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.
27 * This file contains all the routines used when modifying on-disk SPA state.
28 * This includes opening, importing, destroying, exporting a pool, and syncing a
32 #include <sys/zfs_context.h>
33 #include <sys/fm/fs/zfs.h>
34 #include <sys/spa_impl.h>
36 #include <sys/zio_checksum.h>
38 #include <sys/dmu_tx.h>
42 #include <sys/vdev_impl.h>
43 #include <sys/metaslab.h>
44 #include <sys/metaslab_impl.h>
45 #include <sys/uberblock_impl.h>
48 #include <sys/dmu_traverse.h>
49 #include <sys/dmu_objset.h>
50 #include <sys/unique.h>
51 #include <sys/dsl_pool.h>
52 #include <sys/dsl_dataset.h>
53 #include <sys/dsl_dir.h>
54 #include <sys/dsl_prop.h>
55 #include <sys/dsl_synctask.h>
56 #include <sys/fs/zfs.h>
58 #include <sys/callb.h>
59 #include <sys/systeminfo.h>
60 #include <sys/spa_boot.h>
61 #include <sys/zfs_ioctl.h>
62 #include <sys/dsl_scan.h>
65 #include <sys/bootprops.h>
66 #include <sys/callb.h>
67 #include <sys/cpupart.h>
69 #include <sys/sysdc.h>
74 #include "zfs_comutil.h"
76 typedef enum zti_modes {
77 zti_mode_fixed, /* value is # of threads (min 1) */
78 zti_mode_online_percent, /* value is % of online CPUs */
79 zti_mode_batch, /* cpu-intensive; value is ignored */
80 zti_mode_null, /* don't create a taskq */
84 #define ZTI_FIX(n) { zti_mode_fixed, (n) }
85 #define ZTI_PCT(n) { zti_mode_online_percent, (n) }
86 #define ZTI_BATCH { zti_mode_batch, 0 }
87 #define ZTI_NULL { zti_mode_null, 0 }
89 #define ZTI_ONE ZTI_FIX(1)
91 typedef struct zio_taskq_info {
92 enum zti_modes zti_mode;
96 static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
97 "issue", "issue_high", "intr", "intr_high"
101 * Define the taskq threads for the following I/O types:
102 * NULL, READ, WRITE, FREE, CLAIM, and IOCTL
104 const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
105 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
106 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL },
107 { ZTI_FIX(8), ZTI_NULL, ZTI_BATCH, ZTI_NULL },
108 { ZTI_BATCH, ZTI_FIX(5), ZTI_FIX(8), ZTI_FIX(5) },
109 { ZTI_FIX(100), ZTI_NULL, ZTI_ONE, ZTI_NULL },
110 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL },
111 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL },
114 static dsl_syncfunc_t spa_sync_props;
115 static boolean_t spa_has_active_shared_spare(spa_t *spa);
116 static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
117 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
119 static void spa_vdev_resilver_done(spa_t *spa);
121 uint_t zio_taskq_batch_pct = 100; /* 1 thread per cpu in pset */
122 id_t zio_taskq_psrset_bind = PS_NONE;
123 boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */
124 uint_t zio_taskq_basedc = 80; /* base duty cycle */
126 boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */
129 * This (illegal) pool name is used when temporarily importing a spa_t in order
130 * to get the vdev stats associated with the imported devices.
132 #define TRYIMPORT_NAME "$import"
135 * ==========================================================================
136 * SPA properties routines
137 * ==========================================================================
141 * Add a (source=src, propname=propval) list to an nvlist.
144 spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
145 uint64_t intval, zprop_source_t src)
147 const char *propname = zpool_prop_to_name(prop);
150 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
151 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
154 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
156 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
158 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
159 nvlist_free(propval);
163 * Get property values from the spa configuration.
166 spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
170 uint64_t cap, version;
171 zprop_source_t src = ZPROP_SRC_NONE;
172 spa_config_dirent_t *dp;
174 ASSERT(MUTEX_HELD(&spa->spa_props_lock));
176 if (spa->spa_root_vdev != NULL) {
177 alloc = metaslab_class_get_alloc(spa_normal_class(spa));
178 size = metaslab_class_get_space(spa_normal_class(spa));
179 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
180 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
181 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
182 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
184 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
185 (spa_mode(spa) == FREAD), src);
187 cap = (size == 0) ? 0 : (alloc * 100 / size);
188 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
190 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
191 ddt_get_pool_dedup_ratio(spa), src);
193 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
194 spa->spa_root_vdev->vdev_state, src);
196 version = spa_version(spa);
197 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
198 src = ZPROP_SRC_DEFAULT;
200 src = ZPROP_SRC_LOCAL;
201 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
204 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
206 if (spa->spa_root != NULL)
207 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
210 if ((dp = list_head(&spa->spa_config_list)) != NULL) {
211 if (dp->scd_path == NULL) {
212 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
213 "none", 0, ZPROP_SRC_LOCAL);
214 } else if (strcmp(dp->scd_path, spa_config_path) != 0) {
215 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
216 dp->scd_path, 0, ZPROP_SRC_LOCAL);
222 * Get zpool property values.
225 spa_prop_get(spa_t *spa, nvlist_t **nvp)
227 objset_t *mos = spa->spa_meta_objset;
232 VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
234 mutex_enter(&spa->spa_props_lock);
237 * Get properties from the spa config.
239 spa_prop_get_config(spa, nvp);
241 /* If no pool property object, no more prop to get. */
242 if (mos == NULL || spa->spa_pool_props_object == 0) {
243 mutex_exit(&spa->spa_props_lock);
248 * Get properties from the MOS pool property object.
250 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
251 (err = zap_cursor_retrieve(&zc, &za)) == 0;
252 zap_cursor_advance(&zc)) {
255 zprop_source_t src = ZPROP_SRC_DEFAULT;
258 if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
261 switch (za.za_integer_length) {
263 /* integer property */
264 if (za.za_first_integer !=
265 zpool_prop_default_numeric(prop))
266 src = ZPROP_SRC_LOCAL;
268 if (prop == ZPOOL_PROP_BOOTFS) {
270 dsl_dataset_t *ds = NULL;
272 dp = spa_get_dsl(spa);
273 rw_enter(&dp->dp_config_rwlock, RW_READER);
274 if (err = dsl_dataset_hold_obj(dp,
275 za.za_first_integer, FTAG, &ds)) {
276 rw_exit(&dp->dp_config_rwlock);
281 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
283 dsl_dataset_name(ds, strval);
284 dsl_dataset_rele(ds, FTAG);
285 rw_exit(&dp->dp_config_rwlock);
288 intval = za.za_first_integer;
291 spa_prop_add_list(*nvp, prop, strval, intval, src);
295 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
300 /* string property */
301 strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
302 err = zap_lookup(mos, spa->spa_pool_props_object,
303 za.za_name, 1, za.za_num_integers, strval);
305 kmem_free(strval, za.za_num_integers);
308 spa_prop_add_list(*nvp, prop, strval, 0, src);
309 kmem_free(strval, za.za_num_integers);
316 zap_cursor_fini(&zc);
317 mutex_exit(&spa->spa_props_lock);
319 if (err && err != ENOENT) {
329 * Validate the given pool properties nvlist and modify the list
330 * for the property values to be set.
333 spa_prop_validate(spa_t *spa, nvlist_t *props)
336 int error = 0, reset_bootfs = 0;
340 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
342 char *propname, *strval;
347 propname = nvpair_name(elem);
349 if ((prop = zpool_name_to_prop(propname)) == ZPROP_INVAL)
353 case ZPOOL_PROP_VERSION:
354 error = nvpair_value_uint64(elem, &intval);
356 (intval < spa_version(spa) || intval > SPA_VERSION))
360 case ZPOOL_PROP_DELEGATION:
361 case ZPOOL_PROP_AUTOREPLACE:
362 case ZPOOL_PROP_LISTSNAPS:
363 case ZPOOL_PROP_AUTOEXPAND:
364 error = nvpair_value_uint64(elem, &intval);
365 if (!error && intval > 1)
369 case ZPOOL_PROP_BOOTFS:
371 * If the pool version is less than SPA_VERSION_BOOTFS,
372 * or the pool is still being created (version == 0),
373 * the bootfs property cannot be set.
375 if (spa_version(spa) < SPA_VERSION_BOOTFS) {
381 * Make sure the vdev config is bootable
383 if (!vdev_is_bootable(spa->spa_root_vdev)) {
390 error = nvpair_value_string(elem, &strval);
395 if (strval == NULL || strval[0] == '\0') {
396 objnum = zpool_prop_default_numeric(
401 if (error = dmu_objset_hold(strval, FTAG, &os))
404 /* Must be ZPL and not gzip compressed. */
406 if (dmu_objset_type(os) != DMU_OST_ZFS) {
408 } else if ((error = dsl_prop_get_integer(strval,
409 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
410 &compress, NULL)) == 0 &&
411 !BOOTFS_COMPRESS_VALID(compress)) {
414 objnum = dmu_objset_id(os);
416 dmu_objset_rele(os, FTAG);
420 case ZPOOL_PROP_FAILUREMODE:
421 error = nvpair_value_uint64(elem, &intval);
422 if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
423 intval > ZIO_FAILURE_MODE_PANIC))
427 * This is a special case which only occurs when
428 * the pool has completely failed. This allows
429 * the user to change the in-core failmode property
430 * without syncing it out to disk (I/Os might
431 * currently be blocked). We do this by returning
432 * EIO to the caller (spa_prop_set) to trick it
433 * into thinking we encountered a property validation
436 if (!error && spa_suspended(spa)) {
437 spa->spa_failmode = intval;
442 case ZPOOL_PROP_CACHEFILE:
443 if ((error = nvpair_value_string(elem, &strval)) != 0)
446 if (strval[0] == '\0')
449 if (strcmp(strval, "none") == 0)
452 if (strval[0] != '/') {
457 slash = strrchr(strval, '/');
458 ASSERT(slash != NULL);
460 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
461 strcmp(slash, "/..") == 0)
465 case ZPOOL_PROP_DEDUPDITTO:
466 if (spa_version(spa) < SPA_VERSION_DEDUP)
469 error = nvpair_value_uint64(elem, &intval);
471 intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
480 if (!error && reset_bootfs) {
481 error = nvlist_remove(props,
482 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
485 error = nvlist_add_uint64(props,
486 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
494 spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
497 spa_config_dirent_t *dp;
499 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
503 dp = kmem_alloc(sizeof (spa_config_dirent_t),
506 if (cachefile[0] == '\0')
507 dp->scd_path = spa_strdup(spa_config_path);
508 else if (strcmp(cachefile, "none") == 0)
511 dp->scd_path = spa_strdup(cachefile);
513 list_insert_head(&spa->spa_config_list, dp);
515 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
519 spa_prop_set(spa_t *spa, nvlist_t *nvp)
523 boolean_t need_sync = B_FALSE;
526 if ((error = spa_prop_validate(spa, nvp)) != 0)
530 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
531 if ((prop = zpool_name_to_prop(
532 nvpair_name(elem))) == ZPROP_INVAL)
535 if (prop == ZPOOL_PROP_CACHEFILE ||
536 prop == ZPOOL_PROP_ALTROOT ||
537 prop == ZPOOL_PROP_READONLY)
545 return (dsl_sync_task_do(spa_get_dsl(spa), NULL, spa_sync_props,
552 * If the bootfs property value is dsobj, clear it.
555 spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
557 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
558 VERIFY(zap_remove(spa->spa_meta_objset,
559 spa->spa_pool_props_object,
560 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
566 * ==========================================================================
567 * SPA state manipulation (open/create/destroy/import/export)
568 * ==========================================================================
572 spa_error_entry_compare(const void *a, const void *b)
574 spa_error_entry_t *sa = (spa_error_entry_t *)a;
575 spa_error_entry_t *sb = (spa_error_entry_t *)b;
578 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
579 sizeof (zbookmark_t));
590 * Utility function which retrieves copies of the current logs and
591 * re-initializes them in the process.
594 spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
596 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
598 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
599 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
601 avl_create(&spa->spa_errlist_scrub,
602 spa_error_entry_compare, sizeof (spa_error_entry_t),
603 offsetof(spa_error_entry_t, se_avl));
604 avl_create(&spa->spa_errlist_last,
605 spa_error_entry_compare, sizeof (spa_error_entry_t),
606 offsetof(spa_error_entry_t, se_avl));
610 spa_taskq_create(spa_t *spa, const char *name, enum zti_modes mode,
613 uint_t flags = TASKQ_PREPOPULATE;
614 boolean_t batch = B_FALSE;
618 return (NULL); /* no taskq needed */
621 ASSERT3U(value, >=, 1);
622 value = MAX(value, 1);
627 flags |= TASKQ_THREADS_CPU_PCT;
628 value = zio_taskq_batch_pct;
631 case zti_mode_online_percent:
632 flags |= TASKQ_THREADS_CPU_PCT;
636 panic("unrecognized mode for %s taskq (%u:%u) in "
642 if (zio_taskq_sysdc && spa->spa_proc != &p0) {
644 flags |= TASKQ_DC_BATCH;
646 return (taskq_create_sysdc(name, value, 50, INT_MAX,
647 spa->spa_proc, zio_taskq_basedc, flags));
649 return (taskq_create_proc(name, value, maxclsyspri, 50, INT_MAX,
650 spa->spa_proc, flags));
654 spa_create_zio_taskqs(spa_t *spa)
656 for (int t = 0; t < ZIO_TYPES; t++) {
657 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
658 const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
659 enum zti_modes mode = ztip->zti_mode;
660 uint_t value = ztip->zti_value;
663 (void) snprintf(name, sizeof (name),
664 "%s_%s", zio_type_name[t], zio_taskq_types[q]);
666 spa->spa_zio_taskq[t][q] =
667 spa_taskq_create(spa, name, mode, value);
674 spa_thread(void *arg)
679 user_t *pu = PTOU(curproc);
681 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
684 ASSERT(curproc != &p0);
685 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
686 "zpool-%s", spa->spa_name);
687 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
689 /* bind this thread to the requested psrset */
690 if (zio_taskq_psrset_bind != PS_NONE) {
692 mutex_enter(&cpu_lock);
693 mutex_enter(&pidlock);
694 mutex_enter(&curproc->p_lock);
696 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
697 0, NULL, NULL) == 0) {
698 curthread->t_bind_pset = zio_taskq_psrset_bind;
701 "Couldn't bind process for zfs pool \"%s\" to "
702 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
705 mutex_exit(&curproc->p_lock);
706 mutex_exit(&pidlock);
707 mutex_exit(&cpu_lock);
711 if (zio_taskq_sysdc) {
712 sysdc_thread_enter(curthread, 100, 0);
715 spa->spa_proc = curproc;
716 spa->spa_did = curthread->t_did;
718 spa_create_zio_taskqs(spa);
720 mutex_enter(&spa->spa_proc_lock);
721 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
723 spa->spa_proc_state = SPA_PROC_ACTIVE;
724 cv_broadcast(&spa->spa_proc_cv);
726 CALLB_CPR_SAFE_BEGIN(&cprinfo);
727 while (spa->spa_proc_state == SPA_PROC_ACTIVE)
728 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
729 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
731 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
732 spa->spa_proc_state = SPA_PROC_GONE;
734 cv_broadcast(&spa->spa_proc_cv);
735 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */
737 mutex_enter(&curproc->p_lock);
743 * Activate an uninitialized pool.
746 spa_activate(spa_t *spa, int mode)
748 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
750 spa->spa_state = POOL_STATE_ACTIVE;
751 spa->spa_mode = mode;
753 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
754 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
756 /* Try to create a covering process */
757 mutex_enter(&spa->spa_proc_lock);
758 ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
759 ASSERT(spa->spa_proc == &p0);
762 /* Only create a process if we're going to be around a while. */
763 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
764 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
766 spa->spa_proc_state = SPA_PROC_CREATED;
767 while (spa->spa_proc_state == SPA_PROC_CREATED) {
768 cv_wait(&spa->spa_proc_cv,
769 &spa->spa_proc_lock);
771 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
772 ASSERT(spa->spa_proc != &p0);
773 ASSERT(spa->spa_did != 0);
777 "Couldn't create process for zfs pool \"%s\"\n",
782 mutex_exit(&spa->spa_proc_lock);
784 /* If we didn't create a process, we need to create our taskqs. */
785 if (spa->spa_proc == &p0) {
786 spa_create_zio_taskqs(spa);
789 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
790 offsetof(vdev_t, vdev_config_dirty_node));
791 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
792 offsetof(vdev_t, vdev_state_dirty_node));
794 txg_list_create(&spa->spa_vdev_txg_list,
795 offsetof(struct vdev, vdev_txg_node));
797 avl_create(&spa->spa_errlist_scrub,
798 spa_error_entry_compare, sizeof (spa_error_entry_t),
799 offsetof(spa_error_entry_t, se_avl));
800 avl_create(&spa->spa_errlist_last,
801 spa_error_entry_compare, sizeof (spa_error_entry_t),
802 offsetof(spa_error_entry_t, se_avl));
806 * Opposite of spa_activate().
