4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
25 * Copyright (c) 2011 by Delphix. All rights reserved.
29 * This file contains all the routines used when modifying on-disk SPA state.
30 * This includes opening, importing, destroying, exporting a pool, and syncing a
34 #include <sys/zfs_context.h>
35 #include <sys/fm/fs/zfs.h>
36 #include <sys/spa_impl.h>
38 #include <sys/zio_checksum.h>
40 #include <sys/dmu_tx.h>
44 #include <sys/vdev_impl.h>
45 #include <sys/vdev_disk.h>
46 #include <sys/metaslab.h>
47 #include <sys/metaslab_impl.h>
48 #include <sys/uberblock_impl.h>
51 #include <sys/dmu_traverse.h>
52 #include <sys/dmu_objset.h>
53 #include <sys/unique.h>
54 #include <sys/dsl_pool.h>
55 #include <sys/dsl_dataset.h>
56 #include <sys/dsl_dir.h>
57 #include <sys/dsl_prop.h>
58 #include <sys/dsl_synctask.h>
59 #include <sys/fs/zfs.h>
61 #include <sys/callb.h>
62 #include <sys/systeminfo.h>
63 #include <sys/spa_boot.h>
64 #include <sys/zfs_ioctl.h>
65 #include <sys/dsl_scan.h>
68 #include <sys/bootprops.h>
69 #include <sys/callb.h>
70 #include <sys/cpupart.h>
72 #include <sys/sysdc.h>
77 #include "zfs_comutil.h"
79 typedef enum zti_modes {
80 zti_mode_fixed, /* value is # of threads (min 1) */
81 zti_mode_online_percent, /* value is % of online CPUs */
82 zti_mode_batch, /* cpu-intensive; value is ignored */
83 zti_mode_null, /* don't create a taskq */
87 #define ZTI_FIX(n) { zti_mode_fixed, (n) }
88 #define ZTI_PCT(n) { zti_mode_online_percent, (n) }
89 #define ZTI_BATCH { zti_mode_batch, 0 }
90 #define ZTI_NULL { zti_mode_null, 0 }
92 #define ZTI_ONE ZTI_FIX(1)
94 typedef struct zio_taskq_info {
95 enum zti_modes zti_mode;
99 static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
100 "iss", "iss_h", "int", "int_h"
104 * Define the taskq threads for the following I/O types:
105 * NULL, READ, WRITE, FREE, CLAIM, and IOCTL
107 const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
108 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
109 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL },
110 { ZTI_FIX(8), ZTI_NULL, ZTI_BATCH, ZTI_NULL },
111 { ZTI_BATCH, ZTI_FIX(5), ZTI_FIX(16), ZTI_FIX(5) },
112 { ZTI_PCT(100), ZTI_NULL, ZTI_ONE, ZTI_NULL },
113 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL },
114 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL },
117 static dsl_syncfunc_t spa_sync_props;
118 static boolean_t spa_has_active_shared_spare(spa_t *spa);
119 static inline int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
120 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
122 static void spa_vdev_resilver_done(spa_t *spa);
124 uint_t zio_taskq_batch_pct = 100; /* 1 thread per cpu in pset */
125 id_t zio_taskq_psrset_bind = PS_NONE;
126 boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */
127 uint_t zio_taskq_basedc = 80; /* base duty cycle */
129 boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */
132 * This (illegal) pool name is used when temporarily importing a spa_t in order
133 * to get the vdev stats associated with the imported devices.
135 #define TRYIMPORT_NAME "$import"
138 * ==========================================================================
139 * SPA properties routines
140 * ==========================================================================
144 * Add a (source=src, propname=propval) list to an nvlist.
147 spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
148 uint64_t intval, zprop_source_t src)
150 const char *propname = zpool_prop_to_name(prop);
153 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
154 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
157 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
159 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
161 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
162 nvlist_free(propval);
166 * Get property values from the spa configuration.
169 spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
173 uint64_t cap, version;
174 zprop_source_t src = ZPROP_SRC_NONE;
175 spa_config_dirent_t *dp;
177 ASSERT(MUTEX_HELD(&spa->spa_props_lock));
179 if (spa->spa_root_vdev != NULL) {
180 alloc = metaslab_class_get_alloc(spa_normal_class(spa));
181 size = metaslab_class_get_space(spa_normal_class(spa));
182 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
183 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
184 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
185 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
187 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
188 (spa_mode(spa) == FREAD), src);
190 cap = (size == 0) ? 0 : (alloc * 100 / size);
191 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
193 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
194 ddt_get_pool_dedup_ratio(spa), src);
196 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
197 spa->spa_root_vdev->vdev_state, src);
199 version = spa_version(spa);
200 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
201 src = ZPROP_SRC_DEFAULT;
203 src = ZPROP_SRC_LOCAL;
204 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
207 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
209 if (spa->spa_comment != NULL) {
210 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
214 if (spa->spa_root != NULL)
215 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
218 if ((dp = list_head(&spa->spa_config_list)) != NULL) {
219 if (dp->scd_path == NULL) {
220 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
221 "none", 0, ZPROP_SRC_LOCAL);
222 } else if (strcmp(dp->scd_path, spa_config_path) != 0) {
223 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
224 dp->scd_path, 0, ZPROP_SRC_LOCAL);
230 * Get zpool property values.
233 spa_prop_get(spa_t *spa, nvlist_t **nvp)
235 objset_t *mos = spa->spa_meta_objset;
240 err = nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP);
244 mutex_enter(&spa->spa_props_lock);
247 * Get properties from the spa config.
249 spa_prop_get_config(spa, nvp);
251 /* If no pool property object, no more prop to get. */
252 if (mos == NULL || spa->spa_pool_props_object == 0) {
253 mutex_exit(&spa->spa_props_lock);
258 * Get properties from the MOS pool property object.
260 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
261 (err = zap_cursor_retrieve(&zc, &za)) == 0;
262 zap_cursor_advance(&zc)) {
265 zprop_source_t src = ZPROP_SRC_DEFAULT;
268 if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
271 switch (za.za_integer_length) {
273 /* integer property */
274 if (za.za_first_integer !=
275 zpool_prop_default_numeric(prop))
276 src = ZPROP_SRC_LOCAL;
278 if (prop == ZPOOL_PROP_BOOTFS) {
280 dsl_dataset_t *ds = NULL;
282 dp = spa_get_dsl(spa);
283 rw_enter(&dp->dp_config_rwlock, RW_READER);
284 if ((err = dsl_dataset_hold_obj(dp,
285 za.za_first_integer, FTAG, &ds))) {
286 rw_exit(&dp->dp_config_rwlock);
291 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
293 dsl_dataset_name(ds, strval);
294 dsl_dataset_rele(ds, FTAG);
295 rw_exit(&dp->dp_config_rwlock);
298 intval = za.za_first_integer;
301 spa_prop_add_list(*nvp, prop, strval, intval, src);
305 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
310 /* string property */
311 strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
312 err = zap_lookup(mos, spa->spa_pool_props_object,
313 za.za_name, 1, za.za_num_integers, strval);
315 kmem_free(strval, za.za_num_integers);
318 spa_prop_add_list(*nvp, prop, strval, 0, src);
319 kmem_free(strval, za.za_num_integers);
326 zap_cursor_fini(&zc);
327 mutex_exit(&spa->spa_props_lock);
329 if (err && err != ENOENT) {
339 * Validate the given pool properties nvlist and modify the list
340 * for the property values to be set.
343 spa_prop_validate(spa_t *spa, nvlist_t *props)
346 int error = 0, reset_bootfs = 0;
350 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
352 char *propname, *strval;
357 propname = nvpair_name(elem);
359 if ((prop = zpool_name_to_prop(propname)) == ZPROP_INVAL)
363 case ZPOOL_PROP_VERSION:
364 error = nvpair_value_uint64(elem, &intval);
366 (intval < spa_version(spa) || intval > SPA_VERSION))
370 case ZPOOL_PROP_DELEGATION:
371 case ZPOOL_PROP_AUTOREPLACE:
372 case ZPOOL_PROP_LISTSNAPS:
373 case ZPOOL_PROP_AUTOEXPAND:
374 error = nvpair_value_uint64(elem, &intval);
375 if (!error && intval > 1)
379 case ZPOOL_PROP_BOOTFS:
381 * If the pool version is less than SPA_VERSION_BOOTFS,
382 * or the pool is still being created (version == 0),
383 * the bootfs property cannot be set.
385 if (spa_version(spa) < SPA_VERSION_BOOTFS) {
391 * Make sure the vdev config is bootable
393 if (!vdev_is_bootable(spa->spa_root_vdev)) {
400 error = nvpair_value_string(elem, &strval);
405 if (strval == NULL || strval[0] == '\0') {
406 objnum = zpool_prop_default_numeric(
411 if ((error = dmu_objset_hold(strval,FTAG,&os)))
414 /* Must be ZPL and not gzip compressed. */
416 if (dmu_objset_type(os) != DMU_OST_ZFS) {
418 } else if ((error = dsl_prop_get_integer(strval,
419 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
420 &compress, NULL)) == 0 &&
421 !BOOTFS_COMPRESS_VALID(compress)) {
424 objnum = dmu_objset_id(os);
426 dmu_objset_rele(os, FTAG);
430 case ZPOOL_PROP_FAILUREMODE:
431 error = nvpair_value_uint64(elem, &intval);
432 if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
433 intval > ZIO_FAILURE_MODE_PANIC))
437 * This is a special case which only occurs when
438 * the pool has completely failed. This allows
439 * the user to change the in-core failmode property
440 * without syncing it out to disk (I/Os might
441 * currently be blocked). We do this by returning
442 * EIO to the caller (spa_prop_set) to trick it
443 * into thinking we encountered a property validation
446 if (!error && spa_suspended(spa)) {
447 spa->spa_failmode = intval;
452 case ZPOOL_PROP_CACHEFILE:
453 if ((error = nvpair_value_string(elem, &strval)) != 0)
456 if (strval[0] == '\0')
459 if (strcmp(strval, "none") == 0)
462 if (strval[0] != '/') {
467 slash = strrchr(strval, '/');
468 ASSERT(slash != NULL);
470 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
471 strcmp(slash, "/..") == 0)
475 case ZPOOL_PROP_COMMENT:
476 if ((error = nvpair_value_string(elem, &strval)) != 0)
478 for (check = strval; *check != '\0'; check++) {
479 if (!isprint(*check)) {
485 if (strlen(strval) > ZPROP_MAX_COMMENT)
489 case ZPOOL_PROP_DEDUPDITTO:
490 if (spa_version(spa) < SPA_VERSION_DEDUP)
493 error = nvpair_value_uint64(elem, &intval);
495 intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
507 if (!error && reset_bootfs) {
508 error = nvlist_remove(props,
509 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
512 error = nvlist_add_uint64(props,
513 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
521 spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
524 spa_config_dirent_t *dp;
526 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
530 dp = kmem_alloc(sizeof (spa_config_dirent_t),
533 if (cachefile[0] == '\0')
534 dp->scd_path = spa_strdup(spa_config_path);
535 else if (strcmp(cachefile, "none") == 0)
538 dp->scd_path = spa_strdup(cachefile);
540 list_insert_head(&spa->spa_config_list, dp);
542 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
546 spa_prop_set(spa_t *spa, nvlist_t *nvp)
550 boolean_t need_sync = B_FALSE;
553 if ((error = spa_prop_validate(spa, nvp)) != 0)
557 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
558 if ((prop = zpool_name_to_prop(
559 nvpair_name(elem))) == ZPROP_INVAL)
562 if (prop == ZPOOL_PROP_CACHEFILE ||
563 prop == ZPOOL_PROP_ALTROOT ||
564 prop == ZPOOL_PROP_READONLY)
572 return (dsl_sync_task_do(spa_get_dsl(spa), NULL, spa_sync_props,
579 * If the bootfs property value is dsobj, clear it.
582 spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
584 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
585 VERIFY(zap_remove(spa->spa_meta_objset,
586 spa->spa_pool_props_object,
587 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
593 * Change the GUID for the pool. This is done so that we can later
594 * re-import a pool built from a clone of our own vdevs. We will modify
595 * the root vdev's guid, our own pool guid, and then mark all of our
596 * vdevs dirty. Note that we must make sure that all our vdevs are
597 * online when we do this, or else any vdevs that weren't present
598 * would be orphaned from our pool. We are also going to issue a
599 * sysevent to update any watchers.
602 spa_change_guid(spa_t *spa)
604 uint64_t oldguid, newguid;
607 if (!(spa_mode_global & FWRITE))
610 txg = spa_vdev_enter(spa);
612 if (spa->spa_root_vdev->vdev_state != VDEV_STATE_HEALTHY)
613 return (spa_vdev_exit(spa, NULL, txg, ENXIO));
615 oldguid = spa_guid(spa);
616 newguid = spa_generate_guid(NULL);
617 ASSERT3U(oldguid, !=, newguid);
619 spa->spa_root_vdev->vdev_guid = newguid;
620 spa->spa_root_vdev->vdev_guid_sum += (newguid - oldguid);
622 vdev_config_dirty(spa->spa_root_vdev);
624 spa_event_notify(spa, NULL, FM_EREPORT_ZFS_POOL_REGUID);
626 return (spa_vdev_exit(spa, NULL, txg, 0));
630 * ==========================================================================
631 * SPA state manipulation (open/create/destroy/import/export)
632 * ==========================================================================
636 spa_error_entry_compare(const void *a, const void *b)
638 spa_error_entry_t *sa = (spa_error_entry_t *)a;
639 spa_error_entry_t *sb = (spa_error_entry_t *)b;
642 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
643 sizeof (zbookmark_t));
654 * Utility function which retrieves copies of the current logs and
655 * re-initializes them in the process.
658 spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
660 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
662 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
663 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
665 avl_create(&spa->spa_errlist_scrub,
666 spa_error_entry_compare, sizeof (spa_error_entry_t),
667 offsetof(spa_error_entry_t, se_avl));
668 avl_create(&spa->spa_errlist_last,
669 spa_error_entry_compare, sizeof (spa_error_entry_t),
670 offsetof(spa_error_entry_t, se_avl));
674 spa_taskq_create(spa_t *spa, const char *name, enum zti_modes mode,
677 uint_t flags = TASKQ_PREPOPULATE;
678 boolean_t batch = B_FALSE;
682 return (NULL); /* no taskq needed */
685 ASSERT3U(value, >=, 1);
686 value = MAX(value, 1);
691 flags |= TASKQ_THREADS_CPU_PCT;
692 value = zio_taskq_batch_pct;
695 case zti_mode_online_percent:
696 flags |= TASKQ_THREADS_CPU_PCT;
700 panic("unrecognized mode for %s taskq (%u:%u) in "
706 if (zio_taskq_sysdc && spa->spa_proc != &p0) {
708 flags |= TASKQ_DC_BATCH;
710 return (taskq_create_sysdc(name, value, 50, INT_MAX,
711 spa->spa_proc, zio_taskq_basedc, flags));
713 return (taskq_create_proc(name, value, maxclsyspri, 50, INT_MAX,
714 spa->spa_proc, flags));
718 spa_create_zio_taskqs(spa_t *spa)
722 for (t = 0; t < ZIO_TYPES; t++) {
723 for (q = 0; q < ZIO_TASKQ_TYPES; q++) {
724 const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
725 enum zti_modes mode = ztip->zti_mode;
726 uint_t value = ztip->zti_value;
729 (void) snprintf(name, sizeof (name),
730 "%s_%s", zio_type_name[t], zio_taskq_types[q]);
732 spa->spa_zio_taskq[t][q] =
733 spa_taskq_create(spa, name, mode, value);
738 #if defined(_KERNEL) && defined(HAVE_SPA_THREAD)
740 spa_thread(void *arg)
745 user_t *pu = PTOU(curproc);
747 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
750 ASSERT(curproc != &p0);
751 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
752 "zpool-%s", spa->spa_name);
753 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
755 /* bind this thread to the requested psrset */
756 if (zio_taskq_psrset_bind != PS_NONE) {
758 mutex_enter(&cpu_lock);
759 mutex_enter(&pidlock);
760 mutex_enter(&curproc->p_lock);
762 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
763 0, NULL, NULL) == 0) {
764 curthread->t_bind_pset = zio_taskq_psrset_bind;
767 "Couldn't bind process for zfs pool \"%s\" to "
768 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
771 mutex_exit(&curproc->p_lock);
772 mutex_exit(&pidlock);
773 mutex_exit(&cpu_lock);
777 if (zio_taskq_sysdc) {
778 sysdc_thread_enter(curthread, 100, 0);
781 spa->spa_proc = curproc;
782 spa->spa_did = curthread->t_did;
784 spa_create_zio_taskqs(spa);
786 mutex_enter(&spa->spa_proc_lock);
787 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
789 spa->spa_proc_state = SPA_PROC_ACTIVE;
790 cv_broadcast(&spa->spa_proc_cv);
792 CALLB_CPR_SAFE_BEGIN(&cprinfo);
793 while (spa->spa_proc_state == SPA_PROC_ACTIVE)
794 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
795 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
797 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
798 spa->spa_proc_state = SPA_PROC_GONE;
800 cv_broadcast(&spa->spa_proc_cv);
801 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */
803 mutex_enter(&curproc->p_lock);
809 * Activate an uninitialized pool.
