT_WRITE_HOLD | MNT_INTERNAL | MNT_DOOMED |
+			     MNT_SYNC_UMOUNT | MNT_LOCKED
 };
 
 struct vfsmount {
-- 
2.39.5

[PATCH v3 30/48] propagate_one(): get rid of dest_master

[Al Viro]

propagate_mnt() takes the subtree we are about to attach and creates
its copies, setting the propagation between those.  Each copy is cloned
either from the original or from one of the already created copies.
The tricky part is choosing the right copy to serve as a master when we
are starting a new peer group.

The algorithm for doing that selection puts temporary marks on the masters
of mountpoints that already got a copy created for them; since the initial
peer group might have no master at all, we need to special-case that when
looking for the mark.  Currently we do that by memorizing the master of
original peer group.  It works, but we get yet another piece of data to
pass from propagate_mnt() to propagate_one().

Alternative is to mark the master of original peer group if not NULL,
turning the check into "master is NULL or marked".  Less data to pass
around and memory safety is more obvious that way...

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/pnode.c | 13 ++++++++-----
 1 file changed, 8 insertions(+), 5 deletions(-)

diff --git a/fs/pnode.c b/fs/pnode.c
index b997663de6d0..870ebced10aa 100644
--- a/fs/pnode.c
+++ b/fs/pnode.c
@@ -215,7 +215,7 @@ static struct mount *next_group(struct mount *m, struct mount *origin)
 }
 
 /* all accesses are serialized by namespace_sem */
-static struct mount *last_dest, *first_source, *last_source, *dest_master;
+static struct mount *last_dest, *first_source, *last_source;
 static struct hlist_head *list;
 
 static int propagate_one(struct mount *m, struct mountpoint *dest_mp)
@@ -239,7 +239,7 @@ static int propagate_one(struct mount *m, struct mountpoint *dest_mp)
 		bool done;
 		for (n = m; ; n = p) {
 			p = n->mnt_master;
-			if (p == dest_master || IS_MNT_MARKED(p))
+			if (!p || IS_MNT_MARKED(p))
 				break;
 		}
 		do {
@@ -264,7 +264,7 @@ static int propagate_one(struct mount *m, struct mountpoint *dest_mp)
 	read_seqlock_excl(&mount_lock);
 	mnt_set_mountpoint(m, dest_mp, child);
 	read_sequnlock_excl(&mount_lock);
-	if (m->mnt_master != dest_master)
+	if (m->mnt_master)
 		SET_MNT_MARK(m->mnt_master);
 	last_dest = m;
 	last_source = child;
@@ -300,7 +300,8 @@ int propagate_mnt(struct mount *dest_mnt, struct mountpoint *dest_mp,
 	first_source = source_mnt;
 	last_source = source_mnt;
 	list = tree_list;
-	dest_master = dest_mnt->mnt_master;
+	if (dest_mnt->mnt_master)
+		SET_MNT_MARK(dest_mnt->mnt_master);
 
 	/* all peers of dest_mnt, except dest_mnt itself */
 	for (n = next_peer(dest_mnt); n != dest_mnt; n = next_peer(n)) {
@@ -324,9 +325,11 @@ int propagate_mnt(struct mount *dest_mnt, struct mountpoint *dest_mp,
 out:
 	hlist_for_each_entry(n, tree_list, mnt_hash) {
 		m = n->mnt_parent;
-		if (m->mnt_master != dest_mnt->mnt_master)
+		if (m->mnt_master)
 			CLEAR_MNT_MARK(m->mnt_master);
 	}
+	if (dest_mnt->mnt_master)
+		CLEAR_MNT_MARK(dest_mnt->mnt_master);
 	return ret;
 }
 
-- 
2.39.5

[PATCH v3 31/48] propagate_mnt(): handle all peer groups in the same loop

[Al Viro]

the only difference is that for the original group we want to skip
the first element; not worth having the logics twice...

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/pnode.c | 33 +++++++++++++++------------------
 1 file changed, 15 insertions(+), 18 deletions(-)

diff --git a/fs/pnode.c b/fs/pnode.c
index 870ebced10aa..f55295e26217 100644
--- a/fs/pnode.c
+++ b/fs/pnode.c
@@ -289,7 +289,7 @@ int propagate_mnt(struct mount *dest_mnt, struct mountpoint *dest_mp,
 		    struct mount *source_mnt, struct hlist_head *tree_list)
 {
 	struct mount *m, *n;
-	int ret = 0;
+	int err = 0;
 
 	/*
 	 * we don't want to bother passing tons of arguments to
@@ -303,26 +303,23 @@ int propagate_mnt(struct mount *dest_mnt, struct mountpoint *dest_mp,
 	if (dest_mnt->mnt_master)
 		SET_MNT_MARK(dest_mnt->mnt_master);
 
-	/* all peers of dest_mnt, except dest_mnt itself */
-	for (n = next_peer(dest_mnt); n != dest_mnt; n = next_peer(n)) {
-		ret = propagate_one(n, dest_mp);
-		if (ret)
-			goto out;
-	}
-
-	/* all slave groups */
-	for (m = next_group(dest_mnt, dest_mnt); m;
-			m = next_group(m, dest_mnt)) {
-		/* everything in that slave group */
-		n = m;
+	/* iterate over peer groups, depth first */
+	for (m = dest_mnt; m && !err; m = next_group(m, dest_mnt)) {
+		if (m == dest_mnt) { // have one for dest_mnt itself
+			n = next_peer(m);
+			if (n == m)
+				continue;
+		} else {
+			n = m;
+		}
 		do {
-			ret = propagate_one(n, dest_mp);
-			if (ret)
-				goto out;
+			err = propagate_one(n, dest_mp);
+			if (err)
+				break;
 			n = next_peer(n);
 		} while (n != m);
 	}
-out:
+
 	hlist_for_each_entry(n, tree_list, mnt_hash) {
 		m = n->mnt_parent;
 		if (m->mnt_master)
@@ -330,7 +327,7 @@ int propagate_mnt(struct mount *dest_mnt, struct mountpoint *dest_mp,
 	}
 	if (dest_mnt->mnt_master)
 		CLEAR_MNT_MARK(dest_mnt->mnt_master);
-	return ret;
+	return err;
 }
 
 /*
-- 
2.39.5

[PATCH v3 32/48] propagate_one(): separate the "do we need secondary here?" logics

[Al Viro]

take the checks into separate helper - need_secondary(mount, mountpoint).

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/pnode.c | 22 ++++++++++++++--------
 1 file changed, 14 insertions(+), 8 deletions(-)

diff --git a/fs/pnode.c b/fs/pnode.c
index f55295e26217..7c832f98595c 100644
--- a/fs/pnode.c
+++ b/fs/pnode.c
@@ -218,19 +218,24 @@ static struct mount *next_group(struct mount *m, struct mount *origin)
 static struct mount *last_dest, *first_source, *last_source;
 static struct hlist_head *list;
 
-static int propagate_one(struct mount *m, struct mountpoint *dest_mp)
+static bool need_secondary(struct mount *m, struct mountpoint *dest_mp)
 {
-	struct mount *child;
-	int type;
 	/* skip ones added by this propagate_mnt() */
 	if (IS_MNT_NEW(m))
-		return 0;
+		return false;
 	/* skip if mountpoint isn't visible in m */
 	if (!is_subdir(dest_mp->m_dentry, m->mnt.mnt_root))
-		return 0;
+		return false;
 	/* skip if m is in the anon_ns */
 	if (is_anon_ns(m->mnt_ns))
-		return 0;
+		return false;
+	return true;
+}
+
+static int propagate_one(struct mount *m, struct mountpoint *dest_mp)
+{
+	struct mount *child;
+	int type;
 
 	if (peers(m, last_dest)) {
 		type = CL_MAKE_SHARED;
@@ -313,11 +318,12 @@ int propagate_mnt(struct mount *dest_mnt, struct mountpoint *dest_mp,
 			n = m;
 		}
 		do {
+			if (!need_secondary(n, dest_mp))
+				continue;
 			err = propagate_one(n, dest_mp);
 			if (err)
 				break;
-			n = next_peer(n);
-		} while (n != m);
+		} while ((n = next_peer(n)) != m);
 	}
 
 	hlist_for_each_entry(n, tree_list, mnt_hash) {
-- 
2.39.5

[PATCH v3 33/48] propagate_one(): separate the "what should be the master for this copy" part

[Al Viro]

When we create the first copy for a peer group, it becomes a slave of
one of the existing copies; take that logics into a separate helper -
find_master(parent, last_copy, original).

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/pnode.c | 43 ++++++++++++++++++++++++++-----------------
 1 file changed, 26 insertions(+), 17 deletions(-)

diff --git a/fs/pnode.c b/fs/pnode.c
index 7c832f98595c..94de8aad4da5 100644
--- a/fs/pnode.c
+++ b/fs/pnode.c
@@ -232,6 +232,31 @@ static bool need_secondary(struct mount *m, struct mountpoint *dest_mp)
 	return true;
 }
 
+static struct mount *find_master(struct mount *m,
+				struct mount *last_copy,
+				struct mount *original)
+{
+	struct mount *p;
+
+	// ascend until there's a copy for something with the same master
+	for (;;) {
+		p = m->mnt_master;
+		if (!p || IS_MNT_MARKED(p))
+			break;
+		m = p;
+	}
+	while (!peers(last_copy, original)) {
+		struct mount *parent = last_copy->mnt_parent;
+		if (parent->mnt_master == p) {
+			if (!peers(parent, m))
+				last_copy = last_copy->mnt_master;
+			break;
+		}
+		last_copy = last_copy->mnt_master;
+	}
+	return last_copy;
+}
+
 static int propagate_one(struct mount *m, struct mountpoint *dest_mp)
 {
 	struct mount *child;
@@ -240,23 +265,7 @@ static int propagate_one(struct mount *m, struct mountpoint *dest_mp)
 	if (peers(m, last_dest)) {
 		type = CL_MAKE_SHARED;
 	} else {
-		struct mount *n, *p;
-		bool done;
-		for (n = m; ; n = p) {
-			p = n->mnt_master;
-			if (!p || IS_MNT_MARKED(p))
-				break;
-		}
-		do {
-			struct mount *parent = last_source->mnt_parent;
-			if (peers(last_source, first_source))
-				break;
-			done = parent->mnt_master == p;
-			if (done && peers(n, parent))
-				break;
-			last_source = last_source->mnt_master;
-		} while (!done);
-
+		last_source = find_master(m, last_source, first_source);
 		type = CL_SLAVE;
 		/* beginning of peer group among the slaves? */
 		if (IS_MNT_SHARED(m))
-- 
2.39.5

[PATCH v3 34/48] propagate_one(): fold into the sole caller

[Al Viro]

mechanical expansion; will be cleaned up on the next step

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/pnode.c | 57 ++++++++++++++++++++++++------------------------------
 1 file changed, 25 insertions(+), 32 deletions(-)

diff --git a/fs/pnode.c b/fs/pnode.c
index 94de8aad4da5..aeaec24f7456 100644
--- a/fs/pnode.c
+++ b/fs/pnode.c
@@ -257,35 +257,6 @@ static struct mount *find_master(struct mount *m,
 	return last_copy;
 }
 
