1 files changed, 231 insertions, 131 deletions

diff --git a/fs/xfs/xfs_dquot.c b/fs/xfs/xfs_dquot.c
index ff982d983989..612ca682a513 100644
--- a/fs/xfs/xfs_dquot.c
+++ b/fs/xfs/xfs_dquot.c
@@ -31,7 +31,7 @@
  *
  * ip->i_lock
  *   qi->qi_tree_lock
- *     dquot->q_qlock (xfs_dqlock() and friends)
+ *     dquot->q_qlock
  *       dquot->q_flush (xfs_dqflock() and friends)
  *       qi->qi_lru_lock
  *
@@ -69,6 +69,31 @@ xfs_dquot_mark_sick(
 }
 
 /*
+ * Detach the dquot buffer if it's still attached, because we can get called
+ * through dqpurge after a log shutdown.  Caller must hold the dqflock or have
+ * otherwise isolated the dquot.
+ */
+void
+xfs_dquot_detach_buf(
+	struct xfs_dquot	*dqp)
+{
+	struct xfs_dq_logitem	*qlip = &dqp->q_logitem;
+	struct xfs_buf		*bp = NULL;
+
+	spin_lock(&qlip->qli_lock);
+	if (qlip->qli_item.li_buf) {
+		bp = qlip->qli_item.li_buf;
+		qlip->qli_item.li_buf = NULL;
+	}
+	spin_unlock(&qlip->qli_lock);
+	if (bp) {
+		xfs_buf_lock(bp);
+		list_del_init(&qlip->qli_item.li_bio_list);
+		xfs_buf_relse(bp);
+	}
+}
+
+/*
  * This is called to free all the memory associated with a dquot
  */
 void
@@ -76,6 +101,7 @@ xfs_qm_dqdestroy(
 	struct xfs_dquot	*dqp)
 {
 	ASSERT(list_empty(&dqp->q_lru));
+	ASSERT(dqp->q_logitem.qli_item.li_buf == NULL);
 
 	kvfree(dqp->q_logitem.qli_item.li_lv_shadow);
 	mutex_destroy(&dqp->q_qlock);
@@ -775,10 +801,11 @@ xfs_dq_get_next_id(
 static struct xfs_dquot *
 xfs_qm_dqget_cache_lookup(
 	struct xfs_mount	*mp,
-	struct xfs_quotainfo	*qi,
-	struct radix_tree_root	*tree,
-	xfs_dqid_t		id)
+	xfs_dqid_t		id,
+	xfs_dqtype_t		type)
 {
+	struct xfs_quotainfo	*qi = mp->m_quotainfo;
+	struct radix_tree_root	*tree = xfs_dquot_tree(qi, type);
 	struct xfs_dquot	*dqp;
 
 restart:
@@ -790,16 +817,12 @@ restart:
 		return NULL;
 	}
 
-	xfs_dqlock(dqp);
-	if (dqp->q_flags & XFS_DQFLAG_FREEING) {
-		xfs_dqunlock(dqp);
+	if (!lockref_get_not_dead(&dqp->q_lockref)) {
 		mutex_unlock(&qi->qi_tree_lock);
 		trace_xfs_dqget_freeing(dqp);
 		delay(1);
 		goto restart;
 	}
-
-	dqp->q_nrefs++;
 	mutex_unlock(&qi->qi_tree_lock);
 
 	trace_xfs_dqget_hit(dqp);
@@ -810,8 +833,7 @@ restart:
 /*
  * Try to insert a new dquot into the in-core cache.  If an error occurs the
  * caller should throw away the dquot and start over.  Otherwise, the dquot
- * is returned locked (and held by the cache) as if there had been a cache
- * hit.
+ * is returned (and held by the cache) as if there had been a cache hit.
  *
  * The insert needs to be done under memalloc_nofs context because the radix
  * tree can do memory allocation during insert. The qi->qi_tree_lock is taken in
@@ -822,11 +844,12 @@ restart:
 static int
 xfs_qm_dqget_cache_insert(
 	struct xfs_mount	*mp,
-	struct xfs_quotainfo	*qi,
-	struct radix_tree_root	*tree,
 	xfs_dqid_t		id,
+	xfs_dqtype_t		type,
 	struct xfs_dquot	*dqp)
 {
+	struct xfs_quotainfo	*qi = mp->m_quotainfo;
+	struct radix_tree_root	*tree = xfs_dquot_tree(qi, type);
 	unsigned int		nofs_flags;
 	int			error;
 