812 spa_activate(spa_t *spa, int mode)
814 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
816 spa->spa_state = POOL_STATE_ACTIVE;
817 spa->spa_mode = mode;
819 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
820 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
822 /* Try to create a covering process */
823 mutex_enter(&spa->spa_proc_lock);
824 ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
825 ASSERT(spa->spa_proc == &p0);
828 #ifdef HAVE_SPA_THREAD
829 /* Only create a process if we're going to be around a while. */
830 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
831 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
833 spa->spa_proc_state = SPA_PROC_CREATED;
834 while (spa->spa_proc_state == SPA_PROC_CREATED) {
835 cv_wait(&spa->spa_proc_cv,
836 &spa->spa_proc_lock);
838 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
839 ASSERT(spa->spa_proc != &p0);
840 ASSERT(spa->spa_did != 0);
844 "Couldn't create process for zfs pool \"%s\"\n",
849 #endif /* HAVE_SPA_THREAD */
850 mutex_exit(&spa->spa_proc_lock);
852 /* If we didn't create a process, we need to create our taskqs. */
853 if (spa->spa_proc == &p0) {
854 spa_create_zio_taskqs(spa);
857 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
858 offsetof(vdev_t, vdev_config_dirty_node));
859 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
860 offsetof(vdev_t, vdev_state_dirty_node));
862 txg_list_create(&spa->spa_vdev_txg_list,
863 offsetof(struct vdev, vdev_txg_node));
865 avl_create(&spa->spa_errlist_scrub,
866 spa_error_entry_compare, sizeof (spa_error_entry_t),
867 offsetof(spa_error_entry_t, se_avl));
868 avl_create(&spa->spa_errlist_last,
869 spa_error_entry_compare, sizeof (spa_error_entry_t),
870 offsetof(spa_error_entry_t, se_avl));
874 * Opposite of spa_activate().
877 spa_deactivate(spa_t *spa)
881 ASSERT(spa->spa_sync_on == B_FALSE);
882 ASSERT(spa->spa_dsl_pool == NULL);
883 ASSERT(spa->spa_root_vdev == NULL);
884 ASSERT(spa->spa_async_zio_root == NULL);
885 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
887 txg_list_destroy(&spa->spa_vdev_txg_list);
889 list_destroy(&spa->spa_config_dirty_list);
890 list_destroy(&spa->spa_state_dirty_list);
892 for (t = 0; t < ZIO_TYPES; t++) {
893 for (q = 0; q < ZIO_TASKQ_TYPES; q++) {
894 if (spa->spa_zio_taskq[t][q] != NULL)
895 taskq_destroy(spa->spa_zio_taskq[t][q]);
896 spa->spa_zio_taskq[t][q] = NULL;
900 metaslab_class_destroy(spa->spa_normal_class);
901 spa->spa_normal_class = NULL;
903 metaslab_class_destroy(spa->spa_log_class);
904 spa->spa_log_class = NULL;
907 * If this was part of an import or the open otherwise failed, we may
908 * still have errors left in the queues. Empty them just in case.
910 spa_errlog_drain(spa);
912 avl_destroy(&spa->spa_errlist_scrub);
913 avl_destroy(&spa->spa_errlist_last);
915 spa->spa_state = POOL_STATE_UNINITIALIZED;
917 mutex_enter(&spa->spa_proc_lock);
918 if (spa->spa_proc_state != SPA_PROC_NONE) {
919 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
920 spa->spa_proc_state = SPA_PROC_DEACTIVATE;
921 cv_broadcast(&spa->spa_proc_cv);
922 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
923 ASSERT(spa->spa_proc != &p0);
924 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
926 ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
927 spa->spa_proc_state = SPA_PROC_NONE;
929 ASSERT(spa->spa_proc == &p0);
930 mutex_exit(&spa->spa_proc_lock);
933 * We want to make sure spa_thread() has actually exited the ZFS
934 * module, so that the module can't be unloaded out from underneath
937 if (spa->spa_did != 0) {
938 thread_join(spa->spa_did);
944 * Verify a pool configuration, and construct the vdev tree appropriately. This
945 * will create all the necessary vdevs in the appropriate layout, with each vdev
946 * in the CLOSED state. This will prep the pool before open/creation/import.
947 * All vdev validation is done by the vdev_alloc() routine.
950 spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
951 uint_t id, int atype)
958 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
961 if ((*vdp)->vdev_ops->vdev_op_leaf)
964 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
976 for (c = 0; c < children; c++) {
978 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
986 ASSERT(*vdp != NULL);
992 * Opposite of spa_load().
995 spa_unload(spa_t *spa)
999 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1004 spa_async_suspend(spa);
1009 if (spa->spa_sync_on) {
1010 txg_sync_stop(spa->spa_dsl_pool);
1011 spa->spa_sync_on = B_FALSE;
1015 * Wait for any outstanding async I/O to complete.
1017 if (spa->spa_async_zio_root != NULL) {
1018 (void) zio_wait(spa->spa_async_zio_root);
1019 spa->spa_async_zio_root = NULL;
1022 bpobj_close(&spa->spa_deferred_bpobj);
1025 * Close the dsl pool.
1027 if (spa->spa_dsl_pool) {
1028 dsl_pool_close(spa->spa_dsl_pool);
1029 spa->spa_dsl_pool = NULL;
1030 spa->spa_meta_objset = NULL;
1035 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1038 * Drop and purge level 2 cache
1040 spa_l2cache_drop(spa);
1045 if (spa->spa_root_vdev)
1046 vdev_free(spa->spa_root_vdev);
1047 ASSERT(spa->spa_root_vdev == NULL);
1049 for (i = 0; i < spa->spa_spares.sav_count; i++)
1050 vdev_free(spa->spa_spares.sav_vdevs[i]);
1051 if (spa->spa_spares.sav_vdevs) {
1052 kmem_free(spa->spa_spares.sav_vdevs,
1053 spa->spa_spares.sav_count * sizeof (void *));
1054 spa->spa_spares.sav_vdevs = NULL;
1056 if (spa->spa_spares.sav_config) {
1057 nvlist_free(spa->spa_spares.sav_config);
1058 spa->spa_spares.sav_config = NULL;
1060 spa->spa_spares.sav_count = 0;
1062 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1063 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1064 vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1066 if (spa->spa_l2cache.sav_vdevs) {
1067 kmem_free(spa->spa_l2cache.sav_vdevs,
1068 spa->spa_l2cache.sav_count * sizeof (void *));
1069 spa->spa_l2cache.sav_vdevs = NULL;
1071 if (spa->spa_l2cache.sav_config) {
1072 nvlist_free(spa->spa_l2cache.sav_config);
1073 spa->spa_l2cache.sav_config = NULL;
1075 spa->spa_l2cache.sav_count = 0;
1077 spa->spa_async_suspended = 0;
1079 if (spa->spa_comment != NULL) {
1080 spa_strfree(spa->spa_comment);
1081 spa->spa_comment = NULL;
1084 spa_config_exit(spa, SCL_ALL, FTAG);
1088 * Load (or re-load) the current list of vdevs describing the active spares for
1089 * this pool. When this is called, we have some form of basic information in
1090 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1091 * then re-generate a more complete list including status information.
1094 spa_load_spares(spa_t *spa)
1101 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1104 * First, close and free any existing spare vdevs.
1106 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1107 vd = spa->spa_spares.sav_vdevs[i];
1109 /* Undo the call to spa_activate() below */
1110 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1111 B_FALSE)) != NULL && tvd->vdev_isspare)
1112 spa_spare_remove(tvd);
1117 if (spa->spa_spares.sav_vdevs)
1118 kmem_free(spa->spa_spares.sav_vdevs,
1119 spa->spa_spares.sav_count * sizeof (void *));
1121 if (spa->spa_spares.sav_config == NULL)
1124 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1125 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1127 spa->spa_spares.sav_count = (int)nspares;
1128 spa->spa_spares.sav_vdevs = NULL;
1134 * Construct the array of vdevs, opening them to get status in the
1135 * process. For each spare, there is potentially two different vdev_t
1136 * structures associated with it: one in the list of spares (used only
1137 * for basic validation purposes) and one in the active vdev
1138 * configuration (if it's spared in). During this phase we open and
1139 * validate each vdev on the spare list. If the vdev also exists in the
1140 * active configuration, then we also mark this vdev as an active spare.
1142 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1144 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1145 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1146 VDEV_ALLOC_SPARE) == 0);
1149 spa->spa_spares.sav_vdevs[i] = vd;
1151 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1152 B_FALSE)) != NULL) {
1153 if (!tvd->vdev_isspare)
1157 * We only mark the spare active if we were successfully
1158 * able to load the vdev. Otherwise, importing a pool
1159 * with a bad active spare would result in strange
1160 * behavior, because multiple pool would think the spare
1161 * is actively in use.
1163 * There is a vulnerability here to an equally bizarre
1164 * circumstance, where a dead active spare is later
1165 * brought back to life (onlined or otherwise). Given
1166 * the rarity of this scenario, and the extra complexity
1167 * it adds, we ignore the possibility.
1169 if (!vdev_is_dead(tvd))
1170 spa_spare_activate(tvd);
1174 vd->vdev_aux = &spa->spa_spares;
1176 if (vdev_open(vd) != 0)
1179 if (vdev_validate_aux(vd) == 0)
1184 * Recompute the stashed list of spares, with status information
1187 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1188 DATA_TYPE_NVLIST_ARRAY) == 0);
1190 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1192 for (i = 0; i < spa->spa_spares.sav_count; i++)
1193 spares[i] = vdev_config_generate(spa,
1194 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1195 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1196 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1197 for (i = 0; i < spa->spa_spares.sav_count; i++)
1198 nvlist_free(spares[i]);
1199 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1203 * Load (or re-load) the current list of vdevs describing the active l2cache for
1204 * this pool. When this is called, we have some form of basic information in
1205 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1206 * then re-generate a more complete list including status information.
1207 * Devices which are already active have their details maintained, and are
1211 spa_load_l2cache(spa_t *spa)
1215 int i, j, oldnvdevs;
1217 vdev_t *vd, **oldvdevs, **newvdevs = NULL;
1218 spa_aux_vdev_t *sav = &spa->spa_l2cache;
1220 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1222 if (sav->sav_config != NULL) {
1223 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1224 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1225 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1230 oldvdevs = sav->sav_vdevs;
1231 oldnvdevs = sav->sav_count;
1232 sav->sav_vdevs = NULL;
1236 * Process new nvlist of vdevs.
1238 for (i = 0; i < nl2cache; i++) {
1239 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1243 for (j = 0; j < oldnvdevs; j++) {
1245 if (vd != NULL && guid == vd->vdev_guid) {
1247 * Retain previous vdev for add/remove ops.
1255 if (newvdevs[i] == NULL) {
1259 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1260 VDEV_ALLOC_L2CACHE) == 0);
1265 * Commit this vdev as an l2cache device,
1266 * even if it fails to open.
1268 spa_l2cache_add(vd);
1273 spa_l2cache_activate(vd);
1275 if (vdev_open(vd) != 0)
1278 (void) vdev_validate_aux(vd);
1280 if (!vdev_is_dead(vd))
1281 l2arc_add_vdev(spa, vd);
1286 * Purge vdevs that were dropped
1288 for (i = 0; i < oldnvdevs; i++) {
1293 ASSERT(vd->vdev_isl2cache);
1295 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1296 pool != 0ULL && l2arc_vdev_present(vd))
1297 l2arc_remove_vdev(vd);
1298 vdev_clear_stats(vd);
1304 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1306 if (sav->sav_config == NULL)
1309 sav->sav_vdevs = newvdevs;
1310 sav->sav_count = (int)nl2cache;
1313 * Recompute the stashed list of l2cache devices, with status
1314 * information this time.
1316 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1317 DATA_TYPE_NVLIST_ARRAY) == 0);
1319 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1320 for (i = 0; i < sav->sav_count; i++)
1321 l2cache[i] = vdev_config_generate(spa,
1322 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1323 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1324 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1326 for (i = 0; i < sav->sav_count; i++)
1327 nvlist_free(l2cache[i]);
1329 kmem_free(l2cache, sav->sav_count * sizeof (void *));
1333 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1336 char *packed = NULL;
1341 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
1342 nvsize = *(uint64_t *)db->db_data;
1343 dmu_buf_rele(db, FTAG);
1345 packed = kmem_alloc(nvsize, KM_SLEEP | KM_NODEBUG);
1346 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1349 error = nvlist_unpack(packed, nvsize, value, 0);
1350 kmem_free(packed, nvsize);
1356 * Checks to see if the given vdev could not be opened, in which case we post a
1357 * sysevent to notify the autoreplace code that the device has been removed.
1360 spa_check_removed(vdev_t *vd)
1364 for (c = 0; c < vd->vdev_children; c++)
1365 spa_check_removed(vd->vdev_child[c]);
1367 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd)) {
1368 zfs_ereport_post(FM_EREPORT_RESOURCE_AUTOREPLACE,
1369 vd->vdev_spa, vd, NULL, 0, 0);
1370 spa_event_notify(vd->vdev_spa, vd, FM_EREPORT_ZFS_DEVICE_CHECK);
1375 * Validate the current config against the MOS config
1378 spa_config_valid(spa_t *spa, nvlist_t *config)
1380 vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1384 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1386 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1387 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1389 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1392 * If we're doing a normal import, then build up any additional
1393 * diagnostic information about missing devices in this config.
1394 * We'll pass this up to the user for further processing.
1396 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1397 nvlist_t **child, *nv;
1400 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1402 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1404 for (c = 0; c < rvd->vdev_children; c++) {
1405 vdev_t *tvd = rvd->vdev_child[c];
1406 vdev_t *mtvd = mrvd->vdev_child[c];
1408 if (tvd->vdev_ops == &vdev_missing_ops &&
1409 mtvd->vdev_ops != &vdev_missing_ops &&
1411 child[idx++] = vdev_config_generate(spa, mtvd,
1416 VERIFY(nvlist_add_nvlist_array(nv,
1417 ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1418 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1419 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1421 for (i = 0; i < idx; i++)
1422 nvlist_free(child[i]);
1425 kmem_free(child, rvd->vdev_children * sizeof (char **));
1429 * Compare the root vdev tree with the information we have
1430 * from the MOS config (mrvd). Check each top-level vdev
1431 * with the corresponding MOS config top-level (mtvd).
1433 for (c = 0; c < rvd->vdev_children; c++) {
1434 vdev_t *tvd = rvd->vdev_child[c];
1435 vdev_t *mtvd = mrvd->vdev_child[c];
1438 * Resolve any "missing" vdevs in the current configuration.
1439 * If we find that the MOS config has more accurate information
1440 * about the top-level vdev then use that vdev instead.
1442 if (tvd->vdev_ops == &vdev_missing_ops &&
1443 mtvd->vdev_ops != &vdev_missing_ops) {
1445 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1449 * Device specific actions.
1451 if (mtvd->vdev_islog) {
1452 spa_set_log_state(spa, SPA_LOG_CLEAR);
1455 * XXX - once we have 'readonly' pool
1456 * support we should be able to handle
1457 * missing data devices by transitioning
1458 * the pool to readonly.
1464 * Swap the missing vdev with the data we were
1465 * able to obtain from the MOS config.
1467 vdev_remove_child(rvd, tvd);
1468 vdev_remove_child(mrvd, mtvd);
1470 vdev_add_child(rvd, mtvd);
1471 vdev_add_child(mrvd, tvd);
1473 spa_config_exit(spa, SCL_ALL, FTAG);
1475 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1478 } else if (mtvd->vdev_islog) {
1480 * Load the slog device's state from the MOS config
1481 * since it's possible that the label does not
1482 * contain the most up-to-date information.
1484 vdev_load_log_state(tvd, mtvd);
1489 spa_config_exit(spa, SCL_ALL, FTAG);
1492 * Ensure we were able to validate the config.
