4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
25 * Copyright (c) 2011 by Delphix. All rights reserved.
29 * This file contains all the routines used when modifying on-disk SPA state.
30 * This includes opening, importing, destroying, exporting a pool, and syncing a
34 #include <sys/zfs_context.h>
35 #include <sys/fm/fs/zfs.h>
36 #include <sys/spa_impl.h>
38 #include <sys/zio_checksum.h>
40 #include <sys/dmu_tx.h>
44 #include <sys/vdev_impl.h>
45 #include <sys/vdev_disk.h>
46 #include <sys/metaslab.h>
47 #include <sys/metaslab_impl.h>
48 #include <sys/uberblock_impl.h>
51 #include <sys/dmu_traverse.h>
52 #include <sys/dmu_objset.h>
53 #include <sys/unique.h>
54 #include <sys/dsl_pool.h>
55 #include <sys/dsl_dataset.h>
56 #include <sys/dsl_dir.h>
57 #include <sys/dsl_prop.h>
58 #include <sys/dsl_synctask.h>
59 #include <sys/fs/zfs.h>
61 #include <sys/callb.h>
62 #include <sys/systeminfo.h>
63 #include <sys/spa_boot.h>
64 #include <sys/zfs_ioctl.h>
65 #include <sys/dsl_scan.h>
68 #include <sys/bootprops.h>
69 #include <sys/callb.h>
70 #include <sys/cpupart.h>
72 #include <sys/sysdc.h>
77 #include "zfs_comutil.h"
79 typedef enum zti_modes {
80 zti_mode_fixed, /* value is # of threads (min 1) */
81 zti_mode_online_percent, /* value is % of online CPUs */
82 zti_mode_batch, /* cpu-intensive; value is ignored */
83 zti_mode_null, /* don't create a taskq */
87 #define ZTI_FIX(n) { zti_mode_fixed, (n) }
88 #define ZTI_PCT(n) { zti_mode_online_percent, (n) }
89 #define ZTI_BATCH { zti_mode_batch, 0 }
90 #define ZTI_NULL { zti_mode_null, 0 }
92 #define ZTI_ONE ZTI_FIX(1)
94 typedef struct zio_taskq_info {
95 enum zti_modes zti_mode;
99 static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
100 "iss", "iss_h", "int", "int_h"
104 * Define the taskq threads for the following I/O types:
105 * NULL, READ, WRITE, FREE, CLAIM, and IOCTL
107 const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
108 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
109 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL },
110 { ZTI_FIX(8), ZTI_NULL, ZTI_BATCH, ZTI_NULL },
111 { ZTI_BATCH, ZTI_FIX(5), ZTI_FIX(16), ZTI_FIX(5) },
112 { ZTI_PCT(100), ZTI_NULL, ZTI_ONE, ZTI_NULL },
113 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL },
114 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL },
117 static dsl_syncfunc_t spa_sync_props;
118 static boolean_t spa_has_active_shared_spare(spa_t *spa);
119 static inline int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
120 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
122 static void spa_vdev_resilver_done(spa_t *spa);
124 uint_t zio_taskq_batch_pct = 100; /* 1 thread per cpu in pset */
125 id_t zio_taskq_psrset_bind = PS_NONE;
126 boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */
127 uint_t zio_taskq_basedc = 80; /* base duty cycle */
129 boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */
132 * This (illegal) pool name is used when temporarily importing a spa_t in order
133 * to get the vdev stats associated with the imported devices.
135 #define TRYIMPORT_NAME "$import"
138 * ==========================================================================
139 * SPA properties routines
140 * ==========================================================================
144 * Add a (source=src, propname=propval) list to an nvlist.
147 spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
148 uint64_t intval, zprop_source_t src)
150 const char *propname = zpool_prop_to_name(prop);
153 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
154 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
157 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
159 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
161 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
162 nvlist_free(propval);
166 * Get property values from the spa configuration.
169 spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
173 uint64_t cap, version;
174 zprop_source_t src = ZPROP_SRC_NONE;
175 spa_config_dirent_t *dp;
177 ASSERT(MUTEX_HELD(&spa->spa_props_lock));
179 if (spa->spa_root_vdev != NULL) {
180 alloc = metaslab_class_get_alloc(spa_normal_class(spa));
181 size = metaslab_class_get_space(spa_normal_class(spa));
182 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
183 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
184 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
185 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
187 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
188 (spa_mode(spa) == FREAD), src);
190 cap = (size == 0) ? 0 : (alloc * 100 / size);
191 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
193 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
194 ddt_get_pool_dedup_ratio(spa), src);
196 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
197 spa->spa_root_vdev->vdev_state, src);
199 version = spa_version(spa);
200 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
201 src = ZPROP_SRC_DEFAULT;
203 src = ZPROP_SRC_LOCAL;
204 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
207 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
209 if (spa->spa_root != NULL)
210 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
213 if ((dp = list_head(&spa->spa_config_list)) != NULL) {
214 if (dp->scd_path == NULL) {
215 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
216 "none", 0, ZPROP_SRC_LOCAL);
217 } else if (strcmp(dp->scd_path, spa_config_path) != 0) {
218 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
219 dp->scd_path, 0, ZPROP_SRC_LOCAL);
225 * Get zpool property values.
228 spa_prop_get(spa_t *spa, nvlist_t **nvp)
230 objset_t *mos = spa->spa_meta_objset;
235 err = nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP);
239 mutex_enter(&spa->spa_props_lock);
242 * Get properties from the spa config.
244 spa_prop_get_config(spa, nvp);
246 /* If no pool property object, no more prop to get. */
247 if (mos == NULL || spa->spa_pool_props_object == 0) {
248 mutex_exit(&spa->spa_props_lock);
253 * Get properties from the MOS pool property object.
255 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
256 (err = zap_cursor_retrieve(&zc, &za)) == 0;
257 zap_cursor_advance(&zc)) {
260 zprop_source_t src = ZPROP_SRC_DEFAULT;
263 if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
266 switch (za.za_integer_length) {
268 /* integer property */
269 if (za.za_first_integer !=
270 zpool_prop_default_numeric(prop))
271 src = ZPROP_SRC_LOCAL;
273 if (prop == ZPOOL_PROP_BOOTFS) {
275 dsl_dataset_t *ds = NULL;
277 dp = spa_get_dsl(spa);
278 rw_enter(&dp->dp_config_rwlock, RW_READER);
279 if ((err = dsl_dataset_hold_obj(dp,
280 za.za_first_integer, FTAG, &ds))) {
281 rw_exit(&dp->dp_config_rwlock);
286 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
288 dsl_dataset_name(ds, strval);
289 dsl_dataset_rele(ds, FTAG);
290 rw_exit(&dp->dp_config_rwlock);
293 intval = za.za_first_integer;
296 spa_prop_add_list(*nvp, prop, strval, intval, src);
300 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
305 /* string property */
306 strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
307 err = zap_lookup(mos, spa->spa_pool_props_object,
308 za.za_name, 1, za.za_num_integers, strval);
310 kmem_free(strval, za.za_num_integers);
313 spa_prop_add_list(*nvp, prop, strval, 0, src);
314 kmem_free(strval, za.za_num_integers);
321 zap_cursor_fini(&zc);
322 mutex_exit(&spa->spa_props_lock);
324 if (err && err != ENOENT) {
334 * Validate the given pool properties nvlist and modify the list
335 * for the property values to be set.
338 spa_prop_validate(spa_t *spa, nvlist_t *props)
341 int error = 0, reset_bootfs = 0;
345 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
347 char *propname, *strval;
352 propname = nvpair_name(elem);
354 if ((prop = zpool_name_to_prop(propname)) == ZPROP_INVAL)
358 case ZPOOL_PROP_VERSION:
359 error = nvpair_value_uint64(elem, &intval);
361 (intval < spa_version(spa) || intval > SPA_VERSION))
365 case ZPOOL_PROP_DELEGATION:
366 case ZPOOL_PROP_AUTOREPLACE:
367 case ZPOOL_PROP_LISTSNAPS:
368 case ZPOOL_PROP_AUTOEXPAND:
369 error = nvpair_value_uint64(elem, &intval);
370 if (!error && intval > 1)
374 case ZPOOL_PROP_BOOTFS:
376 * If the pool version is less than SPA_VERSION_BOOTFS,
377 * or the pool is still being created (version == 0),
378 * the bootfs property cannot be set.
380 if (spa_version(spa) < SPA_VERSION_BOOTFS) {
386 * Make sure the vdev config is bootable
388 if (!vdev_is_bootable(spa->spa_root_vdev)) {
395 error = nvpair_value_string(elem, &strval);
400 if (strval == NULL || strval[0] == '\0') {
401 objnum = zpool_prop_default_numeric(
406 if ((error = dmu_objset_hold(strval,FTAG,&os)))
409 /* Must be ZPL and not gzip compressed. */
411 if (dmu_objset_type(os) != DMU_OST_ZFS) {
413 } else if ((error = dsl_prop_get_integer(strval,
414 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
415 &compress, NULL)) == 0 &&
416 !BOOTFS_COMPRESS_VALID(compress)) {
419 objnum = dmu_objset_id(os);
421 dmu_objset_rele(os, FTAG);
425 case ZPOOL_PROP_FAILUREMODE:
426 error = nvpair_value_uint64(elem, &intval);
427 if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
428 intval > ZIO_FAILURE_MODE_PANIC))
432 * This is a special case which only occurs when
433 * the pool has completely failed. This allows
434 * the user to change the in-core failmode property
435 * without syncing it out to disk (I/Os might
436 * currently be blocked). We do this by returning
437 * EIO to the caller (spa_prop_set) to trick it
438 * into thinking we encountered a property validation
441 if (!error && spa_suspended(spa)) {
442 spa->spa_failmode = intval;
447 case ZPOOL_PROP_CACHEFILE:
448 if ((error = nvpair_value_string(elem, &strval)) != 0)
451 if (strval[0] == '\0')
454 if (strcmp(strval, "none") == 0)
457 if (strval[0] != '/') {
462 slash = strrchr(strval, '/');
463 ASSERT(slash != NULL);
465 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
466 strcmp(slash, "/..") == 0)
470 case ZPOOL_PROP_DEDUPDITTO:
471 if (spa_version(spa) < SPA_VERSION_DEDUP)
474 error = nvpair_value_uint64(elem, &intval);
476 intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
488 if (!error && reset_bootfs) {
489 error = nvlist_remove(props,
490 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
493 error = nvlist_add_uint64(props,
494 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
502 spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
505 spa_config_dirent_t *dp;
507 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
511 dp = kmem_alloc(sizeof (spa_config_dirent_t),
514 if (cachefile[0] == '\0')
515 dp->scd_path = spa_strdup(spa_config_path);
516 else if (strcmp(cachefile, "none") == 0)
519 dp->scd_path = spa_strdup(cachefile);
521 list_insert_head(&spa->spa_config_list, dp);
523 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
527 spa_prop_set(spa_t *spa, nvlist_t *nvp)
531 boolean_t need_sync = B_FALSE;
534 if ((error = spa_prop_validate(spa, nvp)) != 0)
538 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
539 if ((prop = zpool_name_to_prop(
540 nvpair_name(elem))) == ZPROP_INVAL)
543 if (prop == ZPOOL_PROP_CACHEFILE ||
544 prop == ZPOOL_PROP_ALTROOT ||
545 prop == ZPOOL_PROP_READONLY)
553 return (dsl_sync_task_do(spa_get_dsl(spa), NULL, spa_sync_props,
560 * If the bootfs property value is dsobj, clear it.
563 spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
565 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
566 VERIFY(zap_remove(spa->spa_meta_objset,
567 spa->spa_pool_props_object,
568 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
574 * Change the GUID for the pool. This is done so that we can later
575 * re-import a pool built from a clone of our own vdevs. We will modify
576 * the root vdev's guid, our own pool guid, and then mark all of our
577 * vdevs dirty. Note that we must make sure that all our vdevs are
578 * online when we do this, or else any vdevs that weren't present
579 * would be orphaned from our pool. We are also going to issue a
580 * sysevent to update any watchers.
583 spa_change_guid(spa_t *spa)
585 uint64_t oldguid, newguid;
588 if (!(spa_mode_global & FWRITE))
591 txg = spa_vdev_enter(spa);
593 if (spa->spa_root_vdev->vdev_state != VDEV_STATE_HEALTHY)
594 return (spa_vdev_exit(spa, NULL, txg, ENXIO));
596 oldguid = spa_guid(spa);
597 newguid = spa_generate_guid(NULL);
598 ASSERT3U(oldguid, !=, newguid);
600 spa->spa_root_vdev->vdev_guid = newguid;
601 spa->spa_root_vdev->vdev_guid_sum += (newguid - oldguid);
603 vdev_config_dirty(spa->spa_root_vdev);
605 spa_event_notify(spa, NULL, FM_EREPORT_ZFS_POOL_REGUID);
607 return (spa_vdev_exit(spa, NULL, txg, 0));
611 * ==========================================================================
612 * SPA state manipulation (open/create/destroy/import/export)
613 * ==========================================================================
617 spa_error_entry_compare(const void *a, const void *b)
619 spa_error_entry_t *sa = (spa_error_entry_t *)a;
620 spa_error_entry_t *sb = (spa_error_entry_t *)b;
623 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
624 sizeof (zbookmark_t));
635 * Utility function which retrieves copies of the current logs and
636 * re-initializes them in the process.
639 spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
641 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
643 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
644 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
646 avl_create(&spa->spa_errlist_scrub,
647 spa_error_entry_compare, sizeof (spa_error_entry_t),
648 offsetof(spa_error_entry_t, se_avl));
649 avl_create(&spa->spa_errlist_last,
650 spa_error_entry_compare, sizeof (spa_error_entry_t),
651 offsetof(spa_error_entry_t, se_avl));
655 spa_taskq_create(spa_t *spa, const char *name, enum zti_modes mode,
656 uint_t value, uint_t flags)
658 boolean_t batch = B_FALSE;
662 return (NULL); /* no taskq needed */
665 ASSERT3U(value, >=, 1);
666 value = MAX(value, 1);
671 flags |= TASKQ_THREADS_CPU_PCT;
672 value = zio_taskq_batch_pct;
675 case zti_mode_online_percent:
676 flags |= TASKQ_THREADS_CPU_PCT;
680 panic("unrecognized mode for %s taskq (%u:%u) in "
686 if (zio_taskq_sysdc && spa->spa_proc != &p0) {
688 flags |= TASKQ_DC_BATCH;
690 return (taskq_create_sysdc(name, value, 50, INT_MAX,
691 spa->spa_proc, zio_taskq_basedc, flags));
693 return (taskq_create_proc(name, value, maxclsyspri, 50, INT_MAX,
694 spa->spa_proc, flags));
698 spa_create_zio_taskqs(spa_t *spa)
702 for (t = 0; t < ZIO_TYPES; t++) {
703 for (q = 0; q < ZIO_TASKQ_TYPES; q++) {
704 const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
705 enum zti_modes mode = ztip->zti_mode;
706 uint_t value = ztip->zti_value;
710 if (t == ZIO_TYPE_WRITE)
711 flags |= TASKQ_NORECLAIM;
713 (void) snprintf(name, sizeof (name),
714 "%s_%s", zio_type_name[t], zio_taskq_types[q]);
716 spa->spa_zio_taskq[t][q] =
717 spa_taskq_create(spa, name, mode, value, flags);
722 #if defined(_KERNEL) && defined(HAVE_SPA_THREAD)
724 spa_thread(void *arg)
729 user_t *pu = PTOU(curproc);
731 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
734 ASSERT(curproc != &p0);
735 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
736 "zpool-%s", spa->spa_name);
737 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
739 /* bind this thread to the requested psrset */
740 if (zio_taskq_psrset_bind != PS_NONE) {
742 mutex_enter(&cpu_lock);
743 mutex_enter(&pidlock);
744 mutex_enter(&curproc->p_lock);
746 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
747 0, NULL, NULL) == 0) {
748 curthread->t_bind_pset = zio_taskq_psrset_bind;
751 "Couldn't bind process for zfs pool \"%s\" to "
752 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
755 mutex_exit(&curproc->p_lock);
756 mutex_exit(&pidlock);
757 mutex_exit(&cpu_lock);
761 if (zio_taskq_sysdc) {
762 sysdc_thread_enter(curthread, 100, 0);
765 spa->spa_proc = curproc;
766 spa->spa_did = curthread->t_did;
768 spa_create_zio_taskqs(spa);
770 mutex_enter(&spa->spa_proc_lock);
771 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
773 spa->spa_proc_state = SPA_PROC_ACTIVE;
774 cv_broadcast(&spa->spa_proc_cv);
776 CALLB_CPR_SAFE_BEGIN(&cprinfo);
777 while (spa->spa_proc_state == SPA_PROC_ACTIVE)
778 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
779 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
781 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
782 spa->spa_proc_state = SPA_PROC_GONE;
784 cv_broadcast(&spa->spa_proc_cv);
785 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */
787 mutex_enter(&curproc->p_lock);
793 * Activate an uninitialized pool.
