From: Tyler Hicks tyhicks@linux.microsoft.com
commit 2a3dcbccd64ba35c045fac92272ff981c4cbef44 upstream.
Decrement the refcount of the parent dentry's fid after walking each path component during a full path walk for a lookup. Failure to do so can lead to fids that are not clunked until the filesystem is unmounted, as indicated by this warning:
9pnet: found fid 3 not clunked
The improper refcounting after walking resulted in open(2) returning -EIO on any directories underneath the mount point when using the virtio transport. When using the fd transport, there's no apparent issue until the filesytem is unmounted and the warning above is emitted to the logs.
In some cases, the user may not yet be attached to the filesystem and a new root fid, associated with the user, is created and attached to the root dentry before the full path walk is performed. Increment the new root fid's refcount to two in that situation so that it can be safely decremented to one after it is used for the walk operation. The new fid will still be attached to the root dentry when v9fs_fid_lookup_with_uid() returns so a final refcount of one is correct/expected.
Link: https://lkml.kernel.org/r/20220527000003.355812-2-tyhicks@linux.microsoft.co... Link: https://lkml.kernel.org/r/20220612085330.1451496-4-asmadeus@codewreck.org Fixes: 6636b6dcc3db ("9p: add refcount to p9_fid struct") Cc: stable@vger.kernel.org Signed-off-by: Tyler Hicks tyhicks@linux.microsoft.com Reviewed-by: Christian Schoenebeck linux_oss@crudebyte.com [Dominique: fix clunking fid multiple times discussed in second link] Signed-off-by: Dominique Martinet asmadeus@codewreck.org Signed-off-by: Greg Kroah-Hartman gregkh@linuxfoundation.org --- fs/9p/fid.c | 22 +++++++++------------- 1 file changed, 9 insertions(+), 13 deletions(-)
--- a/fs/9p/fid.c +++ b/fs/9p/fid.c @@ -151,7 +151,7 @@ static struct p9_fid *v9fs_fid_lookup_wi const unsigned char **wnames, *uname; int i, n, l, clone, access; struct v9fs_session_info *v9ses; - struct p9_fid *fid, *old_fid = NULL; + struct p9_fid *fid, *old_fid;
v9ses = v9fs_dentry2v9ses(dentry); access = v9ses->flags & V9FS_ACCESS_MASK; @@ -193,13 +193,12 @@ static struct p9_fid *v9fs_fid_lookup_wi if (IS_ERR(fid)) return fid;
+ refcount_inc(&fid->count); v9fs_fid_add(dentry->d_sb->s_root, fid); } /* If we are root ourself just return that */ - if (dentry->d_sb->s_root == dentry) { - refcount_inc(&fid->count); + if (dentry->d_sb->s_root == dentry) return fid; - } /* * Do a multipath walk with attached root. * When walking parent we need to make sure we @@ -211,6 +210,7 @@ static struct p9_fid *v9fs_fid_lookup_wi fid = ERR_PTR(n); goto err_out; } + old_fid = fid; clone = 1; i = 0; while (i < n) { @@ -220,19 +220,15 @@ static struct p9_fid *v9fs_fid_lookup_wi * walk to ensure none of the patch component change */ fid = p9_client_walk(fid, l, &wnames[i], clone); + /* non-cloning walk will return the same fid */ + if (fid != old_fid) { + p9_client_clunk(old_fid); + old_fid = fid; + } if (IS_ERR(fid)) { - if (old_fid) { - /* - * If we fail, clunk fid which are mapping - * to path component and not the last component - * of the path. - */ - p9_client_clunk(old_fid); - } kfree(wnames); goto err_out; } - old_fid = fid; i += l; clone = 0; }