1494 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1498 * Check for missing log devices
1501 spa_check_logs(spa_t *spa)
1503 switch (spa->spa_log_state) {
1506 case SPA_LOG_MISSING:
1507 /* need to recheck in case slog has been restored */
1508 case SPA_LOG_UNKNOWN:
1509 if (dmu_objset_find(spa->spa_name, zil_check_log_chain, NULL,
1510 DS_FIND_CHILDREN)) {
1511 spa_set_log_state(spa, SPA_LOG_MISSING);
1520 spa_passivate_log(spa_t *spa)
1522 vdev_t *rvd = spa->spa_root_vdev;
1523 boolean_t slog_found = B_FALSE;
1526 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1528 if (!spa_has_slogs(spa))
1531 for (c = 0; c < rvd->vdev_children; c++) {
1532 vdev_t *tvd = rvd->vdev_child[c];
1533 metaslab_group_t *mg = tvd->vdev_mg;
1535 if (tvd->vdev_islog) {
1536 metaslab_group_passivate(mg);
1537 slog_found = B_TRUE;
1541 return (slog_found);
1545 spa_activate_log(spa_t *spa)
1547 vdev_t *rvd = spa->spa_root_vdev;
1550 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1552 for (c = 0; c < rvd->vdev_children; c++) {
1553 vdev_t *tvd = rvd->vdev_child[c];
1554 metaslab_group_t *mg = tvd->vdev_mg;
1556 if (tvd->vdev_islog)
1557 metaslab_group_activate(mg);
1562 spa_offline_log(spa_t *spa)
1566 if ((error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1567 NULL, DS_FIND_CHILDREN)) == 0) {
1570 * We successfully offlined the log device, sync out the
1571 * current txg so that the "stubby" block can be removed
1574 txg_wait_synced(spa->spa_dsl_pool, 0);
1580 spa_aux_check_removed(spa_aux_vdev_t *sav)
1584 for (i = 0; i < sav->sav_count; i++)
1585 spa_check_removed(sav->sav_vdevs[i]);
1589 spa_claim_notify(zio_t *zio)
1591 spa_t *spa = zio->io_spa;
1596 mutex_enter(&spa->spa_props_lock); /* any mutex will do */
1597 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1598 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1599 mutex_exit(&spa->spa_props_lock);
1602 typedef struct spa_load_error {
1603 uint64_t sle_meta_count;
1604 uint64_t sle_data_count;
1608 spa_load_verify_done(zio_t *zio)
1610 blkptr_t *bp = zio->io_bp;
1611 spa_load_error_t *sle = zio->io_private;
1612 dmu_object_type_t type = BP_GET_TYPE(bp);
1613 int error = zio->io_error;
1616 if ((BP_GET_LEVEL(bp) != 0 || dmu_ot[type].ot_metadata) &&
1617 type != DMU_OT_INTENT_LOG)
1618 atomic_add_64(&sle->sle_meta_count, 1);
1620 atomic_add_64(&sle->sle_data_count, 1);
1622 zio_data_buf_free(zio->io_data, zio->io_size);
1627 spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1628 arc_buf_t *pbuf, const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1632 size_t size = BP_GET_PSIZE(bp);
1633 void *data = zio_data_buf_alloc(size);
1635 zio_nowait(zio_read(rio, spa, bp, data, size,
1636 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1637 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1638 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1644 spa_load_verify(spa_t *spa)
1647 spa_load_error_t sle = { 0 };
1648 zpool_rewind_policy_t policy;
1649 boolean_t verify_ok = B_FALSE;
1652 zpool_get_rewind_policy(spa->spa_config, &policy);
1654 if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1657 rio = zio_root(spa, NULL, &sle,
1658 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1660 error = traverse_pool(spa, spa->spa_verify_min_txg,
1661 TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio);
1663 (void) zio_wait(rio);
1665 spa->spa_load_meta_errors = sle.sle_meta_count;
1666 spa->spa_load_data_errors = sle.sle_data_count;
1668 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1669 sle.sle_data_count <= policy.zrp_maxdata) {
1673 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1674 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1676 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1677 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1678 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1679 VERIFY(nvlist_add_int64(spa->spa_load_info,
1680 ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1681 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1682 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1684 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1688 if (error != ENXIO && error != EIO)
1693 return (verify_ok ? 0 : EIO);
1697 * Find a value in the pool props object.
1700 spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1702 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1703 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1707 * Find a value in the pool directory object.
1710 spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1712 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1713 name, sizeof (uint64_t), 1, val));
1717 spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1719 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1724 * Fix up config after a partly-completed split. This is done with the
1725 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
1726 * pool have that entry in their config, but only the splitting one contains
1727 * a list of all the guids of the vdevs that are being split off.
1729 * This function determines what to do with that list: either rejoin
1730 * all the disks to the pool, or complete the splitting process. To attempt
1731 * the rejoin, each disk that is offlined is marked online again, and
1732 * we do a reopen() call. If the vdev label for every disk that was
1733 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1734 * then we call vdev_split() on each disk, and complete the split.
1736 * Otherwise we leave the config alone, with all the vdevs in place in
1737 * the original pool.
1740 spa_try_repair(spa_t *spa, nvlist_t *config)
1747 boolean_t attempt_reopen;
1749 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
1752 /* check that the config is complete */
1753 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
1754 &glist, &gcount) != 0)
1757 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
1759 /* attempt to online all the vdevs & validate */
1760 attempt_reopen = B_TRUE;
1761 for (i = 0; i < gcount; i++) {
1762 if (glist[i] == 0) /* vdev is hole */
1765 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
1766 if (vd[i] == NULL) {
1768 * Don't bother attempting to reopen the disks;
1769 * just do the split.
1771 attempt_reopen = B_FALSE;
1773 /* attempt to re-online it */
1774 vd[i]->vdev_offline = B_FALSE;
1778 if (attempt_reopen) {
1779 vdev_reopen(spa->spa_root_vdev);
1781 /* check each device to see what state it's in */
1782 for (extracted = 0, i = 0; i < gcount; i++) {
1783 if (vd[i] != NULL &&
1784 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
1791 * If every disk has been moved to the new pool, or if we never
1792 * even attempted to look at them, then we split them off for
1795 if (!attempt_reopen || gcount == extracted) {
1796 for (i = 0; i < gcount; i++)
1799 vdev_reopen(spa->spa_root_vdev);
1802 kmem_free(vd, gcount * sizeof (vdev_t *));
1806 spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
1807 boolean_t mosconfig)
1809 nvlist_t *config = spa->spa_config;
1810 char *ereport = FM_EREPORT_ZFS_POOL;
1816 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
1819 ASSERT(spa->spa_comment == NULL);
1820 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
1821 spa->spa_comment = spa_strdup(comment);
1824 * Versioning wasn't explicitly added to the label until later, so if
1825 * it's not present treat it as the initial version.
1827 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
1828 &spa->spa_ubsync.ub_version) != 0)
1829 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
1831 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
1832 &spa->spa_config_txg);
1834 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
1835 spa_guid_exists(pool_guid, 0)) {
1838 spa->spa_config_guid = pool_guid;
1840 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
1842 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
1846 gethrestime(&spa->spa_loaded_ts);
1847 error = spa_load_impl(spa, pool_guid, config, state, type,
1848 mosconfig, &ereport);
1851 spa->spa_minref = refcount_count(&spa->spa_refcount);
1853 if (error != EEXIST) {
1854 spa->spa_loaded_ts.tv_sec = 0;
1855 spa->spa_loaded_ts.tv_nsec = 0;
1857 if (error != EBADF) {
1858 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
1861 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
1868 * Load an existing storage pool, using the pool's builtin spa_config as a
1869 * source of configuration information.
1871 __attribute__((always_inline))
1873 spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
1874 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
1878 nvlist_t *nvroot = NULL;
1880 uberblock_t *ub = &spa->spa_uberblock;
1881 uint64_t children, config_cache_txg = spa->spa_config_txg;
1882 int orig_mode = spa->spa_mode;
1887 * If this is an untrusted config, access the pool in read-only mode.
1888 * This prevents things like resilvering recently removed devices.
1891 spa->spa_mode = FREAD;
1893 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1895 spa->spa_load_state = state;
1897 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
1900 parse = (type == SPA_IMPORT_EXISTING ?
1901 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
1904 * Create "The Godfather" zio to hold all async IOs
1906 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
1907 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
1910 * Parse the configuration into a vdev tree. We explicitly set the
1911 * value that will be returned by spa_version() since parsing the
1912 * configuration requires knowing the version number.
1914 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1915 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
1916 spa_config_exit(spa, SCL_ALL, FTAG);
1921 ASSERT(spa->spa_root_vdev == rvd);
1923 if (type != SPA_IMPORT_ASSEMBLE) {
1924 ASSERT(spa_guid(spa) == pool_guid);
1928 * Try to open all vdevs, loading each label in the process.
1930 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1931 error = vdev_open(rvd);
1932 spa_config_exit(spa, SCL_ALL, FTAG);
1937 * We need to validate the vdev labels against the configuration that
1938 * we have in hand, which is dependent on the setting of mosconfig. If
1939 * mosconfig is true then we're validating the vdev labels based on
1940 * that config. Otherwise, we're validating against the cached config
1941 * (zpool.cache) that was read when we loaded the zfs module, and then
1942 * later we will recursively call spa_load() and validate against
1945 * If we're assembling a new pool that's been split off from an
1946 * existing pool, the labels haven't yet been updated so we skip
1947 * validation for now.
1949 if (type != SPA_IMPORT_ASSEMBLE) {
1950 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1951 error = vdev_validate(rvd, mosconfig);
1952 spa_config_exit(spa, SCL_ALL, FTAG);
1957 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
1962 * Find the best uberblock.
1964 vdev_uberblock_load(NULL, rvd, ub);
1967 * If we weren't able to find a single valid uberblock, return failure.
1969 if (ub->ub_txg == 0)
1970 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
1973 * If the pool is newer than the code, we can't open it.
1975 if (ub->ub_version > SPA_VERSION)
1976 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
1979 * If the vdev guid sum doesn't match the uberblock, we have an
1980 * incomplete configuration. We first check to see if the pool
1981 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
1982 * If it is, defer the vdev_guid_sum check till later so we
1983 * can handle missing vdevs.
1985 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
1986 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
1987 rvd->vdev_guid_sum != ub->ub_guid_sum)
1988 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
1990 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
1991 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1992 spa_try_repair(spa, config);
1993 spa_config_exit(spa, SCL_ALL, FTAG);
1994 nvlist_free(spa->spa_config_splitting);
1995 spa->spa_config_splitting = NULL;
1999 * Initialize internal SPA structures.
2001 spa->spa_state = POOL_STATE_ACTIVE;
2002 spa->spa_ubsync = spa->spa_uberblock;
2003 spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2004 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2005 spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2006 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2007 spa->spa_claim_max_txg = spa->spa_first_txg;
2008 spa->spa_prev_software_version = ub->ub_software_version;
2010 error = dsl_pool_open(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2012 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2013 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2015 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2016 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2020 nvlist_t *policy = NULL, *nvconfig;
2022 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2023 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2025 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2026 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2028 unsigned long myhostid = 0;
2030 VERIFY(nvlist_lookup_string(nvconfig,
2031 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2034 myhostid = zone_get_hostid(NULL);
2037 * We're emulating the system's hostid in userland, so
2038 * we can't use zone_get_hostid().
2040 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2041 #endif /* _KERNEL */
2042 if (hostid != 0 && myhostid != 0 &&
2043 hostid != myhostid) {
2044 nvlist_free(nvconfig);
2045 cmn_err(CE_WARN, "pool '%s' could not be "
2046 "loaded as it was last accessed by "
2047 "another system (host: %s hostid: 0x%lx). "
2048 "See: http://zfsonlinux.org/msg/ZFS-8000-EY",
2049 spa_name(spa), hostname,
2050 (unsigned long)hostid);
2054 if (nvlist_lookup_nvlist(spa->spa_config,
2055 ZPOOL_REWIND_POLICY, &policy) == 0)
2056 VERIFY(nvlist_add_nvlist(nvconfig,
2057 ZPOOL_REWIND_POLICY, policy) == 0);
2059 spa_config_set(spa, nvconfig);
2061 spa_deactivate(spa);
2062 spa_activate(spa, orig_mode);
2064 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2067 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2068 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2069 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2071 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2074 * Load the bit that tells us to use the new accounting function
2075 * (raid-z deflation). If we have an older pool, this will not
2078 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2079 if (error != 0 && error != ENOENT)
2080 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2082 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2083 &spa->spa_creation_version);
2084 if (error != 0 && error != ENOENT)
2085 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2088 * Load the persistent error log. If we have an older pool, this will
2091 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2092 if (error != 0 && error != ENOENT)
2093 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2095 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2096 &spa->spa_errlog_scrub);
2097 if (error != 0 && error != ENOENT)
2098 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2101 * Load the history object. If we have an older pool, this
2102 * will not be present.
2104 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2105 if (error != 0 && error != ENOENT)
2106 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2109 * If we're assembling the pool from the split-off vdevs of
2110 * an existing pool, we don't want to attach the spares & cache
2115 * Load any hot spares for this pool.
2117 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2118 if (error != 0 && error != ENOENT)
2119 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2120 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2121 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2122 if (load_nvlist(spa, spa->spa_spares.sav_object,
2123 &spa->spa_spares.sav_config) != 0)
2124 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2126 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2127 spa_load_spares(spa);
2128 spa_config_exit(spa, SCL_ALL, FTAG);
2129 } else if (error == 0) {
2130 spa->spa_spares.sav_sync = B_TRUE;
2134 * Load any level 2 ARC devices for this pool.
2136 error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2137 &spa->spa_l2cache.sav_object);
2138 if (error != 0 && error != ENOENT)
2139 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2140 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2141 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2142 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2143 &spa->spa_l2cache.sav_config) != 0)
2144 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2146 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2147 spa_load_l2cache(spa);
2148 spa_config_exit(spa, SCL_ALL, FTAG);
2149 } else if (error == 0) {
2150 spa->spa_l2cache.sav_sync = B_TRUE;
2153 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2155 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2156 if (error && error != ENOENT)
2157 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2160 uint64_t autoreplace;
2162 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2163 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2164 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2165 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2166 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2167 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2168 &spa->spa_dedup_ditto);
2170 spa->spa_autoreplace = (autoreplace != 0);
2174 * If the 'autoreplace' property is set, then post a resource notifying
2175 * the ZFS DE that it should not issue any faults for unopenable
2176 * devices. We also iterate over the vdevs, and post a sysevent for any
2177 * unopenable vdevs so that the normal autoreplace handler can take
2180 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2181 spa_check_removed(spa->spa_root_vdev);
2183 * For the import case, this is done in spa_import(), because
2184 * at this point we're using the spare definitions from
2185 * the MOS config, not necessarily from the userland config.
2187 if (state != SPA_LOAD_IMPORT) {
2188 spa_aux_check_removed(&spa->spa_spares);
2189 spa_aux_check_removed(&spa->spa_l2cache);
2194 * Load the vdev state for all toplevel vdevs.
2199 * Propagate the leaf DTLs we just loaded all the way up the tree.
2201 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2202 vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2203 spa_config_exit(spa, SCL_ALL, FTAG);
2206 * Load the DDTs (dedup tables).
2208 error = ddt_load(spa);
2210 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2212 spa_update_dspace(spa);
2215 * Validate the config, using the MOS config to fill in any
2216 * information which might be missing. If we fail to validate
2217 * the config then declare the pool unfit for use. If we're
2218 * assembling a pool from a split, the log is not transferred
2221 if (type != SPA_IMPORT_ASSEMBLE) {
2224 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2225 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2227 if (!spa_config_valid(spa, nvconfig)) {
2228 nvlist_free(nvconfig);
2229 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2232 nvlist_free(nvconfig);
2235 * Now that we've validate the config, check the state of the
2236 * root vdev. If it can't be opened, it indicates one or
2237 * more toplevel vdevs are faulted.
2239 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2242 if (spa_check_logs(spa)) {
2243 *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2244 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2249 * We've successfully opened the pool, verify that we're ready
2250 * to start pushing transactions.
2252 if (state != SPA_LOAD_TRYIMPORT) {
2253 if ((error = spa_load_verify(spa)))
2254 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2258 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2259 spa->spa_load_max_txg == UINT64_MAX)) {
2261 int need_update = B_FALSE;
2264 ASSERT(state != SPA_LOAD_TRYIMPORT);
2267 * Claim log blocks that haven't been committed yet.
2268 * This must all happen in a single txg.
2269 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2270 * invoked from zil_claim_log_block()'s i/o done callback.
2271 * Price of rollback is that we abandon the log.
2273 spa->spa_claiming = B_TRUE;
2275 tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2276 spa_first_txg(spa));
2277 (void) dmu_objset_find(spa_name(spa),
2278 zil_claim, tx, DS_FIND_CHILDREN);
2281 spa->spa_claiming = B_FALSE;
2283 spa_set_log_state(spa, SPA_LOG_GOOD);
2284 spa->spa_sync_on = B_TRUE;
2285 txg_sync_start(spa->spa_dsl_pool);
2288 * Wait for all claims to sync. We sync up to the highest
2289 * claimed log block birth time so that claimed log blocks
2290 * don't appear to be from the future. spa_claim_max_txg
2291 * will have been set for us by either zil_check_log_chain()
2292 * (invoked from spa_check_logs()) or zil_claim() above.
2294 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2297 * If the config cache is stale, or we have uninitialized
2298 * metaslabs (see spa_vdev_add()), then update the config.
2300 * If this is a verbatim import, trust the current
2301 * in-core spa_config and update the disk labels.
2303 if (config_cache_txg != spa->spa_config_txg ||
2304 state == SPA_LOAD_IMPORT ||
2305 state == SPA_LOAD_RECOVER ||
2306 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2307 need_update = B_TRUE;
2309 for (c = 0; c < rvd->vdev_children; c++)
2310 if (rvd->vdev_child[c]->vdev_ms_array == 0)
2311 need_update = B_TRUE;
2314 * Update the config cache asychronously in case we're the
2315 * root pool, in which case the config cache isn't writable yet.
2318 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2321 * Check all DTLs to see if anything needs resilvering.
2323 if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2324 vdev_resilver_needed(rvd, NULL, NULL))
2325 spa_async_request(spa, SPA_ASYNC_RESILVER);
2328 * Delete any inconsistent datasets.
2330 (void) dmu_objset_find(spa_name(spa),
2331 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2334 * Clean up any stale temporary dataset userrefs.