796 spa_activate(spa_t *spa, int mode)
798 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
800 spa->spa_state = POOL_STATE_ACTIVE;
801 spa->spa_mode = mode;
803 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
804 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
806 /* Try to create a covering process */
807 mutex_enter(&spa->spa_proc_lock);
808 ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
809 ASSERT(spa->spa_proc == &p0);
812 #ifdef HAVE_SPA_THREAD
813 /* Only create a process if we're going to be around a while. */
814 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
815 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
817 spa->spa_proc_state = SPA_PROC_CREATED;
818 while (spa->spa_proc_state == SPA_PROC_CREATED) {
819 cv_wait(&spa->spa_proc_cv,
820 &spa->spa_proc_lock);
822 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
823 ASSERT(spa->spa_proc != &p0);
824 ASSERT(spa->spa_did != 0);
828 "Couldn't create process for zfs pool \"%s\"\n",
833 #endif /* HAVE_SPA_THREAD */
834 mutex_exit(&spa->spa_proc_lock);
836 /* If we didn't create a process, we need to create our taskqs. */
837 if (spa->spa_proc == &p0) {
838 spa_create_zio_taskqs(spa);
841 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
842 offsetof(vdev_t, vdev_config_dirty_node));
843 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
844 offsetof(vdev_t, vdev_state_dirty_node));
846 txg_list_create(&spa->spa_vdev_txg_list,
847 offsetof(struct vdev, vdev_txg_node));
849 avl_create(&spa->spa_errlist_scrub,
850 spa_error_entry_compare, sizeof (spa_error_entry_t),
851 offsetof(spa_error_entry_t, se_avl));
852 avl_create(&spa->spa_errlist_last,
853 spa_error_entry_compare, sizeof (spa_error_entry_t),
854 offsetof(spa_error_entry_t, se_avl));
858 * Opposite of spa_activate().
861 spa_deactivate(spa_t *spa)
865 ASSERT(spa->spa_sync_on == B_FALSE);
866 ASSERT(spa->spa_dsl_pool == NULL);
867 ASSERT(spa->spa_root_vdev == NULL);
868 ASSERT(spa->spa_async_zio_root == NULL);
869 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
871 txg_list_destroy(&spa->spa_vdev_txg_list);
873 list_destroy(&spa->spa_config_dirty_list);
874 list_destroy(&spa->spa_state_dirty_list);
876 for (t = 0; t < ZIO_TYPES; t++) {
877 for (q = 0; q < ZIO_TASKQ_TYPES; q++) {
878 if (spa->spa_zio_taskq[t][q] != NULL)
879 taskq_destroy(spa->spa_zio_taskq[t][q]);
880 spa->spa_zio_taskq[t][q] = NULL;
884 metaslab_class_destroy(spa->spa_normal_class);
885 spa->spa_normal_class = NULL;
887 metaslab_class_destroy(spa->spa_log_class);
888 spa->spa_log_class = NULL;
891 * If this was part of an import or the open otherwise failed, we may
892 * still have errors left in the queues. Empty them just in case.
894 spa_errlog_drain(spa);
896 avl_destroy(&spa->spa_errlist_scrub);
897 avl_destroy(&spa->spa_errlist_last);
899 spa->spa_state = POOL_STATE_UNINITIALIZED;
901 mutex_enter(&spa->spa_proc_lock);
902 if (spa->spa_proc_state != SPA_PROC_NONE) {
903 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
904 spa->spa_proc_state = SPA_PROC_DEACTIVATE;
905 cv_broadcast(&spa->spa_proc_cv);
906 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
907 ASSERT(spa->spa_proc != &p0);
908 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
910 ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
911 spa->spa_proc_state = SPA_PROC_NONE;
913 ASSERT(spa->spa_proc == &p0);
914 mutex_exit(&spa->spa_proc_lock);
917 * We want to make sure spa_thread() has actually exited the ZFS
918 * module, so that the module can't be unloaded out from underneath
921 if (spa->spa_did != 0) {
922 thread_join(spa->spa_did);
928 * Verify a pool configuration, and construct the vdev tree appropriately. This
929 * will create all the necessary vdevs in the appropriate layout, with each vdev
930 * in the CLOSED state. This will prep the pool before open/creation/import.
931 * All vdev validation is done by the vdev_alloc() routine.
934 spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
935 uint_t id, int atype)
942 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
945 if ((*vdp)->vdev_ops->vdev_op_leaf)
948 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
960 for (c = 0; c < children; c++) {
962 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
970 ASSERT(*vdp != NULL);
976 * Opposite of spa_load().
979 spa_unload(spa_t *spa)
983 ASSERT(MUTEX_HELD(&spa_namespace_lock));
988 spa_async_suspend(spa);
993 if (spa->spa_sync_on) {
994 txg_sync_stop(spa->spa_dsl_pool);
995 spa->spa_sync_on = B_FALSE;
999 * Wait for any outstanding async I/O to complete.
1001 if (spa->spa_async_zio_root != NULL) {
1002 (void) zio_wait(spa->spa_async_zio_root);
1003 spa->spa_async_zio_root = NULL;
1006 bpobj_close(&spa->spa_deferred_bpobj);
1009 * Close the dsl pool.
1011 if (spa->spa_dsl_pool) {
1012 dsl_pool_close(spa->spa_dsl_pool);
1013 spa->spa_dsl_pool = NULL;
1014 spa->spa_meta_objset = NULL;
1019 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1022 * Drop and purge level 2 cache
1024 spa_l2cache_drop(spa);
1029 if (spa->spa_root_vdev)
1030 vdev_free(spa->spa_root_vdev);
1031 ASSERT(spa->spa_root_vdev == NULL);
1033 for (i = 0; i < spa->spa_spares.sav_count; i++)
1034 vdev_free(spa->spa_spares.sav_vdevs[i]);
1035 if (spa->spa_spares.sav_vdevs) {
1036 kmem_free(spa->spa_spares.sav_vdevs,
1037 spa->spa_spares.sav_count * sizeof (void *));
1038 spa->spa_spares.sav_vdevs = NULL;
1040 if (spa->spa_spares.sav_config) {
1041 nvlist_free(spa->spa_spares.sav_config);
1042 spa->spa_spares.sav_config = NULL;
1044 spa->spa_spares.sav_count = 0;
1046 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1047 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1048 vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1050 if (spa->spa_l2cache.sav_vdevs) {
1051 kmem_free(spa->spa_l2cache.sav_vdevs,
1052 spa->spa_l2cache.sav_count * sizeof (void *));
1053 spa->spa_l2cache.sav_vdevs = NULL;
1055 if (spa->spa_l2cache.sav_config) {
1056 nvlist_free(spa->spa_l2cache.sav_config);
1057 spa->spa_l2cache.sav_config = NULL;
1059 spa->spa_l2cache.sav_count = 0;
1061 spa->spa_async_suspended = 0;
1063 spa_config_exit(spa, SCL_ALL, FTAG);
1067 * Load (or re-load) the current list of vdevs describing the active spares for
1068 * this pool. When this is called, we have some form of basic information in
1069 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1070 * then re-generate a more complete list including status information.
1073 spa_load_spares(spa_t *spa)
1080 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1083 * First, close and free any existing spare vdevs.
1085 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1086 vd = spa->spa_spares.sav_vdevs[i];
1088 /* Undo the call to spa_activate() below */
1089 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1090 B_FALSE)) != NULL && tvd->vdev_isspare)
1091 spa_spare_remove(tvd);
1096 if (spa->spa_spares.sav_vdevs)
1097 kmem_free(spa->spa_spares.sav_vdevs,
1098 spa->spa_spares.sav_count * sizeof (void *));
1100 if (spa->spa_spares.sav_config == NULL)
1103 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1104 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1106 spa->spa_spares.sav_count = (int)nspares;
1107 spa->spa_spares.sav_vdevs = NULL;
1113 * Construct the array of vdevs, opening them to get status in the
1114 * process. For each spare, there is potentially two different vdev_t
1115 * structures associated with it: one in the list of spares (used only
1116 * for basic validation purposes) and one in the active vdev
1117 * configuration (if it's spared in). During this phase we open and
1118 * validate each vdev on the spare list. If the vdev also exists in the
1119 * active configuration, then we also mark this vdev as an active spare.
1121 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1123 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1124 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1125 VDEV_ALLOC_SPARE) == 0);
1128 spa->spa_spares.sav_vdevs[i] = vd;
1130 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1131 B_FALSE)) != NULL) {
1132 if (!tvd->vdev_isspare)
1136 * We only mark the spare active if we were successfully
1137 * able to load the vdev. Otherwise, importing a pool
1138 * with a bad active spare would result in strange
1139 * behavior, because multiple pool would think the spare
1140 * is actively in use.
1142 * There is a vulnerability here to an equally bizarre
1143 * circumstance, where a dead active spare is later
1144 * brought back to life (onlined or otherwise). Given
1145 * the rarity of this scenario, and the extra complexity
1146 * it adds, we ignore the possibility.
1148 if (!vdev_is_dead(tvd))
1149 spa_spare_activate(tvd);
1153 vd->vdev_aux = &spa->spa_spares;
1155 if (vdev_open(vd) != 0)
1158 if (vdev_validate_aux(vd) == 0)
1163 * Recompute the stashed list of spares, with status information
1166 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1167 DATA_TYPE_NVLIST_ARRAY) == 0);
1169 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1171 for (i = 0; i < spa->spa_spares.sav_count; i++)
1172 spares[i] = vdev_config_generate(spa,
1173 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1174 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1175 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1176 for (i = 0; i < spa->spa_spares.sav_count; i++)
1177 nvlist_free(spares[i]);
1178 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1182 * Load (or re-load) the current list of vdevs describing the active l2cache for
1183 * this pool. When this is called, we have some form of basic information in
1184 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1185 * then re-generate a more complete list including status information.
1186 * Devices which are already active have their details maintained, and are
1190 spa_load_l2cache(spa_t *spa)
1194 int i, j, oldnvdevs;
1196 vdev_t *vd, **oldvdevs, **newvdevs = NULL;
1197 spa_aux_vdev_t *sav = &spa->spa_l2cache;
1199 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1201 if (sav->sav_config != NULL) {
1202 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1203 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1204 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1209 oldvdevs = sav->sav_vdevs;
1210 oldnvdevs = sav->sav_count;
1211 sav->sav_vdevs = NULL;
1215 * Process new nvlist of vdevs.
1217 for (i = 0; i < nl2cache; i++) {
1218 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1222 for (j = 0; j < oldnvdevs; j++) {
1224 if (vd != NULL && guid == vd->vdev_guid) {
1226 * Retain previous vdev for add/remove ops.
1234 if (newvdevs[i] == NULL) {
1238 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1239 VDEV_ALLOC_L2CACHE) == 0);
1244 * Commit this vdev as an l2cache device,
1245 * even if it fails to open.
1247 spa_l2cache_add(vd);
1252 spa_l2cache_activate(vd);
1254 if (vdev_open(vd) != 0)
1257 (void) vdev_validate_aux(vd);
1259 if (!vdev_is_dead(vd))
1260 l2arc_add_vdev(spa, vd);
1265 * Purge vdevs that were dropped
1267 for (i = 0; i < oldnvdevs; i++) {
1272 ASSERT(vd->vdev_isl2cache);
1274 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1275 pool != 0ULL && l2arc_vdev_present(vd))
1276 l2arc_remove_vdev(vd);
1277 vdev_clear_stats(vd);
1283 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1285 if (sav->sav_config == NULL)
1288 sav->sav_vdevs = newvdevs;
1289 sav->sav_count = (int)nl2cache;
1292 * Recompute the stashed list of l2cache devices, with status
1293 * information this time.
1295 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1296 DATA_TYPE_NVLIST_ARRAY) == 0);
1298 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1299 for (i = 0; i < sav->sav_count; i++)
1300 l2cache[i] = vdev_config_generate(spa,
1301 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1302 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1303 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1305 for (i = 0; i < sav->sav_count; i++)
1306 nvlist_free(l2cache[i]);
1308 kmem_free(l2cache, sav->sav_count * sizeof (void *));
1312 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1315 char *packed = NULL;
1320 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
1321 nvsize = *(uint64_t *)db->db_data;
1322 dmu_buf_rele(db, FTAG);
1324 packed = kmem_alloc(nvsize, KM_SLEEP | KM_NODEBUG);
1325 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1328 error = nvlist_unpack(packed, nvsize, value, 0);
1329 kmem_free(packed, nvsize);
1335 * Checks to see if the given vdev could not be opened, in which case we post a
1336 * sysevent to notify the autoreplace code that the device has been removed.
1339 spa_check_removed(vdev_t *vd)
1343 for (c = 0; c < vd->vdev_children; c++)
1344 spa_check_removed(vd->vdev_child[c]);
1346 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd)) {
1347 zfs_ereport_post(FM_EREPORT_RESOURCE_AUTOREPLACE,
1348 vd->vdev_spa, vd, NULL, 0, 0);
1349 spa_event_notify(vd->vdev_spa, vd, FM_EREPORT_ZFS_DEVICE_CHECK);
1354 * Validate the current config against the MOS config
1357 spa_config_valid(spa_t *spa, nvlist_t *config)
1359 vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1363 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1365 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1366 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1368 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1371 * If we're doing a normal import, then build up any additional
1372 * diagnostic information about missing devices in this config.
1373 * We'll pass this up to the user for further processing.
1375 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1376 nvlist_t **child, *nv;
1379 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1381 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1383 for (c = 0; c < rvd->vdev_children; c++) {
1384 vdev_t *tvd = rvd->vdev_child[c];
1385 vdev_t *mtvd = mrvd->vdev_child[c];
1387 if (tvd->vdev_ops == &vdev_missing_ops &&
1388 mtvd->vdev_ops != &vdev_missing_ops &&
1390 child[idx++] = vdev_config_generate(spa, mtvd,
1395 VERIFY(nvlist_add_nvlist_array(nv,
1396 ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1397 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1398 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1400 for (i = 0; i < idx; i++)
1401 nvlist_free(child[i]);
1404 kmem_free(child, rvd->vdev_children * sizeof (char **));
1408 * Compare the root vdev tree with the information we have
1409 * from the MOS config (mrvd). Check each top-level vdev
1410 * with the corresponding MOS config top-level (mtvd).
1412 for (c = 0; c < rvd->vdev_children; c++) {
1413 vdev_t *tvd = rvd->vdev_child[c];
1414 vdev_t *mtvd = mrvd->vdev_child[c];
1417 * Resolve any "missing" vdevs in the current configuration.
1418 * If we find that the MOS config has more accurate information
1419 * about the top-level vdev then use that vdev instead.
1421 if (tvd->vdev_ops == &vdev_missing_ops &&
1422 mtvd->vdev_ops != &vdev_missing_ops) {
1424 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1428 * Device specific actions.
1430 if (mtvd->vdev_islog) {
1431 spa_set_log_state(spa, SPA_LOG_CLEAR);
1434 * XXX - once we have 'readonly' pool
1435 * support we should be able to handle
1436 * missing data devices by transitioning
1437 * the pool to readonly.
1443 * Swap the missing vdev with the data we were
1444 * able to obtain from the MOS config.
1446 vdev_remove_child(rvd, tvd);
1447 vdev_remove_child(mrvd, mtvd);
1449 vdev_add_child(rvd, mtvd);
1450 vdev_add_child(mrvd, tvd);
1452 spa_config_exit(spa, SCL_ALL, FTAG);
1454 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1457 } else if (mtvd->vdev_islog) {
1459 * Load the slog device's state from the MOS config
1460 * since it's possible that the label does not
1461 * contain the most up-to-date information.
1463 vdev_load_log_state(tvd, mtvd);
1468 spa_config_exit(spa, SCL_ALL, FTAG);
1471 * Ensure we were able to validate the config.
