[PATCH v4 0/1] vfs: Respect MS_RDONLY at bind mount creation

[PATCH v4 0/1] vfs: Respect MS_RDONLY at bind mount creation

[Richard Yao]

Dear Linus,

I am the ZFSOnLinux developer that ran into you on his way to the airport at
LinuxCon Europe last month. This is the VFS patch that you volunteered to help
me get properly reviewed. My plan was to send it the week after LinuxCon Europe
(last week) as you had asked. However, that was delayed until Monday last week.

Coincidentally, I was in Boston last week and I had the opportunity to chat
with Ted about filesystem stuff in person. He suggested me that this patch
likely got lost in your inbox. He has also volunteered to review the patch and
asked that I resend.

This patch fixes a bug that has annoyed both myself and others for some time
now. I am presently super busy preparing for next week's Open ZFS developer
summit, but I would love to see this patch merged. I will make time to respond
to any emails regarding it.

Yours truly,
Richard Yao

Richard Yao (1):
  vfs: Respect MS_RDONLY at bind mount creation

 fs/namespace.c | 20 ++++++++++++++++----
 fs/pnode.h     | 17 +++++++++--------
 2 files changed, 25 insertions(+), 12 deletions(-)

-- 
2.0.4

[PATCH v4 1/1] vfs: Respect MS_RDONLY at bind mount creation

[Richard Yao]

`mount -o bind,ro ...` suffers from a silent failure where the readonly
flag is ignored. The bind mount will be created rw whenever the target
is rw. Users typically workaround this by remounting readonly, but that
does not work when you want to define readonly bind mounts in fstab.
This is a major annoyance when dealing with recursive bind mounts
because the userland mount command does not expose the option to
recursively remount a subtree as readonly.

Signed-off-by: Richard Yao <ryao@gentoo.org>
---
 fs/namespace.c | 20 ++++++++++++++++----
 fs/pnode.h     | 17 +++++++++--------
 2 files changed, 25 insertions(+), 12 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index 5b66b2b..6f07336 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -990,6 +990,14 @@ static struct mount *clone_mnt(struct mount *old, struct dentry *root,
 	if (flag & CL_MAKE_SHARED)
 		set_mnt_shared(mnt);
 
+	/*
+	 * We set the flag directly because the mount point is not yet visible.
+	 * This means there are no writers that require the checks in
+	 * mnt_make_readonly().
+	 */
+	if (flag & CL_MAKE_RDONLY)
+		mnt->mnt.mnt_flags |= MNT_READONLY;
+
 	/* stick the duplicate mount on the same expiry list
 	 * as the original if that was on one */
 	if (flag & CL_EXPIRE) {
@@ -1992,11 +2000,13 @@ static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
  * do loopback mount.
  */
 static int do_loopback(struct path *path, const char *old_name,
-				int recurse)
+				unsigned long flags)
 {
 	struct path old_path;
 	struct mount *mnt = NULL, *old, *parent;
 	struct mountpoint *mp;
+	int recurse = flags & MS_REC;
+	int clflags = (flags & MS_RDONLY) ? CL_MAKE_RDONLY : 0;
 	int err;
 	if (!old_name || !*old_name)
 		return -EINVAL;
@@ -2027,9 +2037,10 @@ static int do_loopback(struct path *path, const char *old_name,
 		goto out2;
 
 	if (recurse)
-		mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE);
+		mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE |
+			clflags);
 	else
-		mnt = clone_mnt(old, old_path.dentry, 0);
+		mnt = clone_mnt(old, old_path.dentry, clflags);
 
 	if (IS_ERR(mnt)) {
 		err = PTR_ERR(mnt);
@@ -2625,7 +2636,8 @@ long do_mount(const char *dev_name, const char __user *dir_name,
 		retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
 				    data_page);
 	else if (flags & MS_BIND)
-		retval = do_loopback(&path, dev_name, flags & MS_REC);
+		retval = do_loopback(&path, dev_name, flags & (MS_REC |
+							       MS_RDONLY));
 	else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
 		retval = do_change_type(&path, flags);
 	else if (flags & MS_MOVE)
diff --git a/fs/pnode.h b/fs/pnode.h
index 4a24635..326f5be 100644
--- a/fs/pnode.h
+++ b/fs/pnode.h
@@ -20,14 +20,15 @@
 #define SET_MNT_MARK(m) ((m)->mnt.mnt_flags |= MNT_MARKED)
 #define CLEAR_MNT_MARK(m) ((m)->mnt.mnt_flags &= ~MNT_MARKED)
 
