Re: Overlayfs rename bug

From: Miklos Szeredi
Date: Tue Jul 08 2014 - 05:29:35 EST

On Tue, Jun 24, 2014 at 05:46:14PM +0100, David Howells wrote:

> strace shows:
> rename("/mnt/a/foo104", "/mnt/a/foo105") = 0
> lstat("/mnt/a/foo104", {st_mode=S_IFREG|0644, st_size=12, ...}) = 0
> which shouldn't happen.

Sorry for the delay. Following patch fixes it and tests now run fine.

Problem was that copy-up didn't set opaque flag on non-dir. The reason this has
gone unnoticed is that the dentry on overlayfs was unhashed (to get rid of the
unneeded lower dentry reference) so after a new lookup the opaque flag would be
set correctly. Rename, however, rehashed the copied up dentry and so old_opaque
and the opaque flag on old would become out-of-sync.

As a followup patch we could also unhash the copied dentry after the rename, but
that's just an optimization.

Thanks for the report and the great test suite!


diff --git a/fs/overlayfs/copy_up.c b/fs/overlayfs/copy_up.c
index 1670dbe..274c857 100644
--- a/fs/overlayfs/copy_up.c
+++ b/fs/overlayfs/copy_up.c
@@ -255,9 +255,13 @@ static int ovl_copy_up_locked(struct dentry *workdir, struct dentry *upperdir,
* Easiest way to get rid of the lower dentry reference is to
* drop this dentry. This is neither needed nor possible for
* directories.
+ *
+ * Non-directores become opaque when copied up.
- if (!S_ISDIR(stat->mode))
+ if (!S_ISDIR(stat->mode)) {
+ ovl_dentry_set_opaque(dentry, true);
+ }
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at
Please read the FAQ at