Reduce ZIO io_lock contention on sorted scrub.

During sorted scrub multiple threads (one per vdev) are issuing many
ZIOs same time, all using the same scn->scn_zio_root ZIO as parent.
It causes huge lock contention on the single global lock on that ZIO.
Improve it by introducing per-queue null ZIOs, children to that one,
and using them instead as proxy.

For 12 SSD pool storing 1.5TB of 4KB blocks on 80-core system this
dramatically reduces lock contention and reduces scrub time from 21
minutes down to 12.5, while actual read stages (not scan) are about
3x faster, reaching 100K blocks per second per vdev.

Signed-off-by: Alexander Motin <[email protected]>
Sponsored-By: iXsystems, Inc.

module/zfs/dsl_scan.c

@@ -280,6 +280,7 @@ typedef struct scan_io {
 struct dsl_scan_io_queue {
 	dsl_scan_t	*q_scn; /* associated dsl_scan_t */
 	vdev_t		*q_vd; /* top-level vdev that this queue represents */
+	zio_t		*q_zio; /* scn_zio_root child for waiting on IO */
 
 	/* trees used for sorting I/Os and extents of I/Os */
 	range_tree_t	*q_exts_by_addr;
@@ -3044,6 +3045,8 @@ scan_io_queues_run_one(void *arg)
 
 	list_create(&sio_list, sizeof (scan_io_t),
 	    offsetof(scan_io_t, sio_nodes.sio_list_node));
+	queue->q_zio = zio_null(queue->q_scn->scn_zio_root,
+	    queue->q_scn->scn_dp->dp_spa, NULL, NULL, NULL, ZIO_FLAG_CANFAIL);
 	mutex_enter(q_lock);
 
 	/* Calculate maximum in-flight bytes for this vdev. */
@@ -3113,6 +3116,8 @@ scan_io_queues_run_one(void *arg)
 
 	mutex_exit(q_lock);
 	list_destroy(&sio_list);
+	zio_nowait(queue->q_zio);
+	queue->q_zio = NULL;
 }
 
 /*
@@ -4076,6 +4081,7 @@ scan_exec_io(dsl_pool_t *dp, const blkptr_t *bp, int zio_flags,
 	dsl_scan_t *scn = dp->dp_scan;
 	size_t size = BP_GET_PSIZE(bp);
 	abd_t *data = abd_alloc_for_io(size, B_FALSE);
+	zio_t *pio;
 
 	if (queue == NULL) {
 		ASSERT3U(scn->scn_maxinflight_bytes, >, 0);
@@ -4084,6 +4090,7 @@ scan_exec_io(dsl_pool_t *dp, const blkptr_t *bp, int zio_flags,
 			cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
 		spa->spa_scrub_inflight += BP_GET_PSIZE(bp);
 		mutex_exit(&spa->spa_scrub_lock);
+		pio = scn->scn_zio_root;
 	} else {
 		kmutex_t *q_lock = &queue->q_vd->vdev_scan_io_queue_lock;
 
@@ -4093,11 +4100,13 @@ scan_exec_io(dsl_pool_t *dp, const blkptr_t *bp, int zio_flags,
 			cv_wait(&queue->q_zio_cv, q_lock);
 		queue->q_inflight_bytes += BP_GET_PSIZE(bp);
 		mutex_exit(q_lock);
+		pio = queue->q_zio;
+		ASSERT(pio != NULL);
 	}
 
 	count_block(scn, dp->dp_blkstats, bp);
-	zio_nowait(zio_read(scn->scn_zio_root, spa, bp, data, size,
-	    dsl_scan_scrub_done, queue, ZIO_PRIORITY_SCRUB, zio_flags, zb));
+	zio_nowait(zio_read(pio, spa, bp, data, size, dsl_scan_scrub_done,
+	    queue, ZIO_PRIORITY_SCRUB, zio_flags, zb));
 }
 
 /*