An overlay dentry holds a lower dentry that may belong to an NFS mount.
Overlayfs calls ovl_path_open() to get a file descriptor of a lower file
for copying up its data and for a lower directory for listing its
content.
If that lower dentry gets invalidated after ovl_dentry_revalidate() and
before ovl_path_open(), the internal error code 518 (EOPENSTALE), which
is not a POSIX error code, will be exposed to the user.
Check the internal return value -EOPENSTALE in ovl_path_open(), just
the same as it is checked in path_openat() and replace it with the POSIX
error code ESTALE.
Cc: <stable@xxxxxxxxxxxxxxx>
Signed-off-by: Amir Goldstein <amir73il@xxxxxxxxx>
---
fs/overlayfs/util.c | 6 +++++-
1 file changed, 5 insertions(+), 1 deletion(-)
Miklos,
Following a similar bug report and a fix I posted for fanotify,
I realized that overlayfs can also expose EOPENSTALE to userspace.
I did not try to reproduce this with real lower NFS, only tested
patch correctness using a debug patch.
I guess a reproducer could work with ovl_copy_up_data() that is
delayed for a long time without concurrent copy up in kernel <= v4.10
and lower file is pulled underneath while copy up is pending.
However, with concurrent copy up in v4.11 that reproducer will be
of no use.
The only other suspect I found for dentry_open() of an NFS dentry
without checking -EOPENSTALE is ecryptfs.
David,
I think fscache won't be storing it local storage on NFS, right?
Added CC stable. Although I am not sure how important this fix is,
seems simple enough that breaking POSIX would justify patching stable.
Anyway, if you think this calls for a VFS helper I don't mind
doing it, I'm just terrible with names so please suggest one.
In that case, should we make dentry_open() static? non exportable?
Thanks,
Amir.
diff --git a/fs/overlayfs/util.c b/fs/overlayfs/util.c
index 89789bc..931d199 100644
--- a/fs/overlayfs/util.c
+++ b/fs/overlayfs/util.c
@@ -328,7 +328,11 @@ bool ovl_is_whiteout(struct dentry *dentry)
struct file *ovl_path_open(struct path *path, int flags)
{
- return dentry_open(path, flags | O_NOATIME, current_cred());
+ struct file *f = dentry_open(path, flags | O_NOATIME, current_cred());
+
+ if (unlikely(IS_ERR(f) && PTR_ERR(f) == -EOPENSTALE))
+ return ERR_PTR(-ESTALE);
+ return f;
}
int ovl_copy_up_start(struct dentry *dentry)
--
2.7.4
--
To unsubscribe from this list: send the line "unsubscribe linux-unionfs" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