809 spa_deactivate(spa_t *spa)
811 ASSERT(spa->spa_sync_on == B_FALSE);
812 ASSERT(spa->spa_dsl_pool == NULL);
813 ASSERT(spa->spa_root_vdev == NULL);
814 ASSERT(spa->spa_async_zio_root == NULL);
815 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
817 txg_list_destroy(&spa->spa_vdev_txg_list);
819 list_destroy(&spa->spa_config_dirty_list);
820 list_destroy(&spa->spa_state_dirty_list);
822 for (int t = 0; t < ZIO_TYPES; t++) {
823 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
824 if (spa->spa_zio_taskq[t][q] != NULL)
825 taskq_destroy(spa->spa_zio_taskq[t][q]);
826 spa->spa_zio_taskq[t][q] = NULL;
830 metaslab_class_destroy(spa->spa_normal_class);
831 spa->spa_normal_class = NULL;
833 metaslab_class_destroy(spa->spa_log_class);
834 spa->spa_log_class = NULL;
837 * If this was part of an import or the open otherwise failed, we may
838 * still have errors left in the queues. Empty them just in case.
840 spa_errlog_drain(spa);
842 avl_destroy(&spa->spa_errlist_scrub);
843 avl_destroy(&spa->spa_errlist_last);
845 spa->spa_state = POOL_STATE_UNINITIALIZED;
847 mutex_enter(&spa->spa_proc_lock);
848 if (spa->spa_proc_state != SPA_PROC_NONE) {
849 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
850 spa->spa_proc_state = SPA_PROC_DEACTIVATE;
851 cv_broadcast(&spa->spa_proc_cv);
852 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
853 ASSERT(spa->spa_proc != &p0);
854 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
856 ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
857 spa->spa_proc_state = SPA_PROC_NONE;
859 ASSERT(spa->spa_proc == &p0);
860 mutex_exit(&spa->spa_proc_lock);
863 * We want to make sure spa_thread() has actually exited the ZFS
864 * module, so that the module can't be unloaded out from underneath
867 if (spa->spa_did != 0) {
868 thread_join(spa->spa_did);
874 * Verify a pool configuration, and construct the vdev tree appropriately. This
875 * will create all the necessary vdevs in the appropriate layout, with each vdev
876 * in the CLOSED state. This will prep the pool before open/creation/import.
877 * All vdev validation is done by the vdev_alloc() routine.
880 spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
881 uint_t id, int atype)
887 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
890 if ((*vdp)->vdev_ops->vdev_op_leaf)
893 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
905 for (int c = 0; c < children; c++) {
907 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
915 ASSERT(*vdp != NULL);
921 * Opposite of spa_load().
924 spa_unload(spa_t *spa)
928 ASSERT(MUTEX_HELD(&spa_namespace_lock));
933 spa_async_suspend(spa);
938 if (spa->spa_sync_on) {
939 txg_sync_stop(spa->spa_dsl_pool);
940 spa->spa_sync_on = B_FALSE;
944 * Wait for any outstanding async I/O to complete.
946 if (spa->spa_async_zio_root != NULL) {
947 (void) zio_wait(spa->spa_async_zio_root);
948 spa->spa_async_zio_root = NULL;
951 bpobj_close(&spa->spa_deferred_bpobj);
954 * Close the dsl pool.
956 if (spa->spa_dsl_pool) {
957 dsl_pool_close(spa->spa_dsl_pool);
958 spa->spa_dsl_pool = NULL;
959 spa->spa_meta_objset = NULL;
964 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
967 * Drop and purge level 2 cache
969 spa_l2cache_drop(spa);
974 if (spa->spa_root_vdev)
975 vdev_free(spa->spa_root_vdev);
976 ASSERT(spa->spa_root_vdev == NULL);
978 for (i = 0; i < spa->spa_spares.sav_count; i++)
979 vdev_free(spa->spa_spares.sav_vdevs[i]);
980 if (spa->spa_spares.sav_vdevs) {
981 kmem_free(spa->spa_spares.sav_vdevs,
982 spa->spa_spares.sav_count * sizeof (void *));
983 spa->spa_spares.sav_vdevs = NULL;
985 if (spa->spa_spares.sav_config) {
986 nvlist_free(spa->spa_spares.sav_config);
987 spa->spa_spares.sav_config = NULL;
989 spa->spa_spares.sav_count = 0;
991 for (i = 0; i < spa->spa_l2cache.sav_count; i++)
992 vdev_free(spa->spa_l2cache.sav_vdevs[i]);
993 if (spa->spa_l2cache.sav_vdevs) {
994 kmem_free(spa->spa_l2cache.sav_vdevs,
995 spa->spa_l2cache.sav_count * sizeof (void *));
996 spa->spa_l2cache.sav_vdevs = NULL;
998 if (spa->spa_l2cache.sav_config) {
999 nvlist_free(spa->spa_l2cache.sav_config);
1000 spa->spa_l2cache.sav_config = NULL;
1002 spa->spa_l2cache.sav_count = 0;
1004 spa->spa_async_suspended = 0;
1006 spa_config_exit(spa, SCL_ALL, FTAG);
1010 * Load (or re-load) the current list of vdevs describing the active spares for
1011 * this pool. When this is called, we have some form of basic information in
1012 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1013 * then re-generate a more complete list including status information.
1016 spa_load_spares(spa_t *spa)
1023 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1026 * First, close and free any existing spare vdevs.
1028 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1029 vd = spa->spa_spares.sav_vdevs[i];
1031 /* Undo the call to spa_activate() below */
1032 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1033 B_FALSE)) != NULL && tvd->vdev_isspare)
1034 spa_spare_remove(tvd);
1039 if (spa->spa_spares.sav_vdevs)
1040 kmem_free(spa->spa_spares.sav_vdevs,
1041 spa->spa_spares.sav_count * sizeof (void *));
1043 if (spa->spa_spares.sav_config == NULL)
1046 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1047 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1049 spa->spa_spares.sav_count = (int)nspares;
1050 spa->spa_spares.sav_vdevs = NULL;
1056 * Construct the array of vdevs, opening them to get status in the
1057 * process. For each spare, there is potentially two different vdev_t
1058 * structures associated with it: one in the list of spares (used only
1059 * for basic validation purposes) and one in the active vdev
1060 * configuration (if it's spared in). During this phase we open and
1061 * validate each vdev on the spare list. If the vdev also exists in the
1062 * active configuration, then we also mark this vdev as an active spare.
1064 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1066 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1067 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1068 VDEV_ALLOC_SPARE) == 0);
1071 spa->spa_spares.sav_vdevs[i] = vd;
1073 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1074 B_FALSE)) != NULL) {
1075 if (!tvd->vdev_isspare)
1079 * We only mark the spare active if we were successfully
1080 * able to load the vdev. Otherwise, importing a pool
1081 * with a bad active spare would result in strange
1082 * behavior, because multiple pool would think the spare
1083 * is actively in use.
1085 * There is a vulnerability here to an equally bizarre
1086 * circumstance, where a dead active spare is later
1087 * brought back to life (onlined or otherwise). Given
1088 * the rarity of this scenario, and the extra complexity
1089 * it adds, we ignore the possibility.
1091 if (!vdev_is_dead(tvd))
1092 spa_spare_activate(tvd);
1096 vd->vdev_aux = &spa->spa_spares;
1098 if (vdev_open(vd) != 0)
1101 if (vdev_validate_aux(vd) == 0)
1106 * Recompute the stashed list of spares, with status information
1109 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1110 DATA_TYPE_NVLIST_ARRAY) == 0);
1112 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1114 for (i = 0; i < spa->spa_spares.sav_count; i++)
1115 spares[i] = vdev_config_generate(spa,
1116 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1117 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1118 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1119 for (i = 0; i < spa->spa_spares.sav_count; i++)
1120 nvlist_free(spares[i]);
1121 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1125 * Load (or re-load) the current list of vdevs describing the active l2cache for
1126 * this pool. When this is called, we have some form of basic information in
1127 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1128 * then re-generate a more complete list including status information.
1129 * Devices which are already active have their details maintained, and are
1133 spa_load_l2cache(spa_t *spa)
1137 int i, j, oldnvdevs;
1139 vdev_t *vd, **oldvdevs, **newvdevs;
1140 spa_aux_vdev_t *sav = &spa->spa_l2cache;
1142 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1144 if (sav->sav_config != NULL) {
1145 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1146 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1147 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1152 oldvdevs = sav->sav_vdevs;
1153 oldnvdevs = sav->sav_count;
1154 sav->sav_vdevs = NULL;
1158 * Process new nvlist of vdevs.
1160 for (i = 0; i < nl2cache; i++) {
1161 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1165 for (j = 0; j < oldnvdevs; j++) {
1167 if (vd != NULL && guid == vd->vdev_guid) {
1169 * Retain previous vdev for add/remove ops.
1177 if (newvdevs[i] == NULL) {
1181 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1182 VDEV_ALLOC_L2CACHE) == 0);
1187 * Commit this vdev as an l2cache device,
1188 * even if it fails to open.
1190 spa_l2cache_add(vd);
1195 spa_l2cache_activate(vd);
1197 if (vdev_open(vd) != 0)
1200 (void) vdev_validate_aux(vd);
1202 if (!vdev_is_dead(vd))
1203 l2arc_add_vdev(spa, vd);
1208 * Purge vdevs that were dropped
1210 for (i = 0; i < oldnvdevs; i++) {
1215 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1216 pool != 0ULL && l2arc_vdev_present(vd))
1217 l2arc_remove_vdev(vd);
1218 (void) vdev_close(vd);
1219 spa_l2cache_remove(vd);
1224 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1226 if (sav->sav_config == NULL)
1229 sav->sav_vdevs = newvdevs;
1230 sav->sav_count = (int)nl2cache;
1233 * Recompute the stashed list of l2cache devices, with status
1234 * information this time.
1236 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1237 DATA_TYPE_NVLIST_ARRAY) == 0);
1239 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1240 for (i = 0; i < sav->sav_count; i++)
1241 l2cache[i] = vdev_config_generate(spa,
1242 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1243 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1244 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1246 for (i = 0; i < sav->sav_count; i++)
1247 nvlist_free(l2cache[i]);
1249 kmem_free(l2cache, sav->sav_count * sizeof (void *));
1253 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1256 char *packed = NULL;
1261 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
1262 nvsize = *(uint64_t *)db->db_data;
1263 dmu_buf_rele(db, FTAG);
1265 packed = kmem_alloc(nvsize, KM_SLEEP);
1266 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1269 error = nvlist_unpack(packed, nvsize, value, 0);
1270 kmem_free(packed, nvsize);
1276 * Checks to see if the given vdev could not be opened, in which case we post a
1277 * sysevent to notify the autoreplace code that the device has been removed.
1280 spa_check_removed(vdev_t *vd)
1282 for (int c = 0; c < vd->vdev_children; c++)
1283 spa_check_removed(vd->vdev_child[c]);
1285 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd)) {
1286 zfs_post_autoreplace(vd->vdev_spa, vd);
1287 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1292 * Validate the current config against the MOS config
1295 spa_config_valid(spa_t *spa, nvlist_t *config)
1297 vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1300 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1302 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1303 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1305 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1308 * If we're doing a normal import, then build up any additional
1309 * diagnostic information about missing devices in this config.
1310 * We'll pass this up to the user for further processing.
1312 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1313 nvlist_t **child, *nv;
1316 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1318 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1320 for (int c = 0; c < rvd->vdev_children; c++) {
1321 vdev_t *tvd = rvd->vdev_child[c];
1322 vdev_t *mtvd = mrvd->vdev_child[c];
1324 if (tvd->vdev_ops == &vdev_missing_ops &&
1325 mtvd->vdev_ops != &vdev_missing_ops &&
1327 child[idx++] = vdev_config_generate(spa, mtvd,
1332 VERIFY(nvlist_add_nvlist_array(nv,
1333 ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1334 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1335 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1337 for (int i = 0; i < idx; i++)
1338 nvlist_free(child[i]);
1341 kmem_free(child, rvd->vdev_children * sizeof (char **));
1345 * Compare the root vdev tree with the information we have
1346 * from the MOS config (mrvd). Check each top-level vdev
1347 * with the corresponding MOS config top-level (mtvd).
1349 for (int c = 0; c < rvd->vdev_children; c++) {
1350 vdev_t *tvd = rvd->vdev_child[c];
1351 vdev_t *mtvd = mrvd->vdev_child[c];
1354 * Resolve any "missing" vdevs in the current configuration.
1355 * If we find that the MOS config has more accurate information
1356 * about the top-level vdev then use that vdev instead.
1358 if (tvd->vdev_ops == &vdev_missing_ops &&
1359 mtvd->vdev_ops != &vdev_missing_ops) {
1361 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1365 * Device specific actions.
1367 if (mtvd->vdev_islog) {
1368 spa_set_log_state(spa, SPA_LOG_CLEAR);
1371 * XXX - once we have 'readonly' pool
1372 * support we should be able to handle
1373 * missing data devices by transitioning
1374 * the pool to readonly.
1380 * Swap the missing vdev with the data we were
1381 * able to obtain from the MOS config.
1383 vdev_remove_child(rvd, tvd);
1384 vdev_remove_child(mrvd, mtvd);
1386 vdev_add_child(rvd, mtvd);
1387 vdev_add_child(mrvd, tvd);
1389 spa_config_exit(spa, SCL_ALL, FTAG);
1391 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1394 } else if (mtvd->vdev_islog) {
1396 * Load the slog device's state from the MOS config
1397 * since it's possible that the label does not
1398 * contain the most up-to-date information.
1400 vdev_load_log_state(tvd, mtvd);
1405 spa_config_exit(spa, SCL_ALL, FTAG);
1408 * Ensure we were able to validate the config.
1410 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1414 * Check for missing log devices
1417 spa_check_logs(spa_t *spa)
1419 switch (spa->spa_log_state) {
1420 case SPA_LOG_MISSING:
1421 /* need to recheck in case slog has been restored */
1422 case SPA_LOG_UNKNOWN:
1423 if (dmu_objset_find(spa->spa_name, zil_check_log_chain, NULL,
1424 DS_FIND_CHILDREN)) {
1425 spa_set_log_state(spa, SPA_LOG_MISSING);
1434 spa_passivate_log(spa_t *spa)
1436 vdev_t *rvd = spa->spa_root_vdev;
1437 boolean_t slog_found = B_FALSE;
1439 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1441 if (!spa_has_slogs(spa))
1444 for (int c = 0; c < rvd->vdev_children; c++) {
1445 vdev_t *tvd = rvd->vdev_child[c];
1446 metaslab_group_t *mg = tvd->vdev_mg;
1448 if (tvd->vdev_islog) {
1449 metaslab_group_passivate(mg);
1450 slog_found = B_TRUE;
1454 return (slog_found);
1458 spa_activate_log(spa_t *spa)
1460 vdev_t *rvd = spa->spa_root_vdev;
1462 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1464 for (int c = 0; c < rvd->vdev_children; c++) {
1465 vdev_t *tvd = rvd->vdev_child[c];
1466 metaslab_group_t *mg = tvd->vdev_mg;
1468 if (tvd->vdev_islog)
1469 metaslab_group_activate(mg);
1474 spa_offline_log(spa_t *spa)
1478 if ((error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1479 NULL, DS_FIND_CHILDREN)) == 0) {
1482 * We successfully offlined the log device, sync out the
1483 * current txg so that the "stubby" block can be removed
1486 txg_wait_synced(spa->spa_dsl_pool, 0);
1492 spa_aux_check_removed(spa_aux_vdev_t *sav)
1494 for (int i = 0; i < sav->sav_count; i++)
1495 spa_check_removed(sav->sav_vdevs[i]);
1499 spa_claim_notify(zio_t *zio)
1501 spa_t *spa = zio->io_spa;
1506 mutex_enter(&spa->spa_props_lock); /* any mutex will do */
1507 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1508 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1509 mutex_exit(&spa->spa_props_lock);
1512 typedef struct spa_load_error {
1513 uint64_t sle_meta_count;
1514 uint64_t sle_data_count;
1518 spa_load_verify_done(zio_t *zio)
1520 blkptr_t *bp = zio->io_bp;
1521 spa_load_error_t *sle = zio->io_private;
1522 dmu_object_type_t type = BP_GET_TYPE(bp);
1523 int error = zio->io_error;
1526 if ((BP_GET_LEVEL(bp) != 0 || dmu_ot[type].ot_metadata) &&
1527 type != DMU_OT_INTENT_LOG)
1528 atomic_add_64(&sle->sle_meta_count, 1);
1530 atomic_add_64(&sle->sle_data_count, 1);
1532 zio_data_buf_free(zio->io_data, zio->io_size);
1537 spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1538 arc_buf_t *pbuf, const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1542 size_t size = BP_GET_PSIZE(bp);
1543 void *data = zio_data_buf_alloc(size);
1545 zio_nowait(zio_read(rio, spa, bp, data, size,
1546 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1547 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1548 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1554 spa_load_verify(spa_t *spa)
1557 spa_load_error_t sle = { 0 };
1558 zpool_rewind_policy_t policy;
1559 boolean_t verify_ok = B_FALSE;
1562 zpool_get_rewind_policy(spa->spa_config, &policy);
1564 if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1567 rio = zio_root(spa, NULL, &sle,
1568 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1570 error = traverse_pool(spa, spa->spa_verify_min_txg,
1571 TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio);
1573 (void) zio_wait(rio);
1575 spa->spa_load_meta_errors = sle.sle_meta_count;
1576 spa->spa_load_data_errors = sle.sle_data_count;
1578 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1579 sle.sle_data_count <= policy.zrp_maxdata) {
1583 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1584 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1586 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1587 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1588 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1589 VERIFY(nvlist_add_int64(spa->spa_load_info,
1590 ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1591 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1592 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1594 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1598 if (error != ENXIO && error != EIO)
1603 return (verify_ok ? 0 : EIO);
1607 * Find a value in the pool props object.
