4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
27 * This file contains all the routines used when modifying on-disk SPA state.
28 * This includes opening, importing, destroying, exporting a pool, and syncing a
32 #include <sys/zfs_context.h>
33 #include <sys/fm/fs/zfs.h>
34 #include <sys/spa_impl.h>
36 #include <sys/zio_checksum.h>
38 #include <sys/dmu_tx.h>
42 #include <sys/vdev_impl.h>
43 #include <sys/metaslab.h>
44 #include <sys/metaslab_impl.h>
45 #include <sys/uberblock_impl.h>
48 #include <sys/dmu_traverse.h>
49 #include <sys/dmu_objset.h>
50 #include <sys/unique.h>
51 #include <sys/dsl_pool.h>
52 #include <sys/dsl_dataset.h>
53 #include <sys/dsl_dir.h>
54 #include <sys/dsl_prop.h>
55 #include <sys/dsl_synctask.h>
56 #include <sys/fs/zfs.h>
58 #include <sys/callb.h>
59 #include <sys/systeminfo.h>
60 #include <sys/spa_boot.h>
61 #include <sys/zfs_ioctl.h>
62 #include <sys/dsl_scan.h>
65 #include <sys/bootprops.h>
66 #include <sys/callb.h>
67 #include <sys/cpupart.h>
69 #include <sys/sysdc.h>
74 #include "zfs_comutil.h"
76 typedef enum zti_modes {
77 zti_mode_fixed, /* value is # of threads (min 1) */
78 zti_mode_online_percent, /* value is % of online CPUs */
79 zti_mode_batch, /* cpu-intensive; value is ignored */
80 zti_mode_null, /* don't create a taskq */
84 #define ZTI_FIX(n) { zti_mode_fixed, (n) }
85 #define ZTI_PCT(n) { zti_mode_online_percent, (n) }
86 #define ZTI_BATCH { zti_mode_batch, 0 }
87 #define ZTI_NULL { zti_mode_null, 0 }
89 #define ZTI_ONE ZTI_FIX(1)
91 typedef struct zio_taskq_info {
92 enum zti_modes zti_mode;
96 static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
97 "issue", "issue_high", "intr", "intr_high"
101 * Define the taskq threads for the following I/O types:
102 * NULL, READ, WRITE, FREE, CLAIM, and IOCTL
104 const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
105 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
106 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL },
107 { ZTI_FIX(8), ZTI_NULL, ZTI_BATCH, ZTI_NULL },
108 { ZTI_BATCH, ZTI_FIX(5), ZTI_FIX(8), ZTI_FIX(5) },
109 { ZTI_FIX(100), ZTI_NULL, ZTI_ONE, ZTI_NULL },
110 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL },
111 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL },
114 static dsl_syncfunc_t spa_sync_props;
115 static boolean_t spa_has_active_shared_spare(spa_t *spa);
116 static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
117 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
119 static void spa_vdev_resilver_done(spa_t *spa);
121 uint_t zio_taskq_batch_pct = 100; /* 1 thread per cpu in pset */
122 id_t zio_taskq_psrset_bind = PS_NONE;
123 boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */
124 uint_t zio_taskq_basedc = 80; /* base duty cycle */
126 boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */
129 * This (illegal) pool name is used when temporarily importing a spa_t in order
130 * to get the vdev stats associated with the imported devices.
132 #define TRYIMPORT_NAME "$import"
135 * ==========================================================================
136 * SPA properties routines
137 * ==========================================================================
141 * Add a (source=src, propname=propval) list to an nvlist.
144 spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
145 uint64_t intval, zprop_source_t src)
147 const char *propname = zpool_prop_to_name(prop);
150 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
151 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
154 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
156 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
158 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
159 nvlist_free(propval);
163 * Get property values from the spa configuration.
166 spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
170 uint64_t cap, version;
171 zprop_source_t src = ZPROP_SRC_NONE;
172 spa_config_dirent_t *dp;
174 ASSERT(MUTEX_HELD(&spa->spa_props_lock));
176 if (spa->spa_root_vdev != NULL) {
177 alloc = metaslab_class_get_alloc(spa_normal_class(spa));
178 size = metaslab_class_get_space(spa_normal_class(spa));
179 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
180 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
181 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
182 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
184 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
185 (spa_mode(spa) == FREAD), src);
187 cap = (size == 0) ? 0 : (alloc * 100 / size);
188 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
190 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
191 ddt_get_pool_dedup_ratio(spa), src);
193 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
194 spa->spa_root_vdev->vdev_state, src);
196 version = spa_version(spa);
197 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
198 src = ZPROP_SRC_DEFAULT;
200 src = ZPROP_SRC_LOCAL;
201 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
204 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
206 if (spa->spa_root != NULL)
207 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
210 if ((dp = list_head(&spa->spa_config_list)) != NULL) {
211 if (dp->scd_path == NULL) {
212 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
213 "none", 0, ZPROP_SRC_LOCAL);
214 } else if (strcmp(dp->scd_path, spa_config_path) != 0) {
215 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
216 dp->scd_path, 0, ZPROP_SRC_LOCAL);
222 * Get zpool property values.
225 spa_prop_get(spa_t *spa, nvlist_t **nvp)
227 objset_t *mos = spa->spa_meta_objset;
232 VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
234 mutex_enter(&spa->spa_props_lock);
237 * Get properties from the spa config.
239 spa_prop_get_config(spa, nvp);
241 /* If no pool property object, no more prop to get. */
242 if (mos == NULL || spa->spa_pool_props_object == 0) {
243 mutex_exit(&spa->spa_props_lock);
248 * Get properties from the MOS pool property object.
250 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
251 (err = zap_cursor_retrieve(&zc, &za)) == 0;
252 zap_cursor_advance(&zc)) {
255 zprop_source_t src = ZPROP_SRC_DEFAULT;
258 if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
261 switch (za.za_integer_length) {
263 /* integer property */
264 if (za.za_first_integer !=
265 zpool_prop_default_numeric(prop))
266 src = ZPROP_SRC_LOCAL;
268 if (prop == ZPOOL_PROP_BOOTFS) {
270 dsl_dataset_t *ds = NULL;
272 dp = spa_get_dsl(spa);
273 rw_enter(&dp->dp_config_rwlock, RW_READER);
274 if ((err = dsl_dataset_hold_obj(dp,
275 za.za_first_integer, FTAG, &ds))) {
276 rw_exit(&dp->dp_config_rwlock);
281 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
283 dsl_dataset_name(ds, strval);
284 dsl_dataset_rele(ds, FTAG);
285 rw_exit(&dp->dp_config_rwlock);
288 intval = za.za_first_integer;
291 spa_prop_add_list(*nvp, prop, strval, intval, src);
295 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
300 /* string property */
301 strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
302 err = zap_lookup(mos, spa->spa_pool_props_object,
303 za.za_name, 1, za.za_num_integers, strval);
305 kmem_free(strval, za.za_num_integers);
308 spa_prop_add_list(*nvp, prop, strval, 0, src);
309 kmem_free(strval, za.za_num_integers);
316 zap_cursor_fini(&zc);
317 mutex_exit(&spa->spa_props_lock);
319 if (err && err != ENOENT) {
329 * Validate the given pool properties nvlist and modify the list
330 * for the property values to be set.
333 spa_prop_validate(spa_t *spa, nvlist_t *props)
336 int error = 0, reset_bootfs = 0;
340 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
342 char *propname, *strval;
347 propname = nvpair_name(elem);
349 if ((prop = zpool_name_to_prop(propname)) == ZPROP_INVAL)
353 case ZPOOL_PROP_VERSION:
354 error = nvpair_value_uint64(elem, &intval);
356 (intval < spa_version(spa) || intval > SPA_VERSION))
360 case ZPOOL_PROP_DELEGATION:
361 case ZPOOL_PROP_AUTOREPLACE:
362 case ZPOOL_PROP_LISTSNAPS:
363 case ZPOOL_PROP_AUTOEXPAND:
364 error = nvpair_value_uint64(elem, &intval);
365 if (!error && intval > 1)
369 case ZPOOL_PROP_BOOTFS:
371 * If the pool version is less than SPA_VERSION_BOOTFS,
372 * or the pool is still being created (version == 0),
373 * the bootfs property cannot be set.
