linux-fsdevel.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
* [PATCH v2 0/2] virtiofs submounts forgotten after client / guest memory pressure
@ 2023-09-12 18:34 Krister Johansen
  2023-09-12 18:34 ` [PATCH v2 1/2] fuse: revalidate: move lookup into a separate function Krister Johansen
  2023-09-12 18:34 ` [PATCH v2 2/2] fuse: ensure that submounts lookup their root Krister Johansen
  0 siblings, 2 replies; 3+ messages in thread
From: Krister Johansen @ 2023-09-12 18:34 UTC (permalink / raw)
  To: Miklos Szeredi, linux-fsdevel
  Cc: linux-kernel, German Maglione, Greg Kurz, Max Reitz,
	Bernd Schubert

Hi,
I recently ran into a situation where a virtiofs client began
encountering EBADF after the client / guest system had an OOM.  After
reproducing the issue and debugging, it appears that the problem is
caused by a virtiofsd submount being forgotten once the dentry
referencing that submount is killed by the shrinker.  In this particular
case, the submount had been bind mounted into a container's mount
namespace.  The reference count on the original dentry was 0, making it
eligible for eviction.  However, because this dentry was also the last
reference the client knew it had, it sent a forget message to the
server.  This caused all future references to the FUSE node-id from
virtiofsd perspective to become invalid.  Subsequent attempts to used
the node-id received an EBADF from the server.

This pair of patches modifies the virtiofs submount code to perform a
lookup on the nodeid that forms the root of the submount.  The patch
before this pulls the revalidate lookup code into a helper function that
can be used both in revalidate and submount superblock fill.

Tested via:

- fstests for virtiofs
- fstests for fuse (against passthrough_ll)
- manual testing to watch how refcounts change between client and server
  in response to filesytem access, umount, and eviction by the shrinker.

Changes since v1:

- Cleanups to pacify test robot

Changes since RFC:

- Modified fuse_fill_super_submount to always fail if dentry cannot be
  revalidated.  (Feedback from Bernd Schubert)
- Fixed up an edge case where looked up but subsequently declared
  invalid dentries were not correctly tracking nlookup.  (Error was
  introduced in my RFC).

Thanks,

-K


Krister Johansen (2):
  fuse: revalidate: move lookup into a separate function
  fuse: ensure that submounts lookup their root

 fs/fuse/dir.c    | 85 +++++++++++++++++++++++++++++++++---------------
 fs/fuse/fuse_i.h |  6 ++++
 fs/fuse/inode.c  | 43 ++++++++++++++++++++----
 3 files changed, 101 insertions(+), 33 deletions(-)

-- 
2.25.1


^ permalink raw reply	[flat|nested] 3+ messages in thread

* [PATCH v2 1/2] fuse: revalidate: move lookup into a separate function
  2023-09-12 18:34 [PATCH v2 0/2] virtiofs submounts forgotten after client / guest memory pressure Krister Johansen
@ 2023-09-12 18:34 ` Krister Johansen
  2023-09-12 18:34 ` [PATCH v2 2/2] fuse: ensure that submounts lookup their root Krister Johansen
  1 sibling, 0 replies; 3+ messages in thread
From: Krister Johansen @ 2023-09-12 18:34 UTC (permalink / raw)
  To: Miklos Szeredi, linux-fsdevel
  Cc: linux-kernel, German Maglione, Greg Kurz, Max Reitz,
	Bernd Schubert

If this refactoring seems cumbersome, it's because the goal is to move
the lookup parts of fuse_dentry_revalidate into a common function.  This
function will be used elsewhere in a separate commit.  In the meantime,
the new function fuse_dentry_revalidate_lookup is responsible for just
the lookup and validation portions of the revalidate dance.  The
fuse_dentry_revalidate function retains the responsibility for
invalidating and mutating any state associated with the origial
fuse_inode and dentry.

