From: Bob Peterson <rpete...@redhat.com>

When fsck found a duplicate reference to a block it invalidated the dinode's
metadata.  But if it encountered an invalid block, for example, out of range,
the invalidating would stop.  If we encounter a block that isn't valid, we
obviously can't invalidate it.  However, if we return an error, all future
invalidating will stop for that dinode.  That's wrong because we need it to
continue to invalidate the other valid blocks.  If we don't do this, block
references that follow the bad one that are also referenced elsewhere
(duplicates) won't be flagged as such.  As a result, they'll be freed when
this corrupt dinode is deleted, despite being used by another dinode as a
valid block.  This patch makes it return a good return code so the invalidating
continues.

rhbz#675723
---
 gfs2/fsck/pass1.c |   11 +++++++++--
 1 files changed, 9 insertions(+), 2 deletions(-)

diff --git a/gfs2/fsck/pass1.c b/gfs2/fsck/pass1.c
index 49a8f4e..271075c 100644
--- a/gfs2/fsck/pass1.c
+++ b/gfs2/fsck/pass1.c
@@ -825,8 +825,15 @@ static int mark_block_invalid(struct gfs2_inode *ip, 
uint64_t block,
 {
        uint8_t q;
 
-       if (!valid_block(ip->i_sbd, block) != 0)
-               return -EFAULT;
+       /* If the block isn't valid, we obviously can't invalidate it.
+        * However, if we return an error, invalidating will stop, and
+        * we want it to continue to invalidate the valid blocks.  If we
+        * don't do this, block references that follow that are also
+        * referenced elsewhere (duplicates) won't be flagged as such,
+        * and as a result, they'll be freed when this dinode is deleted,
+        * despite being used by another dinode as a valid block. */
+       if (!valid_block(ip->i_sbd, block))
+               return 0;
 
        q = block_type(block);
        if (q != gfs2_block_free) {
-- 
1.7.7.5

Reply via email to