OpenZFS 7968 - multi-threaded spa_sync()

Reviewed by: Pavel Zakharov <[email protected]> Reviewed by: Brad Lewis <[email protected]> Reviewed by: Saso Kiselkov <[email protected]> Reviewed by: Brian Behlendorf <[email protected]> Ported-by: Matthew Ahrens <[email protected]> spa_sync() iterates over all the dirty dnodes and processes each of them by calling dnode_sync(). If there are many dirty dnodes (e.g. because we created or removed a lot of files), the single thread of spa_sync() calling dnode_sync() can become a bottleneck. Additionally, if many dnodes are dirtied concurrently in open context (e.g. due to concurrent file creation), the os_lock will experience lock contention via dnode_setdirty(). The solution is to track dirty dnodes on a multilist_t, and for spa_sync() to use separate threads to process each of the sublists in the multilist. OpenZFS-issue: https://www.illumos.org/issues/7968 OpenZFS-commit: openzfs/openzfs@4a2a54c Closes openzfs#5752
pecker168 · Mar 21, 2017 · 64fc776 · 64fc776
1 parent a3478c0
commit 64fc776
Show file tree

Hide file tree

Showing 16 changed files with 303 additions and 159 deletions.
diff --git a/include/sys/arc_impl.h b/include/sys/arc_impl.h
@@ -70,7 +70,7 @@ typedef struct arc_state {
 	/*
 	 * list of evictable buffers
 	 */
-	multilist_t arcs_list[ARC_BUFC_NUMTYPES];
+	multilist_t *arcs_list[ARC_BUFC_NUMTYPES];
 	/*
 	 * total amount of evictable data in this state
 	 */

diff --git a/include/sys/dmu_objset.h b/include/sys/dmu_objset.h
@@ -20,7 +20,7 @@
  */
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2012, 2016 by Delphix. All rights reserved.
+ * Copyright (c) 2012, 2017 by Delphix. All rights reserved.
  * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
  * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
  */
@@ -113,7 +113,7 @@ struct objset {
 	/* no lock needed: */
 	struct dmu_tx *os_synctx; /* XXX sketchy */
 	zil_header_t os_zil_header;
-	list_t os_synced_dnodes;
+	multilist_t *os_synced_dnodes;
 	uint64_t os_flags;
 	uint64_t os_freed_dnodes;
 	boolean_t os_rescan_dnodes;
@@ -124,11 +124,13 @@ struct objset {
 
 	/* Protected by os_lock */
 	kmutex_t os_lock;
-	list_t os_dirty_dnodes[TXG_SIZE];
-	list_t os_free_dnodes[TXG_SIZE];
+	multilist_t *os_dirty_dnodes[TXG_SIZE];
 	list_t os_dnodes;
 	list_t os_downgraded_dbufs;
 
+	/* Protects changes to DMU_{USER,GROUP}USED_OBJECT */
+	kmutex_t os_userused_lock;
+
 	/* stuff we store for the user */
 	kmutex_t os_user_ptr_lock;
 	void *os_user_ptr;

diff --git a/include/sys/dnode.h b/include/sys/dnode.h
@@ -20,7 +20,7 @@
  */
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2012, 2016 by Delphix. All rights reserved.
+ * Copyright (c) 2012, 2017 by Delphix. All rights reserved.
  * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
  */
 
@@ -35,6 +35,7 @@
 #include <sys/refcount.h>
 #include <sys/dmu_zfetch.h>
 #include <sys/zrlock.h>
+#include <sys/multilist.h>
 
 #ifdef	__cplusplus
 extern "C" {
@@ -243,7 +244,7 @@ struct dnode {
 	uint32_t dn_dbufs_count;	/* count of dn_dbufs */
 
 	/* protected by os_lock: */
-	list_node_t dn_dirty_link[TXG_SIZE];	/* next on dataset's dirty */
+	multilist_node_t dn_dirty_link[TXG_SIZE]; /* next on dataset's dirty */
 
 	/* protected by dn_mtx: */
 	kmutex_t dn_mtx;

diff --git a/include/sys/dsl_pool.h b/include/sys/dsl_pool.h
@@ -20,7 +20,7 @@
  */
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2013 by Delphix. All rights reserved.
+ * Copyright (c) 2013, 2017 by Delphix. All rights reserved.
  * Copyright 2016 Nexenta Systems, Inc.  All rights reserved.
  */
 