1473 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1477 * Check for missing log devices
1480 spa_check_logs(spa_t *spa)
1482 switch (spa->spa_log_state) {
1485 case SPA_LOG_MISSING:
1486 /* need to recheck in case slog has been restored */
1487 case SPA_LOG_UNKNOWN:
1488 if (dmu_objset_find(spa->spa_name, zil_check_log_chain, NULL,
1489 DS_FIND_CHILDREN)) {
1490 spa_set_log_state(spa, SPA_LOG_MISSING);
1499 spa_passivate_log(spa_t *spa)
1501 vdev_t *rvd = spa->spa_root_vdev;
1502 boolean_t slog_found = B_FALSE;
1505 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1507 if (!spa_has_slogs(spa))
1510 for (c = 0; c < rvd->vdev_children; c++) {
1511 vdev_t *tvd = rvd->vdev_child[c];
1512 metaslab_group_t *mg = tvd->vdev_mg;
1514 if (tvd->vdev_islog) {
1515 metaslab_group_passivate(mg);
1516 slog_found = B_TRUE;
1520 return (slog_found);
1524 spa_activate_log(spa_t *spa)
1526 vdev_t *rvd = spa->spa_root_vdev;
1529 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1531 for (c = 0; c < rvd->vdev_children; c++) {
1532 vdev_t *tvd = rvd->vdev_child[c];
1533 metaslab_group_t *mg = tvd->vdev_mg;
1535 if (tvd->vdev_islog)
1536 metaslab_group_activate(mg);
1541 spa_offline_log(spa_t *spa)
1545 if ((error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1546 NULL, DS_FIND_CHILDREN)) == 0) {
1549 * We successfully offlined the log device, sync out the
1550 * current txg so that the "stubby" block can be removed
1553 txg_wait_synced(spa->spa_dsl_pool, 0);
1559 spa_aux_check_removed(spa_aux_vdev_t *sav)
1563 for (i = 0; i < sav->sav_count; i++)
1564 spa_check_removed(sav->sav_vdevs[i]);
1568 spa_claim_notify(zio_t *zio)
1570 spa_t *spa = zio->io_spa;
1575 mutex_enter(&spa->spa_props_lock); /* any mutex will do */
1576 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1577 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1578 mutex_exit(&spa->spa_props_lock);
1581 typedef struct spa_load_error {
1582 uint64_t sle_meta_count;
1583 uint64_t sle_data_count;
1587 spa_load_verify_done(zio_t *zio)
1589 blkptr_t *bp = zio->io_bp;
1590 spa_load_error_t *sle = zio->io_private;
1591 dmu_object_type_t type = BP_GET_TYPE(bp);
1592 int error = zio->io_error;
1595 if ((BP_GET_LEVEL(bp) != 0 || dmu_ot[type].ot_metadata) &&
1596 type != DMU_OT_INTENT_LOG)
1597 atomic_add_64(&sle->sle_meta_count, 1);
1599 atomic_add_64(&sle->sle_data_count, 1);
1601 zio_data_buf_free(zio->io_data, zio->io_size);
1606 spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1607 arc_buf_t *pbuf, const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1611 size_t size = BP_GET_PSIZE(bp);
1612 void *data = zio_data_buf_alloc(size);
1614 zio_nowait(zio_read(rio, spa, bp, data, size,
1615 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1616 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1617 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1623 spa_load_verify(spa_t *spa)
1626 spa_load_error_t sle = { 0 };
1627 zpool_rewind_policy_t policy;
1628 boolean_t verify_ok = B_FALSE;
1631 zpool_get_rewind_policy(spa->spa_config, &policy);
1633 if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1636 rio = zio_root(spa, NULL, &sle,
1637 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1639 error = traverse_pool(spa, spa->spa_verify_min_txg,
1640 TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio);
1642 (void) zio_wait(rio);
1644 spa->spa_load_meta_errors = sle.sle_meta_count;
1645 spa->spa_load_data_errors = sle.sle_data_count;
1647 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1648 sle.sle_data_count <= policy.zrp_maxdata) {
1652 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1653 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1655 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1656 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1657 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1658 VERIFY(nvlist_add_int64(spa->spa_load_info,
1659 ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1660 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1661 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1663 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1667 if (error != ENXIO && error != EIO)
1672 return (verify_ok ? 0 : EIO);
1676 * Find a value in the pool props object.
1679 spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1681 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1682 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1686 * Find a value in the pool directory object.
1689 spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1691 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1692 name, sizeof (uint64_t), 1, val));
1696 spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1698 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1703 * Fix up config after a partly-completed split. This is done with the
1704 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
1705 * pool have that entry in their config, but only the splitting one contains
1706 * a list of all the guids of the vdevs that are being split off.
1708 * This function determines what to do with that list: either rejoin
1709 * all the disks to the pool, or complete the splitting process. To attempt
1710 * the rejoin, each disk that is offlined is marked online again, and
1711 * we do a reopen() call. If the vdev label for every disk that was
1712 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1713 * then we call vdev_split() on each disk, and complete the split.
1715 * Otherwise we leave the config alone, with all the vdevs in place in
1716 * the original pool.
1719 spa_try_repair(spa_t *spa, nvlist_t *config)
1726 boolean_t attempt_reopen;
1728 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
1731 /* check that the config is complete */
1732 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
1733 &glist, &gcount) != 0)
1736 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
1738 /* attempt to online all the vdevs & validate */
1739 attempt_reopen = B_TRUE;
1740 for (i = 0; i < gcount; i++) {
1741 if (glist[i] == 0) /* vdev is hole */
1744 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
1745 if (vd[i] == NULL) {
1747 * Don't bother attempting to reopen the disks;
1748 * just do the split.
1750 attempt_reopen = B_FALSE;
1752 /* attempt to re-online it */
1753 vd[i]->vdev_offline = B_FALSE;
1757 if (attempt_reopen) {
1758 vdev_reopen(spa->spa_root_vdev);
1760 /* check each device to see what state it's in */
1761 for (extracted = 0, i = 0; i < gcount; i++) {
1762 if (vd[i] != NULL &&
1763 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
1770 * If every disk has been moved to the new pool, or if we never
1771 * even attempted to look at them, then we split them off for
1774 if (!attempt_reopen || gcount == extracted) {
1775 for (i = 0; i < gcount; i++)
1778 vdev_reopen(spa->spa_root_vdev);
1781 kmem_free(vd, gcount * sizeof (vdev_t *));
1785 spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
1786 boolean_t mosconfig)
1788 nvlist_t *config = spa->spa_config;
1789 char *ereport = FM_EREPORT_ZFS_POOL;
1794 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
1798 * Versioning wasn't explicitly added to the label until later, so if
1799 * it's not present treat it as the initial version.
1801 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
1802 &spa->spa_ubsync.ub_version) != 0)
1803 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
1805 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
1806 &spa->spa_config_txg);
1808 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
1809 spa_guid_exists(pool_guid, 0)) {
1812 spa->spa_config_guid = pool_guid;
1814 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
1816 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
1820 gethrestime(&spa->spa_loaded_ts);
1821 error = spa_load_impl(spa, pool_guid, config, state, type,
1822 mosconfig, &ereport);
1825 spa->spa_minref = refcount_count(&spa->spa_refcount);
1827 if (error != EEXIST) {
1828 spa->spa_loaded_ts.tv_sec = 0;
1829 spa->spa_loaded_ts.tv_nsec = 0;
1831 if (error != EBADF) {
1832 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
1835 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
1842 * Load an existing storage pool, using the pool's builtin spa_config as a
1843 * source of configuration information.
1845 __attribute__((always_inline))
1847 spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
1848 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
1852 nvlist_t *nvroot = NULL;
1854 uberblock_t *ub = &spa->spa_uberblock;
1855 uint64_t children, config_cache_txg = spa->spa_config_txg;
1856 int orig_mode = spa->spa_mode;
1861 * If this is an untrusted config, access the pool in read-only mode.
1862 * This prevents things like resilvering recently removed devices.
1865 spa->spa_mode = FREAD;
1867 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1869 spa->spa_load_state = state;
1871 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
1874 parse = (type == SPA_IMPORT_EXISTING ?
1875 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
1878 * Create "The Godfather" zio to hold all async IOs
1880 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
1881 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
1884 * Parse the configuration into a vdev tree. We explicitly set the
1885 * value that will be returned by spa_version() since parsing the
1886 * configuration requires knowing the version number.
1888 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1889 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
1890 spa_config_exit(spa, SCL_ALL, FTAG);
1895 ASSERT(spa->spa_root_vdev == rvd);
1897 if (type != SPA_IMPORT_ASSEMBLE) {
1898 ASSERT(spa_guid(spa) == pool_guid);
1902 * Try to open all vdevs, loading each label in the process.
1904 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1905 error = vdev_open(rvd);
1906 spa_config_exit(spa, SCL_ALL, FTAG);
1911 * We need to validate the vdev labels against the configuration that
1912 * we have in hand, which is dependent on the setting of mosconfig. If
1913 * mosconfig is true then we're validating the vdev labels based on
1914 * that config. Otherwise, we're validating against the cached config
1915 * (zpool.cache) that was read when we loaded the zfs module, and then
1916 * later we will recursively call spa_load() and validate against
1919 * If we're assembling a new pool that's been split off from an
1920 * existing pool, the labels haven't yet been updated so we skip
1921 * validation for now.
1923 if (type != SPA_IMPORT_ASSEMBLE) {
1924 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1925 error = vdev_validate(rvd);
1926 spa_config_exit(spa, SCL_ALL, FTAG);
1931 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
1936 * Find the best uberblock.
1938 vdev_uberblock_load(NULL, rvd, ub);
1941 * If we weren't able to find a single valid uberblock, return failure.
1943 if (ub->ub_txg == 0)
1944 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
1947 * If the pool is newer than the code, we can't open it.
1949 if (ub->ub_version > SPA_VERSION)
1950 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
1953 * If the vdev guid sum doesn't match the uberblock, we have an
1954 * incomplete configuration. We first check to see if the pool
1955 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
1956 * If it is, defer the vdev_guid_sum check till later so we
1957 * can handle missing vdevs.
1959 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
1960 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
1961 rvd->vdev_guid_sum != ub->ub_guid_sum)
1962 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
1964 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
1965 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1966 spa_try_repair(spa, config);
1967 spa_config_exit(spa, SCL_ALL, FTAG);
1968 nvlist_free(spa->spa_config_splitting);
1969 spa->spa_config_splitting = NULL;
1973 * Initialize internal SPA structures.
1975 spa->spa_state = POOL_STATE_ACTIVE;
1976 spa->spa_ubsync = spa->spa_uberblock;
1977 spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
1978 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
1979 spa->spa_first_txg = spa->spa_last_ubsync_txg ?
1980 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
1981 spa->spa_claim_max_txg = spa->spa_first_txg;
1982 spa->spa_prev_software_version = ub->ub_software_version;
1984 error = dsl_pool_open(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
1986 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1987 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
1989 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
1990 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1994 nvlist_t *policy = NULL, *nvconfig;
1996 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
1997 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1999 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2000 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2002 unsigned long myhostid = 0;
2004 VERIFY(nvlist_lookup_string(nvconfig,
2005 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2008 myhostid = zone_get_hostid(NULL);
2011 * We're emulating the system's hostid in userland, so
2012 * we can't use zone_get_hostid().
2014 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2015 #endif /* _KERNEL */
2016 if (hostid != 0 && myhostid != 0 &&
2017 hostid != myhostid) {
2018 nvlist_free(nvconfig);
2019 cmn_err(CE_WARN, "pool '%s' could not be "
2020 "loaded as it was last accessed by "
2021 "another system (host: %s hostid: 0x%lx). "
2022 "See: http://zfsonlinux.org/msg/ZFS-8000-EY",
2023 spa_name(spa), hostname,
2024 (unsigned long)hostid);
2028 if (nvlist_lookup_nvlist(spa->spa_config,
2029 ZPOOL_REWIND_POLICY, &policy) == 0)
2030 VERIFY(nvlist_add_nvlist(nvconfig,
2031 ZPOOL_REWIND_POLICY, policy) == 0);
2033 spa_config_set(spa, nvconfig);
2035 spa_deactivate(spa);
2036 spa_activate(spa, orig_mode);
2038 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2041 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2042 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2043 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2045 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2048 * Load the bit that tells us to use the new accounting function
2049 * (raid-z deflation). If we have an older pool, this will not
2052 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2053 if (error != 0 && error != ENOENT)
2054 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2056 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2057 &spa->spa_creation_version);
2058 if (error != 0 && error != ENOENT)
2059 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2062 * Load the persistent error log. If we have an older pool, this will
2065 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2066 if (error != 0 && error != ENOENT)
2067 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2069 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2070 &spa->spa_errlog_scrub);
2071 if (error != 0 && error != ENOENT)
2072 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2075 * Load the history object. If we have an older pool, this
2076 * will not be present.
2078 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2079 if (error != 0 && error != ENOENT)
2080 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2083 * If we're assembling the pool from the split-off vdevs of
2084 * an existing pool, we don't want to attach the spares & cache
2089 * Load any hot spares for this pool.
2091 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2092 if (error != 0 && error != ENOENT)
2093 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2094 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2095 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2096 if (load_nvlist(spa, spa->spa_spares.sav_object,
2097 &spa->spa_spares.sav_config) != 0)
2098 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2100 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2101 spa_load_spares(spa);
2102 spa_config_exit(spa, SCL_ALL, FTAG);
2103 } else if (error == 0) {
2104 spa->spa_spares.sav_sync = B_TRUE;
2108 * Load any level 2 ARC devices for this pool.
2110 error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2111 &spa->spa_l2cache.sav_object);
2112 if (error != 0 && error != ENOENT)
2113 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2114 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2115 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2116 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2117 &spa->spa_l2cache.sav_config) != 0)
2118 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2120 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2121 spa_load_l2cache(spa);
2122 spa_config_exit(spa, SCL_ALL, FTAG);
2123 } else if (error == 0) {
2124 spa->spa_l2cache.sav_sync = B_TRUE;
2127 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2129 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2130 if (error && error != ENOENT)
2131 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2134 uint64_t autoreplace;
2136 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2137 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2138 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2139 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2140 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2141 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2142 &spa->spa_dedup_ditto);
2144 spa->spa_autoreplace = (autoreplace != 0);
2148 * If the 'autoreplace' property is set, then post a resource notifying
2149 * the ZFS DE that it should not issue any faults for unopenable
2150 * devices. We also iterate over the vdevs, and post a sysevent for any
2151 * unopenable vdevs so that the normal autoreplace handler can take
2154 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2155 spa_check_removed(spa->spa_root_vdev);
2157 * For the import case, this is done in spa_import(), because
2158 * at this point we're using the spare definitions from
2159 * the MOS config, not necessarily from the userland config.
2161 if (state != SPA_LOAD_IMPORT) {
2162 spa_aux_check_removed(&spa->spa_spares);
2163 spa_aux_check_removed(&spa->spa_l2cache);
2168 * Load the vdev state for all toplevel vdevs.
2173 * Propagate the leaf DTLs we just loaded all the way up the tree.
2175 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2176 vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2177 spa_config_exit(spa, SCL_ALL, FTAG);
2180 * Load the DDTs (dedup tables).
2182 error = ddt_load(spa);
2184 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2186 spa_update_dspace(spa);
2189 * Validate the config, using the MOS config to fill in any
2190 * information which might be missing. If we fail to validate
2191 * the config then declare the pool unfit for use. If we're
2192 * assembling a pool from a split, the log is not transferred
2195 if (type != SPA_IMPORT_ASSEMBLE) {
2198 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2199 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2201 if (!spa_config_valid(spa, nvconfig)) {
2202 nvlist_free(nvconfig);
2203 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2206 nvlist_free(nvconfig);
2209 * Now that we've validate the config, check the state of the
2210 * root vdev. If it can't be opened, it indicates one or
2211 * more toplevel vdevs are faulted.
2213 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2216 if (spa_check_logs(spa)) {
2217 *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2218 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2223 * We've successfully opened the pool, verify that we're ready
2224 * to start pushing transactions.
2226 if (state != SPA_LOAD_TRYIMPORT) {
2227 if ((error = spa_load_verify(spa)))
2228 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2232 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2233 spa->spa_load_max_txg == UINT64_MAX)) {
2235 int need_update = B_FALSE;
2238 ASSERT(state != SPA_LOAD_TRYIMPORT);
2241 * Claim log blocks that haven't been committed yet.
2242 * This must all happen in a single txg.
2243 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2244 * invoked from zil_claim_log_block()'s i/o done callback.
2245 * Price of rollback is that we abandon the log.
2247 spa->spa_claiming = B_TRUE;
2249 tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2250 spa_first_txg(spa));
2251 (void) dmu_objset_find(spa_name(spa),
2252 zil_claim, tx, DS_FIND_CHILDREN);
2255 spa->spa_claiming = B_FALSE;
2257 spa_set_log_state(spa, SPA_LOG_GOOD);
2258 spa->spa_sync_on = B_TRUE;
2259 txg_sync_start(spa->spa_dsl_pool);
2262 * Wait for all claims to sync. We sync up to the highest
2263 * claimed log block birth time so that claimed log blocks
2264 * don't appear to be from the future. spa_claim_max_txg
2265 * will have been set for us by either zil_check_log_chain()
2266 * (invoked from spa_check_logs()) or zil_claim() above.
2268 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2271 * If the config cache is stale, or we have uninitialized
2272 * metaslabs (see spa_vdev_add()), then update the config.
2274 * If this is a verbatim import, trust the current
2275 * in-core spa_config and update the disk labels.
2277 if (config_cache_txg != spa->spa_config_txg ||
2278 state == SPA_LOAD_IMPORT ||
2279 state == SPA_LOAD_RECOVER ||
2280 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2281 need_update = B_TRUE;
2283 for (c = 0; c < rvd->vdev_children; c++)
2284 if (rvd->vdev_child[c]->vdev_ms_array == 0)
2285 need_update = B_TRUE;
2288 * Update the config cache asychronously in case we're the
2289 * root pool, in which case the config cache isn't writable yet.
2292 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2295 * Check all DTLs to see if anything needs resilvering.
2297 if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2298 vdev_resilver_needed(rvd, NULL, NULL))
2299 spa_async_request(spa, SPA_ASYNC_RESILVER);
2302 * Delete any inconsistent datasets.
2304 (void) dmu_objset_find(spa_name(spa),
2305 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2308 * Clean up any stale temporary dataset userrefs.