-static int propagate_one(struct mount *m, struct mountpoint *dest_mp)
-{
-	struct mount *child;
-	int type;
-
-	if (peers(m, last_dest)) {
-		type = CL_MAKE_SHARED;
-	} else {
-		last_source = find_master(m, last_source, first_source);
-		type = CL_SLAVE;
-		/* beginning of peer group among the slaves? */
-		if (IS_MNT_SHARED(m))
-			type |= CL_MAKE_SHARED;
-	}
-		
-	child = copy_tree(last_source, last_source->mnt.mnt_root, type);
-	if (IS_ERR(child))
-		return PTR_ERR(child);
-	read_seqlock_excl(&mount_lock);
-	mnt_set_mountpoint(m, dest_mp, child);
-	read_sequnlock_excl(&mount_lock);
-	if (m->mnt_master)
-		SET_MNT_MARK(m->mnt_master);
-	last_dest = m;
-	last_source = child;
-	hlist_add_head(&child->mnt_hash, list);
-	return count_mounts(m->mnt_ns, child);
-}
-
 /*
  * mount 'source_mnt' under the destination 'dest_mnt' at
  * dentry 'dest_dentry'. And propagate that mount to
@@ -302,8 +273,8 @@ static int propagate_one(struct mount *m, struct mountpoint *dest_mp)
 int propagate_mnt(struct mount *dest_mnt, struct mountpoint *dest_mp,
 		    struct mount *source_mnt, struct hlist_head *tree_list)
 {
-	struct mount *m, *n;
-	int err = 0;
+	struct mount *m, *n, *child;
+	int err = 0, type;
 
 	/*
 	 * we don't want to bother passing tons of arguments to
@@ -329,7 +300,29 @@ int propagate_mnt(struct mount *dest_mnt, struct mountpoint *dest_mp,
 		do {
 			if (!need_secondary(n, dest_mp))
 				continue;
-			err = propagate_one(n, dest_mp);
+			if (peers(n, last_dest)) {
+				type = CL_MAKE_SHARED;
+			} else {
+				last_source = find_master(n, last_source, first_source);
+				type = CL_SLAVE;
+				/* beginning of peer group among the slaves? */
+				if (IS_MNT_SHARED(n))
+					type |= CL_MAKE_SHARED;
+			}
+			child = copy_tree(last_source, last_source->mnt.mnt_root, type);
+			if (IS_ERR(child)) {
+				err = PTR_ERR(child);
+				break;
+			}
+			read_seqlock_excl(&mount_lock);
+			mnt_set_mountpoint(n, dest_mp, child);
+			read_sequnlock_excl(&mount_lock);
+			if (n->mnt_master)
+				SET_MNT_MARK(n->mnt_master);
+			last_dest = n;
+			last_source = child;
+			hlist_add_head(&child->mnt_hash, list);
+			err = count_mounts(n->mnt_ns, child);
 			if (err)
 				break;
 		} while ((n = next_peer(n)) != m);
-- 
2.39.5

[PATCH v3 35/48] fs/pnode.c: get rid of globals

[Al Viro]

this stuff can be local in propagate_mnt() now (and in some cases
duplicates the existing variables there)

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/pnode.c | 31 ++++++++++---------------------
 1 file changed, 10 insertions(+), 21 deletions(-)

diff --git a/fs/pnode.c b/fs/pnode.c
index aeaec24f7456..e01f43820a93 100644
--- a/fs/pnode.c
+++ b/fs/pnode.c
@@ -214,10 +214,6 @@ static struct mount *next_group(struct mount *m, struct mount *origin)
 	}
 }
 
-/* all accesses are serialized by namespace_sem */
-static struct mount *last_dest, *first_source, *last_source;
-static struct hlist_head *list;
-
 static bool need_secondary(struct mount *m, struct mountpoint *dest_mp)
 {
 	/* skip ones added by this propagate_mnt() */
@@ -273,18 +269,11 @@ static struct mount *find_master(struct mount *m,
 int propagate_mnt(struct mount *dest_mnt, struct mountpoint *dest_mp,
 		    struct mount *source_mnt, struct hlist_head *tree_list)
 {
-	struct mount *m, *n, *child;
+	struct mount *m, *n, *copy, *this, *last_dest;
 	int err = 0, type;
 
-	/*
-	 * we don't want to bother passing tons of arguments to
-	 * propagate_one(); everything is serialized by namespace_sem,
-	 * so globals will do just fine.
-	 */
 	last_dest = dest_mnt;
-	first_source = source_mnt;
-	last_source = source_mnt;
-	list = tree_list;
+	copy = source_mnt;
 	if (dest_mnt->mnt_master)
 		SET_MNT_MARK(dest_mnt->mnt_master);
 
@@ -303,26 +292,26 @@ int propagate_mnt(struct mount *dest_mnt, struct mountpoint *dest_mp,
 			if (peers(n, last_dest)) {
 				type = CL_MAKE_SHARED;
 			} else {
-				last_source = find_master(n, last_source, first_source);
+				copy = find_master(n, copy, source_mnt);
 				type = CL_SLAVE;
 				/* beginning of peer group among the slaves? */
 				if (IS_MNT_SHARED(n))
 					type |= CL_MAKE_SHARED;
 			}
-			child = copy_tree(last_source, last_source->mnt.mnt_root, type);
-			if (IS_ERR(child)) {
-				err = PTR_ERR(child);
+			this = copy_tree(copy, copy->mnt.mnt_root, type);
+			if (IS_ERR(this)) {
+				err = PTR_ERR(this);
 				break;
 			}
 			read_seqlock_excl(&mount_lock);
-			mnt_set_mountpoint(n, dest_mp, child);
+			mnt_set_mountpoint(n, dest_mp, this);
 			read_sequnlock_excl(&mount_lock);
 			if (n->mnt_master)
 				SET_MNT_MARK(n->mnt_master);
 			last_dest = n;
-			last_source = child;
-			hlist_add_head(&child->mnt_hash, list);
-			err = count_mounts(n->mnt_ns, child);
+			copy = this;
+			hlist_add_head(&this->mnt_hash, tree_list);
+			err = count_mounts(n->mnt_ns, this);
 			if (err)
 				break;
 		} while ((n = next_peer(n)) != m);
-- 
2.39.5

[PATCH v3 36/48] propagate_mnt(): get rid of last_dest

[Al Viro]

Its only use is choosing the type of copy - CL_MAKE_SHARED if there
already is a copy in that peer group, CL_SLAVE or CL_SLAVE | CL_MAKE_SHARED
otherwise.

But that's easy to keep track of - just set type in the beginning of group
and reset to CL_MAKE_SHARED after the first created secondary in it...

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/pnode.c | 21 +++++++++------------
 1 file changed, 9 insertions(+), 12 deletions(-)

diff --git a/fs/pnode.c b/fs/pnode.c
index e01f43820a93..b3af55123a82 100644
--- a/fs/pnode.c
+++ b/fs/pnode.c
@@ -269,35 +269,32 @@ static struct mount *find_master(struct mount *m,
 int propagate_mnt(struct mount *dest_mnt, struct mountpoint *dest_mp,
 		    struct mount *source_mnt, struct hlist_head *tree_list)
 {
-	struct mount *m, *n, *copy, *this, *last_dest;
+	struct mount *m, *n, *copy, *this;
 	int err = 0, type;
 
-	last_dest = dest_mnt;
-	copy = source_mnt;
 	if (dest_mnt->mnt_master)
 		SET_MNT_MARK(dest_mnt->mnt_master);
 
 	/* iterate over peer groups, depth first */
 	for (m = dest_mnt; m && !err; m = next_group(m, dest_mnt)) {
 		if (m == dest_mnt) { // have one for dest_mnt itself
+			copy = source_mnt;
+			type = CL_MAKE_SHARED;
 			n = next_peer(m);
 			if (n == m)
 				continue;
 		} else {
+			type = CL_SLAVE;
+			/* beginning of peer group among the slaves? */
+			if (IS_MNT_SHARED(m))
+				type |= CL_MAKE_SHARED;
 			n = m;
 		}
 		do {
 			if (!need_secondary(n, dest_mp))
 				continue;
-			if (peers(n, last_dest)) {
-				type = CL_MAKE_SHARED;
-			} else {
+			if (type & CL_SLAVE) // first in this peer group
 				copy = find_master(n, copy, source_mnt);
-				type = CL_SLAVE;
-				/* beginning of peer group among the slaves? */
-				if (IS_MNT_SHARED(n))
-					type |= CL_MAKE_SHARED;
-			}
 			this = copy_tree(copy, copy->mnt.mnt_root, type);
 			if (IS_ERR(this)) {
 				err = PTR_ERR(this);
@@ -308,12 +305,12 @@ int propagate_mnt(struct mount *dest_mnt, struct mountpoint *dest_mp,
 			read_sequnlock_excl(&mount_lock);
 			if (n->mnt_master)
 				SET_MNT_MARK(n->mnt_master);
-			last_dest = n;
 			copy = this;
 			hlist_add_head(&this->mnt_hash, tree_list);
 			err = count_mounts(n->mnt_ns, this);
 			if (err)
 				break;
+			type = CL_MAKE_SHARED;
 		} while ((n = next_peer(n)) != m);
 	}
 
-- 
2.39.5

[PATCH v3 37/48] propagate_mnt(): fix comment and convert to kernel-doc, while we are at it

[Al Viro]

Mountpoint is passed as struct mountpoint *, not struct dentry *
(and called dest_mp, not dest_dentry) since 2013.

Roots of created copies are linked via mnt_hash, not mnt_list since
a bit before the merge into mainline back in 2005.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/pnode.c | 23 +++++++++++------------
 1 file changed, 11 insertions(+), 12 deletions(-)

diff --git a/fs/pnode.c b/fs/pnode.c
index b3af55123a82..b887116f0041 100644
--- a/fs/pnode.c
+++ b/fs/pnode.c
@@ -253,21 +253,20 @@ static struct mount *find_master(struct mount *m,
 	return last_copy;
 }
 
-/*
- * mount 'source_mnt' under the destination 'dest_mnt' at
- * dentry 'dest_dentry'. And propagate that mount to
- * all the peer and slave mounts of 'dest_mnt'.
- * Link all the new mounts into a propagation tree headed at
- * source_mnt. Also link all the new mounts using ->mnt_list
- * headed at source_mnt's ->mnt_list
+/**
+ * propagate_mnt() - create secondary copies for tree attachment
+ * @dest_mnt:    destination mount.
+ * @dest_mp:     destination mountpoint.
+ * @source_mnt:  source mount.
+ * @tree_list:   list of secondaries to be attached.
  *
- * @dest_mnt: destination mount.
- * @dest_dentry: destination dentry.
- * @source_mnt: source mount.
- * @tree_list : list of heads of trees to be attached.
+ * Create secondary copies for attaching a tree with root @source_mnt
+ * at mount @dest_mnt with mountpoint @dest_mp.  Link all new mounts
+ * into a propagation graph.  Set mountpoints for all secondaries,
+ * link their roots into @tree_list via ->mnt_hash.
  */
 int propagate_mnt(struct mount *dest_mnt, struct mountpoint *dest_mp,
-		    struct mount *source_mnt, struct hlist_head *tree_list)
+		  struct mount *source_mnt, struct hlist_head *tree_list)
 {
 	struct mount *m, *n, *copy, *this;
 	int err = 0, type;
-- 
2.39.5

[PATCH v3 38/48] change_mnt_propagation() cleanups, step 1

[Al Viro]

Lift changing ->mnt_slave from do_make_slave() into the caller.
Simplifies the next steps...

