xfs_repair: junk leaf attribute if count == 0

We have recently seen a case where, during log replay, the
attr3 leaf verifier reported corruption when encountering a
leaf attribute with a count of 0 in the header.

We chalked this up to a transient state when a shortform leaf
was created, the attribute didn't fit, and we promoted the
(empty) attribute to the larger leaf form.

I've recently been given a metadump of unknown provenance which actually
contains a leaf attribute with count 0 on disk.  This causes the
verifier to fire every time xfs_repair is run:

 Metadata corruption detected at xfs_attr3_leaf block 0x480988/0x1000

If this 0-count state is detected, we should just junk the leaf, same
as we would do if the count was too high.  With this change, we now
remedy the problem:

 Metadata corruption detected at xfs_attr3_leaf block 0x480988/0x1000
 bad attribute count 0 in attr block 0, inode 12587828
 problem with attribute contents in inode 12587828
 clearing inode 12587828 attributes
 correcting nblocks for inode 12587828, was 2 - counted 1

Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Signed-off-by: Eric Sandeen <sandeen@sandeen.net>

diff --git a/repair/attr_repair.c b/repair/attr_repair.c
index 40cb5f7f754d82abe67bc88c084d6997e99f4ab8..b855a101864c5a31a3011071e4da6ce963f1954c 100644 (file)
--- a/repair/attr_repair.c
+++ b/repair/attr_repair.c
@@ -593,7 +593,8 @@ process_leaf_attr_block(
        stop = xfs_attr3_leaf_hdr_size(leaf);
 
        /* does the count look sorta valid? */
-       if (leafhdr.count * sizeof(xfs_attr_leaf_entry_t) + stop >
+       if (!leafhdr.count ||
+           leafhdr.count * sizeof(xfs_attr_leaf_entry_t) + stop >
                                                mp->m_sb.sb_blocksize) {
                do_warn(
        _("bad attribute count %d in attr block %u, inode %" PRIu64 "\n"),