2336 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2343 spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2345 int mode = spa->spa_mode;
2348 spa_deactivate(spa);
2350 spa->spa_load_max_txg--;
2352 spa_activate(spa, mode);
2353 spa_async_suspend(spa);
2355 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2359 spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2360 uint64_t max_request, int rewind_flags)
2362 nvlist_t *config = NULL;
2363 int load_error, rewind_error;
2364 uint64_t safe_rewind_txg;
2367 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2368 spa->spa_load_max_txg = spa->spa_load_txg;
2369 spa_set_log_state(spa, SPA_LOG_CLEAR);
2371 spa->spa_load_max_txg = max_request;
2374 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2376 if (load_error == 0)
2379 if (spa->spa_root_vdev != NULL)
2380 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2382 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2383 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2385 if (rewind_flags & ZPOOL_NEVER_REWIND) {
2386 nvlist_free(config);
2387 return (load_error);
2390 /* Price of rolling back is discarding txgs, including log */
2391 if (state == SPA_LOAD_RECOVER)
2392 spa_set_log_state(spa, SPA_LOG_CLEAR);
2394 spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2395 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2396 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2397 TXG_INITIAL : safe_rewind_txg;
2400 * Continue as long as we're finding errors, we're still within
2401 * the acceptable rewind range, and we're still finding uberblocks
2403 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2404 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2405 if (spa->spa_load_max_txg < safe_rewind_txg)
2406 spa->spa_extreme_rewind = B_TRUE;
2407 rewind_error = spa_load_retry(spa, state, mosconfig);
2410 spa->spa_extreme_rewind = B_FALSE;
2411 spa->spa_load_max_txg = UINT64_MAX;
2413 if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2414 spa_config_set(spa, config);
2416 return (state == SPA_LOAD_RECOVER ? rewind_error : load_error);
2422 * The import case is identical to an open except that the configuration is sent
2423 * down from userland, instead of grabbed from the configuration cache. For the
2424 * case of an open, the pool configuration will exist in the
2425 * POOL_STATE_UNINITIALIZED state.
2427 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2428 * the same time open the pool, without having to keep around the spa_t in some
2432 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2436 spa_load_state_t state = SPA_LOAD_OPEN;
2438 int locked = B_FALSE;
2443 * As disgusting as this is, we need to support recursive calls to this
2444 * function because dsl_dir_open() is called during spa_load(), and ends
2445 * up calling spa_open() again. The real fix is to figure out how to
2446 * avoid dsl_dir_open() calling this in the first place.
2448 if (mutex_owner(&spa_namespace_lock) != curthread) {
2449 mutex_enter(&spa_namespace_lock);
2453 if ((spa = spa_lookup(pool)) == NULL) {
2455 mutex_exit(&spa_namespace_lock);
2459 if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2460 zpool_rewind_policy_t policy;
2462 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2464 if (policy.zrp_request & ZPOOL_DO_REWIND)
2465 state = SPA_LOAD_RECOVER;
2467 spa_activate(spa, spa_mode_global);
2469 if (state != SPA_LOAD_RECOVER)
2470 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2472 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2473 policy.zrp_request);
2475 if (error == EBADF) {
2477 * If vdev_validate() returns failure (indicated by
2478 * EBADF), it indicates that one of the vdevs indicates
2479 * that the pool has been exported or destroyed. If
2480 * this is the case, the config cache is out of sync and
2481 * we should remove the pool from the namespace.
2484 spa_deactivate(spa);
2485 spa_config_sync(spa, B_TRUE, B_TRUE);
2488 mutex_exit(&spa_namespace_lock);
2494 * We can't open the pool, but we still have useful
2495 * information: the state of each vdev after the
2496 * attempted vdev_open(). Return this to the user.
2498 if (config != NULL && spa->spa_config) {
2499 VERIFY(nvlist_dup(spa->spa_config, config,
2501 VERIFY(nvlist_add_nvlist(*config,
2502 ZPOOL_CONFIG_LOAD_INFO,
2503 spa->spa_load_info) == 0);
2506 spa_deactivate(spa);
2507 spa->spa_last_open_failed = error;
2509 mutex_exit(&spa_namespace_lock);
2515 spa_open_ref(spa, tag);
2518 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2521 * If we've recovered the pool, pass back any information we
2522 * gathered while doing the load.
2524 if (state == SPA_LOAD_RECOVER) {
2525 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
2526 spa->spa_load_info) == 0);
2530 spa->spa_last_open_failed = 0;
2531 spa->spa_last_ubsync_txg = 0;
2532 spa->spa_load_txg = 0;
2533 mutex_exit(&spa_namespace_lock);
2542 spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
2545 return (spa_open_common(name, spapp, tag, policy, config));
2549 spa_open(const char *name, spa_t **spapp, void *tag)
2551 return (spa_open_common(name, spapp, tag, NULL, NULL));
2555 * Lookup the given spa_t, incrementing the inject count in the process,
2556 * preventing it from being exported or destroyed.
2559 spa_inject_addref(char *name)
2563 mutex_enter(&spa_namespace_lock);
2564 if ((spa = spa_lookup(name)) == NULL) {
2565 mutex_exit(&spa_namespace_lock);
2568 spa->spa_inject_ref++;
2569 mutex_exit(&spa_namespace_lock);
2575 spa_inject_delref(spa_t *spa)
2577 mutex_enter(&spa_namespace_lock);
2578 spa->spa_inject_ref--;
2579 mutex_exit(&spa_namespace_lock);
2583 * Add spares device information to the nvlist.
2586 spa_add_spares(spa_t *spa, nvlist_t *config)
2596 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2598 if (spa->spa_spares.sav_count == 0)
2601 VERIFY(nvlist_lookup_nvlist(config,
2602 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2603 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
2604 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2606 VERIFY(nvlist_add_nvlist_array(nvroot,
2607 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2608 VERIFY(nvlist_lookup_nvlist_array(nvroot,
2609 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2612 * Go through and find any spares which have since been
2613 * repurposed as an active spare. If this is the case, update
2614 * their status appropriately.
2616 for (i = 0; i < nspares; i++) {
2617 VERIFY(nvlist_lookup_uint64(spares[i],
2618 ZPOOL_CONFIG_GUID, &guid) == 0);
2619 if (spa_spare_exists(guid, &pool, NULL) &&
2621 VERIFY(nvlist_lookup_uint64_array(
2622 spares[i], ZPOOL_CONFIG_VDEV_STATS,
2623 (uint64_t **)&vs, &vsc) == 0);
2624 vs->vs_state = VDEV_STATE_CANT_OPEN;
2625 vs->vs_aux = VDEV_AUX_SPARED;
2632 * Add l2cache device information to the nvlist, including vdev stats.
2635 spa_add_l2cache(spa_t *spa, nvlist_t *config)
2638 uint_t i, j, nl2cache;
2645 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2647 if (spa->spa_l2cache.sav_count == 0)
2650 VERIFY(nvlist_lookup_nvlist(config,
2651 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2652 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
2653 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2654 if (nl2cache != 0) {
2655 VERIFY(nvlist_add_nvlist_array(nvroot,
2656 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2657 VERIFY(nvlist_lookup_nvlist_array(nvroot,
2658 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2661 * Update level 2 cache device stats.
2664 for (i = 0; i < nl2cache; i++) {
2665 VERIFY(nvlist_lookup_uint64(l2cache[i],
2666 ZPOOL_CONFIG_GUID, &guid) == 0);
2669 for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
2671 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
2672 vd = spa->spa_l2cache.sav_vdevs[j];
2678 VERIFY(nvlist_lookup_uint64_array(l2cache[i],
2679 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
2681 vdev_get_stats(vd, vs);
2687 spa_get_stats(const char *name, nvlist_t **config, char *altroot, size_t buflen)
2693 error = spa_open_common(name, &spa, FTAG, NULL, config);
2697 * This still leaves a window of inconsistency where the spares
2698 * or l2cache devices could change and the config would be
2699 * self-inconsistent.
2701 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
2703 if (*config != NULL) {
2704 uint64_t loadtimes[2];
2706 loadtimes[0] = spa->spa_loaded_ts.tv_sec;
2707 loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
2708 VERIFY(nvlist_add_uint64_array(*config,
2709 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
2711 VERIFY(nvlist_add_uint64(*config,
2712 ZPOOL_CONFIG_ERRCOUNT,
2713 spa_get_errlog_size(spa)) == 0);
2715 if (spa_suspended(spa))
2716 VERIFY(nvlist_add_uint64(*config,
2717 ZPOOL_CONFIG_SUSPENDED,
2718 spa->spa_failmode) == 0);
2720 spa_add_spares(spa, *config);
2721 spa_add_l2cache(spa, *config);
2726 * We want to get the alternate root even for faulted pools, so we cheat
2727 * and call spa_lookup() directly.
2731 mutex_enter(&spa_namespace_lock);
2732 spa = spa_lookup(name);
2734 spa_altroot(spa, altroot, buflen);
2738 mutex_exit(&spa_namespace_lock);
2740 spa_altroot(spa, altroot, buflen);
2745 spa_config_exit(spa, SCL_CONFIG, FTAG);
2746 spa_close(spa, FTAG);
2753 * Validate that the auxiliary device array is well formed. We must have an
2754 * array of nvlists, each which describes a valid leaf vdev. If this is an
2755 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
2756 * specified, as long as they are well-formed.
2759 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
2760 spa_aux_vdev_t *sav, const char *config, uint64_t version,
2761 vdev_labeltype_t label)
2768 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
2771 * It's acceptable to have no devs specified.
2773 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
2780 * Make sure the pool is formatted with a version that supports this
2783 if (spa_version(spa) < version)
2787 * Set the pending device list so we correctly handle device in-use
2790 sav->sav_pending = dev;
2791 sav->sav_npending = ndev;
2793 for (i = 0; i < ndev; i++) {
2794 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
2798 if (!vd->vdev_ops->vdev_op_leaf) {
2805 * The L2ARC currently only supports disk devices in
2806 * kernel context. For user-level testing, we allow it.
2809 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
2810 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
2818 if ((error = vdev_open(vd)) == 0 &&
2819 (error = vdev_label_init(vd, crtxg, label)) == 0) {
2820 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
2821 vd->vdev_guid) == 0);
2827 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
2834 sav->sav_pending = NULL;
2835 sav->sav_npending = 0;
2840 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
2844 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
2846 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
2847 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
2848 VDEV_LABEL_SPARE)) != 0) {
2852 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
2853 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
2854 VDEV_LABEL_L2CACHE));
2858 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
2863 if (sav->sav_config != NULL) {
2869 * Generate new dev list by concatentating with the
2872 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
2873 &olddevs, &oldndevs) == 0);
2875 newdevs = kmem_alloc(sizeof (void *) *
2876 (ndevs + oldndevs), KM_SLEEP);
2877 for (i = 0; i < oldndevs; i++)
2878 VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
2880 for (i = 0; i < ndevs; i++)
2881 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
2884 VERIFY(nvlist_remove(sav->sav_config, config,
2885 DATA_TYPE_NVLIST_ARRAY) == 0);
2887 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
2888 config, newdevs, ndevs + oldndevs) == 0);
2889 for (i = 0; i < oldndevs + ndevs; i++)
2890 nvlist_free(newdevs[i]);
2891 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
2894 * Generate a new dev list.
2896 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
2898 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
2904 * Stop and drop level 2 ARC devices
2907 spa_l2cache_drop(spa_t *spa)
2911 spa_aux_vdev_t *sav = &spa->spa_l2cache;
2913 for (i = 0; i < sav->sav_count; i++) {
2916 vd = sav->sav_vdevs[i];
2919 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
2920 pool != 0ULL && l2arc_vdev_present(vd))
2921 l2arc_remove_vdev(vd);
2929 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
2930 const char *history_str, nvlist_t *zplprops)
2933 char *altroot = NULL;
2938 uint64_t txg = TXG_INITIAL;
2939 nvlist_t **spares, **l2cache;
2940 uint_t nspares, nl2cache;
2941 uint64_t version, obj;
2945 * If this pool already exists, return failure.
2947 mutex_enter(&spa_namespace_lock);
2948 if (spa_lookup(pool) != NULL) {
2949 mutex_exit(&spa_namespace_lock);
2954 * Allocate a new spa_t structure.
2956 (void) nvlist_lookup_string(props,
2957 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
2958 spa = spa_add(pool, NULL, altroot);
2959 spa_activate(spa, spa_mode_global);
2961 if (props && (error = spa_prop_validate(spa, props))) {
2962 spa_deactivate(spa);
2964 mutex_exit(&spa_namespace_lock);
2968 if (nvlist_lookup_uint64(props, zpool_prop_to_name(ZPOOL_PROP_VERSION),
2970 version = SPA_VERSION;
2971 ASSERT(version <= SPA_VERSION);
2973 spa->spa_first_txg = txg;
2974 spa->spa_uberblock.ub_txg = txg - 1;
2975 spa->spa_uberblock.ub_version = version;
2976 spa->spa_ubsync = spa->spa_uberblock;
2979 * Create "The Godfather" zio to hold all async IOs
2981 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2982 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2985 * Create the root vdev.
2987 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2989 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
2991 ASSERT(error != 0 || rvd != NULL);
2992 ASSERT(error != 0 || spa->spa_root_vdev == rvd);
2994 if (error == 0 && !zfs_allocatable_devs(nvroot))
2998 (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
2999 (error = spa_validate_aux(spa, nvroot, txg,
3000 VDEV_ALLOC_ADD)) == 0) {
3001 for (c = 0; c < rvd->vdev_children; c++) {
3002 vdev_metaslab_set_size(rvd->vdev_child[c]);
3003 vdev_expand(rvd->vdev_child[c], txg);
3007 spa_config_exit(spa, SCL_ALL, FTAG);
3011 spa_deactivate(spa);
3013 mutex_exit(&spa_namespace_lock);
3018 * Get the list of spares, if specified.
3020 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3021 &spares, &nspares) == 0) {
3022 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3024 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3025 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3026 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3027 spa_load_spares(spa);
3028 spa_config_exit(spa, SCL_ALL, FTAG);
3029 spa->spa_spares.sav_sync = B_TRUE;
3033 * Get the list of level 2 cache devices, if specified.
3035 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3036 &l2cache, &nl2cache) == 0) {
3037 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3038 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3039 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3040 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3041 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3042 spa_load_l2cache(spa);
3043 spa_config_exit(spa, SCL_ALL, FTAG);
3044 spa->spa_l2cache.sav_sync = B_TRUE;
3047 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3048 spa->spa_meta_objset = dp->dp_meta_objset;
3051 * Create DDTs (dedup tables).
3055 spa_update_dspace(spa);
3057 tx = dmu_tx_create_assigned(dp, txg);
3060 * Create the pool config object.
3062 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3063 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3064 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3066 if (zap_add(spa->spa_meta_objset,
3067 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3068 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3069 cmn_err(CE_PANIC, "failed to add pool config");
3072 if (zap_add(spa->spa_meta_objset,
3073 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3074 sizeof (uint64_t), 1, &version, tx) != 0) {
3075 cmn_err(CE_PANIC, "failed to add pool version");
3078 /* Newly created pools with the right version are always deflated. */
3079 if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3080 spa->spa_deflate = TRUE;
3081 if (zap_add(spa->spa_meta_objset,
3082 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3083 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3084 cmn_err(CE_PANIC, "failed to add deflate");
3089 * Create the deferred-free bpobj. Turn off compression
3090 * because sync-to-convergence takes longer if the blocksize
3093 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3094 dmu_object_set_compress(spa->spa_meta_objset, obj,
3095 ZIO_COMPRESS_OFF, tx);
3096 if (zap_add(spa->spa_meta_objset,
3097 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3098 sizeof (uint64_t), 1, &obj, tx) != 0) {
3099 cmn_err(CE_PANIC, "failed to add bpobj");
3101 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3102 spa->spa_meta_objset, obj));
3105 * Create the pool's history object.
3107 if (version >= SPA_VERSION_ZPOOL_HISTORY)
3108 spa_history_create_obj(spa, tx);
3111 * Set pool properties.
3113 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3114 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3115 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3116 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3118 if (props != NULL) {
3119 spa_configfile_set(spa, props, B_FALSE);
3120 spa_sync_props(spa, props, tx);
3125 spa->spa_sync_on = B_TRUE;
3126 txg_sync_start(spa->spa_dsl_pool);
3129 * We explicitly wait for the first transaction to complete so that our
3130 * bean counters are appropriately updated.
3132 txg_wait_synced(spa->spa_dsl_pool, txg);
3134 spa_config_sync(spa, B_FALSE, B_TRUE);
3136 if (version >= SPA_VERSION_ZPOOL_HISTORY && history_str != NULL)
3137 (void) spa_history_log(spa, history_str, LOG_CMD_POOL_CREATE);
3138 spa_history_log_version(spa, LOG_POOL_CREATE);
3140 spa->spa_minref = refcount_count(&spa->spa_refcount);
3142 mutex_exit(&spa_namespace_lock);
3149 * Get the root pool information from the root disk, then import the root pool
3150 * during the system boot up time.
3152 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3155 spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3158 nvlist_t *nvtop, *nvroot;
3161 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3165 * Add this top-level vdev to the child array.
3167 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3169 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3171 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3174 * Put this pool's top-level vdevs into a root vdev.
3176 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3177 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3178 VDEV_TYPE_ROOT) == 0);
3179 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3180 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3181 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3185 * Replace the existing vdev_tree with the new root vdev in
3186 * this pool's configuration (remove the old, add the new).
3188 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3189 nvlist_free(nvroot);
3194 * Walk the vdev tree and see if we can find a device with "better"
3195 * configuration. A configuration is "better" if the label on that
3196 * device has a more recent txg.