@@ -834,14 +857,11 @@ xfs_qm_dqget_cache_insert(
 	mutex_lock(&qi->qi_tree_lock);
 	error = radix_tree_insert(tree, id, dqp);
 	if (unlikely(error)) {
-		/* Duplicate found!  Caller must try again. */
 		trace_xfs_dqget_dup(dqp);
 		goto out_unlock;
 	}
 
-	/* Return a locked dquot to the caller, with a reference taken. */
-	xfs_dqlock(dqp);
-	dqp->q_nrefs = 1;
+	lockref_init(&dqp->q_lockref);
 	qi->qi_dquots++;
 
 out_unlock:
@@ -877,7 +897,7 @@ xfs_qm_dqget_checks(
 
 /*
  * Given the file system, id, and type (UDQUOT/GDQUOT/PDQUOT), return a
- * locked dquot, doing an allocation (if requested) as needed.
+ * dquot, doing an allocation (if requested) as needed.
  */
 int
 xfs_qm_dqget(
@@ -887,8 +907,6 @@ xfs_qm_dqget(
 	bool			can_alloc,
 	struct xfs_dquot	**O_dqpp)
 {
-	struct xfs_quotainfo	*qi = mp->m_quotainfo;
-	struct radix_tree_root	*tree = xfs_dquot_tree(qi, type);
 	struct xfs_dquot	*dqp;
 	int			error;
 
@@ -897,28 +915,30 @@ xfs_qm_dqget(
 		return error;
 
 restart:
-	dqp = xfs_qm_dqget_cache_lookup(mp, qi, tree, id);
-	if (dqp) {
-		*O_dqpp = dqp;
-		return 0;
-	}
+	dqp = xfs_qm_dqget_cache_lookup(mp, id, type);
+	if (dqp)
+		goto found;
 
 	error = xfs_qm_dqread(mp, id, type, can_alloc, &dqp);
 	if (error)
 		return error;
 
-	error = xfs_qm_dqget_cache_insert(mp, qi, tree, id, dqp);
+	error = xfs_qm_dqget_cache_insert(mp, id, type, dqp);
 	if (error) {
-		/*
-		 * Duplicate found. Just throw away the new dquot and start
-		 * over.
-		 */
 		xfs_qm_dqdestroy(dqp);
-		XFS_STATS_INC(mp, xs_qm_dquot_dups);
-		goto restart;
+		if (error == -EEXIST) {
+			/*
+			 * Duplicate found. Just throw away the new dquot and
+			 * start over.
+			 */
+			XFS_STATS_INC(mp, xs_qm_dquot_dups);
+			goto restart;
+		}
+		return error;
 	}
 
 	trace_xfs_dqget_miss(dqp);
+found:
 	*O_dqpp = dqp;
 	return 0;
 }
@@ -973,15 +993,16 @@ xfs_qm_dqget_inode(
 	struct xfs_inode	*ip,
 	xfs_dqtype_t		type,
 	bool			can_alloc,
-	struct xfs_dquot	**O_dqpp)
+	struct xfs_dquot	**dqpp)
 {
 	struct xfs_mount	*mp = ip->i_mount;
-	struct xfs_quotainfo	*qi = mp->m_quotainfo;
-	struct radix_tree_root	*tree = xfs_dquot_tree(qi, type);
 	struct xfs_dquot	*dqp;
 	xfs_dqid_t		id;
 	int			error;
 
