OpenZFS 9102 - zfs should be able to initialize storage devices

PROBLEM
========

The first access to a block incurs a performance penalty on some platforms
(e.g. AWS's EBS, VMware VMDKs). Therefore we recommend that volumes are
"thick provisioned", where supported by the platform (VMware). This can
create a large delay in getting a new virtual machines up and running (or
adding storage to an existing Engine). If the thick provision step is
omitted, write performance will be suboptimal until all blocks on the LUN
have been written.

SOLUTION
=========

This feature introduces a way to 'initialize' the disks at install or in the
background to make sure we don't incur this first read penalty.

When an entire LUN is added to ZFS, we make all space available immediately,
and allow ZFS to find unallocated space and zero it out. This works with
concurrent writes to arbitrary offsets, ensuring that we don't zero out
something that has been (or is in the middle of being) written. This scheme
can also be applied to existing pools (affecting only free regions on the
vdev). Detailed design:
        - new subcommand:zpool initialize [-cs] <pool> [<vdev> ...]
                - start, suspend, or cancel initialization
        - Creates new open-context thread for each vdev
        - Thread iterates through all metaslabs in this vdev
        - Each metaslab:
                - select a metaslab
                - load the metaslab
                - mark the metaslab as being zeroed
                - walk all free ranges within that metaslab and translate
                  them to ranges on the leaf vdev
                - issue a "zeroing" I/O on the leaf vdev that corresponds to
                  a free range on the metaslab we're working on
                - continue until all free ranges for this metaslab have been
                  "zeroed"
                - reset/unmark the metaslab being zeroed
                - if more metaslabs exist, then repeat above tasks.
                - if no more metaslabs, then we're done.

        - progress for the initialization is stored on-disk in the vdev’s
          leaf zap object. The following information is stored:
                - the last offset that has been initialized
                - the state of the initialization process (i.e. active,
                  suspended, or canceled)
                - the start time for the initialization

        - progress is reported via the zpool status command and shows
          information for each of the vdevs that are initializing

Porting notes:
- Added zfs_initialize_value module parameter to set the pattern
  written by "zpool initialize".
- Added zfs_vdev_{initializing,removal}_{min,max}_active module options.

Authored by: George Wilson <george.wilson@delphix.com>
Reviewed by: John Wren Kennedy <john.kennedy@delphix.com>
Reviewed by: Matthew Ahrens <mahrens@delphix.com>
Reviewed by: Pavel Zakharov <pavel.zakharov@delphix.com>
Reviewed by: Prakash Surya <prakash.surya@delphix.com>
Reviewed by: loli10K <ezomori.nozomu@gmail.com>
Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov>
Approved by: Richard Lowe <richlowe@richlowe.net>
Signed-off-by: Tim Chase <tim@chase2k.com>
Ported-by: Tim Chase <tim@chase2k.com>

OpenZFS-issue: https://www.illumos.org/issues/9102
OpenZFS-commit: https://github.com/openzfs/openzfs/commit/c3963210eb
Closes #8230

module/zfs/spa.c

@@ -56,6 +56,7 @@
 #include <sys/vdev_removal.h>
 #include <sys/vdev_indirect_mapping.h>
 #include <sys/vdev_indirect_births.h>
+#include <sys/vdev_initialize.h>
 #include <sys/vdev_disk.h>
 #include <sys/metaslab.h>
 #include <sys/metaslab_impl.h>
@@ -434,8 +435,9 @@ spa_prop_get(spa_t *spa, nvlist_t **nvp)
 
 				dp = spa_get_dsl(spa);
 				dsl_pool_config_enter(dp, FTAG);
-				if ((err = dsl_dataset_hold_obj(dp,
-				    za.za_first_integer, FTAG, &ds))) {
+				err = dsl_dataset_hold_obj(dp,
+				    za.za_first_integer, FTAG, &ds);
+				if (err != 0) {
 					dsl_pool_config_exit(dp, FTAG);
 					break;
 				}
@@ -601,7 +603,7 @@ spa_prop_validate(spa_t *spa, nvlist_t *props)
 				}
 
 				error = dmu_objset_hold(strval, FTAG, &os);
-				if (error)
+				if (error != 0)
 					break;
 
 				/*
@@ -1218,8 +1220,10 @@ spa_activate(spa_t *spa, int mode)
 		spa_create_zio_taskqs(spa);
 	}
 