-#define CL_EXPIRE    		0x01
-#define CL_SLAVE     		0x02
-#define CL_COPY_UNBINDABLE	0x04
-#define CL_MAKE_SHARED 		0x08
-#define CL_PRIVATE 		0x10
-#define CL_SHARED_TO_SLAVE	0x20
-#define CL_UNPRIVILEGED		0x40
-#define CL_COPY_MNT_NS_FILE	0x80
+#define CL_EXPIRE		0x001
+#define CL_SLAVE		0x002
+#define CL_COPY_UNBINDABLE	0x004
+#define CL_MAKE_SHARED		0x008
+#define CL_PRIVATE		0x010
+#define CL_SHARED_TO_SLAVE	0x020
+#define CL_UNPRIVILEGED		0x040
+#define CL_COPY_MNT_NS_FILE	0x080
+#define CL_MAKE_RDONLY		0x100
 
 #define CL_COPY_ALL		(CL_COPY_UNBINDABLE | CL_COPY_MNT_NS_FILE)
 
-- 
2.0.4

Re: [PATCH v4 1/1] vfs: Respect MS_RDONLY at bind mount creation

[Mateusz Guzik]

On Wed, Nov 05, 2014 at 08:44:03PM -0500, Richard Yao wrote:
> `mount -o bind,ro ...` suffers from a silent failure where the readonly
> flag is ignored. The bind mount will be created rw whenever the target
> is rw. Users typically workaround this by remounting readonly, but that
> does not work when you want to define readonly bind mounts in fstab.
> This is a major annoyance when dealing with recursive bind mounts
> because the userland mount command does not expose the option to
> recursively remount a subtree as readonly.

This is still a problem in 4.0 kernels.

Can we get this fixed please?

Is there anything wrong with the patch posted here? AFAICT the patch is
still fine (apart from some whitespace issues).

