xfs_db: bug in bmap command?

[Eric Sandeen]

On 8/1/12 10:17 AM, Peter Watkins wrote:
> Hello,
> 
> If you have a moment would you be kind enough to review the test case
> and patch below?
> 
> I ran into this while using xfs_db to dump extents for a large,
> fragmented file. The extents were stored in btree form.

The xfs_db change looks right.  Way back when, commit
b3563c19992d0453db1826ffdbf0be6039998023 did this:

@@ -91,29 +91,27 @@ bmap(
                bno = NULLFSBLOCK;
                rblock = (xfs_bmdr_block_t *)XFS_DFORK_PTR(dip, whichfork);
                fsize = XFS_DFORK_SIZE(dip, mp, whichfork);
-               pp = XFS_BTREE_PTR_ADDR(xfs_bmdr, rblock, 1,
-                       XFS_BTREE_BLOCK_MAXRECS(fsize, xfs_bmdr, 0));
-               kp = XFS_BTREE_KEY_ADDR(xfs_bmdr, rblock, 1);
+               pp = XFS_BMDR_PTR_ADDR(rblock, 1, xfs_bmdr_maxrecs(mp, fsize, 0));
+               kp = XFS_BMDR_KEY_ADDR(rblock, 1);
                bno = select_child(curoffset, kp, pp, 
                                        be16_to_cpu(rblock->bb_numrecs));
                for (;;) {
                        set_cur(&typtab[typ], XFS_FSB_TO_DADDR(mp, bno),
                                blkbb, DB_RING_IGN, NULL);
-                       block = (xfs_bmbt_block_t *)iocur_top->data;
+                       block = (struct xfs_btree_block *)iocur_top->data;
                        if (be16_to_cpu(block->bb_level) == 0)
                                break;
-                       pp = XFS_BTREE_PTR_ADDR(xfs_bmbt, block, 1,
-                               XFS_BTREE_BLOCK_MAXRECS(mp->m_sb.sb_blocksize,
-                                       xfs_bmbt, 0));
-                       kp = XFS_BTREE_KEY_ADDR(xfs_bmbt, block, 1);
+                       pp = XFS_BMDR_PTR_ADDR(block, 1,
+                               xfs_bmbt_maxrecs(mp, mp->m_sb.sb_blocksize, 0));
+                       kp = XFS_BMDR_KEY_ADDR(block, 1);


as part of a kernel<->userspace syncup.  But you can see that some "bmdt's"
turned into "bmdr's" along the way.  Surprised that hasn't been found sooner!

I wonder if there's a way to make to make a testcase which a) doesn't require
256k of "good" output, and b) more robust to failure (a timeout & kill might
let it proceed...)

-Eric

> -Peter
> 
> 
> 
> _______________________________________________
> xfs mailing list
> xfs at oss.sgi.com
> http://oss.sgi.com/mailman/listinfo/xfs
>