-	for (size_t i = 0; i < TXG_SIZE; i++)
-		spa->spa_txg_zio[i] = zio_root(spa, NULL, NULL, 0);
+	for (size_t i = 0; i < TXG_SIZE; i++) {
+		spa->spa_txg_zio[i] = zio_root(spa, NULL, NULL,
+		    ZIO_FLAG_CANFAIL);
+	}
 
 	list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
 	    offsetof(vdev_t, vdev_config_dirty_node));
@@ -1437,6 +1441,11 @@ spa_unload(spa_t *spa)
 	 */
 	spa_async_suspend(spa);
 
+	if (spa->spa_root_vdev) {
+		vdev_initialize_stop_all(spa->spa_root_vdev,
+		    VDEV_INITIALIZE_ACTIVE);
+	}
+
 	/*
 	 * Stop syncing.
 	 */
@@ -1452,10 +1461,10 @@ spa_unload(spa_t *spa)
 	 * calling taskq_wait(mg_taskq).
 	 */
 	if (spa->spa_root_vdev != NULL) {
-		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
+		spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
 		for (int c = 0; c < spa->spa_root_vdev->vdev_children; c++)
 			vdev_metaslab_fini(spa->spa_root_vdev->vdev_child[c]);
-		spa_config_exit(spa, SCL_ALL, FTAG);
+		spa_config_exit(spa, SCL_ALL, spa);
 	}
 
 	if (spa->spa_mmp.mmp_thread)
@@ -1492,7 +1501,7 @@ spa_unload(spa_t *spa)
 
 	bpobj_close(&spa->spa_deferred_bpobj);
 
-	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
+	spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
 
 	/*
 	 * Close all vdevs.
@@ -1554,7 +1563,7 @@ spa_unload(spa_t *spa)
 		spa->spa_comment = NULL;
 	}
 
-	spa_config_exit(spa, SCL_ALL, FTAG);
+	spa_config_exit(spa, SCL_ALL, spa);
 }
 
 /*
@@ -4246,6 +4255,9 @@ spa_load_impl(spa_t *spa, spa_import_type_t type, char **ereport)
 		 */
 		dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
 
+		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
+		vdev_initialize_restart(spa->spa_root_vdev);
+		spa_config_exit(spa, SCL_CONFIG, FTAG);
 	}
 
 	spa_load_note(spa, "LOADED");
@@ -5653,6 +5665,18 @@ spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
 			return (SET_ERROR(EXDEV));
 		}
 
+		/*
+		 * We're about to export or destroy this pool. Make sure
+		 * we stop all initializtion activity here before we
+		 * set the spa_final_txg. This will ensure that all
+		 * dirty data resulting from the initialization is
+		 * committed to disk before we unload the pool.
+		 */
+		if (spa->spa_root_vdev != NULL) {
+			vdev_initialize_stop_all(spa->spa_root_vdev,
+			    VDEV_INITIALIZE_ACTIVE);
+		}
+
 		/*
 		 * We want this to be reflected on every label,
 		 * so mark them all dirty.  spa_unload() will do the
@@ -6357,6 +6381,86 @@ spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
 	return (error);
 }
 