Thanks,
> 
> Signed-off-by: Richard Yao <ryao@gentoo.org>
> ---
>  fs/namespace.c | 20 ++++++++++++++++----
>  fs/pnode.h     | 17 +++++++++--------
>  2 files changed, 25 insertions(+), 12 deletions(-)
> 
> diff --git a/fs/namespace.c b/fs/namespace.c
> index 5b66b2b..6f07336 100644
> --- a/fs/namespace.c
> +++ b/fs/namespace.c
> @@ -990,6 +990,14 @@ static struct mount *clone_mnt(struct mount *old, struct dentry *root,
>  	if (flag & CL_MAKE_SHARED)
>  		set_mnt_shared(mnt);
>  
> +	/*
> +	 * We set the flag directly because the mount point is not yet visible.
> +	 * This means there are no writers that require the checks in
> +	 * mnt_make_readonly().
> +	 */
> +	if (flag & CL_MAKE_RDONLY)
> +		mnt->mnt.mnt_flags |= MNT_READONLY;
> +
>  	/* stick the duplicate mount on the same expiry list
>  	 * as the original if that was on one */
>  	if (flag & CL_EXPIRE) {
> @@ -1992,11 +2000,13 @@ static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
>   * do loopback mount.
>   */
>  static int do_loopback(struct path *path, const char *old_name,
> -				int recurse)
> +				unsigned long flags)
>  {
>  	struct path old_path;
>  	struct mount *mnt = NULL, *old, *parent;
>  	struct mountpoint *mp;
> +	int recurse = flags & MS_REC;
> +	int clflags = (flags & MS_RDONLY) ? CL_MAKE_RDONLY : 0;
>  	int err;
>  	if (!old_name || !*old_name)
>  		return -EINVAL;
> @@ -2027,9 +2037,10 @@ static int do_loopback(struct path *path, const char *old_name,
>  		goto out2;
>  
>  	if (recurse)
> -		mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE);
> +		mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE |
> +			clflags);
>  	else
> -		mnt = clone_mnt(old, old_path.dentry, 0);
> +		mnt = clone_mnt(old, old_path.dentry, clflags);
>  
>  	if (IS_ERR(mnt)) {
>  		err = PTR_ERR(mnt);
> @@ -2625,7 +2636,8 @@ long do_mount(const char *dev_name, const char __user *dir_name,
>  		retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
>  				    data_page);
>  	else if (flags & MS_BIND)
> -		retval = do_loopback(&path, dev_name, flags & MS_REC);
> +		retval = do_loopback(&path, dev_name, flags & (MS_REC |
> +							       MS_RDONLY));
>  	else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
>  		retval = do_change_type(&path, flags);
>  	else if (flags & MS_MOVE)
> diff --git a/fs/pnode.h b/fs/pnode.h
> index 4a24635..326f5be 100644
> --- a/fs/pnode.h
> +++ b/fs/pnode.h
> @@ -20,14 +20,15 @@
>  #define SET_MNT_MARK(m) ((m)->mnt.mnt_flags |= MNT_MARKED)
>  #define CLEAR_MNT_MARK(m) ((m)->mnt.mnt_flags &= ~MNT_MARKED)
>  
> -#define CL_EXPIRE    		0x01
> -#define CL_SLAVE     		0x02
> -#define CL_COPY_UNBINDABLE	0x04
> -#define CL_MAKE_SHARED 		0x08
> -#define CL_PRIVATE 		0x10
> -#define CL_SHARED_TO_SLAVE	0x20
> -#define CL_UNPRIVILEGED		0x40
> -#define CL_COPY_MNT_NS_FILE	0x80
> +#define CL_EXPIRE		0x001
> +#define CL_SLAVE		0x002
> +#define CL_COPY_UNBINDABLE	0x004
> +#define CL_MAKE_SHARED		0x008
> +#define CL_PRIVATE		0x010
> +#define CL_SHARED_TO_SLAVE	0x020
> +#define CL_UNPRIVILEGED		0x040
> +#define CL_COPY_MNT_NS_FILE	0x080
> +#define CL_MAKE_RDONLY		0x100
>  
>  #define CL_COPY_ALL		(CL_COPY_UNBINDABLE | CL_COPY_MNT_NS_FILE)
>  
> -- 
> 2.0.4
> 
> --
> To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html

-- 
Mateusz Guzik

[PATCH v4 0/1] vfs: Respect MS_RDONLY at bind mount creation

[Richard Yao]

Dear Linus,

Here is the patch that we discussed at LinuxCon Europe. My plan was to send it
the week after LinuxCon Europe (last week) as you had asked. However, I caught
somethng at the conference and was sick last week, so that was delayed.

The history of the patch is that the v1 was done in July. I sent v2 shortly
afterward upon realizing that v1 did not take the opportunity atomically close
the race with respect to bind mounting and marking readonly. v3 and v4 followed
with feedback from Mateusz Guzik, but aside from him, no one else replied.

Then at LinuxCon Europe, I spoke briefly to Mauro Carvalho Chehab, Ted T'so and
yourself about it. Mauro thought it looked okay, but he told me not to include
his ACKed by because he is not involved in this subsystem. Ted was surprised
that this was even a bug. The two of us spoke about so many filesystem things
that I neglected to ask if including him as ACKed by was okay. I have CCed
Mateusz, Ted, Mauro, the usual people from get_maintainers.pl and other
interested parties.

Yours truly,
Richard Yao

Richard Yao (1):
  vfs: Respect MS_RDONLY at bind mount creation

 fs/namespace.c | 20 ++++++++++++++++----
 fs/pnode.h     | 17 +++++++++--------
 2 files changed, 25 insertions(+), 12 deletions(-)

-- 
2.0.4

[PATCH v4 1/1] vfs: Respect MS_RDONLY at bind mount creation

[Richard Yao]

`mount -o bind,ro ...` suffers from a silent failure where the readonly
flag is ignored. The bind mount will be created rw whenever the target
is rw. Users typically workaround this by remounting readonly, but that
does not work when you want to define readonly bind mounts in fstab.
This is a major annoyance when dealing with recursive bind mounts
because the userland mount command does not expose the option to
recursively remount a subtree as readonly.

