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.
26 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
30 * This file contains all the routines used when modifying on-disk SPA state.
31 * This includes opening, importing, destroying, exporting a pool, and syncing a
35 #include <sys/zfs_context.h>
36 #include <sys/fm/fs/zfs.h>
37 #include <sys/spa_impl.h>
39 #include <sys/zio_checksum.h>
41 #include <sys/dmu_tx.h>
45 #include <sys/vdev_impl.h>
46 #include <sys/vdev_disk.h>
47 #include <sys/metaslab.h>
48 #include <sys/metaslab_impl.h>
49 #include <sys/uberblock_impl.h>
52 #include <sys/dmu_traverse.h>
53 #include <sys/dmu_objset.h>
54 #include <sys/unique.h>
55 #include <sys/dsl_pool.h>
56 #include <sys/dsl_dataset.h>
57 #include <sys/dsl_dir.h>
58 #include <sys/dsl_prop.h>
59 #include <sys/dsl_synctask.h>
60 #include <sys/fs/zfs.h>
62 #include <sys/callb.h>
63 #include <sys/systeminfo.h>
64 #include <sys/spa_boot.h>
65 #include <sys/zfs_ioctl.h>
66 #include <sys/dsl_scan.h>
69 #include <sys/bootprops.h>
70 #include <sys/callb.h>
71 #include <sys/cpupart.h>
73 #include <sys/sysdc.h>
78 #include "zfs_comutil.h"
80 typedef enum zti_modes {
81 zti_mode_fixed, /* value is # of threads (min 1) */
82 zti_mode_online_percent, /* value is % of online CPUs */
83 zti_mode_batch, /* cpu-intensive; value is ignored */
84 zti_mode_null, /* don't create a taskq */
88 #define ZTI_FIX(n) { zti_mode_fixed, (n) }
89 #define ZTI_PCT(n) { zti_mode_online_percent, (n) }
90 #define ZTI_BATCH { zti_mode_batch, 0 }
91 #define ZTI_NULL { zti_mode_null, 0 }
93 #define ZTI_ONE ZTI_FIX(1)
95 typedef struct zio_taskq_info {
96 enum zti_modes zti_mode;
100 static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
101 "iss", "iss_h", "int", "int_h"
105 * Define the taskq threads for the following I/O types:
106 * NULL, READ, WRITE, FREE, CLAIM, and IOCTL
108 const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
109 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
110 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL },
111 { ZTI_FIX(8), ZTI_NULL, ZTI_BATCH, ZTI_NULL },
112 { ZTI_BATCH, ZTI_FIX(5), ZTI_FIX(16), ZTI_FIX(5) },
113 { ZTI_PCT(100), ZTI_NULL, ZTI_ONE, ZTI_NULL },
114 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL },
115 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL },
118 static dsl_syncfunc_t spa_sync_props;
119 static boolean_t spa_has_active_shared_spare(spa_t *spa);
120 static inline int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
121 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
123 static void spa_vdev_resilver_done(spa_t *spa);
125 uint_t zio_taskq_batch_pct = 100; /* 1 thread per cpu in pset */
126 id_t zio_taskq_psrset_bind = PS_NONE;
127 boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */
128 uint_t zio_taskq_basedc = 80; /* base duty cycle */
130 boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */
133 * This (illegal) pool name is used when temporarily importing a spa_t in order
134 * to get the vdev stats associated with the imported devices.
136 #define TRYIMPORT_NAME "$import"
139 * ==========================================================================
140 * SPA properties routines
141 * ==========================================================================
145 * Add a (source=src, propname=propval) list to an nvlist.
148 spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
149 uint64_t intval, zprop_source_t src)
151 const char *propname = zpool_prop_to_name(prop);
154 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
155 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
158 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
160 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
162 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
163 nvlist_free(propval);
167 * Get property values from the spa configuration.
170 spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
174 uint64_t cap, version;
175 zprop_source_t src = ZPROP_SRC_NONE;
176 spa_config_dirent_t *dp;
178 ASSERT(MUTEX_HELD(&spa->spa_props_lock));
180 if (spa->spa_root_vdev != NULL) {
181 alloc = metaslab_class_get_alloc(spa_normal_class(spa));
182 size = metaslab_class_get_space(spa_normal_class(spa));
183 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
184 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
185 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
186 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
188 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
189 (spa_mode(spa) == FREAD), src);
191 cap = (size == 0) ? 0 : (alloc * 100 / size);
192 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
194 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
195 ddt_get_pool_dedup_ratio(spa), src);
197 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
198 spa->spa_root_vdev->vdev_state, src);
200 version = spa_version(spa);
201 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
202 src = ZPROP_SRC_DEFAULT;
204 src = ZPROP_SRC_LOCAL;
205 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
208 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
210 if (spa->spa_root != NULL)
211 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
214 if ((dp = list_head(&spa->spa_config_list)) != NULL) {
215 if (dp->scd_path == NULL) {
216 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
217 "none", 0, ZPROP_SRC_LOCAL);
218 } else if (strcmp(dp->scd_path, spa_config_path) != 0) {
219 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
220 dp->scd_path, 0, ZPROP_SRC_LOCAL);
226 * Get zpool property values.
229 spa_prop_get(spa_t *spa, nvlist_t **nvp)
231 objset_t *mos = spa->spa_meta_objset;
236 err = nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP);
240 mutex_enter(&spa->spa_props_lock);
243 * Get properties from the spa config.
245 spa_prop_get_config(spa, nvp);
247 /* If no pool property object, no more prop to get. */
248 if (mos == NULL || spa->spa_pool_props_object == 0) {
249 mutex_exit(&spa->spa_props_lock);
254 * Get properties from the MOS pool property object.
256 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
257 (err = zap_cursor_retrieve(&zc, &za)) == 0;
258 zap_cursor_advance(&zc)) {
261 zprop_source_t src = ZPROP_SRC_DEFAULT;
264 if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
267 switch (za.za_integer_length) {
269 /* integer property */
270 if (za.za_first_integer !=
271 zpool_prop_default_numeric(prop))
272 src = ZPROP_SRC_LOCAL;
274 if (prop == ZPOOL_PROP_BOOTFS) {
276 dsl_dataset_t *ds = NULL;
278 dp = spa_get_dsl(spa);
279 rw_enter(&dp->dp_config_rwlock, RW_READER);
280 if ((err = dsl_dataset_hold_obj(dp,
281 za.za_first_integer, FTAG, &ds))) {
282 rw_exit(&dp->dp_config_rwlock);
287 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
289 dsl_dataset_name(ds, strval);
290 dsl_dataset_rele(ds, FTAG);
291 rw_exit(&dp->dp_config_rwlock);
294 intval = za.za_first_integer;
297 spa_prop_add_list(*nvp, prop, strval, intval, src);
301 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
306 /* string property */
307 strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
308 err = zap_lookup(mos, spa->spa_pool_props_object,
309 za.za_name, 1, za.za_num_integers, strval);
311 kmem_free(strval, za.za_num_integers);
314 spa_prop_add_list(*nvp, prop, strval, 0, src);
315 kmem_free(strval, za.za_num_integers);
322 zap_cursor_fini(&zc);
323 mutex_exit(&spa->spa_props_lock);
325 if (err && err != ENOENT) {
335 * Validate the given pool properties nvlist and modify the list
336 * for the property values to be set.
339 spa_prop_validate(spa_t *spa, nvlist_t *props)
342 int error = 0, reset_bootfs = 0;
346 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
348 char *propname, *strval;
353 propname = nvpair_name(elem);
355 if ((prop = zpool_name_to_prop(propname)) == ZPROP_INVAL)
359 case ZPOOL_PROP_VERSION:
360 error = nvpair_value_uint64(elem, &intval);
362 (intval < spa_version(spa) || intval > SPA_VERSION))
366 case ZPOOL_PROP_DELEGATION:
367 case ZPOOL_PROP_AUTOREPLACE:
368 case ZPOOL_PROP_LISTSNAPS:
369 case ZPOOL_PROP_AUTOEXPAND:
370 error = nvpair_value_uint64(elem, &intval);
371 if (!error && intval > 1)
375 case ZPOOL_PROP_BOOTFS:
377 * If the pool version is less than SPA_VERSION_BOOTFS,
378 * or the pool is still being created (version == 0),
379 * the bootfs property cannot be set.
381 if (spa_version(spa) < SPA_VERSION_BOOTFS) {
387 * Make sure the vdev config is bootable
389 if (!vdev_is_bootable(spa->spa_root_vdev)) {
396 error = nvpair_value_string(elem, &strval);
401 if (strval == NULL || strval[0] == '\0') {
402 objnum = zpool_prop_default_numeric(
407 if ((error = dmu_objset_hold(strval,FTAG,&os)))
410 /* Must be ZPL and not gzip compressed. */
412 if (dmu_objset_type(os) != DMU_OST_ZFS) {
414 } else if ((error = dsl_prop_get_integer(strval,
415 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
416 &compress, NULL)) == 0 &&
417 !BOOTFS_COMPRESS_VALID(compress)) {
420 objnum = dmu_objset_id(os);
422 dmu_objset_rele(os, FTAG);
426 case ZPOOL_PROP_FAILUREMODE:
427 error = nvpair_value_uint64(elem, &intval);
428 if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
429 intval > ZIO_FAILURE_MODE_PANIC))
433 * This is a special case which only occurs when
434 * the pool has completely failed. This allows
435 * the user to change the in-core failmode property
436 * without syncing it out to disk (I/Os might
437 * currently be blocked). We do this by returning
438 * EIO to the caller (spa_prop_set) to trick it
439 * into thinking we encountered a property validation
442 if (!error && spa_suspended(spa)) {
443 spa->spa_failmode = intval;
448 case ZPOOL_PROP_CACHEFILE:
449 if ((error = nvpair_value_string(elem, &strval)) != 0)
452 if (strval[0] == '\0')
455 if (strcmp(strval, "none") == 0)
458 if (strval[0] != '/') {
463 slash = strrchr(strval, '/');
464 ASSERT(slash != NULL);
466 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
467 strcmp(slash, "/..") == 0)
471 case ZPOOL_PROP_DEDUPDITTO:
472 if (spa_version(spa) < SPA_VERSION_DEDUP)
475 error = nvpair_value_uint64(elem, &intval);
477 intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
489 if (!error && reset_bootfs) {
490 error = nvlist_remove(props,
491 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
494 error = nvlist_add_uint64(props,
495 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
503 spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
506 spa_config_dirent_t *dp;
508 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
512 dp = kmem_alloc(sizeof (spa_config_dirent_t),
515 if (cachefile[0] == '\0')
516 dp->scd_path = spa_strdup(spa_config_path);
517 else if (strcmp(cachefile, "none") == 0)
520 dp->scd_path = spa_strdup(cachefile);
522 list_insert_head(&spa->spa_config_list, dp);
524 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
528 spa_prop_set(spa_t *spa, nvlist_t *nvp)
532 boolean_t need_sync = B_FALSE;
535 if ((error = spa_prop_validate(spa, nvp)) != 0)
539 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
540 if ((prop = zpool_name_to_prop(
541 nvpair_name(elem))) == ZPROP_INVAL)
544 if (prop == ZPOOL_PROP_CACHEFILE ||
545 prop == ZPOOL_PROP_ALTROOT ||
546 prop == ZPOOL_PROP_READONLY)
554 return (dsl_sync_task_do(spa_get_dsl(spa), NULL, spa_sync_props,
561 * If the bootfs property value is dsobj, clear it.
564 spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
566 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
567 VERIFY(zap_remove(spa->spa_meta_objset,
568 spa->spa_pool_props_object,
569 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
575 * ==========================================================================
576 * SPA state manipulation (open/create/destroy/import/export)
577 * ==========================================================================
581 spa_error_entry_compare(const void *a, const void *b)
583 spa_error_entry_t *sa = (spa_error_entry_t *)a;
584 spa_error_entry_t *sb = (spa_error_entry_t *)b;
587 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
588 sizeof (zbookmark_t));
599 * Utility function which retrieves copies of the current logs and
600 * re-initializes them in the process.
603 spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
605 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
607 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
608 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
610 avl_create(&spa->spa_errlist_scrub,
611 spa_error_entry_compare, sizeof (spa_error_entry_t),
612 offsetof(spa_error_entry_t, se_avl));
613 avl_create(&spa->spa_errlist_last,
614 spa_error_entry_compare, sizeof (spa_error_entry_t),
615 offsetof(spa_error_entry_t, se_avl));
619 spa_taskq_create(spa_t *spa, const char *name, enum zti_modes mode,
620 uint_t value, uint_t flags)
622 boolean_t batch = B_FALSE;
626 return (NULL); /* no taskq needed */
629 ASSERT3U(value, >=, 1);
630 value = MAX(value, 1);
635 flags |= TASKQ_THREADS_CPU_PCT;
636 value = zio_taskq_batch_pct;
639 case zti_mode_online_percent:
640 flags |= TASKQ_THREADS_CPU_PCT;
644 panic("unrecognized mode for %s taskq (%u:%u) in "
650 if (zio_taskq_sysdc && spa->spa_proc != &p0) {
652 flags |= TASKQ_DC_BATCH;
654 return (taskq_create_sysdc(name, value, 50, INT_MAX,
655 spa->spa_proc, zio_taskq_basedc, flags));
657 return (taskq_create_proc(name, value, maxclsyspri, 50, INT_MAX,
658 spa->spa_proc, flags));
662 spa_create_zio_taskqs(spa_t *spa)
666 for (t = 0; t < ZIO_TYPES; t++) {
667 for (q = 0; q < ZIO_TASKQ_TYPES; q++) {
668 const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
669 enum zti_modes mode = ztip->zti_mode;
670 uint_t value = ztip->zti_value;
674 if (t == ZIO_TYPE_WRITE)
675 flags |= TASKQ_NORECLAIM;
677 (void) snprintf(name, sizeof (name),
678 "%s_%s", zio_type_name[t], zio_taskq_types[q]);
680 spa->spa_zio_taskq[t][q] =
681 spa_taskq_create(spa, name, mode, value, flags);
686 #if defined(_KERNEL) && defined(HAVE_SPA_THREAD)
688 spa_thread(void *arg)
693 user_t *pu = PTOU(curproc);
695 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
698 ASSERT(curproc != &p0);
699 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
700 "zpool-%s", spa->spa_name);
701 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
703 /* bind this thread to the requested psrset */
704 if (zio_taskq_psrset_bind != PS_NONE) {
706 mutex_enter(&cpu_lock);
707 mutex_enter(&pidlock);
708 mutex_enter(&curproc->p_lock);
710 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
711 0, NULL, NULL) == 0) {
712 curthread->t_bind_pset = zio_taskq_psrset_bind;
715 "Couldn't bind process for zfs pool \"%s\" to "
716 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
719 mutex_exit(&curproc->p_lock);
720 mutex_exit(&pidlock);
721 mutex_exit(&cpu_lock);
725 if (zio_taskq_sysdc) {
726 sysdc_thread_enter(curthread, 100, 0);
729 spa->spa_proc = curproc;
730 spa->spa_did = curthread->t_did;
732 spa_create_zio_taskqs(spa);
734 mutex_enter(&spa->spa_proc_lock);
735 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
737 spa->spa_proc_state = SPA_PROC_ACTIVE;
738 cv_broadcast(&spa->spa_proc_cv);
740 CALLB_CPR_SAFE_BEGIN(&cprinfo);
741 while (spa->spa_proc_state == SPA_PROC_ACTIVE)
742 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
743 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
745 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
746 spa->spa_proc_state = SPA_PROC_GONE;
748 cv_broadcast(&spa->spa_proc_cv);
749 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */
751 mutex_enter(&curproc->p_lock);
757 * Activate an uninitialized pool.
760 spa_activate(spa_t *spa, int mode)
762 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
764 spa->spa_state = POOL_STATE_ACTIVE;
765 spa->spa_mode = mode;
767 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
768 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
770 /* Try to create a covering process */
771 mutex_enter(&spa->spa_proc_lock);
772 ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
773 ASSERT(spa->spa_proc == &p0);
776 #ifdef HAVE_SPA_THREAD
777 /* Only create a process if we're going to be around a while. */
778 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
779 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
781 spa->spa_proc_state = SPA_PROC_CREATED;
782 while (spa->spa_proc_state == SPA_PROC_CREATED) {
783 cv_wait(&spa->spa_proc_cv,
784 &spa->spa_proc_lock);
786 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
787 ASSERT(spa->spa_proc != &p0);
788 ASSERT(spa->spa_did != 0);
792 "Couldn't create process for zfs pool \"%s\"\n",
797 #endif /* HAVE_SPA_THREAD */
798 mutex_exit(&spa->spa_proc_lock);
800 /* If we didn't create a process, we need to create our taskqs. */
801 if (spa->spa_proc == &p0) {
802 spa_create_zio_taskqs(spa);
805 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
806 offsetof(vdev_t, vdev_config_dirty_node));
807 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
808 offsetof(vdev_t, vdev_state_dirty_node));
810 txg_list_create(&spa->spa_vdev_txg_list,
811 offsetof(struct vdev, vdev_txg_node));
813 avl_create(&spa->spa_errlist_scrub,
814 spa_error_entry_compare, sizeof (spa_error_entry_t),
815 offsetof(spa_error_entry_t, se_avl));
816 avl_create(&spa->spa_errlist_last,
817 spa_error_entry_compare, sizeof (spa_error_entry_t),
818 offsetof(spa_error_entry_t, se_avl));
822 * Opposite of spa_activate().