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/pnode.c | 9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)

diff --git a/fs/pnode.c b/fs/pnode.c
index b887116f0041..14618eac2025 100644
--- a/fs/pnode.c
+++ b/fs/pnode.c
@@ -104,7 +104,6 @@ static int do_make_slave(struct mount *mnt)
 	}
 	list_for_each_entry(slave_mnt, &mnt->mnt_slave_list, mnt_slave)
 		slave_mnt->mnt_master = master;
-	list_move(&mnt->mnt_slave, &master->mnt_slave_list);
 	list_splice(&mnt->mnt_slave_list, master->mnt_slave_list.prev);
 	INIT_LIST_HEAD(&mnt->mnt_slave_list);
 	mnt->mnt_master = master;
@@ -121,8 +120,12 @@ void change_mnt_propagation(struct mount *mnt, int type)
 		return;
 	}
 	do_make_slave(mnt);
-	if (type != MS_SLAVE) {
-		list_del_init(&mnt->mnt_slave);
+	list_del_init(&mnt->mnt_slave);
+	if (type == MS_SLAVE) {
+		if (mnt->mnt_master)
+			list_add(&mnt->mnt_slave,
+				 &mnt->mnt_master->mnt_slave_list);
+	} else {
 		mnt->mnt_master = NULL;
 		if (type == MS_UNBINDABLE)
 			mnt->mnt_t_flags |= T_UNBINDABLE;
-- 
2.39.5

[PATCH v3 39/48] change_mnt_propagation(): do_make_slave() is a no-op unless IS_MNT_SHARED()

[Al Viro]

... since mnt->mnt_share and mnt->mnt_slave_list are guaranteed to be empty unless
IS_MNT_SHARED(mnt).

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/pnode.c | 9 ++++-----
 1 file changed, 4 insertions(+), 5 deletions(-)

diff --git a/fs/pnode.c b/fs/pnode.c
index 14618eac2025..9723f05cda5f 100644
--- a/fs/pnode.c
+++ b/fs/pnode.c
@@ -70,10 +70,8 @@ static int do_make_slave(struct mount *mnt)
 	struct mount *master, *slave_mnt;
 
 	if (list_empty(&mnt->mnt_share)) {
-		if (IS_MNT_SHARED(mnt)) {
-			mnt_release_group_id(mnt);
-			CLEAR_MNT_SHARED(mnt);
-		}
+		mnt_release_group_id(mnt);
+		CLEAR_MNT_SHARED(mnt);
 		master = mnt->mnt_master;
 		if (!master) {
 			struct list_head *p = &mnt->mnt_slave_list;
@@ -119,7 +117,8 @@ void change_mnt_propagation(struct mount *mnt, int type)
 		set_mnt_shared(mnt);
 		return;
 	}
-	do_make_slave(mnt);
+	if (IS_MNT_SHARED(mnt))
+		do_make_slave(mnt);
 	list_del_init(&mnt->mnt_slave);
 	if (type == MS_SLAVE) {
 		if (mnt->mnt_master)
-- 
2.39.5

[PATCH v3 40/48] do_make_slave(): choose new master sanely

[Al Viro]

When mount changes propagation type so that it doesn't propagate
events any more (MS_PRIVATE, MS_SLAVE, MS_UNBINDABLE), we need
to make sure that event propagation between other mounts is
unaffected.

We need to make sure that events from peers and master of that mount
(if any) still reach everything that used to be on its ->mnt_slave_list.

If mount has neither peers nor master, we simply need to dissolve
its ->mnt_slave_list and clear ->mnt_master of everything in there.

If mount has peers, we transfer everything in ->mnt_slave_list of
this mount into that of some of those peers (and adjust ->mnt_master
accordingly).

If mount has a master but no peers, we transfer everything in
->mnt_slave_list of this mount into that of its master (adjusting
->mnt_master, etc.).

There are two problems with the current implementation:
	* there's a long-obsolete logics in choosing the peer -
once upon a time it made sense to prefer the peer that had the
same ->mnt_root as our mount, but that had been pointless since
2014 ("smarter propagate_mnt()")
	* the most common caller of that thing is umount_tree()
taking the mounts out of propagation graph.  In that case it's
possible to have ->mnt_slave_list contents moved many times,
since the replacement master is likely to be taken out by the
same umount_tree(), etc.

Take the choice of replacement master into a separate function
(propagation_source()) and teach it to skip the candidates that
are going to be taken out.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/pnode.c | 62 +++++++++++++++++++++++++++---------------------------
 1 file changed, 31 insertions(+), 31 deletions(-)

diff --git a/fs/pnode.c b/fs/pnode.c
index 9723f05cda5f..91d10af867bd 100644
--- a/fs/pnode.c
+++ b/fs/pnode.c
@@ -65,40 +65,45 @@ int get_dominating_id(struct mount *mnt, const struct path *root)
 	return 0;
 }
 
+static inline bool will_be_unmounted(struct mount *m)
+{
+	return m->mnt.mnt_flags & MNT_UMOUNT;
+}
+
+static struct mount *propagation_source(struct mount *mnt)
+{
+	do {
+		struct mount *m;
+		for (m = next_peer(mnt); m != mnt; m = next_peer(m)) {
+			if (!will_be_unmounted(m))
+				return m;
+		}
+		mnt = mnt->mnt_master;
+	} while (mnt && will_be_unmounted(mnt));
+	return mnt;
+}
+
 static int do_make_slave(struct mount *mnt)
 {
-	struct mount *master, *slave_mnt;
+	struct mount *master = propagation_source(mnt);
+	struct mount *slave_mnt;
 
 	if (list_empty(&mnt->mnt_share)) {
 		mnt_release_group_id(mnt);
-		CLEAR_MNT_SHARED(mnt);
-		master = mnt->mnt_master;
-		if (!master) {
-			struct list_head *p = &mnt->mnt_slave_list;
-			while (!list_empty(p)) {
-				slave_mnt = list_first_entry(p,
-						struct mount, mnt_slave);
-				list_del_init(&slave_mnt->mnt_slave);
-				slave_mnt->mnt_master = NULL;
-			}
-			return 0;
-		}
 	} else {
-		struct mount *m;
-		/*
-		 * slave 'mnt' to a peer mount that has the
-		 * same root dentry. If none is available then
-		 * slave it to anything that is available.
-		 */
-		for (m = master = next_peer(mnt); m != mnt; m = next_peer(m)) {
-			if (m->mnt.mnt_root == mnt->mnt.mnt_root) {
-				master = m;
-				break;
-			}
-		}
 		list_del_init(&mnt->mnt_share);
 		mnt->mnt_group_id = 0;
-		CLEAR_MNT_SHARED(mnt);
+	}
+	CLEAR_MNT_SHARED(mnt);
+	if (!master) {
+		struct list_head *p = &mnt->mnt_slave_list;
+		while (!list_empty(p)) {
+			slave_mnt = list_first_entry(p,
+					struct mount, mnt_slave);
+			list_del_init(&slave_mnt->mnt_slave);
+			slave_mnt->mnt_master = NULL;
+		}
+		return 0;
 	}
 	list_for_each_entry(slave_mnt, &mnt->mnt_slave_list, mnt_slave)
 		slave_mnt->mnt_master = master;
@@ -443,11 +448,6 @@ static inline bool is_candidate(struct mount *m)
 	return m->mnt_t_flags & T_UMOUNT_CANDIDATE;
 }
 
-static inline bool will_be_unmounted(struct mount *m)
-{
-	return m->mnt.mnt_flags & MNT_UMOUNT;
-}
-
 static void umount_one(struct mount *m, struct list_head *to_umount)
 {
 	m->mnt.mnt_flags |= MNT_UMOUNT;
-- 
2.39.5

[PATCH v3 41/48] turn do_make_slave() into transfer_propagation()

[Al Viro]

Lift calculation of replacement propagation source, removal from
peer group and assignment of ->mnt_master from do_make_slave() into
change_mnt_propagation() itself.  What remains is switching of
what used to get propagation *through* mnt to alternative source.
Rename to transfer_propagation(), passing it the replacement source
as the second argument.  Have it return void, while we are at it.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/pnode.c | 36 ++++++++++++++++++------------------
 1 file changed, 18 insertions(+), 18 deletions(-)

diff --git a/fs/pnode.c b/fs/pnode.c
index 91d10af867bd..0a54848cbbd1 100644
--- a/fs/pnode.c
+++ b/fs/pnode.c
@@ -83,19 +83,10 @@ static struct mount *propagation_source(struct mount *mnt)
 	return mnt;
 }
 
-static int do_make_slave(struct mount *mnt)
+static void transfer_propagation(struct mount *mnt, struct mount *to)
 {
-	struct mount *master = propagation_source(mnt);
 	struct mount *slave_mnt;
-
-	if (list_empty(&mnt->mnt_share)) {
-		mnt_release_group_id(mnt);
-	} else {
-		list_del_init(&mnt->mnt_share);
-		mnt->mnt_group_id = 0;
-	}
-	CLEAR_MNT_SHARED(mnt);
-	if (!master) {
+	if (!to) {
 		struct list_head *p = &mnt->mnt_slave_list;
 		while (!list_empty(p)) {
 			slave_mnt = list_first_entry(p,
@@ -103,14 +94,12 @@ static int do_make_slave(struct mount *mnt)
 			list_del_init(&slave_mnt->mnt_slave);
 			slave_mnt->mnt_master = NULL;
 		}
-		return 0;
+		return;
 	}
 	list_for_each_entry(slave_mnt, &mnt->mnt_slave_list, mnt_slave)
-		slave_mnt->mnt_master = master;
-	list_splice(&mnt->mnt_slave_list, master->mnt_slave_list.prev);
+		slave_mnt->mnt_master = to;
+	list_splice(&mnt->mnt_slave_list, to->mnt_slave_list.prev);
 	INIT_LIST_HEAD(&mnt->mnt_slave_list);
-	mnt->mnt_master = master;
-	return 0;
 }
 
 /*
@@ -122,8 +111,19 @@ void change_mnt_propagation(struct mount *mnt, int type)
 		set_mnt_shared(mnt);
 		return;
 	}
-	if (IS_MNT_SHARED(mnt))
-		do_make_slave(mnt);
+	if (IS_MNT_SHARED(mnt)) {
+		struct mount *m = propagation_source(mnt);
+
+		if (list_empty(&mnt->mnt_share)) {
+			mnt_release_group_id(mnt);
+		} else {
+			list_del_init(&mnt->mnt_share);
+			mnt->mnt_group_id = 0;
+		}
+		CLEAR_MNT_SHARED(mnt);
+		transfer_propagation(mnt, m);
+		mnt->mnt_master = m;
+	}
 	list_del_init(&mnt->mnt_slave);
 	if (type == MS_SLAVE) {
 		if (mnt->mnt_master)
-- 
2.39.5

[PATCH v3 42/48] mnt_slave_list/mnt_slave: turn into hlist_head/hlist_node

[Al Viro]