@@ -124,6 +124,7 @@ typedef struct dsl_pool {
 	txg_list_t dp_dirty_zilogs;
 	txg_list_t dp_dirty_dirs;
 	txg_list_t dp_sync_tasks;
+	taskq_t *dp_sync_taskq;
 
 	/*
 	 * Protects administrative changes (properties, namespace)

diff --git a/include/sys/multilist.h b/include/sys/multilist.h
@@ -72,8 +72,7 @@ struct multilist {
 };
 
 void multilist_destroy(multilist_t *);
-void multilist_create(multilist_t *, size_t, size_t,
-    multilist_sublist_index_func_t *);
+multilist_t *multilist_create(size_t, size_t, multilist_sublist_index_func_t *);
 
 void multilist_insert(multilist_t *, void *);
 void multilist_remove(multilist_t *, void *);
@@ -83,6 +82,7 @@ unsigned int multilist_get_num_sublists(multilist_t *);
 unsigned int multilist_get_random_index(multilist_t *);
 
 multilist_sublist_t *multilist_sublist_lock(multilist_t *, unsigned int);
+multilist_sublist_t *multilist_sublist_lock_obj(multilist_t *, void *);
 void multilist_sublist_unlock(multilist_sublist_t *);
 
 void multilist_sublist_insert_head(multilist_sublist_t *, void *);

diff --git a/include/sys/zio.h b/include/sys/zio.h
@@ -22,7 +22,7 @@
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
- * Copyright (c) 2012, 2016 by Delphix. All rights reserved.
+ * Copyright (c) 2012, 2017 by Delphix. All rights reserved.
  * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
  */
 
@@ -454,7 +454,7 @@ struct zio {
 	taskq_ent_t	io_tqent;
 };
 
-extern int zio_timestamp_compare(const void *, const void *);
+extern int zio_bookmark_compare(const void *, const void *);
 
 extern zio_t *zio_null(zio_t *pio, spa_t *spa, vdev_t *vd,
     zio_done_func_t *done, void *private, enum zio_flag flags);

diff --git a/module/zfs/arc.c b/module/zfs/arc.c
@@ -1927,7 +1927,7 @@ add_reference(arc_buf_hdr_t *hdr, void *tag)
 	    (state != arc_anon)) {
 		/* We don't use the L2-only state list. */
 		if (state != arc_l2c_only) {
-			multilist_remove(&state->arcs_list[arc_buf_type(hdr)],
+			multilist_remove(state->arcs_list[arc_buf_type(hdr)],
 			    hdr);
 			arc_evictable_space_decrement(hdr, state);
 		}
@@ -1957,7 +1957,7 @@ remove_reference(arc_buf_hdr_t *hdr, kmutex_t *hash_lock, void *tag)
 	 */
 	if (((cnt = refcount_remove(&hdr->b_l1hdr.b_refcnt, tag)) == 0) &&
 	    (state != arc_anon)) {
-		multilist_insert(&state->arcs_list[arc_buf_type(hdr)], hdr);
+		multilist_insert(state->arcs_list[arc_buf_type(hdr)], hdr);
 		ASSERT3U(hdr->b_l1hdr.b_bufcnt, >, 0);
 		arc_evictable_space_increment(hdr, state);
 	}
@@ -2059,7 +2059,7 @@ arc_change_state(arc_state_t *new_state, arc_buf_hdr_t *hdr,
 	if (refcnt == 0) {
 		if (old_state != arc_anon && old_state != arc_l2c_only) {
 			ASSERT(HDR_HAS_L1HDR(hdr));
-			multilist_remove(&old_state->arcs_list[buftype], hdr);
+			multilist_remove(old_state->arcs_list[buftype], hdr);
 
 			if (GHOST_STATE(old_state)) {
 				ASSERT0(bufcnt);
@@ -2076,7 +2076,7 @@ arc_change_state(arc_state_t *new_state, arc_buf_hdr_t *hdr,
 			 * beforehand.
 			 */
 			ASSERT(HDR_HAS_L1HDR(hdr));
-			multilist_insert(&new_state->arcs_list[buftype], hdr);
+			multilist_insert(new_state->arcs_list[buftype], hdr);
 