375 if (spa_version(spa) < SPA_VERSION_BOOTFS) {
381 * Make sure the vdev config is bootable
383 if (!vdev_is_bootable(spa->spa_root_vdev)) {
390 error = nvpair_value_string(elem, &strval);
395 if (strval == NULL || strval[0] == '\0') {
396 objnum = zpool_prop_default_numeric(
401 if ((error = dmu_objset_hold(strval,FTAG,&os)))
404 /* Must be ZPL and not gzip compressed. */
406 if (dmu_objset_type(os) != DMU_OST_ZFS) {
408 } else if ((error = dsl_prop_get_integer(strval,
409 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
410 &compress, NULL)) == 0 &&
411 !BOOTFS_COMPRESS_VALID(compress)) {
414 objnum = dmu_objset_id(os);
416 dmu_objset_rele(os, FTAG);
420 case ZPOOL_PROP_FAILUREMODE:
421 error = nvpair_value_uint64(elem, &intval);
422 if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
423 intval > ZIO_FAILURE_MODE_PANIC))
427 * This is a special case which only occurs when
428 * the pool has completely failed. This allows
429 * the user to change the in-core failmode property
430 * without syncing it out to disk (I/Os might
431 * currently be blocked). We do this by returning
432 * EIO to the caller (spa_prop_set) to trick it
433 * into thinking we encountered a property validation
436 if (!error && spa_suspended(spa)) {
437 spa->spa_failmode = intval;
442 case ZPOOL_PROP_CACHEFILE:
443 if ((error = nvpair_value_string(elem, &strval)) != 0)
446 if (strval[0] == '\0')
449 if (strcmp(strval, "none") == 0)
452 if (strval[0] != '/') {
457 slash = strrchr(strval, '/');
458 ASSERT(slash != NULL);
460 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
461 strcmp(slash, "/..") == 0)
465 case ZPOOL_PROP_DEDUPDITTO:
466 if (spa_version(spa) < SPA_VERSION_DEDUP)
469 error = nvpair_value_uint64(elem, &intval);
471 intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
483 if (!error && reset_bootfs) {
484 error = nvlist_remove(props,
485 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
488 error = nvlist_add_uint64(props,
489 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
497 spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
500 spa_config_dirent_t *dp;
502 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
506 dp = kmem_alloc(sizeof (spa_config_dirent_t),
509 if (cachefile[0] == '\0')
510 dp->scd_path = spa_strdup(spa_config_path);
511 else if (strcmp(cachefile, "none") == 0)
514 dp->scd_path = spa_strdup(cachefile);
516 list_insert_head(&spa->spa_config_list, dp);
518 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
522 spa_prop_set(spa_t *spa, nvlist_t *nvp)
526 boolean_t need_sync = B_FALSE;
529 if ((error = spa_prop_validate(spa, nvp)) != 0)
533 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
534 if ((prop = zpool_name_to_prop(
535 nvpair_name(elem))) == ZPROP_INVAL)
538 if (prop == ZPOOL_PROP_CACHEFILE ||
539 prop == ZPOOL_PROP_ALTROOT ||
540 prop == ZPOOL_PROP_READONLY)
548 return (dsl_sync_task_do(spa_get_dsl(spa), NULL, spa_sync_props,
555 * If the bootfs property value is dsobj, clear it.
558 spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
560 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
561 VERIFY(zap_remove(spa->spa_meta_objset,
562 spa->spa_pool_props_object,
563 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
569 * ==========================================================================
570 * SPA state manipulation (open/create/destroy/import/export)
571 * ==========================================================================
575 spa_error_entry_compare(const void *a, const void *b)
577 spa_error_entry_t *sa = (spa_error_entry_t *)a;
578 spa_error_entry_t *sb = (spa_error_entry_t *)b;
581 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
582 sizeof (zbookmark_t));
593 * Utility function which retrieves copies of the current logs and
594 * re-initializes them in the process.
597 spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
599 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
601 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
602 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
604 avl_create(&spa->spa_errlist_scrub,
605 spa_error_entry_compare, sizeof (spa_error_entry_t),
606 offsetof(spa_error_entry_t, se_avl));
607 avl_create(&spa->spa_errlist_last,
608 spa_error_entry_compare, sizeof (spa_error_entry_t),
609 offsetof(spa_error_entry_t, se_avl));
613 spa_taskq_create(spa_t *spa, const char *name, enum zti_modes mode,
616 uint_t flags = TASKQ_PREPOPULATE;
617 boolean_t batch = B_FALSE;
621 return (NULL); /* no taskq needed */
624 ASSERT3U(value, >=, 1);
625 value = MAX(value, 1);
630 flags |= TASKQ_THREADS_CPU_PCT;
631 value = zio_taskq_batch_pct;
634 case zti_mode_online_percent:
635 flags |= TASKQ_THREADS_CPU_PCT;
639 panic("unrecognized mode for %s taskq (%u:%u) in "
645 if (zio_taskq_sysdc && spa->spa_proc != &p0) {
647 flags |= TASKQ_DC_BATCH;
649 return (taskq_create_sysdc(name, value, 50, INT_MAX,
650 spa->spa_proc, zio_taskq_basedc, flags));
652 return (taskq_create_proc(name, value, maxclsyspri, 50, INT_MAX,
653 spa->spa_proc, flags));
657 spa_create_zio_taskqs(spa_t *spa)
661 for (t = 0; t < ZIO_TYPES; t++) {
662 for (q = 0; q < ZIO_TASKQ_TYPES; q++) {
663 const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
664 enum zti_modes mode = ztip->zti_mode;
665 uint_t value = ztip->zti_value;
668 (void) snprintf(name, sizeof (name),
669 "%s_%s", zio_type_name[t], zio_taskq_types[q]);
671 spa->spa_zio_taskq[t][q] =
672 spa_taskq_create(spa, name, mode, value);
679 spa_thread(void *arg)
684 user_t *pu = PTOU(curproc);
686 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
689 ASSERT(curproc != &p0);
690 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
691 "zpool-%s", spa->spa_name);
692 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
694 /* bind this thread to the requested psrset */
695 if (zio_taskq_psrset_bind != PS_NONE) {
697 mutex_enter(&cpu_lock);
698 mutex_enter(&pidlock);
699 mutex_enter(&curproc->p_lock);
701 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
702 0, NULL, NULL) == 0) {
703 curthread->t_bind_pset = zio_taskq_psrset_bind;
706 "Couldn't bind process for zfs pool \"%s\" to "
707 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
710 mutex_exit(&curproc->p_lock);
711 mutex_exit(&pidlock);
712 mutex_exit(&cpu_lock);
716 if (zio_taskq_sysdc) {
717 sysdc_thread_enter(curthread, 100, 0);
720 spa->spa_proc = curproc;
721 spa->spa_did = curthread->t_did;
723 spa_create_zio_taskqs(spa);
725 mutex_enter(&spa->spa_proc_lock);
726 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
728 spa->spa_proc_state = SPA_PROC_ACTIVE;
729 cv_broadcast(&spa->spa_proc_cv);
731 CALLB_CPR_SAFE_BEGIN(&cprinfo);
732 while (spa->spa_proc_state == SPA_PROC_ACTIVE)
733 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
734 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
736 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
737 spa->spa_proc_state = SPA_PROC_GONE;
739 cv_broadcast(&spa->spa_proc_cv);
740 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */
742 mutex_enter(&curproc->p_lock);
748 * Activate an uninitialized pool.
751 spa_activate(spa_t *spa, int mode)
753 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
755 spa->spa_state = POOL_STATE_ACTIVE;
756 spa->spa_mode = mode;
758 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
759 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
761 /* Try to create a covering process */
762 mutex_enter(&spa->spa_proc_lock);
763 ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
764 ASSERT(spa->spa_proc == &p0);
767 /* Only create a process if we're going to be around a while. */
768 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
769 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
771 spa->spa_proc_state = SPA_PROC_CREATED;
772 while (spa->spa_proc_state == SPA_PROC_CREATED) {
773 cv_wait(&spa->spa_proc_cv,
774 &spa->spa_proc_lock);
776 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
777 ASSERT(spa->spa_proc != &p0);
778 ASSERT(spa->spa_did != 0);
782 "Couldn't create process for zfs pool \"%s\"\n",
787 mutex_exit(&spa->spa_proc_lock);
789 /* If we didn't create a process, we need to create our taskqs. */
790 if (spa->spa_proc == &p0) {
791 spa_create_zio_taskqs(spa);
794 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
795 offsetof(vdev_t, vdev_config_dirty_node));
796 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
797 offsetof(vdev_t, vdev_state_dirty_node));
799 txg_list_create(&spa->spa_vdev_txg_list,
800 offsetof(struct vdev, vdev_txg_node));
802 avl_create(&spa->spa_errlist_scrub,
803 spa_error_entry_compare, sizeof (spa_error_entry_t),
804 offsetof(spa_error_entry_t, se_avl));
805 avl_create(&spa->spa_errlist_last,
806 spa_error_entry_compare, sizeof (spa_error_entry_t),
807 offsetof(spa_error_entry_t, se_avl));
811 * Opposite of spa_activate().
814 spa_deactivate(spa_t *spa)
818 ASSERT(spa->spa_sync_on == B_FALSE);
819 ASSERT(spa->spa_dsl_pool == NULL);
820 ASSERT(spa->spa_root_vdev == NULL);
821 ASSERT(spa->spa_async_zio_root == NULL);
822 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
824 txg_list_destroy(&spa->spa_vdev_txg_list);
826 list_destroy(&spa->spa_config_dirty_list);
827 list_destroy(&spa->spa_state_dirty_list);
829 for (t = 0; t < ZIO_TYPES; t++) {
830 for (q = 0; q < ZIO_TASKQ_TYPES; q++) {
831 if (spa->spa_zio_taskq[t][q] != NULL)
832 taskq_destroy(spa->spa_zio_taskq[t][q]);
833 spa->spa_zio_taskq[t][q] = NULL;
837 metaslab_class_destroy(spa->spa_normal_class);
838 spa->spa_normal_class = NULL;
840 metaslab_class_destroy(spa->spa_log_class);
841 spa->spa_log_class = NULL;
844 * If this was part of an import or the open otherwise failed, we may
845 * still have errors left in the queues. Empty them just in case.
847 spa_errlog_drain(spa);
849 avl_destroy(&spa->spa_errlist_scrub);
850 avl_destroy(&spa->spa_errlist_last);
852 spa->spa_state = POOL_STATE_UNINITIALIZED;
854 mutex_enter(&spa->spa_proc_lock);
855 if (spa->spa_proc_state != SPA_PROC_NONE) {
856 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
857 spa->spa_proc_state = SPA_PROC_DEACTIVATE;
858 cv_broadcast(&spa->spa_proc_cv);
859 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
860 ASSERT(spa->spa_proc != &p0);
861 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
863 ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
864 spa->spa_proc_state = SPA_PROC_NONE;
866 ASSERT(spa->spa_proc == &p0);
867 mutex_exit(&spa->spa_proc_lock);
870 * We want to make sure spa_thread() has actually exited the ZFS
871 * module, so that the module can't be unloaded out from underneath
874 if (spa->spa_did != 0) {
875 thread_join(spa->spa_did);
881 * Verify a pool configuration, and construct the vdev tree appropriately. This
882 * will create all the necessary vdevs in the appropriate layout, with each vdev
883 * in the CLOSED state. This will prep the pool before open/creation/import.
