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,
120 uint_t zio_taskq_batch_pct = 100; /* 1 thread per cpu in pset */
121 id_t zio_taskq_psrset_bind = PS_NONE;
122 boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */
123 uint_t zio_taskq_basedc = 80; /* base duty cycle */
125 boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */
128 * This (illegal) pool name is used when temporarily importing a spa_t in order
129 * to get the vdev stats associated with the imported devices.
131 #define TRYIMPORT_NAME "$import"
134 * ==========================================================================
135 * SPA properties routines
136 * ==========================================================================
140 * Add a (source=src, propname=propval) list to an nvlist.
143 spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
144 uint64_t intval, zprop_source_t src)
146 const char *propname = zpool_prop_to_name(prop);
149 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
150 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
153 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
155 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
157 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
158 nvlist_free(propval);
162 * Get property values from the spa configuration.
165 spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
169 uint64_t cap, version;
170 zprop_source_t src = ZPROP_SRC_NONE;
171 spa_config_dirent_t *dp;
173 ASSERT(MUTEX_HELD(&spa->spa_props_lock));
175 if (spa->spa_root_vdev != NULL) {
176 alloc = metaslab_class_get_alloc(spa_normal_class(spa));
177 size = metaslab_class_get_space(spa_normal_class(spa));
178 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
179 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
180 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
181 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
184 cap = (size == 0) ? 0 : (alloc * 100 / size);
185 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
187 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
188 ddt_get_pool_dedup_ratio(spa), src);
190 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
191 spa->spa_root_vdev->vdev_state, src);
193 version = spa_version(spa);
194 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
195 src = ZPROP_SRC_DEFAULT;
197 src = ZPROP_SRC_LOCAL;
198 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
201 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
203 if (spa->spa_root != NULL)
204 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
207 if ((dp = list_head(&spa->spa_config_list)) != NULL) {
208 if (dp->scd_path == NULL) {
209 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
210 "none", 0, ZPROP_SRC_LOCAL);
211 } else if (strcmp(dp->scd_path, spa_config_path) != 0) {
212 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
213 dp->scd_path, 0, ZPROP_SRC_LOCAL);
219 * Get zpool property values.
222 spa_prop_get(spa_t *spa, nvlist_t **nvp)
224 objset_t *mos = spa->spa_meta_objset;
229 VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
231 mutex_enter(&spa->spa_props_lock);
234 * Get properties from the spa config.
236 spa_prop_get_config(spa, nvp);
238 /* If no pool property object, no more prop to get. */
239 if (mos == NULL || spa->spa_pool_props_object == 0) {
240 mutex_exit(&spa->spa_props_lock);
245 * Get properties from the MOS pool property object.
247 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
248 (err = zap_cursor_retrieve(&zc, &za)) == 0;
249 zap_cursor_advance(&zc)) {
252 zprop_source_t src = ZPROP_SRC_DEFAULT;
255 if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
258 switch (za.za_integer_length) {
260 /* integer property */
261 if (za.za_first_integer !=
262 zpool_prop_default_numeric(prop))
263 src = ZPROP_SRC_LOCAL;
265 if (prop == ZPOOL_PROP_BOOTFS) {
267 dsl_dataset_t *ds = NULL;
269 dp = spa_get_dsl(spa);
270 rw_enter(&dp->dp_config_rwlock, RW_READER);
271 if (err = dsl_dataset_hold_obj(dp,
272 za.za_first_integer, FTAG, &ds)) {
273 rw_exit(&dp->dp_config_rwlock);
278 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
280 dsl_dataset_name(ds, strval);
281 dsl_dataset_rele(ds, FTAG);
282 rw_exit(&dp->dp_config_rwlock);
285 intval = za.za_first_integer;
288 spa_prop_add_list(*nvp, prop, strval, intval, src);
292 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
297 /* string property */
298 strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
299 err = zap_lookup(mos, spa->spa_pool_props_object,
300 za.za_name, 1, za.za_num_integers, strval);
302 kmem_free(strval, za.za_num_integers);
305 spa_prop_add_list(*nvp, prop, strval, 0, src);
306 kmem_free(strval, za.za_num_integers);
313 zap_cursor_fini(&zc);
314 mutex_exit(&spa->spa_props_lock);
316 if (err && err != ENOENT) {
326 * Validate the given pool properties nvlist and modify the list
327 * for the property values to be set.
330 spa_prop_validate(spa_t *spa, nvlist_t *props)
333 int error = 0, reset_bootfs = 0;
337 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
339 char *propname, *strval;
344 propname = nvpair_name(elem);
346 if ((prop = zpool_name_to_prop(propname)) == ZPROP_INVAL)
350 case ZPOOL_PROP_VERSION:
351 error = nvpair_value_uint64(elem, &intval);
353 (intval < spa_version(spa) || intval > SPA_VERSION))
357 case ZPOOL_PROP_DELEGATION:
358 case ZPOOL_PROP_AUTOREPLACE:
359 case ZPOOL_PROP_LISTSNAPS:
360 case ZPOOL_PROP_AUTOEXPAND:
361 error = nvpair_value_uint64(elem, &intval);
362 if (!error && intval > 1)
366 case ZPOOL_PROP_BOOTFS:
368 * If the pool version is less than SPA_VERSION_BOOTFS,
369 * or the pool is still being created (version == 0),
370 * the bootfs property cannot be set.
372 if (spa_version(spa) < SPA_VERSION_BOOTFS) {
378 * Make sure the vdev config is bootable
380 if (!vdev_is_bootable(spa->spa_root_vdev)) {
387 error = nvpair_value_string(elem, &strval);
392 if (strval == NULL || strval[0] == '\0') {
393 objnum = zpool_prop_default_numeric(
398 if (error = dmu_objset_hold(strval, FTAG, &os))
401 /* Must be ZPL and not gzip compressed. */
403 if (dmu_objset_type(os) != DMU_OST_ZFS) {
405 } else if ((error = dsl_prop_get_integer(strval,
406 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
407 &compress, NULL)) == 0 &&
408 !BOOTFS_COMPRESS_VALID(compress)) {
411 objnum = dmu_objset_id(os);
413 dmu_objset_rele(os, FTAG);
417 case ZPOOL_PROP_FAILUREMODE:
418 error = nvpair_value_uint64(elem, &intval);
419 if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
420 intval > ZIO_FAILURE_MODE_PANIC))
424 * This is a special case which only occurs when
425 * the pool has completely failed. This allows
426 * the user to change the in-core failmode property
427 * without syncing it out to disk (I/Os might
428 * currently be blocked). We do this by returning
429 * EIO to the caller (spa_prop_set) to trick it
430 * into thinking we encountered a property validation
433 if (!error && spa_suspended(spa)) {
434 spa->spa_failmode = intval;
439 case ZPOOL_PROP_CACHEFILE:
440 if ((error = nvpair_value_string(elem, &strval)) != 0)
443 if (strval[0] == '\0')
446 if (strcmp(strval, "none") == 0)
449 if (strval[0] != '/') {
454 slash = strrchr(strval, '/');
455 ASSERT(slash != NULL);
457 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
458 strcmp(slash, "/..") == 0)
462 case ZPOOL_PROP_DEDUPDITTO:
463 if (spa_version(spa) < SPA_VERSION_DEDUP)
466 error = nvpair_value_uint64(elem, &intval);
468 intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
477 if (!error && reset_bootfs) {
478 error = nvlist_remove(props,
479 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
482 error = nvlist_add_uint64(props,
483 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
491 spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
494 spa_config_dirent_t *dp;
496 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
500 dp = kmem_alloc(sizeof (spa_config_dirent_t),
503 if (cachefile[0] == '\0')
504 dp->scd_path = spa_strdup(spa_config_path);
505 else if (strcmp(cachefile, "none") == 0)
508 dp->scd_path = spa_strdup(cachefile);
510 list_insert_head(&spa->spa_config_list, dp);
512 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
516 spa_prop_set(spa_t *spa, nvlist_t *nvp)
520 boolean_t need_sync = B_FALSE;
523 if ((error = spa_prop_validate(spa, nvp)) != 0)
527 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
528 if ((prop = zpool_name_to_prop(
529 nvpair_name(elem))) == ZPROP_INVAL)
532 if (prop == ZPOOL_PROP_CACHEFILE || prop == ZPOOL_PROP_ALTROOT)
540 return (dsl_sync_task_do(spa_get_dsl(spa), NULL, spa_sync_props,
547 * If the bootfs property value is dsobj, clear it.
550 spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
552 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
553 VERIFY(zap_remove(spa->spa_meta_objset,
554 spa->spa_pool_props_object,
555 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
561 * ==========================================================================
562 * SPA state manipulation (open/create/destroy/import/export)
563 * ==========================================================================
567 spa_error_entry_compare(const void *a, const void *b)
569 spa_error_entry_t *sa = (spa_error_entry_t *)a;
570 spa_error_entry_t *sb = (spa_error_entry_t *)b;
573 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
574 sizeof (zbookmark_t));
585 * Utility function which retrieves copies of the current logs and
586 * re-initializes them in the process.
589 spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
591 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
593 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
594 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
596 avl_create(&spa->spa_errlist_scrub,
597 spa_error_entry_compare, sizeof (spa_error_entry_t),
598 offsetof(spa_error_entry_t, se_avl));
599 avl_create(&spa->spa_errlist_last,
600 spa_error_entry_compare, sizeof (spa_error_entry_t),
601 offsetof(spa_error_entry_t, se_avl));
605 spa_taskq_create(spa_t *spa, const char *name, enum zti_modes mode,
608 uint_t flags = TASKQ_PREPOPULATE;
609 boolean_t batch = B_FALSE;
613 return (NULL); /* no taskq needed */
616 ASSERT3U(value, >=, 1);
617 value = MAX(value, 1);
622 flags |= TASKQ_THREADS_CPU_PCT;
623 value = zio_taskq_batch_pct;
626 case zti_mode_online_percent:
627 flags |= TASKQ_THREADS_CPU_PCT;
631 panic("unrecognized mode for %s taskq (%u:%u) in "
637 if (zio_taskq_sysdc && spa->spa_proc != &p0) {
639 flags |= TASKQ_DC_BATCH;
641 return (taskq_create_sysdc(name, value, 50, INT_MAX,
642 spa->spa_proc, zio_taskq_basedc, flags));
644 return (taskq_create_proc(name, value, maxclsyspri, 50, INT_MAX,
645 spa->spa_proc, flags));
649 spa_create_zio_taskqs(spa_t *spa)
651 for (int t = 0; t < ZIO_TYPES; t++) {
652 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
653 const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
654 enum zti_modes mode = ztip->zti_mode;
655 uint_t value = ztip->zti_value;
658 (void) snprintf(name, sizeof (name),
659 "%s_%s", zio_type_name[t], zio_taskq_types[q]);
661 spa->spa_zio_taskq[t][q] =
662 spa_taskq_create(spa, name, mode, value);
669 spa_thread(void *arg)
674 user_t *pu = PTOU(curproc);
676 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
679 ASSERT(curproc != &p0);
680 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
681 "zpool-%s", spa->spa_name);
682 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
684 /* bind this thread to the requested psrset */
685 if (zio_taskq_psrset_bind != PS_NONE) {
687 mutex_enter(&cpu_lock);
688 mutex_enter(&pidlock);
689 mutex_enter(&curproc->p_lock);
691 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
692 0, NULL, NULL) == 0) {
693 curthread->t_bind_pset = zio_taskq_psrset_bind;
696 "Couldn't bind process for zfs pool \"%s\" to "
697 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
700 mutex_exit(&curproc->p_lock);
701 mutex_exit(&pidlock);
702 mutex_exit(&cpu_lock);
706 if (zio_taskq_sysdc) {
707 sysdc_thread_enter(curthread, 100, 0);
710 spa->spa_proc = curproc;
711 spa->spa_did = curthread->t_did;
713 spa_create_zio_taskqs(spa);
715 mutex_enter(&spa->spa_proc_lock);
716 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
718 spa->spa_proc_state = SPA_PROC_ACTIVE;
719 cv_broadcast(&spa->spa_proc_cv);
721 CALLB_CPR_SAFE_BEGIN(&cprinfo);
722 while (spa->spa_proc_state == SPA_PROC_ACTIVE)
723 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
724 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
726 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
727 spa->spa_proc_state = SPA_PROC_GONE;
729 cv_broadcast(&spa->spa_proc_cv);
730 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */
732 mutex_enter(&curproc->p_lock);
738 * Activate an uninitialized pool.
741 spa_activate(spa_t *spa, int mode)
743 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
745 spa->spa_state = POOL_STATE_ACTIVE;
746 spa->spa_mode = mode;
748 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
749 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
751 /* Try to create a covering process */
752 mutex_enter(&spa->spa_proc_lock);
753 ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
754 ASSERT(spa->spa_proc == &p0);
757 /* Only create a process if we're going to be around a while. */
758 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
759 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
761 spa->spa_proc_state = SPA_PROC_CREATED;
762 while (spa->spa_proc_state == SPA_PROC_CREATED) {
763 cv_wait(&spa->spa_proc_cv,
764 &spa->spa_proc_lock);
766 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
767 ASSERT(spa->spa_proc != &p0);
768 ASSERT(spa->spa_did != 0);
772 "Couldn't create process for zfs pool \"%s\"\n",
777 mutex_exit(&spa->spa_proc_lock);
779 /* If we didn't create a process, we need to create our taskqs. */
780 if (spa->spa_proc == &p0) {
781 spa_create_zio_taskqs(spa);
784 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
785 offsetof(vdev_t, vdev_config_dirty_node));
786 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
787 offsetof(vdev_t, vdev_state_dirty_node));
789 txg_list_create(&spa->spa_vdev_txg_list,
790 offsetof(struct vdev, vdev_txg_node));
792 avl_create(&spa->spa_errlist_scrub,
793 spa_error_entry_compare, sizeof (spa_error_entry_t),
794 offsetof(spa_error_entry_t, se_avl));
795 avl_create(&spa->spa_errlist_last,
796 spa_error_entry_compare, sizeof (spa_error_entry_t),
797 offsetof(spa_error_entry_t, se_avl));
801 * Opposite of spa_activate().
804 spa_deactivate(spa_t *spa)
806 ASSERT(spa->spa_sync_on == B_FALSE);
807 ASSERT(spa->spa_dsl_pool == NULL);
808 ASSERT(spa->spa_root_vdev == NULL);
809 ASSERT(spa->spa_async_zio_root == NULL);
810 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
812 txg_list_destroy(&spa->spa_vdev_txg_list);
814 list_destroy(&spa->spa_config_dirty_list);
815 list_destroy(&spa->spa_state_dirty_list);
817 for (int t = 0; t < ZIO_TYPES; t++) {
818 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
819 if (spa->spa_zio_taskq[t][q] != NULL)
820 taskq_destroy(spa->spa_zio_taskq[t][q]);
821 spa->spa_zio_taskq[t][q] = NULL;
825 metaslab_class_destroy(spa->spa_normal_class);
826 spa->spa_normal_class = NULL;
828 metaslab_class_destroy(spa->spa_log_class);
829 spa->spa_log_class = NULL;
832 * If this was part of an import or the open otherwise failed, we may
833 * still have errors left in the queues. Empty them just in case.