 			if (GHOST_STATE(new_state)) {
 				ASSERT0(bufcnt);
@@ -2204,8 +2204,8 @@ arc_change_state(arc_state_t *new_state, arc_buf_hdr_t *hdr,
 	 * L2 headers should never be on the L2 state list since they don't
 	 * have L1 headers allocated.
 	 */
-	ASSERT(multilist_is_empty(&arc_l2c_only->arcs_list[ARC_BUFC_DATA]) &&
-	    multilist_is_empty(&arc_l2c_only->arcs_list[ARC_BUFC_METADATA]));
+	ASSERT(multilist_is_empty(arc_l2c_only->arcs_list[ARC_BUFC_DATA]) &&
+	    multilist_is_empty(arc_l2c_only->arcs_list[ARC_BUFC_METADATA]));
 }
 
 void
@@ -3302,7 +3302,7 @@ arc_evict_state(arc_state_t *state, uint64_t spa, int64_t bytes,
     arc_buf_contents_t type)
 {
 	uint64_t total_evicted = 0;
-	multilist_t *ml = &state->arcs_list[type];
+	multilist_t *ml = state->arcs_list[type];
 	int num_sublists;
 	arc_buf_hdr_t **markers;
 	int i;
@@ -3681,8 +3681,8 @@ arc_adjust_meta(void)
 static arc_buf_contents_t
 arc_adjust_type(arc_state_t *state)
 {
-	multilist_t *data_ml = &state->arcs_list[ARC_BUFC_DATA];
-	multilist_t *meta_ml = &state->arcs_list[ARC_BUFC_METADATA];
+	multilist_t *data_ml = state->arcs_list[ARC_BUFC_DATA];
+	multilist_t *meta_ml = state->arcs_list[ARC_BUFC_METADATA];
 	int data_idx = multilist_get_random_index(data_ml);
 	int meta_idx = multilist_get_random_index(meta_ml);
 	multilist_sublist_t *data_mls;
@@ -6281,44 +6281,44 @@ arc_state_init(void)
 	arc_mfu_ghost = &ARC_mfu_ghost;
 	arc_l2c_only = &ARC_l2c_only;
 
-	multilist_create(&arc_mru->arcs_list[ARC_BUFC_METADATA],
-	    sizeof (arc_buf_hdr_t),
+	arc_mru->arcs_list[ARC_BUFC_METADATA] =
+	    multilist_create(sizeof (arc_buf_hdr_t),
 	    offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node),
 	    arc_state_multilist_index_func);
-	multilist_create(&arc_mru->arcs_list[ARC_BUFC_DATA],
-	    sizeof (arc_buf_hdr_t),
+	arc_mru->arcs_list[ARC_BUFC_DATA] =
+	    multilist_create(sizeof (arc_buf_hdr_t),
 	    offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node),
 	    arc_state_multilist_index_func);
-	multilist_create(&arc_mru_ghost->arcs_list[ARC_BUFC_METADATA],
-	    sizeof (arc_buf_hdr_t),
+	arc_mru_ghost->arcs_list[ARC_BUFC_METADATA] =
+	    multilist_create(sizeof (arc_buf_hdr_t),
 	    offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node),
 	    arc_state_multilist_index_func);
-	multilist_create(&arc_mru_ghost->arcs_list[ARC_BUFC_DATA],
-	    sizeof (arc_buf_hdr_t),
+	arc_mru_ghost->arcs_list[ARC_BUFC_DATA] =
+	    multilist_create(sizeof (arc_buf_hdr_t),
 	    offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node),
 	    arc_state_multilist_index_func);
-	multilist_create(&arc_mfu->arcs_list[ARC_BUFC_METADATA],
-	    sizeof (arc_buf_hdr_t),
+	arc_mfu->arcs_list[ARC_BUFC_METADATA] =
+	    multilist_create(sizeof (arc_buf_hdr_t),
 	    offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node),
 	    arc_state_multilist_index_func);
-	multilist_create(&arc_mfu->arcs_list[ARC_BUFC_DATA],
-	    sizeof (arc_buf_hdr_t),
+	arc_mfu->arcs_list[ARC_BUFC_DATA] =
+	    multilist_create(sizeof (arc_buf_hdr_t),
 	    offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node),
 	    arc_state_multilist_index_func);
-	multilist_create(&arc_mfu_ghost->arcs_list[ARC_BUFC_METADATA],
-	    sizeof (arc_buf_hdr_t),
+	arc_mfu_ghost->arcs_list[ARC_BUFC_METADATA] =
+	    multilist_create(sizeof (arc_buf_hdr_t),
 	    offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node),
 	    arc_state_multilist_index_func);
-	multilist_create(&arc_mfu_ghost->arcs_list[ARC_BUFC_DATA],
-	    sizeof (arc_buf_hdr_t),
+	arc_mfu_ghost->arcs_list[ARC_BUFC_DATA] =
+	    multilist_create(sizeof (arc_buf_hdr_t),
 	    offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node),
 	    arc_state_multilist_index_func);
-	multilist_create(&arc_l2c_only->arcs_list[ARC_BUFC_METADATA],
-	    sizeof (arc_buf_hdr_t),
+	arc_l2c_only->arcs_list[ARC_BUFC_METADATA] =
+	    multilist_create(sizeof (arc_buf_hdr_t),
 	    offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node),
 	    arc_state_multilist_index_func);
-	multilist_create(&arc_l2c_only->arcs_list[ARC_BUFC_DATA],
-	    sizeof (arc_buf_hdr_t),
+	arc_l2c_only->arcs_list[ARC_BUFC_DATA] =
+	    multilist_create(sizeof (arc_buf_hdr_t),
 	    offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node),
 	    arc_state_multilist_index_func);
 