3199 spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3203 for (c = 0; c < vd->vdev_children; c++)
3204 spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3206 if (vd->vdev_ops->vdev_op_leaf) {
3210 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3214 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3218 * Do we have a better boot device?
3220 if (label_txg > *txg) {
3229 * Import a root pool.
3231 * For x86. devpath_list will consist of devid and/or physpath name of
3232 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3233 * The GRUB "findroot" command will return the vdev we should boot.
3235 * For Sparc, devpath_list consists the physpath name of the booting device
3236 * no matter the rootpool is a single device pool or a mirrored pool.
3238 * "/pci@1f,0/ide@d/disk@0,0:a"
3241 spa_import_rootpool(char *devpath, char *devid)
3244 vdev_t *rvd, *bvd, *avd = NULL;
3245 nvlist_t *config, *nvtop;
3251 * Read the label from the boot device and generate a configuration.
3253 config = spa_generate_rootconf(devpath, devid, &guid);
3254 #if defined(_OBP) && defined(_KERNEL)
3255 if (config == NULL) {
3256 if (strstr(devpath, "/iscsi/ssd") != NULL) {
3258 get_iscsi_bootpath_phy(devpath);
3259 config = spa_generate_rootconf(devpath, devid, &guid);
3263 if (config == NULL) {
3264 cmn_err(CE_NOTE, "Can not read the pool label from '%s'",
3269 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3271 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3273 mutex_enter(&spa_namespace_lock);
3274 if ((spa = spa_lookup(pname)) != NULL) {
3276 * Remove the existing root pool from the namespace so that we
3277 * can replace it with the correct config we just read in.
3282 spa = spa_add(pname, config, NULL);
3283 spa->spa_is_root = B_TRUE;
3284 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3287 * Build up a vdev tree based on the boot device's label config.
3289 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3291 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3292 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3293 VDEV_ALLOC_ROOTPOOL);
3294 spa_config_exit(spa, SCL_ALL, FTAG);
3296 mutex_exit(&spa_namespace_lock);
3297 nvlist_free(config);
3298 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3304 * Get the boot vdev.
3306 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3307 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3308 (u_longlong_t)guid);
3314 * Determine if there is a better boot device.
3317 spa_alt_rootvdev(rvd, &avd, &txg);
3319 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3320 "try booting from '%s'", avd->vdev_path);
3326 * If the boot device is part of a spare vdev then ensure that
3327 * we're booting off the active spare.
3329 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3330 !bvd->vdev_isspare) {
3331 cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3332 "try booting from '%s'",
3334 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3340 spa_history_log_version(spa, LOG_POOL_IMPORT);
3342 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3344 spa_config_exit(spa, SCL_ALL, FTAG);
3345 mutex_exit(&spa_namespace_lock);
3347 nvlist_free(config);
3354 * Import a non-root pool into the system.
3357 spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
3360 char *altroot = NULL;
3361 spa_load_state_t state = SPA_LOAD_IMPORT;
3362 zpool_rewind_policy_t policy;
3363 uint64_t mode = spa_mode_global;
3364 uint64_t readonly = B_FALSE;
3367 nvlist_t **spares, **l2cache;
3368 uint_t nspares, nl2cache;
3371 * If a pool with this name exists, return failure.
3373 mutex_enter(&spa_namespace_lock);
3374 if (spa_lookup(pool) != NULL) {
3375 mutex_exit(&spa_namespace_lock);
3380 * Create and initialize the spa structure.
3382 (void) nvlist_lookup_string(props,
3383 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3384 (void) nvlist_lookup_uint64(props,
3385 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
3388 spa = spa_add(pool, config, altroot);
3389 spa->spa_import_flags = flags;
3392 * Verbatim import - Take a pool and insert it into the namespace
3393 * as if it had been loaded at boot.
3395 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
3397 spa_configfile_set(spa, props, B_FALSE);
3399 spa_config_sync(spa, B_FALSE, B_TRUE);
3401 mutex_exit(&spa_namespace_lock);
3402 spa_history_log_version(spa, LOG_POOL_IMPORT);
3407 spa_activate(spa, mode);
3410 * Don't start async tasks until we know everything is healthy.
3412 spa_async_suspend(spa);
3414 zpool_get_rewind_policy(config, &policy);
3415 if (policy.zrp_request & ZPOOL_DO_REWIND)
3416 state = SPA_LOAD_RECOVER;
3419 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
3420 * because the user-supplied config is actually the one to trust when
3423 if (state != SPA_LOAD_RECOVER)
3424 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3426 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
3427 policy.zrp_request);
3430 * Propagate anything learned while loading the pool and pass it
3431 * back to caller (i.e. rewind info, missing devices, etc).
3433 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
3434 spa->spa_load_info) == 0);
3436 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3438 * Toss any existing sparelist, as it doesn't have any validity
3439 * anymore, and conflicts with spa_has_spare().
3441 if (spa->spa_spares.sav_config) {
3442 nvlist_free(spa->spa_spares.sav_config);
3443 spa->spa_spares.sav_config = NULL;
3444 spa_load_spares(spa);
3446 if (spa->spa_l2cache.sav_config) {
3447 nvlist_free(spa->spa_l2cache.sav_config);
3448 spa->spa_l2cache.sav_config = NULL;
3449 spa_load_l2cache(spa);
3452 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3455 error = spa_validate_aux(spa, nvroot, -1ULL,
3458 error = spa_validate_aux(spa, nvroot, -1ULL,
3459 VDEV_ALLOC_L2CACHE);
3460 spa_config_exit(spa, SCL_ALL, FTAG);
3463 spa_configfile_set(spa, props, B_FALSE);
3465 if (error != 0 || (props && spa_writeable(spa) &&
3466 (error = spa_prop_set(spa, props)))) {
3468 spa_deactivate(spa);
3470 mutex_exit(&spa_namespace_lock);
3474 spa_async_resume(spa);
3477 * Override any spares and level 2 cache devices as specified by
3478 * the user, as these may have correct device names/devids, etc.
3480 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3481 &spares, &nspares) == 0) {
3482 if (spa->spa_spares.sav_config)
3483 VERIFY(nvlist_remove(spa->spa_spares.sav_config,
3484 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
3486 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
3487 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3488 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3489 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3490 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3491 spa_load_spares(spa);
3492 spa_config_exit(spa, SCL_ALL, FTAG);
3493 spa->spa_spares.sav_sync = B_TRUE;
3495 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3496 &l2cache, &nl2cache) == 0) {
3497 if (spa->spa_l2cache.sav_config)
3498 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
3499 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
3501 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3502 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3503 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3504 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3505 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3506 spa_load_l2cache(spa);
3507 spa_config_exit(spa, SCL_ALL, FTAG);
3508 spa->spa_l2cache.sav_sync = B_TRUE;
3512 * Check for any removed devices.
3514 if (spa->spa_autoreplace) {
3515 spa_aux_check_removed(&spa->spa_spares);
3516 spa_aux_check_removed(&spa->spa_l2cache);
3519 if (spa_writeable(spa)) {
3521 * Update the config cache to include the newly-imported pool.
3523 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3527 * It's possible that the pool was expanded while it was exported.
3528 * We kick off an async task to handle this for us.
3530 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
3532 mutex_exit(&spa_namespace_lock);
3533 spa_history_log_version(spa, LOG_POOL_IMPORT);
3539 spa_tryimport(nvlist_t *tryconfig)
3541 nvlist_t *config = NULL;
3547 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
3550 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
3554 * Create and initialize the spa structure.
3556 mutex_enter(&spa_namespace_lock);
3557 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
3558 spa_activate(spa, FREAD);
3561 * Pass off the heavy lifting to spa_load().
3562 * Pass TRUE for mosconfig because the user-supplied config
3563 * is actually the one to trust when doing an import.
3565 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
3568 * If 'tryconfig' was at least parsable, return the current config.
3570 if (spa->spa_root_vdev != NULL) {
3571 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
3572 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
3574 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
3576 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
3577 spa->spa_uberblock.ub_timestamp) == 0);
3580 * If the bootfs property exists on this pool then we
3581 * copy it out so that external consumers can tell which
3582 * pools are bootable.
3584 if ((!error || error == EEXIST) && spa->spa_bootfs) {
3585 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3588 * We have to play games with the name since the
3589 * pool was opened as TRYIMPORT_NAME.
3591 if (dsl_dsobj_to_dsname(spa_name(spa),
3592 spa->spa_bootfs, tmpname) == 0) {
3594 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3596 cp = strchr(tmpname, '/');
3598 (void) strlcpy(dsname, tmpname,
3601 (void) snprintf(dsname, MAXPATHLEN,
3602 "%s/%s", poolname, ++cp);
3604 VERIFY(nvlist_add_string(config,
3605 ZPOOL_CONFIG_BOOTFS, dsname) == 0);
3606 kmem_free(dsname, MAXPATHLEN);
3608 kmem_free(tmpname, MAXPATHLEN);
3612 * Add the list of hot spares and level 2 cache devices.
3614 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3615 spa_add_spares(spa, config);
3616 spa_add_l2cache(spa, config);
3617 spa_config_exit(spa, SCL_CONFIG, FTAG);
3621 spa_deactivate(spa);
3623 mutex_exit(&spa_namespace_lock);
3629 * Pool export/destroy
3631 * The act of destroying or exporting a pool is very simple. We make sure there
3632 * is no more pending I/O and any references to the pool are gone. Then, we
3633 * update the pool state and sync all the labels to disk, removing the
3634 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
3635 * we don't sync the labels or remove the configuration cache.
3638 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
3639 boolean_t force, boolean_t hardforce)
3646 if (!(spa_mode_global & FWRITE))
3649 mutex_enter(&spa_namespace_lock);
3650 if ((spa = spa_lookup(pool)) == NULL) {
3651 mutex_exit(&spa_namespace_lock);
3656 * Put a hold on the pool, drop the namespace lock, stop async tasks,
3657 * reacquire the namespace lock, and see if we can export.
3659 spa_open_ref(spa, FTAG);
3660 mutex_exit(&spa_namespace_lock);
3661 spa_async_suspend(spa);
3662 mutex_enter(&spa_namespace_lock);
3663 spa_close(spa, FTAG);
3666 * The pool will be in core if it's openable,
3667 * in which case we can modify its state.
3669 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
3671 * Objsets may be open only because they're dirty, so we
3672 * have to force it to sync before checking spa_refcnt.
3674 txg_wait_synced(spa->spa_dsl_pool, 0);
3677 * A pool cannot be exported or destroyed if there are active
3678 * references. If we are resetting a pool, allow references by
3679 * fault injection handlers.
3681 if (!spa_refcount_zero(spa) ||
3682 (spa->spa_inject_ref != 0 &&
3683 new_state != POOL_STATE_UNINITIALIZED)) {
3684 spa_async_resume(spa);
3685 mutex_exit(&spa_namespace_lock);
3690 * A pool cannot be exported if it has an active shared spare.
3691 * This is to prevent other pools stealing the active spare
3692 * from an exported pool. At user's own will, such pool can
3693 * be forcedly exported.
3695 if (!force && new_state == POOL_STATE_EXPORTED &&
3696 spa_has_active_shared_spare(spa)) {
3697 spa_async_resume(spa);
3698 mutex_exit(&spa_namespace_lock);
3703 * We want this to be reflected on every label,
3704 * so mark them all dirty. spa_unload() will do the
3705 * final sync that pushes these changes out.
3707 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
3708 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3709 spa->spa_state = new_state;
3710 spa->spa_final_txg = spa_last_synced_txg(spa) +
3712 vdev_config_dirty(spa->spa_root_vdev);
3713 spa_config_exit(spa, SCL_ALL, FTAG);
3717 spa_event_notify(spa, NULL, FM_EREPORT_ZFS_POOL_DESTROY);
3719 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
3721 spa_deactivate(spa);
3724 if (oldconfig && spa->spa_config)
3725 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
3727 if (new_state != POOL_STATE_UNINITIALIZED) {
3729 spa_config_sync(spa, B_TRUE, B_TRUE);
3732 mutex_exit(&spa_namespace_lock);
3738 * Destroy a storage pool.
3741 spa_destroy(char *pool)
3743 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
3748 * Export a storage pool.
3751 spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
3752 boolean_t hardforce)
3754 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
3759 * Similar to spa_export(), this unloads the spa_t without actually removing it
3760 * from the namespace in any way.
3763 spa_reset(char *pool)
3765 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
3770 * ==========================================================================
3771 * Device manipulation
3772 * ==========================================================================
3776 * Add a device to a storage pool.
3779 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
3783 vdev_t *rvd = spa->spa_root_vdev;
3785 nvlist_t **spares, **l2cache;
3786 uint_t nspares, nl2cache;
3789 ASSERT(spa_writeable(spa));
3791 txg = spa_vdev_enter(spa);
3793 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
3794 VDEV_ALLOC_ADD)) != 0)
3795 return (spa_vdev_exit(spa, NULL, txg, error));
3797 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */
3799 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
3803 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
3807 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
3808 return (spa_vdev_exit(spa, vd, txg, EINVAL));
3810 if (vd->vdev_children != 0 &&
3811 (error = vdev_create(vd, txg, B_FALSE)) != 0)
3812 return (spa_vdev_exit(spa, vd, txg, error));
3815 * We must validate the spares and l2cache devices after checking the
3816 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
3818 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
3819 return (spa_vdev_exit(spa, vd, txg, error));
3822 * Transfer each new top-level vdev from vd to rvd.
3824 for (c = 0; c < vd->vdev_children; c++) {
3827 * Set the vdev id to the first hole, if one exists.
3829 for (id = 0; id < rvd->vdev_children; id++) {
3830 if (rvd->vdev_child[id]->vdev_ishole) {
3831 vdev_free(rvd->vdev_child[id]);
3835 tvd = vd->vdev_child[c];
3836 vdev_remove_child(vd, tvd);
3838 vdev_add_child(rvd, tvd);
3839 vdev_config_dirty(tvd);
3843 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
3844 ZPOOL_CONFIG_SPARES);
3845 spa_load_spares(spa);
3846 spa->spa_spares.sav_sync = B_TRUE;
3849 if (nl2cache != 0) {
3850 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
3851 ZPOOL_CONFIG_L2CACHE);
3852 spa_load_l2cache(spa);
3853 spa->spa_l2cache.sav_sync = B_TRUE;
3857 * We have to be careful when adding new vdevs to an existing pool.
3858 * If other threads start allocating from these vdevs before we
3859 * sync the config cache, and we lose power, then upon reboot we may
3860 * fail to open the pool because there are DVAs that the config cache
3861 * can't translate. Therefore, we first add the vdevs without
3862 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
3863 * and then let spa_config_update() initialize the new metaslabs.
3865 * spa_load() checks for added-but-not-initialized vdevs, so that
3866 * if we lose power at any point in this sequence, the remaining
3867 * steps will be completed the next time we load the pool.
3869 (void) spa_vdev_exit(spa, vd, txg, 0);
3871 mutex_enter(&spa_namespace_lock);
3872 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3873 mutex_exit(&spa_namespace_lock);
3879 * Attach a device to a mirror. The arguments are the path to any device
3880 * in the mirror, and the nvroot for the new device. If the path specifies
3881 * a device that is not mirrored, we automatically insert the mirror vdev.
3883 * If 'replacing' is specified, the new device is intended to replace the
3884 * existing device; in this case the two devices are made into their own
3885 * mirror using the 'replacing' vdev, which is functionally identical to
3886 * the mirror vdev (it actually reuses all the same ops) but has a few
3887 * extra rules: you can't attach to it after it's been created, and upon
3888 * completion of resilvering, the first disk (the one being replaced)
3889 * is automatically detached.
3892 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
3894 uint64_t txg, dtl_max_txg;
3895 ASSERTV(vdev_t *rvd = spa->spa_root_vdev;)
3896 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
3898 char *oldvdpath, *newvdpath;
3902 ASSERT(spa_writeable(spa));
3904 txg = spa_vdev_enter(spa);
3906 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
3909 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
3911 if (!oldvd->vdev_ops->vdev_op_leaf)
3912 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3914 pvd = oldvd->vdev_parent;
3916 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
3917 VDEV_ALLOC_ATTACH)) != 0)
3918 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
3920 if (newrootvd->vdev_children != 1)
3921 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
3923 newvd = newrootvd->vdev_child[0];
3925 if (!newvd->vdev_ops->vdev_op_leaf)
3926 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
3928 if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
3929 return (spa_vdev_exit(spa, newrootvd, txg, error));
3932 * Spares can't replace logs
3934 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
3935 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3939 * For attach, the only allowable parent is a mirror or the root
3942 if (pvd->vdev_ops != &vdev_mirror_ops &&
3943 pvd->vdev_ops != &vdev_root_ops)
3944 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3946 pvops = &vdev_mirror_ops;
3949 * Active hot spares can only be replaced by inactive hot
3952 if (pvd->vdev_ops == &vdev_spare_ops &&
3953 oldvd->vdev_isspare &&
3954 !spa_has_spare(spa, newvd->vdev_guid))
3955 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3958 * If the source is a hot spare, and the parent isn't already a
3959 * spare, then we want to create a new hot spare. Otherwise, we
3960 * want to create a replacing vdev. The user is not allowed to
3961 * attach to a spared vdev child unless the 'isspare' state is
3962 * the same (spare replaces spare, non-spare replaces
3965 if (pvd->vdev_ops == &vdev_replacing_ops &&
3966 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
3967 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3968 } else if (pvd->vdev_ops == &vdev_spare_ops &&
3969 newvd->vdev_isspare != oldvd->vdev_isspare) {
3970 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3973 if (newvd->vdev_isspare)
3974 pvops = &vdev_spare_ops;
3976 pvops = &vdev_replacing_ops;
3980 * Make sure the new device is big enough.