1610 spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1612 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1613 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1617 * Find a value in the pool directory object.
1620 spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1622 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1623 name, sizeof (uint64_t), 1, val));
1627 spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1629 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1634 * Fix up config after a partly-completed split. This is done with the
1635 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
1636 * pool have that entry in their config, but only the splitting one contains
1637 * a list of all the guids of the vdevs that are being split off.
1639 * This function determines what to do with that list: either rejoin
1640 * all the disks to the pool, or complete the splitting process. To attempt
1641 * the rejoin, each disk that is offlined is marked online again, and
1642 * we do a reopen() call. If the vdev label for every disk that was
1643 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1644 * then we call vdev_split() on each disk, and complete the split.
1646 * Otherwise we leave the config alone, with all the vdevs in place in
1647 * the original pool.
1650 spa_try_repair(spa_t *spa, nvlist_t *config)
1657 boolean_t attempt_reopen;
1659 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
1662 /* check that the config is complete */
1663 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
1664 &glist, &gcount) != 0)
1667 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
1669 /* attempt to online all the vdevs & validate */
1670 attempt_reopen = B_TRUE;
1671 for (i = 0; i < gcount; i++) {
1672 if (glist[i] == 0) /* vdev is hole */
1675 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
1676 if (vd[i] == NULL) {
1678 * Don't bother attempting to reopen the disks;
1679 * just do the split.
1681 attempt_reopen = B_FALSE;
1683 /* attempt to re-online it */
1684 vd[i]->vdev_offline = B_FALSE;
1688 if (attempt_reopen) {
1689 vdev_reopen(spa->spa_root_vdev);
1691 /* check each device to see what state it's in */
1692 for (extracted = 0, i = 0; i < gcount; i++) {
1693 if (vd[i] != NULL &&
1694 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
1701 * If every disk has been moved to the new pool, or if we never
1702 * even attempted to look at them, then we split them off for
1705 if (!attempt_reopen || gcount == extracted) {
1706 for (i = 0; i < gcount; i++)
1709 vdev_reopen(spa->spa_root_vdev);
1712 kmem_free(vd, gcount * sizeof (vdev_t *));
1716 spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
1717 boolean_t mosconfig)
1719 nvlist_t *config = spa->spa_config;
1720 char *ereport = FM_EREPORT_ZFS_POOL;
1725 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
1729 * Versioning wasn't explicitly added to the label until later, so if
1730 * it's not present treat it as the initial version.
1732 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
1733 &spa->spa_ubsync.ub_version) != 0)
1734 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
1736 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
1737 &spa->spa_config_txg);
1739 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
1740 spa_guid_exists(pool_guid, 0)) {
1743 spa->spa_load_guid = pool_guid;
1745 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
1747 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
1751 gethrestime(&spa->spa_loaded_ts);
1752 error = spa_load_impl(spa, pool_guid, config, state, type,
1753 mosconfig, &ereport);
1756 spa->spa_minref = refcount_count(&spa->spa_refcount);
1758 if (error != EEXIST) {
1759 spa->spa_loaded_ts.tv_sec = 0;
1760 spa->spa_loaded_ts.tv_nsec = 0;
1762 if (error != EBADF) {
1763 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
1766 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
1773 * Load an existing storage pool, using the pool's builtin spa_config as a
1774 * source of configuration information.
1777 spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
1778 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
1782 nvlist_t *nvroot = NULL;
1784 uberblock_t *ub = &spa->spa_uberblock;
1785 uint64_t children, config_cache_txg = spa->spa_config_txg;
1786 int orig_mode = spa->spa_mode;
1791 * If this is an untrusted config, access the pool in read-only mode.
1792 * This prevents things like resilvering recently removed devices.
1795 spa->spa_mode = FREAD;
1797 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1799 spa->spa_load_state = state;
1801 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
1804 parse = (type == SPA_IMPORT_EXISTING ?
1805 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
1808 * Create "The Godfather" zio to hold all async IOs
1810 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
1811 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
1814 * Parse the configuration into a vdev tree. We explicitly set the
1815 * value that will be returned by spa_version() since parsing the
1816 * configuration requires knowing the version number.
1818 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1819 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
1820 spa_config_exit(spa, SCL_ALL, FTAG);
1825 ASSERT(spa->spa_root_vdev == rvd);
1827 if (type != SPA_IMPORT_ASSEMBLE) {
1828 ASSERT(spa_guid(spa) == pool_guid);
1832 * Try to open all vdevs, loading each label in the process.
1834 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1835 error = vdev_open(rvd);
1836 spa_config_exit(spa, SCL_ALL, FTAG);
1841 * We need to validate the vdev labels against the configuration that
1842 * we have in hand, which is dependent on the setting of mosconfig. If
1843 * mosconfig is true then we're validating the vdev labels based on
1844 * that config. Otherwise, we're validating against the cached config
1845 * (zpool.cache) that was read when we loaded the zfs module, and then
1846 * later we will recursively call spa_load() and validate against
1849 * If we're assembling a new pool that's been split off from an
1850 * existing pool, the labels haven't yet been updated so we skip
1851 * validation for now.
1853 if (type != SPA_IMPORT_ASSEMBLE) {
1854 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1855 error = vdev_validate(rvd);
1856 spa_config_exit(spa, SCL_ALL, FTAG);
1861 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
1866 * Find the best uberblock.
1868 vdev_uberblock_load(NULL, rvd, ub);
1871 * If we weren't able to find a single valid uberblock, return failure.
1873 if (ub->ub_txg == 0)
1874 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
1877 * If the pool is newer than the code, we can't open it.
1879 if (ub->ub_version > SPA_VERSION)
1880 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
1883 * If the vdev guid sum doesn't match the uberblock, we have an
1884 * incomplete configuration. We first check to see if the pool
1885 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
1886 * If it is, defer the vdev_guid_sum check till later so we
1887 * can handle missing vdevs.
1889 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
1890 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
1891 rvd->vdev_guid_sum != ub->ub_guid_sum)
1892 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
1894 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
1895 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1896 spa_try_repair(spa, config);
1897 spa_config_exit(spa, SCL_ALL, FTAG);
1898 nvlist_free(spa->spa_config_splitting);
1899 spa->spa_config_splitting = NULL;
1903 * Initialize internal SPA structures.
1905 spa->spa_state = POOL_STATE_ACTIVE;
1906 spa->spa_ubsync = spa->spa_uberblock;
1907 spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
1908 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
1909 spa->spa_first_txg = spa->spa_last_ubsync_txg ?
1910 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
1911 spa->spa_claim_max_txg = spa->spa_first_txg;
1912 spa->spa_prev_software_version = ub->ub_software_version;
1914 error = dsl_pool_open(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
1916 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1917 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
1919 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
1920 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1924 nvlist_t *policy = NULL, *nvconfig;
1926 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
1927 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1929 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
1930 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
1932 unsigned long myhostid = 0;
1934 VERIFY(nvlist_lookup_string(nvconfig,
1935 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
1938 myhostid = zone_get_hostid(NULL);
1941 * We're emulating the system's hostid in userland, so
1942 * we can't use zone_get_hostid().
1944 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
1945 #endif /* _KERNEL */
1946 if (hostid != 0 && myhostid != 0 &&
1947 hostid != myhostid) {
1948 nvlist_free(nvconfig);
1949 cmn_err(CE_WARN, "pool '%s' could not be "
1950 "loaded as it was last accessed by "
1951 "another system (host: %s hostid: 0x%lx). "
1952 "See: http://www.sun.com/msg/ZFS-8000-EY",
1953 spa_name(spa), hostname,
1954 (unsigned long)hostid);
1958 if (nvlist_lookup_nvlist(spa->spa_config,
1959 ZPOOL_REWIND_POLICY, &policy) == 0)
1960 VERIFY(nvlist_add_nvlist(nvconfig,
1961 ZPOOL_REWIND_POLICY, policy) == 0);
1963 spa_config_set(spa, nvconfig);
1965 spa_deactivate(spa);
1966 spa_activate(spa, orig_mode);
1968 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
1971 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
1972 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1973 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
1975 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1978 * Load the bit that tells us to use the new accounting function
1979 * (raid-z deflation). If we have an older pool, this will not
1982 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
1983 if (error != 0 && error != ENOENT)
1984 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1986 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
1987 &spa->spa_creation_version);
1988 if (error != 0 && error != ENOENT)
1989 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1992 * Load the persistent error log. If we have an older pool, this will
1995 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
1996 if (error != 0 && error != ENOENT)
1997 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1999 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2000 &spa->spa_errlog_scrub);
2001 if (error != 0 && error != ENOENT)
2002 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2005 * Load the history object. If we have an older pool, this
2006 * will not be present.
2008 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2009 if (error != 0 && error != ENOENT)
2010 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2013 * If we're assembling the pool from the split-off vdevs of
2014 * an existing pool, we don't want to attach the spares & cache
2019 * Load any hot spares for this pool.
2021 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2022 if (error != 0 && error != ENOENT)
2023 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2024 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2025 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2026 if (load_nvlist(spa, spa->spa_spares.sav_object,
2027 &spa->spa_spares.sav_config) != 0)
2028 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2030 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2031 spa_load_spares(spa);
2032 spa_config_exit(spa, SCL_ALL, FTAG);
2033 } else if (error == 0) {
2034 spa->spa_spares.sav_sync = B_TRUE;
2038 * Load any level 2 ARC devices for this pool.
2040 error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2041 &spa->spa_l2cache.sav_object);
2042 if (error != 0 && error != ENOENT)
2043 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2044 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2045 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2046 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2047 &spa->spa_l2cache.sav_config) != 0)
2048 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2050 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2051 spa_load_l2cache(spa);
2052 spa_config_exit(spa, SCL_ALL, FTAG);
2053 } else if (error == 0) {
2054 spa->spa_l2cache.sav_sync = B_TRUE;
2057 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2059 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2060 if (error && error != ENOENT)
2061 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2064 uint64_t autoreplace;
2066 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2067 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2068 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2069 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2070 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2071 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2072 &spa->spa_dedup_ditto);
2074 spa->spa_autoreplace = (autoreplace != 0);
2078 * If the 'autoreplace' property is set, then post a resource notifying
2079 * the ZFS DE that it should not issue any faults for unopenable
2080 * devices. We also iterate over the vdevs, and post a sysevent for any
2081 * unopenable vdevs so that the normal autoreplace handler can take
2084 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2085 spa_check_removed(spa->spa_root_vdev);
2087 * For the import case, this is done in spa_import(), because
2088 * at this point we're using the spare definitions from
2089 * the MOS config, not necessarily from the userland config.
2091 if (state != SPA_LOAD_IMPORT) {
2092 spa_aux_check_removed(&spa->spa_spares);
2093 spa_aux_check_removed(&spa->spa_l2cache);
2098 * Load the vdev state for all toplevel vdevs.
2103 * Propagate the leaf DTLs we just loaded all the way up the tree.
2105 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2106 vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2107 spa_config_exit(spa, SCL_ALL, FTAG);
2110 * Load the DDTs (dedup tables).
2112 error = ddt_load(spa);
2114 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2116 spa_update_dspace(spa);
2119 * Validate the config, using the MOS config to fill in any
2120 * information which might be missing. If we fail to validate
2121 * the config then declare the pool unfit for use. If we're
2122 * assembling a pool from a split, the log is not transferred
2125 if (type != SPA_IMPORT_ASSEMBLE) {
2128 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2129 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2131 if (!spa_config_valid(spa, nvconfig)) {
2132 nvlist_free(nvconfig);
2133 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2136 nvlist_free(nvconfig);
2139 * Now that we've validate the config, check the state of the
2140 * root vdev. If it can't be opened, it indicates one or
2141 * more toplevel vdevs are faulted.
2143 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2146 if (spa_check_logs(spa)) {
2147 *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2148 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2153 * We've successfully opened the pool, verify that we're ready
2154 * to start pushing transactions.
2156 if (state != SPA_LOAD_TRYIMPORT) {
2157 if (error = spa_load_verify(spa))
2158 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2162 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2163 spa->spa_load_max_txg == UINT64_MAX)) {
2165 int need_update = B_FALSE;
2167 ASSERT(state != SPA_LOAD_TRYIMPORT);
2170 * Claim log blocks that haven't been committed yet.
2171 * This must all happen in a single txg.
2172 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2173 * invoked from zil_claim_log_block()'s i/o done callback.
2174 * Price of rollback is that we abandon the log.
2176 spa->spa_claiming = B_TRUE;
2178 tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2179 spa_first_txg(spa));
2180 (void) dmu_objset_find(spa_name(spa),
2181 zil_claim, tx, DS_FIND_CHILDREN);
2184 spa->spa_claiming = B_FALSE;
2186 spa_set_log_state(spa, SPA_LOG_GOOD);
2187 spa->spa_sync_on = B_TRUE;
2188 txg_sync_start(spa->spa_dsl_pool);
2191 * Wait for all claims to sync. We sync up to the highest
2192 * claimed log block birth time so that claimed log blocks
2193 * don't appear to be from the future. spa_claim_max_txg
2194 * will have been set for us by either zil_check_log_chain()
2195 * (invoked from spa_check_logs()) or zil_claim() above.
2197 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2200 * If the config cache is stale, or we have uninitialized
2201 * metaslabs (see spa_vdev_add()), then update the config.
2203 * If this is a verbatim import, trust the current
2204 * in-core spa_config and update the disk labels.
2206 if (config_cache_txg != spa->spa_config_txg ||
2207 state == SPA_LOAD_IMPORT ||
2208 state == SPA_LOAD_RECOVER ||
2209 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2210 need_update = B_TRUE;
2212 for (int c = 0; c < rvd->vdev_children; c++)
2213 if (rvd->vdev_child[c]->vdev_ms_array == 0)
2214 need_update = B_TRUE;
2217 * Update the config cache asychronously in case we're the
2218 * root pool, in which case the config cache isn't writable yet.
2221 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2224 * Check all DTLs to see if anything needs resilvering.
2226 if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2227 vdev_resilver_needed(rvd, NULL, NULL))
2228 spa_async_request(spa, SPA_ASYNC_RESILVER);
2231 * Delete any inconsistent datasets.
2233 (void) dmu_objset_find(spa_name(spa),
2234 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2237 * Clean up any stale temporary dataset userrefs.