@@ -6373,16 +6373,16 @@ arc_state_fini(void)
 	refcount_destroy(&arc_mfu_ghost->arcs_size);
 	refcount_destroy(&arc_l2c_only->arcs_size);
 
-	multilist_destroy(&arc_mru->arcs_list[ARC_BUFC_METADATA]);
-	multilist_destroy(&arc_mru_ghost->arcs_list[ARC_BUFC_METADATA]);
-	multilist_destroy(&arc_mfu->arcs_list[ARC_BUFC_METADATA]);
-	multilist_destroy(&arc_mfu_ghost->arcs_list[ARC_BUFC_METADATA]);
-	multilist_destroy(&arc_mru->arcs_list[ARC_BUFC_DATA]);
-	multilist_destroy(&arc_mru_ghost->arcs_list[ARC_BUFC_DATA]);
-	multilist_destroy(&arc_mfu->arcs_list[ARC_BUFC_DATA]);
-	multilist_destroy(&arc_mfu_ghost->arcs_list[ARC_BUFC_DATA]);
-	multilist_destroy(&arc_l2c_only->arcs_list[ARC_BUFC_METADATA]);
-	multilist_destroy(&arc_l2c_only->arcs_list[ARC_BUFC_DATA]);
+	multilist_destroy(arc_mru->arcs_list[ARC_BUFC_METADATA]);
+	multilist_destroy(arc_mru_ghost->arcs_list[ARC_BUFC_METADATA]);
+	multilist_destroy(arc_mfu->arcs_list[ARC_BUFC_METADATA]);
+	multilist_destroy(arc_mfu_ghost->arcs_list[ARC_BUFC_METADATA]);
+	multilist_destroy(arc_mru->arcs_list[ARC_BUFC_DATA]);
+	multilist_destroy(arc_mru_ghost->arcs_list[ARC_BUFC_DATA]);
+	multilist_destroy(arc_mfu->arcs_list[ARC_BUFC_DATA]);
+	multilist_destroy(arc_mfu_ghost->arcs_list[ARC_BUFC_DATA]);
+	multilist_destroy(arc_l2c_only->arcs_list[ARC_BUFC_METADATA]);
+	multilist_destroy(arc_l2c_only->arcs_list[ARC_BUFC_DATA]);
 }
 
 uint64_t
@@ -7065,16 +7065,16 @@ l2arc_sublist_lock(int list_num)
 