Cc: stable@vger.kernel.org
Fixes: 1866d779d5d2 ("fuse: Allow fuse_fill_super_common() for submounts")
Signed-off-by: Krister Johansen <kjlx@templeofstupid.com>
---
 fs/fuse/dir.c | 85 +++++++++++++++++++++++++++++++++++----------------
 1 file changed, 58 insertions(+), 27 deletions(-)

diff --git a/fs/fuse/dir.c b/fs/fuse/dir.c
index 04a1c62342dc..da5b6079d88c 100644
--- a/fs/fuse/dir.c
+++ b/fs/fuse/dir.c
@@ -183,6 +183,57 @@ static void fuse_lookup_init(struct fuse_conn *fc, struct fuse_args *args,
 	args->out_args[0].value = outarg;
 }
 
+static int fuse_dentry_revalidate_lookup(struct fuse_mount *fm,
+					 struct dentry *entry,
+					 struct inode *inode,
+					 struct fuse_entry_out *outarg,
+					 bool *lookedup)
+{
+	struct dentry *parent;
+	struct fuse_forget_link *forget;
+	FUSE_ARGS(args);
+	int ret;
+
+	forget = fuse_alloc_forget();
+	ret = -ENOMEM;
+	if (!forget)
+		goto out;
+
+	parent = dget_parent(entry);
+	fuse_lookup_init(fm->fc, &args, get_node_id(d_inode(parent)),
+			 &entry->d_name, outarg);
+	ret = fuse_simple_request(fm, &args);
+	dput(parent);
+
+	/* Zero nodeid is same as -ENOENT */
+	if (!ret && !outarg->nodeid)
+		ret = -ENOENT;
+	if (!ret) {
+		if (outarg->nodeid != get_node_id(inode) ||
+		    (bool) IS_AUTOMOUNT(inode) != (bool) (outarg->attr.flags & FUSE_ATTR_SUBMOUNT)) {
+			fuse_queue_forget(fm->fc, forget,
+					  outarg->nodeid, 1);
+			goto invalid;
+		}
+		*lookedup = true;
+	}
+	kfree(forget);
+	if (ret == -ENOMEM || ret == -EINTR)
+		goto out;
+	if (ret || fuse_invalid_attr(&outarg->attr) ||
+	    fuse_stale_inode(inode, outarg->generation, &outarg->attr)) {
+		goto invalid;
+	}
+
+	ret = 1;
+out:
+	return ret;
+
+invalid:
+	ret = 0;
+	goto out;
+}
+
 /*
  * Check whether the dentry is still valid
  *
@@ -206,9 +257,8 @@ static int fuse_dentry_revalidate(struct dentry *entry, unsigned int flags)
 	else if (time_before64(fuse_dentry_time(entry), get_jiffies_64()) ||
 		 (flags & (LOOKUP_EXCL | LOOKUP_REVAL | LOOKUP_RENAME_TARGET))) {
 		struct fuse_entry_out outarg;
-		FUSE_ARGS(args);
-		struct fuse_forget_link *forget;
 		u64 attr_version;
+		bool lookedup = false;
 
 		/* For negative dentries, always do a fresh lookup */
 		if (!inode)
@@ -220,38 +270,19 @@ static int fuse_dentry_revalidate(struct dentry *entry, unsigned int flags)
 
 		fm = get_fuse_mount(inode);
 
-		forget = fuse_alloc_forget();
-		ret = -ENOMEM;
-		if (!forget)
-			goto out;
-
 		attr_version = fuse_get_attr_version(fm->fc);
 