2239 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2246 spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2248 int mode = spa->spa_mode;
2251 spa_deactivate(spa);
2253 spa->spa_load_max_txg--;
2255 spa_activate(spa, mode);
2256 spa_async_suspend(spa);
2258 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2262 spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2263 uint64_t max_request, int rewind_flags)
2265 nvlist_t *config = NULL;
2266 int load_error, rewind_error;
2267 uint64_t safe_rewind_txg;
2270 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2271 spa->spa_load_max_txg = spa->spa_load_txg;
2272 spa_set_log_state(spa, SPA_LOG_CLEAR);
2274 spa->spa_load_max_txg = max_request;
2277 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2279 if (load_error == 0)
2282 if (spa->spa_root_vdev != NULL)
2283 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2285 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2286 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2288 if (rewind_flags & ZPOOL_NEVER_REWIND) {
2289 nvlist_free(config);
2290 return (load_error);
2293 /* Price of rolling back is discarding txgs, including log */
2294 if (state == SPA_LOAD_RECOVER)
2295 spa_set_log_state(spa, SPA_LOG_CLEAR);
2297 spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2298 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2299 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2300 TXG_INITIAL : safe_rewind_txg;
2303 * Continue as long as we're finding errors, we're still within
2304 * the acceptable rewind range, and we're still finding uberblocks
2306 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2307 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2308 if (spa->spa_load_max_txg < safe_rewind_txg)
2309 spa->spa_extreme_rewind = B_TRUE;
2310 rewind_error = spa_load_retry(spa, state, mosconfig);
2313 spa->spa_extreme_rewind = B_FALSE;
2314 spa->spa_load_max_txg = UINT64_MAX;
2316 if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2317 spa_config_set(spa, config);
2319 return (state == SPA_LOAD_RECOVER ? rewind_error : load_error);
2325 * The import case is identical to an open except that the configuration is sent
2326 * down from userland, instead of grabbed from the configuration cache. For the
2327 * case of an open, the pool configuration will exist in the
2328 * POOL_STATE_UNINITIALIZED state.
2330 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2331 * the same time open the pool, without having to keep around the spa_t in some
2335 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2339 spa_load_state_t state = SPA_LOAD_OPEN;
2341 int locked = B_FALSE;
2346 * As disgusting as this is, we need to support recursive calls to this
2347 * function because dsl_dir_open() is called during spa_load(), and ends
2348 * up calling spa_open() again. The real fix is to figure out how to
2349 * avoid dsl_dir_open() calling this in the first place.
2351 if (mutex_owner(&spa_namespace_lock) != curthread) {
2352 mutex_enter(&spa_namespace_lock);
2356 if ((spa = spa_lookup(pool)) == NULL) {
2358 mutex_exit(&spa_namespace_lock);
2362 if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2363 zpool_rewind_policy_t policy;
2365 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2367 if (policy.zrp_request & ZPOOL_DO_REWIND)
2368 state = SPA_LOAD_RECOVER;
2370 spa_activate(spa, spa_mode_global);
2372 if (state != SPA_LOAD_RECOVER)
2373 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2375 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2376 policy.zrp_request);
2378 if (error == EBADF) {
2380 * If vdev_validate() returns failure (indicated by
2381 * EBADF), it indicates that one of the vdevs indicates
2382 * that the pool has been exported or destroyed. If
2383 * this is the case, the config cache is out of sync and
2384 * we should remove the pool from the namespace.
2387 spa_deactivate(spa);
2388 spa_config_sync(spa, B_TRUE, B_TRUE);
2391 mutex_exit(&spa_namespace_lock);
2397 * We can't open the pool, but we still have useful
2398 * information: the state of each vdev after the
2399 * attempted vdev_open(). Return this to the user.
2401 if (config != NULL && spa->spa_config) {
2402 VERIFY(nvlist_dup(spa->spa_config, config,
2404 VERIFY(nvlist_add_nvlist(*config,
2405 ZPOOL_CONFIG_LOAD_INFO,
2406 spa->spa_load_info) == 0);
2409 spa_deactivate(spa);
2410 spa->spa_last_open_failed = error;
2412 mutex_exit(&spa_namespace_lock);
2418 spa_open_ref(spa, tag);
2421 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2424 * If we've recovered the pool, pass back any information we
2425 * gathered while doing the load.
2427 if (state == SPA_LOAD_RECOVER) {
2428 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
2429 spa->spa_load_info) == 0);
2433 spa->spa_last_open_failed = 0;
2434 spa->spa_last_ubsync_txg = 0;
2435 spa->spa_load_txg = 0;
2436 mutex_exit(&spa_namespace_lock);
2445 spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
2448 return (spa_open_common(name, spapp, tag, policy, config));
2452 spa_open(const char *name, spa_t **spapp, void *tag)
2454 return (spa_open_common(name, spapp, tag, NULL, NULL));
2458 * Lookup the given spa_t, incrementing the inject count in the process,
2459 * preventing it from being exported or destroyed.
2462 spa_inject_addref(char *name)
2466 mutex_enter(&spa_namespace_lock);
2467 if ((spa = spa_lookup(name)) == NULL) {
2468 mutex_exit(&spa_namespace_lock);
2471 spa->spa_inject_ref++;
2472 mutex_exit(&spa_namespace_lock);
2478 spa_inject_delref(spa_t *spa)
2480 mutex_enter(&spa_namespace_lock);
2481 spa->spa_inject_ref--;
2482 mutex_exit(&spa_namespace_lock);
2486 * Add spares device information to the nvlist.
2489 spa_add_spares(spa_t *spa, nvlist_t *config)
2499 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2501 if (spa->spa_spares.sav_count == 0)
2504 VERIFY(nvlist_lookup_nvlist(config,
2505 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2506 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
2507 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2509 VERIFY(nvlist_add_nvlist_array(nvroot,
2510 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2511 VERIFY(nvlist_lookup_nvlist_array(nvroot,
2512 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2515 * Go through and find any spares which have since been
2516 * repurposed as an active spare. If this is the case, update
2517 * their status appropriately.
2519 for (i = 0; i < nspares; i++) {
2520 VERIFY(nvlist_lookup_uint64(spares[i],
2521 ZPOOL_CONFIG_GUID, &guid) == 0);
2522 if (spa_spare_exists(guid, &pool, NULL) &&
2524 VERIFY(nvlist_lookup_uint64_array(
2525 spares[i], ZPOOL_CONFIG_VDEV_STATS,
2526 (uint64_t **)&vs, &vsc) == 0);
2527 vs->vs_state = VDEV_STATE_CANT_OPEN;
2528 vs->vs_aux = VDEV_AUX_SPARED;
2535 * Add l2cache device information to the nvlist, including vdev stats.
2538 spa_add_l2cache(spa_t *spa, nvlist_t *config)
2541 uint_t i, j, nl2cache;
2548 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2550 if (spa->spa_l2cache.sav_count == 0)
2553 VERIFY(nvlist_lookup_nvlist(config,
2554 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2555 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
2556 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2557 if (nl2cache != 0) {
2558 VERIFY(nvlist_add_nvlist_array(nvroot,
2559 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2560 VERIFY(nvlist_lookup_nvlist_array(nvroot,
2561 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2564 * Update level 2 cache device stats.
2567 for (i = 0; i < nl2cache; i++) {
2568 VERIFY(nvlist_lookup_uint64(l2cache[i],
2569 ZPOOL_CONFIG_GUID, &guid) == 0);
2572 for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
2574 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
2575 vd = spa->spa_l2cache.sav_vdevs[j];
2581 VERIFY(nvlist_lookup_uint64_array(l2cache[i],
2582 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
2584 vdev_get_stats(vd, vs);
2590 spa_get_stats(const char *name, nvlist_t **config, char *altroot, size_t buflen)
2596 error = spa_open_common(name, &spa, FTAG, NULL, config);
2600 * This still leaves a window of inconsistency where the spares
2601 * or l2cache devices could change and the config would be
2602 * self-inconsistent.
2604 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
2606 if (*config != NULL) {
2607 uint64_t loadtimes[2];
2609 loadtimes[0] = spa->spa_loaded_ts.tv_sec;
2610 loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
2611 VERIFY(nvlist_add_uint64_array(*config,
2612 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
2614 VERIFY(nvlist_add_uint64(*config,
2615 ZPOOL_CONFIG_ERRCOUNT,
2616 spa_get_errlog_size(spa)) == 0);
2618 if (spa_suspended(spa))
2619 VERIFY(nvlist_add_uint64(*config,
2620 ZPOOL_CONFIG_SUSPENDED,
2621 spa->spa_failmode) == 0);
2623 spa_add_spares(spa, *config);
2624 spa_add_l2cache(spa, *config);
2629 * We want to get the alternate root even for faulted pools, so we cheat
2630 * and call spa_lookup() directly.
2634 mutex_enter(&spa_namespace_lock);
2635 spa = spa_lookup(name);
2637 spa_altroot(spa, altroot, buflen);
2641 mutex_exit(&spa_namespace_lock);
2643 spa_altroot(spa, altroot, buflen);
2648 spa_config_exit(spa, SCL_CONFIG, FTAG);
2649 spa_close(spa, FTAG);
2656 * Validate that the auxiliary device array is well formed. We must have an
2657 * array of nvlists, each which describes a valid leaf vdev. If this is an
2658 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
2659 * specified, as long as they are well-formed.
2662 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
2663 spa_aux_vdev_t *sav, const char *config, uint64_t version,
2664 vdev_labeltype_t label)
2671 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
2674 * It's acceptable to have no devs specified.
2676 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
2683 * Make sure the pool is formatted with a version that supports this
2686 if (spa_version(spa) < version)
2690 * Set the pending device list so we correctly handle device in-use
2693 sav->sav_pending = dev;
2694 sav->sav_npending = ndev;
2696 for (i = 0; i < ndev; i++) {
2697 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
2701 if (!vd->vdev_ops->vdev_op_leaf) {
2708 * The L2ARC currently only supports disk devices in
2709 * kernel context. For user-level testing, we allow it.
2712 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
2713 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
2720 if ((error = vdev_open(vd)) == 0 &&
2721 (error = vdev_label_init(vd, crtxg, label)) == 0) {
2722 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
2723 vd->vdev_guid) == 0);
2729 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
2736 sav->sav_pending = NULL;
2737 sav->sav_npending = 0;
2742 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
2746 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
2748 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
2749 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
2750 VDEV_LABEL_SPARE)) != 0) {
2754 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
2755 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
2756 VDEV_LABEL_L2CACHE));
2760 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
2765 if (sav->sav_config != NULL) {
2771 * Generate new dev list by concatentating with the
2774 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
2775 &olddevs, &oldndevs) == 0);
2777 newdevs = kmem_alloc(sizeof (void *) *
2778 (ndevs + oldndevs), KM_SLEEP);
2779 for (i = 0; i < oldndevs; i++)
2780 VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
2782 for (i = 0; i < ndevs; i++)
2783 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
2786 VERIFY(nvlist_remove(sav->sav_config, config,
2787 DATA_TYPE_NVLIST_ARRAY) == 0);
2789 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
2790 config, newdevs, ndevs + oldndevs) == 0);
2791 for (i = 0; i < oldndevs + ndevs; i++)
2792 nvlist_free(newdevs[i]);
2793 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
2796 * Generate a new dev list.
2798 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
2800 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
2806 * Stop and drop level 2 ARC devices
2809 spa_l2cache_drop(spa_t *spa)
2813 spa_aux_vdev_t *sav = &spa->spa_l2cache;
2815 for (i = 0; i < sav->sav_count; i++) {
2818 vd = sav->sav_vdevs[i];
2821 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
2822 pool != 0ULL && l2arc_vdev_present(vd))
2823 l2arc_remove_vdev(vd);
2824 if (vd->vdev_isl2cache)
2825 spa_l2cache_remove(vd);
2826 vdev_clear_stats(vd);
2827 (void) vdev_close(vd);
2835 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
2836 const char *history_str, nvlist_t *zplprops)
2839 char *altroot = NULL;
2844 uint64_t txg = TXG_INITIAL;
2845 nvlist_t **spares, **l2cache;
2846 uint_t nspares, nl2cache;
2847 uint64_t version, obj;
2850 * If this pool already exists, return failure.
2852 mutex_enter(&spa_namespace_lock);
2853 if (spa_lookup(pool) != NULL) {
2854 mutex_exit(&spa_namespace_lock);
2859 * Allocate a new spa_t structure.
2861 (void) nvlist_lookup_string(props,
2862 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
2863 spa = spa_add(pool, NULL, altroot);
2864 spa_activate(spa, spa_mode_global);
2866 if (props && (error = spa_prop_validate(spa, props))) {
2867 spa_deactivate(spa);
2869 mutex_exit(&spa_namespace_lock);
2873 if (nvlist_lookup_uint64(props, zpool_prop_to_name(ZPOOL_PROP_VERSION),
2875 version = SPA_VERSION;
2876 ASSERT(version <= SPA_VERSION);
2878 spa->spa_first_txg = txg;
2879 spa->spa_uberblock.ub_txg = txg - 1;
2880 spa->spa_uberblock.ub_version = version;
2881 spa->spa_ubsync = spa->spa_uberblock;
2884 * Create "The Godfather" zio to hold all async IOs
2886 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2887 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2890 * Create the root vdev.
2892 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2894 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
2896 ASSERT(error != 0 || rvd != NULL);
2897 ASSERT(error != 0 || spa->spa_root_vdev == rvd);
2899 if (error == 0 && !zfs_allocatable_devs(nvroot))
2903 (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
2904 (error = spa_validate_aux(spa, nvroot, txg,
2905 VDEV_ALLOC_ADD)) == 0) {
2906 for (int c = 0; c < rvd->vdev_children; c++) {
2907 vdev_metaslab_set_size(rvd->vdev_child[c]);
2908 vdev_expand(rvd->vdev_child[c], txg);
2912 spa_config_exit(spa, SCL_ALL, FTAG);
2916 spa_deactivate(spa);
2918 mutex_exit(&spa_namespace_lock);
2923 * Get the list of spares, if specified.
2925 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
2926 &spares, &nspares) == 0) {
2927 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
2929 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
2930 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2931 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2932 spa_load_spares(spa);
2933 spa_config_exit(spa, SCL_ALL, FTAG);
2934 spa->spa_spares.sav_sync = B_TRUE;
2938 * Get the list of level 2 cache devices, if specified.
2940 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
2941 &l2cache, &nl2cache) == 0) {
2942 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
2943 NV_UNIQUE_NAME, KM_SLEEP) == 0);
2944 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
2945 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2946 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2947 spa_load_l2cache(spa);
2948 spa_config_exit(spa, SCL_ALL, FTAG);
2949 spa->spa_l2cache.sav_sync = B_TRUE;
2952 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
2953 spa->spa_meta_objset = dp->dp_meta_objset;
2956 * Create DDTs (dedup tables).
2960 spa_update_dspace(spa);
2962 tx = dmu_tx_create_assigned(dp, txg);
2965 * Create the pool config object.
2967 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
2968 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
2969 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
2971 if (zap_add(spa->spa_meta_objset,
2972 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
2973 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
2974 cmn_err(CE_PANIC, "failed to add pool config");
2977 if (zap_add(spa->spa_meta_objset,
2978 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
2979 sizeof (uint64_t), 1, &version, tx) != 0) {
2980 cmn_err(CE_PANIC, "failed to add pool version");
2983 /* Newly created pools with the right version are always deflated. */
2984 if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
2985 spa->spa_deflate = TRUE;
2986 if (zap_add(spa->spa_meta_objset,
2987 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
2988 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
2989 cmn_err(CE_PANIC, "failed to add deflate");
2994 * Create the deferred-free bpobj. Turn off compression
2995 * because sync-to-convergence takes longer if the blocksize
2998 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
2999 dmu_object_set_compress(spa->spa_meta_objset, obj,
3000 ZIO_COMPRESS_OFF, tx);
3001 if (zap_add(spa->spa_meta_objset,
3002 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3003 sizeof (uint64_t), 1, &obj, tx) != 0) {
3004 cmn_err(CE_PANIC, "failed to add bpobj");
3006 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3007 spa->spa_meta_objset, obj));
3010 * Create the pool's history object.
3012 if (version >= SPA_VERSION_ZPOOL_HISTORY)
3013 spa_history_create_obj(spa, tx);
3016 * Set pool properties.
3018 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3019 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3020 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3021 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3023 if (props != NULL) {
3024 spa_configfile_set(spa, props, B_FALSE);
3025 spa_sync_props(spa, props, tx);
3030 spa->spa_sync_on = B_TRUE;
3031 txg_sync_start(spa->spa_dsl_pool);
3034 * We explicitly wait for the first transaction to complete so that our
3035 * bean counters are appropriately updated.
3037 txg_wait_synced(spa->spa_dsl_pool, txg);
3039 spa_config_sync(spa, B_FALSE, B_TRUE);
3041 if (version >= SPA_VERSION_ZPOOL_HISTORY && history_str != NULL)
3042 (void) spa_history_log(spa, history_str, LOG_CMD_POOL_CREATE);
3043 spa_history_log_version(spa, LOG_POOL_CREATE);
3045 spa->spa_minref = refcount_count(&spa->spa_refcount);
3047 mutex_exit(&spa_namespace_lock);
3054 * Get the root pool information from the root disk, then import the root pool
3055 * during the system boot up time.
3057 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3060 spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3063 nvlist_t *nvtop, *nvroot;
3066 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3070 * Add this top-level vdev to the child array.