835 spa_errlog_drain(spa);
837 avl_destroy(&spa->spa_errlist_scrub);
838 avl_destroy(&spa->spa_errlist_last);
840 spa->spa_state = POOL_STATE_UNINITIALIZED;
842 mutex_enter(&spa->spa_proc_lock);
843 if (spa->spa_proc_state != SPA_PROC_NONE) {
844 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
845 spa->spa_proc_state = SPA_PROC_DEACTIVATE;
846 cv_broadcast(&spa->spa_proc_cv);
847 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
848 ASSERT(spa->spa_proc != &p0);
849 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
851 ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
852 spa->spa_proc_state = SPA_PROC_NONE;
854 ASSERT(spa->spa_proc == &p0);
855 mutex_exit(&spa->spa_proc_lock);
858 * We want to make sure spa_thread() has actually exited the ZFS
859 * module, so that the module can't be unloaded out from underneath
862 if (spa->spa_did != 0) {
863 thread_join(spa->spa_did);
869 * Verify a pool configuration, and construct the vdev tree appropriately. This
870 * will create all the necessary vdevs in the appropriate layout, with each vdev
871 * in the CLOSED state. This will prep the pool before open/creation/import.
872 * All vdev validation is done by the vdev_alloc() routine.
875 spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
876 uint_t id, int atype)
882 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
885 if ((*vdp)->vdev_ops->vdev_op_leaf)
888 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
900 for (int c = 0; c < children; c++) {
902 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
910 ASSERT(*vdp != NULL);
916 * Opposite of spa_load().
919 spa_unload(spa_t *spa)
923 ASSERT(MUTEX_HELD(&spa_namespace_lock));
928 spa_async_suspend(spa);
933 if (spa->spa_sync_on) {
934 txg_sync_stop(spa->spa_dsl_pool);
935 spa->spa_sync_on = B_FALSE;
939 * Wait for any outstanding async I/O to complete.
941 if (spa->spa_async_zio_root != NULL) {
942 (void) zio_wait(spa->spa_async_zio_root);
943 spa->spa_async_zio_root = NULL;
946 bpobj_close(&spa->spa_deferred_bpobj);
949 * Close the dsl pool.
951 if (spa->spa_dsl_pool) {
952 dsl_pool_close(spa->spa_dsl_pool);
953 spa->spa_dsl_pool = NULL;
954 spa->spa_meta_objset = NULL;
959 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
962 * Drop and purge level 2 cache
964 spa_l2cache_drop(spa);
969 if (spa->spa_root_vdev)
970 vdev_free(spa->spa_root_vdev);
971 ASSERT(spa->spa_root_vdev == NULL);
973 for (i = 0; i < spa->spa_spares.sav_count; i++)
974 vdev_free(spa->spa_spares.sav_vdevs[i]);
975 if (spa->spa_spares.sav_vdevs) {
976 kmem_free(spa->spa_spares.sav_vdevs,
977 spa->spa_spares.sav_count * sizeof (void *));
978 spa->spa_spares.sav_vdevs = NULL;
980 if (spa->spa_spares.sav_config) {
981 nvlist_free(spa->spa_spares.sav_config);
982 spa->spa_spares.sav_config = NULL;
984 spa->spa_spares.sav_count = 0;
986 for (i = 0; i < spa->spa_l2cache.sav_count; i++)
987 vdev_free(spa->spa_l2cache.sav_vdevs[i]);
988 if (spa->spa_l2cache.sav_vdevs) {
989 kmem_free(spa->spa_l2cache.sav_vdevs,
990 spa->spa_l2cache.sav_count * sizeof (void *));
991 spa->spa_l2cache.sav_vdevs = NULL;
993 if (spa->spa_l2cache.sav_config) {
994 nvlist_free(spa->spa_l2cache.sav_config);
995 spa->spa_l2cache.sav_config = NULL;
997 spa->spa_l2cache.sav_count = 0;
999 spa->spa_async_suspended = 0;
1001 spa_config_exit(spa, SCL_ALL, FTAG);
1005 * Load (or re-load) the current list of vdevs describing the active spares for
1006 * this pool. When this is called, we have some form of basic information in
1007 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1008 * then re-generate a more complete list including status information.
1011 spa_load_spares(spa_t *spa)
1018 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1021 * First, close and free any existing spare vdevs.
1023 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1024 vd = spa->spa_spares.sav_vdevs[i];
1026 /* Undo the call to spa_activate() below */
1027 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1028 B_FALSE)) != NULL && tvd->vdev_isspare)
1029 spa_spare_remove(tvd);
1034 if (spa->spa_spares.sav_vdevs)
1035 kmem_free(spa->spa_spares.sav_vdevs,
1036 spa->spa_spares.sav_count * sizeof (void *));
1038 if (spa->spa_spares.sav_config == NULL)
1041 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1042 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1044 spa->spa_spares.sav_count = (int)nspares;
1045 spa->spa_spares.sav_vdevs = NULL;
1051 * Construct the array of vdevs, opening them to get status in the
1052 * process. For each spare, there is potentially two different vdev_t
1053 * structures associated with it: one in the list of spares (used only
1054 * for basic validation purposes) and one in the active vdev
1055 * configuration (if it's spared in). During this phase we open and
1056 * validate each vdev on the spare list. If the vdev also exists in the
1057 * active configuration, then we also mark this vdev as an active spare.
1059 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1061 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1062 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1063 VDEV_ALLOC_SPARE) == 0);
1066 spa->spa_spares.sav_vdevs[i] = vd;
1068 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1069 B_FALSE)) != NULL) {
1070 if (!tvd->vdev_isspare)
1074 * We only mark the spare active if we were successfully
1075 * able to load the vdev. Otherwise, importing a pool
1076 * with a bad active spare would result in strange
1077 * behavior, because multiple pool would think the spare
1078 * is actively in use.
1080 * There is a vulnerability here to an equally bizarre
1081 * circumstance, where a dead active spare is later
1082 * brought back to life (onlined or otherwise). Given
1083 * the rarity of this scenario, and the extra complexity
1084 * it adds, we ignore the possibility.
1086 if (!vdev_is_dead(tvd))
1087 spa_spare_activate(tvd);
1091 vd->vdev_aux = &spa->spa_spares;
1093 if (vdev_open(vd) != 0)
1096 if (vdev_validate_aux(vd) == 0)
1101 * Recompute the stashed list of spares, with status information
1104 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1105 DATA_TYPE_NVLIST_ARRAY) == 0);
1107 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1109 for (i = 0; i < spa->spa_spares.sav_count; i++)
1110 spares[i] = vdev_config_generate(spa,
1111 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1112 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1113 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1114 for (i = 0; i < spa->spa_spares.sav_count; i++)
1115 nvlist_free(spares[i]);
1116 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1120 * Load (or re-load) the current list of vdevs describing the active l2cache for
1121 * this pool. When this is called, we have some form of basic information in
1122 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1123 * then re-generate a more complete list including status information.
1124 * Devices which are already active have their details maintained, and are
1128 spa_load_l2cache(spa_t *spa)
1132 int i, j, oldnvdevs;
1134 vdev_t *vd, **oldvdevs, **newvdevs;
1135 spa_aux_vdev_t *sav = &spa->spa_l2cache;
1137 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1139 if (sav->sav_config != NULL) {
1140 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1141 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1142 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1147 oldvdevs = sav->sav_vdevs;
1148 oldnvdevs = sav->sav_count;
1149 sav->sav_vdevs = NULL;
1153 * Process new nvlist of vdevs.
1155 for (i = 0; i < nl2cache; i++) {
1156 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1160 for (j = 0; j < oldnvdevs; j++) {
1162 if (vd != NULL && guid == vd->vdev_guid) {
1164 * Retain previous vdev for add/remove ops.
1172 if (newvdevs[i] == NULL) {
1176 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1177 VDEV_ALLOC_L2CACHE) == 0);
1182 * Commit this vdev as an l2cache device,
1183 * even if it fails to open.
1185 spa_l2cache_add(vd);
1190 spa_l2cache_activate(vd);
1192 if (vdev_open(vd) != 0)
1195 (void) vdev_validate_aux(vd);
1197 if (!vdev_is_dead(vd))
1198 l2arc_add_vdev(spa, vd);
1203 * Purge vdevs that were dropped
1205 for (i = 0; i < oldnvdevs; i++) {
1210 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1211 pool != 0ULL && l2arc_vdev_present(vd))
1212 l2arc_remove_vdev(vd);
1213 (void) vdev_close(vd);
1214 spa_l2cache_remove(vd);
1219 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1221 if (sav->sav_config == NULL)
1224 sav->sav_vdevs = newvdevs;
1225 sav->sav_count = (int)nl2cache;
1228 * Recompute the stashed list of l2cache devices, with status
1229 * information this time.
1231 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1232 DATA_TYPE_NVLIST_ARRAY) == 0);
1234 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1235 for (i = 0; i < sav->sav_count; i++)
1236 l2cache[i] = vdev_config_generate(spa,
1237 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1238 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1239 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1241 for (i = 0; i < sav->sav_count; i++)
1242 nvlist_free(l2cache[i]);
1244 kmem_free(l2cache, sav->sav_count * sizeof (void *));
1248 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1251 char *packed = NULL;
1256 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
1257 nvsize = *(uint64_t *)db->db_data;
1258 dmu_buf_rele(db, FTAG);
1260 packed = kmem_alloc(nvsize, KM_SLEEP);
1261 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1264 error = nvlist_unpack(packed, nvsize, value, 0);
1265 kmem_free(packed, nvsize);
1271 * Checks to see if the given vdev could not be opened, in which case we post a
1272 * sysevent to notify the autoreplace code that the device has been removed.
1275 spa_check_removed(vdev_t *vd)
1277 for (int c = 0; c < vd->vdev_children; c++)
1278 spa_check_removed(vd->vdev_child[c]);
1280 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd)) {
1281 zfs_post_autoreplace(vd->vdev_spa, vd);
1282 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1287 * Load the slog device state from the config object since it's possible
1288 * that the label does not contain the most up-to-date information.
1291 spa_load_log_state(spa_t *spa, nvlist_t *nv)
1293 vdev_t *ovd, *rvd = spa->spa_root_vdev;
1296 * Load the original root vdev tree from the passed config.
1298 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1299 VERIFY(spa_config_parse(spa, &ovd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1301 for (int c = 0; c < rvd->vdev_children; c++) {
1302 vdev_t *cvd = rvd->vdev_child[c];
1303 if (cvd->vdev_islog)
1304 vdev_load_log_state(cvd, ovd->vdev_child[c]);
1307 spa_config_exit(spa, SCL_ALL, FTAG);
1311 * Check for missing log devices
1314 spa_check_logs(spa_t *spa)
1316 switch (spa->spa_log_state) {
1317 case SPA_LOG_MISSING:
1318 /* need to recheck in case slog has been restored */
1319 case SPA_LOG_UNKNOWN:
1320 if (dmu_objset_find(spa->spa_name, zil_check_log_chain, NULL,
1321 DS_FIND_CHILDREN)) {
1322 spa_set_log_state(spa, SPA_LOG_MISSING);
1331 spa_passivate_log(spa_t *spa)
1333 vdev_t *rvd = spa->spa_root_vdev;
1334 boolean_t slog_found = B_FALSE;
1336 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1338 if (!spa_has_slogs(spa))
1341 for (int c = 0; c < rvd->vdev_children; c++) {
1342 vdev_t *tvd = rvd->vdev_child[c];
1343 metaslab_group_t *mg = tvd->vdev_mg;
1345 if (tvd->vdev_islog) {
1346 metaslab_group_passivate(mg);
1347 slog_found = B_TRUE;
1351 return (slog_found);
1355 spa_activate_log(spa_t *spa)
1357 vdev_t *rvd = spa->spa_root_vdev;
1359 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1361 for (int c = 0; c < rvd->vdev_children; c++) {
1362 vdev_t *tvd = rvd->vdev_child[c];
1363 metaslab_group_t *mg = tvd->vdev_mg;
1365 if (tvd->vdev_islog)
1366 metaslab_group_activate(mg);
1371 spa_offline_log(spa_t *spa)
1375 if ((error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1376 NULL, DS_FIND_CHILDREN)) == 0) {
1379 * We successfully offlined the log device, sync out the
1380 * current txg so that the "stubby" block can be removed
1383 txg_wait_synced(spa->spa_dsl_pool, 0);
1389 spa_aux_check_removed(spa_aux_vdev_t *sav)
1391 for (int i = 0; i < sav->sav_count; i++)
1392 spa_check_removed(sav->sav_vdevs[i]);
1396 spa_claim_notify(zio_t *zio)
1398 spa_t *spa = zio->io_spa;
1403 mutex_enter(&spa->spa_props_lock); /* any mutex will do */
1404 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1405 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1406 mutex_exit(&spa->spa_props_lock);
1409 typedef struct spa_load_error {
1410 uint64_t sle_meta_count;
1411 uint64_t sle_data_count;
1415 spa_load_verify_done(zio_t *zio)
1417 blkptr_t *bp = zio->io_bp;
1418 spa_load_error_t *sle = zio->io_private;
1419 dmu_object_type_t type = BP_GET_TYPE(bp);
1420 int error = zio->io_error;
1423 if ((BP_GET_LEVEL(bp) != 0 || dmu_ot[type].ot_metadata) &&
1424 type != DMU_OT_INTENT_LOG)
1425 atomic_add_64(&sle->sle_meta_count, 1);
1427 atomic_add_64(&sle->sle_data_count, 1);
1429 zio_data_buf_free(zio->io_data, zio->io_size);
1434 spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1435 arc_buf_t *pbuf, const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1439 size_t size = BP_GET_PSIZE(bp);
1440 void *data = zio_data_buf_alloc(size);
1442 zio_nowait(zio_read(rio, spa, bp, data, size,
1443 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1444 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1445 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1451 spa_load_verify(spa_t *spa)
1454 spa_load_error_t sle = { 0 };
1455 zpool_rewind_policy_t policy;
1456 boolean_t verify_ok = B_FALSE;
1459 zpool_get_rewind_policy(spa->spa_config, &policy);
1461 if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1464 rio = zio_root(spa, NULL, &sle,
1465 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1467 error = traverse_pool(spa, spa->spa_verify_min_txg,
1468 TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio);
1470 (void) zio_wait(rio);
1472 spa->spa_load_meta_errors = sle.sle_meta_count;
1473 spa->spa_load_data_errors = sle.sle_data_count;
1475 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1476 sle.sle_data_count <= policy.zrp_maxdata) {
1478 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1479 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1481 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1485 if (error != ENXIO && error != EIO)
1490 return (verify_ok ? 0 : EIO);
1494 * Find a value in the pool props object.
1497 spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1499 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1500 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1504 * Find a value in the pool directory object.
1507 spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1509 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1510 name, sizeof (uint64_t), 1, val));
1514 spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1516 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1521 * Fix up config after a partly-completed split. This is done with the
1522 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
1523 * pool have that entry in their config, but only the splitting one contains
1524 * a list of all the guids of the vdevs that are being split off.
1526 * This function determines what to do with that list: either rejoin
1527 * all the disks to the pool, or complete the splitting process. To attempt
1528 * the rejoin, each disk that is offlined is marked online again, and
1529 * we do a reopen() call. If the vdev label for every disk that was
1530 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1531 * then we call vdev_split() on each disk, and complete the split.
1533 * Otherwise we leave the config alone, with all the vdevs in place in
1534 * the original pool.
1537 spa_try_repair(spa_t *spa, nvlist_t *config)
1544 boolean_t attempt_reopen;
1546 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
1549 /* check that the config is complete */
1550 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
1551 &glist, &gcount) != 0)
1554 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
1556 /* attempt to online all the vdevs & validate */
1557 attempt_reopen = B_TRUE;
1558 for (i = 0; i < gcount; i++) {
1559 if (glist[i] == 0) /* vdev is hole */
1562 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
1563 if (vd[i] == NULL) {
1565 * Don't bother attempting to reopen the disks;
1566 * just do the split.