884 * All vdev validation is done by the vdev_alloc() routine.
887 spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
888 uint_t id, int atype)
895 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
898 if ((*vdp)->vdev_ops->vdev_op_leaf)
901 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
913 for (c = 0; c < children; c++) {
915 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
923 ASSERT(*vdp != NULL);
929 * Opposite of spa_load().
932 spa_unload(spa_t *spa)
936 ASSERT(MUTEX_HELD(&spa_namespace_lock));
941 spa_async_suspend(spa);
946 if (spa->spa_sync_on) {
947 txg_sync_stop(spa->spa_dsl_pool);
948 spa->spa_sync_on = B_FALSE;
952 * Wait for any outstanding async I/O to complete.
954 if (spa->spa_async_zio_root != NULL) {
955 (void) zio_wait(spa->spa_async_zio_root);
956 spa->spa_async_zio_root = NULL;
959 bpobj_close(&spa->spa_deferred_bpobj);
962 * Close the dsl pool.
964 if (spa->spa_dsl_pool) {
965 dsl_pool_close(spa->spa_dsl_pool);
966 spa->spa_dsl_pool = NULL;
967 spa->spa_meta_objset = NULL;
972 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
975 * Drop and purge level 2 cache
977 spa_l2cache_drop(spa);
982 if (spa->spa_root_vdev)
983 vdev_free(spa->spa_root_vdev);
984 ASSERT(spa->spa_root_vdev == NULL);
986 for (i = 0; i < spa->spa_spares.sav_count; i++)
987 vdev_free(spa->spa_spares.sav_vdevs[i]);
988 if (spa->spa_spares.sav_vdevs) {
989 kmem_free(spa->spa_spares.sav_vdevs,
990 spa->spa_spares.sav_count * sizeof (void *));
991 spa->spa_spares.sav_vdevs = NULL;
993 if (spa->spa_spares.sav_config) {
994 nvlist_free(spa->spa_spares.sav_config);
995 spa->spa_spares.sav_config = NULL;
997 spa->spa_spares.sav_count = 0;
999 for (i = 0; i < spa->spa_l2cache.sav_count; i++)
1000 vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1001 if (spa->spa_l2cache.sav_vdevs) {
1002 kmem_free(spa->spa_l2cache.sav_vdevs,
1003 spa->spa_l2cache.sav_count * sizeof (void *));
1004 spa->spa_l2cache.sav_vdevs = NULL;
1006 if (spa->spa_l2cache.sav_config) {
1007 nvlist_free(spa->spa_l2cache.sav_config);
1008 spa->spa_l2cache.sav_config = NULL;
1010 spa->spa_l2cache.sav_count = 0;
1012 spa->spa_async_suspended = 0;
1014 spa_config_exit(spa, SCL_ALL, FTAG);
1018 * Load (or re-load) the current list of vdevs describing the active spares for
1019 * this pool. When this is called, we have some form of basic information in
1020 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1021 * then re-generate a more complete list including status information.
1024 spa_load_spares(spa_t *spa)
1031 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1034 * First, close and free any existing spare vdevs.
1036 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1037 vd = spa->spa_spares.sav_vdevs[i];
1039 /* Undo the call to spa_activate() below */
1040 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1041 B_FALSE)) != NULL && tvd->vdev_isspare)
1042 spa_spare_remove(tvd);
1047 if (spa->spa_spares.sav_vdevs)
1048 kmem_free(spa->spa_spares.sav_vdevs,
1049 spa->spa_spares.sav_count * sizeof (void *));
1051 if (spa->spa_spares.sav_config == NULL)
1054 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1055 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1057 spa->spa_spares.sav_count = (int)nspares;
1058 spa->spa_spares.sav_vdevs = NULL;
1064 * Construct the array of vdevs, opening them to get status in the
1065 * process. For each spare, there is potentially two different vdev_t
1066 * structures associated with it: one in the list of spares (used only
1067 * for basic validation purposes) and one in the active vdev
1068 * configuration (if it's spared in). During this phase we open and
1069 * validate each vdev on the spare list. If the vdev also exists in the
1070 * active configuration, then we also mark this vdev as an active spare.
1072 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1074 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1075 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1076 VDEV_ALLOC_SPARE) == 0);
1079 spa->spa_spares.sav_vdevs[i] = vd;
1081 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1082 B_FALSE)) != NULL) {
1083 if (!tvd->vdev_isspare)
1087 * We only mark the spare active if we were successfully
1088 * able to load the vdev. Otherwise, importing a pool
1089 * with a bad active spare would result in strange
1090 * behavior, because multiple pool would think the spare
1091 * is actively in use.
1093 * There is a vulnerability here to an equally bizarre
1094 * circumstance, where a dead active spare is later
1095 * brought back to life (onlined or otherwise). Given
1096 * the rarity of this scenario, and the extra complexity
1097 * it adds, we ignore the possibility.
1099 if (!vdev_is_dead(tvd))
1100 spa_spare_activate(tvd);
1104 vd->vdev_aux = &spa->spa_spares;
1106 if (vdev_open(vd) != 0)
1109 if (vdev_validate_aux(vd) == 0)
1114 * Recompute the stashed list of spares, with status information
1117 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1118 DATA_TYPE_NVLIST_ARRAY) == 0);
1120 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1122 for (i = 0; i < spa->spa_spares.sav_count; i++)
1123 spares[i] = vdev_config_generate(spa,
1124 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1125 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1126 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1127 for (i = 0; i < spa->spa_spares.sav_count; i++)
1128 nvlist_free(spares[i]);
1129 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1133 * Load (or re-load) the current list of vdevs describing the active l2cache for
1134 * this pool. When this is called, we have some form of basic information in
1135 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1136 * then re-generate a more complete list including status information.
1137 * Devices which are already active have their details maintained, and are
1141 spa_load_l2cache(spa_t *spa)
1145 int i, j, oldnvdevs;
1147 vdev_t *vd, **oldvdevs, **newvdevs = NULL;
1148 spa_aux_vdev_t *sav = &spa->spa_l2cache;
1150 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1152 if (sav->sav_config != NULL) {
1153 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1154 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1155 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1160 oldvdevs = sav->sav_vdevs;
1161 oldnvdevs = sav->sav_count;
1162 sav->sav_vdevs = NULL;
1166 * Process new nvlist of vdevs.
1168 for (i = 0; i < nl2cache; i++) {
1169 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1173 for (j = 0; j < oldnvdevs; j++) {
1175 if (vd != NULL && guid == vd->vdev_guid) {
1177 * Retain previous vdev for add/remove ops.
1185 if (newvdevs[i] == NULL) {
1189 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1190 VDEV_ALLOC_L2CACHE) == 0);
1195 * Commit this vdev as an l2cache device,
1196 * even if it fails to open.
1198 spa_l2cache_add(vd);
1203 spa_l2cache_activate(vd);
1205 if (vdev_open(vd) != 0)
1208 (void) vdev_validate_aux(vd);
1210 if (!vdev_is_dead(vd))
1211 l2arc_add_vdev(spa, vd);
1216 * Purge vdevs that were dropped
1218 for (i = 0; i < oldnvdevs; i++) {
1223 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1224 pool != 0ULL && l2arc_vdev_present(vd))
1225 l2arc_remove_vdev(vd);
1226 (void) vdev_close(vd);
1227 spa_l2cache_remove(vd);
1232 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1234 if (sav->sav_config == NULL)
1237 sav->sav_vdevs = newvdevs;
1238 sav->sav_count = (int)nl2cache;
1241 * Recompute the stashed list of l2cache devices, with status
1242 * information this time.
1244 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1245 DATA_TYPE_NVLIST_ARRAY) == 0);
1247 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1248 for (i = 0; i < sav->sav_count; i++)
1249 l2cache[i] = vdev_config_generate(spa,
1250 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1251 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1252 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1254 for (i = 0; i < sav->sav_count; i++)
1255 nvlist_free(l2cache[i]);
1257 kmem_free(l2cache, sav->sav_count * sizeof (void *));
1261 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1264 char *packed = NULL;
1269 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
1270 nvsize = *(uint64_t *)db->db_data;
1271 dmu_buf_rele(db, FTAG);
1273 packed = kmem_alloc(nvsize, KM_SLEEP);
1274 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1277 error = nvlist_unpack(packed, nvsize, value, 0);
1278 kmem_free(packed, nvsize);
1284 * Checks to see if the given vdev could not be opened, in which case we post a
1285 * sysevent to notify the autoreplace code that the device has been removed.
1288 spa_check_removed(vdev_t *vd)
1292 for (c = 0; c < vd->vdev_children; c++)
1293 spa_check_removed(vd->vdev_child[c]);
1295 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd)) {
1296 zfs_post_autoreplace(vd->vdev_spa, vd);
1297 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1302 * Validate the current config against the MOS config
1305 spa_config_valid(spa_t *spa, nvlist_t *config)
1307 vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1311 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1313 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1314 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1316 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1319 * If we're doing a normal import, then build up any additional
1320 * diagnostic information about missing devices in this config.
1321 * We'll pass this up to the user for further processing.
