xfs: fix assertion failure at xfs_setattr_nonsize
authorJie Liu <jeff.liu@oracle.com>
Tue, 26 Nov 2013 13:38:34 +0000 (21:38 +0800)
committerBen Myers <bpm@sgi.com>
Tue, 17 Dec 2013 15:16:08 +0000 (09:16 -0600)
For CRC enabled v5 super block, change a file's ownership can simply
trigger an ASSERT failure at xfs_setattr_nonsize() if both group and
project quota are enabled, i.e,

[  305.337609] XFS: Assertion failed: !XFS_IS_PQUOTA_ON(mp), file: fs/xfs/xfs_iops.c, line: 621
[  305.339250] Kernel BUG at ffffffffa0a7fa32 [verbose debug info unavailable]
[  305.383939] Call Trace:
[  305.385536]  [<ffffffffa0a7d95a>] xfs_setattr_nonsize+0x69a/0x720 [xfs]
[  305.387142]  [<ffffffffa0a7dea9>] xfs_vn_setattr+0x29/0x70 [xfs]
[  305.388727]  [<ffffffff811ca388>] notify_change+0x1a8/0x350
[  305.390298]  [<ffffffff811ac39d>] chown_common+0xfd/0x110
[  305.391868]  [<ffffffff811ad6bf>] SyS_fchownat+0xaf/0x110
[  305.393440]  [<ffffffff811ad760>] SyS_lchown+0x20/0x30
[  305.394995]  [<ffffffff8170f7dd>] system_call_fastpath+0x1a/0x1f
[  305.399870] RIP  [<ffffffffa0a7fa32>] assfail+0x22/0x30 [xfs]

This fix adjust the assertion to check if the super block support both
quota inodes or not.

Signed-off-by: Jie Liu <jeff.liu@oracle.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Ben Myers <bpm@sgi.com>
(cherry picked from commit 5a01dd54f4a7fb513062070c5acef20d13cad980)

fs/xfs/xfs_iops.c

index 27e0e54..104455b 100644 (file)
@@ -618,7 +618,8 @@ xfs_setattr_nonsize(
                }
                if (!gid_eq(igid, gid)) {
                        if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_GQUOTA_ON(mp)) {
-                               ASSERT(!XFS_IS_PQUOTA_ON(mp));
+                               ASSERT(xfs_sb_version_has_pquotino(&mp->m_sb) ||
+                                      !XFS_IS_PQUOTA_ON(mp));
                                ASSERT(mask & ATTR_GID);
                                ASSERT(gdqp);
                                olddquot2 = xfs_qm_vop_chown(tp, ip,