 	switch (list_num) {
 	case 0:
-		ml = &arc_mfu->arcs_list[ARC_BUFC_METADATA];
+		ml = arc_mfu->arcs_list[ARC_BUFC_METADATA];
 		break;
 	case 1:
-		ml = &arc_mru->arcs_list[ARC_BUFC_METADATA];
+		ml = arc_mru->arcs_list[ARC_BUFC_METADATA];
 		break;
 	case 2:
-		ml = &arc_mfu->arcs_list[ARC_BUFC_DATA];
+		ml = arc_mfu->arcs_list[ARC_BUFC_DATA];
 		break;
 	case 3:
-		ml = &arc_mru->arcs_list[ARC_BUFC_DATA];
+		ml = arc_mru->arcs_list[ARC_BUFC_DATA];
 		break;
 	default:
 		return (NULL);

diff --git a/module/zfs/dbuf.c b/module/zfs/dbuf.c
@@ -21,7 +21,7 @@
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  * Copyright 2011 Nexenta Systems, Inc.  All rights reserved.
- * Copyright (c) 2012, 2015 by Delphix. All rights reserved.
+ * Copyright (c) 2012, 2017 by Delphix. All rights reserved.
  * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
  * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
  */
@@ -104,7 +104,7 @@ static boolean_t dbuf_evict_thread_exit;
  * Dbufs that are aged out of the cache will be immediately destroyed and
  * become eligible for arc eviction.
  */
-static multilist_t dbuf_cache;
+static multilist_t *dbuf_cache;
 static refcount_t dbuf_cache_size;
 unsigned long  dbuf_cache_max_bytes = 100 * 1024 * 1024;
 
@@ -491,8 +491,8 @@ dbuf_cache_above_lowater(void)
 static void
 dbuf_evict_one(void)
 {
-	int idx = multilist_get_random_index(&dbuf_cache);
-	multilist_sublist_t *mls = multilist_sublist_lock(&dbuf_cache, idx);
+	int idx = multilist_get_random_index(dbuf_cache);
+	multilist_sublist_t *mls = multilist_sublist_lock(dbuf_cache, idx);
 	dmu_buf_impl_t *db;
 	ASSERT(!MUTEX_HELD(&dbuf_evict_lock));
 
@@ -671,7 +671,7 @@ dbuf_init(void)
 	 */
 	dbu_evict_taskq = taskq_create("dbu_evict", 1, defclsyspri, 0, 0, 0);
 
-	multilist_create(&dbuf_cache, sizeof (dmu_buf_impl_t),
+	dbuf_cache = multilist_create(sizeof (dmu_buf_impl_t),
 	    offsetof(dmu_buf_impl_t, db_cache_link),
 	    dbuf_cache_multilist_index_func);
 	refcount_create(&dbuf_cache_size);
@@ -719,7 +719,7 @@ dbuf_fini(void)
 	cv_destroy(&dbuf_evict_cv);
 
 	refcount_destroy(&dbuf_cache_size);
-	multilist_destroy(&dbuf_cache);
+	multilist_destroy(dbuf_cache);
 }
 
 /*
@@ -2120,7 +2120,7 @@ dbuf_destroy(dmu_buf_impl_t *db)
 	dbuf_clear_data(db);
 
 	if (multilist_link_active(&db->db_cache_link)) {
-		multilist_remove(&dbuf_cache, db);
+		multilist_remove(dbuf_cache, db);
 		(void) refcount_remove_many(&dbuf_cache_size,
 		    db->db.db_size, db);
 	}
@@ -2690,7 +2690,7 @@ __dbuf_hold_impl(struct dbuf_hold_impl_data *dh)
 
 	if (multilist_link_active(&dh->dh_db->db_cache_link)) {
 		ASSERT(refcount_is_zero(&dh->dh_db->db_holds));
-		multilist_remove(&dbuf_cache, dh->dh_db);
+		multilist_remove(dbuf_cache, dh->dh_db);
 		(void) refcount_remove_many(&dbuf_cache_size,
 		    dh->dh_db->db.db_size, dh->dh_db);
 	}
@@ -2962,7 +2962,7 @@ dbuf_rele_and_unlock(dmu_buf_impl_t *db, void *tag)
 			    db->db_pending_evict) {
 				dbuf_destroy(db);
 			} else if (!multilist_link_active(&db->db_cache_link)) {
-				multilist_insert(&dbuf_cache, db);
+				multilist_insert(dbuf_cache, db);
 				(void) refcount_add_many(&dbuf_cache_size,
 				    db->db.db_size, db);
 				mutex_exit(&db->db_mtx);