2310 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2317 spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2319 int mode = spa->spa_mode;
2322 spa_deactivate(spa);
2324 spa->spa_load_max_txg--;
2326 spa_activate(spa, mode);
2327 spa_async_suspend(spa);
2329 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2333 spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2334 uint64_t max_request, int rewind_flags)
2336 nvlist_t *config = NULL;
2337 int load_error, rewind_error;
2338 uint64_t safe_rewind_txg;
2341 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2342 spa->spa_load_max_txg = spa->spa_load_txg;
2343 spa_set_log_state(spa, SPA_LOG_CLEAR);
2345 spa->spa_load_max_txg = max_request;
2348 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2350 if (load_error == 0)
2353 if (spa->spa_root_vdev != NULL)
2354 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2356 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2357 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2359 if (rewind_flags & ZPOOL_NEVER_REWIND) {
2360 nvlist_free(config);
2361 return (load_error);
2364 /* Price of rolling back is discarding txgs, including log */
2365 if (state == SPA_LOAD_RECOVER)
2366 spa_set_log_state(spa, SPA_LOG_CLEAR);
2368 spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2369 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2370 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2371 TXG_INITIAL : safe_rewind_txg;
2374 * Continue as long as we're finding errors, we're still within
2375 * the acceptable rewind range, and we're still finding uberblocks
2377 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2378 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2379 if (spa->spa_load_max_txg < safe_rewind_txg)
2380 spa->spa_extreme_rewind = B_TRUE;
2381 rewind_error = spa_load_retry(spa, state, mosconfig);
2384 spa->spa_extreme_rewind = B_FALSE;
2385 spa->spa_load_max_txg = UINT64_MAX;
2387 if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2388 spa_config_set(spa, config);
2390 return (state == SPA_LOAD_RECOVER ? rewind_error : load_error);
2396 * The import case is identical to an open except that the configuration is sent
2397 * down from userland, instead of grabbed from the configuration cache. For the
2398 * case of an open, the pool configuration will exist in the
2399 * POOL_STATE_UNINITIALIZED state.
2401 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2402 * the same time open the pool, without having to keep around the spa_t in some
2406 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2410 spa_load_state_t state = SPA_LOAD_OPEN;
2412 int locked = B_FALSE;
2417 * As disgusting as this is, we need to support recursive calls to this
2418 * function because dsl_dir_open() is called during spa_load(), and ends
2419 * up calling spa_open() again. The real fix is to figure out how to
2420 * avoid dsl_dir_open() calling this in the first place.
2422 if (mutex_owner(&spa_namespace_lock) != curthread) {
2423 mutex_enter(&spa_namespace_lock);
2427 if ((spa = spa_lookup(pool)) == NULL) {
2429 mutex_exit(&spa_namespace_lock);
2433 if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2434 zpool_rewind_policy_t policy;
2436 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2438 if (policy.zrp_request & ZPOOL_DO_REWIND)
2439 state = SPA_LOAD_RECOVER;
2441 spa_activate(spa, spa_mode_global);
2443 if (state != SPA_LOAD_RECOVER)
2444 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2446 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2447 policy.zrp_request);
2449 if (error == EBADF) {
2451 * If vdev_validate() returns failure (indicated by
2452 * EBADF), it indicates that one of the vdevs indicates
2453 * that the pool has been exported or destroyed. If
2454 * this is the case, the config cache is out of sync and
2455 * we should remove the pool from the namespace.
2458 spa_deactivate(spa);
2459 spa_config_sync(spa, B_TRUE, B_TRUE);
2462 mutex_exit(&spa_namespace_lock);
2468 * We can't open the pool, but we still have useful
2469 * information: the state of each vdev after the
2470 * attempted vdev_open(). Return this to the user.
2472 if (config != NULL && spa->spa_config) {
2473 VERIFY(nvlist_dup(spa->spa_config, config,
2475 VERIFY(nvlist_add_nvlist(*config,
2476 ZPOOL_CONFIG_LOAD_INFO,
2477 spa->spa_load_info) == 0);
2480 spa_deactivate(spa);
2481 spa->spa_last_open_failed = error;
2483 mutex_exit(&spa_namespace_lock);
2489 spa_open_ref(spa, tag);
2492 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2495 * If we've recovered the pool, pass back any information we
2496 * gathered while doing the load.
2498 if (state == SPA_LOAD_RECOVER) {
2499 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
2500 spa->spa_load_info) == 0);
2504 spa->spa_last_open_failed = 0;
2505 spa->spa_last_ubsync_txg = 0;
2506 spa->spa_load_txg = 0;
2507 mutex_exit(&spa_namespace_lock);
2516 spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
2519 return (spa_open_common(name, spapp, tag, policy, config));
2523 spa_open(const char *name, spa_t **spapp, void *tag)
2525 return (spa_open_common(name, spapp, tag, NULL, NULL));
2529 * Lookup the given spa_t, incrementing the inject count in the process,
2530 * preventing it from being exported or destroyed.
2533 spa_inject_addref(char *name)
2537 mutex_enter(&spa_namespace_lock);
2538 if ((spa = spa_lookup(name)) == NULL) {
2539 mutex_exit(&spa_namespace_lock);
2542 spa->spa_inject_ref++;
2543 mutex_exit(&spa_namespace_lock);
2549 spa_inject_delref(spa_t *spa)
2551 mutex_enter(&spa_namespace_lock);
2552 spa->spa_inject_ref--;
2553 mutex_exit(&spa_namespace_lock);
2557 * Add spares device information to the nvlist.
2560 spa_add_spares(spa_t *spa, nvlist_t *config)
2570 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2572 if (spa->spa_spares.sav_count == 0)
2575 VERIFY(nvlist_lookup_nvlist(config,
2576 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2577 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
2578 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2580 VERIFY(nvlist_add_nvlist_array(nvroot,
2581 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2582 VERIFY(nvlist_lookup_nvlist_array(nvroot,
2583 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2586 * Go through and find any spares which have since been
2587 * repurposed as an active spare. If this is the case, update
2588 * their status appropriately.
2590 for (i = 0; i < nspares; i++) {
2591 VERIFY(nvlist_lookup_uint64(spares[i],
2592 ZPOOL_CONFIG_GUID, &guid) == 0);
2593 if (spa_spare_exists(guid, &pool, NULL) &&
2595 VERIFY(nvlist_lookup_uint64_array(
2596 spares[i], ZPOOL_CONFIG_VDEV_STATS,
2597 (uint64_t **)&vs, &vsc) == 0);
2598 vs->vs_state = VDEV_STATE_CANT_OPEN;
2599 vs->vs_aux = VDEV_AUX_SPARED;
2606 * Add l2cache device information to the nvlist, including vdev stats.
2609 spa_add_l2cache(spa_t *spa, nvlist_t *config)
2612 uint_t i, j, nl2cache;
2619 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2621 if (spa->spa_l2cache.sav_count == 0)
2624 VERIFY(nvlist_lookup_nvlist(config,
2625 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2626 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
2627 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2628 if (nl2cache != 0) {
2629 VERIFY(nvlist_add_nvlist_array(nvroot,
2630 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2631 VERIFY(nvlist_lookup_nvlist_array(nvroot,
2632 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2635 * Update level 2 cache device stats.
2638 for (i = 0; i < nl2cache; i++) {
2639 VERIFY(nvlist_lookup_uint64(l2cache[i],
2640 ZPOOL_CONFIG_GUID, &guid) == 0);
2643 for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
2645 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
2646 vd = spa->spa_l2cache.sav_vdevs[j];
2652 VERIFY(nvlist_lookup_uint64_array(l2cache[i],
2653 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
2655 vdev_get_stats(vd, vs);
2661 spa_get_stats(const char *name, nvlist_t **config, char *altroot, size_t buflen)
2667 error = spa_open_common(name, &spa, FTAG, NULL, config);
2671 * This still leaves a window of inconsistency where the spares
2672 * or l2cache devices could change and the config would be
2673 * self-inconsistent.
2675 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
2677 if (*config != NULL) {
2678 uint64_t loadtimes[2];
2680 loadtimes[0] = spa->spa_loaded_ts.tv_sec;
2681 loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
2682 VERIFY(nvlist_add_uint64_array(*config,
2683 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
2685 VERIFY(nvlist_add_uint64(*config,
2686 ZPOOL_CONFIG_ERRCOUNT,
2687 spa_get_errlog_size(spa)) == 0);
2689 if (spa_suspended(spa))
2690 VERIFY(nvlist_add_uint64(*config,
2691 ZPOOL_CONFIG_SUSPENDED,
2692 spa->spa_failmode) == 0);
2694 spa_add_spares(spa, *config);
2695 spa_add_l2cache(spa, *config);
2700 * We want to get the alternate root even for faulted pools, so we cheat
2701 * and call spa_lookup() directly.
2705 mutex_enter(&spa_namespace_lock);
2706 spa = spa_lookup(name);
2708 spa_altroot(spa, altroot, buflen);
2712 mutex_exit(&spa_namespace_lock);
2714 spa_altroot(spa, altroot, buflen);
2719 spa_config_exit(spa, SCL_CONFIG, FTAG);
2720 spa_close(spa, FTAG);
2727 * Validate that the auxiliary device array is well formed. We must have an
2728 * array of nvlists, each which describes a valid leaf vdev. If this is an
2729 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
2730 * specified, as long as they are well-formed.
2733 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
2734 spa_aux_vdev_t *sav, const char *config, uint64_t version,
2735 vdev_labeltype_t label)
2742 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
2745 * It's acceptable to have no devs specified.
2747 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
2754 * Make sure the pool is formatted with a version that supports this
2757 if (spa_version(spa) < version)
2761 * Set the pending device list so we correctly handle device in-use
2764 sav->sav_pending = dev;
2765 sav->sav_npending = ndev;
2767 for (i = 0; i < ndev; i++) {
2768 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
2772 if (!vd->vdev_ops->vdev_op_leaf) {
2779 * The L2ARC currently only supports disk devices in
2780 * kernel context. For user-level testing, we allow it.
2783 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
2784 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
2792 if ((error = vdev_open(vd)) == 0 &&
2793 (error = vdev_label_init(vd, crtxg, label)) == 0) {
2794 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
2795 vd->vdev_guid) == 0);
2801 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
2808 sav->sav_pending = NULL;
2809 sav->sav_npending = 0;
2814 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
2818 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
2820 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
2821 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
2822 VDEV_LABEL_SPARE)) != 0) {
2826 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
2827 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
2828 VDEV_LABEL_L2CACHE));
2832 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
2837 if (sav->sav_config != NULL) {
2843 * Generate new dev list by concatentating with the
2846 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
2847 &olddevs, &oldndevs) == 0);
2849 newdevs = kmem_alloc(sizeof (void *) *
2850 (ndevs + oldndevs), KM_SLEEP);
2851 for (i = 0; i < oldndevs; i++)
2852 VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
2854 for (i = 0; i < ndevs; i++)
2855 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
2858 VERIFY(nvlist_remove(sav->sav_config, config,
2859 DATA_TYPE_NVLIST_ARRAY) == 0);
2861 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
2862 config, newdevs, ndevs + oldndevs) == 0);
2863 for (i = 0; i < oldndevs + ndevs; i++)
2864 nvlist_free(newdevs[i]);
2865 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
2868 * Generate a new dev list.
2870 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
2872 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
2878 * Stop and drop level 2 ARC devices
2881 spa_l2cache_drop(spa_t *spa)
2885 spa_aux_vdev_t *sav = &spa->spa_l2cache;
2887 for (i = 0; i < sav->sav_count; i++) {
2890 vd = sav->sav_vdevs[i];
2893 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
2894 pool != 0ULL && l2arc_vdev_present(vd))
2895 l2arc_remove_vdev(vd);
2903 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
2904 const char *history_str, nvlist_t *zplprops)
2907 char *altroot = NULL;
2912 uint64_t txg = TXG_INITIAL;
2913 nvlist_t **spares, **l2cache;
2914 uint_t nspares, nl2cache;
2915 uint64_t version, obj;
2919 * If this pool already exists, return failure.
2921 mutex_enter(&spa_namespace_lock);
2922 if (spa_lookup(pool) != NULL) {
2923 mutex_exit(&spa_namespace_lock);
2928 * Allocate a new spa_t structure.
2930 (void) nvlist_lookup_string(props,
2931 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
2932 spa = spa_add(pool, NULL, altroot);
2933 spa_activate(spa, spa_mode_global);
2935 if (props && (error = spa_prop_validate(spa, props))) {
2936 spa_deactivate(spa);
2938 mutex_exit(&spa_namespace_lock);
2942 if (nvlist_lookup_uint64(props, zpool_prop_to_name(ZPOOL_PROP_VERSION),
2944 version = SPA_VERSION;
2945 ASSERT(version <= SPA_VERSION);
2947 spa->spa_first_txg = txg;
2948 spa->spa_uberblock.ub_txg = txg - 1;
2949 spa->spa_uberblock.ub_version = version;
2950 spa->spa_ubsync = spa->spa_uberblock;
2953 * Create "The Godfather" zio to hold all async IOs
2955 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2956 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2959 * Create the root vdev.
2961 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2963 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
2965 ASSERT(error != 0 || rvd != NULL);
2966 ASSERT(error != 0 || spa->spa_root_vdev == rvd);
2968 if (error == 0 && !zfs_allocatable_devs(nvroot))
2972 (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
2973 (error = spa_validate_aux(spa, nvroot, txg,
2974 VDEV_ALLOC_ADD)) == 0) {
2975 for (c = 0; c < rvd->vdev_children; c++) {
2976 vdev_metaslab_set_size(rvd->vdev_child[c]);
2977 vdev_expand(rvd->vdev_child[c], txg);
2981 spa_config_exit(spa, SCL_ALL, FTAG);
2985 spa_deactivate(spa);
2987 mutex_exit(&spa_namespace_lock);
2992 * Get the list of spares, if specified.
2994 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
2995 &spares, &nspares) == 0) {
2996 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
2998 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
2999 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3000 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3001 spa_load_spares(spa);
3002 spa_config_exit(spa, SCL_ALL, FTAG);
3003 spa->spa_spares.sav_sync = B_TRUE;
3007 * Get the list of level 2 cache devices, if specified.
3009 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3010 &l2cache, &nl2cache) == 0) {
3011 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3012 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3013 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3014 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3015 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3016 spa_load_l2cache(spa);
3017 spa_config_exit(spa, SCL_ALL, FTAG);
3018 spa->spa_l2cache.sav_sync = B_TRUE;
3021 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3022 spa->spa_meta_objset = dp->dp_meta_objset;
3025 * Create DDTs (dedup tables).
3029 spa_update_dspace(spa);
3031 tx = dmu_tx_create_assigned(dp, txg);
3034 * Create the pool config object.
3036 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3037 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3038 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3040 if (zap_add(spa->spa_meta_objset,
3041 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3042 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3043 cmn_err(CE_PANIC, "failed to add pool config");
3046 if (zap_add(spa->spa_meta_objset,
3047 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3048 sizeof (uint64_t), 1, &version, tx) != 0) {
3049 cmn_err(CE_PANIC, "failed to add pool version");
3052 /* Newly created pools with the right version are always deflated. */
3053 if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3054 spa->spa_deflate = TRUE;
3055 if (zap_add(spa->spa_meta_objset,
3056 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3057 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3058 cmn_err(CE_PANIC, "failed to add deflate");
3063 * Create the deferred-free bpobj. Turn off compression
3064 * because sync-to-convergence takes longer if the blocksize
3067 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3068 dmu_object_set_compress(spa->spa_meta_objset, obj,
3069 ZIO_COMPRESS_OFF, tx);
3070 if (zap_add(spa->spa_meta_objset,
3071 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3072 sizeof (uint64_t), 1, &obj, tx) != 0) {
3073 cmn_err(CE_PANIC, "failed to add bpobj");
3075 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3076 spa->spa_meta_objset, obj));
3079 * Create the pool's history object.
3081 if (version >= SPA_VERSION_ZPOOL_HISTORY)
3082 spa_history_create_obj(spa, tx);
3085 * Set pool properties.
3087 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3088 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3089 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3090 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3092 if (props != NULL) {
3093 spa_configfile_set(spa, props, B_FALSE);
3094 spa_sync_props(spa, props, tx);
3099 spa->spa_sync_on = B_TRUE;
3100 txg_sync_start(spa->spa_dsl_pool);
3103 * We explicitly wait for the first transaction to complete so that our
3104 * bean counters are appropriately updated.
3106 txg_wait_synced(spa->spa_dsl_pool, txg);
3108 spa_config_sync(spa, B_FALSE, B_TRUE);
3110 if (version >= SPA_VERSION_ZPOOL_HISTORY && history_str != NULL)
3111 (void) spa_history_log(spa, history_str, LOG_CMD_POOL_CREATE);
3112 spa_history_log_version(spa, LOG_POOL_CREATE);
3114 spa->spa_minref = refcount_count(&spa->spa_refcount);
3116 mutex_exit(&spa_namespace_lock);
3123 * Get the root pool information from the root disk, then import the root pool
3124 * during the system boot up time.
3126 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3129 spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3132 nvlist_t *nvtop, *nvroot;
3135 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3139 * Add this top-level vdev to the child array.
3141 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3143 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3145 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3148 * Put this pool's top-level vdevs into a root vdev.
3150 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3151 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3152 VDEV_TYPE_ROOT) == 0);
3153 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3154 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3155 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3159 * Replace the existing vdev_tree with the new root vdev in
3160 * this pool's configuration (remove the old, add the new).