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/mount.h     |  4 ++--
 fs/namespace.c | 14 ++++++--------
 fs/pnode.c     | 41 +++++++++++++++++++----------------------
 3 files changed, 27 insertions(+), 32 deletions(-)

diff --git a/fs/mount.h b/fs/mount.h
index f299dc85446d..08583428b10b 100644
--- a/fs/mount.h
+++ b/fs/mount.h
@@ -69,8 +69,8 @@ struct mount {
 	struct list_head mnt_list;
 	struct list_head mnt_expire;	/* link in fs-specific expiry list */
 	struct list_head mnt_share;	/* circular list of shared mounts */
-	struct list_head mnt_slave_list;/* list of slave mounts */
-	struct list_head mnt_slave;	/* slave list entry */
+	struct hlist_head mnt_slave_list;/* list of slave mounts */
+	struct hlist_node mnt_slave;	/* slave list entry */
 	struct mount *mnt_master;	/* slave is on master->mnt_slave_list */
 	struct mnt_namespace *mnt_ns;	/* containing namespace */
 	struct mountpoint *mnt_mp;	/* where is it mounted */
diff --git a/fs/namespace.c b/fs/namespace.c
index da27365418a5..38a46b32413d 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -380,8 +380,8 @@ static struct mount *alloc_vfsmnt(const char *name)
 		INIT_LIST_HEAD(&mnt->mnt_list);
 		INIT_LIST_HEAD(&mnt->mnt_expire);
 		INIT_LIST_HEAD(&mnt->mnt_share);
-		INIT_LIST_HEAD(&mnt->mnt_slave_list);
-		INIT_LIST_HEAD(&mnt->mnt_slave);
+		INIT_HLIST_HEAD(&mnt->mnt_slave_list);
+		INIT_HLIST_NODE(&mnt->mnt_slave);
 		INIT_HLIST_NODE(&mnt->mnt_mp_list);
 		INIT_HLIST_HEAD(&mnt->mnt_stuck_children);
 		RB_CLEAR_NODE(&mnt->mnt_node);
@@ -1348,10 +1348,10 @@ static struct mount *clone_mnt(struct mount *old, struct dentry *root,
 
 	if ((flag & CL_SLAVE) ||
 	    ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
-		list_add(&mnt->mnt_slave, &old->mnt_slave_list);
+		hlist_add_head(&mnt->mnt_slave, &old->mnt_slave_list);
 		mnt->mnt_master = old;
 	} else if (IS_MNT_SLAVE(old)) {
-		list_add(&mnt->mnt_slave, &old->mnt_slave);
+		hlist_add_behind(&mnt->mnt_slave, &old->mnt_slave);
 		mnt->mnt_master = old->mnt_master;
 	}
 	return mnt;
@@ -3398,10 +3398,8 @@ static int do_set_group(struct path *from_path, struct path *to_path)
 		goto out;
 
 	if (IS_MNT_SLAVE(from)) {
-		struct mount *m = from->mnt_master;
-
-		list_add(&to->mnt_slave, &from->mnt_slave);
-		to->mnt_master = m;
+		hlist_add_behind(&to->mnt_slave, &from->mnt_slave);
+		to->mnt_master = from->mnt_master;
 	}
 
 	if (IS_MNT_SHARED(from)) {
diff --git a/fs/pnode.c b/fs/pnode.c
index 0a54848cbbd1..69278079faeb 100644
--- a/fs/pnode.c
+++ b/fs/pnode.c
@@ -21,12 +21,12 @@ static inline struct mount *next_peer(struct mount *p)
 
 static inline struct mount *first_slave(struct mount *p)
 {
-	return list_entry(p->mnt_slave_list.next, struct mount, mnt_slave);
+	return hlist_entry(p->mnt_slave_list.first, struct mount, mnt_slave);
 }
 
 static inline struct mount *next_slave(struct mount *p)
 {
-	return list_entry(p->mnt_slave.next, struct mount, mnt_slave);
+	return hlist_entry(p->mnt_slave.next, struct mount, mnt_slave);
 }
 
 static struct mount *get_peer_under_root(struct mount *mnt,
@@ -85,21 +85,18 @@ static struct mount *propagation_source(struct mount *mnt)
 
 static void transfer_propagation(struct mount *mnt, struct mount *to)
 {
-	struct mount *slave_mnt;
-	if (!to) {
-		struct list_head *p = &mnt->mnt_slave_list;
-		while (!list_empty(p)) {
-			slave_mnt = list_first_entry(p,
-					struct mount, mnt_slave);
-			list_del_init(&slave_mnt->mnt_slave);
-			slave_mnt->mnt_master = NULL;
-		}
-		return;
+	struct hlist_node *p = NULL, *n;
+	struct mount *m;
+
+	hlist_for_each_entry_safe(m, n, &mnt->mnt_slave_list, mnt_slave) {
+		m->mnt_master = to;
+		if (!to)
+			hlist_del_init(&m->mnt_slave);
+		else
+			p = &m->mnt_slave;
 	}
-	list_for_each_entry(slave_mnt, &mnt->mnt_slave_list, mnt_slave)
-		slave_mnt->mnt_master = to;
-	list_splice(&mnt->mnt_slave_list, to->mnt_slave_list.prev);
-	INIT_LIST_HEAD(&mnt->mnt_slave_list);
+	if (p)
+		hlist_splice_init(&mnt->mnt_slave_list, p, &to->mnt_slave_list);
 }
 
 /*
@@ -124,10 +121,10 @@ void change_mnt_propagation(struct mount *mnt, int type)
 		transfer_propagation(mnt, m);
 		mnt->mnt_master = m;
 	}
-	list_del_init(&mnt->mnt_slave);
+	hlist_del_init(&mnt->mnt_slave);
 	if (type == MS_SLAVE) {
 		if (mnt->mnt_master)
-			list_add(&mnt->mnt_slave,
+			hlist_add_head(&mnt->mnt_slave,
 				 &mnt->mnt_master->mnt_slave_list);
 	} else {
 		mnt->mnt_master = NULL;
@@ -147,7 +144,7 @@ static struct mount *__propagation_next(struct mount *m,
 		if (master == origin->mnt_master) {
 			struct mount *next = next_peer(m);
 			return (next == origin) ? NULL : next;
-		} else if (m->mnt_slave.next != &master->mnt_slave_list)
+		} else if (m->mnt_slave.next)
 			return next_slave(m);
 
 		/* back at master */
@@ -169,7 +166,7 @@ static struct mount *propagation_next(struct mount *m,
 					 struct mount *origin)
 {
 	/* are there any slaves of this mount? */
-	if (!IS_MNT_NEW(m) && !list_empty(&m->mnt_slave_list))
+	if (!IS_MNT_NEW(m) && !hlist_empty(&m->mnt_slave_list))
 		return first_slave(m);
 
 	return __propagation_next(m, origin);
@@ -194,7 +191,7 @@ static struct mount *next_group(struct mount *m, struct mount *origin)
 	while (1) {
 		while (1) {
 			struct mount *next;
-			if (!IS_MNT_NEW(m) && !list_empty(&m->mnt_slave_list))
+			if (!IS_MNT_NEW(m) && !hlist_empty(&m->mnt_slave_list))
 				return first_slave(m);
 			next = next_peer(m);
 			if (m->mnt_group_id == origin->mnt_group_id) {
@@ -207,7 +204,7 @@ static struct mount *next_group(struct mount *m, struct mount *origin)
 		/* m is the last peer */
 		while (1) {
 			struct mount *master = m->mnt_master;
-			if (m->mnt_slave.next != &master->mnt_slave_list)
+			if (m->mnt_slave.next)
 				return next_slave(m);
 			m = next_peer(master);
 			if (master->mnt_group_id == origin->mnt_group_id)
-- 
2.39.5

[PATCH v3 43/48] change_mnt_propagation(): move ->mnt_master assignment into MS_SLAVE case

[Al Viro]

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/pnode.c | 12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)

diff --git a/fs/pnode.c b/fs/pnode.c
index 69278079faeb..cbf5f5746252 100644
--- a/fs/pnode.c
+++ b/fs/pnode.c
@@ -104,13 +104,14 @@ static void transfer_propagation(struct mount *mnt, struct mount *to)
  */
 void change_mnt_propagation(struct mount *mnt, int type)
 {
+	struct mount *m = mnt->mnt_master;
+
 	if (type == MS_SHARED) {
 		set_mnt_shared(mnt);
 		return;
 	}
 	if (IS_MNT_SHARED(mnt)) {
-		struct mount *m = propagation_source(mnt);
-
+		m = propagation_source(mnt);
 		if (list_empty(&mnt->mnt_share)) {
 			mnt_release_group_id(mnt);
 		} else {
@@ -119,13 +120,12 @@ void change_mnt_propagation(struct mount *mnt, int type)
 		}
 		CLEAR_MNT_SHARED(mnt);
 		transfer_propagation(mnt, m);
-		mnt->mnt_master = m;
 	}
 	hlist_del_init(&mnt->mnt_slave);
 	if (type == MS_SLAVE) {
-		if (mnt->mnt_master)
-			hlist_add_head(&mnt->mnt_slave,
-				 &mnt->mnt_master->mnt_slave_list);
+		mnt->mnt_master = m;
+		if (m)
+			hlist_add_head(&mnt->mnt_slave, &m->mnt_slave_list);
 	} else {
 		mnt->mnt_master = NULL;
 		if (type == MS_UNBINDABLE)
-- 
2.39.5

[PATCH v3 44/48] copy_tree(): don't link the mounts via mnt_list

[Al Viro]

The only place that really needs to be adjusted is commit_tree() -
there we need to iterate through the copy and we might as well
use next_mnt() for that.  However, in case when our tree has been
slid under something already mounted (propagation to a mountpoint
that already has something mounted on it or a 'beneath' move_mount)
we need to take care not to walk into the overmounting tree.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/mount.h     |  3 +--
 fs/namespace.c | 60 ++++++++++++++++++++------------------------------
 fs/pnode.c     |  3 ++-
 3 files changed, 27 insertions(+), 39 deletions(-)

diff --git a/fs/mount.h b/fs/mount.h
index 08583428b10b..97737051a8b9 100644
--- a/fs/mount.h
+++ b/fs/mount.h
@@ -193,7 +193,7 @@ static inline bool mnt_ns_empty(const struct mnt_namespace *ns)
 	return RB_EMPTY_ROOT(&ns->mounts);
 }
 
-static inline void move_from_ns(struct mount *mnt, struct list_head *dt_list)
+static inline void move_from_ns(struct mount *mnt)
 {
 	struct mnt_namespace *ns = mnt->mnt_ns;
 	WARN_ON(!mnt_ns_attached(mnt));
@@ -203,7 +203,6 @@ static inline void move_from_ns(struct mount *mnt, struct list_head *dt_list)
 		ns->mnt_first_node = rb_next(&mnt->mnt_node);
 	rb_erase(&mnt->mnt_node, &ns->mounts);
 	RB_CLEAR_NODE(&mnt->mnt_node);
-	list_add_tail(&mnt->mnt_list, dt_list);
 }
 
 bool has_locked_children(struct mount *mnt, struct dentry *dentry);
diff --git a/fs/namespace.c b/fs/namespace.c
index 38a46b32413d..bd6c7da901fc 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -1161,34 +1161,6 @@ static void mnt_add_to_ns(struct mnt_namespace *ns, struct mount *mnt)
 	mnt_notify_add(mnt);
 }
 