1568 attempt_reopen = B_FALSE;
1570 /* attempt to re-online it */
1571 vd[i]->vdev_offline = B_FALSE;
1575 if (attempt_reopen) {
1576 vdev_reopen(spa->spa_root_vdev);
1578 /* check each device to see what state it's in */
1579 for (extracted = 0, i = 0; i < gcount; i++) {
1580 if (vd[i] != NULL &&
1581 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
1588 * If every disk has been moved to the new pool, or if we never
1589 * even attempted to look at them, then we split them off for
1592 if (!attempt_reopen || gcount == extracted) {
1593 for (i = 0; i < gcount; i++)
1596 vdev_reopen(spa->spa_root_vdev);
1599 kmem_free(vd, gcount * sizeof (vdev_t *));
1603 spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
1604 boolean_t mosconfig)
1606 nvlist_t *config = spa->spa_config;
1607 char *ereport = FM_EREPORT_ZFS_POOL;
1612 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
1616 * Versioning wasn't explicitly added to the label until later, so if
1617 * it's not present treat it as the initial version.
1619 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
1620 &spa->spa_ubsync.ub_version) != 0)
1621 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
1623 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
1624 &spa->spa_config_txg);
1626 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
1627 spa_guid_exists(pool_guid, 0)) {
1630 spa->spa_load_guid = pool_guid;
1632 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
1634 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
1638 error = spa_load_impl(spa, pool_guid, config, state, type,
1639 mosconfig, &ereport);
1642 spa->spa_minref = refcount_count(&spa->spa_refcount);
1643 if (error && error != EBADF)
1644 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
1645 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
1652 * Load an existing storage pool, using the pool's builtin spa_config as a
1653 * source of configuration information.
1656 spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
1657 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
1661 nvlist_t *nvroot = NULL;
1663 uberblock_t *ub = &spa->spa_uberblock;
1664 uint64_t config_cache_txg = spa->spa_config_txg;
1665 int orig_mode = spa->spa_mode;
1670 * If this is an untrusted config, access the pool in read-only mode.
1671 * This prevents things like resilvering recently removed devices.
1674 spa->spa_mode = FREAD;
1676 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1678 spa->spa_load_state = state;
1680 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
1683 parse = (type == SPA_IMPORT_EXISTING ?
1684 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
1687 * Create "The Godfather" zio to hold all async IOs
1689 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
1690 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
1693 * Parse the configuration into a vdev tree. We explicitly set the
1694 * value that will be returned by spa_version() since parsing the
1695 * configuration requires knowing the version number.
1697 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1698 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
1699 spa_config_exit(spa, SCL_ALL, FTAG);
1704 ASSERT(spa->spa_root_vdev == rvd);
1706 if (type != SPA_IMPORT_ASSEMBLE) {
1707 ASSERT(spa_guid(spa) == pool_guid);
1711 * Try to open all vdevs, loading each label in the process.
1713 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1714 error = vdev_open(rvd);
1715 spa_config_exit(spa, SCL_ALL, FTAG);
1720 * We need to validate the vdev labels against the configuration that
1721 * we have in hand, which is dependent on the setting of mosconfig. If
1722 * mosconfig is true then we're validating the vdev labels based on
1723 * that config. Otherwise, we're validating against the cached config
1724 * (zpool.cache) that was read when we loaded the zfs module, and then
1725 * later we will recursively call spa_load() and validate against
1728 * If we're assembling a new pool that's been split off from an
1729 * existing pool, the labels haven't yet been updated so we skip
1730 * validation for now.
1732 if (type != SPA_IMPORT_ASSEMBLE) {
1733 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1734 error = vdev_validate(rvd);
1735 spa_config_exit(spa, SCL_ALL, FTAG);
1740 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
1745 * Find the best uberblock.
1747 vdev_uberblock_load(NULL, rvd, ub);
1750 * If we weren't able to find a single valid uberblock, return failure.
1752 if (ub->ub_txg == 0)
1753 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
1756 * If the pool is newer than the code, we can't open it.
1758 if (ub->ub_version > SPA_VERSION)
1759 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
1762 * If the vdev guid sum doesn't match the uberblock, we have an
1763 * incomplete configuration.
1765 if (mosconfig && type != SPA_IMPORT_ASSEMBLE &&
1766 rvd->vdev_guid_sum != ub->ub_guid_sum)
1767 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
1769 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
1770 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1771 spa_try_repair(spa, config);
1772 spa_config_exit(spa, SCL_ALL, FTAG);
1773 nvlist_free(spa->spa_config_splitting);
1774 spa->spa_config_splitting = NULL;
1778 * Initialize internal SPA structures.
1780 spa->spa_state = POOL_STATE_ACTIVE;
1781 spa->spa_ubsync = spa->spa_uberblock;
1782 spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
1783 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
1784 spa->spa_first_txg = spa->spa_last_ubsync_txg ?
1785 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
1786 spa->spa_claim_max_txg = spa->spa_first_txg;
1787 spa->spa_prev_software_version = ub->ub_software_version;
1789 error = dsl_pool_open(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
1791 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1792 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
1794 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
1795 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1799 nvlist_t *policy = NULL, *nvconfig;
1801 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
1802 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1804 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
1805 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
1807 unsigned long myhostid = 0;
1809 VERIFY(nvlist_lookup_string(nvconfig,
1810 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
1813 myhostid = zone_get_hostid(NULL);
1816 * We're emulating the system's hostid in userland, so
1817 * we can't use zone_get_hostid().
1819 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
1820 #endif /* _KERNEL */
1821 if (hostid != 0 && myhostid != 0 &&
1822 hostid != myhostid) {
1823 nvlist_free(nvconfig);
1824 cmn_err(CE_WARN, "pool '%s' could not be "
1825 "loaded as it was last accessed by "
1826 "another system (host: %s hostid: 0x%lx). "
1827 "See: http://www.sun.com/msg/ZFS-8000-EY",
1828 spa_name(spa), hostname,
1829 (unsigned long)hostid);
1833 if (nvlist_lookup_nvlist(spa->spa_config,
1834 ZPOOL_REWIND_POLICY, &policy) == 0)
1835 VERIFY(nvlist_add_nvlist(nvconfig,
1836 ZPOOL_REWIND_POLICY, policy) == 0);
1838 spa_config_set(spa, nvconfig);
1840 spa_deactivate(spa);
1841 spa_activate(spa, orig_mode);
1843 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
1846 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
1847 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1848 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
1850 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1853 * Load the bit that tells us to use the new accounting function
1854 * (raid-z deflation). If we have an older pool, this will not
1857 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
1858 if (error != 0 && error != ENOENT)
1859 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1861 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
1862 &spa->spa_creation_version);
1863 if (error != 0 && error != ENOENT)
1864 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1867 * Load the persistent error log. If we have an older pool, this will
1870 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
1871 if (error != 0 && error != ENOENT)
1872 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1874 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
1875 &spa->spa_errlog_scrub);
1876 if (error != 0 && error != ENOENT)
1877 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1880 * Load the history object. If we have an older pool, this
1881 * will not be present.
1883 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
1884 if (error != 0 && error != ENOENT)
1885 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1888 * If we're assembling the pool from the split-off vdevs of
1889 * an existing pool, we don't want to attach the spares & cache
1894 * Load any hot spares for this pool.
1896 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
1897 if (error != 0 && error != ENOENT)
1898 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1899 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
1900 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
1901 if (load_nvlist(spa, spa->spa_spares.sav_object,
1902 &spa->spa_spares.sav_config) != 0)
1903 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1905 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1906 spa_load_spares(spa);
1907 spa_config_exit(spa, SCL_ALL, FTAG);
1908 } else if (error == 0) {
1909 spa->spa_spares.sav_sync = B_TRUE;
1913 * Load any level 2 ARC devices for this pool.
1915 error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
1916 &spa->spa_l2cache.sav_object);
1917 if (error != 0 && error != ENOENT)
1918 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1919 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
1920 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
1921 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
1922 &spa->spa_l2cache.sav_config) != 0)
1923 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1925 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1926 spa_load_l2cache(spa);
1927 spa_config_exit(spa, SCL_ALL, FTAG);
1928 } else if (error == 0) {
1929 spa->spa_l2cache.sav_sync = B_TRUE;
1932 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
1934 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
1935 if (error && error != ENOENT)
1936 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1939 uint64_t autoreplace;
1941 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
1942 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
1943 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
1944 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
1945 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
1946 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
1947 &spa->spa_dedup_ditto);
1949 spa->spa_autoreplace = (autoreplace != 0);
1953 * If the 'autoreplace' property is set, then post a resource notifying
1954 * the ZFS DE that it should not issue any faults for unopenable
1955 * devices. We also iterate over the vdevs, and post a sysevent for any
1956 * unopenable vdevs so that the normal autoreplace handler can take
1959 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
1960 spa_check_removed(spa->spa_root_vdev);
1962 * For the import case, this is done in spa_import(), because
1963 * at this point we're using the spare definitions from
1964 * the MOS config, not necessarily from the userland config.
1966 if (state != SPA_LOAD_IMPORT) {
1967 spa_aux_check_removed(&spa->spa_spares);
1968 spa_aux_check_removed(&spa->spa_l2cache);
1973 * Load the vdev state for all toplevel vdevs.
1978 * Propagate the leaf DTLs we just loaded all the way up the tree.
1980 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1981 vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
1982 spa_config_exit(spa, SCL_ALL, FTAG);
1985 * Check the state of the root vdev. If it can't be opened, it
1986 * indicates one or more toplevel vdevs are faulted.
1988 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
1992 * Load the DDTs (dedup tables).
1994 error = ddt_load(spa);
1996 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1998 spa_update_dspace(spa);
2000 if (state != SPA_LOAD_TRYIMPORT) {
2001 error = spa_load_verify(spa);
2003 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2008 * Load the intent log state and check log integrity. If we're
2009 * assembling a pool from a split, the log is not transferred over.
2011 if (type != SPA_IMPORT_ASSEMBLE) {
2014 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2015 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2017 VERIFY(nvlist_lookup_nvlist(nvconfig, ZPOOL_CONFIG_VDEV_TREE,
2019 spa_load_log_state(spa, nvroot);
2020 nvlist_free(nvconfig);
2022 if (spa_check_logs(spa)) {
2023 *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2024 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2028 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2029 spa->spa_load_max_txg == UINT64_MAX)) {
2031 int need_update = B_FALSE;
2033 ASSERT(state != SPA_LOAD_TRYIMPORT);
2036 * Claim log blocks that haven't been committed yet.
2037 * This must all happen in a single txg.
2038 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2039 * invoked from zil_claim_log_block()'s i/o done callback.
2040 * Price of rollback is that we abandon the log.
2042 spa->spa_claiming = B_TRUE;
2044 tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2045 spa_first_txg(spa));
2046 (void) dmu_objset_find(spa_name(spa),
2047 zil_claim, tx, DS_FIND_CHILDREN);
2050 spa->spa_claiming = B_FALSE;
2052 spa_set_log_state(spa, SPA_LOG_GOOD);
2053 spa->spa_sync_on = B_TRUE;
2054 txg_sync_start(spa->spa_dsl_pool);
2057 * Wait for all claims to sync. We sync up to the highest
2058 * claimed log block birth time so that claimed log blocks
2059 * don't appear to be from the future. spa_claim_max_txg
2060 * will have been set for us by either zil_check_log_chain()
2061 * (invoked from spa_check_logs()) or zil_claim() above.
2063 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2066 * If the config cache is stale, or we have uninitialized
2067 * metaslabs (see spa_vdev_add()), then update the config.
2069 * If spa_load_verbatim is true, trust the current
2070 * in-core spa_config and update the disk labels.
2072 if (config_cache_txg != spa->spa_config_txg ||
2073 state == SPA_LOAD_IMPORT || spa->spa_load_verbatim ||
2074 state == SPA_LOAD_RECOVER)
2075 need_update = B_TRUE;
2077 for (int c = 0; c < rvd->vdev_children; c++)
2078 if (rvd->vdev_child[c]->vdev_ms_array == 0)
2079 need_update = B_TRUE;
2082 * Update the config cache asychronously in case we're the
2083 * root pool, in which case the config cache isn't writable yet.
2086 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2089 * Check all DTLs to see if anything needs resilvering.
2091 if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2092 vdev_resilver_needed(rvd, NULL, NULL))
2093 spa_async_request(spa, SPA_ASYNC_RESILVER);
2096 * Delete any inconsistent datasets.
2098 (void) dmu_objset_find(spa_name(spa),
2099 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2102 * Clean up any stale temporary dataset userrefs.
2104 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2111 spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2114 spa_deactivate(spa);
2116 spa->spa_load_max_txg--;
2118 spa_activate(spa, spa_mode_global);
2119 spa_async_suspend(spa);
2121 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2125 spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2126 uint64_t max_request, int rewind_flags)
2128 nvlist_t *config = NULL;
2129 int load_error, rewind_error;
2130 uint64_t safe_rewind_txg;
2133 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2134 spa->spa_load_max_txg = spa->spa_load_txg;
2135 spa_set_log_state(spa, SPA_LOG_CLEAR);
2137 spa->spa_load_max_txg = max_request;
2140 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2142 if (load_error == 0)
2145 if (spa->spa_root_vdev != NULL)
2146 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2148 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2149 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2151 if (rewind_flags & ZPOOL_NEVER_REWIND) {
2152 nvlist_free(config);
2153 return (load_error);
2156 /* Price of rolling back is discarding txgs, including log */
2157 if (state == SPA_LOAD_RECOVER)
2158 spa_set_log_state(spa, SPA_LOG_CLEAR);
2160 spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2161 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2162 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2163 TXG_INITIAL : safe_rewind_txg;
2166 * Continue as long as we're finding errors, we're still within
2167 * the acceptable rewind range, and we're still finding uberblocks
2169 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2170 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2171 if (spa->spa_load_max_txg < safe_rewind_txg)
2172 spa->spa_extreme_rewind = B_TRUE;
2173 rewind_error = spa_load_retry(spa, state, mosconfig);
2177 spa_rewind_data_to_nvlist(spa, config);
2179 spa->spa_extreme_rewind = B_FALSE;
2180 spa->spa_load_max_txg = UINT64_MAX;
2182 if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2183 spa_config_set(spa, config);
2185 return (state == SPA_LOAD_RECOVER ? rewind_error : load_error);
2191 * The import case is identical to an open except that the configuration is sent
2192 * down from userland, instead of grabbed from the configuration cache. For the
2193 * case of an open, the pool configuration will exist in the
2194 * POOL_STATE_UNINITIALIZED state.
2196 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2197 * the same time open the pool, without having to keep around the spa_t in some
2201 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2206 int locked = B_FALSE;
2211 * As disgusting as this is, we need to support recursive calls to this
2212 * function because dsl_dir_open() is called during spa_load(), and ends
2213 * up calling spa_open() again. The real fix is to figure out how to
2214 * avoid dsl_dir_open() calling this in the first place.
2216 if (mutex_owner(&spa_namespace_lock) != curthread) {
2217 mutex_enter(&spa_namespace_lock);
2221 if ((spa = spa_lookup(pool)) == NULL) {
2223 mutex_exit(&spa_namespace_lock);
2227 if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2228 spa_load_state_t state = SPA_LOAD_OPEN;
2229 zpool_rewind_policy_t policy;
2231 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2233 if (policy.zrp_request & ZPOOL_DO_REWIND)
2234 state = SPA_LOAD_RECOVER;
2236 spa_activate(spa, spa_mode_global);
2238 if (state != SPA_LOAD_RECOVER)
2239 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2241 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2242 policy.zrp_request);
2244 if (error == EBADF) {
2246 * If vdev_validate() returns failure (indicated by
2247 * EBADF), it indicates that one of the vdevs indicates
2248 * that the pool has been exported or destroyed. If
2249 * this is the case, the config cache is out of sync and
2250 * we should remove the pool from the namespace.
2253 spa_deactivate(spa);
2254 spa_config_sync(spa, B_TRUE, B_TRUE);
2257 mutex_exit(&spa_namespace_lock);
2263 * We can't open the pool, but we still have useful
2264 * information: the state of each vdev after the
2265 * attempted vdev_open(). Return this to the user.