3162 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3163 nvlist_free(nvroot);
3168 * Walk the vdev tree and see if we can find a device with "better"
3169 * configuration. A configuration is "better" if the label on that
3170 * device has a more recent txg.
3173 spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3177 for (c = 0; c < vd->vdev_children; c++)
3178 spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3180 if (vd->vdev_ops->vdev_op_leaf) {
3184 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3188 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3192 * Do we have a better boot device?
3194 if (label_txg > *txg) {
3203 * Import a root pool.
3205 * For x86. devpath_list will consist of devid and/or physpath name of
3206 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3207 * The GRUB "findroot" command will return the vdev we should boot.
3209 * For Sparc, devpath_list consists the physpath name of the booting device
3210 * no matter the rootpool is a single device pool or a mirrored pool.
3212 * "/pci@1f,0/ide@d/disk@0,0:a"
3215 spa_import_rootpool(char *devpath, char *devid)
3218 vdev_t *rvd, *bvd, *avd = NULL;
3219 nvlist_t *config, *nvtop;
3225 * Read the label from the boot device and generate a configuration.
3227 config = spa_generate_rootconf(devpath, devid, &guid);
3228 #if defined(_OBP) && defined(_KERNEL)
3229 if (config == NULL) {
3230 if (strstr(devpath, "/iscsi/ssd") != NULL) {
3232 get_iscsi_bootpath_phy(devpath);
3233 config = spa_generate_rootconf(devpath, devid, &guid);
3237 if (config == NULL) {
3238 cmn_err(CE_NOTE, "Can not read the pool label from '%s'",
3243 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3245 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3247 mutex_enter(&spa_namespace_lock);
3248 if ((spa = spa_lookup(pname)) != NULL) {
3250 * Remove the existing root pool from the namespace so that we
3251 * can replace it with the correct config we just read in.
3256 spa = spa_add(pname, config, NULL);
3257 spa->spa_is_root = B_TRUE;
3258 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3261 * Build up a vdev tree based on the boot device's label config.
3263 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3265 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3266 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3267 VDEV_ALLOC_ROOTPOOL);
3268 spa_config_exit(spa, SCL_ALL, FTAG);
3270 mutex_exit(&spa_namespace_lock);
3271 nvlist_free(config);
3272 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3278 * Get the boot vdev.
3280 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3281 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3282 (u_longlong_t)guid);
3288 * Determine if there is a better boot device.
3291 spa_alt_rootvdev(rvd, &avd, &txg);
3293 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3294 "try booting from '%s'", avd->vdev_path);
3300 * If the boot device is part of a spare vdev then ensure that
3301 * we're booting off the active spare.
3303 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3304 !bvd->vdev_isspare) {
3305 cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3306 "try booting from '%s'",
3308 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3314 spa_history_log_version(spa, LOG_POOL_IMPORT);
3316 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3318 spa_config_exit(spa, SCL_ALL, FTAG);
3319 mutex_exit(&spa_namespace_lock);
3321 nvlist_free(config);
3328 * Import a non-root pool into the system.
3331 spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
3334 char *altroot = NULL;
3335 spa_load_state_t state = SPA_LOAD_IMPORT;
3336 zpool_rewind_policy_t policy;
3337 uint64_t mode = spa_mode_global;
3338 uint64_t readonly = B_FALSE;
3341 nvlist_t **spares, **l2cache;
3342 uint_t nspares, nl2cache;
3345 * If a pool with this name exists, return failure.
3347 mutex_enter(&spa_namespace_lock);
3348 if (spa_lookup(pool) != NULL) {
3349 mutex_exit(&spa_namespace_lock);
3354 * Create and initialize the spa structure.
3356 (void) nvlist_lookup_string(props,
3357 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3358 (void) nvlist_lookup_uint64(props,
3359 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
3362 spa = spa_add(pool, config, altroot);
3363 spa->spa_import_flags = flags;
3366 * Verbatim import - Take a pool and insert it into the namespace
3367 * as if it had been loaded at boot.
3369 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
3371 spa_configfile_set(spa, props, B_FALSE);
3373 spa_config_sync(spa, B_FALSE, B_TRUE);
3375 mutex_exit(&spa_namespace_lock);
3376 spa_history_log_version(spa, LOG_POOL_IMPORT);
3381 spa_activate(spa, mode);
3384 * Don't start async tasks until we know everything is healthy.
3386 spa_async_suspend(spa);
3388 zpool_get_rewind_policy(config, &policy);
3389 if (policy.zrp_request & ZPOOL_DO_REWIND)
3390 state = SPA_LOAD_RECOVER;
3393 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
3394 * because the user-supplied config is actually the one to trust when
3397 if (state != SPA_LOAD_RECOVER)
3398 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3400 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
3401 policy.zrp_request);
3404 * Propagate anything learned while loading the pool and pass it
3405 * back to caller (i.e. rewind info, missing devices, etc).
3407 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
3408 spa->spa_load_info) == 0);
3410 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3412 * Toss any existing sparelist, as it doesn't have any validity
3413 * anymore, and conflicts with spa_has_spare().
3415 if (spa->spa_spares.sav_config) {
3416 nvlist_free(spa->spa_spares.sav_config);
3417 spa->spa_spares.sav_config = NULL;
3418 spa_load_spares(spa);
3420 if (spa->spa_l2cache.sav_config) {
3421 nvlist_free(spa->spa_l2cache.sav_config);
3422 spa->spa_l2cache.sav_config = NULL;
3423 spa_load_l2cache(spa);
3426 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3429 error = spa_validate_aux(spa, nvroot, -1ULL,
3432 error = spa_validate_aux(spa, nvroot, -1ULL,
3433 VDEV_ALLOC_L2CACHE);
3434 spa_config_exit(spa, SCL_ALL, FTAG);
3437 spa_configfile_set(spa, props, B_FALSE);
3439 if (error != 0 || (props && spa_writeable(spa) &&
3440 (error = spa_prop_set(spa, props)))) {
3442 spa_deactivate(spa);
3444 mutex_exit(&spa_namespace_lock);
3448 spa_async_resume(spa);
3451 * Override any spares and level 2 cache devices as specified by
3452 * the user, as these may have correct device names/devids, etc.
3454 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3455 &spares, &nspares) == 0) {
3456 if (spa->spa_spares.sav_config)
3457 VERIFY(nvlist_remove(spa->spa_spares.sav_config,
3458 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
3460 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
3461 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3462 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3463 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3464 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3465 spa_load_spares(spa);
3466 spa_config_exit(spa, SCL_ALL, FTAG);
3467 spa->spa_spares.sav_sync = B_TRUE;
3469 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3470 &l2cache, &nl2cache) == 0) {
3471 if (spa->spa_l2cache.sav_config)
3472 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
3473 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
3475 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3476 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3477 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3478 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3479 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3480 spa_load_l2cache(spa);
3481 spa_config_exit(spa, SCL_ALL, FTAG);
3482 spa->spa_l2cache.sav_sync = B_TRUE;
3486 * Check for any removed devices.
3488 if (spa->spa_autoreplace) {
3489 spa_aux_check_removed(&spa->spa_spares);
3490 spa_aux_check_removed(&spa->spa_l2cache);
3493 if (spa_writeable(spa)) {
3495 * Update the config cache to include the newly-imported pool.
3497 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3501 * It's possible that the pool was expanded while it was exported.
3502 * We kick off an async task to handle this for us.
3504 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
3506 mutex_exit(&spa_namespace_lock);
3507 spa_history_log_version(spa, LOG_POOL_IMPORT);
3513 spa_tryimport(nvlist_t *tryconfig)
3515 nvlist_t *config = NULL;
3521 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
3524 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
3528 * Create and initialize the spa structure.
3530 mutex_enter(&spa_namespace_lock);
3531 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
3532 spa_activate(spa, FREAD);
3535 * Pass off the heavy lifting to spa_load().
3536 * Pass TRUE for mosconfig because the user-supplied config
3537 * is actually the one to trust when doing an import.
3539 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
3542 * If 'tryconfig' was at least parsable, return the current config.
3544 if (spa->spa_root_vdev != NULL) {
3545 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
3546 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
3548 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
3550 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
3551 spa->spa_uberblock.ub_timestamp) == 0);
3554 * If the bootfs property exists on this pool then we
3555 * copy it out so that external consumers can tell which
3556 * pools are bootable.
3558 if ((!error || error == EEXIST) && spa->spa_bootfs) {
3559 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3562 * We have to play games with the name since the
3563 * pool was opened as TRYIMPORT_NAME.
3565 if (dsl_dsobj_to_dsname(spa_name(spa),
3566 spa->spa_bootfs, tmpname) == 0) {
3568 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3570 cp = strchr(tmpname, '/');
3572 (void) strlcpy(dsname, tmpname,
3575 (void) snprintf(dsname, MAXPATHLEN,
3576 "%s/%s", poolname, ++cp);
3578 VERIFY(nvlist_add_string(config,
3579 ZPOOL_CONFIG_BOOTFS, dsname) == 0);
3580 kmem_free(dsname, MAXPATHLEN);
3582 kmem_free(tmpname, MAXPATHLEN);
3586 * Add the list of hot spares and level 2 cache devices.
3588 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3589 spa_add_spares(spa, config);
3590 spa_add_l2cache(spa, config);
3591 spa_config_exit(spa, SCL_CONFIG, FTAG);
3595 spa_deactivate(spa);
3597 mutex_exit(&spa_namespace_lock);
3603 * Pool export/destroy
3605 * The act of destroying or exporting a pool is very simple. We make sure there
3606 * is no more pending I/O and any references to the pool are gone. Then, we
3607 * update the pool state and sync all the labels to disk, removing the
3608 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
3609 * we don't sync the labels or remove the configuration cache.
3612 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
3613 boolean_t force, boolean_t hardforce)
3620 if (!(spa_mode_global & FWRITE))
3623 mutex_enter(&spa_namespace_lock);
3624 if ((spa = spa_lookup(pool)) == NULL) {
3625 mutex_exit(&spa_namespace_lock);
3630 * Put a hold on the pool, drop the namespace lock, stop async tasks,
3631 * reacquire the namespace lock, and see if we can export.
3633 spa_open_ref(spa, FTAG);
3634 mutex_exit(&spa_namespace_lock);
3635 spa_async_suspend(spa);
3636 mutex_enter(&spa_namespace_lock);
3637 spa_close(spa, FTAG);
3640 * The pool will be in core if it's openable,
3641 * in which case we can modify its state.
3643 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
3645 * Objsets may be open only because they're dirty, so we
3646 * have to force it to sync before checking spa_refcnt.
3648 txg_wait_synced(spa->spa_dsl_pool, 0);
3651 * A pool cannot be exported or destroyed if there are active
3652 * references. If we are resetting a pool, allow references by
3653 * fault injection handlers.
3655 if (!spa_refcount_zero(spa) ||
3656 (spa->spa_inject_ref != 0 &&
3657 new_state != POOL_STATE_UNINITIALIZED)) {
3658 spa_async_resume(spa);
3659 mutex_exit(&spa_namespace_lock);
3664 * A pool cannot be exported if it has an active shared spare.
3665 * This is to prevent other pools stealing the active spare
3666 * from an exported pool. At user's own will, such pool can
3667 * be forcedly exported.
3669 if (!force && new_state == POOL_STATE_EXPORTED &&
3670 spa_has_active_shared_spare(spa)) {
3671 spa_async_resume(spa);
3672 mutex_exit(&spa_namespace_lock);
3677 * We want this to be reflected on every label,
3678 * so mark them all dirty. spa_unload() will do the
3679 * final sync that pushes these changes out.
3681 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
3682 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3683 spa->spa_state = new_state;
3684 spa->spa_final_txg = spa_last_synced_txg(spa) +
3686 vdev_config_dirty(spa->spa_root_vdev);
3687 spa_config_exit(spa, SCL_ALL, FTAG);
3691 spa_event_notify(spa, NULL, FM_EREPORT_ZFS_POOL_DESTROY);
3693 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
3695 spa_deactivate(spa);
3698 if (oldconfig && spa->spa_config)
3699 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
3701 if (new_state != POOL_STATE_UNINITIALIZED) {
3703 spa_config_sync(spa, B_TRUE, B_TRUE);
3706 mutex_exit(&spa_namespace_lock);
3712 * Destroy a storage pool.
3715 spa_destroy(char *pool)
3717 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
3722 * Export a storage pool.
3725 spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
3726 boolean_t hardforce)
3728 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
3733 * Similar to spa_export(), this unloads the spa_t without actually removing it
3734 * from the namespace in any way.
3737 spa_reset(char *pool)
3739 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
3744 * ==========================================================================
3745 * Device manipulation
3746 * ==========================================================================
3750 * Add a device to a storage pool.
3753 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
3757 vdev_t *rvd = spa->spa_root_vdev;
3759 nvlist_t **spares, **l2cache;
3760 uint_t nspares, nl2cache;
3763 ASSERT(spa_writeable(spa));
3765 txg = spa_vdev_enter(spa);
3767 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
3768 VDEV_ALLOC_ADD)) != 0)
3769 return (spa_vdev_exit(spa, NULL, txg, error));
3771 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */
3773 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
3777 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
3781 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
3782 return (spa_vdev_exit(spa, vd, txg, EINVAL));
3784 if (vd->vdev_children != 0 &&
3785 (error = vdev_create(vd, txg, B_FALSE)) != 0)
3786 return (spa_vdev_exit(spa, vd, txg, error));
3789 * We must validate the spares and l2cache devices after checking the
3790 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
3792 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
3793 return (spa_vdev_exit(spa, vd, txg, error));
3796 * Transfer each new top-level vdev from vd to rvd.
3798 for (c = 0; c < vd->vdev_children; c++) {
3801 * Set the vdev id to the first hole, if one exists.
3803 for (id = 0; id < rvd->vdev_children; id++) {
3804 if (rvd->vdev_child[id]->vdev_ishole) {
3805 vdev_free(rvd->vdev_child[id]);
3809 tvd = vd->vdev_child[c];
3810 vdev_remove_child(vd, tvd);
3812 vdev_add_child(rvd, tvd);
3813 vdev_config_dirty(tvd);
3817 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
3818 ZPOOL_CONFIG_SPARES);
3819 spa_load_spares(spa);
3820 spa->spa_spares.sav_sync = B_TRUE;
3823 if (nl2cache != 0) {
3824 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
3825 ZPOOL_CONFIG_L2CACHE);
3826 spa_load_l2cache(spa);
3827 spa->spa_l2cache.sav_sync = B_TRUE;
3831 * We have to be careful when adding new vdevs to an existing pool.
3832 * If other threads start allocating from these vdevs before we
3833 * sync the config cache, and we lose power, then upon reboot we may
3834 * fail to open the pool because there are DVAs that the config cache
3835 * can't translate. Therefore, we first add the vdevs without
3836 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
3837 * and then let spa_config_update() initialize the new metaslabs.
3839 * spa_load() checks for added-but-not-initialized vdevs, so that
3840 * if we lose power at any point in this sequence, the remaining
3841 * steps will be completed the next time we load the pool.
3843 (void) spa_vdev_exit(spa, vd, txg, 0);
3845 mutex_enter(&spa_namespace_lock);
3846 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3847 mutex_exit(&spa_namespace_lock);
3853 * Attach a device to a mirror. The arguments are the path to any device
3854 * in the mirror, and the nvroot for the new device. If the path specifies
3855 * a device that is not mirrored, we automatically insert the mirror vdev.
3857 * If 'replacing' is specified, the new device is intended to replace the
3858 * existing device; in this case the two devices are made into their own
3859 * mirror using the 'replacing' vdev, which is functionally identical to
3860 * the mirror vdev (it actually reuses all the same ops) but has a few
3861 * extra rules: you can't attach to it after it's been created, and upon
3862 * completion of resilvering, the first disk (the one being replaced)
3863 * is automatically detached.
3866 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
3868 uint64_t txg, dtl_max_txg;
3869 ASSERTV(vdev_t *rvd = spa->spa_root_vdev;)
3870 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
3872 char *oldvdpath, *newvdpath;
3876 ASSERT(spa_writeable(spa));
3878 txg = spa_vdev_enter(spa);
3880 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
3883 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
3885 if (!oldvd->vdev_ops->vdev_op_leaf)
3886 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3888 pvd = oldvd->vdev_parent;
3890 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
3891 VDEV_ALLOC_ATTACH)) != 0)
3892 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
3894 if (newrootvd->vdev_children != 1)
3895 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
3897 newvd = newrootvd->vdev_child[0];
3899 if (!newvd->vdev_ops->vdev_op_leaf)
3900 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
3902 if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
3903 return (spa_vdev_exit(spa, newrootvd, txg, error));
3906 * Spares can't replace logs
3908 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
3909 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3913 * For attach, the only allowable parent is a mirror or the root
3916 if (pvd->vdev_ops != &vdev_mirror_ops &&
3917 pvd->vdev_ops != &vdev_root_ops)
3918 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3920 pvops = &vdev_mirror_ops;
3923 * Active hot spares can only be replaced by inactive hot
3926 if (pvd->vdev_ops == &vdev_spare_ops &&
3927 oldvd->vdev_isspare &&
3928 !spa_has_spare(spa, newvd->vdev_guid))
3929 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3932 * If the source is a hot spare, and the parent isn't already a
3933 * spare, then we want to create a new hot spare. Otherwise, we
3934 * want to create a replacing vdev. The user is not allowed to
3935 * attach to a spared vdev child unless the 'isspare' state is
3936 * the same (spare replaces spare, non-spare replaces
3939 if (pvd->vdev_ops == &vdev_replacing_ops &&
3940 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
3941 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3942 } else if (pvd->vdev_ops == &vdev_spare_ops &&
3943 newvd->vdev_isspare != oldvd->vdev_isspare) {
3944 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3947 if (newvd->vdev_isspare)
3948 pvops = &vdev_spare_ops;
3950 pvops = &vdev_replacing_ops;
3954 * Make sure the new device is big enough.
