Merge branch 'for-linus' of git://oss.sgi.com/xfs/xfs

Pull XFS updates from Ben Myers:
 "Scalability improvements for dquots, log grant code cleanups, plus
  bugfixes and cleanups large and small"

Fix up various trivial conflicts that were due to some of the earlier
patches already having been integrated into v3.3 as bugfixes, and then
there were development patches on top of those.  Easily merged by just
taking the newer version from the pulled branch.

* 'for-linus' of git://oss.sgi.com/xfs/xfs: (45 commits)
  xfs: fallback to vmalloc for large buffers in xfs_getbmap
  xfs: fallback to vmalloc for large buffers in xfs_attrmulti_attr_get
  xfs: remove remaining scraps of struct xfs_iomap
  xfs: fix inode lookup race
  xfs: clean up minor sparse warnings
  xfs: remove the global xfs_Gqm structure
  xfs: remove the per-filesystem list of dquots
  xfs: use per-filesystem radix trees for dquot lookup
  xfs: per-filesystem dquot LRU lists
  xfs: use common code for quota statistics
  xfs: reimplement fdatasync support
  xfs: split in-core and on-disk inode log item fields
  xfs: make xfs_inode_item_size idempotent
  xfs: log timestamp updates
  xfs: log file size updates at I/O completion time
  xfs: log file size updates as part of unwritten extent conversion
  xfs: do not require an ioend for new EOF calculation
  xfs: use per-filesystem I/O completion workqueues
  quota: make Q_XQUOTASYNC a noop
  xfs: include reservations in quota reporting
  ...

1 files changed, 124 insertions, 294 deletions

diff --git a/fs/xfs/xfs_dquot.c b/fs/xfs/xfs_dquot.c
index 53db20ee3e77..4be16a0cbe5a 100644
--- a/fs/xfs/xfs_dquot.c
+++ b/fs/xfs/xfs_dquot.c
@@ -43,11 +43,10 @@
  * Lock order:
  *
  * ip->i_lock
- *   qh->qh_lock
- *     qi->qi_dqlist_lock
- *       dquot->q_qlock (xfs_dqlock() and friends)
- *         dquot->q_flush (xfs_dqflock() and friends)
- *         xfs_Gqm->qm_dqfrlist_lock
+ *   qi->qi_tree_lock
+ *     dquot->q_qlock (xfs_dqlock() and friends)
+ *       dquot->q_flush (xfs_dqflock() and friends)
+ *       qi->qi_lru_lock
  *
  * If two dquots need to be locked the order is user before group/project,
  * otherwise by the lowest id first, see xfs_dqlock2.
@@ -60,6 +59,9 @@ int xfs_dqreq_num;
 int xfs_dqerror_mod = 33;
 #endif
 
+struct kmem_zone		*xfs_qm_dqtrxzone;
+static struct kmem_zone		*xfs_qm_dqzone;
+
 static struct lock_class_key xfs_dquot_other_class;
 
 /*
@@ -69,12 +71,12 @@ void
 xfs_qm_dqdestroy(
 	xfs_dquot_t	*dqp)
 {
-	ASSERT(list_empty(&dqp->q_freelist));
+	ASSERT(list_empty(&dqp->q_lru));
 
 	mutex_destroy(&dqp->q_qlock);
-	kmem_zone_free(xfs_Gqm->qm_dqzone, dqp);
+	kmem_zone_free(xfs_qm_dqzone, dqp);
 