825 spa_deactivate(spa_t *spa)
829 ASSERT(spa->spa_sync_on == B_FALSE);
830 ASSERT(spa->spa_dsl_pool == NULL);
831 ASSERT(spa->spa_root_vdev == NULL);
832 ASSERT(spa->spa_async_zio_root == NULL);
833 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
835 txg_list_destroy(&spa->spa_vdev_txg_list);
837 list_destroy(&spa->spa_config_dirty_list);
838 list_destroy(&spa->spa_state_dirty_list);
840 for (t = 0; t < ZIO_TYPES; t++) {
841 for (q = 0; q < ZIO_TASKQ_TYPES; q++) {
842 if (spa->spa_zio_taskq[t][q] != NULL)
843 taskq_destroy(spa->spa_zio_taskq[t][q]);
844 spa->spa_zio_taskq[t][q] = NULL;
848 metaslab_class_destroy(spa->spa_normal_class);
849 spa->spa_normal_class = NULL;
851 metaslab_class_destroy(spa->spa_log_class);
852 spa->spa_log_class = NULL;
855 * If this was part of an import or the open otherwise failed, we may
856 * still have errors left in the queues. Empty them just in case.
858 spa_errlog_drain(spa);
860 avl_destroy(&spa->spa_errlist_scrub);
861 avl_destroy(&spa->spa_errlist_last);
863 spa->spa_state = POOL_STATE_UNINITIALIZED;
865 mutex_enter(&spa->spa_proc_lock);
866 if (spa->spa_proc_state != SPA_PROC_NONE) {
867 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
868 spa->spa_proc_state = SPA_PROC_DEACTIVATE;
869 cv_broadcast(&spa->spa_proc_cv);
870 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
871 ASSERT(spa->spa_proc != &p0);
872 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
874 ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
875 spa->spa_proc_state = SPA_PROC_NONE;
877 ASSERT(spa->spa_proc == &p0);
878 mutex_exit(&spa->spa_proc_lock);
881 * We want to make sure spa_thread() has actually exited the ZFS
882 * module, so that the module can't be unloaded out from underneath
885 if (spa->spa_did != 0) {
886 thread_join(spa->spa_did);
892 * Verify a pool configuration, and construct the vdev tree appropriately. This
893 * will create all the necessary vdevs in the appropriate layout, with each vdev
894 * in the CLOSED state. This will prep the pool before open/creation/import.
895 * All vdev validation is done by the vdev_alloc() routine.
898 spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
899 uint_t id, int atype)
906 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
909 if ((*vdp)->vdev_ops->vdev_op_leaf)
912 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
924 for (c = 0; c < children; c++) {
926 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
934 ASSERT(*vdp != NULL);
940 * Opposite of spa_load().
943 spa_unload(spa_t *spa)
947 ASSERT(MUTEX_HELD(&spa_namespace_lock));
952 spa_async_suspend(spa);
957 if (spa->spa_sync_on) {
958 txg_sync_stop(spa->spa_dsl_pool);
959 spa->spa_sync_on = B_FALSE;
963 * Wait for any outstanding async I/O to complete.
965 if (spa->spa_async_zio_root != NULL) {
966 (void) zio_wait(spa->spa_async_zio_root);
967 spa->spa_async_zio_root = NULL;
970 bpobj_close(&spa->spa_deferred_bpobj);
973 * Close the dsl pool.
975 if (spa->spa_dsl_pool) {
976 dsl_pool_close(spa->spa_dsl_pool);
977 spa->spa_dsl_pool = NULL;
978 spa->spa_meta_objset = NULL;
983 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
986 * Drop and purge level 2 cache
988 spa_l2cache_drop(spa);
993 if (spa->spa_root_vdev)
994 vdev_free(spa->spa_root_vdev);
995 ASSERT(spa->spa_root_vdev == NULL);
997 for (i = 0; i < spa->spa_spares.sav_count; i++)
998 vdev_free(spa->spa_spares.sav_vdevs[i]);
999 if (spa->spa_spares.sav_vdevs) {
1000 kmem_free(spa->spa_spares.sav_vdevs,
1001 spa->spa_spares.sav_count * sizeof (void *));
1002 spa->spa_spares.sav_vdevs = NULL;
1004 if (spa->spa_spares.sav_config) {
1005 nvlist_free(spa->spa_spares.sav_config);
1006 spa->spa_spares.sav_config = NULL;
1008 spa->spa_spares.sav_count = 0;
1010 for (i = 0; i < spa->spa_l2cache.sav_count; i++)
1011 vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1012 if (spa->spa_l2cache.sav_vdevs) {
1013 kmem_free(spa->spa_l2cache.sav_vdevs,
1014 spa->spa_l2cache.sav_count * sizeof (void *));
1015 spa->spa_l2cache.sav_vdevs = NULL;
1017 if (spa->spa_l2cache.sav_config) {
1018 nvlist_free(spa->spa_l2cache.sav_config);
1019 spa->spa_l2cache.sav_config = NULL;
1021 spa->spa_l2cache.sav_count = 0;
1023 spa->spa_async_suspended = 0;
1025 spa_config_exit(spa, SCL_ALL, FTAG);
1029 * Load (or re-load) the current list of vdevs describing the active spares for
1030 * this pool. When this is called, we have some form of basic information in
1031 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1032 * then re-generate a more complete list including status information.
1035 spa_load_spares(spa_t *spa)
1042 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1045 * First, close and free any existing spare vdevs.
1047 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1048 vd = spa->spa_spares.sav_vdevs[i];
1050 /* Undo the call to spa_activate() below */
1051 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1052 B_FALSE)) != NULL && tvd->vdev_isspare)
1053 spa_spare_remove(tvd);
1058 if (spa->spa_spares.sav_vdevs)
1059 kmem_free(spa->spa_spares.sav_vdevs,
1060 spa->spa_spares.sav_count * sizeof (void *));
1062 if (spa->spa_spares.sav_config == NULL)
1065 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1066 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1068 spa->spa_spares.sav_count = (int)nspares;
1069 spa->spa_spares.sav_vdevs = NULL;
1075 * Construct the array of vdevs, opening them to get status in the
1076 * process. For each spare, there is potentially two different vdev_t
1077 * structures associated with it: one in the list of spares (used only
1078 * for basic validation purposes) and one in the active vdev
1079 * configuration (if it's spared in). During this phase we open and
1080 * validate each vdev on the spare list. If the vdev also exists in the
1081 * active configuration, then we also mark this vdev as an active spare.
1083 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1085 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1086 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1087 VDEV_ALLOC_SPARE) == 0);
1090 spa->spa_spares.sav_vdevs[i] = vd;
1092 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1093 B_FALSE)) != NULL) {
1094 if (!tvd->vdev_isspare)
1098 * We only mark the spare active if we were successfully
1099 * able to load the vdev. Otherwise, importing a pool
1100 * with a bad active spare would result in strange
1101 * behavior, because multiple pool would think the spare
1102 * is actively in use.
1104 * There is a vulnerability here to an equally bizarre
1105 * circumstance, where a dead active spare is later
1106 * brought back to life (onlined or otherwise). Given
1107 * the rarity of this scenario, and the extra complexity
1108 * it adds, we ignore the possibility.
1110 if (!vdev_is_dead(tvd))
1111 spa_spare_activate(tvd);
1115 vd->vdev_aux = &spa->spa_spares;
1117 if (vdev_open(vd) != 0)
1120 if (vdev_validate_aux(vd) == 0)
1125 * Recompute the stashed list of spares, with status information
1128 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1129 DATA_TYPE_NVLIST_ARRAY) == 0);
1131 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1133 for (i = 0; i < spa->spa_spares.sav_count; i++)
1134 spares[i] = vdev_config_generate(spa,
1135 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1136 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1137 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1138 for (i = 0; i < spa->spa_spares.sav_count; i++)
1139 nvlist_free(spares[i]);
1140 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1144 * Load (or re-load) the current list of vdevs describing the active l2cache for
1145 * this pool. When this is called, we have some form of basic information in
1146 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1147 * then re-generate a more complete list including status information.
1148 * Devices which are already active have their details maintained, and are
1152 spa_load_l2cache(spa_t *spa)
1156 int i, j, oldnvdevs;
1158 vdev_t *vd, **oldvdevs, **newvdevs = NULL;
1159 spa_aux_vdev_t *sav = &spa->spa_l2cache;
1161 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1163 if (sav->sav_config != NULL) {
1164 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1165 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1166 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1171 oldvdevs = sav->sav_vdevs;
1172 oldnvdevs = sav->sav_count;
1173 sav->sav_vdevs = NULL;
1177 * Process new nvlist of vdevs.
1179 for (i = 0; i < nl2cache; i++) {
1180 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1184 for (j = 0; j < oldnvdevs; j++) {
1186 if (vd != NULL && guid == vd->vdev_guid) {
1188 * Retain previous vdev for add/remove ops.
1196 if (newvdevs[i] == NULL) {
1200 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1201 VDEV_ALLOC_L2CACHE) == 0);
1206 * Commit this vdev as an l2cache device,
1207 * even if it fails to open.
1209 spa_l2cache_add(vd);
1214 spa_l2cache_activate(vd);
1216 if (vdev_open(vd) != 0)
1219 (void) vdev_validate_aux(vd);
1221 if (!vdev_is_dead(vd))
1222 l2arc_add_vdev(spa, vd);
1227 * Purge vdevs that were dropped
1229 for (i = 0; i < oldnvdevs; i++) {
1234 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1235 pool != 0ULL && l2arc_vdev_present(vd))
1236 l2arc_remove_vdev(vd);
1237 (void) vdev_close(vd);
1238 spa_l2cache_remove(vd);
1243 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1245 if (sav->sav_config == NULL)
1248 sav->sav_vdevs = newvdevs;
1249 sav->sav_count = (int)nl2cache;
1252 * Recompute the stashed list of l2cache devices, with status
1253 * information this time.
1255 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1256 DATA_TYPE_NVLIST_ARRAY) == 0);
1258 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1259 for (i = 0; i < sav->sav_count; i++)
1260 l2cache[i] = vdev_config_generate(spa,
1261 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1262 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1263 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1265 for (i = 0; i < sav->sav_count; i++)
1266 nvlist_free(l2cache[i]);
1268 kmem_free(l2cache, sav->sav_count * sizeof (void *));
1272 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1275 char *packed = NULL;
1280 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
1281 nvsize = *(uint64_t *)db->db_data;
1282 dmu_buf_rele(db, FTAG);
1284 packed = kmem_alloc(nvsize, KM_SLEEP | KM_NODEBUG);
1285 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1288 error = nvlist_unpack(packed, nvsize, value, 0);
1289 kmem_free(packed, nvsize);
1295 * Checks to see if the given vdev could not be opened, in which case we post a
1296 * sysevent to notify the autoreplace code that the device has been removed.
1299 spa_check_removed(vdev_t *vd)
1303 for (c = 0; c < vd->vdev_children; c++)
1304 spa_check_removed(vd->vdev_child[c]);
1306 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd)) {
1307 zfs_ereport_post(FM_EREPORT_RESOURCE_AUTOREPLACE,
1308 vd->vdev_spa, vd, NULL, 0, 0);
1309 spa_event_notify(vd->vdev_spa, vd, FM_EREPORT_ZFS_DEVICE_CHECK);
1314 * Validate the current config against the MOS config
1317 spa_config_valid(spa_t *spa, nvlist_t *config)
1319 vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1323 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1325 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1326 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1328 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1331 * If we're doing a normal import, then build up any additional
1332 * diagnostic information about missing devices in this config.
1333 * We'll pass this up to the user for further processing.
1335 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1336 nvlist_t **child, *nv;
1339 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1341 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1343 for (c = 0; c < rvd->vdev_children; c++) {
1344 vdev_t *tvd = rvd->vdev_child[c];
1345 vdev_t *mtvd = mrvd->vdev_child[c];
1347 if (tvd->vdev_ops == &vdev_missing_ops &&
1348 mtvd->vdev_ops != &vdev_missing_ops &&
1350 child[idx++] = vdev_config_generate(spa, mtvd,
1355 VERIFY(nvlist_add_nvlist_array(nv,
1356 ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1357 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1358 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1360 for (i = 0; i < idx; i++)
1361 nvlist_free(child[i]);
1364 kmem_free(child, rvd->vdev_children * sizeof (char **));
1368 * Compare the root vdev tree with the information we have
1369 * from the MOS config (mrvd). Check each top-level vdev
1370 * with the corresponding MOS config top-level (mtvd).
1372 for (c = 0; c < rvd->vdev_children; c++) {
1373 vdev_t *tvd = rvd->vdev_child[c];
1374 vdev_t *mtvd = mrvd->vdev_child[c];
1377 * Resolve any "missing" vdevs in the current configuration.
1378 * If we find that the MOS config has more accurate information
1379 * about the top-level vdev then use that vdev instead.
1381 if (tvd->vdev_ops == &vdev_missing_ops &&
1382 mtvd->vdev_ops != &vdev_missing_ops) {
1384 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1388 * Device specific actions.
1390 if (mtvd->vdev_islog) {
1391 spa_set_log_state(spa, SPA_LOG_CLEAR);
1394 * XXX - once we have 'readonly' pool
1395 * support we should be able to handle
1396 * missing data devices by transitioning
1397 * the pool to readonly.
1403 * Swap the missing vdev with the data we were
1404 * able to obtain from the MOS config.
1406 vdev_remove_child(rvd, tvd);
1407 vdev_remove_child(mrvd, mtvd);
1409 vdev_add_child(rvd, mtvd);
1410 vdev_add_child(mrvd, tvd);
1412 spa_config_exit(spa, SCL_ALL, FTAG);
1414 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1417 } else if (mtvd->vdev_islog) {
1419 * Load the slog device's state from the MOS config
1420 * since it's possible that the label does not
1421 * contain the most up-to-date information.
1423 vdev_load_log_state(tvd, mtvd);
1428 spa_config_exit(spa, SCL_ALL, FTAG);
1431 * Ensure we were able to validate the config.
1433 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1437 * Check for missing log devices
1440 spa_check_logs(spa_t *spa)
1442 switch (spa->spa_log_state) {
1445 case SPA_LOG_MISSING:
1446 /* need to recheck in case slog has been restored */
1447 case SPA_LOG_UNKNOWN:
1448 if (dmu_objset_find(spa->spa_name, zil_check_log_chain, NULL,
1449 DS_FIND_CHILDREN)) {
1450 spa_set_log_state(spa, SPA_LOG_MISSING);
1459 spa_passivate_log(spa_t *spa)
1461 vdev_t *rvd = spa->spa_root_vdev;
1462 boolean_t slog_found = B_FALSE;
1465 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1467 if (!spa_has_slogs(spa))
1470 for (c = 0; c < rvd->vdev_children; c++) {
1471 vdev_t *tvd = rvd->vdev_child[c];
1472 metaslab_group_t *mg = tvd->vdev_mg;
1474 if (tvd->vdev_islog) {
1475 metaslab_group_passivate(mg);
1476 slog_found = B_TRUE;
1480 return (slog_found);
1484 spa_activate_log(spa_t *spa)
1486 vdev_t *rvd = spa->spa_root_vdev;
1489 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1491 for (c = 0; c < rvd->vdev_children; c++) {
1492 vdev_t *tvd = rvd->vdev_child[c];
1493 metaslab_group_t *mg = tvd->vdev_mg;
1495 if (tvd->vdev_islog)
1496 metaslab_group_activate(mg);
1501 spa_offline_log(spa_t *spa)
1505 if ((error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1506 NULL, DS_FIND_CHILDREN)) == 0) {
1509 * We successfully offlined the log device, sync out the
1510 * current txg so that the "stubby" block can be removed
1513 txg_wait_synced(spa->spa_dsl_pool, 0);
1519 spa_aux_check_removed(spa_aux_vdev_t *sav)
1523 for (i = 0; i < sav->sav_count; i++)
1524 spa_check_removed(sav->sav_vdevs[i]);
1528 spa_claim_notify(zio_t *zio)
1530 spa_t *spa = zio->io_spa;
1535 mutex_enter(&spa->spa_props_lock); /* any mutex will do */
1536 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1537 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1538 mutex_exit(&spa->spa_props_lock);
1541 typedef struct spa_load_error {
1542 uint64_t sle_meta_count;
1543 uint64_t sle_data_count;
1547 spa_load_verify_done(zio_t *zio)
1549 blkptr_t *bp = zio->io_bp;
1550 spa_load_error_t *sle = zio->io_private;
1551 dmu_object_type_t type = BP_GET_TYPE(bp);
1552 int error = zio->io_error;
1555 if ((BP_GET_LEVEL(bp) != 0 || dmu_ot[type].ot_metadata) &&
1556 type != DMU_OT_INTENT_LOG)
1557 atomic_add_64(&sle->sle_meta_count, 1);
1559 atomic_add_64(&sle->sle_data_count, 1);
1561 zio_data_buf_free(zio->io_data, zio->io_size);
1566 spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1567 arc_buf_t *pbuf, const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1571 size_t size = BP_GET_PSIZE(bp);
1572 void *data = zio_data_buf_alloc(size);
1574 zio_nowait(zio_read(rio, spa, bp, data, size,
1575 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1576 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1577 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1583 spa_load_verify(spa_t *spa)
1586 spa_load_error_t sle = { 0 };
1587 zpool_rewind_policy_t policy;
1588 boolean_t verify_ok = B_FALSE;
1591 zpool_get_rewind_policy(spa->spa_config, &policy);
1593 if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1596 rio = zio_root(spa, NULL, &sle,
1597 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1599 error = traverse_pool(spa, spa->spa_verify_min_txg,
1600 TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio);
1602 (void) zio_wait(rio);
1604 spa->spa_load_meta_errors = sle.sle_meta_count;
1605 spa->spa_load_data_errors = sle.sle_data_count;
1607 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1608 sle.sle_data_count <= policy.zrp_maxdata) {
1612 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1613 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1615 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1616 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1617 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1618 VERIFY(nvlist_add_int64(spa->spa_load_info,
1619 ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1620 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1621 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1623 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1627 if (error != ENXIO && error != EIO)
1632 return (verify_ok ? 0 : EIO);
1636 * Find a value in the pool props object.