1323 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1324 nvlist_t **child, *nv;
1327 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1329 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1331 for (c = 0; c < rvd->vdev_children; c++) {
1332 vdev_t *tvd = rvd->vdev_child[c];
1333 vdev_t *mtvd = mrvd->vdev_child[c];
1335 if (tvd->vdev_ops == &vdev_missing_ops &&
1336 mtvd->vdev_ops != &vdev_missing_ops &&
1338 child[idx++] = vdev_config_generate(spa, mtvd,
1343 VERIFY(nvlist_add_nvlist_array(nv,
1344 ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1345 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1346 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1348 for (i = 0; i < idx; i++)
1349 nvlist_free(child[i]);
1352 kmem_free(child, rvd->vdev_children * sizeof (char **));
1356 * Compare the root vdev tree with the information we have
1357 * from the MOS config (mrvd). Check each top-level vdev
1358 * with the corresponding MOS config top-level (mtvd).
1360 for (c = 0; c < rvd->vdev_children; c++) {
1361 vdev_t *tvd = rvd->vdev_child[c];
1362 vdev_t *mtvd = mrvd->vdev_child[c];
1365 * Resolve any "missing" vdevs in the current configuration.
1366 * If we find that the MOS config has more accurate information
1367 * about the top-level vdev then use that vdev instead.
1369 if (tvd->vdev_ops == &vdev_missing_ops &&
1370 mtvd->vdev_ops != &vdev_missing_ops) {
1372 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1376 * Device specific actions.
1378 if (mtvd->vdev_islog) {
1379 spa_set_log_state(spa, SPA_LOG_CLEAR);
1382 * XXX - once we have 'readonly' pool
1383 * support we should be able to handle
1384 * missing data devices by transitioning
1385 * the pool to readonly.
1391 * Swap the missing vdev with the data we were
1392 * able to obtain from the MOS config.
1394 vdev_remove_child(rvd, tvd);
1395 vdev_remove_child(mrvd, mtvd);
1397 vdev_add_child(rvd, mtvd);
1398 vdev_add_child(mrvd, tvd);
1400 spa_config_exit(spa, SCL_ALL, FTAG);
1402 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1405 } else if (mtvd->vdev_islog) {
1407 * Load the slog device's state from the MOS config
1408 * since it's possible that the label does not
1409 * contain the most up-to-date information.
1411 vdev_load_log_state(tvd, mtvd);
1416 spa_config_exit(spa, SCL_ALL, FTAG);
1419 * Ensure we were able to validate the config.
1421 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1425 * Check for missing log devices
1428 spa_check_logs(spa_t *spa)
1430 switch (spa->spa_log_state) {
1433 case SPA_LOG_MISSING:
1434 /* need to recheck in case slog has been restored */
1435 case SPA_LOG_UNKNOWN:
1436 if (dmu_objset_find(spa->spa_name, zil_check_log_chain, NULL,
1437 DS_FIND_CHILDREN)) {
1438 spa_set_log_state(spa, SPA_LOG_MISSING);
1447 spa_passivate_log(spa_t *spa)
1449 vdev_t *rvd = spa->spa_root_vdev;
1450 boolean_t slog_found = B_FALSE;
1453 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1455 if (!spa_has_slogs(spa))
1458 for (c = 0; c < rvd->vdev_children; c++) {
1459 vdev_t *tvd = rvd->vdev_child[c];
1460 metaslab_group_t *mg = tvd->vdev_mg;
1462 if (tvd->vdev_islog) {
1463 metaslab_group_passivate(mg);
1464 slog_found = B_TRUE;
1468 return (slog_found);
1472 spa_activate_log(spa_t *spa)
1474 vdev_t *rvd = spa->spa_root_vdev;
1477 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1479 for (c = 0; c < rvd->vdev_children; c++) {
1480 vdev_t *tvd = rvd->vdev_child[c];
1481 metaslab_group_t *mg = tvd->vdev_mg;
1483 if (tvd->vdev_islog)
1484 metaslab_group_activate(mg);
1489 spa_offline_log(spa_t *spa)
1493 if ((error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1494 NULL, DS_FIND_CHILDREN)) == 0) {
1497 * We successfully offlined the log device, sync out the
1498 * current txg so that the "stubby" block can be removed
1501 txg_wait_synced(spa->spa_dsl_pool, 0);
1507 spa_aux_check_removed(spa_aux_vdev_t *sav)
1511 for (i = 0; i < sav->sav_count; i++)
1512 spa_check_removed(sav->sav_vdevs[i]);
1516 spa_claim_notify(zio_t *zio)
1518 spa_t *spa = zio->io_spa;
1523 mutex_enter(&spa->spa_props_lock); /* any mutex will do */
1524 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1525 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1526 mutex_exit(&spa->spa_props_lock);
1529 typedef struct spa_load_error {
1530 uint64_t sle_meta_count;
1531 uint64_t sle_data_count;
1535 spa_load_verify_done(zio_t *zio)
1537 blkptr_t *bp = zio->io_bp;
1538 spa_load_error_t *sle = zio->io_private;
1539 dmu_object_type_t type = BP_GET_TYPE(bp);
1540 int error = zio->io_error;
1543 if ((BP_GET_LEVEL(bp) != 0 || dmu_ot[type].ot_metadata) &&
1544 type != DMU_OT_INTENT_LOG)
1545 atomic_add_64(&sle->sle_meta_count, 1);
1547 atomic_add_64(&sle->sle_data_count, 1);
1549 zio_data_buf_free(zio->io_data, zio->io_size);
1554 spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1555 arc_buf_t *pbuf, const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1559 size_t size = BP_GET_PSIZE(bp);
1560 void *data = zio_data_buf_alloc(size);
1562 zio_nowait(zio_read(rio, spa, bp, data, size,
1563 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1564 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1565 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1571 spa_load_verify(spa_t *spa)
1574 spa_load_error_t sle = { 0 };
1575 zpool_rewind_policy_t policy;
1576 boolean_t verify_ok = B_FALSE;
1579 zpool_get_rewind_policy(spa->spa_config, &policy);
1581 if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1584 rio = zio_root(spa, NULL, &sle,
1585 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1587 error = traverse_pool(spa, spa->spa_verify_min_txg,
1588 TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio);
1590 (void) zio_wait(rio);
1592 spa->spa_load_meta_errors = sle.sle_meta_count;
1593 spa->spa_load_data_errors = sle.sle_data_count;
1595 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1596 sle.sle_data_count <= policy.zrp_maxdata) {
1600 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1601 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1603 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1604 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1605 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1606 VERIFY(nvlist_add_int64(spa->spa_load_info,
1607 ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1608 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1609 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1611 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1615 if (error != ENXIO && error != EIO)
1620 return (verify_ok ? 0 : EIO);
1624 * Find a value in the pool props object.
1627 spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1629 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1630 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1634 * Find a value in the pool directory object.
1637 spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1639 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1640 name, sizeof (uint64_t), 1, val));
1644 spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1646 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1651 * Fix up config after a partly-completed split. This is done with the
1652 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
1653 * pool have that entry in their config, but only the splitting one contains
1654 * a list of all the guids of the vdevs that are being split off.
1656 * This function determines what to do with that list: either rejoin
1657 * all the disks to the pool, or complete the splitting process. To attempt
1658 * the rejoin, each disk that is offlined is marked online again, and
1659 * we do a reopen() call. If the vdev label for every disk that was
1660 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1661 * then we call vdev_split() on each disk, and complete the split.
1663 * Otherwise we leave the config alone, with all the vdevs in place in
1664 * the original pool.
1667 spa_try_repair(spa_t *spa, nvlist_t *config)
1674 boolean_t attempt_reopen;
1676 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
1679 /* check that the config is complete */
1680 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
1681 &glist, &gcount) != 0)
1684 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
1686 /* attempt to online all the vdevs & validate */
1687 attempt_reopen = B_TRUE;
1688 for (i = 0; i < gcount; i++) {
1689 if (glist[i] == 0) /* vdev is hole */
1692 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
1693 if (vd[i] == NULL) {
1695 * Don't bother attempting to reopen the disks;
1696 * just do the split.
1698 attempt_reopen = B_FALSE;
1700 /* attempt to re-online it */
1701 vd[i]->vdev_offline = B_FALSE;
1705 if (attempt_reopen) {
1706 vdev_reopen(spa->spa_root_vdev);
1708 /* check each device to see what state it's in */
1709 for (extracted = 0, i = 0; i < gcount; i++) {
1710 if (vd[i] != NULL &&
1711 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
1718 * If every disk has been moved to the new pool, or if we never
1719 * even attempted to look at them, then we split them off for
1722 if (!attempt_reopen || gcount == extracted) {
1723 for (i = 0; i < gcount; i++)
1726 vdev_reopen(spa->spa_root_vdev);
1729 kmem_free(vd, gcount * sizeof (vdev_t *));
1733 spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
1734 boolean_t mosconfig)
1736 nvlist_t *config = spa->spa_config;
1737 char *ereport = FM_EREPORT_ZFS_POOL;
1742 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
1746 * Versioning wasn't explicitly added to the label until later, so if
1747 * it's not present treat it as the initial version.
1749 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
1750 &spa->spa_ubsync.ub_version) != 0)
1751 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
1753 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
1754 &spa->spa_config_txg);
1756 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
1757 spa_guid_exists(pool_guid, 0)) {
1760 spa->spa_load_guid = pool_guid;
1762 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
1764 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
1768 gethrestime(&spa->spa_loaded_ts);
1769 error = spa_load_impl(spa, pool_guid, config, state, type,
1770 mosconfig, &ereport);
1773 spa->spa_minref = refcount_count(&spa->spa_refcount);
1775 if (error != EEXIST) {
1776 spa->spa_loaded_ts.tv_sec = 0;
1777 spa->spa_loaded_ts.tv_nsec = 0;
1779 if (error != EBADF) {
1780 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
1783 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
1790 * Load an existing storage pool, using the pool's builtin spa_config as a
1791 * source of configuration information.