-		parent = dget_parent(entry);
-		fuse_lookup_init(fm->fc, &args, get_node_id(d_inode(parent)),
-				 &entry->d_name, &outarg);
-		ret = fuse_simple_request(fm, &args);
-		dput(parent);
-		/* Zero nodeid is same as -ENOENT */
-		if (!ret && !outarg.nodeid)
-			ret = -ENOENT;
-		if (!ret) {
+		ret = fuse_dentry_revalidate_lookup(fm, entry, inode, &outarg,
+						    &lookedup);
+		if (ret == -ENOMEM || ret == -EINTR)
+			goto out;
+		if (lookedup) {
 			fi = get_fuse_inode(inode);
-			if (outarg.nodeid != get_node_id(inode) ||
-			    (bool) IS_AUTOMOUNT(inode) != (bool) (outarg.attr.flags & FUSE_ATTR_SUBMOUNT)) {
-				fuse_queue_forget(fm->fc, forget,
-						  outarg.nodeid, 1);
-				goto invalid;
-			}
 			spin_lock(&fi->lock);
 			fi->nlookup++;
 			spin_unlock(&fi->lock);
 		}
-		kfree(forget);
-		if (ret == -ENOMEM || ret == -EINTR)
-			goto out;
-		if (ret || fuse_invalid_attr(&outarg.attr) ||
-		    fuse_stale_inode(inode, outarg.generation, &outarg.attr))
+		if (ret <= 0)
 			goto invalid;
 
 		forget_all_cached_acls(inode);
-- 
2.25.1


^ permalink raw reply related	[flat|nested] 3+ messages in thread

* [PATCH v2 2/2] fuse: ensure that submounts lookup their root
  2023-09-12 18:34 [PATCH v2 0/2] virtiofs submounts forgotten after client / guest memory pressure Krister Johansen
  2023-09-12 18:34 ` [PATCH v2 1/2] fuse: revalidate: move lookup into a separate function Krister Johansen
@ 2023-09-12 18:34 ` Krister Johansen
  1 sibling, 0 replies; 3+ messages in thread
From: Krister Johansen @ 2023-09-12 18:34 UTC (permalink / raw)
  To: Miklos Szeredi, linux-fsdevel
  Cc: linux-kernel, German Maglione, Greg Kurz, Max Reitz,
	Bernd Schubert

Prior to this commit, the submount code assumed that the inode for the
root filesystem could not be evicted.  When eviction occurs the server
may forget the inode.  This author has observed a submount get an EBADF
from a virtiofsd server that resulted from the sole dentry / inode
pair getting evicted from a mount namespace and superblock where they
were originally referenced.  The dentry shrinker triggered a forget
after killing the dentry with the last reference.

As a result, a container that was also using this submount failed to
access its filesystem because it had borrowed the reference instead of
taking its own when setting up its superblock for the submount.

Fix by ensuring that submount superblock configuration looks up the
nodeid for the submount as well.

Cc: stable@vger.kernel.org
Fixes: 1866d779d5d2 ("fuse: Allow fuse_fill_super_common() for submounts")
Signed-off-by: Krister Johansen <kjlx@templeofstupid.com>
---
 fs/fuse/dir.c    | 10 +++++-----
 fs/fuse/fuse_i.h |  6 ++++++
 fs/fuse/inode.c  | 43 +++++++++++++++++++++++++++++++++++++------
 3 files changed, 48 insertions(+), 11 deletions(-)

diff --git a/fs/fuse/dir.c b/fs/fuse/dir.c
index da5b6079d88c..0002667170a9 100644
--- a/fs/fuse/dir.c
+++ b/fs/fuse/dir.c
@@ -183,11 +183,11 @@ static void fuse_lookup_init(struct fuse_conn *fc, struct fuse_args *args,
 	args->out_args[0].value = outarg;
 }
 
-static int fuse_dentry_revalidate_lookup(struct fuse_mount *fm,
-					 struct dentry *entry,
-					 struct inode *inode,
-					 struct fuse_entry_out *outarg,
-					 bool *lookedup)
+int fuse_dentry_revalidate_lookup(struct fuse_mount *fm,
+				  struct dentry *entry,
+				  struct inode *inode,
+				  struct fuse_entry_out *outarg,
+				  bool *lookedup)
 {
 	struct dentry *parent;
 	struct fuse_forget_link *forget;
diff --git a/fs/fuse/fuse_i.h b/fs/fuse/fuse_i.h
index bf0b85d0b95c..f8ba293d5d30 100644
--- a/fs/fuse/fuse_i.h
+++ b/fs/fuse/fuse_i.h
@@ -1325,6 +1325,12 @@ void fuse_dax_dontcache(struct inode *inode, unsigned int flags);
 bool fuse_dax_check_alignment(struct fuse_conn *fc, unsigned int map_alignment);
 void fuse_dax_cancel_work(struct fuse_conn *fc);
 