1639 spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1641 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1642 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1646 * Find a value in the pool directory object.
1649 spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1651 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1652 name, sizeof (uint64_t), 1, val));
1656 spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1658 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1663 * Fix up config after a partly-completed split. This is done with the
1664 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
1665 * pool have that entry in their config, but only the splitting one contains
1666 * a list of all the guids of the vdevs that are being split off.
1668 * This function determines what to do with that list: either rejoin
1669 * all the disks to the pool, or complete the splitting process. To attempt
1670 * the rejoin, each disk that is offlined is marked online again, and
1671 * we do a reopen() call. If the vdev label for every disk that was
1672 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1673 * then we call vdev_split() on each disk, and complete the split.
1675 * Otherwise we leave the config alone, with all the vdevs in place in
1676 * the original pool.
1679 spa_try_repair(spa_t *spa, nvlist_t *config)
1686 boolean_t attempt_reopen;
1688 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
1691 /* check that the config is complete */
1692 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
1693 &glist, &gcount) != 0)
1696 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
1698 /* attempt to online all the vdevs & validate */
1699 attempt_reopen = B_TRUE;
1700 for (i = 0; i < gcount; i++) {
1701 if (glist[i] == 0) /* vdev is hole */
1704 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
1705 if (vd[i] == NULL) {
1707 * Don't bother attempting to reopen the disks;
1708 * just do the split.
1710 attempt_reopen = B_FALSE;
1712 /* attempt to re-online it */
1713 vd[i]->vdev_offline = B_FALSE;
1717 if (attempt_reopen) {
1718 vdev_reopen(spa->spa_root_vdev);
1720 /* check each device to see what state it's in */
1721 for (extracted = 0, i = 0; i < gcount; i++) {
1722 if (vd[i] != NULL &&
1723 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
1730 * If every disk has been moved to the new pool, or if we never
1731 * even attempted to look at them, then we split them off for
1734 if (!attempt_reopen || gcount == extracted) {
1735 for (i = 0; i < gcount; i++)
1738 vdev_reopen(spa->spa_root_vdev);
1741 kmem_free(vd, gcount * sizeof (vdev_t *));
1745 spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
1746 boolean_t mosconfig)
1748 nvlist_t *config = spa->spa_config;
1749 char *ereport = FM_EREPORT_ZFS_POOL;
1754 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
1758 * Versioning wasn't explicitly added to the label until later, so if
1759 * it's not present treat it as the initial version.
1761 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
1762 &spa->spa_ubsync.ub_version) != 0)
1763 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
1765 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
1766 &spa->spa_config_txg);
1768 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
1769 spa_guid_exists(pool_guid, 0)) {
1772 spa->spa_load_guid = pool_guid;
1774 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
1776 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
1780 gethrestime(&spa->spa_loaded_ts);
1781 error = spa_load_impl(spa, pool_guid, config, state, type,
1782 mosconfig, &ereport);
1785 spa->spa_minref = refcount_count(&spa->spa_refcount);
1787 if (error != EEXIST) {
1788 spa->spa_loaded_ts.tv_sec = 0;
1789 spa->spa_loaded_ts.tv_nsec = 0;
1791 if (error != EBADF) {
1792 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
1795 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
1802 * Load an existing storage pool, using the pool's builtin spa_config as a
1803 * source of configuration information.
1805 __attribute__((always_inline))
1807 spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
1808 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
1812 nvlist_t *nvroot = NULL;
1814 uberblock_t *ub = &spa->spa_uberblock;
1815 uint64_t children, config_cache_txg = spa->spa_config_txg;
1816 int orig_mode = spa->spa_mode;
1821 * If this is an untrusted config, access the pool in read-only mode.
1822 * This prevents things like resilvering recently removed devices.
1825 spa->spa_mode = FREAD;
1827 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1829 spa->spa_load_state = state;
1831 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
1834 parse = (type == SPA_IMPORT_EXISTING ?
1835 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
1838 * Create "The Godfather" zio to hold all async IOs
1840 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
1841 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
1844 * Parse the configuration into a vdev tree. We explicitly set the
1845 * value that will be returned by spa_version() since parsing the
1846 * configuration requires knowing the version number.
1848 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1849 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
1850 spa_config_exit(spa, SCL_ALL, FTAG);
1855 ASSERT(spa->spa_root_vdev == rvd);
1857 if (type != SPA_IMPORT_ASSEMBLE) {
1858 ASSERT(spa_guid(spa) == pool_guid);
1862 * Try to open all vdevs, loading each label in the process.
1864 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1865 error = vdev_open(rvd);
1866 spa_config_exit(spa, SCL_ALL, FTAG);
1871 * We need to validate the vdev labels against the configuration that
1872 * we have in hand, which is dependent on the setting of mosconfig. If
1873 * mosconfig is true then we're validating the vdev labels based on
1874 * that config. Otherwise, we're validating against the cached config
1875 * (zpool.cache) that was read when we loaded the zfs module, and then
1876 * later we will recursively call spa_load() and validate against
1879 * If we're assembling a new pool that's been split off from an
1880 * existing pool, the labels haven't yet been updated so we skip
1881 * validation for now.
1883 if (type != SPA_IMPORT_ASSEMBLE) {
1884 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1885 error = vdev_validate(rvd);
1886 spa_config_exit(spa, SCL_ALL, FTAG);
1891 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
1896 * Find the best uberblock.
1898 vdev_uberblock_load(NULL, rvd, ub);
1901 * If we weren't able to find a single valid uberblock, return failure.
1903 if (ub->ub_txg == 0)
1904 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
1907 * If the pool is newer than the code, we can't open it.
1909 if (ub->ub_version > SPA_VERSION)
1910 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
1913 * If the vdev guid sum doesn't match the uberblock, we have an
1914 * incomplete configuration. We first check to see if the pool
1915 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
1916 * If it is, defer the vdev_guid_sum check till later so we
1917 * can handle missing vdevs.
1919 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
1920 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
1921 rvd->vdev_guid_sum != ub->ub_guid_sum)
1922 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
1924 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
1925 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1926 spa_try_repair(spa, config);
1927 spa_config_exit(spa, SCL_ALL, FTAG);
1928 nvlist_free(spa->spa_config_splitting);
1929 spa->spa_config_splitting = NULL;
1933 * Initialize internal SPA structures.
1935 spa->spa_state = POOL_STATE_ACTIVE;
1936 spa->spa_ubsync = spa->spa_uberblock;
1937 spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
1938 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
1939 spa->spa_first_txg = spa->spa_last_ubsync_txg ?
1940 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
1941 spa->spa_claim_max_txg = spa->spa_first_txg;
1942 spa->spa_prev_software_version = ub->ub_software_version;
1944 error = dsl_pool_open(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
1946 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1947 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
1949 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
1950 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1954 nvlist_t *policy = NULL, *nvconfig;
1956 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
1957 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1959 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
1960 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
1962 unsigned long myhostid = 0;
1964 VERIFY(nvlist_lookup_string(nvconfig,
1965 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
1968 myhostid = zone_get_hostid(NULL);
1971 * We're emulating the system's hostid in userland, so
1972 * we can't use zone_get_hostid().
1974 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
1975 #endif /* _KERNEL */
1976 if (hostid != 0 && myhostid != 0 &&
1977 hostid != myhostid) {
1978 nvlist_free(nvconfig);
1979 cmn_err(CE_WARN, "pool '%s' could not be "
1980 "loaded as it was last accessed by "
1981 "another system (host: %s hostid: 0x%lx). "
1982 "See: http://zfsonlinux.org/msg/ZFS-8000-EY",
1983 spa_name(spa), hostname,
1984 (unsigned long)hostid);
1988 if (nvlist_lookup_nvlist(spa->spa_config,
1989 ZPOOL_REWIND_POLICY, &policy) == 0)
1990 VERIFY(nvlist_add_nvlist(nvconfig,
1991 ZPOOL_REWIND_POLICY, policy) == 0);
1993 spa_config_set(spa, nvconfig);
1995 spa_deactivate(spa);
1996 spa_activate(spa, orig_mode);
1998 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2001 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2002 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2003 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2005 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2008 * Load the bit that tells us to use the new accounting function
2009 * (raid-z deflation). If we have an older pool, this will not
2012 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2013 if (error != 0 && error != ENOENT)
2014 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2016 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2017 &spa->spa_creation_version);
2018 if (error != 0 && error != ENOENT)
2019 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2022 * Load the persistent error log. If we have an older pool, this will
2025 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2026 if (error != 0 && error != ENOENT)
2027 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2029 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2030 &spa->spa_errlog_scrub);
2031 if (error != 0 && error != ENOENT)
2032 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2035 * Load the history object. If we have an older pool, this
2036 * will not be present.
2038 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2039 if (error != 0 && error != ENOENT)
2040 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2043 * If we're assembling the pool from the split-off vdevs of
2044 * an existing pool, we don't want to attach the spares & cache
2049 * Load any hot spares for this pool.
2051 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2052 if (error != 0 && error != ENOENT)
2053 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2054 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2055 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2056 if (load_nvlist(spa, spa->spa_spares.sav_object,
2057 &spa->spa_spares.sav_config) != 0)
2058 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2060 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2061 spa_load_spares(spa);
2062 spa_config_exit(spa, SCL_ALL, FTAG);
2063 } else if (error == 0) {
2064 spa->spa_spares.sav_sync = B_TRUE;
2068 * Load any level 2 ARC devices for this pool.
2070 error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2071 &spa->spa_l2cache.sav_object);
2072 if (error != 0 && error != ENOENT)
2073 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2074 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2075 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2076 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2077 &spa->spa_l2cache.sav_config) != 0)
2078 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2080 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2081 spa_load_l2cache(spa);
2082 spa_config_exit(spa, SCL_ALL, FTAG);
2083 } else if (error == 0) {
2084 spa->spa_l2cache.sav_sync = B_TRUE;
2087 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2089 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2090 if (error && error != ENOENT)
2091 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2094 uint64_t autoreplace;
2096 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2097 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2098 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2099 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2100 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2101 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2102 &spa->spa_dedup_ditto);
2104 spa->spa_autoreplace = (autoreplace != 0);
2108 * If the 'autoreplace' property is set, then post a resource notifying
2109 * the ZFS DE that it should not issue any faults for unopenable
2110 * devices. We also iterate over the vdevs, and post a sysevent for any
2111 * unopenable vdevs so that the normal autoreplace handler can take
2114 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2115 spa_check_removed(spa->spa_root_vdev);
2117 * For the import case, this is done in spa_import(), because
2118 * at this point we're using the spare definitions from
2119 * the MOS config, not necessarily from the userland config.
2121 if (state != SPA_LOAD_IMPORT) {
2122 spa_aux_check_removed(&spa->spa_spares);
2123 spa_aux_check_removed(&spa->spa_l2cache);
2128 * Load the vdev state for all toplevel vdevs.
2133 * Propagate the leaf DTLs we just loaded all the way up the tree.
2135 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2136 vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2137 spa_config_exit(spa, SCL_ALL, FTAG);
2140 * Load the DDTs (dedup tables).
2142 error = ddt_load(spa);
2144 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2146 spa_update_dspace(spa);
2149 * Validate the config, using the MOS config to fill in any
2150 * information which might be missing. If we fail to validate
2151 * the config then declare the pool unfit for use. If we're
2152 * assembling a pool from a split, the log is not transferred
2155 if (type != SPA_IMPORT_ASSEMBLE) {
2158 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2159 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2161 if (!spa_config_valid(spa, nvconfig)) {
2162 nvlist_free(nvconfig);
2163 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2166 nvlist_free(nvconfig);
2169 * Now that we've validate the config, check the state of the
2170 * root vdev. If it can't be opened, it indicates one or
2171 * more toplevel vdevs are faulted.
2173 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2176 if (spa_check_logs(spa)) {
2177 *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2178 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2183 * We've successfully opened the pool, verify that we're ready
2184 * to start pushing transactions.
2186 if (state != SPA_LOAD_TRYIMPORT) {
2187 if ((error = spa_load_verify(spa)))
2188 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2192 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2193 spa->spa_load_max_txg == UINT64_MAX)) {
2195 int need_update = B_FALSE;
2198 ASSERT(state != SPA_LOAD_TRYIMPORT);
2201 * Claim log blocks that haven't been committed yet.
2202 * This must all happen in a single txg.
2203 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2204 * invoked from zil_claim_log_block()'s i/o done callback.
2205 * Price of rollback is that we abandon the log.
2207 spa->spa_claiming = B_TRUE;
2209 tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2210 spa_first_txg(spa));
2211 (void) dmu_objset_find(spa_name(spa),
2212 zil_claim, tx, DS_FIND_CHILDREN);
2215 spa->spa_claiming = B_FALSE;
2217 spa_set_log_state(spa, SPA_LOG_GOOD);
2218 spa->spa_sync_on = B_TRUE;
2219 txg_sync_start(spa->spa_dsl_pool);
2222 * Wait for all claims to sync. We sync up to the highest
2223 * claimed log block birth time so that claimed log blocks
2224 * don't appear to be from the future. spa_claim_max_txg
2225 * will have been set for us by either zil_check_log_chain()
2226 * (invoked from spa_check_logs()) or zil_claim() above.
2228 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2231 * If the config cache is stale, or we have uninitialized
2232 * metaslabs (see spa_vdev_add()), then update the config.
2234 * If this is a verbatim import, trust the current
2235 * in-core spa_config and update the disk labels.
2237 if (config_cache_txg != spa->spa_config_txg ||
2238 state == SPA_LOAD_IMPORT ||
2239 state == SPA_LOAD_RECOVER ||
2240 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2241 need_update = B_TRUE;
2243 for (c = 0; c < rvd->vdev_children; c++)
2244 if (rvd->vdev_child[c]->vdev_ms_array == 0)
2245 need_update = B_TRUE;
2248 * Update the config cache asychronously in case we're the
2249 * root pool, in which case the config cache isn't writable yet.
2252 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2255 * Check all DTLs to see if anything needs resilvering.
2257 if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2258 vdev_resilver_needed(rvd, NULL, NULL))
2259 spa_async_request(spa, SPA_ASYNC_RESILVER);
2262 * Delete any inconsistent datasets.
2264 (void) dmu_objset_find(spa_name(spa),
2265 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2268 * Clean up any stale temporary dataset userrefs.
2270 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2277 spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2279 int mode = spa->spa_mode;
2282 spa_deactivate(spa);
2284 spa->spa_load_max_txg--;
2286 spa_activate(spa, mode);
2287 spa_async_suspend(spa);
2289 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2293 spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2294 uint64_t max_request, int rewind_flags)
2296 nvlist_t *config = NULL;
2297 int load_error, rewind_error;
2298 uint64_t safe_rewind_txg;
2301 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2302 spa->spa_load_max_txg = spa->spa_load_txg;
2303 spa_set_log_state(spa, SPA_LOG_CLEAR);
2305 spa->spa_load_max_txg = max_request;
2308 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2310 if (load_error == 0)
2313 if (spa->spa_root_vdev != NULL)
2314 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2316 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2317 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2319 if (rewind_flags & ZPOOL_NEVER_REWIND) {
2320 nvlist_free(config);
2321 return (load_error);
2324 /* Price of rolling back is discarding txgs, including log */
2325 if (state == SPA_LOAD_RECOVER)
2326 spa_set_log_state(spa, SPA_LOG_CLEAR);
2328 spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2329 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2330 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2331 TXG_INITIAL : safe_rewind_txg;
2334 * Continue as long as we're finding errors, we're still within
2335 * the acceptable rewind range, and we're still finding uberblocks
2337 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2338 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2339 if (spa->spa_load_max_txg < safe_rewind_txg)
2340 spa->spa_extreme_rewind = B_TRUE;
2341 rewind_error = spa_load_retry(spa, state, mosconfig);
2344 spa->spa_extreme_rewind = B_FALSE;
2345 spa->spa_load_max_txg = UINT64_MAX;
2347 if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2348 spa_config_set(spa, config);
2350 return (state == SPA_LOAD_RECOVER ? rewind_error : load_error);
2356 * The import case is identical to an open except that the configuration is sent
2357 * down from userland, instead of grabbed from the configuration cache. For the
2358 * case of an open, the pool configuration will exist in the
2359 * POOL_STATE_UNINITIALIZED state.
