xfs: parallelize inode inactivation

Split the inode inactivation work into per-AG work items so that we can
take advantage of parallelization.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>

6 files changed, 78 insertions, 16 deletions

diff --git a/fs/xfs/scrub/common.c b/fs/xfs/scrub/common.c
index f4c6f5062d8c..c8aa42e13692 100644
--- a/fs/xfs/scrub/common.c
+++ b/fs/xfs/scrub/common.c
@@ -911,6 +911,7 @@ xchk_stop_reaping(
 {
 	sc->flags |= XCHK_REAPING_DISABLED;
 	xfs_blockgc_stop(sc->mp);
+	xfs_inactive_cancel_work(sc->mp);
 }
 
 /* Restart background reaping of resources. */
@@ -918,6 +919,7 @@ void
 xchk_start_reaping(
 	struct xfs_scrub	*sc)
 {
+	xfs_inactive_schedule_now(sc->mp);
 	xfs_blockgc_start(sc->mp);
 	sc->flags &= ~XCHK_REAPING_DISABLED;
 }
diff --git a/fs/xfs/xfs_icache.c b/fs/xfs/xfs_icache.c
index 01966f575387..999b8eee02c2 100644
--- a/fs/xfs/xfs_icache.c
+++ b/fs/xfs/xfs_icache.c
@@ -2094,12 +2094,12 @@ xfs_inode_clear_cowblocks_tag(
 /* Queue a new inode inactivation pass if there are reclaimable inodes. */
 static void
 xfs_inactive_work_queue(
-	struct xfs_mount        *mp)
+	struct xfs_perag	*pag)
 {
 	rcu_read_lock();
-	if (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_INACTIVE_TAG))
-		queue_delayed_work(mp->m_inactive_workqueue,
-				&mp->m_inactive_work,
+	if (radix_tree_tagged(&pag->pag_ici_root, XFS_ICI_INACTIVE_TAG))
+		queue_delayed_work(pag->pag_mount->m_inactive_workqueue,
+				&pag->pag_inactive_work,
 				msecs_to_jiffies(xfs_syncd_centisecs / 6 * 10));
 	rcu_read_unlock();
 }
@@ -2122,7 +2122,7 @@ xfs_perag_set_inactive_tag(
 	spin_unlock(&mp->m_perag_lock);
 
 	/* schedule periodic background inode inactivation */
-	xfs_inactive_work_queue(mp);
+	xfs_inactive_work_queue(pag);
 
 	trace_xfs_perag_set_inactive(mp, pag->pag_agno, -1, _RET_IP_);
 }
@@ -2255,8 +2255,9 @@ void
 xfs_inactive_worker(
 	struct work_struct	*work)
 {
-	struct xfs_mount	*mp = container_of(to_delayed_work(work),
-					struct xfs_mount, m_inactive_work);
+	struct xfs_perag	*pag = container_of(to_delayed_work(work),
+					struct xfs_perag, pag_inactive_work);
+	struct xfs_mount	*mp = pag->pag_mount;
 	int			error;
 
 	/*
@@ -2271,12 +2272,32 @@ xfs_inactive_worker(
 	if (!sb_start_write_trylock(mp->m_super))
 		return;
 
-	error = xfs_inactive_inodes(mp, NULL);
+	error = xfs_inode_walk_ag(pag, XFS_INODE_WALK_INACTIVE,
+			xfs_inactive_inode, NULL, XFS_ICI_INACTIVE_TAG);
 	if (error && error != -EAGAIN)
 		xfs_err(mp, "inode inactivation failed, error %d", error);
 
 	sb_end_write(mp->m_super);
-	xfs_inactive_work_queue(mp);
+	xfs_inactive_work_queue(pag);
+}
+
+/* Wait for all background inactivation work to finish. */
+static void
+xfs_inactive_flush(
+	struct xfs_mount	*mp)
+{
+	struct xfs_perag	*pag;
+	xfs_agnumber_t		agno;
+
+	for_each_perag_tag(mp, agno, pag, XFS_ICI_INACTIVE_TAG) {
+		bool		flush;
+
+		spin_lock(&pag->pag_ici_lock);
+		flush = pag->pag_ici_inactive > 0;
+		spin_unlock(&pag->pag_ici_lock);
+		if (flush)
+			flush_delayed_work(&pag->pag_inactive_work);
+	}
 }
 