+/* dir.c */
+int fuse_dentry_revalidate_lookup(struct fuse_mount *fm, struct dentry *entry,
+				  struct inode *inode,
+				  struct fuse_entry_out *outarg,
+				  bool *lookedup);
+
 /* ioctl.c */
 long fuse_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
 long fuse_file_compat_ioctl(struct file *file, unsigned int cmd,
diff --git a/fs/fuse/inode.c b/fs/fuse/inode.c
index d47606206ec3..b3e7b0a397ae 100644
--- a/fs/fuse/inode.c
+++ b/fs/fuse/inode.c
@@ -1464,7 +1464,13 @@ static int fuse_fill_super_submount(struct super_block *sb,
 	struct fuse_mount *fm = get_fuse_mount_super(sb);
 	struct super_block *parent_sb = parent_fi->inode.i_sb;
 	struct fuse_attr root_attr;
+	struct fuse_inode *fi;
 	struct inode *root;
+	struct inode *parent;
+	struct dentry *pdent;
+	struct fuse_entry_out outarg;
+	bool lookedup = false;
+	int ret;
 
 	fuse_sb_defaults(sb);
 	fm->sb = sb;
@@ -1480,14 +1486,39 @@ static int fuse_fill_super_submount(struct super_block *sb,
 	if (parent_sb->s_subtype && !sb->s_subtype)
 		return -ENOMEM;
 
-	fuse_fill_attr_from_inode(&root_attr, parent_fi);
-	root = fuse_iget(sb, parent_fi->nodeid, 0, &root_attr, 0, 0);
 	/*
-	 * This inode is just a duplicate, so it is not looked up and
-	 * its nlookup should not be incremented.  fuse_iget() does
-	 * that, though, so undo it here.
+	 * It is necessary to lookup the parent_if->nodeid in case the dentry
+	 * that triggered the automount of the submount is later evicted.
+	 * If this dentry is evicted without the lookup count getting increased
+	 * on the submount root, then the server can subsequently forget this
+	 * nodeid which leads to errors when trying to access the root of the
+	 * submount.
 	 */
-	get_fuse_inode(root)->nlookup--;
+	parent = &parent_fi->inode;
+	pdent = d_find_alias(parent);
+	if (!pdent)
+		return -EINVAL;
+
+	ret = fuse_dentry_revalidate_lookup(fm, pdent, parent, &outarg,
+	    &lookedup);
+	dput(pdent);
+	/*
+	 * The new root owns this nlookup on success, and it is incremented by
+	 * fuse_iget().  In the case the lookup succeeded but revalidate fails,
+	 * ensure that the lookup count is tracked by the parent.
+	 */
+	if (ret <= 0) {
+		if (lookedup) {
+			fi = get_fuse_inode(parent);
+			spin_lock(&fi->lock);
+			fi->nlookup++;
+			spin_unlock(&fi->lock);
+		}
+		return ret ? ret : -EINVAL;
+	}
+
+	fuse_fill_attr_from_inode(&root_attr, parent_fi);
+	root = fuse_iget(sb, parent_fi->nodeid, 0, &root_attr, 0, 0);
 	sb->s_d_op = &fuse_dentry_operations;
 	sb->s_root = d_make_root(root);
 	if (!sb->s_root)
-- 
2.25.1


^ permalink raw reply related	[flat|nested] 3+ messages in thread

end of thread, other threads:[~2023-09-12 18:43 UTC | newest]

Thread overview: 3+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2023-09-12 18:34 [PATCH v2 0/2] virtiofs submounts forgotten after client / guest memory pressure Krister Johansen
2023-09-12 18:34 ` [PATCH v2 1/2] fuse: revalidate: move lookup into a separate function Krister Johansen
2023-09-12 18:34 ` [PATCH v2 2/2] fuse: ensure that submounts lookup their root Krister Johansen

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).