3956 if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
3957 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
3960 * The new device cannot have a higher alignment requirement
3961 * than the top-level vdev.
3963 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
3964 return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
3967 * If this is an in-place replacement, update oldvd's path and devid
3968 * to make it distinguishable from newvd, and unopenable from now on.
3970 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
3971 spa_strfree(oldvd->vdev_path);
3972 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
3974 (void) sprintf(oldvd->vdev_path, "%s/%s",
3975 newvd->vdev_path, "old");
3976 if (oldvd->vdev_devid != NULL) {
3977 spa_strfree(oldvd->vdev_devid);
3978 oldvd->vdev_devid = NULL;
3982 /* mark the device being resilvered */
3983 newvd->vdev_resilvering = B_TRUE;
3986 * If the parent is not a mirror, or if we're replacing, insert the new
3987 * mirror/replacing/spare vdev above oldvd.
3989 if (pvd->vdev_ops != pvops)
3990 pvd = vdev_add_parent(oldvd, pvops);
3992 ASSERT(pvd->vdev_top->vdev_parent == rvd);
3993 ASSERT(pvd->vdev_ops == pvops);
3994 ASSERT(oldvd->vdev_parent == pvd);
3997 * Extract the new device from its root and add it to pvd.
3999 vdev_remove_child(newrootvd, newvd);
4000 newvd->vdev_id = pvd->vdev_children;
4001 newvd->vdev_crtxg = oldvd->vdev_crtxg;
4002 vdev_add_child(pvd, newvd);
4004 tvd = newvd->vdev_top;
4005 ASSERT(pvd->vdev_top == tvd);
4006 ASSERT(tvd->vdev_parent == rvd);
4008 vdev_config_dirty(tvd);
4011 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4012 * for any dmu_sync-ed blocks. It will propagate upward when
4013 * spa_vdev_exit() calls vdev_dtl_reassess().
4015 dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4017 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4018 dtl_max_txg - TXG_INITIAL);
4020 if (newvd->vdev_isspare) {
4021 spa_spare_activate(newvd);
4022 spa_event_notify(spa, newvd, FM_EREPORT_ZFS_DEVICE_SPARE);
4025 oldvdpath = spa_strdup(oldvd->vdev_path);
4026 newvdpath = spa_strdup(newvd->vdev_path);
4027 newvd_isspare = newvd->vdev_isspare;
4030 * Mark newvd's DTL dirty in this txg.
4032 vdev_dirty(tvd, VDD_DTL, newvd, txg);
4035 * Restart the resilver
4037 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4042 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4044 spa_history_log_internal(LOG_POOL_VDEV_ATTACH, spa, NULL,
4045 "%s vdev=%s %s vdev=%s",
4046 replacing && newvd_isspare ? "spare in" :
4047 replacing ? "replace" : "attach", newvdpath,
4048 replacing ? "for" : "to", oldvdpath);
4050 spa_strfree(oldvdpath);
4051 spa_strfree(newvdpath);
4053 if (spa->spa_bootfs)
4054 spa_event_notify(spa, newvd, FM_EREPORT_ZFS_BOOTFS_VDEV_ATTACH);
4060 * Detach a device from a mirror or replacing vdev.
4061 * If 'replace_done' is specified, only detach if the parent
4062 * is a replacing vdev.
4065 spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4069 ASSERTV(vdev_t *rvd = spa->spa_root_vdev;)
4070 vdev_t *vd, *pvd, *cvd, *tvd;
4071 boolean_t unspare = B_FALSE;
4072 uint64_t unspare_guid = 0;
4076 ASSERT(spa_writeable(spa));
4078 txg = spa_vdev_enter(spa);
4080 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4083 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4085 if (!vd->vdev_ops->vdev_op_leaf)
4086 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4088 pvd = vd->vdev_parent;
4091 * If the parent/child relationship is not as expected, don't do it.
4092 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4093 * vdev that's replacing B with C. The user's intent in replacing
4094 * is to go from M(A,B) to M(A,C). If the user decides to cancel
4095 * the replace by detaching C, the expected behavior is to end up
4096 * M(A,B). But suppose that right after deciding to detach C,
4097 * the replacement of B completes. We would have M(A,C), and then
4098 * ask to detach C, which would leave us with just A -- not what
4099 * the user wanted. To prevent this, we make sure that the
4100 * parent/child relationship hasn't changed -- in this example,
4101 * that C's parent is still the replacing vdev R.
4103 if (pvd->vdev_guid != pguid && pguid != 0)
4104 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4107 * Only 'replacing' or 'spare' vdevs can be replaced.
4109 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4110 pvd->vdev_ops != &vdev_spare_ops)
4111 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4113 ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4114 spa_version(spa) >= SPA_VERSION_SPARES);
4117 * Only mirror, replacing, and spare vdevs support detach.
4119 if (pvd->vdev_ops != &vdev_replacing_ops &&
4120 pvd->vdev_ops != &vdev_mirror_ops &&
4121 pvd->vdev_ops != &vdev_spare_ops)
4122 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4125 * If this device has the only valid copy of some data,
4126 * we cannot safely detach it.
4128 if (vdev_dtl_required(vd))
4129 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4131 ASSERT(pvd->vdev_children >= 2);
4134 * If we are detaching the second disk from a replacing vdev, then
4135 * check to see if we changed the original vdev's path to have "/old"
4136 * at the end in spa_vdev_attach(). If so, undo that change now.
4138 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4139 vd->vdev_path != NULL) {
4140 size_t len = strlen(vd->vdev_path);
4142 for (c = 0; c < pvd->vdev_children; c++) {
4143 cvd = pvd->vdev_child[c];
4145 if (cvd == vd || cvd->vdev_path == NULL)
4148 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4149 strcmp(cvd->vdev_path + len, "/old") == 0) {
4150 spa_strfree(cvd->vdev_path);
4151 cvd->vdev_path = spa_strdup(vd->vdev_path);
4158 * If we are detaching the original disk from a spare, then it implies
4159 * that the spare should become a real disk, and be removed from the
4160 * active spare list for the pool.
4162 if (pvd->vdev_ops == &vdev_spare_ops &&
4164 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4168 * Erase the disk labels so the disk can be used for other things.
4169 * This must be done after all other error cases are handled,
4170 * but before we disembowel vd (so we can still do I/O to it).
4171 * But if we can't do it, don't treat the error as fatal --
4172 * it may be that the unwritability of the disk is the reason
4173 * it's being detached!
4175 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4178 * Remove vd from its parent and compact the parent's children.
4180 vdev_remove_child(pvd, vd);
4181 vdev_compact_children(pvd);
4184 * Remember one of the remaining children so we can get tvd below.
4186 cvd = pvd->vdev_child[pvd->vdev_children - 1];
4189 * If we need to remove the remaining child from the list of hot spares,
4190 * do it now, marking the vdev as no longer a spare in the process.
4191 * We must do this before vdev_remove_parent(), because that can
4192 * change the GUID if it creates a new toplevel GUID. For a similar
4193 * reason, we must remove the spare now, in the same txg as the detach;
4194 * otherwise someone could attach a new sibling, change the GUID, and
4195 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4198 ASSERT(cvd->vdev_isspare);
4199 spa_spare_remove(cvd);
4200 unspare_guid = cvd->vdev_guid;
4201 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4202 cvd->vdev_unspare = B_TRUE;
4206 * If the parent mirror/replacing vdev only has one child,
4207 * the parent is no longer needed. Remove it from the tree.
4209 if (pvd->vdev_children == 1) {
4210 if (pvd->vdev_ops == &vdev_spare_ops)
4211 cvd->vdev_unspare = B_FALSE;
4212 vdev_remove_parent(cvd);
4213 cvd->vdev_resilvering = B_FALSE;
4218 * We don't set tvd until now because the parent we just removed
4219 * may have been the previous top-level vdev.
4221 tvd = cvd->vdev_top;
4222 ASSERT(tvd->vdev_parent == rvd);
4225 * Reevaluate the parent vdev state.
4227 vdev_propagate_state(cvd);
4230 * If the 'autoexpand' property is set on the pool then automatically
4231 * try to expand the size of the pool. For example if the device we
4232 * just detached was smaller than the others, it may be possible to
4233 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4234 * first so that we can obtain the updated sizes of the leaf vdevs.
4236 if (spa->spa_autoexpand) {
4238 vdev_expand(tvd, txg);
4241 vdev_config_dirty(tvd);
4244 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
4245 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4246 * But first make sure we're not on any *other* txg's DTL list, to
4247 * prevent vd from being accessed after it's freed.
4249 vdpath = spa_strdup(vd->vdev_path);
4250 for (t = 0; t < TXG_SIZE; t++)
4251 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4252 vd->vdev_detached = B_TRUE;
4253 vdev_dirty(tvd, VDD_DTL, vd, txg);
4255 spa_event_notify(spa, vd, FM_EREPORT_ZFS_DEVICE_REMOVE);
4257 /* hang on to the spa before we release the lock */
4258 spa_open_ref(spa, FTAG);
4260 error = spa_vdev_exit(spa, vd, txg, 0);
4262 spa_history_log_internal(LOG_POOL_VDEV_DETACH, spa, NULL,
4264 spa_strfree(vdpath);
4267 * If this was the removal of the original device in a hot spare vdev,
4268 * then we want to go through and remove the device from the hot spare
4269 * list of every other pool.
4272 spa_t *altspa = NULL;
4274 mutex_enter(&spa_namespace_lock);
4275 while ((altspa = spa_next(altspa)) != NULL) {
4276 if (altspa->spa_state != POOL_STATE_ACTIVE ||
4280 spa_open_ref(altspa, FTAG);
4281 mutex_exit(&spa_namespace_lock);
4282 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
4283 mutex_enter(&spa_namespace_lock);
4284 spa_close(altspa, FTAG);
4286 mutex_exit(&spa_namespace_lock);
4288 /* search the rest of the vdevs for spares to remove */
4289 spa_vdev_resilver_done(spa);
4292 /* all done with the spa; OK to release */
4293 mutex_enter(&spa_namespace_lock);
4294 spa_close(spa, FTAG);
4295 mutex_exit(&spa_namespace_lock);
4301 * Split a set of devices from their mirrors, and create a new pool from them.
4304 spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
4305 nvlist_t *props, boolean_t exp)
4308 uint64_t txg, *glist;
4310 uint_t c, children, lastlog;
4311 nvlist_t **child, *nvl, *tmp;
4313 char *altroot = NULL;
4314 vdev_t *rvd, **vml = NULL; /* vdev modify list */
4315 boolean_t activate_slog;
4317 ASSERT(spa_writeable(spa));
4319 txg = spa_vdev_enter(spa);
4321 /* clear the log and flush everything up to now */
4322 activate_slog = spa_passivate_log(spa);
4323 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4324 error = spa_offline_log(spa);
4325 txg = spa_vdev_config_enter(spa);
4328 spa_activate_log(spa);
4331 return (spa_vdev_exit(spa, NULL, txg, error));
4333 /* check new spa name before going any further */
4334 if (spa_lookup(newname) != NULL)
4335 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
4338 * scan through all the children to ensure they're all mirrors
4340 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
4341 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
4343 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4345 /* first, check to ensure we've got the right child count */
4346 rvd = spa->spa_root_vdev;
4348 for (c = 0; c < rvd->vdev_children; c++) {
4349 vdev_t *vd = rvd->vdev_child[c];
4351 /* don't count the holes & logs as children */
4352 if (vd->vdev_islog || vd->vdev_ishole) {
4360 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
4361 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4363 /* next, ensure no spare or cache devices are part of the split */
4364 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
4365 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
4366 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4368 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
4369 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
4371 /* then, loop over each vdev and validate it */
4372 for (c = 0; c < children; c++) {
4373 uint64_t is_hole = 0;
4375 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
4379 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
4380 spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
4388 /* which disk is going to be split? */
4389 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
4395 /* look it up in the spa */
4396 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
4397 if (vml[c] == NULL) {
4402 /* make sure there's nothing stopping the split */
4403 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
4404 vml[c]->vdev_islog ||
4405 vml[c]->vdev_ishole ||
4406 vml[c]->vdev_isspare ||
4407 vml[c]->vdev_isl2cache ||
4408 !vdev_writeable(vml[c]) ||
4409 vml[c]->vdev_children != 0 ||
4410 vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
4411 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
4416 if (vdev_dtl_required(vml[c])) {
4421 /* we need certain info from the top level */
4422 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
4423 vml[c]->vdev_top->vdev_ms_array) == 0);
4424 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
4425 vml[c]->vdev_top->vdev_ms_shift) == 0);
4426 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
4427 vml[c]->vdev_top->vdev_asize) == 0);
4428 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
4429 vml[c]->vdev_top->vdev_ashift) == 0);
4433 kmem_free(vml, children * sizeof (vdev_t *));
4434 kmem_free(glist, children * sizeof (uint64_t));
4435 return (spa_vdev_exit(spa, NULL, txg, error));
4438 /* stop writers from using the disks */
4439 for (c = 0; c < children; c++) {
4441 vml[c]->vdev_offline = B_TRUE;
4443 vdev_reopen(spa->spa_root_vdev);
4446 * Temporarily record the splitting vdevs in the spa config. This
4447 * will disappear once the config is regenerated.
4449 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4450 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
4451 glist, children) == 0);
4452 kmem_free(glist, children * sizeof (uint64_t));
4454 mutex_enter(&spa->spa_props_lock);
4455 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
4457 mutex_exit(&spa->spa_props_lock);
4458 spa->spa_config_splitting = nvl;
4459 vdev_config_dirty(spa->spa_root_vdev);
4461 /* configure and create the new pool */
4462 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
4463 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4464 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
4465 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
4466 spa_version(spa)) == 0);
4467 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
4468 spa->spa_config_txg) == 0);
4469 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4470 spa_generate_guid(NULL)) == 0);
4471 (void) nvlist_lookup_string(props,
4472 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4474 /* add the new pool to the namespace */
4475 newspa = spa_add(newname, config, altroot);
4476 newspa->spa_config_txg = spa->spa_config_txg;
4477 spa_set_log_state(newspa, SPA_LOG_CLEAR);
4479 /* release the spa config lock, retaining the namespace lock */
4480 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4482 if (zio_injection_enabled)
4483 zio_handle_panic_injection(spa, FTAG, 1);
4485 spa_activate(newspa, spa_mode_global);
4486 spa_async_suspend(newspa);
4488 /* create the new pool from the disks of the original pool */
4489 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
4493 /* if that worked, generate a real config for the new pool */
4494 if (newspa->spa_root_vdev != NULL) {
4495 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
4496 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4497 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
4498 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
4499 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
4504 if (props != NULL) {
4505 spa_configfile_set(newspa, props, B_FALSE);
4506 error = spa_prop_set(newspa, props);
4511 /* flush everything */
4512 txg = spa_vdev_config_enter(newspa);
4513 vdev_config_dirty(newspa->spa_root_vdev);
4514 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
4516 if (zio_injection_enabled)
4517 zio_handle_panic_injection(spa, FTAG, 2);
4519 spa_async_resume(newspa);
4521 /* finally, update the original pool's config */
4522 txg = spa_vdev_config_enter(spa);
4523 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
4524 error = dmu_tx_assign(tx, TXG_WAIT);
4527 for (c = 0; c < children; c++) {
4528 if (vml[c] != NULL) {
4531 spa_history_log_internal(LOG_POOL_VDEV_DETACH,
4537 vdev_config_dirty(spa->spa_root_vdev);
4538 spa->spa_config_splitting = NULL;
4542 (void) spa_vdev_exit(spa, NULL, txg, 0);
4544 if (zio_injection_enabled)
4545 zio_handle_panic_injection(spa, FTAG, 3);
4547 /* split is complete; log a history record */
4548 spa_history_log_internal(LOG_POOL_SPLIT, newspa, NULL,
4549 "split new pool %s from pool %s", newname, spa_name(spa));
4551 kmem_free(vml, children * sizeof (vdev_t *));
4553 /* if we're not going to mount the filesystems in userland, export */
4555 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
4562 spa_deactivate(newspa);
4565 txg = spa_vdev_config_enter(spa);
4567 /* re-online all offlined disks */
4568 for (c = 0; c < children; c++) {
4570 vml[c]->vdev_offline = B_FALSE;
4572 vdev_reopen(spa->spa_root_vdev);
4574 nvlist_free(spa->spa_config_splitting);
4575 spa->spa_config_splitting = NULL;
4576 (void) spa_vdev_exit(spa, NULL, txg, error);
4578 kmem_free(vml, children * sizeof (vdev_t *));
4583 spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
4587 for (i = 0; i < count; i++) {
4590 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
4593 if (guid == target_guid)
4601 spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
4602 nvlist_t *dev_to_remove)
4604 nvlist_t **newdev = NULL;
4608 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
4610 for (i = 0, j = 0; i < count; i++) {
4611 if (dev[i] == dev_to_remove)
4613 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
4616 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
4617 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
4619 for (i = 0; i < count - 1; i++)
4620 nvlist_free(newdev[i]);
4623 kmem_free(newdev, (count - 1) * sizeof (void *));
4627 * Evacuate the device.