-	atomic_dec(&xfs_Gqm->qm_totaldquots);
+	XFS_STATS_DEC(xs_qm_dquot);
 }
 
 /*
@@ -282,7 +284,7 @@ xfs_qm_dqalloc(
 	 * Return if this type of quotas is turned off while we didn't
 	 * have an inode lock
 	 */
-	if (XFS_IS_THIS_QUOTA_OFF(dqp)) {
+	if (!xfs_this_quota_on(dqp->q_mount, dqp->dq_flags)) {
 		xfs_iunlock(quotip, XFS_ILOCK_EXCL);
 		return (ESRCH);
 	}
@@ -384,7 +386,7 @@ xfs_qm_dqtobp(
 	dqp->q_fileoffset = (xfs_fileoff_t)id / mp->m_quotainfo->qi_dqperchunk;
 
 	xfs_ilock(quotip, XFS_ILOCK_SHARED);
-	if (XFS_IS_THIS_QUOTA_OFF(dqp)) {
+	if (!xfs_this_quota_on(dqp->q_mount, dqp->dq_flags)) {
 		/*
 		 * Return if this type of quotas is turned off while we
 		 * didn't have the quota inode lock.
@@ -492,12 +494,12 @@ xfs_qm_dqread(
 	int			cancelflags = 0;
 
 
-	dqp = kmem_zone_zalloc(xfs_Gqm->qm_dqzone, KM_SLEEP);
+	dqp = kmem_zone_zalloc(xfs_qm_dqzone, KM_SLEEP);
 
 	dqp->dq_flags = type;
 	dqp->q_core.d_id = cpu_to_be32(id);
 	dqp->q_mount = mp;
-	INIT_LIST_HEAD(&dqp->q_freelist);
+	INIT_LIST_HEAD(&dqp->q_lru);
 	mutex_init(&dqp->q_qlock);
 	init_waitqueue_head(&dqp->q_pinwait);
 
@@ -516,7 +518,7 @@ xfs_qm_dqread(
 	if (!(type & XFS_DQ_USER))
 		lockdep_set_class(&dqp->q_qlock, &xfs_dquot_other_class);
 
-	atomic_inc(&xfs_Gqm->qm_totaldquots);
+	XFS_STATS_INC(xs_qm_dquot);
 
 	trace_xfs_dqread(dqp);
 
@@ -602,60 +604,6 @@ error0:
 }
 
 /*
- * Lookup a dquot in the incore dquot hashtable. We keep two separate
- * hashtables for user and group dquots; and, these are global tables
- * inside the XQM, not per-filesystem tables.
- * The hash chain must be locked by caller, and it is left locked
- * on return. Returning dquot is locked.
- */
-STATIC int
-xfs_qm_dqlookup(
-	xfs_mount_t		*mp,
-	xfs_dqid_t		id,
-	xfs_dqhash_t		*qh,
-	xfs_dquot_t		**O_dqpp)
-{
-	xfs_dquot_t		*dqp;
-
-	ASSERT(mutex_is_locked(&qh->qh_lock));
-
-	/*
-	 * Traverse the hashchain looking for a match
-	 */
-	list_for_each_entry(dqp, &qh->qh_list, q_hashlist) {
-		/*
-		 * We already have the hashlock. We don't need the
-		 * dqlock to look at the id field of the dquot, since the
-		 * id can't be modified without the hashlock anyway.
-		 */
-		if (be32_to_cpu(dqp->q_core.d_id) != id || dqp->q_mount != mp)
-			continue;
-
-		trace_xfs_dqlookup_found(dqp);
-
-		xfs_dqlock(dqp);
-		if (dqp->dq_flags & XFS_DQ_FREEING) {
-			*O_dqpp = NULL;
-			xfs_dqunlock(dqp);
-			return -1;
-		}
-
-		dqp->q_nrefs++;
-
-		/*
-		 * move the dquot to the front of the hashchain
-		 */
-		list_move(&dqp->q_hashlist, &qh->qh_list);
-		trace_xfs_dqlookup_done(dqp);
-		*O_dqpp = dqp;
-		return 0;
-	}
-
-	*O_dqpp = NULL;
-	return 1;
-}
-
-/*
  * Given the file system, inode OR id, and type (UDQUOT/GDQUOT), return a
  * a locked dquot, doing an allocation (if requested) as needed.
  * When both an inode and an id are given, the inode's id takes precedence.
@@ -672,10 +620,10 @@ xfs_qm_dqget(
 	uint		flags,	  /* DQALLOC, DQSUSER, DQREPAIR, DOWARN */
 	xfs_dquot_t	**O_dqpp) /* OUT : locked incore dquot */
 {
-	xfs_dquot_t	*dqp;
-	xfs_dqhash_t	*h;
-	uint		version;
-	int		error;
+	struct xfs_quotainfo	*qi = mp->m_quotainfo;
+	struct radix_tree_root *tree = XFS_DQUOT_TREE(qi, type);
+	struct xfs_dquot	*dqp;
+	int			error;
 