3072 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3074 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3076 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3079 * Put this pool's top-level vdevs into a root vdev.
3081 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3082 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3083 VDEV_TYPE_ROOT) == 0);
3084 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3085 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3086 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3090 * Replace the existing vdev_tree with the new root vdev in
3091 * this pool's configuration (remove the old, add the new).
3093 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3094 nvlist_free(nvroot);
3099 * Walk the vdev tree and see if we can find a device with "better"
3100 * configuration. A configuration is "better" if the label on that
3101 * device has a more recent txg.
3104 spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3106 for (int c = 0; c < vd->vdev_children; c++)
3107 spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3109 if (vd->vdev_ops->vdev_op_leaf) {
3113 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3117 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3121 * Do we have a better boot device?
3123 if (label_txg > *txg) {
3132 * Import a root pool.
3134 * For x86. devpath_list will consist of devid and/or physpath name of
3135 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3136 * The GRUB "findroot" command will return the vdev we should boot.
3138 * For Sparc, devpath_list consists the physpath name of the booting device
3139 * no matter the rootpool is a single device pool or a mirrored pool.
3141 * "/pci@1f,0/ide@d/disk@0,0:a"
3144 spa_import_rootpool(char *devpath, char *devid)
3147 vdev_t *rvd, *bvd, *avd = NULL;
3148 nvlist_t *config, *nvtop;
3154 * Read the label from the boot device and generate a configuration.
3156 config = spa_generate_rootconf(devpath, devid, &guid);
3157 #if defined(_OBP) && defined(_KERNEL)
3158 if (config == NULL) {
3159 if (strstr(devpath, "/iscsi/ssd") != NULL) {
3161 get_iscsi_bootpath_phy(devpath);
3162 config = spa_generate_rootconf(devpath, devid, &guid);
3166 if (config == NULL) {
3167 cmn_err(CE_NOTE, "Can not read the pool label from '%s'",
3172 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3174 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3176 mutex_enter(&spa_namespace_lock);
3177 if ((spa = spa_lookup(pname)) != NULL) {
3179 * Remove the existing root pool from the namespace so that we
3180 * can replace it with the correct config we just read in.
3185 spa = spa_add(pname, config, NULL);
3186 spa->spa_is_root = B_TRUE;
3187 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3190 * Build up a vdev tree based on the boot device's label config.
3192 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3194 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3195 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3196 VDEV_ALLOC_ROOTPOOL);
3197 spa_config_exit(spa, SCL_ALL, FTAG);
3199 mutex_exit(&spa_namespace_lock);
3200 nvlist_free(config);
3201 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3207 * Get the boot vdev.
3209 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3210 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3211 (u_longlong_t)guid);
3217 * Determine if there is a better boot device.
3220 spa_alt_rootvdev(rvd, &avd, &txg);
3222 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3223 "try booting from '%s'", avd->vdev_path);
3229 * If the boot device is part of a spare vdev then ensure that
3230 * we're booting off the active spare.
3232 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3233 !bvd->vdev_isspare) {
3234 cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3235 "try booting from '%s'",
3237 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3243 spa_history_log_version(spa, LOG_POOL_IMPORT);
3245 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3247 spa_config_exit(spa, SCL_ALL, FTAG);
3248 mutex_exit(&spa_namespace_lock);
3250 nvlist_free(config);
3257 * Import a non-root pool into the system.
3260 spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
3263 char *altroot = NULL;
3264 spa_load_state_t state = SPA_LOAD_IMPORT;
3265 zpool_rewind_policy_t policy;
3266 uint64_t mode = spa_mode_global;
3267 uint64_t readonly = B_FALSE;
3270 nvlist_t **spares, **l2cache;
3271 uint_t nspares, nl2cache;
3274 * If a pool with this name exists, return failure.
3276 mutex_enter(&spa_namespace_lock);
3277 if (spa_lookup(pool) != NULL) {
3278 mutex_exit(&spa_namespace_lock);
3283 * Create and initialize the spa structure.
3285 (void) nvlist_lookup_string(props,
3286 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3287 (void) nvlist_lookup_uint64(props,
3288 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
3291 spa = spa_add(pool, config, altroot);
3292 spa->spa_import_flags = flags;
3295 * Verbatim import - Take a pool and insert it into the namespace
3296 * as if it had been loaded at boot.
3298 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
3300 spa_configfile_set(spa, props, B_FALSE);
3302 spa_config_sync(spa, B_FALSE, B_TRUE);
3304 mutex_exit(&spa_namespace_lock);
3305 spa_history_log_version(spa, LOG_POOL_IMPORT);
3310 spa_activate(spa, mode);
3313 * Don't start async tasks until we know everything is healthy.
3315 spa_async_suspend(spa);
3317 zpool_get_rewind_policy(config, &policy);
3318 if (policy.zrp_request & ZPOOL_DO_REWIND)
3319 state = SPA_LOAD_RECOVER;
3322 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
3323 * because the user-supplied config is actually the one to trust when
3326 if (state != SPA_LOAD_RECOVER)
3327 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3329 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
3330 policy.zrp_request);
3333 * Propagate anything learned while loading the pool and pass it
3334 * back to caller (i.e. rewind info, missing devices, etc).
3336 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
3337 spa->spa_load_info) == 0);
3339 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3341 * Toss any existing sparelist, as it doesn't have any validity
3342 * anymore, and conflicts with spa_has_spare().
3344 if (spa->spa_spares.sav_config) {
3345 nvlist_free(spa->spa_spares.sav_config);
3346 spa->spa_spares.sav_config = NULL;
3347 spa_load_spares(spa);
3349 if (spa->spa_l2cache.sav_config) {
3350 nvlist_free(spa->spa_l2cache.sav_config);
3351 spa->spa_l2cache.sav_config = NULL;
3352 spa_load_l2cache(spa);
3355 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3358 error = spa_validate_aux(spa, nvroot, -1ULL,
3361 error = spa_validate_aux(spa, nvroot, -1ULL,
3362 VDEV_ALLOC_L2CACHE);
3363 spa_config_exit(spa, SCL_ALL, FTAG);
3366 spa_configfile_set(spa, props, B_FALSE);
3368 if (error != 0 || (props && spa_writeable(spa) &&
3369 (error = spa_prop_set(spa, props)))) {
3371 spa_deactivate(spa);
3373 mutex_exit(&spa_namespace_lock);
3377 spa_async_resume(spa);
3380 * Override any spares and level 2 cache devices as specified by
3381 * the user, as these may have correct device names/devids, etc.
3383 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3384 &spares, &nspares) == 0) {
3385 if (spa->spa_spares.sav_config)
3386 VERIFY(nvlist_remove(spa->spa_spares.sav_config,
3387 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
3389 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
3390 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3391 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3392 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3393 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3394 spa_load_spares(spa);
3395 spa_config_exit(spa, SCL_ALL, FTAG);
3396 spa->spa_spares.sav_sync = B_TRUE;
3398 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3399 &l2cache, &nl2cache) == 0) {
3400 if (spa->spa_l2cache.sav_config)
3401 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
3402 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
3404 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3405 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3406 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3407 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3408 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3409 spa_load_l2cache(spa);
3410 spa_config_exit(spa, SCL_ALL, FTAG);
3411 spa->spa_l2cache.sav_sync = B_TRUE;
3415 * Check for any removed devices.
3417 if (spa->spa_autoreplace) {
3418 spa_aux_check_removed(&spa->spa_spares);
3419 spa_aux_check_removed(&spa->spa_l2cache);
3422 if (spa_writeable(spa)) {
3424 * Update the config cache to include the newly-imported pool.
3426 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3430 * It's possible that the pool was expanded while it was exported.
3431 * We kick off an async task to handle this for us.
3433 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
3435 mutex_exit(&spa_namespace_lock);
3436 spa_history_log_version(spa, LOG_POOL_IMPORT);
3442 spa_tryimport(nvlist_t *tryconfig)
3444 nvlist_t *config = NULL;
3450 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
3453 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
3457 * Create and initialize the spa structure.
3459 mutex_enter(&spa_namespace_lock);
3460 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
3461 spa_activate(spa, FREAD);
3464 * Pass off the heavy lifting to spa_load().
3465 * Pass TRUE for mosconfig because the user-supplied config
3466 * is actually the one to trust when doing an import.
3468 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
3471 * If 'tryconfig' was at least parsable, return the current config.
3473 if (spa->spa_root_vdev != NULL) {
3474 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
3475 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
3477 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
3479 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
3480 spa->spa_uberblock.ub_timestamp) == 0);
3483 * If the bootfs property exists on this pool then we
3484 * copy it out so that external consumers can tell which
3485 * pools are bootable.
3487 if ((!error || error == EEXIST) && spa->spa_bootfs) {
3488 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3491 * We have to play games with the name since the
3492 * pool was opened as TRYIMPORT_NAME.
3494 if (dsl_dsobj_to_dsname(spa_name(spa),
3495 spa->spa_bootfs, tmpname) == 0) {
3497 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3499 cp = strchr(tmpname, '/');
3501 (void) strlcpy(dsname, tmpname,
3504 (void) snprintf(dsname, MAXPATHLEN,
3505 "%s/%s", poolname, ++cp);
3507 VERIFY(nvlist_add_string(config,
3508 ZPOOL_CONFIG_BOOTFS, dsname) == 0);
3509 kmem_free(dsname, MAXPATHLEN);
3511 kmem_free(tmpname, MAXPATHLEN);
3515 * Add the list of hot spares and level 2 cache devices.
3517 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3518 spa_add_spares(spa, config);
3519 spa_add_l2cache(spa, config);
3520 spa_config_exit(spa, SCL_CONFIG, FTAG);
3524 spa_deactivate(spa);
3526 mutex_exit(&spa_namespace_lock);
3532 * Pool export/destroy
3534 * The act of destroying or exporting a pool is very simple. We make sure there
3535 * is no more pending I/O and any references to the pool are gone. Then, we
3536 * update the pool state and sync all the labels to disk, removing the
3537 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
3538 * we don't sync the labels or remove the configuration cache.
3541 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
3542 boolean_t force, boolean_t hardforce)
3549 if (!(spa_mode_global & FWRITE))
3552 mutex_enter(&spa_namespace_lock);
3553 if ((spa = spa_lookup(pool)) == NULL) {
3554 mutex_exit(&spa_namespace_lock);
3559 * Put a hold on the pool, drop the namespace lock, stop async tasks,
3560 * reacquire the namespace lock, and see if we can export.
3562 spa_open_ref(spa, FTAG);
3563 mutex_exit(&spa_namespace_lock);
3564 spa_async_suspend(spa);
3565 mutex_enter(&spa_namespace_lock);
3566 spa_close(spa, FTAG);
3569 * The pool will be in core if it's openable,
3570 * in which case we can modify its state.
3572 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
3574 * Objsets may be open only because they're dirty, so we
3575 * have to force it to sync before checking spa_refcnt.
3577 txg_wait_synced(spa->spa_dsl_pool, 0);
3580 * A pool cannot be exported or destroyed if there are active
3581 * references. If we are resetting a pool, allow references by
3582 * fault injection handlers.
3584 if (!spa_refcount_zero(spa) ||
3585 (spa->spa_inject_ref != 0 &&
3586 new_state != POOL_STATE_UNINITIALIZED)) {
3587 spa_async_resume(spa);
3588 mutex_exit(&spa_namespace_lock);
3593 * A pool cannot be exported if it has an active shared spare.
3594 * This is to prevent other pools stealing the active spare
3595 * from an exported pool. At user's own will, such pool can
3596 * be forcedly exported.
3598 if (!force && new_state == POOL_STATE_EXPORTED &&
3599 spa_has_active_shared_spare(spa)) {
3600 spa_async_resume(spa);
3601 mutex_exit(&spa_namespace_lock);
3606 * We want this to be reflected on every label,
3607 * so mark them all dirty. spa_unload() will do the
3608 * final sync that pushes these changes out.
3610 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
3611 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3612 spa->spa_state = new_state;
3613 spa->spa_final_txg = spa_last_synced_txg(spa) +
3615 vdev_config_dirty(spa->spa_root_vdev);
3616 spa_config_exit(spa, SCL_ALL, FTAG);
3620 spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
3622 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
3624 spa_deactivate(spa);
3627 if (oldconfig && spa->spa_config)
3628 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
3630 if (new_state != POOL_STATE_UNINITIALIZED) {
3632 spa_config_sync(spa, B_TRUE, B_TRUE);
3635 mutex_exit(&spa_namespace_lock);
3641 * Destroy a storage pool.
3644 spa_destroy(char *pool)
3646 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
3651 * Export a storage pool.
3654 spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
3655 boolean_t hardforce)
3657 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
3662 * Similar to spa_export(), this unloads the spa_t without actually removing it
3663 * from the namespace in any way.
3666 spa_reset(char *pool)
3668 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
3673 * ==========================================================================
3674 * Device manipulation
3675 * ==========================================================================
3679 * Add a device to a storage pool.
3682 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
3686 vdev_t *rvd = spa->spa_root_vdev;
3688 nvlist_t **spares, **l2cache;
3689 uint_t nspares, nl2cache;
3691 ASSERT(spa_writeable(spa));
3693 txg = spa_vdev_enter(spa);
3695 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
3696 VDEV_ALLOC_ADD)) != 0)
3697 return (spa_vdev_exit(spa, NULL, txg, error));
3699 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */
3701 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
3705 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
3709 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
3710 return (spa_vdev_exit(spa, vd, txg, EINVAL));
3712 if (vd->vdev_children != 0 &&
3713 (error = vdev_create(vd, txg, B_FALSE)) != 0)
3714 return (spa_vdev_exit(spa, vd, txg, error));
3717 * We must validate the spares and l2cache devices after checking the
3718 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
3720 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
3721 return (spa_vdev_exit(spa, vd, txg, error));
3724 * Transfer each new top-level vdev from vd to rvd.
3726 for (int c = 0; c < vd->vdev_children; c++) {
3729 * Set the vdev id to the first hole, if one exists.
3731 for (id = 0; id < rvd->vdev_children; id++) {
3732 if (rvd->vdev_child[id]->vdev_ishole) {
3733 vdev_free(rvd->vdev_child[id]);
3737 tvd = vd->vdev_child[c];
3738 vdev_remove_child(vd, tvd);
3740 vdev_add_child(rvd, tvd);
3741 vdev_config_dirty(tvd);
3745 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
3746 ZPOOL_CONFIG_SPARES);
3747 spa_load_spares(spa);
3748 spa->spa_spares.sav_sync = B_TRUE;
3751 if (nl2cache != 0) {
3752 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
3753 ZPOOL_CONFIG_L2CACHE);
3754 spa_load_l2cache(spa);
3755 spa->spa_l2cache.sav_sync = B_TRUE;
3759 * We have to be careful when adding new vdevs to an existing pool.
3760 * If other threads start allocating from these vdevs before we
3761 * sync the config cache, and we lose power, then upon reboot we may
3762 * fail to open the pool because there are DVAs that the config cache
3763 * can't translate. Therefore, we first add the vdevs without
3764 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
3765 * and then let spa_config_update() initialize the new metaslabs.
3767 * spa_load() checks for added-but-not-initialized vdevs, so that
3768 * if we lose power at any point in this sequence, the remaining
3769 * steps will be completed the next time we load the pool.
3771 (void) spa_vdev_exit(spa, vd, txg, 0);
3773 mutex_enter(&spa_namespace_lock);
3774 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3775 mutex_exit(&spa_namespace_lock);
3781 * Attach a device to a mirror. The arguments are the path to any device
3782 * in the mirror, and the nvroot for the new device. If the path specifies
3783 * a device that is not mirrored, we automatically insert the mirror vdev.
3785 * If 'replacing' is specified, the new device is intended to replace the
3786 * existing device; in this case the two devices are made into their own
3787 * mirror using the 'replacing' vdev, which is functionally identical to
3788 * the mirror vdev (it actually reuses all the same ops) but has a few
3789 * extra rules: you can't attach to it after it's been created, and upon
3790 * completion of resilvering, the first disk (the one being replaced)
3791 * is automatically detached.