+	ASSERT(!*dqpp);
+	xfs_assert_ilocked(ip, XFS_ILOCK_EXCL);
+
 	error = xfs_qm_dqget_checks(mp, type);
 	if (error)
 		return error;
@@ -993,11 +1014,9 @@ xfs_qm_dqget_inode(
 	id = xfs_qm_id_for_quotatype(ip, type);
 
 restart:
-	dqp = xfs_qm_dqget_cache_lookup(mp, qi, tree, id);
-	if (dqp) {
-		*O_dqpp = dqp;
-		return 0;
-	}
+	dqp = xfs_qm_dqget_cache_lookup(mp, id, type);
+	if (dqp)
+		goto found;
 
 	/*
 	 * Dquot cache miss. We don't want to keep the inode lock across
@@ -1023,7 +1042,6 @@ restart:
 		if (dqp1) {
 			xfs_qm_dqdestroy(dqp);
 			dqp = dqp1;
-			xfs_dqlock(dqp);
 			goto dqret;
 		}
 	} else {
@@ -1032,21 +1050,26 @@ restart:
 		return -ESRCH;
 	}
 
-	error = xfs_qm_dqget_cache_insert(mp, qi, tree, id, dqp);
+	error = xfs_qm_dqget_cache_insert(mp, id, type, dqp);
 	if (error) {
-		/*
-		 * Duplicate found. Just throw away the new dquot and start
-		 * over.
-		 */
 		xfs_qm_dqdestroy(dqp);
-		XFS_STATS_INC(mp, xs_qm_dquot_dups);
-		goto restart;
+		if (error == -EEXIST) {
+			/*
+			 * Duplicate found. Just throw away the new dquot and
+			 * start over.
+			 */
+			XFS_STATS_INC(mp, xs_qm_dquot_dups);
+			goto restart;
+		}
+		return error;
 	}
 
 dqret:
 	xfs_assert_ilocked(ip, XFS_ILOCK_EXCL);
 	trace_xfs_dqget_miss(dqp);
-	*O_dqpp = dqp;
+found:
+	trace_xfs_dqattach_get(dqp);
+	*dqpp = dqp;
 	return 0;
 }
 
@@ -1072,63 +1095,41 @@ xfs_qm_dqget_next(
 		else if (error != 0)
 			break;
 
+		mutex_lock(&dqp->q_qlock);
 		if (!XFS_IS_DQUOT_UNINITIALIZED(dqp)) {
 			*dqpp = dqp;
 			return 0;
 		}
 
-		xfs_qm_dqput(dqp);
+		mutex_unlock(&dqp->q_qlock);
+		xfs_qm_dqrele(dqp);
 	}
 
 	return error;
 }
 
 /*
- * Release a reference to the dquot (decrement ref-count) and unlock it.
- *
- * If there is a group quota attached to this dquot, carefully release that
- * too without tripping over deadlocks'n'stuff.
+ * Release a reference to the dquot.
  */
 void
-xfs_qm_dqput(
+xfs_qm_dqrele(
 	struct xfs_dquot	*dqp)
 {
-	ASSERT(dqp->q_nrefs > 0);
-	ASSERT(XFS_DQ_IS_LOCKED(dqp));
+	if (!dqp)
+		return;
 
-	trace_xfs_dqput(dqp);
+	trace_xfs_dqrele(dqp);
 
-	if (--dqp->q_nrefs == 0) {
+	if (lockref_put_or_lock(&dqp->q_lockref))
+		return;
+	if (!--dqp->q_lockref.count) {
 		struct xfs_quotainfo	*qi = dqp->q_mount->m_quotainfo;
-		trace_xfs_dqput_free(dqp);
 