2267 if (config != NULL && spa->spa_config)
2268 VERIFY(nvlist_dup(spa->spa_config, config,
2271 spa_deactivate(spa);
2272 spa->spa_last_open_failed = error;
2274 mutex_exit(&spa_namespace_lock);
2281 spa_open_ref(spa, tag);
2285 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2288 spa->spa_last_open_failed = 0;
2289 spa->spa_last_ubsync_txg = 0;
2290 spa->spa_load_txg = 0;
2291 mutex_exit(&spa_namespace_lock);
2300 spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
2303 return (spa_open_common(name, spapp, tag, policy, config));
2307 spa_open(const char *name, spa_t **spapp, void *tag)
2309 return (spa_open_common(name, spapp, tag, NULL, NULL));
2313 * Lookup the given spa_t, incrementing the inject count in the process,
2314 * preventing it from being exported or destroyed.
2317 spa_inject_addref(char *name)
2321 mutex_enter(&spa_namespace_lock);
2322 if ((spa = spa_lookup(name)) == NULL) {
2323 mutex_exit(&spa_namespace_lock);
2326 spa->spa_inject_ref++;
2327 mutex_exit(&spa_namespace_lock);
2333 spa_inject_delref(spa_t *spa)
2335 mutex_enter(&spa_namespace_lock);
2336 spa->spa_inject_ref--;
2337 mutex_exit(&spa_namespace_lock);
2341 * Add spares device information to the nvlist.
2344 spa_add_spares(spa_t *spa, nvlist_t *config)
2354 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2356 if (spa->spa_spares.sav_count == 0)
2359 VERIFY(nvlist_lookup_nvlist(config,
2360 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2361 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
2362 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2364 VERIFY(nvlist_add_nvlist_array(nvroot,
2365 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2366 VERIFY(nvlist_lookup_nvlist_array(nvroot,
2367 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2370 * Go through and find any spares which have since been
2371 * repurposed as an active spare. If this is the case, update
2372 * their status appropriately.
2374 for (i = 0; i < nspares; i++) {
2375 VERIFY(nvlist_lookup_uint64(spares[i],
2376 ZPOOL_CONFIG_GUID, &guid) == 0);
2377 if (spa_spare_exists(guid, &pool, NULL) &&
2379 VERIFY(nvlist_lookup_uint64_array(
2380 spares[i], ZPOOL_CONFIG_VDEV_STATS,
2381 (uint64_t **)&vs, &vsc) == 0);
2382 vs->vs_state = VDEV_STATE_CANT_OPEN;
2383 vs->vs_aux = VDEV_AUX_SPARED;
2390 * Add l2cache device information to the nvlist, including vdev stats.
2393 spa_add_l2cache(spa_t *spa, nvlist_t *config)
2396 uint_t i, j, nl2cache;
2403 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2405 if (spa->spa_l2cache.sav_count == 0)
2408 VERIFY(nvlist_lookup_nvlist(config,
2409 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2410 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
2411 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2412 if (nl2cache != 0) {
2413 VERIFY(nvlist_add_nvlist_array(nvroot,
2414 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2415 VERIFY(nvlist_lookup_nvlist_array(nvroot,
2416 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2419 * Update level 2 cache device stats.
2422 for (i = 0; i < nl2cache; i++) {
2423 VERIFY(nvlist_lookup_uint64(l2cache[i],
2424 ZPOOL_CONFIG_GUID, &guid) == 0);
2427 for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
2429 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
2430 vd = spa->spa_l2cache.sav_vdevs[j];
2436 VERIFY(nvlist_lookup_uint64_array(l2cache[i],
2437 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
2439 vdev_get_stats(vd, vs);
2445 spa_get_stats(const char *name, nvlist_t **config, char *altroot, size_t buflen)
2451 error = spa_open_common(name, &spa, FTAG, NULL, config);
2455 * This still leaves a window of inconsistency where the spares
2456 * or l2cache devices could change and the config would be
2457 * self-inconsistent.
2459 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
2461 if (*config != NULL) {
2462 VERIFY(nvlist_add_uint64(*config,
2463 ZPOOL_CONFIG_ERRCOUNT,
2464 spa_get_errlog_size(spa)) == 0);
2466 if (spa_suspended(spa))
2467 VERIFY(nvlist_add_uint64(*config,
2468 ZPOOL_CONFIG_SUSPENDED,
2469 spa->spa_failmode) == 0);
2471 spa_add_spares(spa, *config);
2472 spa_add_l2cache(spa, *config);
2477 * We want to get the alternate root even for faulted pools, so we cheat
2478 * and call spa_lookup() directly.
2482 mutex_enter(&spa_namespace_lock);
2483 spa = spa_lookup(name);
2485 spa_altroot(spa, altroot, buflen);
2489 mutex_exit(&spa_namespace_lock);
2491 spa_altroot(spa, altroot, buflen);
2496 spa_config_exit(spa, SCL_CONFIG, FTAG);
2497 spa_close(spa, FTAG);
2504 * Validate that the auxiliary device array is well formed. We must have an
2505 * array of nvlists, each which describes a valid leaf vdev. If this is an
2506 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
2507 * specified, as long as they are well-formed.
2510 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
2511 spa_aux_vdev_t *sav, const char *config, uint64_t version,
2512 vdev_labeltype_t label)
2519 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
2522 * It's acceptable to have no devs specified.
2524 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
2531 * Make sure the pool is formatted with a version that supports this
2534 if (spa_version(spa) < version)
2538 * Set the pending device list so we correctly handle device in-use
2541 sav->sav_pending = dev;
2542 sav->sav_npending = ndev;
2544 for (i = 0; i < ndev; i++) {
2545 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
2549 if (!vd->vdev_ops->vdev_op_leaf) {
2556 * The L2ARC currently only supports disk devices in
2557 * kernel context. For user-level testing, we allow it.
2560 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
2561 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
2568 if ((error = vdev_open(vd)) == 0 &&
2569 (error = vdev_label_init(vd, crtxg, label)) == 0) {
2570 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
2571 vd->vdev_guid) == 0);
2577 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
2584 sav->sav_pending = NULL;
2585 sav->sav_npending = 0;
2590 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
2594 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
2596 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
2597 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
2598 VDEV_LABEL_SPARE)) != 0) {
2602 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
2603 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
2604 VDEV_LABEL_L2CACHE));
2608 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
2613 if (sav->sav_config != NULL) {
2619 * Generate new dev list by concatentating with the
2622 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
2623 &olddevs, &oldndevs) == 0);
2625 newdevs = kmem_alloc(sizeof (void *) *
2626 (ndevs + oldndevs), KM_SLEEP);
2627 for (i = 0; i < oldndevs; i++)
2628 VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
2630 for (i = 0; i < ndevs; i++)
2631 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
2634 VERIFY(nvlist_remove(sav->sav_config, config,
2635 DATA_TYPE_NVLIST_ARRAY) == 0);
2637 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
2638 config, newdevs, ndevs + oldndevs) == 0);
2639 for (i = 0; i < oldndevs + ndevs; i++)
2640 nvlist_free(newdevs[i]);
2641 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
2644 * Generate a new dev list.
2646 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
2648 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
2654 * Stop and drop level 2 ARC devices
2657 spa_l2cache_drop(spa_t *spa)
2661 spa_aux_vdev_t *sav = &spa->spa_l2cache;
2663 for (i = 0; i < sav->sav_count; i++) {
2666 vd = sav->sav_vdevs[i];
2669 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
2670 pool != 0ULL && l2arc_vdev_present(vd))
2671 l2arc_remove_vdev(vd);
2672 if (vd->vdev_isl2cache)
2673 spa_l2cache_remove(vd);
2674 vdev_clear_stats(vd);
2675 (void) vdev_close(vd);
2683 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
2684 const char *history_str, nvlist_t *zplprops)
2687 char *altroot = NULL;
2692 uint64_t txg = TXG_INITIAL;
2693 nvlist_t **spares, **l2cache;
2694 uint_t nspares, nl2cache;
2695 uint64_t version, obj;
2698 * If this pool already exists, return failure.
2700 mutex_enter(&spa_namespace_lock);
2701 if (spa_lookup(pool) != NULL) {
2702 mutex_exit(&spa_namespace_lock);
2707 * Allocate a new spa_t structure.
2709 (void) nvlist_lookup_string(props,
2710 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
2711 spa = spa_add(pool, NULL, altroot);
2712 spa_activate(spa, spa_mode_global);
2714 if (props && (error = spa_prop_validate(spa, props))) {
2715 spa_deactivate(spa);
2717 mutex_exit(&spa_namespace_lock);
2721 if (nvlist_lookup_uint64(props, zpool_prop_to_name(ZPOOL_PROP_VERSION),
2723 version = SPA_VERSION;
2724 ASSERT(version <= SPA_VERSION);
2726 spa->spa_first_txg = txg;
2727 spa->spa_uberblock.ub_txg = txg - 1;
2728 spa->spa_uberblock.ub_version = version;
2729 spa->spa_ubsync = spa->spa_uberblock;
2732 * Create "The Godfather" zio to hold all async IOs
2734 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2735 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2738 * Create the root vdev.
2740 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2742 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
2744 ASSERT(error != 0 || rvd != NULL);
2745 ASSERT(error != 0 || spa->spa_root_vdev == rvd);
2747 if (error == 0 && !zfs_allocatable_devs(nvroot))
2751 (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
2752 (error = spa_validate_aux(spa, nvroot, txg,
2753 VDEV_ALLOC_ADD)) == 0) {
2754 for (int c = 0; c < rvd->vdev_children; c++) {
2755 vdev_metaslab_set_size(rvd->vdev_child[c]);
2756 vdev_expand(rvd->vdev_child[c], txg);
2760 spa_config_exit(spa, SCL_ALL, FTAG);
2764 spa_deactivate(spa);
2766 mutex_exit(&spa_namespace_lock);
2771 * Get the list of spares, if specified.
2773 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
2774 &spares, &nspares) == 0) {
2775 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
2777 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
2778 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2779 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2780 spa_load_spares(spa);
2781 spa_config_exit(spa, SCL_ALL, FTAG);
2782 spa->spa_spares.sav_sync = B_TRUE;
2786 * Get the list of level 2 cache devices, if specified.
2788 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
2789 &l2cache, &nl2cache) == 0) {
2790 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
2791 NV_UNIQUE_NAME, KM_SLEEP) == 0);
2792 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
2793 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2794 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2795 spa_load_l2cache(spa);
2796 spa_config_exit(spa, SCL_ALL, FTAG);
2797 spa->spa_l2cache.sav_sync = B_TRUE;
2800 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
2801 spa->spa_meta_objset = dp->dp_meta_objset;
2804 * Create DDTs (dedup tables).
2808 spa_update_dspace(spa);
2810 tx = dmu_tx_create_assigned(dp, txg);
2813 * Create the pool config object.
2815 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
2816 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
2817 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
2819 if (zap_add(spa->spa_meta_objset,
2820 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
2821 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
2822 cmn_err(CE_PANIC, "failed to add pool config");
2825 if (zap_add(spa->spa_meta_objset,
2826 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
2827 sizeof (uint64_t), 1, &version, tx) != 0) {
2828 cmn_err(CE_PANIC, "failed to add pool version");
2831 /* Newly created pools with the right version are always deflated. */
2832 if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
2833 spa->spa_deflate = TRUE;
2834 if (zap_add(spa->spa_meta_objset,
2835 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
2836 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
2837 cmn_err(CE_PANIC, "failed to add deflate");
2842 * Create the deferred-free bpobj. Turn off compression
2843 * because sync-to-convergence takes longer if the blocksize
2846 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
2847 dmu_object_set_compress(spa->spa_meta_objset, obj,
2848 ZIO_COMPRESS_OFF, tx);
2849 if (zap_add(spa->spa_meta_objset,
2850 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
2851 sizeof (uint64_t), 1, &obj, tx) != 0) {
2852 cmn_err(CE_PANIC, "failed to add bpobj");
2854 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
2855 spa->spa_meta_objset, obj));
2858 * Create the pool's history object.
2860 if (version >= SPA_VERSION_ZPOOL_HISTORY)
2861 spa_history_create_obj(spa, tx);
2864 * Set pool properties.
2866 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
2867 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2868 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
2869 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
2871 if (props != NULL) {
2872 spa_configfile_set(spa, props, B_FALSE);
2873 spa_sync_props(spa, props, tx);
2878 spa->spa_sync_on = B_TRUE;
2879 txg_sync_start(spa->spa_dsl_pool);
2882 * We explicitly wait for the first transaction to complete so that our
2883 * bean counters are appropriately updated.
2885 txg_wait_synced(spa->spa_dsl_pool, txg);
2887 spa_config_sync(spa, B_FALSE, B_TRUE);
2889 if (version >= SPA_VERSION_ZPOOL_HISTORY && history_str != NULL)
2890 (void) spa_history_log(spa, history_str, LOG_CMD_POOL_CREATE);
2891 spa_history_log_version(spa, LOG_POOL_CREATE);
2893 spa->spa_minref = refcount_count(&spa->spa_refcount);
2895 mutex_exit(&spa_namespace_lock);
2902 * Get the root pool information from the root disk, then import the root pool
2903 * during the system boot up time.
2905 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
2908 spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
2911 nvlist_t *nvtop, *nvroot;
2914 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
2918 * Add this top-level vdev to the child array.
2920 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
2922 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
2924 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
2927 * Put this pool's top-level vdevs into a root vdev.
2929 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2930 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
2931 VDEV_TYPE_ROOT) == 0);
2932 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
2933 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
2934 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
2938 * Replace the existing vdev_tree with the new root vdev in
2939 * this pool's configuration (remove the old, add the new).
2941 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
2942 nvlist_free(nvroot);
2947 * Walk the vdev tree and see if we can find a device with "better"
2948 * configuration. A configuration is "better" if the label on that
2949 * device has a more recent txg.
2952 spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
2954 for (int c = 0; c < vd->vdev_children; c++)
2955 spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
2957 if (vd->vdev_ops->vdev_op_leaf) {
2961 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
2965 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
2969 * Do we have a better boot device?
2971 if (label_txg > *txg) {
2980 * Import a root pool.
2982 * For x86. devpath_list will consist of devid and/or physpath name of
2983 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
2984 * The GRUB "findroot" command will return the vdev we should boot.
2986 * For Sparc, devpath_list consists the physpath name of the booting device
2987 * no matter the rootpool is a single device pool or a mirrored pool.
2989 * "/pci@1f,0/ide@d/disk@0,0:a"
2992 spa_import_rootpool(char *devpath, char *devid)
2995 vdev_t *rvd, *bvd, *avd = NULL;
2996 nvlist_t *config, *nvtop;
3002 * Read the label from the boot device and generate a configuration.
3004 config = spa_generate_rootconf(devpath, devid, &guid);
3005 #if defined(_OBP) && defined(_KERNEL)
3006 if (config == NULL) {
3007 if (strstr(devpath, "/iscsi/ssd") != NULL) {
3009 get_iscsi_bootpath_phy(devpath);
3010 config = spa_generate_rootconf(devpath, devid, &guid);
3014 if (config == NULL) {
3015 cmn_err(CE_NOTE, "Can not read the pool label from '%s'",
3020 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3022 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3024 mutex_enter(&spa_namespace_lock);
3025 if ((spa = spa_lookup(pname)) != NULL) {
3027 * Remove the existing root pool from the namespace so that we
3028 * can replace it with the correct config we just read in.
3033 spa = spa_add(pname, config, NULL);
3034 spa->spa_is_root = B_TRUE;
3035 spa->spa_load_verbatim = B_TRUE;
3038 * Build up a vdev tree based on the boot device's label config.