3794 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
3796 uint64_t txg, dtl_max_txg;
3797 vdev_t *rvd = spa->spa_root_vdev;
3798 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
3800 char *oldvdpath, *newvdpath;
3804 ASSERT(spa_writeable(spa));
3806 txg = spa_vdev_enter(spa);
3808 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
3811 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
3813 if (!oldvd->vdev_ops->vdev_op_leaf)
3814 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3816 pvd = oldvd->vdev_parent;
3818 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
3819 VDEV_ALLOC_ADD)) != 0)
3820 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
3822 if (newrootvd->vdev_children != 1)
3823 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
3825 newvd = newrootvd->vdev_child[0];
3827 if (!newvd->vdev_ops->vdev_op_leaf)
3828 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
3830 if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
3831 return (spa_vdev_exit(spa, newrootvd, txg, error));
3834 * Spares can't replace logs
3836 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
3837 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3841 * For attach, the only allowable parent is a mirror or the root
3844 if (pvd->vdev_ops != &vdev_mirror_ops &&
3845 pvd->vdev_ops != &vdev_root_ops)
3846 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3848 pvops = &vdev_mirror_ops;
3851 * Active hot spares can only be replaced by inactive hot
3854 if (pvd->vdev_ops == &vdev_spare_ops &&
3855 oldvd->vdev_isspare &&
3856 !spa_has_spare(spa, newvd->vdev_guid))
3857 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3860 * If the source is a hot spare, and the parent isn't already a
3861 * spare, then we want to create a new hot spare. Otherwise, we
3862 * want to create a replacing vdev. The user is not allowed to
3863 * attach to a spared vdev child unless the 'isspare' state is
3864 * the same (spare replaces spare, non-spare replaces
3867 if (pvd->vdev_ops == &vdev_replacing_ops &&
3868 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
3869 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3870 } else if (pvd->vdev_ops == &vdev_spare_ops &&
3871 newvd->vdev_isspare != oldvd->vdev_isspare) {
3872 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3875 if (newvd->vdev_isspare)
3876 pvops = &vdev_spare_ops;
3878 pvops = &vdev_replacing_ops;
3882 * Make sure the new device is big enough.
3884 if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
3885 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
3888 * The new device cannot have a higher alignment requirement
3889 * than the top-level vdev.
3891 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
3892 return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
3895 * If this is an in-place replacement, update oldvd's path and devid
3896 * to make it distinguishable from newvd, and unopenable from now on.
3898 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
3899 spa_strfree(oldvd->vdev_path);
3900 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
3902 (void) sprintf(oldvd->vdev_path, "%s/%s",
3903 newvd->vdev_path, "old");
3904 if (oldvd->vdev_devid != NULL) {
3905 spa_strfree(oldvd->vdev_devid);
3906 oldvd->vdev_devid = NULL;
3910 /* mark the device being resilvered */
3911 newvd->vdev_resilvering = B_TRUE;
3914 * If the parent is not a mirror, or if we're replacing, insert the new
3915 * mirror/replacing/spare vdev above oldvd.
3917 if (pvd->vdev_ops != pvops)
3918 pvd = vdev_add_parent(oldvd, pvops);
3920 ASSERT(pvd->vdev_top->vdev_parent == rvd);
3921 ASSERT(pvd->vdev_ops == pvops);
3922 ASSERT(oldvd->vdev_parent == pvd);
3925 * Extract the new device from its root and add it to pvd.
3927 vdev_remove_child(newrootvd, newvd);
3928 newvd->vdev_id = pvd->vdev_children;
3929 newvd->vdev_crtxg = oldvd->vdev_crtxg;
3930 vdev_add_child(pvd, newvd);
3932 tvd = newvd->vdev_top;
3933 ASSERT(pvd->vdev_top == tvd);
3934 ASSERT(tvd->vdev_parent == rvd);
3936 vdev_config_dirty(tvd);
3939 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
3940 * for any dmu_sync-ed blocks. It will propagate upward when
3941 * spa_vdev_exit() calls vdev_dtl_reassess().
3943 dtl_max_txg = txg + TXG_CONCURRENT_STATES;
3945 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
3946 dtl_max_txg - TXG_INITIAL);
3948 if (newvd->vdev_isspare) {
3949 spa_spare_activate(newvd);
3950 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
3953 oldvdpath = spa_strdup(oldvd->vdev_path);
3954 newvdpath = spa_strdup(newvd->vdev_path);
3955 newvd_isspare = newvd->vdev_isspare;
3958 * Mark newvd's DTL dirty in this txg.
3960 vdev_dirty(tvd, VDD_DTL, newvd, txg);
3963 * Restart the resilver
3965 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
3970 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
3972 spa_history_log_internal(LOG_POOL_VDEV_ATTACH, spa, NULL,
3973 "%s vdev=%s %s vdev=%s",
3974 replacing && newvd_isspare ? "spare in" :
3975 replacing ? "replace" : "attach", newvdpath,
3976 replacing ? "for" : "to", oldvdpath);
3978 spa_strfree(oldvdpath);
3979 spa_strfree(newvdpath);
3981 if (spa->spa_bootfs)
3982 spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
3988 * Detach a device from a mirror or replacing vdev.
3989 * If 'replace_done' is specified, only detach if the parent
3990 * is a replacing vdev.
3993 spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
3997 vdev_t *rvd = spa->spa_root_vdev;
3998 vdev_t *vd, *pvd, *cvd, *tvd;
3999 boolean_t unspare = B_FALSE;
4000 uint64_t unspare_guid;
4003 ASSERT(spa_writeable(spa));
4005 txg = spa_vdev_enter(spa);
4007 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4010 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4012 if (!vd->vdev_ops->vdev_op_leaf)
4013 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4015 pvd = vd->vdev_parent;
4018 * If the parent/child relationship is not as expected, don't do it.
4019 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4020 * vdev that's replacing B with C. The user's intent in replacing
4021 * is to go from M(A,B) to M(A,C). If the user decides to cancel
4022 * the replace by detaching C, the expected behavior is to end up
4023 * M(A,B). But suppose that right after deciding to detach C,
4024 * the replacement of B completes. We would have M(A,C), and then
4025 * ask to detach C, which would leave us with just A -- not what
4026 * the user wanted. To prevent this, we make sure that the
4027 * parent/child relationship hasn't changed -- in this example,
4028 * that C's parent is still the replacing vdev R.
4030 if (pvd->vdev_guid != pguid && pguid != 0)
4031 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4034 * Only 'replacing' or 'spare' vdevs can be replaced.
4036 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4037 pvd->vdev_ops != &vdev_spare_ops)
4038 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4040 ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4041 spa_version(spa) >= SPA_VERSION_SPARES);
4044 * Only mirror, replacing, and spare vdevs support detach.
4046 if (pvd->vdev_ops != &vdev_replacing_ops &&
4047 pvd->vdev_ops != &vdev_mirror_ops &&
4048 pvd->vdev_ops != &vdev_spare_ops)
4049 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4052 * If this device has the only valid copy of some data,
4053 * we cannot safely detach it.
4055 if (vdev_dtl_required(vd))
4056 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4058 ASSERT(pvd->vdev_children >= 2);
4061 * If we are detaching the second disk from a replacing vdev, then
4062 * check to see if we changed the original vdev's path to have "/old"
4063 * at the end in spa_vdev_attach(). If so, undo that change now.
4065 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4066 vd->vdev_path != NULL) {
4067 size_t len = strlen(vd->vdev_path);
4069 for (int c = 0; c < pvd->vdev_children; c++) {
4070 cvd = pvd->vdev_child[c];
4072 if (cvd == vd || cvd->vdev_path == NULL)
4075 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4076 strcmp(cvd->vdev_path + len, "/old") == 0) {
4077 spa_strfree(cvd->vdev_path);
4078 cvd->vdev_path = spa_strdup(vd->vdev_path);
4085 * If we are detaching the original disk from a spare, then it implies
4086 * that the spare should become a real disk, and be removed from the
4087 * active spare list for the pool.
4089 if (pvd->vdev_ops == &vdev_spare_ops &&
4091 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4095 * Erase the disk labels so the disk can be used for other things.
4096 * This must be done after all other error cases are handled,
4097 * but before we disembowel vd (so we can still do I/O to it).
4098 * But if we can't do it, don't treat the error as fatal --
4099 * it may be that the unwritability of the disk is the reason
4100 * it's being detached!
4102 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4105 * Remove vd from its parent and compact the parent's children.
4107 vdev_remove_child(pvd, vd);
4108 vdev_compact_children(pvd);
4111 * Remember one of the remaining children so we can get tvd below.
4113 cvd = pvd->vdev_child[pvd->vdev_children - 1];
4116 * If we need to remove the remaining child from the list of hot spares,
4117 * do it now, marking the vdev as no longer a spare in the process.
4118 * We must do this before vdev_remove_parent(), because that can
4119 * change the GUID if it creates a new toplevel GUID. For a similar
4120 * reason, we must remove the spare now, in the same txg as the detach;
4121 * otherwise someone could attach a new sibling, change the GUID, and
4122 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4125 ASSERT(cvd->vdev_isspare);
4126 spa_spare_remove(cvd);
4127 unspare_guid = cvd->vdev_guid;
4128 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4129 cvd->vdev_unspare = B_TRUE;
4133 * If the parent mirror/replacing vdev only has one child,
4134 * the parent is no longer needed. Remove it from the tree.
4136 if (pvd->vdev_children == 1) {
4137 if (pvd->vdev_ops == &vdev_spare_ops)
4138 cvd->vdev_unspare = B_FALSE;
4139 vdev_remove_parent(cvd);
4140 cvd->vdev_resilvering = B_FALSE;
4145 * We don't set tvd until now because the parent we just removed
4146 * may have been the previous top-level vdev.
4148 tvd = cvd->vdev_top;
4149 ASSERT(tvd->vdev_parent == rvd);
4152 * Reevaluate the parent vdev state.
4154 vdev_propagate_state(cvd);
4157 * If the 'autoexpand' property is set on the pool then automatically
4158 * try to expand the size of the pool. For example if the device we
4159 * just detached was smaller than the others, it may be possible to
4160 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4161 * first so that we can obtain the updated sizes of the leaf vdevs.
4163 if (spa->spa_autoexpand) {
4165 vdev_expand(tvd, txg);
4168 vdev_config_dirty(tvd);
4171 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
4172 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4173 * But first make sure we're not on any *other* txg's DTL list, to
4174 * prevent vd from being accessed after it's freed.
4176 vdpath = spa_strdup(vd->vdev_path);
4177 for (int t = 0; t < TXG_SIZE; t++)
4178 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4179 vd->vdev_detached = B_TRUE;
4180 vdev_dirty(tvd, VDD_DTL, vd, txg);
4182 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
4184 /* hang on to the spa before we release the lock */
4185 spa_open_ref(spa, FTAG);
4187 error = spa_vdev_exit(spa, vd, txg, 0);
4189 spa_history_log_internal(LOG_POOL_VDEV_DETACH, spa, NULL,
4191 spa_strfree(vdpath);
4194 * If this was the removal of the original device in a hot spare vdev,
4195 * then we want to go through and remove the device from the hot spare
4196 * list of every other pool.
4199 spa_t *altspa = NULL;
4201 mutex_enter(&spa_namespace_lock);
4202 while ((altspa = spa_next(altspa)) != NULL) {
4203 if (altspa->spa_state != POOL_STATE_ACTIVE ||
4207 spa_open_ref(altspa, FTAG);
4208 mutex_exit(&spa_namespace_lock);
4209 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
4210 mutex_enter(&spa_namespace_lock);
4211 spa_close(altspa, FTAG);
4213 mutex_exit(&spa_namespace_lock);
4215 /* search the rest of the vdevs for spares to remove */
4216 spa_vdev_resilver_done(spa);
4219 /* all done with the spa; OK to release */
4220 mutex_enter(&spa_namespace_lock);
4221 spa_close(spa, FTAG);
4222 mutex_exit(&spa_namespace_lock);
4228 * Split a set of devices from their mirrors, and create a new pool from them.
4231 spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
4232 nvlist_t *props, boolean_t exp)
4235 uint64_t txg, *glist;
4237 uint_t c, children, lastlog;
4238 nvlist_t **child, *nvl, *tmp;
4240 char *altroot = NULL;
4241 vdev_t *rvd, **vml = NULL; /* vdev modify list */
4242 boolean_t activate_slog;
4244 ASSERT(spa_writeable(spa));
4246 txg = spa_vdev_enter(spa);
4248 /* clear the log and flush everything up to now */
4249 activate_slog = spa_passivate_log(spa);
4250 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4251 error = spa_offline_log(spa);
4252 txg = spa_vdev_config_enter(spa);
4255 spa_activate_log(spa);
4258 return (spa_vdev_exit(spa, NULL, txg, error));
4260 /* check new spa name before going any further */
4261 if (spa_lookup(newname) != NULL)
4262 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
4265 * scan through all the children to ensure they're all mirrors
4267 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
4268 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
4270 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4272 /* first, check to ensure we've got the right child count */
4273 rvd = spa->spa_root_vdev;
4275 for (c = 0; c < rvd->vdev_children; c++) {
4276 vdev_t *vd = rvd->vdev_child[c];
4278 /* don't count the holes & logs as children */
4279 if (vd->vdev_islog || vd->vdev_ishole) {
4287 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
4288 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4290 /* next, ensure no spare or cache devices are part of the split */
4291 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
4292 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
4293 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4295 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
4296 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
4298 /* then, loop over each vdev and validate it */
4299 for (c = 0; c < children; c++) {
4300 uint64_t is_hole = 0;
4302 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
4306 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
4307 spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
4315 /* which disk is going to be split? */
4316 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
4322 /* look it up in the spa */
4323 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
4324 if (vml[c] == NULL) {
4329 /* make sure there's nothing stopping the split */
4330 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
4331 vml[c]->vdev_islog ||
4332 vml[c]->vdev_ishole ||
4333 vml[c]->vdev_isspare ||
4334 vml[c]->vdev_isl2cache ||
4335 !vdev_writeable(vml[c]) ||
4336 vml[c]->vdev_children != 0 ||
4337 vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
4338 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
4343 if (vdev_dtl_required(vml[c])) {
4348 /* we need certain info from the top level */
4349 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
4350 vml[c]->vdev_top->vdev_ms_array) == 0);
4351 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
4352 vml[c]->vdev_top->vdev_ms_shift) == 0);
4353 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
4354 vml[c]->vdev_top->vdev_asize) == 0);
4355 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
4356 vml[c]->vdev_top->vdev_ashift) == 0);
4360 kmem_free(vml, children * sizeof (vdev_t *));
4361 kmem_free(glist, children * sizeof (uint64_t));
4362 return (spa_vdev_exit(spa, NULL, txg, error));
4365 /* stop writers from using the disks */
4366 for (c = 0; c < children; c++) {
4368 vml[c]->vdev_offline = B_TRUE;
4370 vdev_reopen(spa->spa_root_vdev);
4373 * Temporarily record the splitting vdevs in the spa config. This
4374 * will disappear once the config is regenerated.
4376 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4377 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
4378 glist, children) == 0);
4379 kmem_free(glist, children * sizeof (uint64_t));
4381 mutex_enter(&spa->spa_props_lock);
4382 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
4384 mutex_exit(&spa->spa_props_lock);
4385 spa->spa_config_splitting = nvl;
4386 vdev_config_dirty(spa->spa_root_vdev);
4388 /* configure and create the new pool */
4389 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
4390 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4391 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
4392 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
4393 spa_version(spa)) == 0);
4394 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
4395 spa->spa_config_txg) == 0);
4396 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4397 spa_generate_guid(NULL)) == 0);
4398 (void) nvlist_lookup_string(props,
4399 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4401 /* add the new pool to the namespace */
4402 newspa = spa_add(newname, config, altroot);
4403 newspa->spa_config_txg = spa->spa_config_txg;
4404 spa_set_log_state(newspa, SPA_LOG_CLEAR);
4406 /* release the spa config lock, retaining the namespace lock */
4407 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4409 if (zio_injection_enabled)
4410 zio_handle_panic_injection(spa, FTAG, 1);
4412 spa_activate(newspa, spa_mode_global);
4413 spa_async_suspend(newspa);
4415 /* create the new pool from the disks of the original pool */
4416 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
4420 /* if that worked, generate a real config for the new pool */
4421 if (newspa->spa_root_vdev != NULL) {
4422 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
4423 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4424 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
4425 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
4426 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
4431 if (props != NULL) {
4432 spa_configfile_set(newspa, props, B_FALSE);
4433 error = spa_prop_set(newspa, props);
4438 /* flush everything */
4439 txg = spa_vdev_config_enter(newspa);
4440 vdev_config_dirty(newspa->spa_root_vdev);
4441 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
4443 if (zio_injection_enabled)
4444 zio_handle_panic_injection(spa, FTAG, 2);
4446 spa_async_resume(newspa);
4448 /* finally, update the original pool's config */
4449 txg = spa_vdev_config_enter(spa);
4450 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
4451 error = dmu_tx_assign(tx, TXG_WAIT);
4454 for (c = 0; c < children; c++) {
4455 if (vml[c] != NULL) {
4458 spa_history_log_internal(LOG_POOL_VDEV_DETACH,
4464 vdev_config_dirty(spa->spa_root_vdev);
4465 spa->spa_config_splitting = NULL;
4469 (void) spa_vdev_exit(spa, NULL, txg, 0);
4471 if (zio_injection_enabled)
4472 zio_handle_panic_injection(spa, FTAG, 3);
4474 /* split is complete; log a history record */
4475 spa_history_log_internal(LOG_POOL_SPLIT, newspa, NULL,
4476 "split new pool %s from pool %s", newname, spa_name(spa));
4478 kmem_free(vml, children * sizeof (vdev_t *));
4480 /* if we're not going to mount the filesystems in userland, export */
4482 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
4489 spa_deactivate(newspa);
4492 txg = spa_vdev_config_enter(spa);
4494 /* re-online all offlined disks */
4495 for (c = 0; c < children; c++) {
4497 vml[c]->vdev_offline = B_FALSE;
4499 vdev_reopen(spa->spa_root_vdev);
4501 nvlist_free(spa->spa_config_splitting);
4502 spa->spa_config_splitting = NULL;
4503 (void) spa_vdev_exit(spa, NULL, txg, error);
4505 kmem_free(vml, children * sizeof (vdev_t *));
4510 spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
4512 for (int i = 0; i < count; i++) {
4515 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
4518 if (guid == target_guid)
4526 spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
4527 nvlist_t *dev_to_remove)
4529 nvlist_t **newdev = NULL;
4532 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
4534 for (int i = 0, j = 0; i < count; i++) {
4535 if (dev[i] == dev_to_remove)
4537 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
4540 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
4541 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
4543 for (int i = 0; i < count - 1; i++)
4544 nvlist_free(newdev[i]);
4547 kmem_free(newdev, (count - 1) * sizeof (void *));
4551 * Evacuate the device.