4630 spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
4635 ASSERT(MUTEX_HELD(&spa_namespace_lock));
4636 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
4637 ASSERT(vd == vd->vdev_top);
4640 * Evacuate the device. We don't hold the config lock as writer
4641 * since we need to do I/O but we do keep the
4642 * spa_namespace_lock held. Once this completes the device
4643 * should no longer have any blocks allocated on it.
4645 if (vd->vdev_islog) {
4646 if (vd->vdev_stat.vs_alloc != 0)
4647 error = spa_offline_log(spa);
4656 * The evacuation succeeded. Remove any remaining MOS metadata
4657 * associated with this vdev, and wait for these changes to sync.
4659 ASSERT3U(vd->vdev_stat.vs_alloc, ==, 0);
4660 txg = spa_vdev_config_enter(spa);
4661 vd->vdev_removing = B_TRUE;
4662 vdev_dirty(vd, 0, NULL, txg);
4663 vdev_config_dirty(vd);
4664 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4670 * Complete the removal by cleaning up the namespace.
4673 spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
4675 vdev_t *rvd = spa->spa_root_vdev;
4676 uint64_t id = vd->vdev_id;
4677 boolean_t last_vdev = (id == (rvd->vdev_children - 1));
4679 ASSERT(MUTEX_HELD(&spa_namespace_lock));
4680 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
4681 ASSERT(vd == vd->vdev_top);
4684 * Only remove any devices which are empty.
4686 if (vd->vdev_stat.vs_alloc != 0)
4689 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4691 if (list_link_active(&vd->vdev_state_dirty_node))
4692 vdev_state_clean(vd);
4693 if (list_link_active(&vd->vdev_config_dirty_node))
4694 vdev_config_clean(vd);
4699 vdev_compact_children(rvd);
4701 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
4702 vdev_add_child(rvd, vd);
4704 vdev_config_dirty(rvd);
4707 * Reassess the health of our root vdev.
4713 * Remove a device from the pool -
4715 * Removing a device from the vdev namespace requires several steps
4716 * and can take a significant amount of time. As a result we use
4717 * the spa_vdev_config_[enter/exit] functions which allow us to
4718 * grab and release the spa_config_lock while still holding the namespace
4719 * lock. During each step the configuration is synced out.
4723 * Remove a device from the pool. Currently, this supports removing only hot
4724 * spares, slogs, and level 2 ARC devices.
4727 spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
4730 metaslab_group_t *mg;
4731 nvlist_t **spares, **l2cache, *nv;
4733 uint_t nspares, nl2cache;
4735 boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
4737 ASSERT(spa_writeable(spa));
4740 txg = spa_vdev_enter(spa);
4742 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4744 if (spa->spa_spares.sav_vdevs != NULL &&
4745 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
4746 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
4747 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
4749 * Only remove the hot spare if it's not currently in use
4752 if (vd == NULL || unspare) {
4753 spa_vdev_remove_aux(spa->spa_spares.sav_config,
4754 ZPOOL_CONFIG_SPARES, spares, nspares, nv);
4755 spa_load_spares(spa);
4756 spa->spa_spares.sav_sync = B_TRUE;
4760 } else if (spa->spa_l2cache.sav_vdevs != NULL &&
4761 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
4762 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
4763 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
4765 * Cache devices can always be removed.
4767 spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
4768 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
4769 spa_load_l2cache(spa);
4770 spa->spa_l2cache.sav_sync = B_TRUE;
4771 } else if (vd != NULL && vd->vdev_islog) {
4773 ASSERT(vd == vd->vdev_top);
4776 * XXX - Once we have bp-rewrite this should
4777 * become the common case.
4783 * Stop allocating from this vdev.
4785 metaslab_group_passivate(mg);
4788 * Wait for the youngest allocations and frees to sync,
4789 * and then wait for the deferral of those frees to finish.
4791 spa_vdev_config_exit(spa, NULL,
4792 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
4795 * Attempt to evacuate the vdev.
4797 error = spa_vdev_remove_evacuate(spa, vd);
4799 txg = spa_vdev_config_enter(spa);
4802 * If we couldn't evacuate the vdev, unwind.
4805 metaslab_group_activate(mg);
4806 return (spa_vdev_exit(spa, NULL, txg, error));
4810 * Clean up the vdev namespace.
4812 spa_vdev_remove_from_namespace(spa, vd);
4814 } else if (vd != NULL) {
4816 * Normal vdevs cannot be removed (yet).
4821 * There is no vdev of any kind with the specified guid.
4827 return (spa_vdev_exit(spa, NULL, txg, error));
4833 * Find any device that's done replacing, or a vdev marked 'unspare' that's
4834 * current spared, so we can detach it.
4837 spa_vdev_resilver_done_hunt(vdev_t *vd)
4839 vdev_t *newvd, *oldvd;
4842 for (c = 0; c < vd->vdev_children; c++) {
4843 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
4849 * Check for a completed replacement. We always consider the first
4850 * vdev in the list to be the oldest vdev, and the last one to be
4851 * the newest (see spa_vdev_attach() for how that works). In
4852 * the case where the newest vdev is faulted, we will not automatically
4853 * remove it after a resilver completes. This is OK as it will require
4854 * user intervention to determine which disk the admin wishes to keep.
4856 if (vd->vdev_ops == &vdev_replacing_ops) {
4857 ASSERT(vd->vdev_children > 1);
4859 newvd = vd->vdev_child[vd->vdev_children - 1];
4860 oldvd = vd->vdev_child[0];
4862 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
4863 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
4864 !vdev_dtl_required(oldvd))
4869 * Check for a completed resilver with the 'unspare' flag set.
4871 if (vd->vdev_ops == &vdev_spare_ops) {
4872 vdev_t *first = vd->vdev_child[0];
4873 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
4875 if (last->vdev_unspare) {
4878 } else if (first->vdev_unspare) {
4885 if (oldvd != NULL &&
4886 vdev_dtl_empty(newvd, DTL_MISSING) &&
4887 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
4888 !vdev_dtl_required(oldvd))
4892 * If there are more than two spares attached to a disk,
4893 * and those spares are not required, then we want to
4894 * attempt to free them up now so that they can be used
4895 * by other pools. Once we're back down to a single
4896 * disk+spare, we stop removing them.
4898 if (vd->vdev_children > 2) {
4899 newvd = vd->vdev_child[1];
4901 if (newvd->vdev_isspare && last->vdev_isspare &&
4902 vdev_dtl_empty(last, DTL_MISSING) &&
4903 vdev_dtl_empty(last, DTL_OUTAGE) &&
4904 !vdev_dtl_required(newvd))
4913 spa_vdev_resilver_done(spa_t *spa)
4915 vdev_t *vd, *pvd, *ppvd;
4916 uint64_t guid, sguid, pguid, ppguid;
4918 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4920 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
4921 pvd = vd->vdev_parent;
4922 ppvd = pvd->vdev_parent;
4923 guid = vd->vdev_guid;
4924 pguid = pvd->vdev_guid;
4925 ppguid = ppvd->vdev_guid;
4928 * If we have just finished replacing a hot spared device, then
4929 * we need to detach the parent's first child (the original hot
4932 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
4933 ppvd->vdev_children == 2) {
4934 ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
4935 sguid = ppvd->vdev_child[1]->vdev_guid;
4937 spa_config_exit(spa, SCL_ALL, FTAG);
4938 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
4940 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
4942 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4945 spa_config_exit(spa, SCL_ALL, FTAG);
4949 * Update the stored path or FRU for this vdev.
4952 spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
4956 boolean_t sync = B_FALSE;
4958 ASSERT(spa_writeable(spa));
4960 spa_vdev_state_enter(spa, SCL_ALL);
4962 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
4963 return (spa_vdev_state_exit(spa, NULL, ENOENT));
4965 if (!vd->vdev_ops->vdev_op_leaf)
4966 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
4969 if (strcmp(value, vd->vdev_path) != 0) {
4970 spa_strfree(vd->vdev_path);
4971 vd->vdev_path = spa_strdup(value);
4975 if (vd->vdev_fru == NULL) {
4976 vd->vdev_fru = spa_strdup(value);
4978 } else if (strcmp(value, vd->vdev_fru) != 0) {
4979 spa_strfree(vd->vdev_fru);
4980 vd->vdev_fru = spa_strdup(value);
4985 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
4989 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
4991 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
4995 spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
4997 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5001 * ==========================================================================
5003 * ==========================================================================
5007 spa_scan_stop(spa_t *spa)
5009 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5010 if (dsl_scan_resilvering(spa->spa_dsl_pool))
5012 return (dsl_scan_cancel(spa->spa_dsl_pool));
5016 spa_scan(spa_t *spa, pool_scan_func_t func)
5018 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5020 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5024 * If a resilver was requested, but there is no DTL on a
5025 * writeable leaf device, we have nothing to do.
5027 if (func == POOL_SCAN_RESILVER &&
5028 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5029 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5033 return (dsl_scan(spa->spa_dsl_pool, func));
5037 * ==========================================================================
5038 * SPA async task processing
5039 * ==========================================================================
5043 spa_async_remove(spa_t *spa, vdev_t *vd)
5047 if (vd->vdev_remove_wanted) {
5048 vd->vdev_remove_wanted = B_FALSE;
5049 vd->vdev_delayed_close = B_FALSE;
5050 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5053 * We want to clear the stats, but we don't want to do a full
5054 * vdev_clear() as that will cause us to throw away
5055 * degraded/faulted state as well as attempt to reopen the
5056 * device, all of which is a waste.
5058 vd->vdev_stat.vs_read_errors = 0;
5059 vd->vdev_stat.vs_write_errors = 0;
5060 vd->vdev_stat.vs_checksum_errors = 0;
5062 vdev_state_dirty(vd->vdev_top);
5065 for (c = 0; c < vd->vdev_children; c++)
5066 spa_async_remove(spa, vd->vdev_child[c]);
5070 spa_async_probe(spa_t *spa, vdev_t *vd)
5074 if (vd->vdev_probe_wanted) {
5075 vd->vdev_probe_wanted = B_FALSE;
5076 vdev_reopen(vd); /* vdev_open() does the actual probe */
5079 for (c = 0; c < vd->vdev_children; c++)
5080 spa_async_probe(spa, vd->vdev_child[c]);
5084 spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5088 if (!spa->spa_autoexpand)
5091 for (c = 0; c < vd->vdev_children; c++) {
5092 vdev_t *cvd = vd->vdev_child[c];
5093 spa_async_autoexpand(spa, cvd);
5096 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5099 spa_event_notify(vd->vdev_spa, vd, FM_EREPORT_ZFS_DEVICE_AUTOEXPAND);
5103 spa_async_thread(spa_t *spa)
5107 ASSERT(spa->spa_sync_on);
5109 mutex_enter(&spa->spa_async_lock);
5110 tasks = spa->spa_async_tasks;
5111 spa->spa_async_tasks = 0;
5112 mutex_exit(&spa->spa_async_lock);
5115 * See if the config needs to be updated.
5117 if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5118 uint64_t old_space, new_space;
5120 mutex_enter(&spa_namespace_lock);
5121 old_space = metaslab_class_get_space(spa_normal_class(spa));
5122 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5123 new_space = metaslab_class_get_space(spa_normal_class(spa));
5124 mutex_exit(&spa_namespace_lock);
5127 * If the pool grew as a result of the config update,
5128 * then log an internal history event.
5130 if (new_space != old_space) {
5131 spa_history_log_internal(LOG_POOL_VDEV_ONLINE,
5133 "pool '%s' size: %llu(+%llu)",
5134 spa_name(spa), new_space, new_space - old_space);
5139 * See if any devices need to be marked REMOVED.
5141 if (tasks & SPA_ASYNC_REMOVE) {
5142 spa_vdev_state_enter(spa, SCL_NONE);
5143 spa_async_remove(spa, spa->spa_root_vdev);
5144 for (i = 0; i < spa->spa_l2cache.sav_count; i++)
5145 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5146 for (i = 0; i < spa->spa_spares.sav_count; i++)
5147 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5148 (void) spa_vdev_state_exit(spa, NULL, 0);
5151 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5152 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5153 spa_async_autoexpand(spa, spa->spa_root_vdev);
5154 spa_config_exit(spa, SCL_CONFIG, FTAG);
5158 * See if any devices need to be probed.
5160 if (tasks & SPA_ASYNC_PROBE) {
5161 spa_vdev_state_enter(spa, SCL_NONE);
5162 spa_async_probe(spa, spa->spa_root_vdev);
5163 (void) spa_vdev_state_exit(spa, NULL, 0);
5167 * If any devices are done replacing, detach them.
5169 if (tasks & SPA_ASYNC_RESILVER_DONE)
5170 spa_vdev_resilver_done(spa);
5173 * Kick off a resilver.
5175 if (tasks & SPA_ASYNC_RESILVER)
5176 dsl_resilver_restart(spa->spa_dsl_pool, 0);
5179 * Let the world know that we're done.
5181 mutex_enter(&spa->spa_async_lock);
5182 spa->spa_async_thread = NULL;
5183 cv_broadcast(&spa->spa_async_cv);
5184 mutex_exit(&spa->spa_async_lock);
5189 spa_async_suspend(spa_t *spa)
5191 mutex_enter(&spa->spa_async_lock);
5192 spa->spa_async_suspended++;
5193 while (spa->spa_async_thread != NULL)
5194 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5195 mutex_exit(&spa->spa_async_lock);
5199 spa_async_resume(spa_t *spa)
5201 mutex_enter(&spa->spa_async_lock);
5202 ASSERT(spa->spa_async_suspended != 0);
5203 spa->spa_async_suspended--;
5204 mutex_exit(&spa->spa_async_lock);
5208 spa_async_dispatch(spa_t *spa)
5210 mutex_enter(&spa->spa_async_lock);
5211 if (spa->spa_async_tasks && !spa->spa_async_suspended &&
5212 spa->spa_async_thread == NULL &&
5213 rootdir != NULL && !vn_is_readonly(rootdir))
5214 spa->spa_async_thread = thread_create(NULL, 0,
5215 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
5216 mutex_exit(&spa->spa_async_lock);
5220 spa_async_request(spa_t *spa, int task)
5222 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
5223 mutex_enter(&spa->spa_async_lock);
5224 spa->spa_async_tasks |= task;
5225 mutex_exit(&spa->spa_async_lock);
5229 * ==========================================================================
5230 * SPA syncing routines
5231 * ==========================================================================
5235 bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5238 bpobj_enqueue(bpo, bp, tx);
5243 spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5247 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
5253 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
5255 char *packed = NULL;
5260 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
5263 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5264 * information. This avoids the dbuf_will_dirty() path and
5265 * saves us a pre-read to get data we don't actually care about.
5267 bufsize = P2ROUNDUP(nvsize, SPA_CONFIG_BLOCKSIZE);
5268 packed = vmem_alloc(bufsize, KM_SLEEP);
5270 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
5272 bzero(packed + nvsize, bufsize - nvsize);
5274 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
5276 vmem_free(packed, bufsize);
5278 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
5279 dmu_buf_will_dirty(db, tx);
5280 *(uint64_t *)db->db_data = nvsize;
5281 dmu_buf_rele(db, FTAG);
5285 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
5286 const char *config, const char *entry)
5296 * Update the MOS nvlist describing the list of available devices.
5297 * spa_validate_aux() will have already made sure this nvlist is
5298 * valid and the vdevs are labeled appropriately.
5300 if (sav->sav_object == 0) {
5301 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
5302 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
5303 sizeof (uint64_t), tx);
5304 VERIFY(zap_update(spa->spa_meta_objset,
5305 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
5306 &sav->sav_object, tx) == 0);
5309 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5310 if (sav->sav_count == 0) {
5311 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
5313 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
5314 for (i = 0; i < sav->sav_count; i++)
5315 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
5316 B_FALSE, VDEV_CONFIG_L2CACHE);
5317 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
5318 sav->sav_count) == 0);
5319 for (i = 0; i < sav->sav_count; i++)
5320 nvlist_free(list[i]);
5321 kmem_free(list, sav->sav_count * sizeof (void *));
5324 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
5325 nvlist_free(nvroot);
5327 sav->sav_sync = B_FALSE;
5331 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
5335 if (list_is_empty(&spa->spa_config_dirty_list))
5338 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5340 config = spa_config_generate(spa, spa->spa_root_vdev,
5341 dmu_tx_get_txg(tx), B_FALSE);
5343 spa_config_exit(spa, SCL_STATE, FTAG);
5345 if (spa->spa_config_syncing)
5346 nvlist_free(spa->spa_config_syncing);
5347 spa->spa_config_syncing = config;
5349 spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
5353 * Set zpool properties.