3982 if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
3983 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
3986 * The new device cannot have a higher alignment requirement
3987 * than the top-level vdev.
3989 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
3990 return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
3993 * If this is an in-place replacement, update oldvd's path and devid
3994 * to make it distinguishable from newvd, and unopenable from now on.
3996 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
3997 spa_strfree(oldvd->vdev_path);
3998 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4000 (void) sprintf(oldvd->vdev_path, "%s/%s",
4001 newvd->vdev_path, "old");
4002 if (oldvd->vdev_devid != NULL) {
4003 spa_strfree(oldvd->vdev_devid);
4004 oldvd->vdev_devid = NULL;
4008 /* mark the device being resilvered */
4009 newvd->vdev_resilvering = B_TRUE;
4012 * If the parent is not a mirror, or if we're replacing, insert the new
4013 * mirror/replacing/spare vdev above oldvd.
4015 if (pvd->vdev_ops != pvops)
4016 pvd = vdev_add_parent(oldvd, pvops);
4018 ASSERT(pvd->vdev_top->vdev_parent == rvd);
4019 ASSERT(pvd->vdev_ops == pvops);
4020 ASSERT(oldvd->vdev_parent == pvd);
4023 * Extract the new device from its root and add it to pvd.
4025 vdev_remove_child(newrootvd, newvd);
4026 newvd->vdev_id = pvd->vdev_children;
4027 newvd->vdev_crtxg = oldvd->vdev_crtxg;
4028 vdev_add_child(pvd, newvd);
4030 tvd = newvd->vdev_top;
4031 ASSERT(pvd->vdev_top == tvd);
4032 ASSERT(tvd->vdev_parent == rvd);
4034 vdev_config_dirty(tvd);
4037 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4038 * for any dmu_sync-ed blocks. It will propagate upward when
4039 * spa_vdev_exit() calls vdev_dtl_reassess().
4041 dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4043 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4044 dtl_max_txg - TXG_INITIAL);
4046 if (newvd->vdev_isspare) {
4047 spa_spare_activate(newvd);
4048 spa_event_notify(spa, newvd, FM_EREPORT_ZFS_DEVICE_SPARE);
4051 oldvdpath = spa_strdup(oldvd->vdev_path);
4052 newvdpath = spa_strdup(newvd->vdev_path);
4053 newvd_isspare = newvd->vdev_isspare;
4056 * Mark newvd's DTL dirty in this txg.
4058 vdev_dirty(tvd, VDD_DTL, newvd, txg);
4061 * Restart the resilver
4063 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4068 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4070 spa_history_log_internal(LOG_POOL_VDEV_ATTACH, spa, NULL,
4071 "%s vdev=%s %s vdev=%s",
4072 replacing && newvd_isspare ? "spare in" :
4073 replacing ? "replace" : "attach", newvdpath,
4074 replacing ? "for" : "to", oldvdpath);
4076 spa_strfree(oldvdpath);
4077 spa_strfree(newvdpath);
4079 if (spa->spa_bootfs)
4080 spa_event_notify(spa, newvd, FM_EREPORT_ZFS_BOOTFS_VDEV_ATTACH);
4086 * Detach a device from a mirror or replacing vdev.
4087 * If 'replace_done' is specified, only detach if the parent
4088 * is a replacing vdev.
4091 spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4095 ASSERTV(vdev_t *rvd = spa->spa_root_vdev;)
4096 vdev_t *vd, *pvd, *cvd, *tvd;
4097 boolean_t unspare = B_FALSE;
4098 uint64_t unspare_guid = 0;
4102 ASSERT(spa_writeable(spa));
4104 txg = spa_vdev_enter(spa);
4106 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4109 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4111 if (!vd->vdev_ops->vdev_op_leaf)
4112 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4114 pvd = vd->vdev_parent;
4117 * If the parent/child relationship is not as expected, don't do it.
4118 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4119 * vdev that's replacing B with C. The user's intent in replacing
4120 * is to go from M(A,B) to M(A,C). If the user decides to cancel
4121 * the replace by detaching C, the expected behavior is to end up
4122 * M(A,B). But suppose that right after deciding to detach C,
4123 * the replacement of B completes. We would have M(A,C), and then
4124 * ask to detach C, which would leave us with just A -- not what
4125 * the user wanted. To prevent this, we make sure that the
4126 * parent/child relationship hasn't changed -- in this example,
4127 * that C's parent is still the replacing vdev R.
4129 if (pvd->vdev_guid != pguid && pguid != 0)
4130 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4133 * Only 'replacing' or 'spare' vdevs can be replaced.
4135 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4136 pvd->vdev_ops != &vdev_spare_ops)
4137 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4139 ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4140 spa_version(spa) >= SPA_VERSION_SPARES);
4143 * Only mirror, replacing, and spare vdevs support detach.
4145 if (pvd->vdev_ops != &vdev_replacing_ops &&
4146 pvd->vdev_ops != &vdev_mirror_ops &&
4147 pvd->vdev_ops != &vdev_spare_ops)
4148 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4151 * If this device has the only valid copy of some data,
4152 * we cannot safely detach it.
4154 if (vdev_dtl_required(vd))
4155 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4157 ASSERT(pvd->vdev_children >= 2);
4160 * If we are detaching the second disk from a replacing vdev, then
4161 * check to see if we changed the original vdev's path to have "/old"
4162 * at the end in spa_vdev_attach(). If so, undo that change now.
4164 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4165 vd->vdev_path != NULL) {
4166 size_t len = strlen(vd->vdev_path);
4168 for (c = 0; c < pvd->vdev_children; c++) {
4169 cvd = pvd->vdev_child[c];
4171 if (cvd == vd || cvd->vdev_path == NULL)
4174 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4175 strcmp(cvd->vdev_path + len, "/old") == 0) {
4176 spa_strfree(cvd->vdev_path);
4177 cvd->vdev_path = spa_strdup(vd->vdev_path);
4184 * If we are detaching the original disk from a spare, then it implies
4185 * that the spare should become a real disk, and be removed from the
4186 * active spare list for the pool.
4188 if (pvd->vdev_ops == &vdev_spare_ops &&
4190 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4194 * Erase the disk labels so the disk can be used for other things.
4195 * This must be done after all other error cases are handled,
4196 * but before we disembowel vd (so we can still do I/O to it).
4197 * But if we can't do it, don't treat the error as fatal --
4198 * it may be that the unwritability of the disk is the reason
4199 * it's being detached!
4201 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4204 * Remove vd from its parent and compact the parent's children.
4206 vdev_remove_child(pvd, vd);
4207 vdev_compact_children(pvd);
4210 * Remember one of the remaining children so we can get tvd below.
4212 cvd = pvd->vdev_child[pvd->vdev_children - 1];
4215 * If we need to remove the remaining child from the list of hot spares,
4216 * do it now, marking the vdev as no longer a spare in the process.
4217 * We must do this before vdev_remove_parent(), because that can
4218 * change the GUID if it creates a new toplevel GUID. For a similar
4219 * reason, we must remove the spare now, in the same txg as the detach;
4220 * otherwise someone could attach a new sibling, change the GUID, and
4221 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4224 ASSERT(cvd->vdev_isspare);
4225 spa_spare_remove(cvd);
4226 unspare_guid = cvd->vdev_guid;
4227 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4228 cvd->vdev_unspare = B_TRUE;
4232 * If the parent mirror/replacing vdev only has one child,
4233 * the parent is no longer needed. Remove it from the tree.
4235 if (pvd->vdev_children == 1) {
4236 if (pvd->vdev_ops == &vdev_spare_ops)
4237 cvd->vdev_unspare = B_FALSE;
4238 vdev_remove_parent(cvd);
4239 cvd->vdev_resilvering = B_FALSE;
4244 * We don't set tvd until now because the parent we just removed
4245 * may have been the previous top-level vdev.
4247 tvd = cvd->vdev_top;
4248 ASSERT(tvd->vdev_parent == rvd);
4251 * Reevaluate the parent vdev state.
4253 vdev_propagate_state(cvd);
4256 * If the 'autoexpand' property is set on the pool then automatically
4257 * try to expand the size of the pool. For example if the device we
4258 * just detached was smaller than the others, it may be possible to
4259 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4260 * first so that we can obtain the updated sizes of the leaf vdevs.
4262 if (spa->spa_autoexpand) {
4264 vdev_expand(tvd, txg);
4267 vdev_config_dirty(tvd);
4270 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
4271 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4272 * But first make sure we're not on any *other* txg's DTL list, to
4273 * prevent vd from being accessed after it's freed.
4275 vdpath = spa_strdup(vd->vdev_path);
4276 for (t = 0; t < TXG_SIZE; t++)
4277 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4278 vd->vdev_detached = B_TRUE;
4279 vdev_dirty(tvd, VDD_DTL, vd, txg);
4281 spa_event_notify(spa, vd, FM_EREPORT_ZFS_DEVICE_REMOVE);
4283 /* hang on to the spa before we release the lock */
4284 spa_open_ref(spa, FTAG);
4286 error = spa_vdev_exit(spa, vd, txg, 0);
4288 spa_history_log_internal(LOG_POOL_VDEV_DETACH, spa, NULL,
4290 spa_strfree(vdpath);
4293 * If this was the removal of the original device in a hot spare vdev,
4294 * then we want to go through and remove the device from the hot spare
4295 * list of every other pool.
4298 spa_t *altspa = NULL;
4300 mutex_enter(&spa_namespace_lock);
4301 while ((altspa = spa_next(altspa)) != NULL) {
4302 if (altspa->spa_state != POOL_STATE_ACTIVE ||
4306 spa_open_ref(altspa, FTAG);
4307 mutex_exit(&spa_namespace_lock);
4308 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
4309 mutex_enter(&spa_namespace_lock);
4310 spa_close(altspa, FTAG);
4312 mutex_exit(&spa_namespace_lock);
4314 /* search the rest of the vdevs for spares to remove */
4315 spa_vdev_resilver_done(spa);
4318 /* all done with the spa; OK to release */
4319 mutex_enter(&spa_namespace_lock);
4320 spa_close(spa, FTAG);
4321 mutex_exit(&spa_namespace_lock);
4327 * Split a set of devices from their mirrors, and create a new pool from them.
4330 spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
4331 nvlist_t *props, boolean_t exp)
4334 uint64_t txg, *glist;
4336 uint_t c, children, lastlog;
4337 nvlist_t **child, *nvl, *tmp;
4339 char *altroot = NULL;
4340 vdev_t *rvd, **vml = NULL; /* vdev modify list */
4341 boolean_t activate_slog;
4343 ASSERT(spa_writeable(spa));
4345 txg = spa_vdev_enter(spa);
4347 /* clear the log and flush everything up to now */
4348 activate_slog = spa_passivate_log(spa);
4349 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4350 error = spa_offline_log(spa);
4351 txg = spa_vdev_config_enter(spa);
4354 spa_activate_log(spa);
4357 return (spa_vdev_exit(spa, NULL, txg, error));
4359 /* check new spa name before going any further */
4360 if (spa_lookup(newname) != NULL)
4361 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
4364 * scan through all the children to ensure they're all mirrors
4366 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
4367 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
4369 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4371 /* first, check to ensure we've got the right child count */
4372 rvd = spa->spa_root_vdev;
4374 for (c = 0; c < rvd->vdev_children; c++) {
4375 vdev_t *vd = rvd->vdev_child[c];
4377 /* don't count the holes & logs as children */
4378 if (vd->vdev_islog || vd->vdev_ishole) {
4386 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
4387 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4389 /* next, ensure no spare or cache devices are part of the split */
4390 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
4391 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
4392 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4394 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
4395 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
4397 /* then, loop over each vdev and validate it */
4398 for (c = 0; c < children; c++) {
4399 uint64_t is_hole = 0;
4401 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
4405 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
4406 spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
4414 /* which disk is going to be split? */
4415 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
4421 /* look it up in the spa */
4422 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
4423 if (vml[c] == NULL) {
4428 /* make sure there's nothing stopping the split */
4429 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
4430 vml[c]->vdev_islog ||
4431 vml[c]->vdev_ishole ||
4432 vml[c]->vdev_isspare ||
4433 vml[c]->vdev_isl2cache ||
4434 !vdev_writeable(vml[c]) ||
4435 vml[c]->vdev_children != 0 ||
4436 vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
4437 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
4442 if (vdev_dtl_required(vml[c])) {
4447 /* we need certain info from the top level */
4448 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
4449 vml[c]->vdev_top->vdev_ms_array) == 0);
4450 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
4451 vml[c]->vdev_top->vdev_ms_shift) == 0);
4452 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
4453 vml[c]->vdev_top->vdev_asize) == 0);
4454 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
4455 vml[c]->vdev_top->vdev_ashift) == 0);
4459 kmem_free(vml, children * sizeof (vdev_t *));
4460 kmem_free(glist, children * sizeof (uint64_t));
4461 return (spa_vdev_exit(spa, NULL, txg, error));
4464 /* stop writers from using the disks */
4465 for (c = 0; c < children; c++) {
4467 vml[c]->vdev_offline = B_TRUE;
4469 vdev_reopen(spa->spa_root_vdev);
4472 * Temporarily record the splitting vdevs in the spa config. This
4473 * will disappear once the config is regenerated.
4475 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4476 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
4477 glist, children) == 0);
4478 kmem_free(glist, children * sizeof (uint64_t));
4480 mutex_enter(&spa->spa_props_lock);
4481 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
4483 mutex_exit(&spa->spa_props_lock);
4484 spa->spa_config_splitting = nvl;
4485 vdev_config_dirty(spa->spa_root_vdev);
4487 /* configure and create the new pool */
4488 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
4489 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4490 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
4491 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
4492 spa_version(spa)) == 0);
4493 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
4494 spa->spa_config_txg) == 0);
4495 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4496 spa_generate_guid(NULL)) == 0);
4497 (void) nvlist_lookup_string(props,
4498 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4500 /* add the new pool to the namespace */
4501 newspa = spa_add(newname, config, altroot);
4502 newspa->spa_config_txg = spa->spa_config_txg;
4503 spa_set_log_state(newspa, SPA_LOG_CLEAR);
4505 /* release the spa config lock, retaining the namespace lock */
4506 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4508 if (zio_injection_enabled)
4509 zio_handle_panic_injection(spa, FTAG, 1);
4511 spa_activate(newspa, spa_mode_global);
4512 spa_async_suspend(newspa);
4514 /* create the new pool from the disks of the original pool */
4515 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
4519 /* if that worked, generate a real config for the new pool */
4520 if (newspa->spa_root_vdev != NULL) {
4521 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
4522 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4523 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
4524 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
4525 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
4530 if (props != NULL) {
4531 spa_configfile_set(newspa, props, B_FALSE);
4532 error = spa_prop_set(newspa, props);
4537 /* flush everything */
4538 txg = spa_vdev_config_enter(newspa);
4539 vdev_config_dirty(newspa->spa_root_vdev);
4540 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
4542 if (zio_injection_enabled)
4543 zio_handle_panic_injection(spa, FTAG, 2);
4545 spa_async_resume(newspa);
4547 /* finally, update the original pool's config */
4548 txg = spa_vdev_config_enter(spa);
4549 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
4550 error = dmu_tx_assign(tx, TXG_WAIT);
4553 for (c = 0; c < children; c++) {
4554 if (vml[c] != NULL) {
4557 spa_history_log_internal(LOG_POOL_VDEV_DETACH,
4563 vdev_config_dirty(spa->spa_root_vdev);
4564 spa->spa_config_splitting = NULL;
4568 (void) spa_vdev_exit(spa, NULL, txg, 0);
4570 if (zio_injection_enabled)
4571 zio_handle_panic_injection(spa, FTAG, 3);
4573 /* split is complete; log a history record */
4574 spa_history_log_internal(LOG_POOL_SPLIT, newspa, NULL,
4575 "split new pool %s from pool %s", newname, spa_name(spa));
4577 kmem_free(vml, children * sizeof (vdev_t *));
4579 /* if we're not going to mount the filesystems in userland, export */
4581 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
4588 spa_deactivate(newspa);
4591 txg = spa_vdev_config_enter(spa);
4593 /* re-online all offlined disks */
4594 for (c = 0; c < children; c++) {
4596 vml[c]->vdev_offline = B_FALSE;
4598 vdev_reopen(spa->spa_root_vdev);
4600 nvlist_free(spa->spa_config_splitting);
4601 spa->spa_config_splitting = NULL;
4602 (void) spa_vdev_exit(spa, NULL, txg, error);
4604 kmem_free(vml, children * sizeof (vdev_t *));
4609 spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
4613 for (i = 0; i < count; i++) {
4616 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
4619 if (guid == target_guid)
4627 spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
4628 nvlist_t *dev_to_remove)
4630 nvlist_t **newdev = NULL;
4634 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
4636 for (i = 0, j = 0; i < count; i++) {
4637 if (dev[i] == dev_to_remove)
4639 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
4642 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
4643 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
4645 for (i = 0; i < count - 1; i++)
4646 nvlist_free(newdev[i]);
4649 kmem_free(newdev, (count - 1) * sizeof (void *));
4653 * Evacuate the device.