+int
+spa_vdev_initialize(spa_t *spa, uint64_t guid, uint64_t cmd_type)
+{
+	/*
+	 * We hold the namespace lock through the whole function
+	 * to prevent any changes to the pool while we're starting or
+	 * stopping initialization. The config and state locks are held so that
+	 * we can properly assess the vdev state before we commit to
+	 * the initializing operation.
+	 */
+	mutex_enter(&spa_namespace_lock);
+	spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
+
+	/* Look up vdev and ensure it's a leaf. */
+	vdev_t *vd = spa_lookup_by_guid(spa, guid, B_FALSE);
+	if (vd == NULL || vd->vdev_detached) {
+		spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
+		mutex_exit(&spa_namespace_lock);
+		return (SET_ERROR(ENODEV));
+	} else if (!vd->vdev_ops->vdev_op_leaf || !vdev_is_concrete(vd)) {
+		spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
+		mutex_exit(&spa_namespace_lock);
+		return (SET_ERROR(EINVAL));
+	} else if (!vdev_writeable(vd)) {
+		spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
+		mutex_exit(&spa_namespace_lock);
+		return (SET_ERROR(EROFS));
+	}
+	mutex_enter(&vd->vdev_initialize_lock);
+	spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
+
+	/*
+	 * When we activate an initialize action we check to see
+	 * if the vdev_initialize_thread is NULL. We do this instead
+	 * of using the vdev_initialize_state since there might be
+	 * a previous initialization process which has completed but
+	 * the thread is not exited.
+	 */
+	if (cmd_type == POOL_INITIALIZE_DO &&
+	    (vd->vdev_initialize_thread != NULL ||
+	    vd->vdev_top->vdev_removing)) {
+		mutex_exit(&vd->vdev_initialize_lock);
+		mutex_exit(&spa_namespace_lock);
+		return (SET_ERROR(EBUSY));
+	} else if (cmd_type == POOL_INITIALIZE_CANCEL &&
+	    (vd->vdev_initialize_state != VDEV_INITIALIZE_ACTIVE &&
+	    vd->vdev_initialize_state != VDEV_INITIALIZE_SUSPENDED)) {
+		mutex_exit(&vd->vdev_initialize_lock);
+		mutex_exit(&spa_namespace_lock);
+		return (SET_ERROR(ESRCH));
+	} else if (cmd_type == POOL_INITIALIZE_SUSPEND &&
+	    vd->vdev_initialize_state != VDEV_INITIALIZE_ACTIVE) {
+		mutex_exit(&vd->vdev_initialize_lock);
+		mutex_exit(&spa_namespace_lock);
+		return (SET_ERROR(ESRCH));
+	}
+
+	switch (cmd_type) {
+	case POOL_INITIALIZE_DO:
+		vdev_initialize(vd);
+		break;
+	case POOL_INITIALIZE_CANCEL:
+		vdev_initialize_stop(vd, VDEV_INITIALIZE_CANCELED);
+		break;
+	case POOL_INITIALIZE_SUSPEND:
+		vdev_initialize_stop(vd, VDEV_INITIALIZE_SUSPENDED);
+		break;
+	default:
+		panic("invalid cmd_type %llu", (unsigned long long)cmd_type);
+	}
+	mutex_exit(&vd->vdev_initialize_lock);
+
+	/* Sync out the initializing state */
+	txg_wait_synced(spa->spa_dsl_pool, 0);
+	mutex_exit(&spa_namespace_lock);
+
+	return (0);
+}
+
+
 /*
  * Split a set of devices from their mirrors, and create a new pool from them.
  */
@@ -6565,6 +6669,19 @@ spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
 	spa_activate(newspa, spa_mode_global);
 	spa_async_suspend(newspa);
 
+	for (c = 0; c < children; c++) {
+		if (vml[c] != NULL) {
+			/*
+			 * Temporarily stop the initializing activity. We set
+			 * the state to ACTIVE so that we know to resume
+			 * the initializing once the split has completed.
+			 */
+			mutex_enter(&vml[c]->vdev_initialize_lock);
+			vdev_initialize_stop(vml[c], VDEV_INITIALIZE_ACTIVE);
+			mutex_exit(&vml[c]->vdev_initialize_lock);
+		}
+	}
+
 	newspa->spa_config_source = SPA_CONFIG_SRC_SPLIT;
 
 	/* create the new pool from the disks of the original pool */
@@ -6652,6 +6769,10 @@ out:
 		if (vml[c] != NULL)
 			vml[c]->vdev_offline = B_FALSE;
 	}
+
+	/* restart initializing disks as necessary */
+	spa_async_request(spa, SPA_ASYNC_INITIALIZE_RESTART);
+
 	vdev_reopen(spa->spa_root_vdev);
 
 	nvlist_free(spa->spa_config_splitting);
@@ -7025,6 +7146,14 @@ spa_async_thread(void *arg)
 	    !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_RESILVER_DEFER)))
 		dsl_resilver_restart(dp, 0);
 
+	if (tasks & SPA_ASYNC_INITIALIZE_RESTART) {
+		mutex_enter(&spa_namespace_lock);
+		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
+		vdev_initialize_restart(spa->spa_root_vdev);
+		spa_config_exit(spa, SCL_CONFIG, FTAG);
+		mutex_exit(&spa_namespace_lock);
+	}
+
 	/*
 	 * Let the world know that we're done.
 	 */
@@ -7677,8 +7806,9 @@ spa_sync(spa_t *spa, uint64_t txg)
 	 * Wait for i/os issued in open context that need to complete
 	 * before this txg syncs.
 	 */
-	VERIFY0(zio_wait(spa->spa_txg_zio[txg & TXG_MASK]));
-	spa->spa_txg_zio[txg & TXG_MASK] = zio_root(spa, NULL, NULL, 0);
+	(void) zio_wait(spa->spa_txg_zio[txg & TXG_MASK]);
+	spa->spa_txg_zio[txg & TXG_MASK] = zio_root(spa, NULL, NULL,
+	    ZIO_FLAG_CANFAIL);
 
 	/*
 	 * Lock out configuration changes.
@@ -7983,7 +8113,8 @@ spa_sync(spa_t *spa, uint64_t txg)
 	/*
 	 * Update usable space statistics.
 	 */
-	while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg))))
+	while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
+	    != NULL)
 		vdev_sync_done(vd, txg);
 
 	spa_update_dspace(spa);