-/*
- * vfsmount lock must be held for write
- */
-static void commit_tree(struct mount *mnt)
-{
-	struct mount *parent = mnt->mnt_parent;
-	struct mount *m;
-	LIST_HEAD(head);
-	struct mnt_namespace *n = parent->mnt_ns;
-
-	BUG_ON(parent == mnt);
-
-	if (!mnt_ns_attached(mnt)) {
-		list_add_tail(&head, &mnt->mnt_list);
-		while (!list_empty(&head)) {
-			m = list_first_entry(&head, typeof(*m), mnt_list);
-			list_del(&m->mnt_list);
-
-			mnt_add_to_ns(n, m);
-		}
-		n->nr_mounts += n->pending_mounts;
-		n->pending_mounts = 0;
-	}
-
-	make_visible(mnt);
-	touch_mnt_namespace(n);
-}
-
 static struct mount *next_mnt(struct mount *p, struct mount *root)
 {
 	struct list_head *next = p->mnt_mounts.next;
@@ -1215,6 +1187,27 @@ static struct mount *skip_mnt_tree(struct mount *p)
 	return p;
 }
 
+/*
+ * vfsmount lock must be held for write
+ */
+static void commit_tree(struct mount *mnt)
+{
+	struct mnt_namespace *n = mnt->mnt_parent->mnt_ns;
+
+	if (!mnt_ns_attached(mnt)) {
+		for (struct mount *m = mnt; m; m = next_mnt(m, mnt))
+			if (unlikely(mnt_ns_attached(m)))
+				m = skip_mnt_tree(m);
+			else
+				mnt_add_to_ns(n, m);
+		n->nr_mounts += n->pending_mounts;
+		n->pending_mounts = 0;
+	}
+
+	make_visible(mnt);
+	touch_mnt_namespace(n);
+}
+
 /**
  * vfs_create_mount - Create a mount for a configured superblock
  * @fc: The configuration context with the superblock attached
@@ -1831,9 +1824,8 @@ static void umount_tree(struct mount *mnt, enum umount_tree_flags how)
 	for (p = mnt; p; p = next_mnt(p, mnt)) {
 		p->mnt.mnt_flags |= MNT_UMOUNT;
 		if (mnt_ns_attached(p))
-			move_from_ns(p, &tmp_list);
-		else
-			list_move(&p->mnt_list, &tmp_list);
+			move_from_ns(p);
+		list_add_tail(&p->mnt_list, &tmp_list);
 	}
 
 	/* Hide the mounts from mnt_mounts */
@@ -2270,7 +2262,6 @@ struct mount *copy_tree(struct mount *src_root, struct dentry *dentry,
 					list_add(&dst_mnt->mnt_expire,
 						 &src_mnt->mnt_expire);
 			}
-			list_add_tail(&dst_mnt->mnt_list, &res->mnt_list);
 			attach_mnt(dst_mnt, dst_parent, src_parent->mnt_mp);
 			unlock_mount_hash();
 		}
@@ -2686,12 +2677,9 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 		list_del_init(&source_mnt->mnt_expire);
 	} else {
 		if (source_mnt->mnt_ns) {
-			LIST_HEAD(head);
-
 			/* move from anon - the caller will destroy */
 			for (p = source_mnt; p; p = next_mnt(p, source_mnt))
-				move_from_ns(p, &head);
-			list_del_init(&head);
+				move_from_ns(p);
 		}
 	}
 
diff --git a/fs/pnode.c b/fs/pnode.c
index cbf5f5746252..81f7599bdac4 100644
--- a/fs/pnode.c
+++ b/fs/pnode.c
@@ -449,7 +449,8 @@ static void umount_one(struct mount *m, struct list_head *to_umount)
 {
 	m->mnt.mnt_flags |= MNT_UMOUNT;
 	list_del_init(&m->mnt_child);
-	move_from_ns(m, to_umount);
+	move_from_ns(m);
+	list_add_tail(&m->mnt_list, to_umount);
 }
 
 static void remove_from_candidate_list(struct mount *m)
-- 
2.39.5

Re: [PATCH v3 44/48] copy_tree(): don't link the mounts via mnt_list

[Lai, Yi]

Hi Al Viro,

Greetings!

I used Syzkaller and found that there is BUG: soft lockup in attach_recursive_mnt in linux-next next-20250812.

After bisection and the first bad commit is:
"
663206854f02 copy_tree(): don't link the mounts via mnt_list
"

All detailed into can be found at:
https://github.com/laifryiee/syzkaller_logs/tree/main/250813_093835_attach_recursive_mnt
Syzkaller repro code:
https://github.com/laifryiee/syzkaller_logs/tree/main/250813_093835_attach_recursive_mnt/repro.c
Kconfig(make olddefconfig):
https://github.com/laifryiee/syzkaller_logs/tree/main/250813_093835_attach_recursive_mnt/kconfig_origin
Bisect info:
https://github.com/laifryiee/syzkaller_logs/tree/main/250813_093835_attach_recursive_mnt/bisect_info.log
bzImage:
https://github.com/laifryiee/syzkaller_logs/raw/refs/heads/main/250813_093835_attach_recursive_mnt/bzImage_next-20250812
Issue dmesg:
https://github.com/laifryiee/syzkaller_logs/blob/main/250813_093835_attach_recursive_mnt/next-20250812_dmesg.log

"
[   48.594242] watchdog: BUG: soft lockup - CPU#0 stuck for 21s! [repro:724]
[   48.594264] Modules linked in:
[   48.594270] irq event stamp: 442328
[   48.594273] hardirqs last  enabled at (442327): [<ffffffff85da4465>] _raw_spin_unlock_irqrestore+0x35/0x70
[   48.594297] hardirqs last disabled at (442328): [<ffffffff85d749a4>] sysvec_apic_timer_interrupt+0x14/0xd0
[   48.594311] softirqs last  enabled at (442262): [<ffffffff814814ee>] __irq_exit_rcu+0x10e/0x170
[   48.594336] softirqs last disabled at (442257): [<ffffffff814814ee>] __irq_exit_rcu+0x10e/0x170
[   48.594354] CPU: 0 UID: 0 PID: 724 Comm: repro Tainted: G        W           6.17.0-rc1-next-20250812-next-2025081 #1 P
[   48.594367] Tainted: [W]=WARN
[   48.594370] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.16.0-0-gd239552ce722-prebuilt.qemu.org 044
[   48.594376] RIP: 0010:attach_recursive_mnt+0xf9a/0x1990
[   48.594394] Code: c1 e8 03 80 3c 18 00 0f 85 07 06 00 00 49 8d 7f 10 4d 8b 6f 70 48 89 f8 48 c1 e8 03 80 3c 18 00 0f 855
[   48.594402] RSP: 0018:ff1100001282fbb0 EFLAGS: 00000246
[   48.594409] RAX: 1fe220000537eea2 RBX: dffffc0000000000 RCX: ffffffff820f5a86
[   48.594414] RDX: ff11000014570000 RSI: ffffffff820f5ad8 RDI: ff11000029bf7510
[   48.594419] RBP: ff1100001282fcd0 R08: 0000000000000001 R09: 0000000000000001
[   48.594424] R10: ff11000010902200 R11: ff11000014570e58 R12: ff11000029017c00
[   48.594429] R13: ff11000026b98220 R14: ff110000290b4070 R15: ff11000026b981c0
[   48.594435] FS:  00007f5edc748800(0000) GS:ff110000e3940000(0000) knlGS:0000000000000000
[   48.594441] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[   48.594445] CR2: 00007f5edc7491a8 CR3: 000000002110e001 CR4: 0000000000771ef0
[   48.594454] PKRU: 55555554
[   48.594456] Call Trace:
[   48.594460]  <TASK>
[   48.594474]  ? __pfx_attach_recursive_mnt+0x10/0x10
[   48.594488]  ? do_raw_spin_unlock+0x15c/0x210
[   48.594501]  ? _raw_spin_unlock+0x31/0x60
[   48.594511]  ? __sanitizer_cov_trace_const_cmp4+0x1a/0x20
[   48.594526]  ? clone_mnt+0x755/0xbd0
[   48.594541]  graft_tree+0x190/0x220
[   48.594552]  ? graft_tree+0x190/0x220
[   48.594566]  path_mount+0x1b0a/0x1f70
[   48.594578]  ? lockdep_hardirqs_on+0x89/0x110
[   48.594590]  ? trace_hardirqs_on+0x51/0x60
[   48.594607]  ? __pfx_path_mount+0x10/0x10
[   48.594618]  ? __kasan_slab_free+0x4f/0x60
[   48.594631]  ? kmem_cache_free+0x2ea/0x520
[   48.594643]  ? putname.part.0+0x132/0x180
[   48.594657]  ? putname.part.0+0x137/0x180
[   48.594669]  __x64_sys_mount+0x2a6/0x330
[   48.594680]  ? __x64_sys_mount+0x2a6/0x330
[   48.594694]  ? __pfx___x64_sys_mount+0x10/0x10
[   48.594713]  x64_sys_call+0x2127/0x2180
[   48.594722]  do_syscall_64+0x6d/0x2e0
[   48.594736]  entry_SYSCALL_64_after_hwframe+0x76/0x7e
[   48.594744] RIP: 0033:0x7f5edc43ee5d
[   48.594755] Code: ff c3 66 2e 0f 1f 84 00 00 00 00 00 90 f3 0f 1e fa 48 89 f8 48 89 f7 48 89 d6 48 89 ca 4d 89 c2 4d 898
[   48.594762] RSP: 002b:00007ffd1a883988 EFLAGS: 00000246 ORIG_RAX: 00000000000000a5
[   48.594768] RAX: ffffffffffffffda RBX: 0000000000000000 RCX: 00007f5edc43ee5d
[   48.594773] RDX: 0000000000000000 RSI: 0000000020000080 RDI: 0000000020000040
[   48.594778] RBP: 00007ffd1a883aa0 R08: 0000000000000000 R09: 0000000000000000
[   48.594782] R10: 0000000000001400 R11: 0000000000000246 R12: 00007ffd1a883bf8
[   48.594787] R13: 00000000004043b3 R14: 0000000000406e08 R15: 00007f5edc795000
[   48.594803]  </TASK>
[   48.594808] Kernel panic - not syncing: softlockup: hung tasks
[   48.615257] CPU: 0 UID: 0 PID: 724 Comm: repro Tainted: G        W    L      6.17.0-rc1-next-20250812-next-2025081 #1 P
[   48.616012] Tainted: [W]=WARN, [L]=SOFTLOCKUP
[   48.616285] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.16.0-0-gd239552ce722-prebuilt.qemu.org 044
[   48.616983] Call Trace:
[   48.617160]  <IRQ>
[   48.617299]  dump_stack_lvl+0x42/0x150
[   48.617544]  dump_stack+0x19/0x20
[   48.617762]  vpanic+0x6dd/0x770
[   48.617987]  ? __pfx_vpanic+0x10/0x10
[   48.618229]  ? show_trace_log_lvl+0x2c1/0x3c0
[   48.618512]  panic+0xc7/0xd0
[   48.618711]  ? __pfx_panic+0x10/0x10
[   48.618952]  ? watchdog_timer_fn+0x5a0/0x6c0
[   48.619242]  ? watchdog_timer_fn+0x593/0x6c0
[   48.619519]  watchdog_timer_fn+0x5b1/0x6c0
[   48.619779]  ? __pfx_watchdog_timer_fn+0x10/0x10
[   48.620088]  __hrtimer_run_queues+0x6aa/0xb70
[   48.620379]  ? __pfx___hrtimer_run_queues+0x10/0x10
[   48.620696]  hrtimer_interrupt+0x397/0x870
[   48.620979]  __sysvec_apic_timer_interrupt+0x108/0x3b0
[   48.621330]  sysvec_apic_timer_interrupt+0xaf/0xd0
[   48.621631]  </IRQ>
[   48.621773]  <TASK>
[   48.621915]  asm_sysvec_apic_timer_interrupt+0x1f/0x30
[   48.622253] RIP: 0010:attach_recursive_mnt+0xf9a/0x1990
[   48.622584] Code: c1 e8 03 80 3c 18 00 0f 85 07 06 00 00 49 8d 7f 10 4d 8b 6f 70 48 89 f8 48 c1 e8 03 80 3c 18 00 0f 855
[   48.623711] RSP: 0018:ff1100001282fbb0 EFLAGS: 00000246
[   48.624042] RAX: 1fe220000537eea2 RBX: dffffc0000000000 RCX: ffffffff820f5a86
[   48.624480] RDX: ff11000014570000 RSI: ffffffff820f5ad8 RDI: ff11000029bf7510
[   48.624912] RBP: ff1100001282fcd0 R08: 0000000000000001 R09: 0000000000000001
[   48.625356] R10: ff11000010902200 R11: ff11000014570e58 R12: ff11000029017c00
[   48.625792] R13: ff11000026b98220 R14: ff110000290b4070 R15: ff11000026b981c0
[   48.626262]  ? attach_recursive_mnt+0xf16/0x1990
[   48.626559]  ? attach_recursive_mnt+0xf68/0x1990
[   48.626863]  ? __pfx_attach_recursive_mnt+0x10/0x10
[   48.627188]  ? do_raw_spin_unlock+0x15c/0x210
[   48.627465]  ? _raw_spin_unlock+0x31/0x60
[   48.627724]  ? __sanitizer_cov_trace_const_cmp4+0x1a/0x20
[   48.628076]  ? clone_mnt+0x755/0xbd0
[   48.628313]  graft_tree+0x190/0x220
[   48.628539]  ? graft_tree+0x190/0x220
[   48.628783]  path_mount+0x1b0a/0x1f70
[   48.629034]  ? lockdep_hardirqs_on+0x89/0x110
[   48.629313]  ? trace_hardirqs_on+0x51/0x60
[   48.629579]  ? __pfx_path_mount+0x10/0x10
[   48.629837]  ? __kasan_slab_free+0x4f/0x60
[   48.630121]  ? kmem_cache_free+0x2ea/0x520
[   48.630384]  ? putname.part.0+0x132/0x180
[   48.630644]  ? putname.part.0+0x137/0x180
[   48.630901]  __x64_sys_mount+0x2a6/0x330
[   48.631186]  ? __x64_sys_mount+0x2a6/0x330
[   48.631452]  ? __pfx___x64_sys_mount+0x10/0x10
[   48.631744]  x64_sys_call+0x2127/0x2180
[   48.632000]  do_syscall_64+0x6d/0x2e0
[   48.632239]  entry_SYSCALL_64_after_hwframe+0x76/0x7e
[   48.632555] RIP: 0033:0x7f5edc43ee5d
[   48.632787] Code: ff c3 66 2e 0f 1f 84 00 00 00 00 00 90 f3 0f 1e fa 48 89 f8 48 89 f7 48 89 d6 48 89 ca 4d 89 c2 4d 898
[   48.633901] RSP: 002b:00007ffd1a883988 EFLAGS: 00000246 ORIG_RAX: 00000000000000a5
[   48.634375] RAX: ffffffffffffffda RBX: 0000000000000000 RCX: 00007f5edc43ee5d
[   48.634812] RDX: 0000000000000000 RSI: 0000000020000080 RDI: 0000000020000040
[   48.635282] RBP: 00007ffd1a883aa0 R08: 0000000000000000 R09: 0000000000000000
[   48.635711] R10: 0000000000001400 R11: 0000000000000246 R12: 00007ffd1a883bf8
[   48.636154] R13: 00000000004043b3 R14: 0000000000406e08 R15: 00007f5edc795000
[   48.636600]  </TASK>
[   48.636838] Kernel Offset: disabled
[   48.637081] ---[ end Kernel panic - not syncing: softlockup: hung tasks ]---
"