4554 spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
4559 ASSERT(MUTEX_HELD(&spa_namespace_lock));
4560 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
4561 ASSERT(vd == vd->vdev_top);
4564 * Evacuate the device. We don't hold the config lock as writer
4565 * since we need to do I/O but we do keep the
4566 * spa_namespace_lock held. Once this completes the device
4567 * should no longer have any blocks allocated on it.
4569 if (vd->vdev_islog) {
4570 if (vd->vdev_stat.vs_alloc != 0)
4571 error = spa_offline_log(spa);
4580 * The evacuation succeeded. Remove any remaining MOS metadata
4581 * associated with this vdev, and wait for these changes to sync.
4583 ASSERT3U(vd->vdev_stat.vs_alloc, ==, 0);
4584 txg = spa_vdev_config_enter(spa);
4585 vd->vdev_removing = B_TRUE;
4586 vdev_dirty(vd, 0, NULL, txg);
4587 vdev_config_dirty(vd);
4588 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4594 * Complete the removal by cleaning up the namespace.
4597 spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
4599 vdev_t *rvd = spa->spa_root_vdev;
4600 uint64_t id = vd->vdev_id;
4601 boolean_t last_vdev = (id == (rvd->vdev_children - 1));
4603 ASSERT(MUTEX_HELD(&spa_namespace_lock));
4604 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
4605 ASSERT(vd == vd->vdev_top);
4608 * Only remove any devices which are empty.
4610 if (vd->vdev_stat.vs_alloc != 0)
4613 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4615 if (list_link_active(&vd->vdev_state_dirty_node))
4616 vdev_state_clean(vd);
4617 if (list_link_active(&vd->vdev_config_dirty_node))
4618 vdev_config_clean(vd);
4623 vdev_compact_children(rvd);
4625 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
4626 vdev_add_child(rvd, vd);
4628 vdev_config_dirty(rvd);
4631 * Reassess the health of our root vdev.
4637 * Remove a device from the pool -
4639 * Removing a device from the vdev namespace requires several steps
4640 * and can take a significant amount of time. As a result we use
4641 * the spa_vdev_config_[enter/exit] functions which allow us to
4642 * grab and release the spa_config_lock while still holding the namespace
4643 * lock. During each step the configuration is synced out.
4647 * Remove a device from the pool. Currently, this supports removing only hot
4648 * spares, slogs, and level 2 ARC devices.
4651 spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
4654 metaslab_group_t *mg;
4655 nvlist_t **spares, **l2cache, *nv;
4657 uint_t nspares, nl2cache;
4659 boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
4661 ASSERT(spa_writeable(spa));
4664 txg = spa_vdev_enter(spa);
4666 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4668 if (spa->spa_spares.sav_vdevs != NULL &&
4669 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
4670 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
4671 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
4673 * Only remove the hot spare if it's not currently in use
4676 if (vd == NULL || unspare) {
4677 spa_vdev_remove_aux(spa->spa_spares.sav_config,
4678 ZPOOL_CONFIG_SPARES, spares, nspares, nv);
4679 spa_load_spares(spa);
4680 spa->spa_spares.sav_sync = B_TRUE;
4684 } else if (spa->spa_l2cache.sav_vdevs != NULL &&
4685 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
4686 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
4687 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
4689 * Cache devices can always be removed.
4691 spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
4692 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
4693 spa_load_l2cache(spa);
4694 spa->spa_l2cache.sav_sync = B_TRUE;
4695 } else if (vd != NULL && vd->vdev_islog) {
4697 ASSERT(vd == vd->vdev_top);
4700 * XXX - Once we have bp-rewrite this should
4701 * become the common case.
4707 * Stop allocating from this vdev.
4709 metaslab_group_passivate(mg);
4712 * Wait for the youngest allocations and frees to sync,
4713 * and then wait for the deferral of those frees to finish.
4715 spa_vdev_config_exit(spa, NULL,
4716 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
4719 * Attempt to evacuate the vdev.
4721 error = spa_vdev_remove_evacuate(spa, vd);
4723 txg = spa_vdev_config_enter(spa);
4726 * If we couldn't evacuate the vdev, unwind.
4729 metaslab_group_activate(mg);
4730 return (spa_vdev_exit(spa, NULL, txg, error));
4734 * Clean up the vdev namespace.
4736 spa_vdev_remove_from_namespace(spa, vd);
4738 } else if (vd != NULL) {
4740 * Normal vdevs cannot be removed (yet).
4745 * There is no vdev of any kind with the specified guid.
4751 return (spa_vdev_exit(spa, NULL, txg, error));
4757 * Find any device that's done replacing, or a vdev marked 'unspare' that's
4758 * current spared, so we can detach it.
4761 spa_vdev_resilver_done_hunt(vdev_t *vd)
4763 vdev_t *newvd, *oldvd;
4765 for (int c = 0; c < vd->vdev_children; c++) {
4766 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
4772 * Check for a completed replacement. We always consider the first
4773 * vdev in the list to be the oldest vdev, and the last one to be
4774 * the newest (see spa_vdev_attach() for how that works). In
4775 * the case where the newest vdev is faulted, we will not automatically
4776 * remove it after a resilver completes. This is OK as it will require
4777 * user intervention to determine which disk the admin wishes to keep.
4779 if (vd->vdev_ops == &vdev_replacing_ops) {
4780 ASSERT(vd->vdev_children > 1);
4782 newvd = vd->vdev_child[vd->vdev_children - 1];
4783 oldvd = vd->vdev_child[0];
4785 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
4786 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
4787 !vdev_dtl_required(oldvd))
4792 * Check for a completed resilver with the 'unspare' flag set.
4794 if (vd->vdev_ops == &vdev_spare_ops) {
4795 vdev_t *first = vd->vdev_child[0];
4796 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
4798 if (last->vdev_unspare) {
4801 } else if (first->vdev_unspare) {
4808 if (oldvd != NULL &&
4809 vdev_dtl_empty(newvd, DTL_MISSING) &&
4810 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
4811 !vdev_dtl_required(oldvd))
4815 * If there are more than two spares attached to a disk,
4816 * and those spares are not required, then we want to
4817 * attempt to free them up now so that they can be used
4818 * by other pools. Once we're back down to a single
4819 * disk+spare, we stop removing them.
4821 if (vd->vdev_children > 2) {
4822 newvd = vd->vdev_child[1];
4824 if (newvd->vdev_isspare && last->vdev_isspare &&
4825 vdev_dtl_empty(last, DTL_MISSING) &&
4826 vdev_dtl_empty(last, DTL_OUTAGE) &&
4827 !vdev_dtl_required(newvd))
4836 spa_vdev_resilver_done(spa_t *spa)
4838 vdev_t *vd, *pvd, *ppvd;
4839 uint64_t guid, sguid, pguid, ppguid;
4841 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4843 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
4844 pvd = vd->vdev_parent;
4845 ppvd = pvd->vdev_parent;
4846 guid = vd->vdev_guid;
4847 pguid = pvd->vdev_guid;
4848 ppguid = ppvd->vdev_guid;
4851 * If we have just finished replacing a hot spared device, then
4852 * we need to detach the parent's first child (the original hot
4855 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
4856 ppvd->vdev_children == 2) {
4857 ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
4858 sguid = ppvd->vdev_child[1]->vdev_guid;
4860 spa_config_exit(spa, SCL_ALL, FTAG);
4861 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
4863 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
4865 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4868 spa_config_exit(spa, SCL_ALL, FTAG);
4872 * Update the stored path or FRU for this vdev.
4875 spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
4879 boolean_t sync = B_FALSE;
4881 ASSERT(spa_writeable(spa));
4883 spa_vdev_state_enter(spa, SCL_ALL);
4885 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
4886 return (spa_vdev_state_exit(spa, NULL, ENOENT));
4888 if (!vd->vdev_ops->vdev_op_leaf)
4889 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
4892 if (strcmp(value, vd->vdev_path) != 0) {
4893 spa_strfree(vd->vdev_path);
4894 vd->vdev_path = spa_strdup(value);
4898 if (vd->vdev_fru == NULL) {
4899 vd->vdev_fru = spa_strdup(value);
4901 } else if (strcmp(value, vd->vdev_fru) != 0) {
4902 spa_strfree(vd->vdev_fru);
4903 vd->vdev_fru = spa_strdup(value);
4908 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
4912 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
4914 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
4918 spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
4920 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
4924 * ==========================================================================
4926 * ==========================================================================
4930 spa_scan_stop(spa_t *spa)
4932 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
4933 if (dsl_scan_resilvering(spa->spa_dsl_pool))
4935 return (dsl_scan_cancel(spa->spa_dsl_pool));
4939 spa_scan(spa_t *spa, pool_scan_func_t func)
4941 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
4943 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
4947 * If a resilver was requested, but there is no DTL on a
4948 * writeable leaf device, we have nothing to do.
4950 if (func == POOL_SCAN_RESILVER &&
4951 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
4952 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
4956 return (dsl_scan(spa->spa_dsl_pool, func));
4960 * ==========================================================================
4961 * SPA async task processing
4962 * ==========================================================================
4966 spa_async_remove(spa_t *spa, vdev_t *vd)
4968 if (vd->vdev_remove_wanted) {
4969 vd->vdev_remove_wanted = B_FALSE;
4970 vd->vdev_delayed_close = B_FALSE;
4971 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
4974 * We want to clear the stats, but we don't want to do a full
4975 * vdev_clear() as that will cause us to throw away
4976 * degraded/faulted state as well as attempt to reopen the
4977 * device, all of which is a waste.
4979 vd->vdev_stat.vs_read_errors = 0;
4980 vd->vdev_stat.vs_write_errors = 0;
4981 vd->vdev_stat.vs_checksum_errors = 0;
4983 vdev_state_dirty(vd->vdev_top);
4986 for (int c = 0; c < vd->vdev_children; c++)
4987 spa_async_remove(spa, vd->vdev_child[c]);
4991 spa_async_probe(spa_t *spa, vdev_t *vd)
4993 if (vd->vdev_probe_wanted) {
4994 vd->vdev_probe_wanted = B_FALSE;
4995 vdev_reopen(vd); /* vdev_open() does the actual probe */
4998 for (int c = 0; c < vd->vdev_children; c++)
4999 spa_async_probe(spa, vd->vdev_child[c]);
5003 spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5009 if (!spa->spa_autoexpand)
5012 for (int c = 0; c < vd->vdev_children; c++) {
5013 vdev_t *cvd = vd->vdev_child[c];
5014 spa_async_autoexpand(spa, cvd);
5017 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5020 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
5021 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
5023 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5024 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
5026 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
5027 ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
5030 kmem_free(physpath, MAXPATHLEN);
5034 spa_async_thread(spa_t *spa)
5038 ASSERT(spa->spa_sync_on);
5040 mutex_enter(&spa->spa_async_lock);
5041 tasks = spa->spa_async_tasks;
5042 spa->spa_async_tasks = 0;
5043 mutex_exit(&spa->spa_async_lock);
5046 * See if the config needs to be updated.
5048 if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5049 uint64_t old_space, new_space;
5051 mutex_enter(&spa_namespace_lock);
5052 old_space = metaslab_class_get_space(spa_normal_class(spa));
5053 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5054 new_space = metaslab_class_get_space(spa_normal_class(spa));
5055 mutex_exit(&spa_namespace_lock);
5058 * If the pool grew as a result of the config update,
5059 * then log an internal history event.
5061 if (new_space != old_space) {
5062 spa_history_log_internal(LOG_POOL_VDEV_ONLINE,
5064 "pool '%s' size: %llu(+%llu)",
5065 spa_name(spa), new_space, new_space - old_space);
5070 * See if any devices need to be marked REMOVED.
5072 if (tasks & SPA_ASYNC_REMOVE) {
5073 spa_vdev_state_enter(spa, SCL_NONE);
5074 spa_async_remove(spa, spa->spa_root_vdev);
5075 for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
5076 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5077 for (int i = 0; i < spa->spa_spares.sav_count; i++)
5078 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5079 (void) spa_vdev_state_exit(spa, NULL, 0);
5082 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5083 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5084 spa_async_autoexpand(spa, spa->spa_root_vdev);
5085 spa_config_exit(spa, SCL_CONFIG, FTAG);
5089 * See if any devices need to be probed.
5091 if (tasks & SPA_ASYNC_PROBE) {
5092 spa_vdev_state_enter(spa, SCL_NONE);
5093 spa_async_probe(spa, spa->spa_root_vdev);
5094 (void) spa_vdev_state_exit(spa, NULL, 0);
5098 * If any devices are done replacing, detach them.
5100 if (tasks & SPA_ASYNC_RESILVER_DONE)
5101 spa_vdev_resilver_done(spa);
5104 * Kick off a resilver.
5106 if (tasks & SPA_ASYNC_RESILVER)
5107 dsl_resilver_restart(spa->spa_dsl_pool, 0);
5110 * Let the world know that we're done.
5112 mutex_enter(&spa->spa_async_lock);
5113 spa->spa_async_thread = NULL;
5114 cv_broadcast(&spa->spa_async_cv);
5115 mutex_exit(&spa->spa_async_lock);
5120 spa_async_suspend(spa_t *spa)
5122 mutex_enter(&spa->spa_async_lock);
5123 spa->spa_async_suspended++;
5124 while (spa->spa_async_thread != NULL)
5125 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5126 mutex_exit(&spa->spa_async_lock);
5130 spa_async_resume(spa_t *spa)
5132 mutex_enter(&spa->spa_async_lock);
5133 ASSERT(spa->spa_async_suspended != 0);
5134 spa->spa_async_suspended--;
5135 mutex_exit(&spa->spa_async_lock);
5139 spa_async_dispatch(spa_t *spa)
5141 mutex_enter(&spa->spa_async_lock);
5142 if (spa->spa_async_tasks && !spa->spa_async_suspended &&
5143 spa->spa_async_thread == NULL &&
5144 rootdir != NULL && !vn_is_readonly(rootdir))
5145 spa->spa_async_thread = thread_create(NULL, 0,
5146 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
5147 mutex_exit(&spa->spa_async_lock);
5151 spa_async_request(spa_t *spa, int task)
5153 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
5154 mutex_enter(&spa->spa_async_lock);
5155 spa->spa_async_tasks |= task;
5156 mutex_exit(&spa->spa_async_lock);
5160 * ==========================================================================
5161 * SPA syncing routines
5162 * ==========================================================================
5166 bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5169 bpobj_enqueue(bpo, bp, tx);
5174 spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5178 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
5184 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
5186 char *packed = NULL;
5191 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
5194 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5195 * information. This avoids the dbuf_will_dirty() path and
5196 * saves us a pre-read to get data we don't actually care about.
5198 bufsize = P2ROUNDUP(nvsize, SPA_CONFIG_BLOCKSIZE);
5199 packed = kmem_alloc(bufsize, KM_SLEEP);
5201 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
5203 bzero(packed + nvsize, bufsize - nvsize);
5205 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
5207 kmem_free(packed, bufsize);
5209 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
5210 dmu_buf_will_dirty(db, tx);
5211 *(uint64_t *)db->db_data = nvsize;
5212 dmu_buf_rele(db, FTAG);
5216 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
5217 const char *config, const char *entry)
5227 * Update the MOS nvlist describing the list of available devices.