1794 spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
1795 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
1799 nvlist_t *nvroot = NULL;
1801 uberblock_t *ub = &spa->spa_uberblock;
1802 uint64_t children, config_cache_txg = spa->spa_config_txg;
1803 int orig_mode = spa->spa_mode;
1808 * If this is an untrusted config, access the pool in read-only mode.
1809 * This prevents things like resilvering recently removed devices.
1812 spa->spa_mode = FREAD;
1814 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1816 spa->spa_load_state = state;
1818 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
1821 parse = (type == SPA_IMPORT_EXISTING ?
1822 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
1825 * Create "The Godfather" zio to hold all async IOs
1827 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
1828 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
1831 * Parse the configuration into a vdev tree. We explicitly set the
1832 * value that will be returned by spa_version() since parsing the
1833 * configuration requires knowing the version number.
1835 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1836 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
1837 spa_config_exit(spa, SCL_ALL, FTAG);
1842 ASSERT(spa->spa_root_vdev == rvd);
1844 if (type != SPA_IMPORT_ASSEMBLE) {
1845 ASSERT(spa_guid(spa) == pool_guid);
1849 * Try to open all vdevs, loading each label in the process.
1851 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1852 error = vdev_open(rvd);
1853 spa_config_exit(spa, SCL_ALL, FTAG);
1858 * We need to validate the vdev labels against the configuration that
1859 * we have in hand, which is dependent on the setting of mosconfig. If
1860 * mosconfig is true then we're validating the vdev labels based on
1861 * that config. Otherwise, we're validating against the cached config
1862 * (zpool.cache) that was read when we loaded the zfs module, and then
1863 * later we will recursively call spa_load() and validate against
1866 * If we're assembling a new pool that's been split off from an
1867 * existing pool, the labels haven't yet been updated so we skip
1868 * validation for now.
1870 if (type != SPA_IMPORT_ASSEMBLE) {
1871 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1872 error = vdev_validate(rvd);
1873 spa_config_exit(spa, SCL_ALL, FTAG);
1878 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
1883 * Find the best uberblock.
1885 vdev_uberblock_load(NULL, rvd, ub);
1888 * If we weren't able to find a single valid uberblock, return failure.
1890 if (ub->ub_txg == 0)
1891 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
1894 * If the pool is newer than the code, we can't open it.
1896 if (ub->ub_version > SPA_VERSION)
1897 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
1900 * If the vdev guid sum doesn't match the uberblock, we have an
1901 * incomplete configuration. We first check to see if the pool
1902 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
1903 * If it is, defer the vdev_guid_sum check till later so we
1904 * can handle missing vdevs.
1906 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
1907 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
1908 rvd->vdev_guid_sum != ub->ub_guid_sum)
1909 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
1911 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
1912 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1913 spa_try_repair(spa, config);
1914 spa_config_exit(spa, SCL_ALL, FTAG);
1915 nvlist_free(spa->spa_config_splitting);
1916 spa->spa_config_splitting = NULL;
1920 * Initialize internal SPA structures.
1922 spa->spa_state = POOL_STATE_ACTIVE;
1923 spa->spa_ubsync = spa->spa_uberblock;
1924 spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
1925 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
1926 spa->spa_first_txg = spa->spa_last_ubsync_txg ?
1927 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
1928 spa->spa_claim_max_txg = spa->spa_first_txg;
1929 spa->spa_prev_software_version = ub->ub_software_version;
1931 error = dsl_pool_open(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
1933 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1934 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
1936 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
1937 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1941 nvlist_t *policy = NULL, *nvconfig;
1943 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
1944 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1946 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
1947 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
1949 unsigned long myhostid = 0;
1951 VERIFY(nvlist_lookup_string(nvconfig,
1952 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
1955 myhostid = zone_get_hostid(NULL);
1958 * We're emulating the system's hostid in userland, so
1959 * we can't use zone_get_hostid().
1961 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
1962 #endif /* _KERNEL */
1963 if (hostid != 0 && myhostid != 0 &&
1964 hostid != myhostid) {
1965 nvlist_free(nvconfig);
1966 cmn_err(CE_WARN, "pool '%s' could not be "
1967 "loaded as it was last accessed by "
1968 "another system (host: %s hostid: 0x%lx). "
1969 "See: http://www.sun.com/msg/ZFS-8000-EY",
1970 spa_name(spa), hostname,
1971 (unsigned long)hostid);
1975 if (nvlist_lookup_nvlist(spa->spa_config,
1976 ZPOOL_REWIND_POLICY, &policy) == 0)
1977 VERIFY(nvlist_add_nvlist(nvconfig,
1978 ZPOOL_REWIND_POLICY, policy) == 0);
1980 spa_config_set(spa, nvconfig);
1982 spa_deactivate(spa);
1983 spa_activate(spa, orig_mode);
1985 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
1988 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
1989 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1990 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
1992 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1995 * Load the bit that tells us to use the new accounting function
1996 * (raid-z deflation). If we have an older pool, this will not
1999 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2000 if (error != 0 && error != ENOENT)
2001 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2003 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2004 &spa->spa_creation_version);
2005 if (error != 0 && error != ENOENT)
2006 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2009 * Load the persistent error log. If we have an older pool, this will
2012 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2013 if (error != 0 && error != ENOENT)
2014 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2016 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2017 &spa->spa_errlog_scrub);
2018 if (error != 0 && error != ENOENT)
2019 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2022 * Load the history object. If we have an older pool, this
2023 * will not be present.
2025 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2026 if (error != 0 && error != ENOENT)
2027 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2030 * If we're assembling the pool from the split-off vdevs of
2031 * an existing pool, we don't want to attach the spares & cache
2036 * Load any hot spares for this pool.
2038 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2039 if (error != 0 && error != ENOENT)
2040 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2041 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2042 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2043 if (load_nvlist(spa, spa->spa_spares.sav_object,
2044 &spa->spa_spares.sav_config) != 0)
2045 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2047 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2048 spa_load_spares(spa);
2049 spa_config_exit(spa, SCL_ALL, FTAG);
2050 } else if (error == 0) {
2051 spa->spa_spares.sav_sync = B_TRUE;
2055 * Load any level 2 ARC devices for this pool.
2057 error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2058 &spa->spa_l2cache.sav_object);
2059 if (error != 0 && error != ENOENT)
2060 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2061 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2062 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2063 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2064 &spa->spa_l2cache.sav_config) != 0)
2065 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2067 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2068 spa_load_l2cache(spa);
2069 spa_config_exit(spa, SCL_ALL, FTAG);
2070 } else if (error == 0) {
2071 spa->spa_l2cache.sav_sync = B_TRUE;
2074 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2076 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2077 if (error && error != ENOENT)
2078 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2081 uint64_t autoreplace;
2083 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2084 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2085 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2086 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2087 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2088 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2089 &spa->spa_dedup_ditto);
2091 spa->spa_autoreplace = (autoreplace != 0);
2095 * If the 'autoreplace' property is set, then post a resource notifying
2096 * the ZFS DE that it should not issue any faults for unopenable
2097 * devices. We also iterate over the vdevs, and post a sysevent for any
2098 * unopenable vdevs so that the normal autoreplace handler can take
2101 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2102 spa_check_removed(spa->spa_root_vdev);
2104 * For the import case, this is done in spa_import(), because
2105 * at this point we're using the spare definitions from
2106 * the MOS config, not necessarily from the userland config.
2108 if (state != SPA_LOAD_IMPORT) {
2109 spa_aux_check_removed(&spa->spa_spares);
2110 spa_aux_check_removed(&spa->spa_l2cache);
2115 * Load the vdev state for all toplevel vdevs.
2120 * Propagate the leaf DTLs we just loaded all the way up the tree.
2122 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2123 vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2124 spa_config_exit(spa, SCL_ALL, FTAG);
2127 * Load the DDTs (dedup tables).
2129 error = ddt_load(spa);
2131 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2133 spa_update_dspace(spa);
2136 * Validate the config, using the MOS config to fill in any
2137 * information which might be missing. If we fail to validate
2138 * the config then declare the pool unfit for use. If we're
2139 * assembling a pool from a split, the log is not transferred
2142 if (type != SPA_IMPORT_ASSEMBLE) {
2145 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2146 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2148 if (!spa_config_valid(spa, nvconfig)) {
2149 nvlist_free(nvconfig);
2150 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2153 nvlist_free(nvconfig);
2156 * Now that we've validate the config, check the state of the
2157 * root vdev. If it can't be opened, it indicates one or
2158 * more toplevel vdevs are faulted.
2160 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2163 if (spa_check_logs(spa)) {
2164 *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2165 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2170 * We've successfully opened the pool, verify that we're ready
2171 * to start pushing transactions.
2173 if (state != SPA_LOAD_TRYIMPORT) {
2174 if ((error = spa_load_verify(spa)))
2175 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2179 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2180 spa->spa_load_max_txg == UINT64_MAX)) {
2182 int need_update = B_FALSE;
2185 ASSERT(state != SPA_LOAD_TRYIMPORT);
2188 * Claim log blocks that haven't been committed yet.
2189 * This must all happen in a single txg.
2190 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2191 * invoked from zil_claim_log_block()'s i/o done callback.
2192 * Price of rollback is that we abandon the log.