2361 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2362 * the same time open the pool, without having to keep around the spa_t in some
2366 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2370 spa_load_state_t state = SPA_LOAD_OPEN;
2372 int locked = B_FALSE;
2377 * As disgusting as this is, we need to support recursive calls to this
2378 * function because dsl_dir_open() is called during spa_load(), and ends
2379 * up calling spa_open() again. The real fix is to figure out how to
2380 * avoid dsl_dir_open() calling this in the first place.
2382 if (mutex_owner(&spa_namespace_lock) != curthread) {
2383 mutex_enter(&spa_namespace_lock);
2387 if ((spa = spa_lookup(pool)) == NULL) {
2389 mutex_exit(&spa_namespace_lock);
2393 if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2394 zpool_rewind_policy_t policy;
2396 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2398 if (policy.zrp_request & ZPOOL_DO_REWIND)
2399 state = SPA_LOAD_RECOVER;
2401 spa_activate(spa, spa_mode_global);
2403 if (state != SPA_LOAD_RECOVER)
2404 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2406 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2407 policy.zrp_request);
2409 if (error == EBADF) {
2411 * If vdev_validate() returns failure (indicated by
2412 * EBADF), it indicates that one of the vdevs indicates
2413 * that the pool has been exported or destroyed. If
2414 * this is the case, the config cache is out of sync and
2415 * we should remove the pool from the namespace.
2418 spa_deactivate(spa);
2419 spa_config_sync(spa, B_TRUE, B_TRUE);
2422 mutex_exit(&spa_namespace_lock);
2428 * We can't open the pool, but we still have useful
2429 * information: the state of each vdev after the
2430 * attempted vdev_open(). Return this to the user.
2432 if (config != NULL && spa->spa_config) {
2433 VERIFY(nvlist_dup(spa->spa_config, config,
2435 VERIFY(nvlist_add_nvlist(*config,
2436 ZPOOL_CONFIG_LOAD_INFO,
2437 spa->spa_load_info) == 0);
2440 spa_deactivate(spa);
2441 spa->spa_last_open_failed = error;
2443 mutex_exit(&spa_namespace_lock);
2449 spa_open_ref(spa, tag);
2452 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2455 * If we've recovered the pool, pass back any information we
2456 * gathered while doing the load.
2458 if (state == SPA_LOAD_RECOVER) {
2459 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
2460 spa->spa_load_info) == 0);
2464 spa->spa_last_open_failed = 0;
2465 spa->spa_last_ubsync_txg = 0;
2466 spa->spa_load_txg = 0;
2467 mutex_exit(&spa_namespace_lock);
2476 spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
2479 return (spa_open_common(name, spapp, tag, policy, config));
2483 spa_open(const char *name, spa_t **spapp, void *tag)
2485 return (spa_open_common(name, spapp, tag, NULL, NULL));
2489 * Lookup the given spa_t, incrementing the inject count in the process,
2490 * preventing it from being exported or destroyed.
2493 spa_inject_addref(char *name)
2497 mutex_enter(&spa_namespace_lock);
2498 if ((spa = spa_lookup(name)) == NULL) {
2499 mutex_exit(&spa_namespace_lock);
2502 spa->spa_inject_ref++;
2503 mutex_exit(&spa_namespace_lock);
2509 spa_inject_delref(spa_t *spa)
2511 mutex_enter(&spa_namespace_lock);
2512 spa->spa_inject_ref--;
2513 mutex_exit(&spa_namespace_lock);
2517 * Add spares device information to the nvlist.
2520 spa_add_spares(spa_t *spa, nvlist_t *config)
2530 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2532 if (spa->spa_spares.sav_count == 0)
2535 VERIFY(nvlist_lookup_nvlist(config,
2536 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2537 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
2538 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2540 VERIFY(nvlist_add_nvlist_array(nvroot,
2541 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2542 VERIFY(nvlist_lookup_nvlist_array(nvroot,
2543 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2546 * Go through and find any spares which have since been
2547 * repurposed as an active spare. If this is the case, update
2548 * their status appropriately.
2550 for (i = 0; i < nspares; i++) {
2551 VERIFY(nvlist_lookup_uint64(spares[i],
2552 ZPOOL_CONFIG_GUID, &guid) == 0);
2553 if (spa_spare_exists(guid, &pool, NULL) &&
2555 VERIFY(nvlist_lookup_uint64_array(
2556 spares[i], ZPOOL_CONFIG_VDEV_STATS,
2557 (uint64_t **)&vs, &vsc) == 0);
2558 vs->vs_state = VDEV_STATE_CANT_OPEN;
2559 vs->vs_aux = VDEV_AUX_SPARED;
2566 * Add l2cache device information to the nvlist, including vdev stats.
2569 spa_add_l2cache(spa_t *spa, nvlist_t *config)
2572 uint_t i, j, nl2cache;
2579 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2581 if (spa->spa_l2cache.sav_count == 0)
2584 VERIFY(nvlist_lookup_nvlist(config,
2585 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2586 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
2587 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2588 if (nl2cache != 0) {
2589 VERIFY(nvlist_add_nvlist_array(nvroot,
2590 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2591 VERIFY(nvlist_lookup_nvlist_array(nvroot,
2592 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2595 * Update level 2 cache device stats.
2598 for (i = 0; i < nl2cache; i++) {
2599 VERIFY(nvlist_lookup_uint64(l2cache[i],
2600 ZPOOL_CONFIG_GUID, &guid) == 0);
2603 for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
2605 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
2606 vd = spa->spa_l2cache.sav_vdevs[j];
2612 VERIFY(nvlist_lookup_uint64_array(l2cache[i],
2613 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
2615 vdev_get_stats(vd, vs);
2621 spa_get_stats(const char *name, nvlist_t **config, char *altroot, size_t buflen)
2627 error = spa_open_common(name, &spa, FTAG, NULL, config);
2631 * This still leaves a window of inconsistency where the spares
2632 * or l2cache devices could change and the config would be
2633 * self-inconsistent.
2635 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
2637 if (*config != NULL) {
2638 uint64_t loadtimes[2];
2640 loadtimes[0] = spa->spa_loaded_ts.tv_sec;
2641 loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
2642 VERIFY(nvlist_add_uint64_array(*config,
2643 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
2645 VERIFY(nvlist_add_uint64(*config,
2646 ZPOOL_CONFIG_ERRCOUNT,
2647 spa_get_errlog_size(spa)) == 0);
2649 if (spa_suspended(spa))
2650 VERIFY(nvlist_add_uint64(*config,
2651 ZPOOL_CONFIG_SUSPENDED,
2652 spa->spa_failmode) == 0);
2654 spa_add_spares(spa, *config);
2655 spa_add_l2cache(spa, *config);
2660 * We want to get the alternate root even for faulted pools, so we cheat
2661 * and call spa_lookup() directly.
2665 mutex_enter(&spa_namespace_lock);
2666 spa = spa_lookup(name);
2668 spa_altroot(spa, altroot, buflen);
2672 mutex_exit(&spa_namespace_lock);
2674 spa_altroot(spa, altroot, buflen);
2679 spa_config_exit(spa, SCL_CONFIG, FTAG);
2680 spa_close(spa, FTAG);
2687 * Validate that the auxiliary device array is well formed. We must have an
2688 * array of nvlists, each which describes a valid leaf vdev. If this is an
2689 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
2690 * specified, as long as they are well-formed.
2693 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
2694 spa_aux_vdev_t *sav, const char *config, uint64_t version,
2695 vdev_labeltype_t label)
2702 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
2705 * It's acceptable to have no devs specified.
2707 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
2714 * Make sure the pool is formatted with a version that supports this
2717 if (spa_version(spa) < version)
2721 * Set the pending device list so we correctly handle device in-use
2724 sav->sav_pending = dev;
2725 sav->sav_npending = ndev;
2727 for (i = 0; i < ndev; i++) {
2728 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
2732 if (!vd->vdev_ops->vdev_op_leaf) {
2739 * The L2ARC currently only supports disk devices in
2740 * kernel context. For user-level testing, we allow it.
2743 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
2744 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
2751 if ((error = vdev_open(vd)) == 0 &&
2752 (error = vdev_label_init(vd, crtxg, label)) == 0) {
2753 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
2754 vd->vdev_guid) == 0);
2760 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
2767 sav->sav_pending = NULL;
2768 sav->sav_npending = 0;
2773 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
2777 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
2779 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
2780 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
2781 VDEV_LABEL_SPARE)) != 0) {
2785 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
2786 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
2787 VDEV_LABEL_L2CACHE));
2791 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
2796 if (sav->sav_config != NULL) {
2802 * Generate new dev list by concatentating with the
2805 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
2806 &olddevs, &oldndevs) == 0);
2808 newdevs = kmem_alloc(sizeof (void *) *
2809 (ndevs + oldndevs), KM_SLEEP);
2810 for (i = 0; i < oldndevs; i++)
2811 VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
2813 for (i = 0; i < ndevs; i++)
2814 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
2817 VERIFY(nvlist_remove(sav->sav_config, config,
2818 DATA_TYPE_NVLIST_ARRAY) == 0);
2820 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
2821 config, newdevs, ndevs + oldndevs) == 0);
2822 for (i = 0; i < oldndevs + ndevs; i++)
2823 nvlist_free(newdevs[i]);
2824 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
2827 * Generate a new dev list.
2829 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
2831 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
2837 * Stop and drop level 2 ARC devices
2840 spa_l2cache_drop(spa_t *spa)
2844 spa_aux_vdev_t *sav = &spa->spa_l2cache;
2846 for (i = 0; i < sav->sav_count; i++) {
2849 vd = sav->sav_vdevs[i];
2852 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
2853 pool != 0ULL && l2arc_vdev_present(vd))
2854 l2arc_remove_vdev(vd);
2855 if (vd->vdev_isl2cache)
2856 spa_l2cache_remove(vd);
2857 vdev_clear_stats(vd);
2858 (void) vdev_close(vd);
2866 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
2867 const char *history_str, nvlist_t *zplprops)
2870 char *altroot = NULL;
2875 uint64_t txg = TXG_INITIAL;
2876 nvlist_t **spares, **l2cache;
2877 uint_t nspares, nl2cache;
2878 uint64_t version, obj;
2882 * If this pool already exists, return failure.
2884 mutex_enter(&spa_namespace_lock);
2885 if (spa_lookup(pool) != NULL) {
2886 mutex_exit(&spa_namespace_lock);
2891 * Allocate a new spa_t structure.
2893 (void) nvlist_lookup_string(props,
2894 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
2895 spa = spa_add(pool, NULL, altroot);
2896 spa_activate(spa, spa_mode_global);
2898 if (props && (error = spa_prop_validate(spa, props))) {
2899 spa_deactivate(spa);
2901 mutex_exit(&spa_namespace_lock);
2905 if (nvlist_lookup_uint64(props, zpool_prop_to_name(ZPOOL_PROP_VERSION),
2907 version = SPA_VERSION;
2908 ASSERT(version <= SPA_VERSION);
2910 spa->spa_first_txg = txg;
2911 spa->spa_uberblock.ub_txg = txg - 1;
2912 spa->spa_uberblock.ub_version = version;
2913 spa->spa_ubsync = spa->spa_uberblock;
2916 * Create "The Godfather" zio to hold all async IOs
2918 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2919 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2922 * Create the root vdev.
2924 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2926 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
2928 ASSERT(error != 0 || rvd != NULL);
2929 ASSERT(error != 0 || spa->spa_root_vdev == rvd);
2931 if (error == 0 && !zfs_allocatable_devs(nvroot))
2935 (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
2936 (error = spa_validate_aux(spa, nvroot, txg,
2937 VDEV_ALLOC_ADD)) == 0) {
2938 for (c = 0; c < rvd->vdev_children; c++) {
2939 vdev_metaslab_set_size(rvd->vdev_child[c]);
2940 vdev_expand(rvd->vdev_child[c], txg);
2944 spa_config_exit(spa, SCL_ALL, FTAG);
2948 spa_deactivate(spa);
2950 mutex_exit(&spa_namespace_lock);
2955 * Get the list of spares, if specified.
2957 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
2958 &spares, &nspares) == 0) {
2959 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
2961 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
2962 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2963 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2964 spa_load_spares(spa);
2965 spa_config_exit(spa, SCL_ALL, FTAG);
2966 spa->spa_spares.sav_sync = B_TRUE;
2970 * Get the list of level 2 cache devices, if specified.
2972 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
2973 &l2cache, &nl2cache) == 0) {
2974 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
2975 NV_UNIQUE_NAME, KM_SLEEP) == 0);
2976 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
2977 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2978 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2979 spa_load_l2cache(spa);
2980 spa_config_exit(spa, SCL_ALL, FTAG);
2981 spa->spa_l2cache.sav_sync = B_TRUE;
2984 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
2985 spa->spa_meta_objset = dp->dp_meta_objset;
2988 * Create DDTs (dedup tables).
2992 spa_update_dspace(spa);
2994 tx = dmu_tx_create_assigned(dp, txg);
2997 * Create the pool config object.
2999 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3000 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3001 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3003 if (zap_add(spa->spa_meta_objset,
3004 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3005 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3006 cmn_err(CE_PANIC, "failed to add pool config");
3009 if (zap_add(spa->spa_meta_objset,
3010 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3011 sizeof (uint64_t), 1, &version, tx) != 0) {
3012 cmn_err(CE_PANIC, "failed to add pool version");
3015 /* Newly created pools with the right version are always deflated. */
3016 if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3017 spa->spa_deflate = TRUE;
3018 if (zap_add(spa->spa_meta_objset,
3019 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3020 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3021 cmn_err(CE_PANIC, "failed to add deflate");
3026 * Create the deferred-free bpobj. Turn off compression
3027 * because sync-to-convergence takes longer if the blocksize
3030 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3031 dmu_object_set_compress(spa->spa_meta_objset, obj,
3032 ZIO_COMPRESS_OFF, tx);
3033 if (zap_add(spa->spa_meta_objset,
3034 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3035 sizeof (uint64_t), 1, &obj, tx) != 0) {
3036 cmn_err(CE_PANIC, "failed to add bpobj");
3038 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3039 spa->spa_meta_objset, obj));
3042 * Create the pool's history object.
3044 if (version >= SPA_VERSION_ZPOOL_HISTORY)
3045 spa_history_create_obj(spa, tx);
3048 * Set pool properties.
3050 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3051 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3052 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3053 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3055 if (props != NULL) {
3056 spa_configfile_set(spa, props, B_FALSE);
3057 spa_sync_props(spa, props, tx);
3062 spa->spa_sync_on = B_TRUE;
3063 txg_sync_start(spa->spa_dsl_pool);
3066 * We explicitly wait for the first transaction to complete so that our
3067 * bean counters are appropriately updated.
3069 txg_wait_synced(spa->spa_dsl_pool, txg);
3071 spa_config_sync(spa, B_FALSE, B_TRUE);
3073 if (version >= SPA_VERSION_ZPOOL_HISTORY && history_str != NULL)
3074 (void) spa_history_log(spa, history_str, LOG_CMD_POOL_CREATE);
3075 spa_history_log_version(spa, LOG_POOL_CREATE);
3077 spa->spa_minref = refcount_count(&spa->spa_refcount);
3079 mutex_exit(&spa_namespace_lock);
3086 * Get the root pool information from the root disk, then import the root pool
3087 * during the system boot up time.
3089 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3092 spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3095 nvlist_t *nvtop, *nvroot;
3098 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3102 * Add this top-level vdev to the child array.
3104 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3106 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3108 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3111 * Put this pool's top-level vdevs into a root vdev.
3113 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3114 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3115 VDEV_TYPE_ROOT) == 0);
3116 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3117 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3118 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3122 * Replace the existing vdev_tree with the new root vdev in
3123 * this pool's configuration (remove the old, add the new).
3125 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3126 nvlist_free(nvroot);
3131 * Walk the vdev tree and see if we can find a device with "better"
3132 * configuration. A configuration is "better" if the label on that
3133 * device has a more recent txg.
3136 spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3140 for (c = 0; c < vd->vdev_children; c++)
3141 spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3143 if (vd->vdev_ops->vdev_op_leaf) {
3147 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3151 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3155 * Do we have a better boot device?
3157 if (label_txg > *txg) {
3166 * Import a root pool.
3168 * For x86. devpath_list will consist of devid and/or physpath name of
3169 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3170 * The GRUB "findroot" command will return the vdev we should boot.