3040 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3042 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3043 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3044 VDEV_ALLOC_ROOTPOOL);
3045 spa_config_exit(spa, SCL_ALL, FTAG);
3047 mutex_exit(&spa_namespace_lock);
3048 nvlist_free(config);
3049 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3055 * Get the boot vdev.
3057 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3058 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3059 (u_longlong_t)guid);
3065 * Determine if there is a better boot device.
3068 spa_alt_rootvdev(rvd, &avd, &txg);
3070 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3071 "try booting from '%s'", avd->vdev_path);
3077 * If the boot device is part of a spare vdev then ensure that
3078 * we're booting off the active spare.
3080 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3081 !bvd->vdev_isspare) {
3082 cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3083 "try booting from '%s'",
3084 bvd->vdev_parent->vdev_child[1]->vdev_path);
3090 spa_history_log_version(spa, LOG_POOL_IMPORT);
3092 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3094 spa_config_exit(spa, SCL_ALL, FTAG);
3095 mutex_exit(&spa_namespace_lock);
3097 nvlist_free(config);
3104 * Take a pool and insert it into the namespace as if it had been loaded at
3108 spa_import_verbatim(const char *pool, nvlist_t *config, nvlist_t *props)
3111 char *altroot = NULL;
3113 mutex_enter(&spa_namespace_lock);
3114 if (spa_lookup(pool) != NULL) {
3115 mutex_exit(&spa_namespace_lock);
3119 (void) nvlist_lookup_string(props,
3120 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3121 spa = spa_add(pool, config, altroot);
3123 spa->spa_load_verbatim = B_TRUE;
3126 spa_configfile_set(spa, props, B_FALSE);
3128 spa_config_sync(spa, B_FALSE, B_TRUE);
3130 mutex_exit(&spa_namespace_lock);
3131 spa_history_log_version(spa, LOG_POOL_IMPORT);
3137 * Import a non-root pool into the system.
3140 spa_import(const char *pool, nvlist_t *config, nvlist_t *props)
3143 char *altroot = NULL;
3144 spa_load_state_t state = SPA_LOAD_IMPORT;
3145 zpool_rewind_policy_t policy;
3148 nvlist_t **spares, **l2cache;
3149 uint_t nspares, nl2cache;
3152 * If a pool with this name exists, return failure.
3154 mutex_enter(&spa_namespace_lock);
3155 if (spa_lookup(pool) != NULL) {
3156 mutex_exit(&spa_namespace_lock);
3160 zpool_get_rewind_policy(config, &policy);
3161 if (policy.zrp_request & ZPOOL_DO_REWIND)
3162 state = SPA_LOAD_RECOVER;
3165 * Create and initialize the spa structure.
3167 (void) nvlist_lookup_string(props,
3168 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3169 spa = spa_add(pool, config, altroot);
3170 spa_activate(spa, spa_mode_global);
3173 * Don't start async tasks until we know everything is healthy.
3175 spa_async_suspend(spa);
3178 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
3179 * because the user-supplied config is actually the one to trust when
3182 if (state != SPA_LOAD_RECOVER)
3183 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3184 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
3185 policy.zrp_request);
3188 * Propagate anything learned about failing or best txgs
3191 spa_rewind_data_to_nvlist(spa, config);
3193 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3195 * Toss any existing sparelist, as it doesn't have any validity
3196 * anymore, and conflicts with spa_has_spare().
3198 if (spa->spa_spares.sav_config) {
3199 nvlist_free(spa->spa_spares.sav_config);
3200 spa->spa_spares.sav_config = NULL;
3201 spa_load_spares(spa);
3203 if (spa->spa_l2cache.sav_config) {
3204 nvlist_free(spa->spa_l2cache.sav_config);
3205 spa->spa_l2cache.sav_config = NULL;
3206 spa_load_l2cache(spa);
3209 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3212 error = spa_validate_aux(spa, nvroot, -1ULL,
3215 error = spa_validate_aux(spa, nvroot, -1ULL,
3216 VDEV_ALLOC_L2CACHE);
3217 spa_config_exit(spa, SCL_ALL, FTAG);
3220 spa_configfile_set(spa, props, B_FALSE);
3222 if (error != 0 || (props && spa_writeable(spa) &&
3223 (error = spa_prop_set(spa, props)))) {
3225 spa_deactivate(spa);
3227 mutex_exit(&spa_namespace_lock);
3232 * Override any spares and level 2 cache devices as specified by
3233 * the user, as these may have correct device names/devids, etc.
3235 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3236 &spares, &nspares) == 0) {
3237 if (spa->spa_spares.sav_config)
3238 VERIFY(nvlist_remove(spa->spa_spares.sav_config,
3239 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
3241 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
3242 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3243 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3244 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3245 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3246 spa_load_spares(spa);
3247 spa_config_exit(spa, SCL_ALL, FTAG);
3248 spa->spa_spares.sav_sync = B_TRUE;
3250 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3251 &l2cache, &nl2cache) == 0) {
3252 if (spa->spa_l2cache.sav_config)
3253 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
3254 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
3256 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3257 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3258 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3259 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3260 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3261 spa_load_l2cache(spa);
3262 spa_config_exit(spa, SCL_ALL, FTAG);
3263 spa->spa_l2cache.sav_sync = B_TRUE;
3267 * Check for any removed devices.
3269 if (spa->spa_autoreplace) {
3270 spa_aux_check_removed(&spa->spa_spares);
3271 spa_aux_check_removed(&spa->spa_l2cache);
3274 if (spa_writeable(spa)) {
3276 * Update the config cache to include the newly-imported pool.
3278 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3281 spa_async_resume(spa);
3284 * It's possible that the pool was expanded while it was exported.
3285 * We kick off an async task to handle this for us.
3287 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
3289 mutex_exit(&spa_namespace_lock);
3290 spa_history_log_version(spa, LOG_POOL_IMPORT);
3296 spa_tryimport(nvlist_t *tryconfig)
3298 nvlist_t *config = NULL;
3304 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
3307 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
3311 * Create and initialize the spa structure.
3313 mutex_enter(&spa_namespace_lock);
3314 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
3315 spa_activate(spa, FREAD);
3318 * Pass off the heavy lifting to spa_load().
3319 * Pass TRUE for mosconfig because the user-supplied config
3320 * is actually the one to trust when doing an import.
3322 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
3325 * If 'tryconfig' was at least parsable, return the current config.
3327 if (spa->spa_root_vdev != NULL) {
3328 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
3329 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
3331 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
3333 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
3334 spa->spa_uberblock.ub_timestamp) == 0);
3337 * If the bootfs property exists on this pool then we
3338 * copy it out so that external consumers can tell which
3339 * pools are bootable.
3341 if ((!error || error == EEXIST) && spa->spa_bootfs) {
3342 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3345 * We have to play games with the name since the
3346 * pool was opened as TRYIMPORT_NAME.
3348 if (dsl_dsobj_to_dsname(spa_name(spa),
3349 spa->spa_bootfs, tmpname) == 0) {
3351 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3353 cp = strchr(tmpname, '/');
3355 (void) strlcpy(dsname, tmpname,
3358 (void) snprintf(dsname, MAXPATHLEN,
3359 "%s/%s", poolname, ++cp);
3361 VERIFY(nvlist_add_string(config,
3362 ZPOOL_CONFIG_BOOTFS, dsname) == 0);
3363 kmem_free(dsname, MAXPATHLEN);
3365 kmem_free(tmpname, MAXPATHLEN);
3369 * Add the list of hot spares and level 2 cache devices.
3371 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3372 spa_add_spares(spa, config);
3373 spa_add_l2cache(spa, config);
3374 spa_config_exit(spa, SCL_CONFIG, FTAG);
3378 spa_deactivate(spa);
3380 mutex_exit(&spa_namespace_lock);
3386 * Pool export/destroy
3388 * The act of destroying or exporting a pool is very simple. We make sure there
3389 * is no more pending I/O and any references to the pool are gone. Then, we
3390 * update the pool state and sync all the labels to disk, removing the
3391 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
3392 * we don't sync the labels or remove the configuration cache.
3395 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
3396 boolean_t force, boolean_t hardforce)
3403 if (!(spa_mode_global & FWRITE))
3406 mutex_enter(&spa_namespace_lock);
3407 if ((spa = spa_lookup(pool)) == NULL) {
3408 mutex_exit(&spa_namespace_lock);
3413 * Put a hold on the pool, drop the namespace lock, stop async tasks,
3414 * reacquire the namespace lock, and see if we can export.
3416 spa_open_ref(spa, FTAG);
3417 mutex_exit(&spa_namespace_lock);
3418 spa_async_suspend(spa);
3419 mutex_enter(&spa_namespace_lock);
3420 spa_close(spa, FTAG);
3423 * The pool will be in core if it's openable,
3424 * in which case we can modify its state.
3426 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
3428 * Objsets may be open only because they're dirty, so we
3429 * have to force it to sync before checking spa_refcnt.
3431 txg_wait_synced(spa->spa_dsl_pool, 0);
3434 * A pool cannot be exported or destroyed if there are active
3435 * references. If we are resetting a pool, allow references by
3436 * fault injection handlers.
3438 if (!spa_refcount_zero(spa) ||
3439 (spa->spa_inject_ref != 0 &&
3440 new_state != POOL_STATE_UNINITIALIZED)) {
3441 spa_async_resume(spa);
3442 mutex_exit(&spa_namespace_lock);
3447 * A pool cannot be exported if it has an active shared spare.
3448 * This is to prevent other pools stealing the active spare
3449 * from an exported pool. At user's own will, such pool can
3450 * be forcedly exported.
3452 if (!force && new_state == POOL_STATE_EXPORTED &&
3453 spa_has_active_shared_spare(spa)) {
3454 spa_async_resume(spa);
3455 mutex_exit(&spa_namespace_lock);
3460 * We want this to be reflected on every label,
3461 * so mark them all dirty. spa_unload() will do the
3462 * final sync that pushes these changes out.
3464 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
3465 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3466 spa->spa_state = new_state;
3467 spa->spa_final_txg = spa_last_synced_txg(spa) +
3469 vdev_config_dirty(spa->spa_root_vdev);
3470 spa_config_exit(spa, SCL_ALL, FTAG);
3474 spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
3476 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
3478 spa_deactivate(spa);
3481 if (oldconfig && spa->spa_config)
3482 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
3484 if (new_state != POOL_STATE_UNINITIALIZED) {
3486 spa_config_sync(spa, B_TRUE, B_TRUE);
3489 mutex_exit(&spa_namespace_lock);
3495 * Destroy a storage pool.
3498 spa_destroy(char *pool)
3500 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
3505 * Export a storage pool.
3508 spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
3509 boolean_t hardforce)
3511 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
3516 * Similar to spa_export(), this unloads the spa_t without actually removing it
3517 * from the namespace in any way.
3520 spa_reset(char *pool)
3522 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
3527 * ==========================================================================
3528 * Device manipulation
3529 * ==========================================================================
3533 * Add a device to a storage pool.
3536 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
3540 vdev_t *rvd = spa->spa_root_vdev;
3542 nvlist_t **spares, **l2cache;
3543 uint_t nspares, nl2cache;
3545 txg = spa_vdev_enter(spa);
3547 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
3548 VDEV_ALLOC_ADD)) != 0)
3549 return (spa_vdev_exit(spa, NULL, txg, error));
3551 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */
3553 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
3557 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
3561 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
3562 return (spa_vdev_exit(spa, vd, txg, EINVAL));
3564 if (vd->vdev_children != 0 &&
3565 (error = vdev_create(vd, txg, B_FALSE)) != 0)
3566 return (spa_vdev_exit(spa, vd, txg, error));
3569 * We must validate the spares and l2cache devices after checking the
3570 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
3572 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
3573 return (spa_vdev_exit(spa, vd, txg, error));
3576 * Transfer each new top-level vdev from vd to rvd.
3578 for (int c = 0; c < vd->vdev_children; c++) {
3581 * Set the vdev id to the first hole, if one exists.
3583 for (id = 0; id < rvd->vdev_children; id++) {
3584 if (rvd->vdev_child[id]->vdev_ishole) {
3585 vdev_free(rvd->vdev_child[id]);
3589 tvd = vd->vdev_child[c];
3590 vdev_remove_child(vd, tvd);
3592 vdev_add_child(rvd, tvd);
3593 vdev_config_dirty(tvd);
3597 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
3598 ZPOOL_CONFIG_SPARES);
3599 spa_load_spares(spa);
3600 spa->spa_spares.sav_sync = B_TRUE;
3603 if (nl2cache != 0) {
3604 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
3605 ZPOOL_CONFIG_L2CACHE);
3606 spa_load_l2cache(spa);
3607 spa->spa_l2cache.sav_sync = B_TRUE;
3611 * We have to be careful when adding new vdevs to an existing pool.
3612 * If other threads start allocating from these vdevs before we
3613 * sync the config cache, and we lose power, then upon reboot we may
3614 * fail to open the pool because there are DVAs that the config cache
3615 * can't translate. Therefore, we first add the vdevs without
3616 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
3617 * and then let spa_config_update() initialize the new metaslabs.
3619 * spa_load() checks for added-but-not-initialized vdevs, so that
3620 * if we lose power at any point in this sequence, the remaining
3621 * steps will be completed the next time we load the pool.
3623 (void) spa_vdev_exit(spa, vd, txg, 0);
3625 mutex_enter(&spa_namespace_lock);
3626 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3627 mutex_exit(&spa_namespace_lock);
3633 * Attach a device to a mirror. The arguments are the path to any device
3634 * in the mirror, and the nvroot for the new device. If the path specifies
3635 * a device that is not mirrored, we automatically insert the mirror vdev.
3637 * If 'replacing' is specified, the new device is intended to replace the
3638 * existing device; in this case the two devices are made into their own
3639 * mirror using the 'replacing' vdev, which is functionally identical to
3640 * the mirror vdev (it actually reuses all the same ops) but has a few
3641 * extra rules: you can't attach to it after it's been created, and upon
3642 * completion of resilvering, the first disk (the one being replaced)
3643 * is automatically detached.
3646 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
3648 uint64_t txg, dtl_max_txg;
3649 vdev_t *rvd = spa->spa_root_vdev;
3650 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
3652 char *oldvdpath, *newvdpath;
3656 txg = spa_vdev_enter(spa);
3658 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
3661 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
3663 if (!oldvd->vdev_ops->vdev_op_leaf)
3664 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3666 pvd = oldvd->vdev_parent;
3668 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
3669 VDEV_ALLOC_ADD)) != 0)
3670 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
3672 if (newrootvd->vdev_children != 1)
3673 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
3675 newvd = newrootvd->vdev_child[0];
3677 if (!newvd->vdev_ops->vdev_op_leaf)
3678 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
3680 if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
3681 return (spa_vdev_exit(spa, newrootvd, txg, error));
3684 * Spares can't replace logs
3686 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
3687 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3691 * For attach, the only allowable parent is a mirror or the root
3694 if (pvd->vdev_ops != &vdev_mirror_ops &&
3695 pvd->vdev_ops != &vdev_root_ops)
3696 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3698 pvops = &vdev_mirror_ops;
3701 * Active hot spares can only be replaced by inactive hot
3704 if (pvd->vdev_ops == &vdev_spare_ops &&
3705 pvd->vdev_child[1] == oldvd &&
3706 !spa_has_spare(spa, newvd->vdev_guid))
3707 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3710 * If the source is a hot spare, and the parent isn't already a
3711 * spare, then we want to create a new hot spare. Otherwise, we
3712 * want to create a replacing vdev. The user is not allowed to
3713 * attach to a spared vdev child unless the 'isspare' state is
3714 * the same (spare replaces spare, non-spare replaces
3717 if (pvd->vdev_ops == &vdev_replacing_ops)
3718 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3719 else if (pvd->vdev_ops == &vdev_spare_ops &&
3720 newvd->vdev_isspare != oldvd->vdev_isspare)
3721 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3722 else if (pvd->vdev_ops != &vdev_spare_ops &&
3723 newvd->vdev_isspare)
3724 pvops = &vdev_spare_ops;
3726 pvops = &vdev_replacing_ops;
3730 * Make sure the new device is big enough.