5228 * spa_validate_aux() will have already made sure this nvlist is
5229 * valid and the vdevs are labeled appropriately.
5231 if (sav->sav_object == 0) {
5232 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
5233 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
5234 sizeof (uint64_t), tx);
5235 VERIFY(zap_update(spa->spa_meta_objset,
5236 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
5237 &sav->sav_object, tx) == 0);
5240 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5241 if (sav->sav_count == 0) {
5242 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
5244 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
5245 for (i = 0; i < sav->sav_count; i++)
5246 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
5247 B_FALSE, VDEV_CONFIG_L2CACHE);
5248 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
5249 sav->sav_count) == 0);
5250 for (i = 0; i < sav->sav_count; i++)
5251 nvlist_free(list[i]);
5252 kmem_free(list, sav->sav_count * sizeof (void *));
5255 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
5256 nvlist_free(nvroot);
5258 sav->sav_sync = B_FALSE;
5262 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
5266 if (list_is_empty(&spa->spa_config_dirty_list))
5269 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5271 config = spa_config_generate(spa, spa->spa_root_vdev,
5272 dmu_tx_get_txg(tx), B_FALSE);
5274 spa_config_exit(spa, SCL_STATE, FTAG);
5276 if (spa->spa_config_syncing)
5277 nvlist_free(spa->spa_config_syncing);
5278 spa->spa_config_syncing = config;
5280 spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
5284 * Set zpool properties.
5287 spa_sync_props(void *arg1, void *arg2, dmu_tx_t *tx)
5290 objset_t *mos = spa->spa_meta_objset;
5291 nvlist_t *nvp = arg2;
5296 const char *propname;
5297 zprop_type_t proptype;
5299 mutex_enter(&spa->spa_props_lock);
5302 while ((elem = nvlist_next_nvpair(nvp, elem))) {
5303 switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
5304 case ZPOOL_PROP_VERSION:
5306 * Only set version for non-zpool-creation cases
5307 * (set/import). spa_create() needs special care
5308 * for version setting.
5310 if (tx->tx_txg != TXG_INITIAL) {
5311 VERIFY(nvpair_value_uint64(elem,
5313 ASSERT(intval <= SPA_VERSION);
5314 ASSERT(intval >= spa_version(spa));
5315 spa->spa_uberblock.ub_version = intval;
5316 vdev_config_dirty(spa->spa_root_vdev);
5320 case ZPOOL_PROP_ALTROOT:
5322 * 'altroot' is a non-persistent property. It should
5323 * have been set temporarily at creation or import time.
5325 ASSERT(spa->spa_root != NULL);
5328 case ZPOOL_PROP_READONLY:
5329 case ZPOOL_PROP_CACHEFILE:
5331 * 'readonly' and 'cachefile' are also non-persisitent
5337 * Set pool property values in the poolprops mos object.
5339 if (spa->spa_pool_props_object == 0) {
5340 VERIFY((spa->spa_pool_props_object =
5341 zap_create(mos, DMU_OT_POOL_PROPS,
5342 DMU_OT_NONE, 0, tx)) > 0);
5344 VERIFY(zap_update(mos,
5345 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
5346 8, 1, &spa->spa_pool_props_object, tx)
5350 /* normalize the property name */
5351 propname = zpool_prop_to_name(prop);
5352 proptype = zpool_prop_get_type(prop);
5354 if (nvpair_type(elem) == DATA_TYPE_STRING) {
5355 ASSERT(proptype == PROP_TYPE_STRING);
5356 VERIFY(nvpair_value_string(elem, &strval) == 0);
5357 VERIFY(zap_update(mos,
5358 spa->spa_pool_props_object, propname,
5359 1, strlen(strval) + 1, strval, tx) == 0);
5361 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
5362 VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5364 if (proptype == PROP_TYPE_INDEX) {
5366 VERIFY(zpool_prop_index_to_string(
5367 prop, intval, &unused) == 0);
5369 VERIFY(zap_update(mos,
5370 spa->spa_pool_props_object, propname,
5371 8, 1, &intval, tx) == 0);
5373 ASSERT(0); /* not allowed */
5377 case ZPOOL_PROP_DELEGATION:
5378 spa->spa_delegation = intval;
5380 case ZPOOL_PROP_BOOTFS:
5381 spa->spa_bootfs = intval;
5383 case ZPOOL_PROP_FAILUREMODE:
5384 spa->spa_failmode = intval;
5386 case ZPOOL_PROP_AUTOEXPAND:
5387 spa->spa_autoexpand = intval;
5388 if (tx->tx_txg != TXG_INITIAL)
5389 spa_async_request(spa,
5390 SPA_ASYNC_AUTOEXPAND);
5392 case ZPOOL_PROP_DEDUPDITTO:
5393 spa->spa_dedup_ditto = intval;
5400 /* log internal history if this is not a zpool create */
5401 if (spa_version(spa) >= SPA_VERSION_ZPOOL_HISTORY &&
5402 tx->tx_txg != TXG_INITIAL) {
5403 spa_history_log_internal(LOG_POOL_PROPSET,
5404 spa, tx, "%s %lld %s",
5405 nvpair_name(elem), intval, spa_name(spa));
5409 mutex_exit(&spa->spa_props_lock);
5413 * Perform one-time upgrade on-disk changes. spa_version() does not
5414 * reflect the new version this txg, so there must be no changes this
5415 * txg to anything that the upgrade code depends on after it executes.
5416 * Therefore this must be called after dsl_pool_sync() does the sync
5420 spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
5422 dsl_pool_t *dp = spa->spa_dsl_pool;
5424 ASSERT(spa->spa_sync_pass == 1);
5426 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
5427 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
5428 dsl_pool_create_origin(dp, tx);
5430 /* Keeping the origin open increases spa_minref */
5431 spa->spa_minref += 3;
5434 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
5435 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
5436 dsl_pool_upgrade_clones(dp, tx);
5439 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
5440 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
5441 dsl_pool_upgrade_dir_clones(dp, tx);
5443 /* Keeping the freedir open increases spa_minref */
5444 spa->spa_minref += 3;
5449 * Sync the specified transaction group. New blocks may be dirtied as
5450 * part of the process, so we iterate until it converges.
5453 spa_sync(spa_t *spa, uint64_t txg)
5455 dsl_pool_t *dp = spa->spa_dsl_pool;
5456 objset_t *mos = spa->spa_meta_objset;
5457 bpobj_t *defer_bpo = &spa->spa_deferred_bpobj;
5458 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
5459 vdev_t *rvd = spa->spa_root_vdev;
5464 VERIFY(spa_writeable(spa));
5467 * Lock out configuration changes.
5469 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5471 spa->spa_syncing_txg = txg;
5472 spa->spa_sync_pass = 0;
5475 * If there are any pending vdev state changes, convert them
5476 * into config changes that go out with this transaction group.
5478 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5479 while (list_head(&spa->spa_state_dirty_list) != NULL) {
5481 * We need the write lock here because, for aux vdevs,
5482 * calling vdev_config_dirty() modifies sav_config.
5483 * This is ugly and will become unnecessary when we
5484 * eliminate the aux vdev wart by integrating all vdevs
5485 * into the root vdev tree.
5487 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
5488 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
5489 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
5490 vdev_state_clean(vd);
5491 vdev_config_dirty(vd);
5493 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
5494 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
5496 spa_config_exit(spa, SCL_STATE, FTAG);
5498 tx = dmu_tx_create_assigned(dp, txg);
5501 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
5502 * set spa_deflate if we have no raid-z vdevs.
5504 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
5505 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
5508 for (i = 0; i < rvd->vdev_children; i++) {
5509 vd = rvd->vdev_child[i];
5510 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
5513 if (i == rvd->vdev_children) {
5514 spa->spa_deflate = TRUE;
5515 VERIFY(0 == zap_add(spa->spa_meta_objset,
5516 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
5517 sizeof (uint64_t), 1, &spa->spa_deflate, tx));
5522 * If anything has changed in this txg, or if someone is waiting
5523 * for this txg to sync (eg, spa_vdev_remove()), push the
5524 * deferred frees from the previous txg. If not, leave them
5525 * alone so that we don't generate work on an otherwise idle
5528 if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
5529 !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
5530 !txg_list_empty(&dp->dp_sync_tasks, txg) ||
5531 ((dsl_scan_active(dp->dp_scan) ||
5532 txg_sync_waiting(dp)) && !spa_shutting_down(spa))) {
5533 zio_t *zio = zio_root(spa, NULL, NULL, 0);
5534 VERIFY3U(bpobj_iterate(defer_bpo,
5535 spa_free_sync_cb, zio, tx), ==, 0);
5536 VERIFY3U(zio_wait(zio), ==, 0);
5540 * Iterate to convergence.
5543 int pass = ++spa->spa_sync_pass;
5545 spa_sync_config_object(spa, tx);
5546 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
5547 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
5548 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
5549 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
5550 spa_errlog_sync(spa, txg);
5551 dsl_pool_sync(dp, txg);
5553 if (pass <= SYNC_PASS_DEFERRED_FREE) {
5554 zio_t *zio = zio_root(spa, NULL, NULL, 0);
5555 bplist_iterate(free_bpl, spa_free_sync_cb,
5557 VERIFY(zio_wait(zio) == 0);
5559 bplist_iterate(free_bpl, bpobj_enqueue_cb,
5564 dsl_scan_sync(dp, tx);
5566 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
5570 spa_sync_upgrades(spa, tx);
5572 } while (dmu_objset_is_dirty(mos, txg));
5575 * Rewrite the vdev configuration (which includes the uberblock)
5576 * to commit the transaction group.
5578 * If there are no dirty vdevs, we sync the uberblock to a few
5579 * random top-level vdevs that are known to be visible in the
5580 * config cache (see spa_vdev_add() for a complete description).
5581 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
5585 * We hold SCL_STATE to prevent vdev open/close/etc.
5586 * while we're attempting to write the vdev labels.
5588 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5590 if (list_is_empty(&spa->spa_config_dirty_list)) {
5591 vdev_t *svd[SPA_DVAS_PER_BP];
5593 int children = rvd->vdev_children;
5594 int c0 = spa_get_random(children);
5596 for (int c = 0; c < children; c++) {
5597 vd = rvd->vdev_child[(c0 + c) % children];
5598 if (vd->vdev_ms_array == 0 || vd->vdev_islog)
5600 svd[svdcount++] = vd;
5601 if (svdcount == SPA_DVAS_PER_BP)
5604 error = vdev_config_sync(svd, svdcount, txg, B_FALSE);
5606 error = vdev_config_sync(svd, svdcount, txg,
5609 error = vdev_config_sync(rvd->vdev_child,
5610 rvd->vdev_children, txg, B_FALSE);
5612 error = vdev_config_sync(rvd->vdev_child,
5613 rvd->vdev_children, txg, B_TRUE);
5616 spa_config_exit(spa, SCL_STATE, FTAG);
5620 zio_suspend(spa, NULL);
5621 zio_resume_wait(spa);
5626 * Clear the dirty config list.
5628 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
5629 vdev_config_clean(vd);
5632 * Now that the new config has synced transactionally,
5633 * let it become visible to the config cache.
5635 if (spa->spa_config_syncing != NULL) {
5636 spa_config_set(spa, spa->spa_config_syncing);
5637 spa->spa_config_txg = txg;
5638 spa->spa_config_syncing = NULL;
5641 spa->spa_ubsync = spa->spa_uberblock;
5643 dsl_pool_sync_done(dp, txg);
5646 * Update usable space statistics.
5648 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
5649 vdev_sync_done(vd, txg);
5651 spa_update_dspace(spa);
5654 * It had better be the case that we didn't dirty anything
5655 * since vdev_config_sync().
5657 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
5658 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
5659 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
5661 spa->spa_sync_pass = 0;
5663 spa_config_exit(spa, SCL_CONFIG, FTAG);
5665 spa_handle_ignored_writes(spa);
5668 * If any async tasks have been requested, kick them off.
5670 spa_async_dispatch(spa);
5674 * Sync all pools. We don't want to hold the namespace lock across these
5675 * operations, so we take a reference on the spa_t and drop the lock during the
5679 spa_sync_allpools(void)
5682 mutex_enter(&spa_namespace_lock);
5683 while ((spa = spa_next(spa)) != NULL) {
5684 if (spa_state(spa) != POOL_STATE_ACTIVE ||
5685 !spa_writeable(spa) || spa_suspended(spa))
5687 spa_open_ref(spa, FTAG);
5688 mutex_exit(&spa_namespace_lock);
5689 txg_wait_synced(spa_get_dsl(spa), 0);
5690 mutex_enter(&spa_namespace_lock);
5691 spa_close(spa, FTAG);
5693 mutex_exit(&spa_namespace_lock);
5697 * ==========================================================================
5698 * Miscellaneous routines
5699 * ==========================================================================
5703 * Remove all pools in the system.
5711 * Remove all cached state. All pools should be closed now,
5712 * so every spa in the AVL tree should be unreferenced.
5714 mutex_enter(&spa_namespace_lock);
5715 while ((spa = spa_next(NULL)) != NULL) {
5717 * Stop async tasks. The async thread may need to detach
5718 * a device that's been replaced, which requires grabbing
5719 * spa_namespace_lock, so we must drop it here.
5721 spa_open_ref(spa, FTAG);
5722 mutex_exit(&spa_namespace_lock);
5723 spa_async_suspend(spa);
5724 mutex_enter(&spa_namespace_lock);
5725 spa_close(spa, FTAG);
5727 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
5729 spa_deactivate(spa);
5733 mutex_exit(&spa_namespace_lock);
5737 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
5742 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
5746 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
5747 vd = spa->spa_l2cache.sav_vdevs[i];
5748 if (vd->vdev_guid == guid)
5752 for (i = 0; i < spa->spa_spares.sav_count; i++) {
5753 vd = spa->spa_spares.sav_vdevs[i];
5754 if (vd->vdev_guid == guid)
5763 spa_upgrade(spa_t *spa, uint64_t version)
5765 ASSERT(spa_writeable(spa));
5767 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5770 * This should only be called for a non-faulted pool, and since a
5771 * future version would result in an unopenable pool, this shouldn't be
5774 ASSERT(spa->spa_uberblock.ub_version <= SPA_VERSION);
5775 ASSERT(version >= spa->spa_uberblock.ub_version);
5777 spa->spa_uberblock.ub_version = version;
5778 vdev_config_dirty(spa->spa_root_vdev);
5780 spa_config_exit(spa, SCL_ALL, FTAG);
5782 txg_wait_synced(spa_get_dsl(spa), 0);
5786 spa_has_spare(spa_t *spa, uint64_t guid)
5790 spa_aux_vdev_t *sav = &spa->spa_spares;
5792 for (i = 0; i < sav->sav_count; i++)
5793 if (sav->sav_vdevs[i]->vdev_guid == guid)
5796 for (i = 0; i < sav->sav_npending; i++) {
5797 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
5798 &spareguid) == 0 && spareguid == guid)
5806 * Check if a pool has an active shared spare device.
5807 * Note: reference count of an active spare is 2, as a spare and as a replace
5810 spa_has_active_shared_spare(spa_t *spa)
5814 spa_aux_vdev_t *sav = &spa->spa_spares;
5816 for (i = 0; i < sav->sav_count; i++) {
5817 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
5818 &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
5827 * Post a sysevent corresponding to the given event. The 'name' must be one of
5828 * the event definitions in sys/sysevent/eventdefs.h. The payload will be
5829 * filled in from the spa and (optionally) the vdev. This doesn't do anything
5830 * in the userland libzpool, as we don't want consumers to misinterpret ztest
5831 * or zdb as real changes.
5834 spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
5838 sysevent_attr_list_t *attr = NULL;
5839 sysevent_value_t value;
5842 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
5845 value.value_type = SE_DATA_TYPE_STRING;
5846 value.value.sv_string = spa_name(spa);
5847 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
5850 value.value_type = SE_DATA_TYPE_UINT64;
5851 value.value.sv_uint64 = spa_guid(spa);
5852 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
5856 value.value_type = SE_DATA_TYPE_UINT64;
5857 value.value.sv_uint64 = vd->vdev_guid;
5858 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
5862 if (vd->vdev_path) {
5863 value.value_type = SE_DATA_TYPE_STRING;
5864 value.value.sv_string = vd->vdev_path;
5865 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
5866 &value, SE_SLEEP) != 0)
5871 if (sysevent_attach_attributes(ev, attr) != 0)
5875 (void) log_sysevent(ev, SE_SLEEP, &eid);
5879 sysevent_free_attr(attr);