3172 * For Sparc, devpath_list consists the physpath name of the booting device
3173 * no matter the rootpool is a single device pool or a mirrored pool.
3175 * "/pci@1f,0/ide@d/disk@0,0:a"
3178 spa_import_rootpool(char *devpath, char *devid)
3181 vdev_t *rvd, *bvd, *avd = NULL;
3182 nvlist_t *config, *nvtop;
3188 * Read the label from the boot device and generate a configuration.
3190 config = spa_generate_rootconf(devpath, devid, &guid);
3191 #if defined(_OBP) && defined(_KERNEL)
3192 if (config == NULL) {
3193 if (strstr(devpath, "/iscsi/ssd") != NULL) {
3195 get_iscsi_bootpath_phy(devpath);
3196 config = spa_generate_rootconf(devpath, devid, &guid);
3200 if (config == NULL) {
3201 cmn_err(CE_NOTE, "Can not read the pool label from '%s'",
3206 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3208 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3210 mutex_enter(&spa_namespace_lock);
3211 if ((spa = spa_lookup(pname)) != NULL) {
3213 * Remove the existing root pool from the namespace so that we
3214 * can replace it with the correct config we just read in.
3219 spa = spa_add(pname, config, NULL);
3220 spa->spa_is_root = B_TRUE;
3221 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3224 * Build up a vdev tree based on the boot device's label config.
3226 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3228 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3229 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3230 VDEV_ALLOC_ROOTPOOL);
3231 spa_config_exit(spa, SCL_ALL, FTAG);
3233 mutex_exit(&spa_namespace_lock);
3234 nvlist_free(config);
3235 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3241 * Get the boot vdev.
3243 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3244 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3245 (u_longlong_t)guid);
3251 * Determine if there is a better boot device.
3254 spa_alt_rootvdev(rvd, &avd, &txg);
3256 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3257 "try booting from '%s'", avd->vdev_path);
3263 * If the boot device is part of a spare vdev then ensure that
3264 * we're booting off the active spare.
3266 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3267 !bvd->vdev_isspare) {
3268 cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3269 "try booting from '%s'",
3271 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3277 spa_history_log_version(spa, LOG_POOL_IMPORT);
3279 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3281 spa_config_exit(spa, SCL_ALL, FTAG);
3282 mutex_exit(&spa_namespace_lock);
3284 nvlist_free(config);
3291 * Import a non-root pool into the system.
3294 spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
3297 char *altroot = NULL;
3298 spa_load_state_t state = SPA_LOAD_IMPORT;
3299 zpool_rewind_policy_t policy;
3300 uint64_t mode = spa_mode_global;
3301 uint64_t readonly = B_FALSE;
3304 nvlist_t **spares, **l2cache;
3305 uint_t nspares, nl2cache;
3308 * If a pool with this name exists, return failure.
3310 mutex_enter(&spa_namespace_lock);
3311 if (spa_lookup(pool) != NULL) {
3312 mutex_exit(&spa_namespace_lock);
3317 * Create and initialize the spa structure.
3319 (void) nvlist_lookup_string(props,
3320 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3321 (void) nvlist_lookup_uint64(props,
3322 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
3325 spa = spa_add(pool, config, altroot);
3326 spa->spa_import_flags = flags;
3329 * Verbatim import - Take a pool and insert it into the namespace
3330 * as if it had been loaded at boot.
3332 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
3334 spa_configfile_set(spa, props, B_FALSE);
3336 spa_config_sync(spa, B_FALSE, B_TRUE);
3338 mutex_exit(&spa_namespace_lock);
3339 spa_history_log_version(spa, LOG_POOL_IMPORT);
3344 spa_activate(spa, mode);
3347 * Don't start async tasks until we know everything is healthy.
3349 spa_async_suspend(spa);
3351 zpool_get_rewind_policy(config, &policy);
3352 if (policy.zrp_request & ZPOOL_DO_REWIND)
3353 state = SPA_LOAD_RECOVER;
3356 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
3357 * because the user-supplied config is actually the one to trust when
3360 if (state != SPA_LOAD_RECOVER)
3361 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3363 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
3364 policy.zrp_request);
3367 * Propagate anything learned while loading the pool and pass it
3368 * back to caller (i.e. rewind info, missing devices, etc).
3370 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
3371 spa->spa_load_info) == 0);
3373 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3375 * Toss any existing sparelist, as it doesn't have any validity
3376 * anymore, and conflicts with spa_has_spare().
3378 if (spa->spa_spares.sav_config) {
3379 nvlist_free(spa->spa_spares.sav_config);
3380 spa->spa_spares.sav_config = NULL;
3381 spa_load_spares(spa);
3383 if (spa->spa_l2cache.sav_config) {
3384 nvlist_free(spa->spa_l2cache.sav_config);
3385 spa->spa_l2cache.sav_config = NULL;
3386 spa_load_l2cache(spa);
3389 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3392 error = spa_validate_aux(spa, nvroot, -1ULL,
3395 error = spa_validate_aux(spa, nvroot, -1ULL,
3396 VDEV_ALLOC_L2CACHE);
3397 spa_config_exit(spa, SCL_ALL, FTAG);
3400 spa_configfile_set(spa, props, B_FALSE);
3402 if (error != 0 || (props && spa_writeable(spa) &&
3403 (error = spa_prop_set(spa, props)))) {
3405 spa_deactivate(spa);
3407 mutex_exit(&spa_namespace_lock);
3411 spa_async_resume(spa);
3414 * Override any spares and level 2 cache devices as specified by
3415 * the user, as these may have correct device names/devids, etc.
3417 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3418 &spares, &nspares) == 0) {
3419 if (spa->spa_spares.sav_config)
3420 VERIFY(nvlist_remove(spa->spa_spares.sav_config,
3421 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
3423 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
3424 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3425 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3426 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3427 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3428 spa_load_spares(spa);
3429 spa_config_exit(spa, SCL_ALL, FTAG);
3430 spa->spa_spares.sav_sync = B_TRUE;
3432 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3433 &l2cache, &nl2cache) == 0) {
3434 if (spa->spa_l2cache.sav_config)
3435 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
3436 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
3438 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3439 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3440 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3441 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3442 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3443 spa_load_l2cache(spa);
3444 spa_config_exit(spa, SCL_ALL, FTAG);
3445 spa->spa_l2cache.sav_sync = B_TRUE;
3449 * Check for any removed devices.
3451 if (spa->spa_autoreplace) {
3452 spa_aux_check_removed(&spa->spa_spares);
3453 spa_aux_check_removed(&spa->spa_l2cache);
3456 if (spa_writeable(spa)) {
3458 * Update the config cache to include the newly-imported pool.
3460 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3464 * It's possible that the pool was expanded while it was exported.
3465 * We kick off an async task to handle this for us.
3467 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
3469 mutex_exit(&spa_namespace_lock);
3470 spa_history_log_version(spa, LOG_POOL_IMPORT);
3476 spa_tryimport(nvlist_t *tryconfig)
3478 nvlist_t *config = NULL;
3484 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
3487 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
3491 * Create and initialize the spa structure.
3493 mutex_enter(&spa_namespace_lock);
3494 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
3495 spa_activate(spa, FREAD);
3498 * Pass off the heavy lifting to spa_load().
3499 * Pass TRUE for mosconfig because the user-supplied config
3500 * is actually the one to trust when doing an import.
3502 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
3505 * If 'tryconfig' was at least parsable, return the current config.
3507 if (spa->spa_root_vdev != NULL) {
3508 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
3509 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
3511 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
3513 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
3514 spa->spa_uberblock.ub_timestamp) == 0);
3517 * If the bootfs property exists on this pool then we
3518 * copy it out so that external consumers can tell which
3519 * pools are bootable.
3521 if ((!error || error == EEXIST) && spa->spa_bootfs) {
3522 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3525 * We have to play games with the name since the
3526 * pool was opened as TRYIMPORT_NAME.
3528 if (dsl_dsobj_to_dsname(spa_name(spa),
3529 spa->spa_bootfs, tmpname) == 0) {
3531 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3533 cp = strchr(tmpname, '/');
3535 (void) strlcpy(dsname, tmpname,
3538 (void) snprintf(dsname, MAXPATHLEN,
3539 "%s/%s", poolname, ++cp);
3541 VERIFY(nvlist_add_string(config,
3542 ZPOOL_CONFIG_BOOTFS, dsname) == 0);
3543 kmem_free(dsname, MAXPATHLEN);
3545 kmem_free(tmpname, MAXPATHLEN);
3549 * Add the list of hot spares and level 2 cache devices.
3551 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3552 spa_add_spares(spa, config);
3553 spa_add_l2cache(spa, config);
3554 spa_config_exit(spa, SCL_CONFIG, FTAG);
3558 spa_deactivate(spa);
3560 mutex_exit(&spa_namespace_lock);
3566 * Pool export/destroy
3568 * The act of destroying or exporting a pool is very simple. We make sure there
3569 * is no more pending I/O and any references to the pool are gone. Then, we
3570 * update the pool state and sync all the labels to disk, removing the
3571 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
3572 * we don't sync the labels or remove the configuration cache.
3575 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
3576 boolean_t force, boolean_t hardforce)
3583 if (!(spa_mode_global & FWRITE))
3586 mutex_enter(&spa_namespace_lock);
3587 if ((spa = spa_lookup(pool)) == NULL) {
3588 mutex_exit(&spa_namespace_lock);
3593 * Put a hold on the pool, drop the namespace lock, stop async tasks,
3594 * reacquire the namespace lock, and see if we can export.
3596 spa_open_ref(spa, FTAG);
3597 mutex_exit(&spa_namespace_lock);
3598 spa_async_suspend(spa);
3599 mutex_enter(&spa_namespace_lock);
3600 spa_close(spa, FTAG);
3603 * The pool will be in core if it's openable,
3604 * in which case we can modify its state.
3606 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
3608 * Objsets may be open only because they're dirty, so we
3609 * have to force it to sync before checking spa_refcnt.
3611 txg_wait_synced(spa->spa_dsl_pool, 0);
3614 * A pool cannot be exported or destroyed if there are active
3615 * references. If we are resetting a pool, allow references by
3616 * fault injection handlers.
3618 if (!spa_refcount_zero(spa) ||
3619 (spa->spa_inject_ref != 0 &&
3620 new_state != POOL_STATE_UNINITIALIZED)) {
3621 spa_async_resume(spa);
3622 mutex_exit(&spa_namespace_lock);
3627 * A pool cannot be exported if it has an active shared spare.
3628 * This is to prevent other pools stealing the active spare
3629 * from an exported pool. At user's own will, such pool can
3630 * be forcedly exported.
3632 if (!force && new_state == POOL_STATE_EXPORTED &&
3633 spa_has_active_shared_spare(spa)) {
3634 spa_async_resume(spa);
3635 mutex_exit(&spa_namespace_lock);
3640 * We want this to be reflected on every label,
3641 * so mark them all dirty. spa_unload() will do the
3642 * final sync that pushes these changes out.
3644 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
3645 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3646 spa->spa_state = new_state;
3647 spa->spa_final_txg = spa_last_synced_txg(spa) +
3649 vdev_config_dirty(spa->spa_root_vdev);
3650 spa_config_exit(spa, SCL_ALL, FTAG);
3654 spa_event_notify(spa, NULL, FM_EREPORT_ZFS_POOL_DESTROY);
3656 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
3658 spa_deactivate(spa);
3661 if (oldconfig && spa->spa_config)
3662 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
3664 if (new_state != POOL_STATE_UNINITIALIZED) {
3666 spa_config_sync(spa, B_TRUE, B_TRUE);
3669 mutex_exit(&spa_namespace_lock);
3675 * Destroy a storage pool.
3678 spa_destroy(char *pool)
3680 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
3685 * Export a storage pool.
3688 spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
3689 boolean_t hardforce)
3691 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
3696 * Similar to spa_export(), this unloads the spa_t without actually removing it
3697 * from the namespace in any way.
3700 spa_reset(char *pool)
3702 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
3707 * ==========================================================================
3708 * Device manipulation
3709 * ==========================================================================
3713 * Add a device to a storage pool.
3716 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
3720 vdev_t *rvd = spa->spa_root_vdev;
3722 nvlist_t **spares, **l2cache;
3723 uint_t nspares, nl2cache;
3726 ASSERT(spa_writeable(spa));
3728 txg = spa_vdev_enter(spa);
3730 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
3731 VDEV_ALLOC_ADD)) != 0)
3732 return (spa_vdev_exit(spa, NULL, txg, error));
3734 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */
3736 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
3740 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
3744 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
3745 return (spa_vdev_exit(spa, vd, txg, EINVAL));
3747 if (vd->vdev_children != 0 &&
3748 (error = vdev_create(vd, txg, B_FALSE)) != 0)
3749 return (spa_vdev_exit(spa, vd, txg, error));
3752 * We must validate the spares and l2cache devices after checking the
3753 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
3755 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
3756 return (spa_vdev_exit(spa, vd, txg, error));
3759 * Transfer each new top-level vdev from vd to rvd.
3761 for (c = 0; c < vd->vdev_children; c++) {
3764 * Set the vdev id to the first hole, if one exists.
3766 for (id = 0; id < rvd->vdev_children; id++) {
3767 if (rvd->vdev_child[id]->vdev_ishole) {
3768 vdev_free(rvd->vdev_child[id]);
3772 tvd = vd->vdev_child[c];
3773 vdev_remove_child(vd, tvd);
3775 vdev_add_child(rvd, tvd);
3776 vdev_config_dirty(tvd);
3780 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
3781 ZPOOL_CONFIG_SPARES);
3782 spa_load_spares(spa);
3783 spa->spa_spares.sav_sync = B_TRUE;
3786 if (nl2cache != 0) {
3787 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
3788 ZPOOL_CONFIG_L2CACHE);
3789 spa_load_l2cache(spa);
3790 spa->spa_l2cache.sav_sync = B_TRUE;
3794 * We have to be careful when adding new vdevs to an existing pool.
3795 * If other threads start allocating from these vdevs before we
3796 * sync the config cache, and we lose power, then upon reboot we may
3797 * fail to open the pool because there are DVAs that the config cache
3798 * can't translate. Therefore, we first add the vdevs without
3799 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
3800 * and then let spa_config_update() initialize the new metaslabs.
3802 * spa_load() checks for added-but-not-initialized vdevs, so that
3803 * if we lose power at any point in this sequence, the remaining
3804 * steps will be completed the next time we load the pool.
3806 (void) spa_vdev_exit(spa, vd, txg, 0);
3808 mutex_enter(&spa_namespace_lock);
3809 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3810 mutex_exit(&spa_namespace_lock);
3816 * Attach a device to a mirror. The arguments are the path to any device
3817 * in the mirror, and the nvroot for the new device. If the path specifies
3818 * a device that is not mirrored, we automatically insert the mirror vdev.
3820 * If 'replacing' is specified, the new device is intended to replace the
3821 * existing device; in this case the two devices are made into their own
3822 * mirror using the 'replacing' vdev, which is functionally identical to
3823 * the mirror vdev (it actually reuses all the same ops) but has a few
3824 * extra rules: you can't attach to it after it's been created, and upon
3825 * completion of resilvering, the first disk (the one being replaced)
3826 * is automatically detached.
3829 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
3831 uint64_t txg, dtl_max_txg;
3832 ASSERTV(vdev_t *rvd = spa->spa_root_vdev;)
3833 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
3835 char *oldvdpath, *newvdpath;
3839 ASSERT(spa_writeable(spa));
3841 txg = spa_vdev_enter(spa);
3843 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
3846 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
3848 if (!oldvd->vdev_ops->vdev_op_leaf)
3849 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3851 pvd = oldvd->vdev_parent;
3853 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
3854 VDEV_ALLOC_ADD)) != 0)
3855 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
3857 if (newrootvd->vdev_children != 1)
3858 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
3860 newvd = newrootvd->vdev_child[0];
3862 if (!newvd->vdev_ops->vdev_op_leaf)
3863 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
3865 if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
3866 return (spa_vdev_exit(spa, newrootvd, txg, error));
3869 * Spares can't replace logs
3871 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
3872 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3876 * For attach, the only allowable parent is a mirror or the root
3879 if (pvd->vdev_ops != &vdev_mirror_ops &&
3880 pvd->vdev_ops != &vdev_root_ops)
3881 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3883 pvops = &vdev_mirror_ops;
3886 * Active hot spares can only be replaced by inactive hot
3889 if (pvd->vdev_ops == &vdev_spare_ops &&
3890 oldvd->vdev_isspare &&
3891 !spa_has_spare(spa, newvd->vdev_guid))
3892 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3895 * If the source is a hot spare, and the parent isn't already a
3896 * spare, then we want to create a new hot spare. Otherwise, we
3897 * want to create a replacing vdev. The user is not allowed to
3898 * attach to a spared vdev child unless the 'isspare' state is
3899 * the same (spare replaces spare, non-spare replaces
3902 if (pvd->vdev_ops == &vdev_replacing_ops &&
3903 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
3904 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3905 } else if (pvd->vdev_ops == &vdev_spare_ops &&
3906 newvd->vdev_isspare != oldvd->vdev_isspare) {
3907 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3910 if (newvd->vdev_isspare)
3911 pvops = &vdev_spare_ops;
3913 pvops = &vdev_replacing_ops;
3917 * Make sure the new device is big enough.
