1 files changed, 27 insertions, 16 deletions

diff --git a/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/vdev_queue.c b/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/vdev_queue.c
index 5a3ba1b3e983..78b725a37b68 100644
--- a/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/vdev_queue.c
+++ b/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/vdev_queue.c
@@ -24,7 +24,7 @@
  */
 
 /*
- * Copyright (c) 2012, 2017 by Delphix. All rights reserved.
+ * Copyright (c) 2012, 2018 by Delphix. All rights reserved.
  * Copyright (c) 2014 Integros [integros.com]
  */
 
@@ -157,6 +157,8 @@ uint32_t zfs_vdev_trim_min_active = 1;
 uint32_t zfs_vdev_trim_max_active = 64;
 uint32_t zfs_vdev_removal_min_active = 1;
 uint32_t zfs_vdev_removal_max_active = 2;
+uint32_t zfs_vdev_initializing_min_active = 1;
+uint32_t zfs_vdev_initializing_max_active = 1;
 
 
 /*
@@ -195,6 +197,14 @@ int zfs_vdev_queue_depth_pct = 1000;
 int zfs_vdev_queue_depth_pct = 300;
 #endif
 
+/*
+ * When performing allocations for a given metaslab, we want to make sure that
+ * there are enough IOs to aggregate together to improve throughput. We want to
+ * ensure that there are at least 128k worth of IOs that can be aggregated, and
+ * we assume that the average allocation size is 4k, so we need the queue depth
+ * to be 32 per allocator to get good aggregation of sequential writes.
+ */
+int zfs_vdev_def_queue_depth = 32;
 
 #ifdef __FreeBSD__
 #ifdef _KERNEL
@@ -301,20 +311,15 @@ sysctl_zfs_async_write_active_max_dirty_percent(SYSCTL_HANDLER_ARGS)
 int
 vdev_queue_offset_compare(const void *x1, const void *x2)
 {
-	const zio_t *z1 = x1;
-	const zio_t *z2 = x2;
+	const zio_t *z1 = (const zio_t *)x1;
+	const zio_t *z2 = (const zio_t *)x2;
 
-	if (z1->io_offset < z2->io_offset)
-		return (-1);
-	if (z1->io_offset > z2->io_offset)
-		return (1);
+	int cmp = AVL_CMP(z1->io_offset, z2->io_offset);
 
-	if (z1 < z2)
-		return (-1);
-	if (z1 > z2)
-		return (1);
+	if (likely(cmp))
+		return (cmp);
 
-	return (0);
+	return (AVL_PCMP(z1, z2));
 }
 
 static inline avl_tree_t *
@@ -534,6 +539,8 @@ vdev_queue_class_min_active(zio_priority_t p)
 		return (zfs_vdev_trim_min_active);
 	case ZIO_PRIORITY_REMOVAL:
 		return (zfs_vdev_removal_min_active);
+	case ZIO_PRIORITY_INITIALIZING:
+		return (zfs_vdev_initializing_min_active);
 	default:
 		panic("invalid priority %u", p);
 		return (0);
@@ -597,6 +604,8 @@ vdev_queue_class_max_active(spa_t *spa, zio_priority_t p)
 		return (zfs_vdev_trim_max_active);
 	case ZIO_PRIORITY_REMOVAL:
 		return (zfs_vdev_removal_max_active);
+	case ZIO_PRIORITY_INITIALIZING:
+		return (zfs_vdev_initializing_max_active);
 	default:
 		panic("invalid priority %u", p);
 		return (0);
@@ -824,8 +833,8 @@ again:
 	}
 
 	/*
-	 * For LBA-ordered queues (async / scrub), issue the i/o which follows
-	 * the most recently issued i/o in LBA (offset) order.
+	 * For LBA-ordered queues (async / scrub / initializing), issue the
+	 * i/o which follows the most recently issued i/o in LBA (offset) order.
 	 *
 	 * For FIFO queues (sync), issue the i/o with the lowest timestamp.
 	 */
@@ -881,12 +890,14 @@ vdev_queue_io(zio_t *zio)
 		if (zio->io_priority != ZIO_PRIORITY_SYNC_READ &&
 		    zio->io_priority != ZIO_PRIORITY_ASYNC_READ &&
 		    zio->io_priority != ZIO_PRIORITY_SCRUB &&
-		    zio->io_priority != ZIO_PRIORITY_REMOVAL)
+		    zio->io_priority != ZIO_PRIORITY_REMOVAL &&
+		    zio->io_priority != ZIO_PRIORITY_INITIALIZING)
 			zio->io_priority = ZIO_PRIORITY_ASYNC_READ;
 	} else if (zio->io_type == ZIO_TYPE_WRITE) {
 		if (zio->io_priority != ZIO_PRIORITY_SYNC_WRITE &&
 		    zio->io_priority != ZIO_PRIORITY_ASYNC_WRITE &&
-		    zio->io_priority != ZIO_PRIORITY_REMOVAL)
+		    zio->io_priority != ZIO_PRIORITY_REMOVAL &&
+		    zio->io_priority != ZIO_PRIORITY_INITIALIZING)
 			zio->io_priority = ZIO_PRIORITY_ASYNC_WRITE;
 	} else {
 		ASSERT(zio->io_type == ZIO_TYPE_FREE);