Hope this cound be insightful to you.

Regards,
Yi Lai

---

If you don't need the following environment to reproduce the problem or if you
already have one reproduced environment, please ignore the following information.

How to reproduce:
git clone https://gitlab.com/xupengfe/repro_vm_env.git
cd repro_vm_env
tar -xvf repro_vm_env.tar.gz
cd repro_vm_env; ./start3.sh  // it needs qemu-system-x86_64 and I used v7.1.0
  // start3.sh will load bzImage_2241ab53cbb5cdb08a6b2d4688feb13971058f65 v6.2-rc5 kernel
  // You could change the bzImage_xxx as you want
  // Maybe you need to remove line "-drive if=pflash,format=raw,readonly=on,file=./OVMF_CODE.fd \" for different qemu version
You could use below command to log in, there is no password for root.
ssh -p 10023 root@localhost

After login vm(virtual machine) successfully, you could transfer reproduced
binary to the vm by below way, and reproduce the problem in vm:
gcc -pthread -o repro repro.c
scp -P 10023 repro root@localhost:/root/

Get the bzImage for target kernel:
Please use target kconfig and copy it to kernel_src/.config
make olddefconfig
make -jx bzImage           //x should equal or less than cpu num your pc has

Fill the bzImage file into above start3.sh to load the target kernel in vm.


Tips:
If you already have qemu-system-x86_64, please ignore below info.
If you want to install qemu v7.1.0 version:
git clone https://github.com/qemu/qemu.git
cd qemu
git checkout -f v7.1.0
mkdir build
cd build
yum install -y ninja-build.x86_64
yum -y install libslirp-devel.x86_64
../configure --target-list=x86_64-softmmu --enable-kvm --enable-vnc --enable-gtk --enable-sdl --enable-usb-redir --enable-slirp
make
make install
 
On Mon, Jun 30, 2025 at 03:52:51AM +0100, Al Viro wrote:
> The only place that really needs to be adjusted is commit_tree() -
> there we need to iterate through the copy and we might as well
> use next_mnt() for that.  However, in case when our tree has been
> slid under something already mounted (propagation to a mountpoint
> that already has something mounted on it or a 'beneath' move_mount)
> we need to take care not to walk into the overmounting tree.
> 
> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
> ---
>  fs/mount.h     |  3 +--
>  fs/namespace.c | 60 ++++++++++++++++++++------------------------------
>  fs/pnode.c     |  3 ++-
>  3 files changed, 27 insertions(+), 39 deletions(-)
> 
> diff --git a/fs/mount.h b/fs/mount.h
> index 08583428b10b..97737051a8b9 100644
> --- a/fs/mount.h
> +++ b/fs/mount.h
> @@ -193,7 +193,7 @@ static inline bool mnt_ns_empty(const struct mnt_namespace *ns)
>  	return RB_EMPTY_ROOT(&ns->mounts);
>  }
>  
> -static inline void move_from_ns(struct mount *mnt, struct list_head *dt_list)
> +static inline void move_from_ns(struct mount *mnt)
>  {
>  	struct mnt_namespace *ns = mnt->mnt_ns;
>  	WARN_ON(!mnt_ns_attached(mnt));
> @@ -203,7 +203,6 @@ static inline void move_from_ns(struct mount *mnt, struct list_head *dt_list)
>  		ns->mnt_first_node = rb_next(&mnt->mnt_node);
>  	rb_erase(&mnt->mnt_node, &ns->mounts);
>  	RB_CLEAR_NODE(&mnt->mnt_node);
> -	list_add_tail(&mnt->mnt_list, dt_list);
>  }
>  
>  bool has_locked_children(struct mount *mnt, struct dentry *dentry);
> diff --git a/fs/namespace.c b/fs/namespace.c
> index 38a46b32413d..bd6c7da901fc 100644
> --- a/fs/namespace.c
> +++ b/fs/namespace.c
> @@ -1161,34 +1161,6 @@ static void mnt_add_to_ns(struct mnt_namespace *ns, struct mount *mnt)
>  	mnt_notify_add(mnt);
>  }
>  
> -/*
> - * vfsmount lock must be held for write
> - */
> -static void commit_tree(struct mount *mnt)
> -{
> -	struct mount *parent = mnt->mnt_parent;
> -	struct mount *m;
> -	LIST_HEAD(head);
> -	struct mnt_namespace *n = parent->mnt_ns;
> -
> -	BUG_ON(parent == mnt);
> -
> -	if (!mnt_ns_attached(mnt)) {
> -		list_add_tail(&head, &mnt->mnt_list);
> -		while (!list_empty(&head)) {
> -			m = list_first_entry(&head, typeof(*m), mnt_list);
> -			list_del(&m->mnt_list);
> -
> -			mnt_add_to_ns(n, m);
> -		}
> -		n->nr_mounts += n->pending_mounts;
> -		n->pending_mounts = 0;
> -	}
> -
> -	make_visible(mnt);
> -	touch_mnt_namespace(n);
> -}
> -
>  static struct mount *next_mnt(struct mount *p, struct mount *root)
>  {
>  	struct list_head *next = p->mnt_mounts.next;
> @@ -1215,6 +1187,27 @@ static struct mount *skip_mnt_tree(struct mount *p)
>  	return p;
>  }
>  
> +/*
> + * vfsmount lock must be held for write
> + */
> +static void commit_tree(struct mount *mnt)
> +{
> +	struct mnt_namespace *n = mnt->mnt_parent->mnt_ns;
> +
> +	if (!mnt_ns_attached(mnt)) {
> +		for (struct mount *m = mnt; m; m = next_mnt(m, mnt))
> +			if (unlikely(mnt_ns_attached(m)))
> +				m = skip_mnt_tree(m);
> +			else
> +				mnt_add_to_ns(n, m);
> +		n->nr_mounts += n->pending_mounts;
> +		n->pending_mounts = 0;
> +	}
> +
> +	make_visible(mnt);
> +	touch_mnt_namespace(n);
> +}
> +
>  /**
>   * vfs_create_mount - Create a mount for a configured superblock
>   * @fc: The configuration context with the superblock attached
> @@ -1831,9 +1824,8 @@ static void umount_tree(struct mount *mnt, enum umount_tree_flags how)
>  	for (p = mnt; p; p = next_mnt(p, mnt)) {
>  		p->mnt.mnt_flags |= MNT_UMOUNT;
>  		if (mnt_ns_attached(p))
> -			move_from_ns(p, &tmp_list);
> -		else
> -			list_move(&p->mnt_list, &tmp_list);
> +			move_from_ns(p);
> +		list_add_tail(&p->mnt_list, &tmp_list);
>  	}
>  
>  	/* Hide the mounts from mnt_mounts */
> @@ -2270,7 +2262,6 @@ struct mount *copy_tree(struct mount *src_root, struct dentry *dentry,
>  					list_add(&dst_mnt->mnt_expire,
>  						 &src_mnt->mnt_expire);
>  			}
> -			list_add_tail(&dst_mnt->mnt_list, &res->mnt_list);
>  			attach_mnt(dst_mnt, dst_parent, src_parent->mnt_mp);
>  			unlock_mount_hash();
>  		}
> @@ -2686,12 +2677,9 @@ static int attach_recursive_mnt(struct mount *source_mnt,
>  		list_del_init(&source_mnt->mnt_expire);
>  	} else {
>  		if (source_mnt->mnt_ns) {
> -			LIST_HEAD(head);
> -
>  			/* move from anon - the caller will destroy */
>  			for (p = source_mnt; p; p = next_mnt(p, source_mnt))
> -				move_from_ns(p, &head);
> -			list_del_init(&head);
> +				move_from_ns(p);
>  		}
>  	}
>  
> diff --git a/fs/pnode.c b/fs/pnode.c
> index cbf5f5746252..81f7599bdac4 100644
> --- a/fs/pnode.c
> +++ b/fs/pnode.c
> @@ -449,7 +449,8 @@ static void umount_one(struct mount *m, struct list_head *to_umount)
>  {
>  	m->mnt.mnt_flags |= MNT_UMOUNT;
>  	list_del_init(&m->mnt_child);
> -	move_from_ns(m, to_umount);
> +	move_from_ns(m);
> +	list_add_tail(&m->mnt_list, to_umount);
>  }
>  
>  static void remove_from_candidate_list(struct mount *m)
> -- 
> 2.39.5
> 