 	ASSERT(XFS_IS_QUOTA_RUNNING(mp));
 	if ((! XFS_IS_UQUOTA_ON(mp) && type == XFS_DQ_USER) ||
@@ -683,7 +631,6 @@ xfs_qm_dqget(
 	    (! XFS_IS_GQUOTA_ON(mp) && type == XFS_DQ_GROUP)) {
 		return (ESRCH);
 	}
-	h = XFS_DQ_HASH(mp, id, type);
 
 #ifdef DEBUG
 	if (xfs_do_dqerror) {
@@ -699,42 +646,33 @@ xfs_qm_dqget(
 	       type == XFS_DQ_GROUP);
 	if (ip) {
 		ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
-		if (type == XFS_DQ_USER)
-			ASSERT(ip->i_udquot == NULL);
-		else
-			ASSERT(ip->i_gdquot == NULL);
+		ASSERT(xfs_inode_dquot(ip, type) == NULL);
 	}
 #endif
 
 restart:
-	mutex_lock(&h->qh_lock);
+	mutex_lock(&qi->qi_tree_lock);
+	dqp = radix_tree_lookup(tree, id);
+	if (dqp) {
+		xfs_dqlock(dqp);
+		if (dqp->dq_flags & XFS_DQ_FREEING) {
+			xfs_dqunlock(dqp);
+			mutex_unlock(&qi->qi_tree_lock);
+			trace_xfs_dqget_freeing(dqp);
+			delay(1);
+			goto restart;
+		}
 
-	/*
-	 * Look in the cache (hashtable).
-	 * The chain is kept locked during lookup.
-	 */
-	switch (xfs_qm_dqlookup(mp, id, h, O_dqpp)) {
-	case -1:
-		XQM_STATS_INC(xqmstats.xs_qm_dquot_dups);
-		mutex_unlock(&h->qh_lock);
-		delay(1);
-		goto restart;
-	case 0:
-		XQM_STATS_INC(xqmstats.xs_qm_dqcachehits);
-		/*
-		 * The dquot was found, moved to the front of the chain,
-		 * taken off the freelist if it was on it, and locked
-		 * at this point. Just unlock the hashchain and return.
-		 */
-		ASSERT(*O_dqpp);
-		ASSERT(XFS_DQ_IS_LOCKED(*O_dqpp));
-		mutex_unlock(&h->qh_lock);
-		trace_xfs_dqget_hit(*O_dqpp);
-		return 0;	/* success */
-	default:
-		XQM_STATS_INC(xqmstats.xs_qm_dqcachemisses);
-		break;
+		dqp->q_nrefs++;
+		mutex_unlock(&qi->qi_tree_lock);
+
+		trace_xfs_dqget_hit(dqp);
+		XFS_STATS_INC(xs_qm_dqcachehits);
+		*O_dqpp = dqp;
+		return 0;
 	}
+	mutex_unlock(&qi->qi_tree_lock);
+	XFS_STATS_INC(xs_qm_dqcachemisses);
 
 	/*
 	 * Dquot cache miss. We don't want to keep the inode lock across
@@ -745,12 +683,6 @@ restart:
 	 */
 	if (ip)
 		xfs_iunlock(ip, XFS_ILOCK_EXCL);
-	/*
-	 * Save the hashchain version stamp, and unlock the chain, so that
-	 * we don't keep the lock across a disk read
-	 */
-	version = h->qh_version;
-	mutex_unlock(&h->qh_lock);
 
 	error = xfs_qm_dqread(mp, id, type, flags, &dqp);
 
