xfs: avoid nesting transactions in xfs_qm_scall_setqlim()
authorDave Chinner <dchinner@redhat.com>
Tue, 21 May 2013 08:02:00 +0000 (18:02 +1000)
committerBen Myers <bpm@sgi.com>
Tue, 21 May 2013 18:57:05 +0000 (13:57 -0500)
Lockdep reports:

=============================================
[ INFO: possible recursive locking detected ]
3.9.0+ #3 Not tainted
---------------------------------------------
setquota/28368 is trying to acquire lock:
 (sb_internal){++++.?}, at: [<c11e8846>] xfs_trans_alloc+0x26/0x50

but task is already holding lock:
 (sb_internal){++++.?}, at: [<c11e8846>] xfs_trans_alloc+0x26/0x50

from xfs_qm_scall_setqlim()->xfs_dqread() when a dquot needs to be
allocated.

xfs_qm_scall_setqlim() is starting a transaction and then not
passing it into xfs_qm_dqet() and so it starts it's own transaction
when allocating the dquot.  Splat!

Fix this by not allocating the dquot in xfs_qm_scall_setqlim()
inside the setqlim transaction. This requires getting the dquot
first (and allocating it if necessary) then dropping and relocking
the dquot before joining it to the setqlim transaction.

Reported-by: Michael L. Semon <mlsemon35@gmail.com>
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Ben Myers <bpm@sgi.com>
Signed-off-by: Ben Myers <bpm@sgi.com>
fs/xfs/xfs_qm_syscalls.c

index c41190c..6cdf6ff 100644 (file)
@@ -489,31 +489,36 @@ xfs_qm_scall_setqlim(
        if ((newlim->d_fieldmask & XFS_DQ_MASK) == 0)
                return 0;
 
-       tp = xfs_trans_alloc(mp, XFS_TRANS_QM_SETQLIM);
-       error = xfs_trans_reserve(tp, 0, XFS_QM_SETQLIM_LOG_RES(mp),
-                                 0, 0, XFS_DEFAULT_LOG_COUNT);
-       if (error) {
-               xfs_trans_cancel(tp, 0);
-               return (error);
-       }
-
        /*
         * We don't want to race with a quotaoff so take the quotaoff lock.
-        * (We don't hold an inode lock, so there's nothing else to stop
-        * a quotaoff from happening). (XXXThis doesn't currently happen
-        * because we take the vfslock before calling xfs_qm_sysent).
+        * We don't hold an inode lock, so there's nothing else to stop
+        * a quotaoff from happening.
         */
        mutex_lock(&q->qi_quotaofflock);
 
        /*
-        * Get the dquot (locked), and join it to the transaction.
-        * Allocate the dquot if this doesn't exist.
+        * Get the dquot (locked) before we start, as we need to do a
+        * transaction to allocate it if it doesn't exist. Once we have the
+        * dquot, unlock it so we can start the next transaction safely. We hold
+        * a reference to the dquot, so it's safe to do this unlock/lock without
+        * it being reclaimed in the mean time.
         */
-       if ((error = xfs_qm_dqget(mp, NULL, id, type, XFS_QMOPT_DQALLOC, &dqp))) {
-               xfs_trans_cancel(tp, XFS_TRANS_ABORT);
+       error = xfs_qm_dqget(mp, NULL, id, type, XFS_QMOPT_DQALLOC, &dqp);
+       if (error) {
                ASSERT(error != ENOENT);
                goto out_unlock;
        }
+       xfs_dqunlock(dqp);
+
+       tp = xfs_trans_alloc(mp, XFS_TRANS_QM_SETQLIM);
+       error = xfs_trans_reserve(tp, 0, XFS_QM_SETQLIM_LOG_RES(mp),
+                                 0, 0, XFS_DEFAULT_LOG_COUNT);
+       if (error) {
+               xfs_trans_cancel(tp, 0);
+               goto out_rele;
+       }
+
+       xfs_dqlock(dqp);
        xfs_trans_dqjoin(tp, dqp);
        ddq = &dqp->q_core;
 
@@ -621,9 +626,10 @@ xfs_qm_scall_setqlim(
        xfs_trans_log_dquot(tp, dqp);
 
        error = xfs_trans_commit(tp, 0);
-       xfs_qm_dqrele(dqp);
 
- out_unlock:
+out_rele:
+       xfs_qm_dqrele(dqp);
+out_unlock:
        mutex_unlock(&q->qi_quotaofflock);
        return error;
 }