3732 if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
3733 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
3736 * The new device cannot have a higher alignment requirement
3737 * than the top-level vdev.
3739 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
3740 return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
3743 * If this is an in-place replacement, update oldvd's path and devid
3744 * to make it distinguishable from newvd, and unopenable from now on.
3746 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
3747 spa_strfree(oldvd->vdev_path);
3748 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
3750 (void) sprintf(oldvd->vdev_path, "%s/%s",
3751 newvd->vdev_path, "old");
3752 if (oldvd->vdev_devid != NULL) {
3753 spa_strfree(oldvd->vdev_devid);
3754 oldvd->vdev_devid = NULL;
3759 * If the parent is not a mirror, or if we're replacing, insert the new
3760 * mirror/replacing/spare vdev above oldvd.
3762 if (pvd->vdev_ops != pvops)
3763 pvd = vdev_add_parent(oldvd, pvops);
3765 ASSERT(pvd->vdev_top->vdev_parent == rvd);
3766 ASSERT(pvd->vdev_ops == pvops);
3767 ASSERT(oldvd->vdev_parent == pvd);
3770 * Extract the new device from its root and add it to pvd.
3772 vdev_remove_child(newrootvd, newvd);
3773 newvd->vdev_id = pvd->vdev_children;
3774 newvd->vdev_crtxg = oldvd->vdev_crtxg;
3775 vdev_add_child(pvd, newvd);
3777 tvd = newvd->vdev_top;
3778 ASSERT(pvd->vdev_top == tvd);
3779 ASSERT(tvd->vdev_parent == rvd);
3781 vdev_config_dirty(tvd);
3784 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
3785 * for any dmu_sync-ed blocks. It will propagate upward when
3786 * spa_vdev_exit() calls vdev_dtl_reassess().
3788 dtl_max_txg = txg + TXG_CONCURRENT_STATES;
3790 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
3791 dtl_max_txg - TXG_INITIAL);
3793 if (newvd->vdev_isspare) {
3794 spa_spare_activate(newvd);
3795 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
3798 oldvdpath = spa_strdup(oldvd->vdev_path);
3799 newvdpath = spa_strdup(newvd->vdev_path);
3800 newvd_isspare = newvd->vdev_isspare;
3803 * Mark newvd's DTL dirty in this txg.
3805 vdev_dirty(tvd, VDD_DTL, newvd, txg);
3808 * Restart the resilver
3810 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
3815 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
3817 spa_history_log_internal(LOG_POOL_VDEV_ATTACH, spa, NULL,
3818 "%s vdev=%s %s vdev=%s",
3819 replacing && newvd_isspare ? "spare in" :
3820 replacing ? "replace" : "attach", newvdpath,
3821 replacing ? "for" : "to", oldvdpath);
3823 spa_strfree(oldvdpath);
3824 spa_strfree(newvdpath);
3830 * Detach a device from a mirror or replacing vdev.
3831 * If 'replace_done' is specified, only detach if the parent
3832 * is a replacing vdev.
3835 spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
3839 vdev_t *rvd = spa->spa_root_vdev;
3840 vdev_t *vd, *pvd, *cvd, *tvd;
3841 boolean_t unspare = B_FALSE;
3842 uint64_t unspare_guid;
3846 txg = spa_vdev_enter(spa);
3848 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
3851 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
3853 if (!vd->vdev_ops->vdev_op_leaf)
3854 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3856 pvd = vd->vdev_parent;
3859 * If the parent/child relationship is not as expected, don't do it.
3860 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
3861 * vdev that's replacing B with C. The user's intent in replacing
3862 * is to go from M(A,B) to M(A,C). If the user decides to cancel
3863 * the replace by detaching C, the expected behavior is to end up
3864 * M(A,B). But suppose that right after deciding to detach C,
3865 * the replacement of B completes. We would have M(A,C), and then
3866 * ask to detach C, which would leave us with just A -- not what
3867 * the user wanted. To prevent this, we make sure that the
3868 * parent/child relationship hasn't changed -- in this example,
3869 * that C's parent is still the replacing vdev R.
3871 if (pvd->vdev_guid != pguid && pguid != 0)
3872 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
3875 * If replace_done is specified, only remove this device if it's
3876 * the first child of a replacing vdev. For the 'spare' vdev, either
3877 * disk can be removed.
3880 if (pvd->vdev_ops == &vdev_replacing_ops) {
3881 if (vd->vdev_id != 0)
3882 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3883 } else if (pvd->vdev_ops != &vdev_spare_ops) {
3884 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3888 ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
3889 spa_version(spa) >= SPA_VERSION_SPARES);
3892 * Only mirror, replacing, and spare vdevs support detach.
3894 if (pvd->vdev_ops != &vdev_replacing_ops &&
3895 pvd->vdev_ops != &vdev_mirror_ops &&
3896 pvd->vdev_ops != &vdev_spare_ops)
3897 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3900 * If this device has the only valid copy of some data,
3901 * we cannot safely detach it.
3903 if (vdev_dtl_required(vd))
3904 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
3906 ASSERT(pvd->vdev_children >= 2);
3909 * If we are detaching the second disk from a replacing vdev, then
3910 * check to see if we changed the original vdev's path to have "/old"
3911 * at the end in spa_vdev_attach(). If so, undo that change now.
3913 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id == 1 &&
3914 pvd->vdev_child[0]->vdev_path != NULL &&
3915 pvd->vdev_child[1]->vdev_path != NULL) {
3916 ASSERT(pvd->vdev_child[1] == vd);
3917 cvd = pvd->vdev_child[0];
3918 len = strlen(vd->vdev_path);
3919 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
3920 strcmp(cvd->vdev_path + len, "/old") == 0) {
3921 spa_strfree(cvd->vdev_path);
3922 cvd->vdev_path = spa_strdup(vd->vdev_path);
3927 * If we are detaching the original disk from a spare, then it implies
3928 * that the spare should become a real disk, and be removed from the
3929 * active spare list for the pool.
3931 if (pvd->vdev_ops == &vdev_spare_ops &&
3932 vd->vdev_id == 0 && pvd->vdev_child[1]->vdev_isspare)
3936 * Erase the disk labels so the disk can be used for other things.
3937 * This must be done after all other error cases are handled,
3938 * but before we disembowel vd (so we can still do I/O to it).
3939 * But if we can't do it, don't treat the error as fatal --
3940 * it may be that the unwritability of the disk is the reason
3941 * it's being detached!
3943 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
3946 * Remove vd from its parent and compact the parent's children.
3948 vdev_remove_child(pvd, vd);
3949 vdev_compact_children(pvd);
3952 * Remember one of the remaining children so we can get tvd below.
3954 cvd = pvd->vdev_child[0];
3957 * If we need to remove the remaining child from the list of hot spares,
3958 * do it now, marking the vdev as no longer a spare in the process.
3959 * We must do this before vdev_remove_parent(), because that can
3960 * change the GUID if it creates a new toplevel GUID. For a similar
3961 * reason, we must remove the spare now, in the same txg as the detach;
3962 * otherwise someone could attach a new sibling, change the GUID, and
3963 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
3966 ASSERT(cvd->vdev_isspare);
3967 spa_spare_remove(cvd);
3968 unspare_guid = cvd->vdev_guid;
3969 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
3973 * If the parent mirror/replacing vdev only has one child,
3974 * the parent is no longer needed. Remove it from the tree.
3976 if (pvd->vdev_children == 1)
3977 vdev_remove_parent(cvd);
3980 * We don't set tvd until now because the parent we just removed
3981 * may have been the previous top-level vdev.
3983 tvd = cvd->vdev_top;
3984 ASSERT(tvd->vdev_parent == rvd);
3987 * Reevaluate the parent vdev state.
3989 vdev_propagate_state(cvd);
3992 * If the 'autoexpand' property is set on the pool then automatically
3993 * try to expand the size of the pool. For example if the device we
3994 * just detached was smaller than the others, it may be possible to
3995 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
3996 * first so that we can obtain the updated sizes of the leaf vdevs.
3998 if (spa->spa_autoexpand) {
4000 vdev_expand(tvd, txg);
4003 vdev_config_dirty(tvd);
4006 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
4007 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4008 * But first make sure we're not on any *other* txg's DTL list, to
4009 * prevent vd from being accessed after it's freed.
4011 vdpath = spa_strdup(vd->vdev_path);
4012 for (int t = 0; t < TXG_SIZE; t++)
4013 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4014 vd->vdev_detached = B_TRUE;
4015 vdev_dirty(tvd, VDD_DTL, vd, txg);
4017 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
4019 error = spa_vdev_exit(spa, vd, txg, 0);
4021 spa_history_log_internal(LOG_POOL_VDEV_DETACH, spa, NULL,
4023 spa_strfree(vdpath);
4026 * If this was the removal of the original device in a hot spare vdev,
4027 * then we want to go through and remove the device from the hot spare
4028 * list of every other pool.
4033 mutex_enter(&spa_namespace_lock);
4034 while ((spa = spa_next(spa)) != NULL) {
4035 if (spa->spa_state != POOL_STATE_ACTIVE)
4039 spa_open_ref(spa, FTAG);
4040 mutex_exit(&spa_namespace_lock);
4041 (void) spa_vdev_remove(spa, unspare_guid,
4043 mutex_enter(&spa_namespace_lock);
4044 spa_close(spa, FTAG);
4046 mutex_exit(&spa_namespace_lock);
4053 * Split a set of devices from their mirrors, and create a new pool from them.
4056 spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
4057 nvlist_t *props, boolean_t exp)
4060 uint64_t txg, *glist;
4062 uint_t c, children, lastlog;
4063 nvlist_t **child, *nvl, *tmp;
4065 char *altroot = NULL;
4066 vdev_t *rvd, **vml = NULL; /* vdev modify list */
4067 boolean_t activate_slog;
4069 if (!spa_writeable(spa))
4072 txg = spa_vdev_enter(spa);
4074 /* clear the log and flush everything up to now */
4075 activate_slog = spa_passivate_log(spa);
4076 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4077 error = spa_offline_log(spa);
4078 txg = spa_vdev_config_enter(spa);
4081 spa_activate_log(spa);
4084 return (spa_vdev_exit(spa, NULL, txg, error));
4086 /* check new spa name before going any further */
4087 if (spa_lookup(newname) != NULL)
4088 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
4091 * scan through all the children to ensure they're all mirrors
4093 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
4094 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
4096 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4098 /* first, check to ensure we've got the right child count */
4099 rvd = spa->spa_root_vdev;
4101 for (c = 0; c < rvd->vdev_children; c++) {
4102 vdev_t *vd = rvd->vdev_child[c];
4104 /* don't count the holes & logs as children */
4105 if (vd->vdev_islog || vd->vdev_ishole) {
4113 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
4114 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4116 /* next, ensure no spare or cache devices are part of the split */
4117 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
4118 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
4119 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4121 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
4122 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
4124 /* then, loop over each vdev and validate it */
4125 for (c = 0; c < children; c++) {
4126 uint64_t is_hole = 0;
4128 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
4132 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
4133 spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
4141 /* which disk is going to be split? */
4142 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
4148 /* look it up in the spa */
4149 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
4150 if (vml[c] == NULL) {
4155 /* make sure there's nothing stopping the split */
4156 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
4157 vml[c]->vdev_islog ||
4158 vml[c]->vdev_ishole ||
4159 vml[c]->vdev_isspare ||
4160 vml[c]->vdev_isl2cache ||
4161 !vdev_writeable(vml[c]) ||
4162 vml[c]->vdev_children != 0 ||
4163 vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
4164 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
4169 if (vdev_dtl_required(vml[c])) {
4174 /* we need certain info from the top level */
4175 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
4176 vml[c]->vdev_top->vdev_ms_array) == 0);
4177 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
4178 vml[c]->vdev_top->vdev_ms_shift) == 0);
4179 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
4180 vml[c]->vdev_top->vdev_asize) == 0);
4181 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
4182 vml[c]->vdev_top->vdev_ashift) == 0);
4186 kmem_free(vml, children * sizeof (vdev_t *));
4187 kmem_free(glist, children * sizeof (uint64_t));
4188 return (spa_vdev_exit(spa, NULL, txg, error));
4191 /* stop writers from using the disks */
4192 for (c = 0; c < children; c++) {
4194 vml[c]->vdev_offline = B_TRUE;
4196 vdev_reopen(spa->spa_root_vdev);
4199 * Temporarily record the splitting vdevs in the spa config. This
4200 * will disappear once the config is regenerated.
4202 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4203 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
4204 glist, children) == 0);
4205 kmem_free(glist, children * sizeof (uint64_t));
4207 mutex_enter(&spa->spa_props_lock);
4208 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
4210 mutex_exit(&spa->spa_props_lock);
4211 spa->spa_config_splitting = nvl;
4212 vdev_config_dirty(spa->spa_root_vdev);
4214 /* configure and create the new pool */
4215 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
4216 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4217 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
4218 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
4219 spa_version(spa)) == 0);
4220 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
4221 spa->spa_config_txg) == 0);
4222 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4223 spa_generate_guid(NULL)) == 0);
4224 (void) nvlist_lookup_string(props,
4225 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4227 /* add the new pool to the namespace */
4228 newspa = spa_add(newname, config, altroot);
4229 newspa->spa_config_txg = spa->spa_config_txg;
4230 spa_set_log_state(newspa, SPA_LOG_CLEAR);
4232 /* release the spa config lock, retaining the namespace lock */
4233 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4235 if (zio_injection_enabled)
4236 zio_handle_panic_injection(spa, FTAG, 1);
4238 spa_activate(newspa, spa_mode_global);
4239 spa_async_suspend(newspa);
4241 /* create the new pool from the disks of the original pool */
4242 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
4246 /* if that worked, generate a real config for the new pool */
4247 if (newspa->spa_root_vdev != NULL) {
4248 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
4249 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4250 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
4251 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
4252 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
4257 if (props != NULL) {
4258 spa_configfile_set(newspa, props, B_FALSE);
4259 error = spa_prop_set(newspa, props);
4264 /* flush everything */
4265 txg = spa_vdev_config_enter(newspa);
4266 vdev_config_dirty(newspa->spa_root_vdev);
4267 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
4269 if (zio_injection_enabled)
4270 zio_handle_panic_injection(spa, FTAG, 2);
4272 spa_async_resume(newspa);
4274 /* finally, update the original pool's config */
4275 txg = spa_vdev_config_enter(spa);
4276 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
4277 error = dmu_tx_assign(tx, TXG_WAIT);
4280 for (c = 0; c < children; c++) {
4281 if (vml[c] != NULL) {
4284 spa_history_log_internal(LOG_POOL_VDEV_DETACH,
4290 vdev_config_dirty(spa->spa_root_vdev);
4291 spa->spa_config_splitting = NULL;
4295 (void) spa_vdev_exit(spa, NULL, txg, 0);
4297 if (zio_injection_enabled)
4298 zio_handle_panic_injection(spa, FTAG, 3);
4300 /* split is complete; log a history record */
4301 spa_history_log_internal(LOG_POOL_SPLIT, newspa, NULL,
4302 "split new pool %s from pool %s", newname, spa_name(spa));
4304 kmem_free(vml, children * sizeof (vdev_t *));
4306 /* if we're not going to mount the filesystems in userland, export */
4308 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
4315 spa_deactivate(newspa);
4318 txg = spa_vdev_config_enter(spa);
4320 /* re-online all offlined disks */
4321 for (c = 0; c < children; c++) {
4323 vml[c]->vdev_offline = B_FALSE;
4325 vdev_reopen(spa->spa_root_vdev);
4327 nvlist_free(spa->spa_config_splitting);
4328 spa->spa_config_splitting = NULL;
4329 (void) spa_vdev_exit(spa, NULL, txg, error);
4331 kmem_free(vml, children * sizeof (vdev_t *));
4336 spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
4338 for (int i = 0; i < count; i++) {
4341 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
4344 if (guid == target_guid)
4352 spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
4353 nvlist_t *dev_to_remove)
4355 nvlist_t **newdev = NULL;
4358 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
4360 for (int i = 0, j = 0; i < count; i++) {
4361 if (dev[i] == dev_to_remove)
4363 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
4366 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
4367 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
4369 for (int i = 0; i < count - 1; i++)
4370 nvlist_free(newdev[i]);
4373 kmem_free(newdev, (count - 1) * sizeof (void *));
4377 * Evacuate the device.