4656 spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
4661 ASSERT(MUTEX_HELD(&spa_namespace_lock));
4662 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
4663 ASSERT(vd == vd->vdev_top);
4666 * Evacuate the device. We don't hold the config lock as writer
4667 * since we need to do I/O but we do keep the
4668 * spa_namespace_lock held. Once this completes the device
4669 * should no longer have any blocks allocated on it.
4671 if (vd->vdev_islog) {
4672 if (vd->vdev_stat.vs_alloc != 0)
4673 error = spa_offline_log(spa);
4682 * The evacuation succeeded. Remove any remaining MOS metadata
4683 * associated with this vdev, and wait for these changes to sync.
4685 ASSERT3U(vd->vdev_stat.vs_alloc, ==, 0);
4686 txg = spa_vdev_config_enter(spa);
4687 vd->vdev_removing = B_TRUE;
4688 vdev_dirty(vd, 0, NULL, txg);
4689 vdev_config_dirty(vd);
4690 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4696 * Complete the removal by cleaning up the namespace.
4699 spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
4701 vdev_t *rvd = spa->spa_root_vdev;
4702 uint64_t id = vd->vdev_id;
4703 boolean_t last_vdev = (id == (rvd->vdev_children - 1));
4705 ASSERT(MUTEX_HELD(&spa_namespace_lock));
4706 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
4707 ASSERT(vd == vd->vdev_top);
4710 * Only remove any devices which are empty.
4712 if (vd->vdev_stat.vs_alloc != 0)
4715 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4717 if (list_link_active(&vd->vdev_state_dirty_node))
4718 vdev_state_clean(vd);
4719 if (list_link_active(&vd->vdev_config_dirty_node))
4720 vdev_config_clean(vd);
4725 vdev_compact_children(rvd);
4727 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
4728 vdev_add_child(rvd, vd);
4730 vdev_config_dirty(rvd);
4733 * Reassess the health of our root vdev.
4739 * Remove a device from the pool -
4741 * Removing a device from the vdev namespace requires several steps
4742 * and can take a significant amount of time. As a result we use
4743 * the spa_vdev_config_[enter/exit] functions which allow us to
4744 * grab and release the spa_config_lock while still holding the namespace
4745 * lock. During each step the configuration is synced out.
4749 * Remove a device from the pool. Currently, this supports removing only hot
4750 * spares, slogs, and level 2 ARC devices.
4753 spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
4756 metaslab_group_t *mg;
4757 nvlist_t **spares, **l2cache, *nv;
4759 uint_t nspares, nl2cache;
4761 boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
4763 ASSERT(spa_writeable(spa));
4766 txg = spa_vdev_enter(spa);
4768 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4770 if (spa->spa_spares.sav_vdevs != NULL &&
4771 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
4772 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
4773 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
4775 * Only remove the hot spare if it's not currently in use
4778 if (vd == NULL || unspare) {
4779 spa_vdev_remove_aux(spa->spa_spares.sav_config,
4780 ZPOOL_CONFIG_SPARES, spares, nspares, nv);
4781 spa_load_spares(spa);
4782 spa->spa_spares.sav_sync = B_TRUE;
4786 } else if (spa->spa_l2cache.sav_vdevs != NULL &&
4787 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
4788 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
4789 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
4791 * Cache devices can always be removed.
4793 spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
4794 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
4795 spa_load_l2cache(spa);
4796 spa->spa_l2cache.sav_sync = B_TRUE;
4797 } else if (vd != NULL && vd->vdev_islog) {
4799 ASSERT(vd == vd->vdev_top);
4802 * XXX - Once we have bp-rewrite this should
4803 * become the common case.
4809 * Stop allocating from this vdev.
4811 metaslab_group_passivate(mg);
4814 * Wait for the youngest allocations and frees to sync,
4815 * and then wait for the deferral of those frees to finish.
4817 spa_vdev_config_exit(spa, NULL,
4818 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
4821 * Attempt to evacuate the vdev.
4823 error = spa_vdev_remove_evacuate(spa, vd);
4825 txg = spa_vdev_config_enter(spa);
4828 * If we couldn't evacuate the vdev, unwind.
4831 metaslab_group_activate(mg);
4832 return (spa_vdev_exit(spa, NULL, txg, error));
4836 * Clean up the vdev namespace.
4838 spa_vdev_remove_from_namespace(spa, vd);
4840 } else if (vd != NULL) {
4842 * Normal vdevs cannot be removed (yet).
4847 * There is no vdev of any kind with the specified guid.
4853 return (spa_vdev_exit(spa, NULL, txg, error));
4859 * Find any device that's done replacing, or a vdev marked 'unspare' that's
4860 * current spared, so we can detach it.
4863 spa_vdev_resilver_done_hunt(vdev_t *vd)
4865 vdev_t *newvd, *oldvd;
4868 for (c = 0; c < vd->vdev_children; c++) {
4869 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
4875 * Check for a completed replacement. We always consider the first
4876 * vdev in the list to be the oldest vdev, and the last one to be
4877 * the newest (see spa_vdev_attach() for how that works). In
4878 * the case where the newest vdev is faulted, we will not automatically
4879 * remove it after a resilver completes. This is OK as it will require
4880 * user intervention to determine which disk the admin wishes to keep.
4882 if (vd->vdev_ops == &vdev_replacing_ops) {
4883 ASSERT(vd->vdev_children > 1);
4885 newvd = vd->vdev_child[vd->vdev_children - 1];
4886 oldvd = vd->vdev_child[0];
4888 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
4889 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
4890 !vdev_dtl_required(oldvd))
4895 * Check for a completed resilver with the 'unspare' flag set.
4897 if (vd->vdev_ops == &vdev_spare_ops) {
4898 vdev_t *first = vd->vdev_child[0];
4899 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
4901 if (last->vdev_unspare) {
4904 } else if (first->vdev_unspare) {
4911 if (oldvd != NULL &&
4912 vdev_dtl_empty(newvd, DTL_MISSING) &&
4913 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
4914 !vdev_dtl_required(oldvd))
4918 * If there are more than two spares attached to a disk,
4919 * and those spares are not required, then we want to
4920 * attempt to free them up now so that they can be used
4921 * by other pools. Once we're back down to a single
4922 * disk+spare, we stop removing them.
4924 if (vd->vdev_children > 2) {
4925 newvd = vd->vdev_child[1];
4927 if (newvd->vdev_isspare && last->vdev_isspare &&
4928 vdev_dtl_empty(last, DTL_MISSING) &&
4929 vdev_dtl_empty(last, DTL_OUTAGE) &&
4930 !vdev_dtl_required(newvd))
4939 spa_vdev_resilver_done(spa_t *spa)
4941 vdev_t *vd, *pvd, *ppvd;
4942 uint64_t guid, sguid, pguid, ppguid;
4944 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4946 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
4947 pvd = vd->vdev_parent;
4948 ppvd = pvd->vdev_parent;
4949 guid = vd->vdev_guid;
4950 pguid = pvd->vdev_guid;
4951 ppguid = ppvd->vdev_guid;
4954 * If we have just finished replacing a hot spared device, then
4955 * we need to detach the parent's first child (the original hot
4958 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
4959 ppvd->vdev_children == 2) {
4960 ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
4961 sguid = ppvd->vdev_child[1]->vdev_guid;
4963 spa_config_exit(spa, SCL_ALL, FTAG);
4964 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
4966 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
4968 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4971 spa_config_exit(spa, SCL_ALL, FTAG);
4975 * Update the stored path or FRU for this vdev.
4978 spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
4982 boolean_t sync = B_FALSE;
4984 ASSERT(spa_writeable(spa));
4986 spa_vdev_state_enter(spa, SCL_ALL);
4988 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
4989 return (spa_vdev_state_exit(spa, NULL, ENOENT));
4991 if (!vd->vdev_ops->vdev_op_leaf)
4992 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
4995 if (strcmp(value, vd->vdev_path) != 0) {
4996 spa_strfree(vd->vdev_path);
4997 vd->vdev_path = spa_strdup(value);
5001 if (vd->vdev_fru == NULL) {
5002 vd->vdev_fru = spa_strdup(value);
5004 } else if (strcmp(value, vd->vdev_fru) != 0) {
5005 spa_strfree(vd->vdev_fru);
5006 vd->vdev_fru = spa_strdup(value);
5011 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
5015 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
5017 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5021 spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5023 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5027 * ==========================================================================
5029 * ==========================================================================
5033 spa_scan_stop(spa_t *spa)
5035 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5036 if (dsl_scan_resilvering(spa->spa_dsl_pool))
5038 return (dsl_scan_cancel(spa->spa_dsl_pool));
5042 spa_scan(spa_t *spa, pool_scan_func_t func)
5044 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5046 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5050 * If a resilver was requested, but there is no DTL on a
5051 * writeable leaf device, we have nothing to do.
5053 if (func == POOL_SCAN_RESILVER &&
5054 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5055 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5059 return (dsl_scan(spa->spa_dsl_pool, func));
5063 * ==========================================================================
5064 * SPA async task processing
5065 * ==========================================================================
5069 spa_async_remove(spa_t *spa, vdev_t *vd)
5073 if (vd->vdev_remove_wanted) {
5074 vd->vdev_remove_wanted = B_FALSE;
5075 vd->vdev_delayed_close = B_FALSE;
5076 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5079 * We want to clear the stats, but we don't want to do a full
5080 * vdev_clear() as that will cause us to throw away
5081 * degraded/faulted state as well as attempt to reopen the
5082 * device, all of which is a waste.
5084 vd->vdev_stat.vs_read_errors = 0;
5085 vd->vdev_stat.vs_write_errors = 0;
5086 vd->vdev_stat.vs_checksum_errors = 0;
5088 vdev_state_dirty(vd->vdev_top);
5091 for (c = 0; c < vd->vdev_children; c++)
5092 spa_async_remove(spa, vd->vdev_child[c]);
5096 spa_async_probe(spa_t *spa, vdev_t *vd)
5100 if (vd->vdev_probe_wanted) {
5101 vd->vdev_probe_wanted = B_FALSE;
5102 vdev_reopen(vd); /* vdev_open() does the actual probe */
5105 for (c = 0; c < vd->vdev_children; c++)
5106 spa_async_probe(spa, vd->vdev_child[c]);
5110 spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5114 if (!spa->spa_autoexpand)
5117 for (c = 0; c < vd->vdev_children; c++) {
5118 vdev_t *cvd = vd->vdev_child[c];
5119 spa_async_autoexpand(spa, cvd);
5122 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5125 spa_event_notify(vd->vdev_spa, vd, FM_EREPORT_ZFS_DEVICE_AUTOEXPAND);
5129 spa_async_thread(spa_t *spa)
5133 ASSERT(spa->spa_sync_on);
5135 mutex_enter(&spa->spa_async_lock);
5136 tasks = spa->spa_async_tasks;
5137 spa->spa_async_tasks = 0;
5138 mutex_exit(&spa->spa_async_lock);
5141 * See if the config needs to be updated.
5143 if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5144 uint64_t old_space, new_space;
5146 mutex_enter(&spa_namespace_lock);
5147 old_space = metaslab_class_get_space(spa_normal_class(spa));
5148 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5149 new_space = metaslab_class_get_space(spa_normal_class(spa));
5150 mutex_exit(&spa_namespace_lock);
5153 * If the pool grew as a result of the config update,
5154 * then log an internal history event.
5156 if (new_space != old_space) {
5157 spa_history_log_internal(LOG_POOL_VDEV_ONLINE,
5159 "pool '%s' size: %llu(+%llu)",
5160 spa_name(spa), new_space, new_space - old_space);
5165 * See if any devices need to be marked REMOVED.
5167 if (tasks & SPA_ASYNC_REMOVE) {
5168 spa_vdev_state_enter(spa, SCL_NONE);
5169 spa_async_remove(spa, spa->spa_root_vdev);
5170 for (i = 0; i < spa->spa_l2cache.sav_count; i++)
5171 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5172 for (i = 0; i < spa->spa_spares.sav_count; i++)
5173 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5174 (void) spa_vdev_state_exit(spa, NULL, 0);
5177 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5178 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5179 spa_async_autoexpand(spa, spa->spa_root_vdev);
5180 spa_config_exit(spa, SCL_CONFIG, FTAG);
5184 * See if any devices need to be probed.
5186 if (tasks & SPA_ASYNC_PROBE) {
5187 spa_vdev_state_enter(spa, SCL_NONE);
5188 spa_async_probe(spa, spa->spa_root_vdev);
5189 (void) spa_vdev_state_exit(spa, NULL, 0);
5193 * If any devices are done replacing, detach them.
5195 if (tasks & SPA_ASYNC_RESILVER_DONE)
5196 spa_vdev_resilver_done(spa);
5199 * Kick off a resilver.
5201 if (tasks & SPA_ASYNC_RESILVER)
5202 dsl_resilver_restart(spa->spa_dsl_pool, 0);
5205 * Let the world know that we're done.
5207 mutex_enter(&spa->spa_async_lock);
5208 spa->spa_async_thread = NULL;
5209 cv_broadcast(&spa->spa_async_cv);
5210 mutex_exit(&spa->spa_async_lock);
5215 spa_async_suspend(spa_t *spa)
5217 mutex_enter(&spa->spa_async_lock);
5218 spa->spa_async_suspended++;
5219 while (spa->spa_async_thread != NULL)
5220 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5221 mutex_exit(&spa->spa_async_lock);
5225 spa_async_resume(spa_t *spa)
5227 mutex_enter(&spa->spa_async_lock);
5228 ASSERT(spa->spa_async_suspended != 0);
5229 spa->spa_async_suspended--;
5230 mutex_exit(&spa->spa_async_lock);
5234 spa_async_dispatch(spa_t *spa)
5236 mutex_enter(&spa->spa_async_lock);
5237 if (spa->spa_async_tasks && !spa->spa_async_suspended &&
5238 spa->spa_async_thread == NULL &&
5239 rootdir != NULL && !vn_is_readonly(rootdir))
5240 spa->spa_async_thread = thread_create(NULL, 0,
5241 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
5242 mutex_exit(&spa->spa_async_lock);
5246 spa_async_request(spa_t *spa, int task)
5248 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
5249 mutex_enter(&spa->spa_async_lock);
5250 spa->spa_async_tasks |= task;
5251 mutex_exit(&spa->spa_async_lock);
5255 * ==========================================================================
5256 * SPA syncing routines
5257 * ==========================================================================
5261 bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5264 bpobj_enqueue(bpo, bp, tx);
5269 spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5273 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
5279 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
5281 char *packed = NULL;
5286 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
5289 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5290 * information. This avoids the dbuf_will_dirty() path and
5291 * saves us a pre-read to get data we don't actually care about.
5293 bufsize = P2ROUNDUP(nvsize, SPA_CONFIG_BLOCKSIZE);
5294 packed = vmem_alloc(bufsize, KM_SLEEP);
5296 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
5298 bzero(packed + nvsize, bufsize - nvsize);
5300 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
5302 vmem_free(packed, bufsize);
5304 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
5305 dmu_buf_will_dirty(db, tx);
5306 *(uint64_t *)db->db_data = nvsize;
5307 dmu_buf_rele(db, FTAG);
5311 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
5312 const char *config, const char *entry)
5322 * Update the MOS nvlist describing the list of available devices.
5323 * spa_validate_aux() will have already made sure this nvlist is
5324 * valid and the vdevs are labeled appropriately.
5326 if (sav->sav_object == 0) {
5327 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
5328 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
5329 sizeof (uint64_t), tx);
5330 VERIFY(zap_update(spa->spa_meta_objset,
5331 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
5332 &sav->sav_object, tx) == 0);
5335 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5336 if (sav->sav_count == 0) {
5337 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
5339 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
5340 for (i = 0; i < sav->sav_count; i++)
5341 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
5342 B_FALSE, VDEV_CONFIG_L2CACHE);
5343 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
5344 sav->sav_count) == 0);
5345 for (i = 0; i < sav->sav_count; i++)
5346 nvlist_free(list[i]);
5347 kmem_free(list, sav->sav_count * sizeof (void *));
5350 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
5351 nvlist_free(nvroot);
5353 sav->sav_sync = B_FALSE;
5357 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
5361 if (list_is_empty(&spa->spa_config_dirty_list))
5364 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5366 config = spa_config_generate(spa, spa->spa_root_vdev,
5367 dmu_tx_get_txg(tx), B_FALSE);
5369 spa_config_exit(spa, SCL_STATE, FTAG);
5371 if (spa->spa_config_syncing)
5372 nvlist_free(spa->spa_config_syncing);
5373 spa->spa_config_syncing = config;
5375 spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
5379 * Set zpool properties.
5382 spa_sync_props(void *arg1, void *arg2, dmu_tx_t *tx)
5385 objset_t *mos = spa->spa_meta_objset;
5386 nvlist_t *nvp = arg2;
5391 const char *propname;
5392 zprop_type_t proptype;
5394 mutex_enter(&spa->spa_props_lock);
5397 while ((elem = nvlist_next_nvpair(nvp, elem))) {
5398 switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
5399 case ZPOOL_PROP_VERSION:
5401 * Only set version for non-zpool-creation cases
5402 * (set/import). spa_create() needs special care
5403 * for version setting.