2194 spa->spa_claiming = B_TRUE;
2196 tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2197 spa_first_txg(spa));
2198 (void) dmu_objset_find(spa_name(spa),
2199 zil_claim, tx, DS_FIND_CHILDREN);
2202 spa->spa_claiming = B_FALSE;
2204 spa_set_log_state(spa, SPA_LOG_GOOD);
2205 spa->spa_sync_on = B_TRUE;
2206 txg_sync_start(spa->spa_dsl_pool);
2209 * Wait for all claims to sync. We sync up to the highest
2210 * claimed log block birth time so that claimed log blocks
2211 * don't appear to be from the future. spa_claim_max_txg
2212 * will have been set for us by either zil_check_log_chain()
2213 * (invoked from spa_check_logs()) or zil_claim() above.
2215 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2218 * If the config cache is stale, or we have uninitialized
2219 * metaslabs (see spa_vdev_add()), then update the config.
2221 * If this is a verbatim import, trust the current
2222 * in-core spa_config and update the disk labels.
2224 if (config_cache_txg != spa->spa_config_txg ||
2225 state == SPA_LOAD_IMPORT ||
2226 state == SPA_LOAD_RECOVER ||
2227 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2228 need_update = B_TRUE;
2230 for (c = 0; c < rvd->vdev_children; c++)
2231 if (rvd->vdev_child[c]->vdev_ms_array == 0)
2232 need_update = B_TRUE;
2235 * Update the config cache asychronously in case we're the
2236 * root pool, in which case the config cache isn't writable yet.
2239 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2242 * Check all DTLs to see if anything needs resilvering.
2244 if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2245 vdev_resilver_needed(rvd, NULL, NULL))
2246 spa_async_request(spa, SPA_ASYNC_RESILVER);
2249 * Delete any inconsistent datasets.
2251 (void) dmu_objset_find(spa_name(spa),
2252 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2255 * Clean up any stale temporary dataset userrefs.
2257 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2264 spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2266 int mode = spa->spa_mode;
2269 spa_deactivate(spa);
2271 spa->spa_load_max_txg--;
2273 spa_activate(spa, mode);
2274 spa_async_suspend(spa);
2276 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2280 spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2281 uint64_t max_request, int rewind_flags)
2283 nvlist_t *config = NULL;
2284 int load_error, rewind_error;
2285 uint64_t safe_rewind_txg;
2288 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2289 spa->spa_load_max_txg = spa->spa_load_txg;
2290 spa_set_log_state(spa, SPA_LOG_CLEAR);
2292 spa->spa_load_max_txg = max_request;
2295 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2297 if (load_error == 0)
2300 if (spa->spa_root_vdev != NULL)
2301 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2303 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2304 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2306 if (rewind_flags & ZPOOL_NEVER_REWIND) {
2307 nvlist_free(config);
2308 return (load_error);
2311 /* Price of rolling back is discarding txgs, including log */
2312 if (state == SPA_LOAD_RECOVER)
2313 spa_set_log_state(spa, SPA_LOG_CLEAR);
2315 spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2316 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2317 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2318 TXG_INITIAL : safe_rewind_txg;
2321 * Continue as long as we're finding errors, we're still within
2322 * the acceptable rewind range, and we're still finding uberblocks
2324 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2325 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2326 if (spa->spa_load_max_txg < safe_rewind_txg)
2327 spa->spa_extreme_rewind = B_TRUE;
2328 rewind_error = spa_load_retry(spa, state, mosconfig);
2331 spa->spa_extreme_rewind = B_FALSE;
2332 spa->spa_load_max_txg = UINT64_MAX;
2334 if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2335 spa_config_set(spa, config);
2337 return (state == SPA_LOAD_RECOVER ? rewind_error : load_error);
2343 * The import case is identical to an open except that the configuration is sent
2344 * down from userland, instead of grabbed from the configuration cache. For the
2345 * case of an open, the pool configuration will exist in the
2346 * POOL_STATE_UNINITIALIZED state.
2348 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2349 * the same time open the pool, without having to keep around the spa_t in some
2353 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2357 spa_load_state_t state = SPA_LOAD_OPEN;
2359 int locked = B_FALSE;
2364 * As disgusting as this is, we need to support recursive calls to this
2365 * function because dsl_dir_open() is called during spa_load(), and ends
2366 * up calling spa_open() again. The real fix is to figure out how to
2367 * avoid dsl_dir_open() calling this in the first place.
2369 if (mutex_owner(&spa_namespace_lock) != curthread) {
2370 mutex_enter(&spa_namespace_lock);
2374 if ((spa = spa_lookup(pool)) == NULL) {
2376 mutex_exit(&spa_namespace_lock);
2380 if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2381 zpool_rewind_policy_t policy;
2383 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2385 if (policy.zrp_request & ZPOOL_DO_REWIND)
2386 state = SPA_LOAD_RECOVER;
2388 spa_activate(spa, spa_mode_global);
2390 if (state != SPA_LOAD_RECOVER)
2391 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2393 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2394 policy.zrp_request);
2396 if (error == EBADF) {
2398 * If vdev_validate() returns failure (indicated by
2399 * EBADF), it indicates that one of the vdevs indicates
2400 * that the pool has been exported or destroyed. If
2401 * this is the case, the config cache is out of sync and
2402 * we should remove the pool from the namespace.
2405 spa_deactivate(spa);
2406 spa_config_sync(spa, B_TRUE, B_TRUE);
2409 mutex_exit(&spa_namespace_lock);
2415 * We can't open the pool, but we still have useful
2416 * information: the state of each vdev after the
2417 * attempted vdev_open(). Return this to the user.
2419 if (config != NULL && spa->spa_config) {
2420 VERIFY(nvlist_dup(spa->spa_config, config,
2422 VERIFY(nvlist_add_nvlist(*config,
2423 ZPOOL_CONFIG_LOAD_INFO,
2424 spa->spa_load_info) == 0);
2427 spa_deactivate(spa);
2428 spa->spa_last_open_failed = error;
2430 mutex_exit(&spa_namespace_lock);
2436 spa_open_ref(spa, tag);
2439 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2442 * If we've recovered the pool, pass back any information we
2443 * gathered while doing the load.
2445 if (state == SPA_LOAD_RECOVER) {
2446 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
2447 spa->spa_load_info) == 0);
2451 spa->spa_last_open_failed = 0;
2452 spa->spa_last_ubsync_txg = 0;
2453 spa->spa_load_txg = 0;
2454 mutex_exit(&spa_namespace_lock);
2463 spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
2466 return (spa_open_common(name, spapp, tag, policy, config));
2470 spa_open(const char *name, spa_t **spapp, void *tag)
2472 return (spa_open_common(name, spapp, tag, NULL, NULL));
2476 * Lookup the given spa_t, incrementing the inject count in the process,
2477 * preventing it from being exported or destroyed.
2480 spa_inject_addref(char *name)
2484 mutex_enter(&spa_namespace_lock);
2485 if ((spa = spa_lookup(name)) == NULL) {
2486 mutex_exit(&spa_namespace_lock);
2489 spa->spa_inject_ref++;
2490 mutex_exit(&spa_namespace_lock);
2496 spa_inject_delref(spa_t *spa)
2498 mutex_enter(&spa_namespace_lock);
2499 spa->spa_inject_ref--;
2500 mutex_exit(&spa_namespace_lock);
2504 * Add spares device information to the nvlist.
2507 spa_add_spares(spa_t *spa, nvlist_t *config)
2517 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2519 if (spa->spa_spares.sav_count == 0)
2522 VERIFY(nvlist_lookup_nvlist(config,
2523 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2524 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
2525 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2527 VERIFY(nvlist_add_nvlist_array(nvroot,
2528 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2529 VERIFY(nvlist_lookup_nvlist_array(nvroot,
2530 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2533 * Go through and find any spares which have since been
2534 * repurposed as an active spare. If this is the case, update
2535 * their status appropriately.
2537 for (i = 0; i < nspares; i++) {
2538 VERIFY(nvlist_lookup_uint64(spares[i],
2539 ZPOOL_CONFIG_GUID, &guid) == 0);
2540 if (spa_spare_exists(guid, &pool, NULL) &&
2542 VERIFY(nvlist_lookup_uint64_array(
2543 spares[i], ZPOOL_CONFIG_VDEV_STATS,
2544 (uint64_t **)&vs, &vsc) == 0);
2545 vs->vs_state = VDEV_STATE_CANT_OPEN;
2546 vs->vs_aux = VDEV_AUX_SPARED;
2553 * Add l2cache device information to the nvlist, including vdev stats.
2556 spa_add_l2cache(spa_t *spa, nvlist_t *config)
2559 uint_t i, j, nl2cache;
2566 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2568 if (spa->spa_l2cache.sav_count == 0)
2571 VERIFY(nvlist_lookup_nvlist(config,
2572 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2573 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
2574 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2575 if (nl2cache != 0) {
2576 VERIFY(nvlist_add_nvlist_array(nvroot,
2577 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2578 VERIFY(nvlist_lookup_nvlist_array(nvroot,
2579 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2582 * Update level 2 cache device stats.
2585 for (i = 0; i < nl2cache; i++) {
2586 VERIFY(nvlist_lookup_uint64(l2cache[i],
2587 ZPOOL_CONFIG_GUID, &guid) == 0);
2590 for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
2592 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
2593 vd = spa->spa_l2cache.sav_vdevs[j];
2599 VERIFY(nvlist_lookup_uint64_array(l2cache[i],
2600 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
2602 vdev_get_stats(vd, vs);
2608 spa_get_stats(const char *name, nvlist_t **config, char *altroot, size_t buflen)
2614 error = spa_open_common(name, &spa, FTAG, NULL, config);
2618 * This still leaves a window of inconsistency where the spares
2619 * or l2cache devices could change and the config would be
2620 * self-inconsistent.