@@ -760,97 +692,53 @@ restart:
 	if (error)
 		return error;
 
-	/*
-	 * Dquot lock comes after hashlock in the lock ordering
-	 */
 	if (ip) {
 		/*
 		 * A dquot could be attached to this inode by now, since
 		 * we had dropped the ilock.
 		 */
-		if (type == XFS_DQ_USER) {
-			if (!XFS_IS_UQUOTA_ON(mp)) {
-				/* inode stays locked on return */
-				xfs_qm_dqdestroy(dqp);
-				return XFS_ERROR(ESRCH);
-			}
-			if (ip->i_udquot) {
+		if (xfs_this_quota_on(mp, type)) {
+			struct xfs_dquot	*dqp1;
+
+			dqp1 = xfs_inode_dquot(ip, type);
+			if (dqp1) {
 				xfs_qm_dqdestroy(dqp);
-				dqp = ip->i_udquot;
+				dqp = dqp1;
 				xfs_dqlock(dqp);
 				goto dqret;
 			}
 		} else {
-			if (!XFS_IS_OQUOTA_ON(mp)) {
-				/* inode stays locked on return */
-				xfs_qm_dqdestroy(dqp);
-				return XFS_ERROR(ESRCH);
-			}
-			if (ip->i_gdquot) {
-				xfs_qm_dqdestroy(dqp);
-				dqp = ip->i_gdquot;
-				xfs_dqlock(dqp);
-				goto dqret;
-			}
+			/* inode stays locked on return */
+			xfs_qm_dqdestroy(dqp);
+			return XFS_ERROR(ESRCH);
 		}
 	}
 
-	/*
-	 * Hashlock comes after ilock in lock order
-	 */
-	mutex_lock(&h->qh_lock);
-	if (version != h->qh_version) {
-		xfs_dquot_t *tmpdqp;
+	mutex_lock(&qi->qi_tree_lock);
+	error = -radix_tree_insert(tree, id, dqp);
+	if (unlikely(error)) {
+		WARN_ON(error != EEXIST);
+
 		/*
-		 * Now, see if somebody else put the dquot in the
-		 * hashtable before us. This can happen because we didn't
-		 * keep the hashchain lock. We don't have to worry about
-		 * lock order between the two dquots here since dqp isn't
-		 * on any findable lists yet.
+		 * Duplicate found. Just throw away the new dquot and start
+		 * over.
 		 */
-		switch (xfs_qm_dqlookup(mp, id, h, &tmpdqp)) {
-		case 0:
-		case -1:
-			/*
-			 * Duplicate found, either in cache or on its way out.
-			 * Just throw away the new dquot and start over.
-			 */
-			if (tmpdqp)
-				xfs_qm_dqput(tmpdqp);
-			mutex_unlock(&h->qh_lock);
-			xfs_qm_dqdestroy(dqp);
-			XQM_STATS_INC(xqmstats.xs_qm_dquot_dups);
-			goto restart;
-		default:
-			break;
-		}
+		mutex_unlock(&qi->qi_tree_lock);
+		trace_xfs_dqget_dup(dqp);
+		xfs_qm_dqdestroy(dqp);
+		XFS_STATS_INC(xs_qm_dquot_dups);
+		goto restart;
 	}
 
 	/*
-	 * Put the dquot at the beginning of the hash-chain and mp's list
-	 * LOCK ORDER: hashlock, freelistlock, mplistlock, udqlock, gdqlock ..
-	 */
-	ASSERT(mutex_is_locked(&h->qh_lock));
-	dqp->q_hash = h;
-	list_add(&dqp->q_hashlist, &h->qh_list);
-	h->qh_version++;
-
-	/*
-	 * Attach this dquot to this filesystem's list of all dquots,
-	 * kept inside the mount structure in m_quotainfo field
-	 */
-	mutex_lock(&mp->m_quotainfo->qi_dqlist_lock);
-
-	/*
 	 * We return a locked dquot to the caller, with a reference taken
 	 */
 	xfs_dqlock(dqp);
 	dqp->q_nrefs = 1;
 