4380 spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
4385 ASSERT(MUTEX_HELD(&spa_namespace_lock));
4386 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
4387 ASSERT(vd == vd->vdev_top);
4390 * Evacuate the device. We don't hold the config lock as writer
4391 * since we need to do I/O but we do keep the
4392 * spa_namespace_lock held. Once this completes the device
4393 * should no longer have any blocks allocated on it.
4395 if (vd->vdev_islog) {
4396 if (vd->vdev_stat.vs_alloc != 0)
4397 error = spa_offline_log(spa);
4406 * The evacuation succeeded. Remove any remaining MOS metadata
4407 * associated with this vdev, and wait for these changes to sync.
4409 ASSERT3U(vd->vdev_stat.vs_alloc, ==, 0);
4410 txg = spa_vdev_config_enter(spa);
4411 vd->vdev_removing = B_TRUE;
4412 vdev_dirty(vd, 0, NULL, txg);
4413 vdev_config_dirty(vd);
4414 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4420 * Complete the removal by cleaning up the namespace.
4423 spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
4425 vdev_t *rvd = spa->spa_root_vdev;
4426 uint64_t id = vd->vdev_id;
4427 boolean_t last_vdev = (id == (rvd->vdev_children - 1));
4429 ASSERT(MUTEX_HELD(&spa_namespace_lock));
4430 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
4431 ASSERT(vd == vd->vdev_top);
4434 * Only remove any devices which are empty.
4436 if (vd->vdev_stat.vs_alloc != 0)
4439 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4441 if (list_link_active(&vd->vdev_state_dirty_node))
4442 vdev_state_clean(vd);
4443 if (list_link_active(&vd->vdev_config_dirty_node))
4444 vdev_config_clean(vd);
4449 vdev_compact_children(rvd);
4451 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
4452 vdev_add_child(rvd, vd);
4454 vdev_config_dirty(rvd);
4457 * Reassess the health of our root vdev.
4463 * Remove a device from the pool -
4465 * Removing a device from the vdev namespace requires several steps
4466 * and can take a significant amount of time. As a result we use
4467 * the spa_vdev_config_[enter/exit] functions which allow us to
4468 * grab and release the spa_config_lock while still holding the namespace
4469 * lock. During each step the configuration is synced out.
4473 * Remove a device from the pool. Currently, this supports removing only hot
4474 * spares, slogs, and level 2 ARC devices.
4477 spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
4480 metaslab_group_t *mg;
4481 nvlist_t **spares, **l2cache, *nv;
4483 uint_t nspares, nl2cache;
4485 boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
4488 txg = spa_vdev_enter(spa);
4490 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4492 if (spa->spa_spares.sav_vdevs != NULL &&
4493 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
4494 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
4495 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
4497 * Only remove the hot spare if it's not currently in use
4500 if (vd == NULL || unspare) {
4501 spa_vdev_remove_aux(spa->spa_spares.sav_config,
4502 ZPOOL_CONFIG_SPARES, spares, nspares, nv);
4503 spa_load_spares(spa);
4504 spa->spa_spares.sav_sync = B_TRUE;
4508 } else if (spa->spa_l2cache.sav_vdevs != NULL &&
4509 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
4510 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
4511 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
4513 * Cache devices can always be removed.
4515 spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
4516 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
4517 spa_load_l2cache(spa);
4518 spa->spa_l2cache.sav_sync = B_TRUE;
4519 } else if (vd != NULL && vd->vdev_islog) {
4521 ASSERT(vd == vd->vdev_top);
4524 * XXX - Once we have bp-rewrite this should
4525 * become the common case.
4531 * Stop allocating from this vdev.
4533 metaslab_group_passivate(mg);
4536 * Wait for the youngest allocations and frees to sync,
4537 * and then wait for the deferral of those frees to finish.
4539 spa_vdev_config_exit(spa, NULL,
4540 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
4543 * Attempt to evacuate the vdev.
4545 error = spa_vdev_remove_evacuate(spa, vd);
4547 txg = spa_vdev_config_enter(spa);
4550 * If we couldn't evacuate the vdev, unwind.
4553 metaslab_group_activate(mg);
4554 return (spa_vdev_exit(spa, NULL, txg, error));
4558 * Clean up the vdev namespace.
4560 spa_vdev_remove_from_namespace(spa, vd);
4562 } else if (vd != NULL) {
4564 * Normal vdevs cannot be removed (yet).
4569 * There is no vdev of any kind with the specified guid.
4575 return (spa_vdev_exit(spa, NULL, txg, error));
4581 * Find any device that's done replacing, or a vdev marked 'unspare' that's
4582 * current spared, so we can detach it.
4585 spa_vdev_resilver_done_hunt(vdev_t *vd)
4587 vdev_t *newvd, *oldvd;
4589 for (int c = 0; c < vd->vdev_children; c++) {
4590 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
4596 * Check for a completed replacement.
4598 if (vd->vdev_ops == &vdev_replacing_ops && vd->vdev_children == 2) {
4599 oldvd = vd->vdev_child[0];
4600 newvd = vd->vdev_child[1];
4602 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
4603 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
4604 !vdev_dtl_required(oldvd))
4609 * Check for a completed resilver with the 'unspare' flag set.
4611 if (vd->vdev_ops == &vdev_spare_ops && vd->vdev_children == 2) {
4612 newvd = vd->vdev_child[0];
4613 oldvd = vd->vdev_child[1];
4615 if (newvd->vdev_unspare &&
4616 vdev_dtl_empty(newvd, DTL_MISSING) &&
4617 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
4618 !vdev_dtl_required(oldvd)) {
4619 newvd->vdev_unspare = 0;
4628 spa_vdev_resilver_done(spa_t *spa)
4630 vdev_t *vd, *pvd, *ppvd;
4631 uint64_t guid, sguid, pguid, ppguid;
4633 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4635 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
4636 pvd = vd->vdev_parent;
4637 ppvd = pvd->vdev_parent;
4638 guid = vd->vdev_guid;
4639 pguid = pvd->vdev_guid;
4640 ppguid = ppvd->vdev_guid;
4643 * If we have just finished replacing a hot spared device, then
4644 * we need to detach the parent's first child (the original hot
4647 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0) {
4648 ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
4649 ASSERT(ppvd->vdev_children == 2);
4650 sguid = ppvd->vdev_child[1]->vdev_guid;
4652 spa_config_exit(spa, SCL_ALL, FTAG);
4653 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
4655 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
4657 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4660 spa_config_exit(spa, SCL_ALL, FTAG);
4664 * Update the stored path or FRU for this vdev.
4667 spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
4671 boolean_t sync = B_FALSE;
4673 spa_vdev_state_enter(spa, SCL_ALL);
4675 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
4676 return (spa_vdev_state_exit(spa, NULL, ENOENT));
4678 if (!vd->vdev_ops->vdev_op_leaf)
4679 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
4682 if (strcmp(value, vd->vdev_path) != 0) {
4683 spa_strfree(vd->vdev_path);
4684 vd->vdev_path = spa_strdup(value);
4688 if (vd->vdev_fru == NULL) {
4689 vd->vdev_fru = spa_strdup(value);
4691 } else if (strcmp(value, vd->vdev_fru) != 0) {
4692 spa_strfree(vd->vdev_fru);
4693 vd->vdev_fru = spa_strdup(value);
4698 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
4702 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
4704 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
4708 spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
4710 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
4714 * ==========================================================================
4716 * ==========================================================================
4720 spa_scan_stop(spa_t *spa)
4722 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
4723 if (dsl_scan_resilvering(spa->spa_dsl_pool))
4725 return (dsl_scan_cancel(spa->spa_dsl_pool));
4729 spa_scan(spa_t *spa, pool_scan_func_t func)
4731 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
4733 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
4737 * If a resilver was requested, but there is no DTL on a
4738 * writeable leaf device, we have nothing to do.
4740 if (func == POOL_SCAN_RESILVER &&
4741 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
4742 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
4746 return (dsl_scan(spa->spa_dsl_pool, func));
4750 * ==========================================================================
4751 * SPA async task processing
4752 * ==========================================================================
4756 spa_async_remove(spa_t *spa, vdev_t *vd)
4758 if (vd->vdev_remove_wanted) {
4759 vd->vdev_remove_wanted = B_FALSE;
4760 vd->vdev_delayed_close = B_FALSE;
4761 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
4764 * We want to clear the stats, but we don't want to do a full
4765 * vdev_clear() as that will cause us to throw away
4766 * degraded/faulted state as well as attempt to reopen the
4767 * device, all of which is a waste.
4769 vd->vdev_stat.vs_read_errors = 0;
4770 vd->vdev_stat.vs_write_errors = 0;
4771 vd->vdev_stat.vs_checksum_errors = 0;
4773 vdev_state_dirty(vd->vdev_top);
4776 for (int c = 0; c < vd->vdev_children; c++)
4777 spa_async_remove(spa, vd->vdev_child[c]);
4781 spa_async_probe(spa_t *spa, vdev_t *vd)
4783 if (vd->vdev_probe_wanted) {
4784 vd->vdev_probe_wanted = B_FALSE;
4785 vdev_reopen(vd); /* vdev_open() does the actual probe */
4788 for (int c = 0; c < vd->vdev_children; c++)
4789 spa_async_probe(spa, vd->vdev_child[c]);
4793 spa_async_autoexpand(spa_t *spa, vdev_t *vd)
4799 if (!spa->spa_autoexpand)
4802 for (int c = 0; c < vd->vdev_children; c++) {
4803 vdev_t *cvd = vd->vdev_child[c];
4804 spa_async_autoexpand(spa, cvd);
4807 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
4810 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
4811 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
4813 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4814 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
4816 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
4817 ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
4820 kmem_free(physpath, MAXPATHLEN);
4824 spa_async_thread(spa_t *spa)
4828 ASSERT(spa->spa_sync_on);
4830 mutex_enter(&spa->spa_async_lock);
4831 tasks = spa->spa_async_tasks;
4832 spa->spa_async_tasks = 0;
4833 mutex_exit(&spa->spa_async_lock);
4836 * See if the config needs to be updated.
4838 if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
4839 uint64_t old_space, new_space;
4841 mutex_enter(&spa_namespace_lock);
4842 old_space = metaslab_class_get_space(spa_normal_class(spa));
4843 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4844 new_space = metaslab_class_get_space(spa_normal_class(spa));
4845 mutex_exit(&spa_namespace_lock);
4848 * If the pool grew as a result of the config update,
4849 * then log an internal history event.
4851 if (new_space != old_space) {
4852 spa_history_log_internal(LOG_POOL_VDEV_ONLINE,
4854 "pool '%s' size: %llu(+%llu)",
4855 spa_name(spa), new_space, new_space - old_space);
4860 * See if any devices need to be marked REMOVED.
4862 if (tasks & SPA_ASYNC_REMOVE) {
4863 spa_vdev_state_enter(spa, SCL_NONE);
4864 spa_async_remove(spa, spa->spa_root_vdev);
4865 for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
4866 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
4867 for (int i = 0; i < spa->spa_spares.sav_count; i++)
4868 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
4869 (void) spa_vdev_state_exit(spa, NULL, 0);
4872 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
4873 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4874 spa_async_autoexpand(spa, spa->spa_root_vdev);
4875 spa_config_exit(spa, SCL_CONFIG, FTAG);
4879 * See if any devices need to be probed.
4881 if (tasks & SPA_ASYNC_PROBE) {
4882 spa_vdev_state_enter(spa, SCL_NONE);
4883 spa_async_probe(spa, spa->spa_root_vdev);
4884 (void) spa_vdev_state_exit(spa, NULL, 0);
4888 * If any devices are done replacing, detach them.
4890 if (tasks & SPA_ASYNC_RESILVER_DONE)
4891 spa_vdev_resilver_done(spa);
4894 * Kick off a resilver.
4896 if (tasks & SPA_ASYNC_RESILVER)
4897 dsl_resilver_restart(spa->spa_dsl_pool, 0);
4900 * Let the world know that we're done.
4902 mutex_enter(&spa->spa_async_lock);
4903 spa->spa_async_thread = NULL;
4904 cv_broadcast(&spa->spa_async_cv);
4905 mutex_exit(&spa->spa_async_lock);
4910 spa_async_suspend(spa_t *spa)
4912 mutex_enter(&spa->spa_async_lock);
4913 spa->spa_async_suspended++;
4914 while (spa->spa_async_thread != NULL)
4915 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
4916 mutex_exit(&spa->spa_async_lock);
4920 spa_async_resume(spa_t *spa)
4922 mutex_enter(&spa->spa_async_lock);
4923 ASSERT(spa->spa_async_suspended != 0);
4924 spa->spa_async_suspended--;
4925 mutex_exit(&spa->spa_async_lock);
4929 spa_async_dispatch(spa_t *spa)
4931 mutex_enter(&spa->spa_async_lock);
4932 if (spa->spa_async_tasks && !spa->spa_async_suspended &&
4933 spa->spa_async_thread == NULL &&
4934 rootdir != NULL && !vn_is_readonly(rootdir))
4935 spa->spa_async_thread = thread_create(NULL, 0,
4936 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
4937 mutex_exit(&spa->spa_async_lock);
4941 spa_async_request(spa_t *spa, int task)
4943 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
4944 mutex_enter(&spa->spa_async_lock);
4945 spa->spa_async_tasks |= task;
4946 mutex_exit(&spa->spa_async_lock);
4950 * ==========================================================================
4951 * SPA syncing routines
4952 * ==========================================================================
4956 bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
4959 bpobj_enqueue(bpo, bp, tx);
4964 spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
4968 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
4974 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
4976 char *packed = NULL;
4981 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
4984 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
4985 * information. This avoids the dbuf_will_dirty() path and
4986 * saves us a pre-read to get data we don't actually care about.
4988 bufsize = P2ROUNDUP(nvsize, SPA_CONFIG_BLOCKSIZE);
4989 packed = kmem_alloc(bufsize, KM_SLEEP);
4991 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
4993 bzero(packed + nvsize, bufsize - nvsize);
4995 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
4997 kmem_free(packed, bufsize);
4999 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
5000 dmu_buf_will_dirty(db, tx);
5001 *(uint64_t *)db->db_data = nvsize;
5002 dmu_buf_rele(db, FTAG);
5006 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
5007 const char *config, const char *entry)
5017 * Update the MOS nvlist describing the list of available devices.
5018 * spa_validate_aux() will have already made sure this nvlist is
5019 * valid and the vdevs are labeled appropriately.