2622 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
2624 if (*config != NULL) {
2625 uint64_t loadtimes[2];
2627 loadtimes[0] = spa->spa_loaded_ts.tv_sec;
2628 loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
2629 VERIFY(nvlist_add_uint64_array(*config,
2630 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
2632 VERIFY(nvlist_add_uint64(*config,
2633 ZPOOL_CONFIG_ERRCOUNT,
2634 spa_get_errlog_size(spa)) == 0);
2636 if (spa_suspended(spa))
2637 VERIFY(nvlist_add_uint64(*config,
2638 ZPOOL_CONFIG_SUSPENDED,
2639 spa->spa_failmode) == 0);
2641 spa_add_spares(spa, *config);
2642 spa_add_l2cache(spa, *config);
2647 * We want to get the alternate root even for faulted pools, so we cheat
2648 * and call spa_lookup() directly.
2652 mutex_enter(&spa_namespace_lock);
2653 spa = spa_lookup(name);
2655 spa_altroot(spa, altroot, buflen);
2659 mutex_exit(&spa_namespace_lock);
2661 spa_altroot(spa, altroot, buflen);
2666 spa_config_exit(spa, SCL_CONFIG, FTAG);
2667 spa_close(spa, FTAG);
2674 * Validate that the auxiliary device array is well formed. We must have an
2675 * array of nvlists, each which describes a valid leaf vdev. If this is an
2676 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
2677 * specified, as long as they are well-formed.
2680 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
2681 spa_aux_vdev_t *sav, const char *config, uint64_t version,
2682 vdev_labeltype_t label)
2689 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
2692 * It's acceptable to have no devs specified.
2694 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
2701 * Make sure the pool is formatted with a version that supports this
2704 if (spa_version(spa) < version)
2708 * Set the pending device list so we correctly handle device in-use
2711 sav->sav_pending = dev;
2712 sav->sav_npending = ndev;
2714 for (i = 0; i < ndev; i++) {
2715 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
2719 if (!vd->vdev_ops->vdev_op_leaf) {
2726 * The L2ARC currently only supports disk devices in
2727 * kernel context. For user-level testing, we allow it.
2730 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
2731 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
2738 if ((error = vdev_open(vd)) == 0 &&
2739 (error = vdev_label_init(vd, crtxg, label)) == 0) {
2740 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
2741 vd->vdev_guid) == 0);
2747 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
2754 sav->sav_pending = NULL;
2755 sav->sav_npending = 0;
2760 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
2764 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
2766 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
2767 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
2768 VDEV_LABEL_SPARE)) != 0) {
2772 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
2773 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
2774 VDEV_LABEL_L2CACHE));
2778 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
2783 if (sav->sav_config != NULL) {
2789 * Generate new dev list by concatentating with the
2792 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
2793 &olddevs, &oldndevs) == 0);
2795 newdevs = kmem_alloc(sizeof (void *) *
2796 (ndevs + oldndevs), KM_SLEEP);
2797 for (i = 0; i < oldndevs; i++)
2798 VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
2800 for (i = 0; i < ndevs; i++)
2801 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
2804 VERIFY(nvlist_remove(sav->sav_config, config,
2805 DATA_TYPE_NVLIST_ARRAY) == 0);
2807 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
2808 config, newdevs, ndevs + oldndevs) == 0);
2809 for (i = 0; i < oldndevs + ndevs; i++)
2810 nvlist_free(newdevs[i]);
2811 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
2814 * Generate a new dev list.
2816 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
2818 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
2824 * Stop and drop level 2 ARC devices
2827 spa_l2cache_drop(spa_t *spa)
2831 spa_aux_vdev_t *sav = &spa->spa_l2cache;
2833 for (i = 0; i < sav->sav_count; i++) {
2836 vd = sav->sav_vdevs[i];
2839 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
2840 pool != 0ULL && l2arc_vdev_present(vd))
2841 l2arc_remove_vdev(vd);
2842 if (vd->vdev_isl2cache)
2843 spa_l2cache_remove(vd);
2844 vdev_clear_stats(vd);
2845 (void) vdev_close(vd);
2853 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
2854 const char *history_str, nvlist_t *zplprops)
2857 char *altroot = NULL;
2862 uint64_t txg = TXG_INITIAL;
2863 nvlist_t **spares, **l2cache;
2864 uint_t nspares, nl2cache;
2865 uint64_t version, obj;
2869 * If this pool already exists, return failure.
2871 mutex_enter(&spa_namespace_lock);
2872 if (spa_lookup(pool) != NULL) {
2873 mutex_exit(&spa_namespace_lock);
2878 * Allocate a new spa_t structure.
2880 (void) nvlist_lookup_string(props,
2881 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
2882 spa = spa_add(pool, NULL, altroot);
2883 spa_activate(spa, spa_mode_global);
2885 if (props && (error = spa_prop_validate(spa, props))) {
2886 spa_deactivate(spa);
2888 mutex_exit(&spa_namespace_lock);
2892 if (nvlist_lookup_uint64(props, zpool_prop_to_name(ZPOOL_PROP_VERSION),
2894 version = SPA_VERSION;
2895 ASSERT(version <= SPA_VERSION);
2897 spa->spa_first_txg = txg;
2898 spa->spa_uberblock.ub_txg = txg - 1;
2899 spa->spa_uberblock.ub_version = version;
2900 spa->spa_ubsync = spa->spa_uberblock;
2903 * Create "The Godfather" zio to hold all async IOs
2905 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2906 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2909 * Create the root vdev.
2911 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2913 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
2915 ASSERT(error != 0 || rvd != NULL);
2916 ASSERT(error != 0 || spa->spa_root_vdev == rvd);
2918 if (error == 0 && !zfs_allocatable_devs(nvroot))
2922 (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
2923 (error = spa_validate_aux(spa, nvroot, txg,
2924 VDEV_ALLOC_ADD)) == 0) {
2925 for (c = 0; c < rvd->vdev_children; c++) {
2926 vdev_metaslab_set_size(rvd->vdev_child[c]);
2927 vdev_expand(rvd->vdev_child[c], txg);
2931 spa_config_exit(spa, SCL_ALL, FTAG);
2935 spa_deactivate(spa);
2937 mutex_exit(&spa_namespace_lock);
2942 * Get the list of spares, if specified.
2944 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
2945 &spares, &nspares) == 0) {
2946 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
2948 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
2949 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2950 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2951 spa_load_spares(spa);
2952 spa_config_exit(spa, SCL_ALL, FTAG);
2953 spa->spa_spares.sav_sync = B_TRUE;
2957 * Get the list of level 2 cache devices, if specified.
2959 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
2960 &l2cache, &nl2cache) == 0) {
2961 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
2962 NV_UNIQUE_NAME, KM_SLEEP) == 0);
2963 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
2964 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2965 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2966 spa_load_l2cache(spa);
2967 spa_config_exit(spa, SCL_ALL, FTAG);
2968 spa->spa_l2cache.sav_sync = B_TRUE;
2971 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
2972 spa->spa_meta_objset = dp->dp_meta_objset;
2975 * Create DDTs (dedup tables).
2979 spa_update_dspace(spa);
2981 tx = dmu_tx_create_assigned(dp, txg);
2984 * Create the pool config object.
2986 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
2987 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
2988 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
2990 if (zap_add(spa->spa_meta_objset,
2991 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
2992 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
2993 cmn_err(CE_PANIC, "failed to add pool config");
2996 if (zap_add(spa->spa_meta_objset,
2997 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
2998 sizeof (uint64_t), 1, &version, tx) != 0) {
2999 cmn_err(CE_PANIC, "failed to add pool version");
3002 /* Newly created pools with the right version are always deflated. */
3003 if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3004 spa->spa_deflate = TRUE;
3005 if (zap_add(spa->spa_meta_objset,
3006 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3007 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3008 cmn_err(CE_PANIC, "failed to add deflate");
3013 * Create the deferred-free bpobj. Turn off compression
3014 * because sync-to-convergence takes longer if the blocksize
3017 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3018 dmu_object_set_compress(spa->spa_meta_objset, obj,
3019 ZIO_COMPRESS_OFF, tx);
3020 if (zap_add(spa->spa_meta_objset,
3021 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3022 sizeof (uint64_t), 1, &obj, tx) != 0) {
3023 cmn_err(CE_PANIC, "failed to add bpobj");
3025 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3026 spa->spa_meta_objset, obj));
3029 * Create the pool's history object.
3031 if (version >= SPA_VERSION_ZPOOL_HISTORY)
3032 spa_history_create_obj(spa, tx);
3035 * Set pool properties.
3037 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3038 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3039 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3040 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3042 if (props != NULL) {
3043 spa_configfile_set(spa, props, B_FALSE);
3044 spa_sync_props(spa, props, tx);
3049 spa->spa_sync_on = B_TRUE;
3050 txg_sync_start(spa->spa_dsl_pool);
3053 * We explicitly wait for the first transaction to complete so that our
3054 * bean counters are appropriately updated.
3056 txg_wait_synced(spa->spa_dsl_pool, txg);
3058 spa_config_sync(spa, B_FALSE, B_TRUE);
3060 if (version >= SPA_VERSION_ZPOOL_HISTORY && history_str != NULL)
3061 (void) spa_history_log(spa, history_str, LOG_CMD_POOL_CREATE);
3062 spa_history_log_version(spa, LOG_POOL_CREATE);
3064 spa->spa_minref = refcount_count(&spa->spa_refcount);
3066 mutex_exit(&spa_namespace_lock);
3073 * Get the root pool information from the root disk, then import the root pool
3074 * during the system boot up time.