+		trace_xfs_dqrele_free(dqp);
 		if (list_lru_add_obj(&qi->qi_lru, &dqp->q_lru))
 			XFS_STATS_INC(dqp->q_mount, xs_qm_dquot_unused);
 	}
-	xfs_dqunlock(dqp);
-}
-
-/*
- * Release a dquot. Flush it if dirty, then dqput() it.
- * dquot must not be locked.
- */
-void
-xfs_qm_dqrele(
-	struct xfs_dquot	*dqp)
-{
-	if (!dqp)
-		return;
-
-	trace_xfs_dqrele(dqp);
-
-	xfs_dqlock(dqp);
-	/*
-	 * We don't care to flush it if the dquot is dirty here.
-	 * That will create stutters that we want to avoid.
-	 * Instead we do a delayed write when we try to reclaim
-	 * a dirty dquot. Also xfs_sync will take part of the burden...
-	 */
-	xfs_qm_dqput(dqp);
+	spin_unlock(&dqp->q_lockref.lock);
 }
 
 /*
@@ -1142,9 +1143,11 @@ static void
 xfs_qm_dqflush_done(
 	struct xfs_log_item	*lip)
 {
-	struct xfs_dq_logitem	*qip = (struct xfs_dq_logitem *)lip;
-	struct xfs_dquot	*dqp = qip->qli_dquot;
+	struct xfs_dq_logitem	*qlip =
+			container_of(lip, struct xfs_dq_logitem, qli_item);
+	struct xfs_dquot	*dqp = qlip->qli_dquot;
 	struct xfs_ail		*ailp = lip->li_ailp;
+	struct xfs_buf		*bp = NULL;
 	xfs_lsn_t		tail_lsn;
 
 	/*
@@ -1156,12 +1159,11 @@ xfs_qm_dqflush_done(
 	 * holding the lock before removing the dquot from the AIL.
 	 */
 	if (test_bit(XFS_LI_IN_AIL, &lip->li_flags) &&
-	    ((lip->li_lsn == qip->qli_flush_lsn) ||
+	    (lip->li_lsn == qlip->qli_flush_lsn ||
 	     test_bit(XFS_LI_FAILED, &lip->li_flags))) {
-
 		spin_lock(&ailp->ail_lock);
-		xfs_clear_li_failed(lip);
-		if (lip->li_lsn == qip->qli_flush_lsn) {
+		clear_bit(XFS_LI_FAILED, &lip->li_flags);
+		if (lip->li_lsn == qlip->qli_flush_lsn) {
 			/* xfs_ail_update_finish() drops the AIL lock */
 			tail_lsn = xfs_ail_delete_one(ailp, lip);
 			xfs_ail_update_finish(ailp, tail_lsn);
@@ -1171,6 +1173,20 @@ xfs_qm_dqflush_done(
 	}
 
 	/*
+	 * If this dquot hasn't been dirtied since initiating the last dqflush,
+	 * release the buffer reference.  We already unlinked this dquot item
+	 * from the buffer.
+	 */
+	spin_lock(&qlip->qli_lock);
+	if (!qlip->qli_dirty) {
+		bp = lip->li_buf;
+		lip->li_buf = NULL;
+	}
+	spin_unlock(&qlip->qli_lock);
+	if (bp)
+		xfs_buf_rele(bp);
+
+	/*
 	 * Release the dq's flush lock since we're done with it.
 	 */
 	xfs_dqfunlock(dqp);
@@ -1188,18 +1204,6 @@ xfs_buf_dquot_iodone(
 	}
 }
 
-void
-xfs_buf_dquot_io_fail(
-	struct xfs_buf		*bp)
-{
-	struct xfs_log_item	*lip;
-
-	spin_lock(&bp->b_mount->m_ail->ail_lock);
-	list_for_each_entry(lip, &bp->b_li_list, li_bio_list)
-		xfs_set_li_failed(lip, bp);
-	spin_unlock(&bp->b_mount->m_ail->ail_lock);
-}
-
 /* Check incore dquot for errors before we flush. */
 static xfs_failaddr_t
 xfs_qm_dqflush_check(
@@ -1239,6 +1243,115 @@ xfs_qm_dqflush_check(
 }
 