5021 if (sav->sav_object == 0) {
5022 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
5023 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
5024 sizeof (uint64_t), tx);
5025 VERIFY(zap_update(spa->spa_meta_objset,
5026 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
5027 &sav->sav_object, tx) == 0);
5030 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5031 if (sav->sav_count == 0) {
5032 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
5034 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
5035 for (i = 0; i < sav->sav_count; i++)
5036 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
5037 B_FALSE, VDEV_CONFIG_L2CACHE);
5038 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
5039 sav->sav_count) == 0);
5040 for (i = 0; i < sav->sav_count; i++)
5041 nvlist_free(list[i]);
5042 kmem_free(list, sav->sav_count * sizeof (void *));
5045 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
5046 nvlist_free(nvroot);
5048 sav->sav_sync = B_FALSE;
5052 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
5056 if (list_is_empty(&spa->spa_config_dirty_list))
5059 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5061 config = spa_config_generate(spa, spa->spa_root_vdev,
5062 dmu_tx_get_txg(tx), B_FALSE);
5064 spa_config_exit(spa, SCL_STATE, FTAG);
5066 if (spa->spa_config_syncing)
5067 nvlist_free(spa->spa_config_syncing);
5068 spa->spa_config_syncing = config;
5070 spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
5074 * Set zpool properties.
5077 spa_sync_props(void *arg1, void *arg2, dmu_tx_t *tx)
5080 objset_t *mos = spa->spa_meta_objset;
5081 nvlist_t *nvp = arg2;
5086 const char *propname;
5087 zprop_type_t proptype;
5089 mutex_enter(&spa->spa_props_lock);
5092 while ((elem = nvlist_next_nvpair(nvp, elem))) {
5093 switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
5094 case ZPOOL_PROP_VERSION:
5096 * Only set version for non-zpool-creation cases
5097 * (set/import). spa_create() needs special care
5098 * for version setting.
5100 if (tx->tx_txg != TXG_INITIAL) {
5101 VERIFY(nvpair_value_uint64(elem,
5103 ASSERT(intval <= SPA_VERSION);
5104 ASSERT(intval >= spa_version(spa));
5105 spa->spa_uberblock.ub_version = intval;
5106 vdev_config_dirty(spa->spa_root_vdev);
5110 case ZPOOL_PROP_ALTROOT:
5112 * 'altroot' is a non-persistent property. It should
5113 * have been set temporarily at creation or import time.
5115 ASSERT(spa->spa_root != NULL);
5118 case ZPOOL_PROP_CACHEFILE:
5120 * 'cachefile' is also a non-persisitent property.
5125 * Set pool property values in the poolprops mos object.
5127 if (spa->spa_pool_props_object == 0) {
5128 VERIFY((spa->spa_pool_props_object =
5129 zap_create(mos, DMU_OT_POOL_PROPS,
5130 DMU_OT_NONE, 0, tx)) > 0);
5132 VERIFY(zap_update(mos,
5133 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
5134 8, 1, &spa->spa_pool_props_object, tx)
5138 /* normalize the property name */
5139 propname = zpool_prop_to_name(prop);
5140 proptype = zpool_prop_get_type(prop);
5142 if (nvpair_type(elem) == DATA_TYPE_STRING) {
5143 ASSERT(proptype == PROP_TYPE_STRING);
5144 VERIFY(nvpair_value_string(elem, &strval) == 0);
5145 VERIFY(zap_update(mos,
5146 spa->spa_pool_props_object, propname,
5147 1, strlen(strval) + 1, strval, tx) == 0);
5149 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
5150 VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5152 if (proptype == PROP_TYPE_INDEX) {
5154 VERIFY(zpool_prop_index_to_string(
5155 prop, intval, &unused) == 0);
5157 VERIFY(zap_update(mos,
5158 spa->spa_pool_props_object, propname,
5159 8, 1, &intval, tx) == 0);
5161 ASSERT(0); /* not allowed */
5165 case ZPOOL_PROP_DELEGATION:
5166 spa->spa_delegation = intval;
5168 case ZPOOL_PROP_BOOTFS:
5169 spa->spa_bootfs = intval;
5171 case ZPOOL_PROP_FAILUREMODE:
5172 spa->spa_failmode = intval;
5174 case ZPOOL_PROP_AUTOEXPAND:
5175 spa->spa_autoexpand = intval;
5176 if (tx->tx_txg != TXG_INITIAL)
5177 spa_async_request(spa,
5178 SPA_ASYNC_AUTOEXPAND);
5180 case ZPOOL_PROP_DEDUPDITTO:
5181 spa->spa_dedup_ditto = intval;
5188 /* log internal history if this is not a zpool create */
5189 if (spa_version(spa) >= SPA_VERSION_ZPOOL_HISTORY &&
5190 tx->tx_txg != TXG_INITIAL) {
5191 spa_history_log_internal(LOG_POOL_PROPSET,
5192 spa, tx, "%s %lld %s",
5193 nvpair_name(elem), intval, spa_name(spa));
5197 mutex_exit(&spa->spa_props_lock);
5201 * Perform one-time upgrade on-disk changes. spa_version() does not
5202 * reflect the new version this txg, so there must be no changes this
5203 * txg to anything that the upgrade code depends on after it executes.
5204 * Therefore this must be called after dsl_pool_sync() does the sync
5208 spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
5210 dsl_pool_t *dp = spa->spa_dsl_pool;
5212 ASSERT(spa->spa_sync_pass == 1);
5214 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
5215 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
5216 dsl_pool_create_origin(dp, tx);
5218 /* Keeping the origin open increases spa_minref */
5219 spa->spa_minref += 3;
5222 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
5223 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
5224 dsl_pool_upgrade_clones(dp, tx);
5227 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
5228 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
5229 dsl_pool_upgrade_dir_clones(dp, tx);
5231 /* Keeping the freedir open increases spa_minref */
5232 spa->spa_minref += 3;
5237 * Sync the specified transaction group. New blocks may be dirtied as
5238 * part of the process, so we iterate until it converges.
5241 spa_sync(spa_t *spa, uint64_t txg)
5243 dsl_pool_t *dp = spa->spa_dsl_pool;
5244 objset_t *mos = spa->spa_meta_objset;
5245 bpobj_t *defer_bpo = &spa->spa_deferred_bpobj;
5246 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
5247 vdev_t *rvd = spa->spa_root_vdev;
5253 * Lock out configuration changes.
5255 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5257 spa->spa_syncing_txg = txg;
5258 spa->spa_sync_pass = 0;
5261 * If there are any pending vdev state changes, convert them
5262 * into config changes that go out with this transaction group.
5264 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5265 while (list_head(&spa->spa_state_dirty_list) != NULL) {
5267 * We need the write lock here because, for aux vdevs,
5268 * calling vdev_config_dirty() modifies sav_config.
5269 * This is ugly and will become unnecessary when we
5270 * eliminate the aux vdev wart by integrating all vdevs
5271 * into the root vdev tree.
5273 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
5274 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
5275 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
5276 vdev_state_clean(vd);
5277 vdev_config_dirty(vd);
5279 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
5280 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
5282 spa_config_exit(spa, SCL_STATE, FTAG);
5284 tx = dmu_tx_create_assigned(dp, txg);
5287 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
5288 * set spa_deflate if we have no raid-z vdevs.
5290 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
5291 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
5294 for (i = 0; i < rvd->vdev_children; i++) {
5295 vd = rvd->vdev_child[i];
5296 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
5299 if (i == rvd->vdev_children) {
5300 spa->spa_deflate = TRUE;
5301 VERIFY(0 == zap_add(spa->spa_meta_objset,
5302 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
5303 sizeof (uint64_t), 1, &spa->spa_deflate, tx));
5308 * If anything has changed in this txg, or if someone is waiting
5309 * for this txg to sync (eg, spa_vdev_remove()), push the
5310 * deferred frees from the previous txg. If not, leave them
5311 * alone so that we don't generate work on an otherwise idle
5314 if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
5315 !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
5316 !txg_list_empty(&dp->dp_sync_tasks, txg) ||
5317 ((dsl_scan_active(dp->dp_scan) ||
5318 txg_sync_waiting(dp)) && !spa_shutting_down(spa))) {
5319 zio_t *zio = zio_root(spa, NULL, NULL, 0);
5320 VERIFY3U(bpobj_iterate(defer_bpo,
5321 spa_free_sync_cb, zio, tx), ==, 0);
5322 VERIFY3U(zio_wait(zio), ==, 0);
5326 * Iterate to convergence.
5329 int pass = ++spa->spa_sync_pass;
5331 spa_sync_config_object(spa, tx);
5332 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
5333 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
5334 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
5335 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
5336 spa_errlog_sync(spa, txg);
5337 dsl_pool_sync(dp, txg);
5339 if (pass <= SYNC_PASS_DEFERRED_FREE) {
5340 zio_t *zio = zio_root(spa, NULL, NULL, 0);
5341 bplist_iterate(free_bpl, spa_free_sync_cb,
5343 VERIFY(zio_wait(zio) == 0);
5345 bplist_iterate(free_bpl, bpobj_enqueue_cb,
5350 dsl_scan_sync(dp, tx);
5352 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
5356 spa_sync_upgrades(spa, tx);
5358 } while (dmu_objset_is_dirty(mos, txg));
5361 * Rewrite the vdev configuration (which includes the uberblock)
5362 * to commit the transaction group.
5364 * If there are no dirty vdevs, we sync the uberblock to a few
5365 * random top-level vdevs that are known to be visible in the
5366 * config cache (see spa_vdev_add() for a complete description).
5367 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
5371 * We hold SCL_STATE to prevent vdev open/close/etc.
5372 * while we're attempting to write the vdev labels.
5374 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5376 if (list_is_empty(&spa->spa_config_dirty_list)) {
5377 vdev_t *svd[SPA_DVAS_PER_BP];
5379 int children = rvd->vdev_children;
5380 int c0 = spa_get_random(children);
5382 for (int c = 0; c < children; c++) {
5383 vd = rvd->vdev_child[(c0 + c) % children];
5384 if (vd->vdev_ms_array == 0 || vd->vdev_islog)
5386 svd[svdcount++] = vd;
5387 if (svdcount == SPA_DVAS_PER_BP)
5390 error = vdev_config_sync(svd, svdcount, txg, B_FALSE);
5392 error = vdev_config_sync(svd, svdcount, txg,
5395 error = vdev_config_sync(rvd->vdev_child,
5396 rvd->vdev_children, txg, B_FALSE);
5398 error = vdev_config_sync(rvd->vdev_child,
5399 rvd->vdev_children, txg, B_TRUE);
5402 spa_config_exit(spa, SCL_STATE, FTAG);
5406 zio_suspend(spa, NULL);
5407 zio_resume_wait(spa);
5412 * Clear the dirty config list.
5414 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
5415 vdev_config_clean(vd);
5418 * Now that the new config has synced transactionally,
5419 * let it become visible to the config cache.
5421 if (spa->spa_config_syncing != NULL) {
5422 spa_config_set(spa, spa->spa_config_syncing);
5423 spa->spa_config_txg = txg;
5424 spa->spa_config_syncing = NULL;
5427 spa->spa_ubsync = spa->spa_uberblock;
5429 dsl_pool_sync_done(dp, txg);
5432 * Update usable space statistics.
5434 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
5435 vdev_sync_done(vd, txg);
5437 spa_update_dspace(spa);
5440 * It had better be the case that we didn't dirty anything
5441 * since vdev_config_sync().
5443 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
5444 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
5445 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
5447 spa->spa_sync_pass = 0;
5449 spa_config_exit(spa, SCL_CONFIG, FTAG);
5451 spa_handle_ignored_writes(spa);
5454 * If any async tasks have been requested, kick them off.
5456 spa_async_dispatch(spa);
5460 * Sync all pools. We don't want to hold the namespace lock across these
5461 * operations, so we take a reference on the spa_t and drop the lock during the
5465 spa_sync_allpools(void)
5468 mutex_enter(&spa_namespace_lock);
5469 while ((spa = spa_next(spa)) != NULL) {
5470 if (spa_state(spa) != POOL_STATE_ACTIVE || spa_suspended(spa))
5472 spa_open_ref(spa, FTAG);
5473 mutex_exit(&spa_namespace_lock);
5474 txg_wait_synced(spa_get_dsl(spa), 0);
5475 mutex_enter(&spa_namespace_lock);
5476 spa_close(spa, FTAG);
5478 mutex_exit(&spa_namespace_lock);
5482 * ==========================================================================
5483 * Miscellaneous routines
5484 * ==========================================================================
5488 * Remove all pools in the system.
5496 * Remove all cached state. All pools should be closed now,
5497 * so every spa in the AVL tree should be unreferenced.
5499 mutex_enter(&spa_namespace_lock);
5500 while ((spa = spa_next(NULL)) != NULL) {
5502 * Stop async tasks. The async thread may need to detach
5503 * a device that's been replaced, which requires grabbing
5504 * spa_namespace_lock, so we must drop it here.
5506 spa_open_ref(spa, FTAG);
5507 mutex_exit(&spa_namespace_lock);
5508 spa_async_suspend(spa);
5509 mutex_enter(&spa_namespace_lock);
5510 spa_close(spa, FTAG);
5512 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
5514 spa_deactivate(spa);
5518 mutex_exit(&spa_namespace_lock);
5522 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
5527 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
5531 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
5532 vd = spa->spa_l2cache.sav_vdevs[i];
5533 if (vd->vdev_guid == guid)
5537 for (i = 0; i < spa->spa_spares.sav_count; i++) {
5538 vd = spa->spa_spares.sav_vdevs[i];
5539 if (vd->vdev_guid == guid)
5548 spa_upgrade(spa_t *spa, uint64_t version)
5550 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5553 * This should only be called for a non-faulted pool, and since a
5554 * future version would result in an unopenable pool, this shouldn't be
5557 ASSERT(spa->spa_uberblock.ub_version <= SPA_VERSION);
5558 ASSERT(version >= spa->spa_uberblock.ub_version);
5560 spa->spa_uberblock.ub_version = version;
5561 vdev_config_dirty(spa->spa_root_vdev);
5563 spa_config_exit(spa, SCL_ALL, FTAG);
5565 txg_wait_synced(spa_get_dsl(spa), 0);
5569 spa_has_spare(spa_t *spa, uint64_t guid)
5573 spa_aux_vdev_t *sav = &spa->spa_spares;
5575 for (i = 0; i < sav->sav_count; i++)
5576 if (sav->sav_vdevs[i]->vdev_guid == guid)
5579 for (i = 0; i < sav->sav_npending; i++) {
5580 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
5581 &spareguid) == 0 && spareguid == guid)
5589 * Check if a pool has an active shared spare device.
5590 * Note: reference count of an active spare is 2, as a spare and as a replace
5593 spa_has_active_shared_spare(spa_t *spa)
5597 spa_aux_vdev_t *sav = &spa->spa_spares;
5599 for (i = 0; i < sav->sav_count; i++) {
5600 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
5601 &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
5610 * Post a sysevent corresponding to the given event. The 'name' must be one of
5611 * the event definitions in sys/sysevent/eventdefs.h. The payload will be
5612 * filled in from the spa and (optionally) the vdev. This doesn't do anything
5613 * in the userland libzpool, as we don't want consumers to misinterpret ztest
5614 * or zdb as real changes.
5617 spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
5621 sysevent_attr_list_t *attr = NULL;
5622 sysevent_value_t value;
5625 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
5628 value.value_type = SE_DATA_TYPE_STRING;
5629 value.value.sv_string = spa_name(spa);
5630 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
5633 value.value_type = SE_DATA_TYPE_UINT64;
5634 value.value.sv_uint64 = spa_guid(spa);
5635 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
5639 value.value_type = SE_DATA_TYPE_UINT64;
5640 value.value.sv_uint64 = vd->vdev_guid;
5641 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
5645 if (vd->vdev_path) {
5646 value.value_type = SE_DATA_TYPE_STRING;
5647 value.value.sv_string = vd->vdev_path;
5648 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
5649 &value, SE_SLEEP) != 0)
5654 if (sysevent_attach_attributes(ev, attr) != 0)
5658 (void) log_sysevent(ev, SE_SLEEP, &eid);
5662 sysevent_free_attr(attr);