 /* Flush all inode inactivation work that might be queued. */
@@ -2284,8 +2305,8 @@ void
 xfs_inactive_force(
 	struct xfs_mount	*mp)
 {
-	queue_delayed_work(mp->m_inactive_workqueue, &mp->m_inactive_work, 0);
-	flush_delayed_work(&mp->m_inactive_work);
+	xfs_inactive_schedule_now(mp);
+	xfs_inactive_flush(mp);
 }
 
 /*
@@ -2297,9 +2318,40 @@ void
 xfs_inactive_shutdown(
 	struct xfs_mount	*mp)
 {
-	cancel_delayed_work_sync(&mp->m_inactive_work);
-	flush_workqueue(mp->m_inactive_workqueue);
+	xfs_inactive_cancel_work(mp);
 	xfs_inactive_inodes(mp, NULL);
 	cancel_delayed_work_sync(&mp->m_reclaim_work);
 	xfs_reclaim_inodes(mp, SYNC_WAIT);
 }
+
+/* Cancel all queued inactivation work. */
+void
+xfs_inactive_cancel_work(
+	struct xfs_mount	*mp)
+{
+	struct xfs_perag	*pag;
+	xfs_agnumber_t		agno;
+
+	for_each_perag_tag(mp, agno, pag, XFS_ICI_INACTIVE_TAG)
+		cancel_delayed_work_sync(&pag->pag_inactive_work);
+	flush_workqueue(mp->m_inactive_workqueue);
+}
+
+/* Cancel all pending deferred inactivation work and reschedule it now. */
+void
+xfs_inactive_schedule_now(
+	struct xfs_mount	*mp)
+{
+	struct xfs_perag	*pag;
+	xfs_agnumber_t		agno;
+
+	for_each_perag_tag(mp, agno, pag, XFS_ICI_INACTIVE_TAG) {
+		spin_lock(&pag->pag_ici_lock);
+		if (pag->pag_ici_inactive) {
+			cancel_delayed_work(&pag->pag_inactive_work);
+			queue_delayed_work(mp->m_inactive_workqueue,
+					&pag->pag_inactive_work, 0);
+		}
+		spin_unlock(&pag->pag_ici_lock);
+	}
+}
diff --git a/fs/xfs/xfs_icache.h b/fs/xfs/xfs_icache.h
index 15271ba87144..5850bf885452 100644
--- a/fs/xfs/xfs_icache.h
+++ b/fs/xfs/xfs_icache.h
@@ -81,5 +81,7 @@ void xfs_inactive_worker(struct work_struct *work);
 int xfs_inactive_inodes(struct xfs_mount *mp, struct xfs_eofblocks *eofb);
 void xfs_inactive_force(struct xfs_mount *mp);
 void xfs_inactive_shutdown(struct xfs_mount *mp);
+void xfs_inactive_cancel_work(struct xfs_mount *mp);
+void xfs_inactive_schedule_now(struct xfs_mount *mp);
 
 #endif
diff --git a/fs/xfs/xfs_mount.c b/fs/xfs/xfs_mount.c
index 1a34bf507c86..702a6541c376 100644
--- a/fs/xfs/xfs_mount.c
+++ b/fs/xfs/xfs_mount.c
@@ -127,6 +127,7 @@ __xfs_free_perag(
 	struct xfs_perag *pag = container_of(head, struct xfs_perag, rcu_head);
 