3919 if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
3920 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
3923 * The new device cannot have a higher alignment requirement
3924 * than the top-level vdev.
3926 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
3927 return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
3930 * If this is an in-place replacement, update oldvd's path and devid
3931 * to make it distinguishable from newvd, and unopenable from now on.
3933 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
3934 spa_strfree(oldvd->vdev_path);
3935 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
3937 (void) sprintf(oldvd->vdev_path, "%s/%s",
3938 newvd->vdev_path, "old");
3939 if (oldvd->vdev_devid != NULL) {
3940 spa_strfree(oldvd->vdev_devid);
3941 oldvd->vdev_devid = NULL;
3945 /* mark the device being resilvered */
3946 newvd->vdev_resilvering = B_TRUE;
3949 * If the parent is not a mirror, or if we're replacing, insert the new
3950 * mirror/replacing/spare vdev above oldvd.
3952 if (pvd->vdev_ops != pvops)
3953 pvd = vdev_add_parent(oldvd, pvops);
3955 ASSERT(pvd->vdev_top->vdev_parent == rvd);
3956 ASSERT(pvd->vdev_ops == pvops);
3957 ASSERT(oldvd->vdev_parent == pvd);
3960 * Extract the new device from its root and add it to pvd.
3962 vdev_remove_child(newrootvd, newvd);
3963 newvd->vdev_id = pvd->vdev_children;
3964 newvd->vdev_crtxg = oldvd->vdev_crtxg;
3965 vdev_add_child(pvd, newvd);
3967 tvd = newvd->vdev_top;
3968 ASSERT(pvd->vdev_top == tvd);
3969 ASSERT(tvd->vdev_parent == rvd);
3971 vdev_config_dirty(tvd);
3974 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
3975 * for any dmu_sync-ed blocks. It will propagate upward when
3976 * spa_vdev_exit() calls vdev_dtl_reassess().
3978 dtl_max_txg = txg + TXG_CONCURRENT_STATES;
3980 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
3981 dtl_max_txg - TXG_INITIAL);
3983 if (newvd->vdev_isspare) {
3984 spa_spare_activate(newvd);
3985 spa_event_notify(spa, newvd, FM_EREPORT_ZFS_DEVICE_SPARE);
3988 oldvdpath = spa_strdup(oldvd->vdev_path);
3989 newvdpath = spa_strdup(newvd->vdev_path);
3990 newvd_isspare = newvd->vdev_isspare;
3993 * Mark newvd's DTL dirty in this txg.
3995 vdev_dirty(tvd, VDD_DTL, newvd, txg);
3998 * Restart the resilver
4000 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4005 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4007 spa_history_log_internal(LOG_POOL_VDEV_ATTACH, spa, NULL,
4008 "%s vdev=%s %s vdev=%s",
4009 replacing && newvd_isspare ? "spare in" :
4010 replacing ? "replace" : "attach", newvdpath,
4011 replacing ? "for" : "to", oldvdpath);
4013 spa_strfree(oldvdpath);
4014 spa_strfree(newvdpath);
4016 if (spa->spa_bootfs)
4017 spa_event_notify(spa, newvd, FM_EREPORT_ZFS_BOOTFS_VDEV_ATTACH);
4023 * Detach a device from a mirror or replacing vdev.
4024 * If 'replace_done' is specified, only detach if the parent
4025 * is a replacing vdev.
4028 spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4032 ASSERTV(vdev_t *rvd = spa->spa_root_vdev;)
4033 vdev_t *vd, *pvd, *cvd, *tvd;
4034 boolean_t unspare = B_FALSE;
4035 uint64_t unspare_guid = 0;
4039 ASSERT(spa_writeable(spa));
4041 txg = spa_vdev_enter(spa);
4043 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4046 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4048 if (!vd->vdev_ops->vdev_op_leaf)
4049 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4051 pvd = vd->vdev_parent;
4054 * If the parent/child relationship is not as expected, don't do it.
4055 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4056 * vdev that's replacing B with C. The user's intent in replacing
4057 * is to go from M(A,B) to M(A,C). If the user decides to cancel
4058 * the replace by detaching C, the expected behavior is to end up
4059 * M(A,B). But suppose that right after deciding to detach C,
4060 * the replacement of B completes. We would have M(A,C), and then
4061 * ask to detach C, which would leave us with just A -- not what
4062 * the user wanted. To prevent this, we make sure that the
4063 * parent/child relationship hasn't changed -- in this example,
4064 * that C's parent is still the replacing vdev R.
4066 if (pvd->vdev_guid != pguid && pguid != 0)
4067 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4070 * Only 'replacing' or 'spare' vdevs can be replaced.
4072 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4073 pvd->vdev_ops != &vdev_spare_ops)
4074 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4076 ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4077 spa_version(spa) >= SPA_VERSION_SPARES);
4080 * Only mirror, replacing, and spare vdevs support detach.
4082 if (pvd->vdev_ops != &vdev_replacing_ops &&
4083 pvd->vdev_ops != &vdev_mirror_ops &&
4084 pvd->vdev_ops != &vdev_spare_ops)
4085 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4088 * If this device has the only valid copy of some data,
4089 * we cannot safely detach it.
4091 if (vdev_dtl_required(vd))
4092 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4094 ASSERT(pvd->vdev_children >= 2);
4097 * If we are detaching the second disk from a replacing vdev, then
4098 * check to see if we changed the original vdev's path to have "/old"
4099 * at the end in spa_vdev_attach(). If so, undo that change now.
4101 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4102 vd->vdev_path != NULL) {
4103 size_t len = strlen(vd->vdev_path);
4105 for (c = 0; c < pvd->vdev_children; c++) {
4106 cvd = pvd->vdev_child[c];
4108 if (cvd == vd || cvd->vdev_path == NULL)
4111 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4112 strcmp(cvd->vdev_path + len, "/old") == 0) {
4113 spa_strfree(cvd->vdev_path);
4114 cvd->vdev_path = spa_strdup(vd->vdev_path);
4121 * If we are detaching the original disk from a spare, then it implies
4122 * that the spare should become a real disk, and be removed from the
4123 * active spare list for the pool.
4125 if (pvd->vdev_ops == &vdev_spare_ops &&
4127 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4131 * Erase the disk labels so the disk can be used for other things.
4132 * This must be done after all other error cases are handled,
4133 * but before we disembowel vd (so we can still do I/O to it).
4134 * But if we can't do it, don't treat the error as fatal --
4135 * it may be that the unwritability of the disk is the reason
4136 * it's being detached!
4138 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4141 * Remove vd from its parent and compact the parent's children.
4143 vdev_remove_child(pvd, vd);
4144 vdev_compact_children(pvd);
4147 * Remember one of the remaining children so we can get tvd below.
4149 cvd = pvd->vdev_child[pvd->vdev_children - 1];
4152 * If we need to remove the remaining child from the list of hot spares,
4153 * do it now, marking the vdev as no longer a spare in the process.
4154 * We must do this before vdev_remove_parent(), because that can
4155 * change the GUID if it creates a new toplevel GUID. For a similar
4156 * reason, we must remove the spare now, in the same txg as the detach;
4157 * otherwise someone could attach a new sibling, change the GUID, and
4158 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4161 ASSERT(cvd->vdev_isspare);
4162 spa_spare_remove(cvd);
4163 unspare_guid = cvd->vdev_guid;
4164 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4165 cvd->vdev_unspare = B_TRUE;
4169 * If the parent mirror/replacing vdev only has one child,
4170 * the parent is no longer needed. Remove it from the tree.
4172 if (pvd->vdev_children == 1) {
4173 if (pvd->vdev_ops == &vdev_spare_ops)
4174 cvd->vdev_unspare = B_FALSE;
4175 vdev_remove_parent(cvd);
4176 cvd->vdev_resilvering = B_FALSE;
4181 * We don't set tvd until now because the parent we just removed
4182 * may have been the previous top-level vdev.
4184 tvd = cvd->vdev_top;
4185 ASSERT(tvd->vdev_parent == rvd);
4188 * Reevaluate the parent vdev state.
4190 vdev_propagate_state(cvd);
4193 * If the 'autoexpand' property is set on the pool then automatically
4194 * try to expand the size of the pool. For example if the device we
4195 * just detached was smaller than the others, it may be possible to
4196 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4197 * first so that we can obtain the updated sizes of the leaf vdevs.
4199 if (spa->spa_autoexpand) {
4201 vdev_expand(tvd, txg);
4204 vdev_config_dirty(tvd);
4207 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
4208 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4209 * But first make sure we're not on any *other* txg's DTL list, to
4210 * prevent vd from being accessed after it's freed.
4212 vdpath = spa_strdup(vd->vdev_path);
4213 for (t = 0; t < TXG_SIZE; t++)
4214 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4215 vd->vdev_detached = B_TRUE;
4216 vdev_dirty(tvd, VDD_DTL, vd, txg);
4218 spa_event_notify(spa, vd, FM_EREPORT_ZFS_DEVICE_REMOVE);
4220 /* hang on to the spa before we release the lock */
4221 spa_open_ref(spa, FTAG);
4223 error = spa_vdev_exit(spa, vd, txg, 0);
4225 spa_history_log_internal(LOG_POOL_VDEV_DETACH, spa, NULL,
4227 spa_strfree(vdpath);
4230 * If this was the removal of the original device in a hot spare vdev,
4231 * then we want to go through and remove the device from the hot spare
4232 * list of every other pool.
4235 spa_t *altspa = NULL;
4237 mutex_enter(&spa_namespace_lock);
4238 while ((altspa = spa_next(altspa)) != NULL) {
4239 if (altspa->spa_state != POOL_STATE_ACTIVE ||
4243 spa_open_ref(altspa, FTAG);
4244 mutex_exit(&spa_namespace_lock);
4245 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
4246 mutex_enter(&spa_namespace_lock);
4247 spa_close(altspa, FTAG);
4249 mutex_exit(&spa_namespace_lock);
4251 /* search the rest of the vdevs for spares to remove */
4252 spa_vdev_resilver_done(spa);
4255 /* all done with the spa; OK to release */
4256 mutex_enter(&spa_namespace_lock);
4257 spa_close(spa, FTAG);
4258 mutex_exit(&spa_namespace_lock);
4264 * Split a set of devices from their mirrors, and create a new pool from them.
4267 spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
4268 nvlist_t *props, boolean_t exp)
4271 uint64_t txg, *glist;
4273 uint_t c, children, lastlog;
4274 nvlist_t **child, *nvl, *tmp;
4276 char *altroot = NULL;
4277 vdev_t *rvd, **vml = NULL; /* vdev modify list */
4278 boolean_t activate_slog;
4280 ASSERT(spa_writeable(spa));
4282 txg = spa_vdev_enter(spa);
4284 /* clear the log and flush everything up to now */
4285 activate_slog = spa_passivate_log(spa);
4286 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4287 error = spa_offline_log(spa);
4288 txg = spa_vdev_config_enter(spa);
4291 spa_activate_log(spa);
4294 return (spa_vdev_exit(spa, NULL, txg, error));
4296 /* check new spa name before going any further */
4297 if (spa_lookup(newname) != NULL)
4298 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
4301 * scan through all the children to ensure they're all mirrors
4303 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
4304 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
4306 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4308 /* first, check to ensure we've got the right child count */
4309 rvd = spa->spa_root_vdev;
4311 for (c = 0; c < rvd->vdev_children; c++) {
4312 vdev_t *vd = rvd->vdev_child[c];
4314 /* don't count the holes & logs as children */
4315 if (vd->vdev_islog || vd->vdev_ishole) {
4323 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
4324 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4326 /* next, ensure no spare or cache devices are part of the split */
4327 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
4328 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
4329 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4331 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
4332 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
4334 /* then, loop over each vdev and validate it */
4335 for (c = 0; c < children; c++) {
4336 uint64_t is_hole = 0;
4338 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
4342 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
4343 spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
4351 /* which disk is going to be split? */
4352 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
4358 /* look it up in the spa */
4359 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
4360 if (vml[c] == NULL) {
4365 /* make sure there's nothing stopping the split */
4366 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
4367 vml[c]->vdev_islog ||
4368 vml[c]->vdev_ishole ||
4369 vml[c]->vdev_isspare ||
4370 vml[c]->vdev_isl2cache ||
4371 !vdev_writeable(vml[c]) ||
4372 vml[c]->vdev_children != 0 ||
4373 vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
4374 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
4379 if (vdev_dtl_required(vml[c])) {
4384 /* we need certain info from the top level */
4385 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
4386 vml[c]->vdev_top->vdev_ms_array) == 0);
4387 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
4388 vml[c]->vdev_top->vdev_ms_shift) == 0);
4389 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
4390 vml[c]->vdev_top->vdev_asize) == 0);
4391 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
4392 vml[c]->vdev_top->vdev_ashift) == 0);
4396 kmem_free(vml, children * sizeof (vdev_t *));
4397 kmem_free(glist, children * sizeof (uint64_t));
4398 return (spa_vdev_exit(spa, NULL, txg, error));
4401 /* stop writers from using the disks */
4402 for (c = 0; c < children; c++) {
4404 vml[c]->vdev_offline = B_TRUE;
4406 vdev_reopen(spa->spa_root_vdev);
4409 * Temporarily record the splitting vdevs in the spa config. This
4410 * will disappear once the config is regenerated.
4412 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4413 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
4414 glist, children) == 0);
4415 kmem_free(glist, children * sizeof (uint64_t));
4417 mutex_enter(&spa->spa_props_lock);
4418 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
4420 mutex_exit(&spa->spa_props_lock);
4421 spa->spa_config_splitting = nvl;
4422 vdev_config_dirty(spa->spa_root_vdev);
4424 /* configure and create the new pool */
4425 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
4426 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4427 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
4428 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
4429 spa_version(spa)) == 0);
4430 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
4431 spa->spa_config_txg) == 0);
4432 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4433 spa_generate_guid(NULL)) == 0);
4434 (void) nvlist_lookup_string(props,
4435 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4437 /* add the new pool to the namespace */
4438 newspa = spa_add(newname, config, altroot);
4439 newspa->spa_config_txg = spa->spa_config_txg;
4440 spa_set_log_state(newspa, SPA_LOG_CLEAR);
4442 /* release the spa config lock, retaining the namespace lock */
4443 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4445 if (zio_injection_enabled)
4446 zio_handle_panic_injection(spa, FTAG, 1);
4448 spa_activate(newspa, spa_mode_global);
4449 spa_async_suspend(newspa);
4451 /* create the new pool from the disks of the original pool */
4452 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
4456 /* if that worked, generate a real config for the new pool */
4457 if (newspa->spa_root_vdev != NULL) {
4458 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
4459 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4460 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
4461 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
4462 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
4467 if (props != NULL) {
4468 spa_configfile_set(newspa, props, B_FALSE);
4469 error = spa_prop_set(newspa, props);
4474 /* flush everything */
4475 txg = spa_vdev_config_enter(newspa);
4476 vdev_config_dirty(newspa->spa_root_vdev);
4477 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
4479 if (zio_injection_enabled)
4480 zio_handle_panic_injection(spa, FTAG, 2);
4482 spa_async_resume(newspa);
4484 /* finally, update the original pool's config */
4485 txg = spa_vdev_config_enter(spa);
4486 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
4487 error = dmu_tx_assign(tx, TXG_WAIT);
4490 for (c = 0; c < children; c++) {
4491 if (vml[c] != NULL) {
4494 spa_history_log_internal(LOG_POOL_VDEV_DETACH,
4500 vdev_config_dirty(spa->spa_root_vdev);
4501 spa->spa_config_splitting = NULL;
4505 (void) spa_vdev_exit(spa, NULL, txg, 0);
4507 if (zio_injection_enabled)
4508 zio_handle_panic_injection(spa, FTAG, 3);
4510 /* split is complete; log a history record */
4511 spa_history_log_internal(LOG_POOL_SPLIT, newspa, NULL,
4512 "split new pool %s from pool %s", newname, spa_name(spa));
4514 kmem_free(vml, children * sizeof (vdev_t *));
4516 /* if we're not going to mount the filesystems in userland, export */
4518 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
4525 spa_deactivate(newspa);
4528 txg = spa_vdev_config_enter(spa);
4530 /* re-online all offlined disks */
4531 for (c = 0; c < children; c++) {
4533 vml[c]->vdev_offline = B_FALSE;
4535 vdev_reopen(spa->spa_root_vdev);
4537 nvlist_free(spa->spa_config_splitting);
4538 spa->spa_config_splitting = NULL;
4539 (void) spa_vdev_exit(spa, NULL, txg, error);
4541 kmem_free(vml, children * sizeof (vdev_t *));
4546 spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
4550 for (i = 0; i < count; i++) {
4553 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
4556 if (guid == target_guid)
4564 spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
4565 nvlist_t *dev_to_remove)
4567 nvlist_t **newdev = NULL;
4571 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
4573 for (i = 0, j = 0; i < count; i++) {
4574 if (dev[i] == dev_to_remove)
4576 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
4579 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
4580 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
4582 for (i = 0; i < count - 1; i++)
4583 nvlist_free(newdev[i]);
4586 kmem_free(newdev, (count - 1) * sizeof (void *));
4590 * Evacuate the device.