Signed-off-by: Richard Yao <ryao@gentoo.org>
---
 fs/namespace.c | 20 ++++++++++++++++----
 fs/pnode.h     | 17 +++++++++--------
 2 files changed, 25 insertions(+), 12 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index 5b66b2b..6f07336 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -990,6 +990,14 @@ static struct mount *clone_mnt(struct mount *old, struct dentry *root,
 	if (flag & CL_MAKE_SHARED)
 		set_mnt_shared(mnt);
 
+	/*
+	 * We set the flag directly because the mount point is not yet visible.
+	 * This means there are no writers that require the checks in
+	 * mnt_make_readonly().
+	 */
+	if (flag & CL_MAKE_RDONLY)
+		mnt->mnt.mnt_flags |= MNT_READONLY;
+
 	/* stick the duplicate mount on the same expiry list
 	 * as the original if that was on one */
 	if (flag & CL_EXPIRE) {
@@ -1992,11 +2000,13 @@ static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
  * do loopback mount.
  */
 static int do_loopback(struct path *path, const char *old_name,
-				int recurse)
+				unsigned long flags)
 {
 	struct path old_path;
 	struct mount *mnt = NULL, *old, *parent;
 	struct mountpoint *mp;
+	int recurse = flags & MS_REC;
+	int clflags = (flags & MS_RDONLY) ? CL_MAKE_RDONLY : 0;
 	int err;
 	if (!old_name || !*old_name)
 		return -EINVAL;
@@ -2027,9 +2037,10 @@ static int do_loopback(struct path *path, const char *old_name,
 		goto out2;
 
 	if (recurse)
-		mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE);
+		mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE |
+			clflags);
 	else
-		mnt = clone_mnt(old, old_path.dentry, 0);
+		mnt = clone_mnt(old, old_path.dentry, clflags);
 
 	if (IS_ERR(mnt)) {
 		err = PTR_ERR(mnt);
@@ -2625,7 +2636,8 @@ long do_mount(const char *dev_name, const char __user *dir_name,
 		retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
 				    data_page);
 	else if (flags & MS_BIND)
-		retval = do_loopback(&path, dev_name, flags & MS_REC);
+		retval = do_loopback(&path, dev_name, flags & (MS_REC |
+							       MS_RDONLY));
 	else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
 		retval = do_change_type(&path, flags);
 	else if (flags & MS_MOVE)
diff --git a/fs/pnode.h b/fs/pnode.h
index 4a24635..326f5be 100644
--- a/fs/pnode.h
+++ b/fs/pnode.h
@@ -20,14 +20,15 @@
 #define SET_MNT_MARK(m) ((m)->mnt.mnt_flags |= MNT_MARKED)
 #define CLEAR_MNT_MARK(m) ((m)->mnt.mnt_flags &= ~MNT_MARKED)
 
-#define CL_EXPIRE    		0x01
-#define CL_SLAVE     		0x02
-#define CL_COPY_UNBINDABLE	0x04
-#define CL_MAKE_SHARED 		0x08
-#define CL_PRIVATE 		0x10
-#define CL_SHARED_TO_SLAVE	0x20
-#define CL_UNPRIVILEGED		0x40
-#define CL_COPY_MNT_NS_FILE	0x80
+#define CL_EXPIRE		0x001
+#define CL_SLAVE		0x002
+#define CL_COPY_UNBINDABLE	0x004
+#define CL_MAKE_SHARED		0x008
+#define CL_PRIVATE		0x010
+#define CL_SHARED_TO_SLAVE	0x020
+#define CL_UNPRIVILEGED		0x040
+#define CL_COPY_MNT_NS_FILE	0x080
+#define CL_MAKE_RDONLY		0x100
 
 #define CL_COPY_ALL		(CL_COPY_UNBINDABLE | CL_COPY_MNT_NS_FILE)
 
-- 
2.0.4