Re: [PATCH v3 44/48] copy_tree(): don't link the mounts via mnt_list

[Al Viro]

On Wed, Aug 13, 2025 at 02:45:25PM +0800, Lai, Yi wrote:
> Syzkaller repro code:
> https://github.com/laifryiee/syzkaller_logs/tree/main/250813_093835_attach_recursive_mnt/repro.c

404: The main branch of syzkaller_logs does not contain the path 250813_093835_attach_recursive_mnt/repro.c.

Re: [PATCH v3 44/48] copy_tree(): don't link the mounts via mnt_list

[Al Viro]

On Wed, Aug 13, 2025 at 08:13:03AM +0100, Al Viro wrote:
> On Wed, Aug 13, 2025 at 02:45:25PM +0800, Lai, Yi wrote:
> > Syzkaller repro code:
> > https://github.com/laifryiee/syzkaller_logs/tree/main/250813_093835_attach_recursive_mnt/repro.c
> 
> 404: The main branch of syzkaller_logs does not contain the path 250813_093835_attach_recursive_mnt/repro.c.

https://github.com/laifryiee/syzkaller_logs/blob/main/250813_093835_attach_recursive_mnt/repro.c

does get it...  Anyway, I'm about to fall down right now (half past 3am here),
will take a look once I get some sleep...

Re: [PATCH v3 44/48] copy_tree(): don't link the mounts via mnt_list

[Al Viro]

On Wed, Aug 13, 2025 at 08:32:24AM +0100, Al Viro wrote:
> On Wed, Aug 13, 2025 at 08:13:03AM +0100, Al Viro wrote:
> > On Wed, Aug 13, 2025 at 02:45:25PM +0800, Lai, Yi wrote:
> > > Syzkaller repro code:
> > > https://github.com/laifryiee/syzkaller_logs/tree/main/250813_093835_attach_recursive_mnt/repro.c
> > 
> > 404: The main branch of syzkaller_logs does not contain the path 250813_093835_attach_recursive_mnt/repro.c.
> 
> https://github.com/laifryiee/syzkaller_logs/blob/main/250813_093835_attach_recursive_mnt/repro.c
> 
> does get it...  Anyway, I'm about to fall down right now (half past 3am here),
> will take a look once I get some sleep...

OK, I think I understand what's going on there.  FWIW, reproducer can be
greatly simplified:

cd /tmp
mkdir a
mount --bind a a
mount --make-shared a
while mount --bind a a do echo splat; done

Beginning of that thing is to make it possible to clean the resulting mess
out, when after about 16 iterations you run out of limit on the number of
mounts - you are explicitly asking to double the number under /tmp/a
on each iteration.  And default /proc/sys/fs/mount-max is set to 100000...

As for cleaning up, umount2("/tmp/a", MNT_DETACH); will do it...

The minimal fix should be to do commit_tree() just *before* the preceding
if (q) {...} in attach_recursive_mnt().

Said that, this is not the only problem exposed by that reproducer - with
that kind of long chain of overmounts, all peers to each other, we hit
two more stupidities on the umount side - reparent() shouldn't fucking
bother if the overmount is also going to be taken out and change_mnt_type()
only needs to look for propagation source if the victim has slaves (those
will need to be moved to new master) *or* if the victim is getting turned
into a slave.

See if the following recovers the performance:

diff --git a/fs/namespace.c b/fs/namespace.c
index a191c6519e36..88db58061919 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -1197,10 +1197,7 @@ static void commit_tree(struct mount *mnt)
 
 	if (!mnt_ns_attached(mnt)) {
 		for (struct mount *m = mnt; m; m = next_mnt(m, mnt))
-			if (unlikely(mnt_ns_attached(m)))
-				m = skip_mnt_tree(m);
-			else
-				mnt_add_to_ns(n, m);
+			mnt_add_to_ns(n, m);
 		n->nr_mounts += n->pending_mounts;
 		n->pending_mounts = 0;
 	}
@@ -2704,6 +2701,7 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 			lock_mnt_tree(child);
 		q = __lookup_mnt(&child->mnt_parent->mnt,
 				 child->mnt_mountpoint);
+		commit_tree(child);
 		if (q) {
 			struct mountpoint *mp = root.mp;
 			struct mount *r = child;
@@ -2713,7 +2711,6 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 				mp = shorter;
 			mnt_change_mountpoint(r, mp, q);
 		}
-		commit_tree(child);
 	}
 	unpin_mountpoint(&root);
 	unlock_mount_hash();
diff --git a/fs/pnode.c b/fs/pnode.c
index 81f7599bdac4..040a8559b8f5 100644
--- a/fs/pnode.c
+++ b/fs/pnode.c
@@ -111,7 +111,8 @@ void change_mnt_propagation(struct mount *mnt, int type)
 		return;
 	}
 	if (IS_MNT_SHARED(mnt)) {
-		m = propagation_source(mnt);
+		if (type == MS_SLAVE || !hlist_empty(&mnt->mnt_slave_list))
+			m = propagation_source(mnt);
 		if (list_empty(&mnt->mnt_share)) {
 			mnt_release_group_id(mnt);
 		} else {
@@ -595,6 +596,8 @@ static void reparent(struct mount *m)
 	struct mount *p = m;
 	struct mountpoint *mp;
 
+	if (will_be_unmounted(m))
+		return;
 	do {
 		mp = p->mnt_mp;
 		p = p->mnt_parent;

Re: [PATCH v3 44/48] copy_tree(): don't link the mounts via mnt_list

[Al Viro]

On Fri, Aug 15, 2025 at 12:21:14AM +0100, Al Viro wrote:

> while mount --bind a a do echo splat; done

that, of course, should've been

while mount --bind a a; do echo splat; done

Re: [PATCH v3 44/48] copy_tree(): don't link the mounts via mnt_list

[Lai, Yi]

On Fri, Aug 15, 2025 at 12:21:14AM +0100, Al Viro wrote:
> On Wed, Aug 13, 2025 at 08:32:24AM +0100, Al Viro wrote:
> > On Wed, Aug 13, 2025 at 08:13:03AM +0100, Al Viro wrote:
> > > On Wed, Aug 13, 2025 at 02:45:25PM +0800, Lai, Yi wrote:
> > > > Syzkaller repro code:
> > > > https://github.com/laifryiee/syzkaller_logs/tree/main/250813_093835_attach_recursive_mnt/repro.c
> > > 
> > > 404: The main branch of syzkaller_logs does not contain the path 250813_093835_attach_recursive_mnt/repro.c.
> > 
> > https://github.com/laifryiee/syzkaller_logs/blob/main/250813_093835_attach_recursive_mnt/repro.c
> > 
> > does get it...  Anyway, I'm about to fall down right now (half past 3am here),
> > will take a look once I get some sleep...
> 
> OK, I think I understand what's going on there.  FWIW, reproducer can be
> greatly simplified:
> 
> cd /tmp
> mkdir a
> mount --bind a a
> mount --make-shared a
> while mount --bind a a do echo splat; done
> 
> Beginning of that thing is to make it possible to clean the resulting mess
> out, when after about 16 iterations you run out of limit on the number of
> mounts - you are explicitly asking to double the number under /tmp/a
> on each iteration.  And default /proc/sys/fs/mount-max is set to 100000...
> 
> As for cleaning up, umount2("/tmp/a", MNT_DETACH); will do it...
> 
> The minimal fix should be to do commit_tree() just *before* the preceding
> if (q) {...} in attach_recursive_mnt().
> 
> Said that, this is not the only problem exposed by that reproducer - with
> that kind of long chain of overmounts, all peers to each other, we hit
> two more stupidities on the umount side - reparent() shouldn't fucking
> bother if the overmount is also going to be taken out and change_mnt_type()
> only needs to look for propagation source if the victim has slaves (those
> will need to be moved to new master) *or* if the victim is getting turned
> into a slave.
> 
> See if the following recovers the performance:
> 
> diff --git a/fs/namespace.c b/fs/namespace.c
> index a191c6519e36..88db58061919 100644
> --- a/fs/namespace.c
> +++ b/fs/namespace.c
> @@ -1197,10 +1197,7 @@ static void commit_tree(struct mount *mnt)
>  
>  	if (!mnt_ns_attached(mnt)) {
>  		for (struct mount *m = mnt; m; m = next_mnt(m, mnt))
> -			if (unlikely(mnt_ns_attached(m)))
> -				m = skip_mnt_tree(m);
> -			else
> -				mnt_add_to_ns(n, m);
> +			mnt_add_to_ns(n, m);
>  		n->nr_mounts += n->pending_mounts;
>  		n->pending_mounts = 0;
>  	}
> @@ -2704,6 +2701,7 @@ static int attach_recursive_mnt(struct mount *source_mnt,
>  			lock_mnt_tree(child);
>  		q = __lookup_mnt(&child->mnt_parent->mnt,
>  				 child->mnt_mountpoint);
> +		commit_tree(child);
>  		if (q) {
>  			struct mountpoint *mp = root.mp;
>  			struct mount *r = child;
> @@ -2713,7 +2711,6 @@ static int attach_recursive_mnt(struct mount *source_mnt,
>  				mp = shorter;
>  			mnt_change_mountpoint(r, mp, q);
>  		}
> -		commit_tree(child);
>  	}
>  	unpin_mountpoint(&root);
>  	unlock_mount_hash();
> diff --git a/fs/pnode.c b/fs/pnode.c
> index 81f7599bdac4..040a8559b8f5 100644
> --- a/fs/pnode.c
> +++ b/fs/pnode.c
> @@ -111,7 +111,8 @@ void change_mnt_propagation(struct mount *mnt, int type)
>  		return;
>  	}
>  	if (IS_MNT_SHARED(mnt)) {
> -		m = propagation_source(mnt);
> +		if (type == MS_SLAVE || !hlist_empty(&mnt->mnt_slave_list))
> +			m = propagation_source(mnt);
>  		if (list_empty(&mnt->mnt_share)) {
>  			mnt_release_group_id(mnt);
>  		} else {
> @@ -595,6 +596,8 @@ static void reparent(struct mount *m)
>  	struct mount *p = m;
>  	struct mountpoint *mp;
>  
> +	if (will_be_unmounted(m))
> +		return;
>  	do {
>  		mp = p->mnt_mp;
>  		p = p->mnt_parent;

After applying this patch on top of linux-next, issue cannot be reproduced.