4593 spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
4598 ASSERT(MUTEX_HELD(&spa_namespace_lock));
4599 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
4600 ASSERT(vd == vd->vdev_top);
4603 * Evacuate the device. We don't hold the config lock as writer
4604 * since we need to do I/O but we do keep the
4605 * spa_namespace_lock held. Once this completes the device
4606 * should no longer have any blocks allocated on it.
4608 if (vd->vdev_islog) {
4609 if (vd->vdev_stat.vs_alloc != 0)
4610 error = spa_offline_log(spa);
4619 * The evacuation succeeded. Remove any remaining MOS metadata
4620 * associated with this vdev, and wait for these changes to sync.
4622 ASSERT3U(vd->vdev_stat.vs_alloc, ==, 0);
4623 txg = spa_vdev_config_enter(spa);
4624 vd->vdev_removing = B_TRUE;
4625 vdev_dirty(vd, 0, NULL, txg);
4626 vdev_config_dirty(vd);
4627 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4633 * Complete the removal by cleaning up the namespace.
4636 spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
4638 vdev_t *rvd = spa->spa_root_vdev;
4639 uint64_t id = vd->vdev_id;
4640 boolean_t last_vdev = (id == (rvd->vdev_children - 1));
4642 ASSERT(MUTEX_HELD(&spa_namespace_lock));
4643 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
4644 ASSERT(vd == vd->vdev_top);
4647 * Only remove any devices which are empty.
4649 if (vd->vdev_stat.vs_alloc != 0)
4652 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4654 if (list_link_active(&vd->vdev_state_dirty_node))
4655 vdev_state_clean(vd);
4656 if (list_link_active(&vd->vdev_config_dirty_node))
4657 vdev_config_clean(vd);
4662 vdev_compact_children(rvd);
4664 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
4665 vdev_add_child(rvd, vd);
4667 vdev_config_dirty(rvd);
4670 * Reassess the health of our root vdev.
4676 * Remove a device from the pool -
4678 * Removing a device from the vdev namespace requires several steps
4679 * and can take a significant amount of time. As a result we use
4680 * the spa_vdev_config_[enter/exit] functions which allow us to
4681 * grab and release the spa_config_lock while still holding the namespace
4682 * lock. During each step the configuration is synced out.
4686 * Remove a device from the pool. Currently, this supports removing only hot
4687 * spares, slogs, and level 2 ARC devices.
4690 spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
4693 metaslab_group_t *mg;
4694 nvlist_t **spares, **l2cache, *nv;
4696 uint_t nspares, nl2cache;
4698 boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
4700 ASSERT(spa_writeable(spa));
4703 txg = spa_vdev_enter(spa);
4705 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4707 if (spa->spa_spares.sav_vdevs != NULL &&
4708 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
4709 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
4710 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
4712 * Only remove the hot spare if it's not currently in use
4715 if (vd == NULL || unspare) {
4716 spa_vdev_remove_aux(spa->spa_spares.sav_config,
4717 ZPOOL_CONFIG_SPARES, spares, nspares, nv);
4718 spa_load_spares(spa);
4719 spa->spa_spares.sav_sync = B_TRUE;
4723 } else if (spa->spa_l2cache.sav_vdevs != NULL &&
4724 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
4725 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
4726 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
4728 * Cache devices can always be removed.
4730 spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
4731 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
4732 spa_load_l2cache(spa);
4733 spa->spa_l2cache.sav_sync = B_TRUE;
4734 } else if (vd != NULL && vd->vdev_islog) {
4736 ASSERT(vd == vd->vdev_top);
4739 * XXX - Once we have bp-rewrite this should
4740 * become the common case.
4746 * Stop allocating from this vdev.
4748 metaslab_group_passivate(mg);
4751 * Wait for the youngest allocations and frees to sync,
4752 * and then wait for the deferral of those frees to finish.
4754 spa_vdev_config_exit(spa, NULL,
4755 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
4758 * Attempt to evacuate the vdev.
4760 error = spa_vdev_remove_evacuate(spa, vd);
4762 txg = spa_vdev_config_enter(spa);
4765 * If we couldn't evacuate the vdev, unwind.
4768 metaslab_group_activate(mg);
4769 return (spa_vdev_exit(spa, NULL, txg, error));
4773 * Clean up the vdev namespace.
4775 spa_vdev_remove_from_namespace(spa, vd);
4777 } else if (vd != NULL) {
4779 * Normal vdevs cannot be removed (yet).
4784 * There is no vdev of any kind with the specified guid.
4790 return (spa_vdev_exit(spa, NULL, txg, error));
4796 * Find any device that's done replacing, or a vdev marked 'unspare' that's
4797 * current spared, so we can detach it.
4800 spa_vdev_resilver_done_hunt(vdev_t *vd)
4802 vdev_t *newvd, *oldvd;
4805 for (c = 0; c < vd->vdev_children; c++) {
4806 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
4812 * Check for a completed replacement. We always consider the first
4813 * vdev in the list to be the oldest vdev, and the last one to be
4814 * the newest (see spa_vdev_attach() for how that works). In
4815 * the case where the newest vdev is faulted, we will not automatically
4816 * remove it after a resilver completes. This is OK as it will require
4817 * user intervention to determine which disk the admin wishes to keep.
4819 if (vd->vdev_ops == &vdev_replacing_ops) {
4820 ASSERT(vd->vdev_children > 1);
4822 newvd = vd->vdev_child[vd->vdev_children - 1];
4823 oldvd = vd->vdev_child[0];
4825 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
4826 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
4827 !vdev_dtl_required(oldvd))
4832 * Check for a completed resilver with the 'unspare' flag set.
4834 if (vd->vdev_ops == &vdev_spare_ops) {
4835 vdev_t *first = vd->vdev_child[0];
4836 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
4838 if (last->vdev_unspare) {
4841 } else if (first->vdev_unspare) {
4848 if (oldvd != NULL &&
4849 vdev_dtl_empty(newvd, DTL_MISSING) &&
4850 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
4851 !vdev_dtl_required(oldvd))
4855 * If there are more than two spares attached to a disk,
4856 * and those spares are not required, then we want to
4857 * attempt to free them up now so that they can be used
4858 * by other pools. Once we're back down to a single
4859 * disk+spare, we stop removing them.
4861 if (vd->vdev_children > 2) {
4862 newvd = vd->vdev_child[1];
4864 if (newvd->vdev_isspare && last->vdev_isspare &&
4865 vdev_dtl_empty(last, DTL_MISSING) &&
4866 vdev_dtl_empty(last, DTL_OUTAGE) &&
4867 !vdev_dtl_required(newvd))
4876 spa_vdev_resilver_done(spa_t *spa)
4878 vdev_t *vd, *pvd, *ppvd;
4879 uint64_t guid, sguid, pguid, ppguid;
4881 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4883 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
4884 pvd = vd->vdev_parent;
4885 ppvd = pvd->vdev_parent;
4886 guid = vd->vdev_guid;
4887 pguid = pvd->vdev_guid;
4888 ppguid = ppvd->vdev_guid;
4891 * If we have just finished replacing a hot spared device, then
4892 * we need to detach the parent's first child (the original hot
4895 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
4896 ppvd->vdev_children == 2) {
4897 ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
4898 sguid = ppvd->vdev_child[1]->vdev_guid;
4900 spa_config_exit(spa, SCL_ALL, FTAG);
4901 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
4903 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
4905 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4908 spa_config_exit(spa, SCL_ALL, FTAG);
4912 * Update the stored path or FRU for this vdev.
4915 spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
4919 boolean_t sync = B_FALSE;
4921 ASSERT(spa_writeable(spa));
4923 spa_vdev_state_enter(spa, SCL_ALL);
4925 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
4926 return (spa_vdev_state_exit(spa, NULL, ENOENT));
4928 if (!vd->vdev_ops->vdev_op_leaf)
4929 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
4932 if (strcmp(value, vd->vdev_path) != 0) {
4933 spa_strfree(vd->vdev_path);
4934 vd->vdev_path = spa_strdup(value);
4938 if (vd->vdev_fru == NULL) {
4939 vd->vdev_fru = spa_strdup(value);
4941 } else if (strcmp(value, vd->vdev_fru) != 0) {
4942 spa_strfree(vd->vdev_fru);
4943 vd->vdev_fru = spa_strdup(value);
4948 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
4952 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
4954 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
4958 spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
4960 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
4964 * ==========================================================================
4966 * ==========================================================================
4970 spa_scan_stop(spa_t *spa)
4972 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
4973 if (dsl_scan_resilvering(spa->spa_dsl_pool))
4975 return (dsl_scan_cancel(spa->spa_dsl_pool));
4979 spa_scan(spa_t *spa, pool_scan_func_t func)
4981 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
4983 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
4987 * If a resilver was requested, but there is no DTL on a
4988 * writeable leaf device, we have nothing to do.
4990 if (func == POOL_SCAN_RESILVER &&
4991 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
4992 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
4996 return (dsl_scan(spa->spa_dsl_pool, func));
5000 * ==========================================================================
5001 * SPA async task processing
5002 * ==========================================================================
5006 spa_async_remove(spa_t *spa, vdev_t *vd)
5010 if (vd->vdev_remove_wanted) {
5011 vd->vdev_remove_wanted = B_FALSE;
5012 vd->vdev_delayed_close = B_FALSE;
5013 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5016 * We want to clear the stats, but we don't want to do a full
5017 * vdev_clear() as that will cause us to throw away
5018 * degraded/faulted state as well as attempt to reopen the
5019 * device, all of which is a waste.
5021 vd->vdev_stat.vs_read_errors = 0;
5022 vd->vdev_stat.vs_write_errors = 0;
5023 vd->vdev_stat.vs_checksum_errors = 0;
5025 vdev_state_dirty(vd->vdev_top);
5028 for (c = 0; c < vd->vdev_children; c++)
5029 spa_async_remove(spa, vd->vdev_child[c]);
5033 spa_async_probe(spa_t *spa, vdev_t *vd)
5037 if (vd->vdev_probe_wanted) {
5038 vd->vdev_probe_wanted = B_FALSE;
5039 vdev_reopen(vd); /* vdev_open() does the actual probe */
5042 for (c = 0; c < vd->vdev_children; c++)
5043 spa_async_probe(spa, vd->vdev_child[c]);
5047 spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5051 if (!spa->spa_autoexpand)
5054 for (c = 0; c < vd->vdev_children; c++) {
5055 vdev_t *cvd = vd->vdev_child[c];
5056 spa_async_autoexpand(spa, cvd);
5059 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5062 spa_event_notify(vd->vdev_spa, vd, FM_EREPORT_ZFS_DEVICE_AUTOEXPAND);
5066 spa_async_thread(spa_t *spa)
5070 ASSERT(spa->spa_sync_on);
5072 mutex_enter(&spa->spa_async_lock);
5073 tasks = spa->spa_async_tasks;
5074 spa->spa_async_tasks = 0;
5075 mutex_exit(&spa->spa_async_lock);
5078 * See if the config needs to be updated.
5080 if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5081 uint64_t old_space, new_space;
5083 mutex_enter(&spa_namespace_lock);
5084 old_space = metaslab_class_get_space(spa_normal_class(spa));
5085 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5086 new_space = metaslab_class_get_space(spa_normal_class(spa));
5087 mutex_exit(&spa_namespace_lock);
5090 * If the pool grew as a result of the config update,
5091 * then log an internal history event.
5093 if (new_space != old_space) {
5094 spa_history_log_internal(LOG_POOL_VDEV_ONLINE,
5096 "pool '%s' size: %llu(+%llu)",
5097 spa_name(spa), new_space, new_space - old_space);
5102 * See if any devices need to be marked REMOVED.
5104 if (tasks & SPA_ASYNC_REMOVE) {
5105 spa_vdev_state_enter(spa, SCL_NONE);
5106 spa_async_remove(spa, spa->spa_root_vdev);
5107 for (i = 0; i < spa->spa_l2cache.sav_count; i++)
5108 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5109 for (i = 0; i < spa->spa_spares.sav_count; i++)
5110 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5111 (void) spa_vdev_state_exit(spa, NULL, 0);
5114 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5115 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5116 spa_async_autoexpand(spa, spa->spa_root_vdev);
5117 spa_config_exit(spa, SCL_CONFIG, FTAG);
5121 * See if any devices need to be probed.
5123 if (tasks & SPA_ASYNC_PROBE) {
5124 spa_vdev_state_enter(spa, SCL_NONE);
5125 spa_async_probe(spa, spa->spa_root_vdev);
5126 (void) spa_vdev_state_exit(spa, NULL, 0);
5130 * If any devices are done replacing, detach them.
5132 if (tasks & SPA_ASYNC_RESILVER_DONE)
5133 spa_vdev_resilver_done(spa);
5136 * Kick off a resilver.
5138 if (tasks & SPA_ASYNC_RESILVER)
5139 dsl_resilver_restart(spa->spa_dsl_pool, 0);
5142 * Let the world know that we're done.
5144 mutex_enter(&spa->spa_async_lock);
5145 spa->spa_async_thread = NULL;
5146 cv_broadcast(&spa->spa_async_cv);
5147 mutex_exit(&spa->spa_async_lock);
5152 spa_async_suspend(spa_t *spa)
5154 mutex_enter(&spa->spa_async_lock);
5155 spa->spa_async_suspended++;
5156 while (spa->spa_async_thread != NULL)
5157 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5158 mutex_exit(&spa->spa_async_lock);
5162 spa_async_resume(spa_t *spa)
5164 mutex_enter(&spa->spa_async_lock);
5165 ASSERT(spa->spa_async_suspended != 0);
5166 spa->spa_async_suspended--;
5167 mutex_exit(&spa->spa_async_lock);
5171 spa_async_dispatch(spa_t *spa)
5173 mutex_enter(&spa->spa_async_lock);
5174 if (spa->spa_async_tasks && !spa->spa_async_suspended &&
5175 spa->spa_async_thread == NULL &&
5176 rootdir != NULL && !vn_is_readonly(rootdir))
5177 spa->spa_async_thread = thread_create(NULL, 0,
5178 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
5179 mutex_exit(&spa->spa_async_lock);
5183 spa_async_request(spa_t *spa, int task)
5185 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
5186 mutex_enter(&spa->spa_async_lock);
5187 spa->spa_async_tasks |= task;
5188 mutex_exit(&spa->spa_async_lock);
5192 * ==========================================================================
5193 * SPA syncing routines
5194 * ==========================================================================
5198 bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5201 bpobj_enqueue(bpo, bp, tx);
5206 spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5210 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
5216 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
5218 char *packed = NULL;
5223 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
5226 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5227 * information. This avoids the dbuf_will_dirty() path and
5228 * saves us a pre-read to get data we don't actually care about.
5230 bufsize = P2ROUNDUP(nvsize, SPA_CONFIG_BLOCKSIZE);
5231 packed = vmem_alloc(bufsize, KM_SLEEP);
5233 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
5235 bzero(packed + nvsize, bufsize - nvsize);
5237 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
5239 vmem_free(packed, bufsize);
5241 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
5242 dmu_buf_will_dirty(db, tx);
5243 *(uint64_t *)db->db_data = nvsize;
5244 dmu_buf_rele(db, FTAG);
5248 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
5249 const char *config, const char *entry)
5259 * Update the MOS nvlist describing the list of available devices.
5260 * spa_validate_aux() will have already made sure this nvlist is
5261 * valid and the vdevs are labeled appropriately.
5263 if (sav->sav_object == 0) {
5264 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
5265 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
5266 sizeof (uint64_t), tx);
5267 VERIFY(zap_update(spa->spa_meta_objset,
5268 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
5269 &sav->sav_object, tx) == 0);
5272 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5273 if (sav->sav_count == 0) {
5274 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
5276 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
5277 for (i = 0; i < sav->sav_count; i++)
5278 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
5279 B_FALSE, VDEV_CONFIG_L2CACHE);
5280 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
5281 sav->sav_count) == 0);
5282 for (i = 0; i < sav->sav_count; i++)
5283 nvlist_free(list[i]);
5284 kmem_free(list, sav->sav_count * sizeof (void *));
5287 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
5288 nvlist_free(nvroot);
5290 sav->sav_sync = B_FALSE;
5294 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
5298 if (list_is_empty(&spa->spa_config_dirty_list))
5301 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5303 config = spa_config_generate(spa, spa->spa_root_vdev,
5304 dmu_tx_get_txg(tx), B_FALSE);
5306 spa_config_exit(spa, SCL_STATE, FTAG);
5308 if (spa->spa_config_syncing)
5309 nvlist_free(spa->spa_config_syncing);
5310 spa->spa_config_syncing = config;
5312 spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
5316 * Set zpool properties.