-	list_add(&dqp->q_mplist, &mp->m_quotainfo->qi_dqlist);
-	mp->m_quotainfo->qi_dquots++;
-	mutex_unlock(&mp->m_quotainfo->qi_dqlist_lock);
-	mutex_unlock(&h->qh_lock);
+	qi->qi_dquots++;
+	mutex_unlock(&qi->qi_tree_lock);
+
  dqret:
 	ASSERT((ip == NULL) || xfs_isilocked(ip, XFS_ILOCK_EXCL));
 	trace_xfs_dqget_miss(dqp);
@@ -859,37 +747,22 @@ restart:
 }
 
 
-/*
- * 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.
- */
-void
-xfs_qm_dqput(
+STATIC void
+xfs_qm_dqput_final(
 	struct xfs_dquot	*dqp)
 {
+	struct xfs_quotainfo	*qi = dqp->q_mount->m_quotainfo;
 	struct xfs_dquot	*gdqp;
 
-	ASSERT(dqp->q_nrefs > 0);
-	ASSERT(XFS_DQ_IS_LOCKED(dqp));
-
-	trace_xfs_dqput(dqp);
-
-recurse:
-	if (--dqp->q_nrefs > 0) {
-		xfs_dqunlock(dqp);
-		return;
-	}
-
 	trace_xfs_dqput_free(dqp);
 
-	mutex_lock(&xfs_Gqm->qm_dqfrlist_lock);
-	if (list_empty(&dqp->q_freelist)) {
-		list_add_tail(&dqp->q_freelist, &xfs_Gqm->qm_dqfrlist);
-		xfs_Gqm->qm_dqfrlist_cnt++;
+	mutex_lock(&qi->qi_lru_lock);
+	if (list_empty(&dqp->q_lru)) {
+		list_add_tail(&dqp->q_lru, &qi->qi_lru_list);
+		qi->qi_lru_count++;
+		XFS_STATS_INC(xs_qm_dquot_unused);
 	}
-	mutex_unlock(&xfs_Gqm->qm_dqfrlist_lock);
+	mutex_unlock(&qi->qi_lru_lock);
 
 	/*
 	 * If we just added a udquot to the freelist, then we want to release
@@ -906,10 +779,29 @@ recurse:
 	/*
 	 * If we had a group quota hint, release it now.
 	 */
-	if (gdqp) {
-		dqp = gdqp;
-		goto recurse;
-	}
+	if (gdqp)
+		xfs_qm_dqput(gdqp);
+}
+
+/*
+ * 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.
+ */
+void
+xfs_qm_dqput(
+	struct xfs_dquot	*dqp)
+{
+	ASSERT(dqp->q_nrefs > 0);
+	ASSERT(XFS_DQ_IS_LOCKED(dqp));
+
+	trace_xfs_dqput(dqp);
+
+	if (--dqp->q_nrefs > 0)
+		xfs_dqunlock(dqp);
+	else
+		xfs_qm_dqput_final(dqp);
 }
 
 /*
@@ -1091,17 +983,6 @@ xfs_qm_dqflush(
 
 }
 
-void
-xfs_dqunlock(
-	xfs_dquot_t *dqp)
-{
-	xfs_dqunlock_nonotify(dqp);
-	if (dqp->q_logitem.qli_dquot == dqp) {
-		xfs_trans_unlocked_item(dqp->q_logitem.qli_item.li_ailp,
-					&dqp->q_logitem.qli_item);
-	}
-}
-
 /*
  * Lock two xfs_dquot structures.
  *
@@ -1131,85 +1012,6 @@ xfs_dqlock2(
 }
 