5405 if (tx->tx_txg != TXG_INITIAL) {
5406 VERIFY(nvpair_value_uint64(elem,
5408 ASSERT(intval <= SPA_VERSION);
5409 ASSERT(intval >= spa_version(spa));
5410 spa->spa_uberblock.ub_version = intval;
5411 vdev_config_dirty(spa->spa_root_vdev);
5415 case ZPOOL_PROP_ALTROOT:
5417 * 'altroot' is a non-persistent property. It should
5418 * have been set temporarily at creation or import time.
5420 ASSERT(spa->spa_root != NULL);
5423 case ZPOOL_PROP_READONLY:
5424 case ZPOOL_PROP_CACHEFILE:
5426 * 'readonly' and 'cachefile' are also non-persisitent
5430 case ZPOOL_PROP_COMMENT:
5431 VERIFY(nvpair_value_string(elem, &strval) == 0);
5432 if (spa->spa_comment != NULL)
5433 spa_strfree(spa->spa_comment);
5434 spa->spa_comment = spa_strdup(strval);
5436 * We need to dirty the configuration on all the vdevs
5437 * so that their labels get updated. It's unnecessary
5438 * to do this for pool creation since the vdev's
5439 * configuratoin has already been dirtied.
5441 if (tx->tx_txg != TXG_INITIAL)
5442 vdev_config_dirty(spa->spa_root_vdev);
5446 * Set pool property values in the poolprops mos object.
5448 if (spa->spa_pool_props_object == 0) {
5449 VERIFY((spa->spa_pool_props_object =
5450 zap_create(mos, DMU_OT_POOL_PROPS,
5451 DMU_OT_NONE, 0, tx)) > 0);
5453 VERIFY(zap_update(mos,
5454 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
5455 8, 1, &spa->spa_pool_props_object, tx)
5459 /* normalize the property name */
5460 propname = zpool_prop_to_name(prop);
5461 proptype = zpool_prop_get_type(prop);
5463 if (nvpair_type(elem) == DATA_TYPE_STRING) {
5464 ASSERT(proptype == PROP_TYPE_STRING);
5465 VERIFY(nvpair_value_string(elem, &strval) == 0);
5466 VERIFY(zap_update(mos,
5467 spa->spa_pool_props_object, propname,
5468 1, strlen(strval) + 1, strval, tx) == 0);
5470 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
5471 VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5473 if (proptype == PROP_TYPE_INDEX) {
5475 VERIFY(zpool_prop_index_to_string(
5476 prop, intval, &unused) == 0);
5478 VERIFY(zap_update(mos,
5479 spa->spa_pool_props_object, propname,
5480 8, 1, &intval, tx) == 0);
5482 ASSERT(0); /* not allowed */
5486 case ZPOOL_PROP_DELEGATION:
5487 spa->spa_delegation = intval;
5489 case ZPOOL_PROP_BOOTFS:
5490 spa->spa_bootfs = intval;
5492 case ZPOOL_PROP_FAILUREMODE:
5493 spa->spa_failmode = intval;
5495 case ZPOOL_PROP_AUTOEXPAND:
5496 spa->spa_autoexpand = intval;
5497 if (tx->tx_txg != TXG_INITIAL)
5498 spa_async_request(spa,
5499 SPA_ASYNC_AUTOEXPAND);
5501 case ZPOOL_PROP_DEDUPDITTO:
5502 spa->spa_dedup_ditto = intval;
5509 /* log internal history if this is not a zpool create */
5510 if (spa_version(spa) >= SPA_VERSION_ZPOOL_HISTORY &&
5511 tx->tx_txg != TXG_INITIAL) {
5512 spa_history_log_internal(LOG_POOL_PROPSET,
5513 spa, tx, "%s %lld %s",
5514 nvpair_name(elem), intval, spa_name(spa));
5518 mutex_exit(&spa->spa_props_lock);
5522 * Perform one-time upgrade on-disk changes. spa_version() does not
5523 * reflect the new version this txg, so there must be no changes this
5524 * txg to anything that the upgrade code depends on after it executes.
5525 * Therefore this must be called after dsl_pool_sync() does the sync
5529 spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
5531 dsl_pool_t *dp = spa->spa_dsl_pool;
5533 ASSERT(spa->spa_sync_pass == 1);
5535 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
5536 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
5537 dsl_pool_create_origin(dp, tx);
5539 /* Keeping the origin open increases spa_minref */
5540 spa->spa_minref += 3;
5543 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
5544 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
5545 dsl_pool_upgrade_clones(dp, tx);
5548 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
5549 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
5550 dsl_pool_upgrade_dir_clones(dp, tx);
5552 /* Keeping the freedir open increases spa_minref */
5553 spa->spa_minref += 3;
5558 * Sync the specified transaction group. New blocks may be dirtied as
5559 * part of the process, so we iterate until it converges.
5562 spa_sync(spa_t *spa, uint64_t txg)
5564 dsl_pool_t *dp = spa->spa_dsl_pool;
5565 objset_t *mos = spa->spa_meta_objset;
5566 bpobj_t *defer_bpo = &spa->spa_deferred_bpobj;
5567 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
5568 vdev_t *rvd = spa->spa_root_vdev;
5574 VERIFY(spa_writeable(spa));
5577 * Lock out configuration changes.
5579 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5581 spa->spa_syncing_txg = txg;
5582 spa->spa_sync_pass = 0;
5585 * If there are any pending vdev state changes, convert them
5586 * into config changes that go out with this transaction group.
5588 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5589 while (list_head(&spa->spa_state_dirty_list) != NULL) {
5591 * We need the write lock here because, for aux vdevs,
5592 * calling vdev_config_dirty() modifies sav_config.
5593 * This is ugly and will become unnecessary when we
5594 * eliminate the aux vdev wart by integrating all vdevs
5595 * into the root vdev tree.
5597 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
5598 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
5599 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
5600 vdev_state_clean(vd);
5601 vdev_config_dirty(vd);
5603 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
5604 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
5606 spa_config_exit(spa, SCL_STATE, FTAG);
5608 tx = dmu_tx_create_assigned(dp, txg);
5611 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
5612 * set spa_deflate if we have no raid-z vdevs.
5614 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
5615 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
5618 for (i = 0; i < rvd->vdev_children; i++) {
5619 vd = rvd->vdev_child[i];
5620 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
5623 if (i == rvd->vdev_children) {
5624 spa->spa_deflate = TRUE;
5625 VERIFY(0 == zap_add(spa->spa_meta_objset,
5626 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
5627 sizeof (uint64_t), 1, &spa->spa_deflate, tx));
5632 * If anything has changed in this txg, or if someone is waiting
5633 * for this txg to sync (eg, spa_vdev_remove()), push the
5634 * deferred frees from the previous txg. If not, leave them
5635 * alone so that we don't generate work on an otherwise idle
5638 if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
5639 !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
5640 !txg_list_empty(&dp->dp_sync_tasks, txg) ||
5641 ((dsl_scan_active(dp->dp_scan) ||
5642 txg_sync_waiting(dp)) && !spa_shutting_down(spa))) {
5643 zio_t *zio = zio_root(spa, NULL, NULL, 0);
5644 VERIFY3U(bpobj_iterate(defer_bpo,
5645 spa_free_sync_cb, zio, tx), ==, 0);
5646 VERIFY3U(zio_wait(zio), ==, 0);
5650 * Iterate to convergence.
5653 int pass = ++spa->spa_sync_pass;
5655 spa_sync_config_object(spa, tx);
5656 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
5657 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
5658 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
5659 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
5660 spa_errlog_sync(spa, txg);
5661 dsl_pool_sync(dp, txg);
5663 if (pass <= SYNC_PASS_DEFERRED_FREE) {
5664 zio_t *zio = zio_root(spa, NULL, NULL, 0);
5665 bplist_iterate(free_bpl, spa_free_sync_cb,
5667 VERIFY(zio_wait(zio) == 0);
5669 bplist_iterate(free_bpl, bpobj_enqueue_cb,
5674 dsl_scan_sync(dp, tx);
5676 while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)))
5680 spa_sync_upgrades(spa, tx);
5682 } while (dmu_objset_is_dirty(mos, txg));
5685 * Rewrite the vdev configuration (which includes the uberblock)
5686 * to commit the transaction group.
5688 * If there are no dirty vdevs, we sync the uberblock to a few
5689 * random top-level vdevs that are known to be visible in the
5690 * config cache (see spa_vdev_add() for a complete description).
5691 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
5695 * We hold SCL_STATE to prevent vdev open/close/etc.
5696 * while we're attempting to write the vdev labels.
5698 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5700 if (list_is_empty(&spa->spa_config_dirty_list)) {
5701 vdev_t *svd[SPA_DVAS_PER_BP];
5703 int children = rvd->vdev_children;
5704 int c0 = spa_get_random(children);
5706 for (c = 0; c < children; c++) {
5707 vd = rvd->vdev_child[(c0 + c) % children];
5708 if (vd->vdev_ms_array == 0 || vd->vdev_islog)
5710 svd[svdcount++] = vd;
5711 if (svdcount == SPA_DVAS_PER_BP)
5714 error = vdev_config_sync(svd, svdcount, txg, B_FALSE);
5716 error = vdev_config_sync(svd, svdcount, txg,
5719 error = vdev_config_sync(rvd->vdev_child,
5720 rvd->vdev_children, txg, B_FALSE);
5722 error = vdev_config_sync(rvd->vdev_child,
5723 rvd->vdev_children, txg, B_TRUE);
5726 spa_config_exit(spa, SCL_STATE, FTAG);
5730 zio_suspend(spa, NULL);
5731 zio_resume_wait(spa);
5736 * Clear the dirty config list.
5738 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
5739 vdev_config_clean(vd);
5742 * Now that the new config has synced transactionally,
5743 * let it become visible to the config cache.
5745 if (spa->spa_config_syncing != NULL) {
5746 spa_config_set(spa, spa->spa_config_syncing);
5747 spa->spa_config_txg = txg;
5748 spa->spa_config_syncing = NULL;
5751 spa->spa_ubsync = spa->spa_uberblock;
5753 dsl_pool_sync_done(dp, txg);
5756 * Update usable space statistics.
5758 while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg))))
5759 vdev_sync_done(vd, txg);
5761 spa_update_dspace(spa);
5764 * It had better be the case that we didn't dirty anything
5765 * since vdev_config_sync().
5767 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
5768 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
5769 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
5771 spa->spa_sync_pass = 0;
5773 spa_config_exit(spa, SCL_CONFIG, FTAG);
5775 spa_handle_ignored_writes(spa);
5778 * If any async tasks have been requested, kick them off.
5780 spa_async_dispatch(spa);
5784 * Sync all pools. We don't want to hold the namespace lock across these
5785 * operations, so we take a reference on the spa_t and drop the lock during the
5789 spa_sync_allpools(void)
5792 mutex_enter(&spa_namespace_lock);
5793 while ((spa = spa_next(spa)) != NULL) {
5794 if (spa_state(spa) != POOL_STATE_ACTIVE ||
5795 !spa_writeable(spa) || spa_suspended(spa))
5797 spa_open_ref(spa, FTAG);
5798 mutex_exit(&spa_namespace_lock);
5799 txg_wait_synced(spa_get_dsl(spa), 0);
5800 mutex_enter(&spa_namespace_lock);
5801 spa_close(spa, FTAG);
5803 mutex_exit(&spa_namespace_lock);
5807 * ==========================================================================
5808 * Miscellaneous routines
5809 * ==========================================================================
5813 * Remove all pools in the system.
5821 * Remove all cached state. All pools should be closed now,
5822 * so every spa in the AVL tree should be unreferenced.
5824 mutex_enter(&spa_namespace_lock);
5825 while ((spa = spa_next(NULL)) != NULL) {
5827 * Stop async tasks. The async thread may need to detach
5828 * a device that's been replaced, which requires grabbing
5829 * spa_namespace_lock, so we must drop it here.
5831 spa_open_ref(spa, FTAG);
5832 mutex_exit(&spa_namespace_lock);
5833 spa_async_suspend(spa);
5834 mutex_enter(&spa_namespace_lock);
5835 spa_close(spa, FTAG);
5837 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
5839 spa_deactivate(spa);
5843 mutex_exit(&spa_namespace_lock);
5847 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
5852 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
5856 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
5857 vd = spa->spa_l2cache.sav_vdevs[i];
5858 if (vd->vdev_guid == guid)
5862 for (i = 0; i < spa->spa_spares.sav_count; i++) {
5863 vd = spa->spa_spares.sav_vdevs[i];
5864 if (vd->vdev_guid == guid)
5873 spa_upgrade(spa_t *spa, uint64_t version)
5875 ASSERT(spa_writeable(spa));
5877 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5880 * This should only be called for a non-faulted pool, and since a
5881 * future version would result in an unopenable pool, this shouldn't be
5884 ASSERT(spa->spa_uberblock.ub_version <= SPA_VERSION);
5885 ASSERT(version >= spa->spa_uberblock.ub_version);
5887 spa->spa_uberblock.ub_version = version;
5888 vdev_config_dirty(spa->spa_root_vdev);
5890 spa_config_exit(spa, SCL_ALL, FTAG);
5892 txg_wait_synced(spa_get_dsl(spa), 0);
5896 spa_has_spare(spa_t *spa, uint64_t guid)
5900 spa_aux_vdev_t *sav = &spa->spa_spares;
5902 for (i = 0; i < sav->sav_count; i++)
5903 if (sav->sav_vdevs[i]->vdev_guid == guid)
5906 for (i = 0; i < sav->sav_npending; i++) {
5907 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
5908 &spareguid) == 0 && spareguid == guid)
5916 * Check if a pool has an active shared spare device.
5917 * Note: reference count of an active spare is 2, as a spare and as a replace
5920 spa_has_active_shared_spare(spa_t *spa)
5924 spa_aux_vdev_t *sav = &spa->spa_spares;
5926 for (i = 0; i < sav->sav_count; i++) {
5927 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
5928 &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
5937 * Post a FM_EREPORT_ZFS_* event from sys/fm/fs/zfs.h. The payload will be
5938 * filled in from the spa and (optionally) the vdev. This doesn't do anything
5939 * in the userland libzpool, as we don't want consumers to misinterpret ztest
5940 * or zdb as real changes.
5943 spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
5946 zfs_ereport_post(name, spa, vd, NULL, 0, 0);
5950 #if defined(_KERNEL) && defined(HAVE_SPL)
5951 /* state manipulation functions */
5952 EXPORT_SYMBOL(spa_open);
5953 EXPORT_SYMBOL(spa_open_rewind);
5954 EXPORT_SYMBOL(spa_get_stats);
5955 EXPORT_SYMBOL(spa_create);
5956 EXPORT_SYMBOL(spa_import_rootpool);
5957 EXPORT_SYMBOL(spa_import);
5958 EXPORT_SYMBOL(spa_tryimport);
5959 EXPORT_SYMBOL(spa_destroy);
5960 EXPORT_SYMBOL(spa_export);
5961 EXPORT_SYMBOL(spa_reset);
5962 EXPORT_SYMBOL(spa_async_request);
5963 EXPORT_SYMBOL(spa_async_suspend);
5964 EXPORT_SYMBOL(spa_async_resume);
5965 EXPORT_SYMBOL(spa_inject_addref);
5966 EXPORT_SYMBOL(spa_inject_delref);
5967 EXPORT_SYMBOL(spa_scan_stat_init);
5968 EXPORT_SYMBOL(spa_scan_get_stats);
5970 /* device maniion */
5971 EXPORT_SYMBOL(spa_vdev_add);
5972 EXPORT_SYMBOL(spa_vdev_attach);
5973 EXPORT_SYMBOL(spa_vdev_detach);
5974 EXPORT_SYMBOL(spa_vdev_remove);
5975 EXPORT_SYMBOL(spa_vdev_setpath);
5976 EXPORT_SYMBOL(spa_vdev_setfru);
5977 EXPORT_SYMBOL(spa_vdev_split_mirror);
5979 /* spare statech is global across all pools) */
5980 EXPORT_SYMBOL(spa_spare_add);
5981 EXPORT_SYMBOL(spa_spare_remove);
5982 EXPORT_SYMBOL(spa_spare_exists);
5983 EXPORT_SYMBOL(spa_spare_activate);
5985 /* L2ARC statech is global across all pools) */
5986 EXPORT_SYMBOL(spa_l2cache_add);
5987 EXPORT_SYMBOL(spa_l2cache_remove);
5988 EXPORT_SYMBOL(spa_l2cache_exists);
5989 EXPORT_SYMBOL(spa_l2cache_activate);
5990 EXPORT_SYMBOL(spa_l2cache_drop);
5993 EXPORT_SYMBOL(spa_scan);
5994 EXPORT_SYMBOL(spa_scan_stop);
5997 EXPORT_SYMBOL(spa_sync); /* only for DMU use */
5998 EXPORT_SYMBOL(spa_sync_allpools);
6001 EXPORT_SYMBOL(spa_prop_set);
6002 EXPORT_SYMBOL(spa_prop_get);
6003 EXPORT_SYMBOL(spa_prop_clear_bootfs);
6005 /* asynchronous event notification */
6006 EXPORT_SYMBOL(spa_event_notify);