5319 spa_sync_props(void *arg1, void *arg2, dmu_tx_t *tx)
5322 objset_t *mos = spa->spa_meta_objset;
5323 nvlist_t *nvp = arg2;
5328 const char *propname;
5329 zprop_type_t proptype;
5331 mutex_enter(&spa->spa_props_lock);
5334 while ((elem = nvlist_next_nvpair(nvp, elem))) {
5335 switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
5336 case ZPOOL_PROP_VERSION:
5338 * Only set version for non-zpool-creation cases
5339 * (set/import). spa_create() needs special care
5340 * for version setting.
5342 if (tx->tx_txg != TXG_INITIAL) {
5343 VERIFY(nvpair_value_uint64(elem,
5345 ASSERT(intval <= SPA_VERSION);
5346 ASSERT(intval >= spa_version(spa));
5347 spa->spa_uberblock.ub_version = intval;
5348 vdev_config_dirty(spa->spa_root_vdev);
5352 case ZPOOL_PROP_ALTROOT:
5354 * 'altroot' is a non-persistent property. It should
5355 * have been set temporarily at creation or import time.
5357 ASSERT(spa->spa_root != NULL);
5360 case ZPOOL_PROP_READONLY:
5361 case ZPOOL_PROP_CACHEFILE:
5363 * 'readonly' and 'cachefile' are also non-persisitent
5369 * Set pool property values in the poolprops mos object.
5371 if (spa->spa_pool_props_object == 0) {
5372 VERIFY((spa->spa_pool_props_object =
5373 zap_create(mos, DMU_OT_POOL_PROPS,
5374 DMU_OT_NONE, 0, tx)) > 0);
5376 VERIFY(zap_update(mos,
5377 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
5378 8, 1, &spa->spa_pool_props_object, tx)
5382 /* normalize the property name */
5383 propname = zpool_prop_to_name(prop);
5384 proptype = zpool_prop_get_type(prop);
5386 if (nvpair_type(elem) == DATA_TYPE_STRING) {
5387 ASSERT(proptype == PROP_TYPE_STRING);
5388 VERIFY(nvpair_value_string(elem, &strval) == 0);
5389 VERIFY(zap_update(mos,
5390 spa->spa_pool_props_object, propname,
5391 1, strlen(strval) + 1, strval, tx) == 0);
5393 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
5394 VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5396 if (proptype == PROP_TYPE_INDEX) {
5398 VERIFY(zpool_prop_index_to_string(
5399 prop, intval, &unused) == 0);
5401 VERIFY(zap_update(mos,
5402 spa->spa_pool_props_object, propname,
5403 8, 1, &intval, tx) == 0);
5405 ASSERT(0); /* not allowed */
5409 case ZPOOL_PROP_DELEGATION:
5410 spa->spa_delegation = intval;
5412 case ZPOOL_PROP_BOOTFS:
5413 spa->spa_bootfs = intval;
5415 case ZPOOL_PROP_FAILUREMODE:
5416 spa->spa_failmode = intval;
5418 case ZPOOL_PROP_AUTOEXPAND:
5419 spa->spa_autoexpand = intval;
5420 if (tx->tx_txg != TXG_INITIAL)
5421 spa_async_request(spa,
5422 SPA_ASYNC_AUTOEXPAND);
5424 case ZPOOL_PROP_DEDUPDITTO:
5425 spa->spa_dedup_ditto = intval;
5432 /* log internal history if this is not a zpool create */
5433 if (spa_version(spa) >= SPA_VERSION_ZPOOL_HISTORY &&
5434 tx->tx_txg != TXG_INITIAL) {
5435 spa_history_log_internal(LOG_POOL_PROPSET,
5436 spa, tx, "%s %lld %s",
5437 nvpair_name(elem), intval, spa_name(spa));
5441 mutex_exit(&spa->spa_props_lock);
5445 * Perform one-time upgrade on-disk changes. spa_version() does not
5446 * reflect the new version this txg, so there must be no changes this
5447 * txg to anything that the upgrade code depends on after it executes.
5448 * Therefore this must be called after dsl_pool_sync() does the sync
5452 spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
5454 dsl_pool_t *dp = spa->spa_dsl_pool;
5456 ASSERT(spa->spa_sync_pass == 1);
5458 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
5459 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
5460 dsl_pool_create_origin(dp, tx);
5462 /* Keeping the origin open increases spa_minref */
5463 spa->spa_minref += 3;
5466 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
5467 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
5468 dsl_pool_upgrade_clones(dp, tx);
5471 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
5472 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
5473 dsl_pool_upgrade_dir_clones(dp, tx);
5475 /* Keeping the freedir open increases spa_minref */
5476 spa->spa_minref += 3;
5481 * Sync the specified transaction group. New blocks may be dirtied as
5482 * part of the process, so we iterate until it converges.
5485 spa_sync(spa_t *spa, uint64_t txg)
5487 dsl_pool_t *dp = spa->spa_dsl_pool;
5488 objset_t *mos = spa->spa_meta_objset;
5489 bpobj_t *defer_bpo = &spa->spa_deferred_bpobj;
5490 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
5491 vdev_t *rvd = spa->spa_root_vdev;
5497 VERIFY(spa_writeable(spa));
5500 * Lock out configuration changes.
5502 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5504 spa->spa_syncing_txg = txg;
5505 spa->spa_sync_pass = 0;
5508 * If there are any pending vdev state changes, convert them
5509 * into config changes that go out with this transaction group.
5511 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5512 while (list_head(&spa->spa_state_dirty_list) != NULL) {
5514 * We need the write lock here because, for aux vdevs,
5515 * calling vdev_config_dirty() modifies sav_config.
5516 * This is ugly and will become unnecessary when we
5517 * eliminate the aux vdev wart by integrating all vdevs
5518 * into the root vdev tree.
5520 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
5521 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
5522 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
5523 vdev_state_clean(vd);
5524 vdev_config_dirty(vd);
5526 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
5527 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
5529 spa_config_exit(spa, SCL_STATE, FTAG);
5531 tx = dmu_tx_create_assigned(dp, txg);
5534 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
5535 * set spa_deflate if we have no raid-z vdevs.
5537 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
5538 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
5541 for (i = 0; i < rvd->vdev_children; i++) {
5542 vd = rvd->vdev_child[i];
5543 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
5546 if (i == rvd->vdev_children) {
5547 spa->spa_deflate = TRUE;
5548 VERIFY(0 == zap_add(spa->spa_meta_objset,
5549 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
5550 sizeof (uint64_t), 1, &spa->spa_deflate, tx));
5555 * If anything has changed in this txg, or if someone is waiting
5556 * for this txg to sync (eg, spa_vdev_remove()), push the
5557 * deferred frees from the previous txg. If not, leave them
5558 * alone so that we don't generate work on an otherwise idle
5561 if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
5562 !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
5563 !txg_list_empty(&dp->dp_sync_tasks, txg) ||
5564 ((dsl_scan_active(dp->dp_scan) ||
5565 txg_sync_waiting(dp)) && !spa_shutting_down(spa))) {
5566 zio_t *zio = zio_root(spa, NULL, NULL, 0);
5567 VERIFY3U(bpobj_iterate(defer_bpo,
5568 spa_free_sync_cb, zio, tx), ==, 0);
5569 VERIFY3U(zio_wait(zio), ==, 0);
5573 * Iterate to convergence.
5576 int pass = ++spa->spa_sync_pass;
5578 spa_sync_config_object(spa, tx);
5579 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
5580 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
5581 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
5582 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
5583 spa_errlog_sync(spa, txg);
5584 dsl_pool_sync(dp, txg);
5586 if (pass <= SYNC_PASS_DEFERRED_FREE) {
5587 zio_t *zio = zio_root(spa, NULL, NULL, 0);
5588 bplist_iterate(free_bpl, spa_free_sync_cb,
5590 VERIFY(zio_wait(zio) == 0);
5592 bplist_iterate(free_bpl, bpobj_enqueue_cb,
5597 dsl_scan_sync(dp, tx);
5599 while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)))
5603 spa_sync_upgrades(spa, tx);
5605 } while (dmu_objset_is_dirty(mos, txg));
5608 * Rewrite the vdev configuration (which includes the uberblock)
5609 * to commit the transaction group.
5611 * If there are no dirty vdevs, we sync the uberblock to a few
5612 * random top-level vdevs that are known to be visible in the
5613 * config cache (see spa_vdev_add() for a complete description).
5614 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
5618 * We hold SCL_STATE to prevent vdev open/close/etc.
5619 * while we're attempting to write the vdev labels.
5621 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5623 if (list_is_empty(&spa->spa_config_dirty_list)) {
5624 vdev_t *svd[SPA_DVAS_PER_BP];
5626 int children = rvd->vdev_children;
5627 int c0 = spa_get_random(children);
5629 for (c = 0; c < children; c++) {
5630 vd = rvd->vdev_child[(c0 + c) % children];
5631 if (vd->vdev_ms_array == 0 || vd->vdev_islog)
5633 svd[svdcount++] = vd;
5634 if (svdcount == SPA_DVAS_PER_BP)
5637 error = vdev_config_sync(svd, svdcount, txg, B_FALSE);
5639 error = vdev_config_sync(svd, svdcount, txg,
5642 error = vdev_config_sync(rvd->vdev_child,
5643 rvd->vdev_children, txg, B_FALSE);
5645 error = vdev_config_sync(rvd->vdev_child,
5646 rvd->vdev_children, txg, B_TRUE);
5649 spa_config_exit(spa, SCL_STATE, FTAG);
5653 zio_suspend(spa, NULL);
5654 zio_resume_wait(spa);
5659 * Clear the dirty config list.
5661 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
5662 vdev_config_clean(vd);
5665 * Now that the new config has synced transactionally,
5666 * let it become visible to the config cache.
5668 if (spa->spa_config_syncing != NULL) {
5669 spa_config_set(spa, spa->spa_config_syncing);
5670 spa->spa_config_txg = txg;
5671 spa->spa_config_syncing = NULL;
5674 spa->spa_ubsync = spa->spa_uberblock;
5676 dsl_pool_sync_done(dp, txg);
5679 * Update usable space statistics.
5681 while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg))))
5682 vdev_sync_done(vd, txg);
5684 spa_update_dspace(spa);
5687 * It had better be the case that we didn't dirty anything
5688 * since vdev_config_sync().
5690 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
5691 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
5692 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
5694 spa->spa_sync_pass = 0;
5696 spa_config_exit(spa, SCL_CONFIG, FTAG);
5698 spa_handle_ignored_writes(spa);
5701 * If any async tasks have been requested, kick them off.
5703 spa_async_dispatch(spa);
5707 * Sync all pools. We don't want to hold the namespace lock across these
5708 * operations, so we take a reference on the spa_t and drop the lock during the
5712 spa_sync_allpools(void)
5715 mutex_enter(&spa_namespace_lock);
5716 while ((spa = spa_next(spa)) != NULL) {
5717 if (spa_state(spa) != POOL_STATE_ACTIVE ||
5718 !spa_writeable(spa) || spa_suspended(spa))
5720 spa_open_ref(spa, FTAG);
5721 mutex_exit(&spa_namespace_lock);
5722 txg_wait_synced(spa_get_dsl(spa), 0);
5723 mutex_enter(&spa_namespace_lock);
5724 spa_close(spa, FTAG);
5726 mutex_exit(&spa_namespace_lock);
5730 * ==========================================================================
5731 * Miscellaneous routines
5732 * ==========================================================================
5736 * Remove all pools in the system.
5744 * Remove all cached state. All pools should be closed now,
5745 * so every spa in the AVL tree should be unreferenced.
5747 mutex_enter(&spa_namespace_lock);
5748 while ((spa = spa_next(NULL)) != NULL) {
5750 * Stop async tasks. The async thread may need to detach
5751 * a device that's been replaced, which requires grabbing
5752 * spa_namespace_lock, so we must drop it here.
5754 spa_open_ref(spa, FTAG);
5755 mutex_exit(&spa_namespace_lock);
5756 spa_async_suspend(spa);
5757 mutex_enter(&spa_namespace_lock);
5758 spa_close(spa, FTAG);
5760 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
5762 spa_deactivate(spa);
5766 mutex_exit(&spa_namespace_lock);
5770 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
5775 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
5779 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
5780 vd = spa->spa_l2cache.sav_vdevs[i];
5781 if (vd->vdev_guid == guid)
5785 for (i = 0; i < spa->spa_spares.sav_count; i++) {
5786 vd = spa->spa_spares.sav_vdevs[i];
5787 if (vd->vdev_guid == guid)
5796 spa_upgrade(spa_t *spa, uint64_t version)
5798 ASSERT(spa_writeable(spa));
5800 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5803 * This should only be called for a non-faulted pool, and since a
5804 * future version would result in an unopenable pool, this shouldn't be
5807 ASSERT(spa->spa_uberblock.ub_version <= SPA_VERSION);
5808 ASSERT(version >= spa->spa_uberblock.ub_version);
5810 spa->spa_uberblock.ub_version = version;
5811 vdev_config_dirty(spa->spa_root_vdev);
5813 spa_config_exit(spa, SCL_ALL, FTAG);
5815 txg_wait_synced(spa_get_dsl(spa), 0);
5819 spa_has_spare(spa_t *spa, uint64_t guid)
5823 spa_aux_vdev_t *sav = &spa->spa_spares;
5825 for (i = 0; i < sav->sav_count; i++)
5826 if (sav->sav_vdevs[i]->vdev_guid == guid)
5829 for (i = 0; i < sav->sav_npending; i++) {
5830 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
5831 &spareguid) == 0 && spareguid == guid)
5839 * Check if a pool has an active shared spare device.
5840 * Note: reference count of an active spare is 2, as a spare and as a replace
5843 spa_has_active_shared_spare(spa_t *spa)
5847 spa_aux_vdev_t *sav = &spa->spa_spares;
5849 for (i = 0; i < sav->sav_count; i++) {
5850 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
5851 &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
5860 * Post a FM_EREPORT_ZFS_* event from sys/fm/fs/zfs.h. The payload will be
5861 * filled in from the spa and (optionally) the vdev. This doesn't do anything
5862 * in the userland libzpool, as we don't want consumers to misinterpret ztest
5863 * or zdb as real changes.
5866 spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
5869 zfs_ereport_post(name, spa, vd, NULL, 0, 0);
5873 #if defined(_KERNEL) && defined(HAVE_SPL)
5874 /* state manipulation functions */
5875 EXPORT_SYMBOL(spa_open);
5876 EXPORT_SYMBOL(spa_open_rewind);
5877 EXPORT_SYMBOL(spa_get_stats);
5878 EXPORT_SYMBOL(spa_create);
5879 EXPORT_SYMBOL(spa_import_rootpool);
5880 EXPORT_SYMBOL(spa_import);
5881 EXPORT_SYMBOL(spa_tryimport);
5882 EXPORT_SYMBOL(spa_destroy);
5883 EXPORT_SYMBOL(spa_export);
5884 EXPORT_SYMBOL(spa_reset);
5885 EXPORT_SYMBOL(spa_async_request);
5886 EXPORT_SYMBOL(spa_async_suspend);
5887 EXPORT_SYMBOL(spa_async_resume);
5888 EXPORT_SYMBOL(spa_inject_addref);
5889 EXPORT_SYMBOL(spa_inject_delref);
5890 EXPORT_SYMBOL(spa_scan_stat_init);
5891 EXPORT_SYMBOL(spa_scan_get_stats);
5893 /* device maniion */
5894 EXPORT_SYMBOL(spa_vdev_add);
5895 EXPORT_SYMBOL(spa_vdev_attach);
5896 EXPORT_SYMBOL(spa_vdev_detach);
5897 EXPORT_SYMBOL(spa_vdev_remove);
5898 EXPORT_SYMBOL(spa_vdev_setpath);
5899 EXPORT_SYMBOL(spa_vdev_setfru);
5900 EXPORT_SYMBOL(spa_vdev_split_mirror);
5902 /* spare statech is global across all pools) */
5903 EXPORT_SYMBOL(spa_spare_add);
5904 EXPORT_SYMBOL(spa_spare_remove);
5905 EXPORT_SYMBOL(spa_spare_exists);
5906 EXPORT_SYMBOL(spa_spare_activate);
5908 /* L2ARC statech is global across all pools) */
5909 EXPORT_SYMBOL(spa_l2cache_add);
5910 EXPORT_SYMBOL(spa_l2cache_remove);
5911 EXPORT_SYMBOL(spa_l2cache_exists);
5912 EXPORT_SYMBOL(spa_l2cache_activate);
5913 EXPORT_SYMBOL(spa_l2cache_drop);
5916 EXPORT_SYMBOL(spa_scan);
5917 EXPORT_SYMBOL(spa_scan_stop);
5920 EXPORT_SYMBOL(spa_sync); /* only for DMU use */
5921 EXPORT_SYMBOL(spa_sync_allpools);
5924 EXPORT_SYMBOL(spa_prop_set);
5925 EXPORT_SYMBOL(spa_prop_get);
5926 EXPORT_SYMBOL(spa_prop_clear_bootfs);
5928 /* asynchronous event notification */
5929 EXPORT_SYMBOL(spa_event_notify);