 /*
- * Take a dquot out of the mount's dqlist as well as the hashlist.  This is
- * called via unmount as well as quotaoff, and the purge will always succeed.
- */
-void
-xfs_qm_dqpurge(
-	struct xfs_dquot	*dqp)
-{
-	struct xfs_mount	*mp = dqp->q_mount;
-	struct xfs_dqhash	*qh = dqp->q_hash;
-
-	xfs_dqlock(dqp);
-
-	/*
-	 * If we're turning off quotas, we have to make sure that, for
-	 * example, we don't delete quota disk blocks while dquots are
-	 * in the process of getting written to those disk blocks.
-	 * This dquot might well be on AIL, and we can't leave it there
-	 * if we're turning off quotas. Basically, we need this flush
-	 * lock, and are willing to block on it.
-	 */
-	if (!xfs_dqflock_nowait(dqp)) {
-		/*
-		 * Block on the flush lock after nudging dquot buffer,
-		 * if it is incore.
-		 */
-		xfs_dqflock_pushbuf_wait(dqp);
-	}
-
-	/*
-	 * If we are turning this type of quotas off, we don't care
-	 * about the dirty metadata sitting in this dquot. OTOH, if
-	 * we're unmounting, we do care, so we flush it and wait.
-	 */
-	if (XFS_DQ_IS_DIRTY(dqp)) {
-		int	error;
-
-		/*
-		 * We don't care about getting disk errors here. We need
-		 * to purge this dquot anyway, so we go ahead regardless.
-		 */
-		error = xfs_qm_dqflush(dqp, SYNC_WAIT);
-		if (error)
-			xfs_warn(mp, "%s: dquot %p flush failed",
-				__func__, dqp);
-		xfs_dqflock(dqp);
-	}
-
-	ASSERT(atomic_read(&dqp->q_pincount) == 0);
-	ASSERT(XFS_FORCED_SHUTDOWN(mp) ||
-	       !(dqp->q_logitem.qli_item.li_flags & XFS_LI_IN_AIL));
-
-	xfs_dqfunlock(dqp);
-	xfs_dqunlock(dqp);
-
-	mutex_lock(&qh->qh_lock);
-	list_del_init(&dqp->q_hashlist);
-	qh->qh_version++;
-	mutex_unlock(&qh->qh_lock);
-
-	mutex_lock(&mp->m_quotainfo->qi_dqlist_lock);
-	list_del_init(&dqp->q_mplist);
-	mp->m_quotainfo->qi_dqreclaims++;
-	mp->m_quotainfo->qi_dquots--;
-	mutex_unlock(&mp->m_quotainfo->qi_dqlist_lock);
-
-	/*
-	 * We move dquots to the freelist as soon as their reference count
-	 * hits zero, so it really should be on the freelist here.
-	 */
-	mutex_lock(&xfs_Gqm->qm_dqfrlist_lock);
-	ASSERT(!list_empty(&dqp->q_freelist));
-	list_del_init(&dqp->q_freelist);
-	xfs_Gqm->qm_dqfrlist_cnt--;
-	mutex_unlock(&xfs_Gqm->qm_dqfrlist_lock);
-
-	xfs_qm_dqdestroy(dqp);
-}
-
-/*
  * Give the buffer a little push if it is incore and
  * wait on the flush lock.
  */
@@ -1241,3 +1043,31 @@ xfs_dqflock_pushbuf_wait(
 out_lock:
 	xfs_dqflock(dqp);
 }
+
+int __init
+xfs_qm_init(void)
+{
+	xfs_qm_dqzone =
+		kmem_zone_init(sizeof(struct xfs_dquot), "xfs_dquot");
+	if (!xfs_qm_dqzone)
+		goto out;
+
+	xfs_qm_dqtrxzone =
+		kmem_zone_init(sizeof(struct xfs_dquot_acct), "xfs_dqtrx");
+	if (!xfs_qm_dqtrxzone)
+		goto out_free_dqzone;
+
+	return 0;
+
+out_free_dqzone:
+	kmem_zone_destroy(xfs_qm_dqzone);
+out:
+	return -ENOMEM;
+}
+
+void __exit
+xfs_qm_exit(void)
+{
+	kmem_zone_destroy(xfs_qm_dqtrxzone);
+	kmem_zone_destroy(xfs_qm_dqzone);
+}