xfs: fix node forward in xfs_node_toosmall
authorMark Tinguely <tinguely@sgi.com>
Mon, 23 Sep 2013 17:18:58 +0000 (12:18 -0500)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Sun, 13 Oct 2013 23:08:34 +0000 (16:08 -0700)
commit 997def25e4b9cee3b01609e18a52f926bca8bd2b upstream.

Commit f5ea1100 cleans up the disk to host conversions for
node directory entries, but because a variable is reused in
xfs_node_toosmall() the next node is not correctly found.
If the original node is small enough (<= 3/8 of the node size),
this change may incorrectly cause a node collapse when it should
not. That will cause an assert in xfstest generic/319:

   Assertion failed: first <= last && last < BBTOB(bp->b_length),
   file: /root/newest/xfs/fs/xfs/xfs_trans_buf.c, line: 569

Keep the original node header to get the correct forward node.

(When a node is considered for a merge with a sibling, it overwrites the
 sibling pointers of the original incore nodehdr with the sibling's
 pointers.  This leads to loop considering the original node as a merge
 candidate with itself in the second pass, and so it incorrectly
 determines a merge should occur.)

[v3: added Dave Chinner's (slightly modified) suggestion to the commit header,
cleaned up whitespace.  -bpm]

Signed-off-by: Mark Tinguely <tinguely@sgi.com>
Reviewed-by: Ben Myers <bpm@sgi.com>
Signed-off-by: Ben Myers <bpm@sgi.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
fs/xfs/xfs_da_btree.c

index 0b8b2a13cd24debe493c8982679a2c565ebae5a1..eca6f9d8a26301697e1995845de0184204ce9c25 100644 (file)
@@ -1223,6 +1223,7 @@ xfs_da3_node_toosmall(
        /* start with smaller blk num */
        forward = nodehdr.forw < nodehdr.back;
        for (i = 0; i < 2; forward = !forward, i++) {
+               struct xfs_da3_icnode_hdr thdr;
                if (forward)
                        blkno = nodehdr.forw;
                else
@@ -1235,10 +1236,10 @@ xfs_da3_node_toosmall(
                        return(error);
 
                node = bp->b_addr;
-               xfs_da3_node_hdr_from_disk(&nodehdr, node);
+               xfs_da3_node_hdr_from_disk(&thdr, node);
                xfs_trans_brelse(state->args->trans, bp);
 
-               if (count - nodehdr.count >= 0)
+               if (count - thdr.count >= 0)
                        break;  /* fits with at least 25% to spare */
        }
        if (i >= 2) {