Regards,
Yi Lai

[PATCH v3 45/48] take freeing of emptied mnt_namespace to namespace_unlock()

[Al Viro]

Freeing of a namespace must be delayed until after we'd dealt with mount
notifications (in namespace_unlock()).  The reasons are not immediately
obvious (they are buried in ->prev_ns handling in mnt_notify()), and
having that free_mnt_ns() explicitly called after namespace_unlock()
is asking for trouble - it does feel like they should be OK to free
as soon as they've been emptied.

Make the things more explicit by setting 'emptied_ns' under namespace_sem
and having namespace_unlock() free the sucker as soon as it's safe to free.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 27 ++++++++++++---------------
 1 file changed, 12 insertions(+), 15 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index bd6c7da901fc..85db0de5fb53 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -79,6 +79,7 @@ static struct kmem_cache *mnt_cache __ro_after_init;
 static DECLARE_RWSEM(namespace_sem);
 static HLIST_HEAD(unmounted);	/* protected by namespace_sem */
 static LIST_HEAD(ex_mountpoints); /* protected by namespace_sem */
+static struct mnt_namespace *emptied_ns; /* protected by namespace_sem */
 static DEFINE_SEQLOCK(mnt_ns_tree_lock);
 
 #ifdef CONFIG_FSNOTIFY
@@ -1730,15 +1731,18 @@ static bool need_notify_mnt_list(void)
 }
 #endif
 
+static void free_mnt_ns(struct mnt_namespace *);
 static void namespace_unlock(void)
 {
 	struct hlist_head head;
 	struct hlist_node *p;
 	struct mount *m;
+	struct mnt_namespace *ns = emptied_ns;
 	LIST_HEAD(list);
 
 	hlist_move_list(&unmounted, &head);
 	list_splice_init(&ex_mountpoints, &list);
+	emptied_ns = NULL;
 
 	if (need_notify_mnt_list()) {
 		/*
@@ -1752,6 +1756,11 @@ static void namespace_unlock(void)
 	} else {
 		up_write(&namespace_sem);
 	}
+	if (unlikely(ns)) {
+		/* Make sure we notice when we leak mounts. */
+		VFS_WARN_ON_ONCE(!mnt_ns_empty(ns));
+		free_mnt_ns(ns);
+	}
 
 	shrink_dentry_list(&list);
 
@@ -2335,12 +2344,10 @@ void drop_collected_paths(struct path *paths, struct path *prealloc)
 		kfree(paths);
 }
 
-static void free_mnt_ns(struct mnt_namespace *);
 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *, bool);
 
 void dissolve_on_fput(struct vfsmount *mnt)
 {
-	struct mnt_namespace *ns;
 	struct mount *m = real_mount(mnt);
 
 	/*
@@ -2362,15 +2369,11 @@ void dissolve_on_fput(struct vfsmount *mnt)
 		if (!anon_ns_root(m))
 			return;
 
-		ns = m->mnt_ns;
+		emptied_ns = m->mnt_ns;
 		lock_mount_hash();
 		umount_tree(m, UMOUNT_CONNECTED);
 		unlock_mount_hash();
 	}
-
-	/* Make sure we notice when we leak mounts. */
-	VFS_WARN_ON_ONCE(!mnt_ns_empty(ns));
-	free_mnt_ns(ns);
 }
 
 static bool __has_locked_children(struct mount *mnt, struct dentry *dentry)
@@ -2678,6 +2681,7 @@ static int attach_recursive_mnt(struct mount *source_mnt,
 	} else {
 		if (source_mnt->mnt_ns) {
 			/* move from anon - the caller will destroy */
+			emptied_ns = source_mnt->mnt_ns;
 			for (p = source_mnt; p; p = next_mnt(p, source_mnt))
 				move_from_ns(p);
 		}
@@ -3656,13 +3660,6 @@ static int do_move_mount(struct path *old_path,
 	err = attach_recursive_mnt(old, p, mp.mp);
 out:
 	unlock_mount(&mp);
-	if (!err) {
-		if (is_anon_ns(ns)) {
-			/* Make sure we notice when we leak mounts. */
-			VFS_WARN_ON_ONCE(!mnt_ns_empty(ns));
-			free_mnt_ns(ns);
-		}
-	}
 	return err;
 }
 
@@ -6153,11 +6150,11 @@ void put_mnt_ns(struct mnt_namespace *ns)
 	if (!refcount_dec_and_test(&ns->ns.count))
 		return;
 	namespace_lock();
+	emptied_ns = ns;
 	lock_mount_hash();
 	umount_tree(ns->root, 0);
 	unlock_mount_hash();
 	namespace_unlock();
-	free_mnt_ns(ns);
 }
 
 struct vfsmount *kern_mount(struct file_system_type *type)
-- 
2.39.5

[PATCH v3 46/48] get rid of CL_SHARE_TO_SLAVE

[Al Viro]

the only difference between it and CL_SLAVE is in this predicate
in clone_mnt():
	if ((flag & CL_SLAVE) ||
	    ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
However, in case of CL_SHARED_TO_SLAVE we have not allocated any
mount group ids since the time we'd grabbed namespace_sem, so
IS_MNT_SHARED() is equivalent to non-zero ->mnt_group_id.  And
in case of CL_SLAVE old has come either from the original tree,
which had ->mnt_group_id allocated for all nodes or from result
of sequence of CL_MAKE_SHARED or CL_MAKE_SHARED|CL_SLAVE copies,
ultimately going back to the original tree.  In both cases we are
guaranteed that old->mnt_group_id will be non-zero.

In other words, the predicate is always equal to
	(flags & (CL_SLAVE | CL_SHARED_TO_SLAVE)) && old->mnt_group_id
and with that replacement CL_SLAVE and CL_SHARED_TO_SLAVE have exact
same behaviour.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 7 +++----
 fs/pnode.h     | 1 -
 2 files changed, 3 insertions(+), 5 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index 85db0de5fb53..ca36c4a6a143 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -1309,7 +1309,7 @@ static struct mount *clone_mnt(struct mount *old, struct dentry *root,
 	mnt->mnt.mnt_flags = READ_ONCE(old->mnt.mnt_flags) &
 			     ~MNT_INTERNAL_FLAGS;
 
-	if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
+	if (flag & (CL_SLAVE | CL_PRIVATE))
 		mnt->mnt_group_id = 0; /* not a peer of original */
 	else
 		mnt->mnt_group_id = old->mnt_group_id;
@@ -1340,8 +1340,7 @@ static struct mount *clone_mnt(struct mount *old, struct dentry *root,
 	if (peers(mnt, old))
 		list_add(&mnt->mnt_share, &old->mnt_share);
 
-	if ((flag & CL_SLAVE) ||
-	    ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
+	if ((flag & CL_SLAVE) && old->mnt_group_id) {
 		hlist_add_head(&mnt->mnt_slave, &old->mnt_slave_list);
 		mnt->mnt_master = old;
 	} else if (IS_MNT_SLAVE(old)) {
@@ -4228,7 +4227,7 @@ struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
 	/* First pass: copy the tree topology */
 	copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
 	if (user_ns != ns->user_ns)
-		copy_flags |= CL_SHARED_TO_SLAVE;
+		copy_flags |= CL_SLAVE;
 	new = copy_tree(old, old->mnt.mnt_root, copy_flags);
 	if (IS_ERR(new)) {
 		namespace_unlock();
diff --git a/fs/pnode.h b/fs/pnode.h
index 507e30e7a420..00ab153e3e9d 100644
--- a/fs/pnode.h
+++ b/fs/pnode.h
@@ -25,7 +25,6 @@
 #define CL_COPY_UNBINDABLE	0x04
 #define CL_MAKE_SHARED 		0x08
 #define CL_PRIVATE 		0x10
-#define CL_SHARED_TO_SLAVE	0x20
 #define CL_COPY_MNT_NS_FILE	0x40
 
 /*
-- 
2.39.5

[PATCH v3 47/48] invent_group_ids(): zero ->mnt_group_id always implies !IS_MNT_SHARED()

[Al Viro]

All places where we call set_mnt_shared() are guaranteed to have
non-zero ->mnt_group_id - either by explicit test, or by having
done successful invent_group_ids() covering the same mount since
we'd grabbed namespace_sem.

The opposite combination (non-zero ->mnt_group_id and !IS_MNT_SHARED())
*is* possible - it means that we have allocated group id, but didn't
get around to set_mnt_shared() yet; such state is transient -
by the time we do namespace_unlock(), we must either do set_mnt_shared()
or unroll the group id allocations by cleanup_group_ids().

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index ca36c4a6a143..a75438121417 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -2516,7 +2516,7 @@ static int invent_group_ids(struct mount *mnt, bool recurse)
 	struct mount *p;
 
 	for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
-		if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
+		if (!p->mnt_group_id) {
 			int err = mnt_alloc_group_id(p);
 			if (err) {
 				cleanup_group_ids(mnt, p);
-- 
2.39.5

[PATCH v3 48/48] statmount_mnt_basic(): simplify the logics for group id

[Al Viro]

We are holding namespace_sem shared and we have not done any group
id allocations since we grabbed it.  Therefore IS_MNT_SHARED(m)
is equivalent to non-zero m->mnt_group_id.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/namespace.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index a75438121417..c549bd39c210 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -5298,7 +5298,7 @@ static void statmount_mnt_basic(struct kstatmount *s)
 	s->sm.mnt_parent_id_old = m->mnt_parent->mnt_id;
 	s->sm.mnt_attr = mnt_to_attr_flags(&m->mnt);
 	s->sm.mnt_propagation = mnt_to_propagation_flags(m);
-	s->sm.mnt_peer_group = IS_MNT_SHARED(m) ? m->mnt_group_id : 0;
+	s->sm.mnt_peer_group = m->mnt_group_id;
 	s->sm.mnt_master = IS_MNT_SLAVE(m) ? m->mnt_master->mnt_group_id : 0;
 }
 
-- 
2.39.5

Re: [PATCHES v3][RFC][CFR] mount-related stuff

[Al Viro]

On Mon, Jun 30, 2025 at 03:51:48AM +0100, Al Viro wrote:
> Updated variant (-rc4-based) force-pushed to
> git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs.git #work.mount
> individual patches in followups.  It seems to survive testing here, but
> more testing and review would be very welcome.  Again, that is not all -
> there's more stuff coming...
> 
> Folks, please review - if nobody objects, it goes into #for-next in
> a day or two.

... and there it goes.