 /*
+ * Get the buffer containing the on-disk dquot.
+ *
+ * Requires dquot flush lock, will clear the dirty flag, delete the quota log
+ * item from the AIL, and shut down the system if something goes wrong.
+ */
+static int
+xfs_dquot_read_buf(
+	struct xfs_trans	*tp,
+	struct xfs_dquot	*dqp,
+	struct xfs_buf		**bpp)
+{
+	struct xfs_mount	*mp = dqp->q_mount;
+	struct xfs_buf		*bp = NULL;
+	int			error;
+
+	error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, dqp->q_blkno,
+				   mp->m_quotainfo->qi_dqchunklen, 0,
+				   &bp, &xfs_dquot_buf_ops);
+	if (xfs_metadata_is_sick(error))
+		xfs_dquot_mark_sick(dqp);
+	if (error)
+		goto out_abort;
+
+	*bpp = bp;
+	return 0;
+
+out_abort:
+	dqp->q_flags &= ~XFS_DQFLAG_DIRTY;
+	xfs_trans_ail_delete(&dqp->q_logitem.qli_item, 0);
+	xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
+	return error;
+}
+
+/*
+ * Attach a dquot buffer to this dquot to avoid allocating a buffer during a
+ * dqflush, since dqflush can be called from reclaim context.  Caller must hold
+ * the dqlock.
+ */
+int
+xfs_dquot_attach_buf(
+	struct xfs_trans	*tp,
+	struct xfs_dquot	*dqp)
+{
+	struct xfs_dq_logitem	*qlip = &dqp->q_logitem;
+	struct xfs_log_item	*lip = &qlip->qli_item;
+	int			error;
+
+	spin_lock(&qlip->qli_lock);
+	if (!lip->li_buf) {
+		struct xfs_buf	*bp = NULL;
+
+		spin_unlock(&qlip->qli_lock);
+		error = xfs_dquot_read_buf(tp, dqp, &bp);
+		if (error)
+			return error;
+
+		/*
+		 * Hold the dquot buffer so that we retain our ref to it after
+		 * detaching it from the transaction, then give that ref to the
+		 * dquot log item so that the AIL does not have to read the
+		 * dquot buffer to push this item.
+		 */
+		xfs_buf_hold(bp);
+		xfs_trans_brelse(tp, bp);
+
+		spin_lock(&qlip->qli_lock);
+		lip->li_buf = bp;
+	}
+	qlip->qli_dirty = true;
+	spin_unlock(&qlip->qli_lock);
+
+	return 0;
+}
+
+/*
+ * Get a new reference the dquot buffer attached to this dquot for a dqflush
+ * operation.
+ *
+ * Returns 0 and a NULL bp if none was attached to the dquot; 0 and a locked
+ * bp; or -EAGAIN if the buffer could not be locked.
+ */
+int
+xfs_dquot_use_attached_buf(
+	struct xfs_dquot	*dqp,
+	struct xfs_buf		**bpp)
+{
+	struct xfs_buf		*bp = dqp->q_logitem.qli_item.li_buf;
+
+	/*
+	 * A NULL buffer can happen if the dquot dirty flag was set but the
+	 * filesystem shut down before transaction commit happened.  In that
+	 * case we're not going to flush anyway.
+	 */
+	if (!bp) {
+		ASSERT(xfs_is_shutdown(dqp->q_mount));
+
+		*bpp = NULL;
+		return 0;
+	}
+
+	if (!xfs_buf_trylock(bp))
+		return -EAGAIN;
+
+	xfs_buf_hold(bp);
+	*bpp = bp;
+	return 0;
+}
+
+/*
  * Write a modified dquot to disk.
  * The dquot must be locked and the flush lock too taken by caller.
  * The flush lock will not be unlocked until the dquot reaches the disk,
@@ -1249,42 +1362,24 @@ xfs_qm_dqflush_check(
 int
 xfs_qm_dqflush(
 	struct xfs_dquot	*dqp,
-	struct xfs_buf		**bpp)
+	struct xfs_buf		*bp)
 {
 	struct xfs_mount	*mp = dqp->q_mount;
-	struct xfs_log_item	*lip = &dqp->q_logitem.qli_item;
-	struct xfs_buf		*bp;
+	struct xfs_dq_logitem	*qlip = &dqp->q_logitem;
+	struct xfs_log_item	*lip = &qlip->qli_item;
 	struct xfs_dqblk	*dqblk;
 	xfs_failaddr_t		fa;
 	int			error;
 