 	ASSERT(!delayed_work_pending(&pag->pag_blockgc_work));
+	ASSERT(!delayed_work_pending(&pag->pag_inactive_work));
 	ASSERT(atomic_read(&pag->pag_ref) == 0);
 	kmem_free(pag);
 }
@@ -148,6 +149,7 @@ xfs_free_perag(
 		ASSERT(pag);
 		ASSERT(atomic_read(&pag->pag_ref) == 0);
 		cancel_delayed_work_sync(&pag->pag_blockgc_work);
+		cancel_delayed_work_sync(&pag->pag_inactive_work);
 		xfs_iunlink_destroy(pag);
 		xfs_buf_hash_destroy(pag);
 		mutex_destroy(&pag->pag_ici_reclaim_lock);
@@ -204,6 +206,7 @@ xfs_initialize_perag(
 		spin_lock_init(&pag->pag_ici_lock);
 		mutex_init(&pag->pag_ici_reclaim_lock);
 		INIT_DELAYED_WORK(&pag->pag_blockgc_work, xfs_blockgc_worker);
+		INIT_DELAYED_WORK(&pag->pag_inactive_work, xfs_inactive_worker);
 		INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC);
 		if (xfs_buf_hash_init(pag))
 			goto out_free_pag;
diff --git a/fs/xfs/xfs_mount.h b/fs/xfs/xfs_mount.h
index f3405eae9cc1..c67cf932ef0c 100644
--- a/fs/xfs/xfs_mount.h
+++ b/fs/xfs/xfs_mount.h
@@ -162,7 +162,6 @@ typedef struct xfs_mount {
 	atomic_t		m_active_trans;	/* number trans frozen */
 	struct xfs_mru_cache	*m_filestream;  /* per-mount filestream data */
 	struct delayed_work	m_reclaim_work;	/* background inode reclaim */
-	struct delayed_work	m_inactive_work; /* background inode inactive */
 	bool			m_update_sb;	/* sb needs update in mount */
 	int64_t			m_low_space[XFS_LOWSP_MAX];
 						/* low free space thresholds */
@@ -366,6 +365,9 @@ typedef struct xfs_perag {
 	/* background prealloc block trimming */
 	struct delayed_work	pag_blockgc_work;
 
+	/* background inode inactivation */
+	struct delayed_work	pag_inactive_work;
+
 	/* reference count */
 	uint8_t			pagf_refcount_level;
 
diff --git a/fs/xfs/xfs_super.c b/fs/xfs/xfs_super.c
index 33764de2c76a..c021618f7588 100644
--- a/fs/xfs/xfs_super.c
+++ b/fs/xfs/xfs_super.c
@@ -490,7 +490,8 @@ xfs_init_mount_workqueues(
 		goto out_destroy_eofb;
 
 	mp->m_inactive_workqueue = alloc_workqueue("xfs-inactive/%s",
-			WQ_MEM_RECLAIM | WQ_FREEZABLE, 0, mp->m_super->s_id);
+			WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_FREEZABLE, 0,
+			mp->m_super->s_id);
 	if (!mp->m_inactive_workqueue)
 		goto out_destroy_sync;
 
@@ -1418,6 +1419,7 @@ xfs_configure_background_workqueues(
 
 	max_active = min_t(unsigned int, max_active, WQ_UNBOUND_MAX_ACTIVE);
 	workqueue_set_max_active(mp->m_blockgc_workqueue, max_active);
+	workqueue_set_max_active(mp->m_inactive_workqueue, max_active);
 }
 
 static int
@@ -1845,7 +1847,6 @@ static int xfs_init_fs_context(
 	atomic_set(&mp->m_active_trans, 0);
 	INIT_WORK(&mp->m_flush_inodes_work, xfs_flush_inodes_worker);
 	INIT_DELAYED_WORK(&mp->m_reclaim_work, xfs_reclaim_worker);
-	INIT_DELAYED_WORK(&mp->m_inactive_work, xfs_inactive_worker);
 	mp->m_kobj.kobject.kset = xfs_kset;
 	/*
 	 * We don't create the finobt per-ag space reservation until after log