5356 spa_sync_props(void *arg1, void *arg2, dmu_tx_t *tx)
5359 objset_t *mos = spa->spa_meta_objset;
5360 nvlist_t *nvp = arg2;
5365 const char *propname;
5366 zprop_type_t proptype;
5368 mutex_enter(&spa->spa_props_lock);
5371 while ((elem = nvlist_next_nvpair(nvp, elem))) {
5372 switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
5373 case ZPOOL_PROP_VERSION:
5375 * Only set version for non-zpool-creation cases
5376 * (set/import). spa_create() needs special care
5377 * for version setting.
5379 if (tx->tx_txg != TXG_INITIAL) {
5380 VERIFY(nvpair_value_uint64(elem,
5382 ASSERT(intval <= SPA_VERSION);
5383 ASSERT(intval >= spa_version(spa));
5384 spa->spa_uberblock.ub_version = intval;
5385 vdev_config_dirty(spa->spa_root_vdev);
5389 case ZPOOL_PROP_ALTROOT:
5391 * 'altroot' is a non-persistent property. It should
5392 * have been set temporarily at creation or import time.
5394 ASSERT(spa->spa_root != NULL);
5397 case ZPOOL_PROP_READONLY:
5398 case ZPOOL_PROP_CACHEFILE:
5400 * 'readonly' and 'cachefile' are also non-persisitent
5406 * Set pool property values in the poolprops mos object.
5408 if (spa->spa_pool_props_object == 0) {
5409 VERIFY((spa->spa_pool_props_object =
5410 zap_create(mos, DMU_OT_POOL_PROPS,
5411 DMU_OT_NONE, 0, tx)) > 0);
5413 VERIFY(zap_update(mos,
5414 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
5415 8, 1, &spa->spa_pool_props_object, tx)
5419 /* normalize the property name */
5420 propname = zpool_prop_to_name(prop);
5421 proptype = zpool_prop_get_type(prop);
5423 if (nvpair_type(elem) == DATA_TYPE_STRING) {
5424 ASSERT(proptype == PROP_TYPE_STRING);
5425 VERIFY(nvpair_value_string(elem, &strval) == 0);
5426 VERIFY(zap_update(mos,
5427 spa->spa_pool_props_object, propname,
5428 1, strlen(strval) + 1, strval, tx) == 0);
5430 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
5431 VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5433 if (proptype == PROP_TYPE_INDEX) {
5435 VERIFY(zpool_prop_index_to_string(
5436 prop, intval, &unused) == 0);
5438 VERIFY(zap_update(mos,
5439 spa->spa_pool_props_object, propname,
5440 8, 1, &intval, tx) == 0);
5442 ASSERT(0); /* not allowed */
5446 case ZPOOL_PROP_DELEGATION:
5447 spa->spa_delegation = intval;
5449 case ZPOOL_PROP_BOOTFS:
5450 spa->spa_bootfs = intval;
5452 case ZPOOL_PROP_FAILUREMODE:
5453 spa->spa_failmode = intval;
5455 case ZPOOL_PROP_AUTOEXPAND:
5456 spa->spa_autoexpand = intval;
5457 if (tx->tx_txg != TXG_INITIAL)
5458 spa_async_request(spa,
5459 SPA_ASYNC_AUTOEXPAND);
5461 case ZPOOL_PROP_DEDUPDITTO:
5462 spa->spa_dedup_ditto = intval;
5469 /* log internal history if this is not a zpool create */
5470 if (spa_version(spa) >= SPA_VERSION_ZPOOL_HISTORY &&
5471 tx->tx_txg != TXG_INITIAL) {
5472 spa_history_log_internal(LOG_POOL_PROPSET,
5473 spa, tx, "%s %lld %s",
5474 nvpair_name(elem), intval, spa_name(spa));
5478 mutex_exit(&spa->spa_props_lock);
5482 * Perform one-time upgrade on-disk changes. spa_version() does not
5483 * reflect the new version this txg, so there must be no changes this
5484 * txg to anything that the upgrade code depends on after it executes.
5485 * Therefore this must be called after dsl_pool_sync() does the sync
5489 spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
5491 dsl_pool_t *dp = spa->spa_dsl_pool;
5493 ASSERT(spa->spa_sync_pass == 1);
5495 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
5496 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
5497 dsl_pool_create_origin(dp, tx);
5499 /* Keeping the origin open increases spa_minref */
5500 spa->spa_minref += 3;
5503 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
5504 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
5505 dsl_pool_upgrade_clones(dp, tx);
5508 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
5509 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
5510 dsl_pool_upgrade_dir_clones(dp, tx);
5512 /* Keeping the freedir open increases spa_minref */
5513 spa->spa_minref += 3;
5518 * Sync the specified transaction group. New blocks may be dirtied as
5519 * part of the process, so we iterate until it converges.
5522 spa_sync(spa_t *spa, uint64_t txg)
5524 dsl_pool_t *dp = spa->spa_dsl_pool;
5525 objset_t *mos = spa->spa_meta_objset;
5526 bpobj_t *defer_bpo = &spa->spa_deferred_bpobj;
5527 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
5528 vdev_t *rvd = spa->spa_root_vdev;
5534 VERIFY(spa_writeable(spa));
5537 * Lock out configuration changes.
5539 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5541 spa->spa_syncing_txg = txg;
5542 spa->spa_sync_pass = 0;
5545 * If there are any pending vdev state changes, convert them
5546 * into config changes that go out with this transaction group.
5548 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5549 while (list_head(&spa->spa_state_dirty_list) != NULL) {
5551 * We need the write lock here because, for aux vdevs,
5552 * calling vdev_config_dirty() modifies sav_config.
5553 * This is ugly and will become unnecessary when we
5554 * eliminate the aux vdev wart by integrating all vdevs
5555 * into the root vdev tree.
5557 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
5558 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
5559 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
5560 vdev_state_clean(vd);
5561 vdev_config_dirty(vd);
5563 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
5564 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
5566 spa_config_exit(spa, SCL_STATE, FTAG);
5568 tx = dmu_tx_create_assigned(dp, txg);
5571 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
5572 * set spa_deflate if we have no raid-z vdevs.
5574 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
5575 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
5578 for (i = 0; i < rvd->vdev_children; i++) {
5579 vd = rvd->vdev_child[i];
5580 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
5583 if (i == rvd->vdev_children) {
5584 spa->spa_deflate = TRUE;
5585 VERIFY(0 == zap_add(spa->spa_meta_objset,
5586 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
5587 sizeof (uint64_t), 1, &spa->spa_deflate, tx));
5592 * If anything has changed in this txg, or if someone is waiting
5593 * for this txg to sync (eg, spa_vdev_remove()), push the
5594 * deferred frees from the previous txg. If not, leave them
5595 * alone so that we don't generate work on an otherwise idle
5598 if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
5599 !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
5600 !txg_list_empty(&dp->dp_sync_tasks, txg) ||
5601 ((dsl_scan_active(dp->dp_scan) ||
5602 txg_sync_waiting(dp)) && !spa_shutting_down(spa))) {
5603 zio_t *zio = zio_root(spa, NULL, NULL, 0);
5604 VERIFY3U(bpobj_iterate(defer_bpo,
5605 spa_free_sync_cb, zio, tx), ==, 0);
5606 VERIFY3U(zio_wait(zio), ==, 0);
5610 * Iterate to convergence.
5613 int pass = ++spa->spa_sync_pass;
5615 spa_sync_config_object(spa, tx);
5616 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
5617 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
5618 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
5619 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
5620 spa_errlog_sync(spa, txg);
5621 dsl_pool_sync(dp, txg);
5623 if (pass <= SYNC_PASS_DEFERRED_FREE) {
5624 zio_t *zio = zio_root(spa, NULL, NULL, 0);
5625 bplist_iterate(free_bpl, spa_free_sync_cb,
5627 VERIFY(zio_wait(zio) == 0);
5629 bplist_iterate(free_bpl, bpobj_enqueue_cb,
5634 dsl_scan_sync(dp, tx);
5636 while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)))
5640 spa_sync_upgrades(spa, tx);
5642 } while (dmu_objset_is_dirty(mos, txg));
5645 * Rewrite the vdev configuration (which includes the uberblock)
5646 * to commit the transaction group.
5648 * If there are no dirty vdevs, we sync the uberblock to a few
5649 * random top-level vdevs that are known to be visible in the
5650 * config cache (see spa_vdev_add() for a complete description).
5651 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
5655 * We hold SCL_STATE to prevent vdev open/close/etc.
5656 * while we're attempting to write the vdev labels.
5658 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5660 if (list_is_empty(&spa->spa_config_dirty_list)) {
5661 vdev_t *svd[SPA_DVAS_PER_BP];
5663 int children = rvd->vdev_children;
5664 int c0 = spa_get_random(children);
5666 for (c = 0; c < children; c++) {
5667 vd = rvd->vdev_child[(c0 + c) % children];
5668 if (vd->vdev_ms_array == 0 || vd->vdev_islog)
5670 svd[svdcount++] = vd;
5671 if (svdcount == SPA_DVAS_PER_BP)
5674 error = vdev_config_sync(svd, svdcount, txg, B_FALSE);
5676 error = vdev_config_sync(svd, svdcount, txg,
5679 error = vdev_config_sync(rvd->vdev_child,
5680 rvd->vdev_children, txg, B_FALSE);
5682 error = vdev_config_sync(rvd->vdev_child,
5683 rvd->vdev_children, txg, B_TRUE);
5686 spa_config_exit(spa, SCL_STATE, FTAG);
5690 zio_suspend(spa, NULL);
5691 zio_resume_wait(spa);
5696 * Clear the dirty config list.
5698 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
5699 vdev_config_clean(vd);
5702 * Now that the new config has synced transactionally,
5703 * let it become visible to the config cache.
5705 if (spa->spa_config_syncing != NULL) {
5706 spa_config_set(spa, spa->spa_config_syncing);
5707 spa->spa_config_txg = txg;
5708 spa->spa_config_syncing = NULL;
5711 spa->spa_ubsync = spa->spa_uberblock;
5713 dsl_pool_sync_done(dp, txg);
5716 * Update usable space statistics.
5718 while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg))))
5719 vdev_sync_done(vd, txg);
5721 spa_update_dspace(spa);
5724 * It had better be the case that we didn't dirty anything
5725 * since vdev_config_sync().
5727 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
5728 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
5729 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
5731 spa->spa_sync_pass = 0;
5733 spa_config_exit(spa, SCL_CONFIG, FTAG);
5735 spa_handle_ignored_writes(spa);
5738 * If any async tasks have been requested, kick them off.
5740 spa_async_dispatch(spa);
5744 * Sync all pools. We don't want to hold the namespace lock across these
5745 * operations, so we take a reference on the spa_t and drop the lock during the
5749 spa_sync_allpools(void)
5752 mutex_enter(&spa_namespace_lock);
5753 while ((spa = spa_next(spa)) != NULL) {
5754 if (spa_state(spa) != POOL_STATE_ACTIVE ||
5755 !spa_writeable(spa) || spa_suspended(spa))
5757 spa_open_ref(spa, FTAG);
5758 mutex_exit(&spa_namespace_lock);
5759 txg_wait_synced(spa_get_dsl(spa), 0);
5760 mutex_enter(&spa_namespace_lock);
5761 spa_close(spa, FTAG);
5763 mutex_exit(&spa_namespace_lock);
5767 * ==========================================================================
5768 * Miscellaneous routines
5769 * ==========================================================================
5773 * Remove all pools in the system.
5781 * Remove all cached state. All pools should be closed now,
5782 * so every spa in the AVL tree should be unreferenced.
5784 mutex_enter(&spa_namespace_lock);
5785 while ((spa = spa_next(NULL)) != NULL) {
5787 * Stop async tasks. The async thread may need to detach
5788 * a device that's been replaced, which requires grabbing
5789 * spa_namespace_lock, so we must drop it here.
5791 spa_open_ref(spa, FTAG);
5792 mutex_exit(&spa_namespace_lock);
5793 spa_async_suspend(spa);
5794 mutex_enter(&spa_namespace_lock);
5795 spa_close(spa, FTAG);
5797 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
5799 spa_deactivate(spa);
5803 mutex_exit(&spa_namespace_lock);
5807 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
5812 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
5816 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
5817 vd = spa->spa_l2cache.sav_vdevs[i];
5818 if (vd->vdev_guid == guid)
5822 for (i = 0; i < spa->spa_spares.sav_count; i++) {
5823 vd = spa->spa_spares.sav_vdevs[i];
5824 if (vd->vdev_guid == guid)
5833 spa_upgrade(spa_t *spa, uint64_t version)
5835 ASSERT(spa_writeable(spa));
5837 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5840 * This should only be called for a non-faulted pool, and since a
5841 * future version would result in an unopenable pool, this shouldn't be
5844 ASSERT(spa->spa_uberblock.ub_version <= SPA_VERSION);
5845 ASSERT(version >= spa->spa_uberblock.ub_version);
5847 spa->spa_uberblock.ub_version = version;
5848 vdev_config_dirty(spa->spa_root_vdev);
5850 spa_config_exit(spa, SCL_ALL, FTAG);
5852 txg_wait_synced(spa_get_dsl(spa), 0);
5856 spa_has_spare(spa_t *spa, uint64_t guid)
5860 spa_aux_vdev_t *sav = &spa->spa_spares;
5862 for (i = 0; i < sav->sav_count; i++)
5863 if (sav->sav_vdevs[i]->vdev_guid == guid)
5866 for (i = 0; i < sav->sav_npending; i++) {
5867 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
5868 &spareguid) == 0 && spareguid == guid)
5876 * Check if a pool has an active shared spare device.
5877 * Note: reference count of an active spare is 2, as a spare and as a replace
5880 spa_has_active_shared_spare(spa_t *spa)
5884 spa_aux_vdev_t *sav = &spa->spa_spares;
5886 for (i = 0; i < sav->sav_count; i++) {
5887 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
5888 &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
5897 * Post a FM_EREPORT_ZFS_* event from sys/fm/fs/zfs.h. The payload will be
5898 * filled in from the spa and (optionally) the vdev. This doesn't do anything
5899 * in the userland libzpool, as we don't want consumers to misinterpret ztest
5900 * or zdb as real changes.
5903 spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
5906 zfs_ereport_post(name, spa, vd, NULL, 0, 0);
5910 #if defined(_KERNEL) && defined(HAVE_SPL)
5911 /* state manipulation functions */
5912 EXPORT_SYMBOL(spa_open);
5913 EXPORT_SYMBOL(spa_open_rewind);
5914 EXPORT_SYMBOL(spa_get_stats);
5915 EXPORT_SYMBOL(spa_create);
5916 EXPORT_SYMBOL(spa_import_rootpool);
5917 EXPORT_SYMBOL(spa_import);
5918 EXPORT_SYMBOL(spa_tryimport);
5919 EXPORT_SYMBOL(spa_destroy);
5920 EXPORT_SYMBOL(spa_export);
5921 EXPORT_SYMBOL(spa_reset);
5922 EXPORT_SYMBOL(spa_async_request);
5923 EXPORT_SYMBOL(spa_async_suspend);
5924 EXPORT_SYMBOL(spa_async_resume);
5925 EXPORT_SYMBOL(spa_inject_addref);
5926 EXPORT_SYMBOL(spa_inject_delref);
5927 EXPORT_SYMBOL(spa_scan_stat_init);
5928 EXPORT_SYMBOL(spa_scan_get_stats);
5930 /* device maniion */
5931 EXPORT_SYMBOL(spa_vdev_add);
5932 EXPORT_SYMBOL(spa_vdev_attach);
5933 EXPORT_SYMBOL(spa_vdev_detach);
5934 EXPORT_SYMBOL(spa_vdev_remove);
5935 EXPORT_SYMBOL(spa_vdev_setpath);
5936 EXPORT_SYMBOL(spa_vdev_setfru);
5937 EXPORT_SYMBOL(spa_vdev_split_mirror);
5939 /* spare statech is global across all pools) */
5940 EXPORT_SYMBOL(spa_spare_add);
5941 EXPORT_SYMBOL(spa_spare_remove);
5942 EXPORT_SYMBOL(spa_spare_exists);
5943 EXPORT_SYMBOL(spa_spare_activate);
5945 /* L2ARC statech is global across all pools) */
5946 EXPORT_SYMBOL(spa_l2cache_add);
5947 EXPORT_SYMBOL(spa_l2cache_remove);
5948 EXPORT_SYMBOL(spa_l2cache_exists);
5949 EXPORT_SYMBOL(spa_l2cache_activate);
5950 EXPORT_SYMBOL(spa_l2cache_drop);
5953 EXPORT_SYMBOL(spa_scan);
5954 EXPORT_SYMBOL(spa_scan_stop);
5957 EXPORT_SYMBOL(spa_sync); /* only for DMU use */
5958 EXPORT_SYMBOL(spa_sync_allpools);
5961 EXPORT_SYMBOL(spa_prop_set);
5962 EXPORT_SYMBOL(spa_prop_get);
5963 EXPORT_SYMBOL(spa_prop_clear_bootfs);
5965 /* asynchronous event notification */
5966 EXPORT_SYMBOL(spa_event_notify);