 	ASSERT(XFS_DQ_IS_LOCKED(dqp));
 	ASSERT(!completion_done(&dqp->q_flush));
+	ASSERT(atomic_read(&dqp->q_pincount) == 0);
 
 	trace_xfs_dqflush(dqp);
-
-	*bpp = NULL;
-
-	xfs_qm_dqunpin_wait(dqp);
-
-	/*
-	 * Get the buffer containing the on-disk dquot
-	 */
-	error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dqp->q_blkno,
-				   mp->m_quotainfo->qi_dqchunklen, XBF_TRYLOCK,
-				   &bp, &xfs_dquot_buf_ops);
-	if (error == -EAGAIN)
-		goto out_unlock;
-	if (xfs_metadata_is_sick(error))
-		xfs_dquot_mark_sick(dqp);
-	if (error)
-		goto out_abort;
-
 	fa = xfs_qm_dqflush_check(dqp);
 	if (fa) {
 		xfs_alert(mp, "corrupt dquot ID 0x%x in memory at %pS",
 				dqp->q_id, fa);
-		xfs_buf_relse(bp);
 		xfs_dquot_mark_sick(dqp);
 		error = -EFSCORRUPTED;
 		goto out_abort;
@@ -1299,8 +1394,15 @@ xfs_qm_dqflush(
 	 */
 	dqp->q_flags &= ~XFS_DQFLAG_DIRTY;
 
-	xfs_trans_ail_copy_lsn(mp->m_ail, &dqp->q_logitem.qli_flush_lsn,
-					&dqp->q_logitem.qli_item.li_lsn);
+	/*
+	 * We hold the dquot lock, so nobody can dirty it while we're
+	 * scheduling the write out.  Clear the dirty-since-flush flag.
+	 */
+	spin_lock(&qlip->qli_lock);
+	qlip->qli_dirty = false;
+	spin_unlock(&qlip->qli_lock);
+
+	xfs_trans_ail_copy_lsn(mp->m_ail, &qlip->qli_flush_lsn, &lip->li_lsn);
 
 	/*
 	 * copy the lsn into the on-disk dquot now while we have the in memory
@@ -1312,7 +1414,7 @@ xfs_qm_dqflush(
 	 * of a dquot without an up-to-date CRC getting to disk.
 	 */
 	if (xfs_has_crc(mp)) {
-		dqblk->dd_lsn = cpu_to_be64(dqp->q_logitem.qli_item.li_lsn);
+		dqblk->dd_lsn = cpu_to_be64(lip->li_lsn);
 		xfs_update_cksum((char *)dqblk, sizeof(struct xfs_dqblk),
 				 XFS_DQUOT_CRC_OFF);
 	}
@@ -1321,8 +1423,8 @@ xfs_qm_dqflush(
 	 * Attach the dquot to the buffer so that we can remove this dquot from
 	 * the AIL and release the flush lock once the dquot is synced to disk.
 	 */
-	bp->b_flags |= _XBF_DQUOTS;
-	list_add_tail(&dqp->q_logitem.qli_item.li_bio_list, &bp->b_li_list);
+	bp->b_iodone = xfs_buf_dquot_iodone;
+	list_add_tail(&lip->li_bio_list, &bp->b_li_list);
 
 	/*
 	 * If the buffer is pinned then push on the log so we won't
@@ -1334,14 +1436,12 @@ xfs_qm_dqflush(
 	}
 
 	trace_xfs_dqflush_done(dqp);
-	*bpp = bp;
 	return 0;
 
 out_abort:
 	dqp->q_flags &= ~XFS_DQFLAG_DIRTY;
 	xfs_trans_ail_delete(lip, 0);
 	xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
-out_unlock:
 	xfs_dqfunlock(dqp);
 	return error;
 }