3076 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3079 spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3082 nvlist_t *nvtop, *nvroot;
3085 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3089 * Add this top-level vdev to the child array.
3091 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3093 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3095 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3098 * Put this pool's top-level vdevs into a root vdev.
3100 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3101 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3102 VDEV_TYPE_ROOT) == 0);
3103 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3104 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3105 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3109 * Replace the existing vdev_tree with the new root vdev in
3110 * this pool's configuration (remove the old, add the new).
3112 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3113 nvlist_free(nvroot);
3118 * Walk the vdev tree and see if we can find a device with "better"
3119 * configuration. A configuration is "better" if the label on that
3120 * device has a more recent txg.
3123 spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3127 for (c = 0; c < vd->vdev_children; c++)
3128 spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3130 if (vd->vdev_ops->vdev_op_leaf) {
3134 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3138 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3142 * Do we have a better boot device?
3144 if (label_txg > *txg) {
3153 * Import a root pool.
3155 * For x86. devpath_list will consist of devid and/or physpath name of
3156 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3157 * The GRUB "findroot" command will return the vdev we should boot.
3159 * For Sparc, devpath_list consists the physpath name of the booting device
3160 * no matter the rootpool is a single device pool or a mirrored pool.
3162 * "/pci@1f,0/ide@d/disk@0,0:a"
3165 spa_import_rootpool(char *devpath, char *devid)
3168 vdev_t *rvd, *bvd, *avd = NULL;
3169 nvlist_t *config, *nvtop;
3175 * Read the label from the boot device and generate a configuration.
3177 config = spa_generate_rootconf(devpath, devid, &guid);
3178 #if defined(_OBP) && defined(_KERNEL)
3179 if (config == NULL) {
3180 if (strstr(devpath, "/iscsi/ssd") != NULL) {
3182 get_iscsi_bootpath_phy(devpath);
3183 config = spa_generate_rootconf(devpath, devid, &guid);
3187 if (config == NULL) {
3188 cmn_err(CE_NOTE, "Can not read the pool label from '%s'",
3193 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3195 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3197 mutex_enter(&spa_namespace_lock);
3198 if ((spa = spa_lookup(pname)) != NULL) {
3200 * Remove the existing root pool from the namespace so that we
3201 * can replace it with the correct config we just read in.
3206 spa = spa_add(pname, config, NULL);
3207 spa->spa_is_root = B_TRUE;
3208 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3211 * Build up a vdev tree based on the boot device's label config.
3213 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3215 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3216 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3217 VDEV_ALLOC_ROOTPOOL);
3218 spa_config_exit(spa, SCL_ALL, FTAG);
3220 mutex_exit(&spa_namespace_lock);
3221 nvlist_free(config);
3222 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3228 * Get the boot vdev.
3230 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3231 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3232 (u_longlong_t)guid);
3238 * Determine if there is a better boot device.
3241 spa_alt_rootvdev(rvd, &avd, &txg);
3243 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3244 "try booting from '%s'", avd->vdev_path);
3250 * If the boot device is part of a spare vdev then ensure that
3251 * we're booting off the active spare.
3253 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3254 !bvd->vdev_isspare) {
3255 cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3256 "try booting from '%s'",
3258 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3264 spa_history_log_version(spa, LOG_POOL_IMPORT);
3266 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3268 spa_config_exit(spa, SCL_ALL, FTAG);
3269 mutex_exit(&spa_namespace_lock);
3271 nvlist_free(config);
3278 * Import a non-root pool into the system.
3281 spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
3284 char *altroot = NULL;
3285 spa_load_state_t state = SPA_LOAD_IMPORT;
3286 zpool_rewind_policy_t policy;
3287 uint64_t mode = spa_mode_global;
3288 uint64_t readonly = B_FALSE;
3291 nvlist_t **spares, **l2cache;
3292 uint_t nspares, nl2cache;
3295 * If a pool with this name exists, return failure.
3297 mutex_enter(&spa_namespace_lock);
3298 if (spa_lookup(pool) != NULL) {
3299 mutex_exit(&spa_namespace_lock);
3304 * Create and initialize the spa structure.
3306 (void) nvlist_lookup_string(props,
3307 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3308 (void) nvlist_lookup_uint64(props,
3309 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
3312 spa = spa_add(pool, config, altroot);
3313 spa->spa_import_flags = flags;
3316 * Verbatim import - Take a pool and insert it into the namespace
3317 * as if it had been loaded at boot.
3319 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
3321 spa_configfile_set(spa, props, B_FALSE);
3323 spa_config_sync(spa, B_FALSE, B_TRUE);
3325 mutex_exit(&spa_namespace_lock);
3326 spa_history_log_version(spa, LOG_POOL_IMPORT);
3331 spa_activate(spa, mode);
3334 * Don't start async tasks until we know everything is healthy.
3336 spa_async_suspend(spa);
3338 zpool_get_rewind_policy(config, &policy);
3339 if (policy.zrp_request & ZPOOL_DO_REWIND)
3340 state = SPA_LOAD_RECOVER;
3343 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
3344 * because the user-supplied config is actually the one to trust when
3347 if (state != SPA_LOAD_RECOVER)
3348 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3350 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
3351 policy.zrp_request);
3354 * Propagate anything learned while loading the pool and pass it
3355 * back to caller (i.e. rewind info, missing devices, etc).
3357 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
3358 spa->spa_load_info) == 0);
3360 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3362 * Toss any existing sparelist, as it doesn't have any validity
3363 * anymore, and conflicts with spa_has_spare().
3365 if (spa->spa_spares.sav_config) {
3366 nvlist_free(spa->spa_spares.sav_config);
3367 spa->spa_spares.sav_config = NULL;
3368 spa_load_spares(spa);
3370 if (spa->spa_l2cache.sav_config) {
3371 nvlist_free(spa->spa_l2cache.sav_config);
3372 spa->spa_l2cache.sav_config = NULL;
3373 spa_load_l2cache(spa);
3376 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3379 error = spa_validate_aux(spa, nvroot, -1ULL,
3382 error = spa_validate_aux(spa, nvroot, -1ULL,
3383 VDEV_ALLOC_L2CACHE);
3384 spa_config_exit(spa, SCL_ALL, FTAG);
3387 spa_configfile_set(spa, props, B_FALSE);
3389 if (error != 0 || (props && spa_writeable(spa) &&
3390 (error = spa_prop_set(spa, props)))) {
3392 spa_deactivate(spa);
3394 mutex_exit(&spa_namespace_lock);
3398 spa_async_resume(spa);
3401 * Override any spares and level 2 cache devices as specified by
3402 * the user, as these may have correct device names/devids, etc.
3404 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3405 &spares, &nspares) == 0) {
3406 if (spa->spa_spares.sav_config)
3407 VERIFY(nvlist_remove(spa->spa_spares.sav_config,
3408 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
3410 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
3411 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3412 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3413 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3414 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3415 spa_load_spares(spa);
3416 spa_config_exit(spa, SCL_ALL, FTAG);
3417 spa->spa_spares.sav_sync = B_TRUE;
3419 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3420 &l2cache, &nl2cache) == 0) {
3421 if (spa->spa_l2cache.sav_config)
3422 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
3423 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
3425 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3426 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3427 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3428 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3429 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3430 spa_load_l2cache(spa);
3431 spa_config_exit(spa, SCL_ALL, FTAG);
3432 spa->spa_l2cache.sav_sync = B_TRUE;
3436 * Check for any removed devices.
3438 if (spa->spa_autoreplace) {
3439 spa_aux_check_removed(&spa->spa_spares);
3440 spa_aux_check_removed(&spa->spa_l2cache);
3443 if (spa_writeable(spa)) {
3445 * Update the config cache to include the newly-imported pool.
3447 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3451 * It's possible that the pool was expanded while it was exported.
3452 * We kick off an async task to handle this for us.
3454 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
3456 mutex_exit(&spa_namespace_lock);
3457 spa_history_log_version(spa, LOG_POOL_IMPORT);
3463 spa_tryimport(nvlist_t *tryconfig)
3465 nvlist_t *config = NULL;
3471 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
3474 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
3478 * Create and initialize the spa structure.
3480 mutex_enter(&spa_namespace_lock);
3481 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
3482 spa_activate(spa, FREAD);
3485 * Pass off the heavy lifting to spa_load().
3486 * Pass TRUE for mosconfig because the user-supplied config
3487 * is actually the one to trust when doing an import.
3489 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
3492 * If 'tryconfig' was at least parsable, return the current config.
3494 if (spa->spa_root_vdev != NULL) {
3495 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
3496 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
3498 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
3500 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
3501 spa->spa_uberblock.ub_timestamp) == 0);
3504 * If the bootfs property exists on this pool then we
3505 * copy it out so that external consumers can tell which
3506 * pools are bootable.
3508 if ((!error || error == EEXIST) && spa->spa_bootfs) {
3509 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3512 * We have to play games with the name since the
3513 * pool was opened as TRYIMPORT_NAME.
3515 if (dsl_dsobj_to_dsname(spa_name(spa),
3516 spa->spa_bootfs, tmpname) == 0) {
3518 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3520 cp = strchr(tmpname, '/');
3522 (void) strlcpy(dsname, tmpname,
3525 (void) snprintf(dsname, MAXPATHLEN,
3526 "%s/%s", poolname, ++cp);
3528 VERIFY(nvlist_add_string(config,
3529 ZPOOL_CONFIG_BOOTFS, dsname) == 0);
3530 kmem_free(dsname, MAXPATHLEN);
3532 kmem_free(tmpname, MAXPATHLEN);
3536 * Add the list of hot spares and level 2 cache devices.
3538 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3539 spa_add_spares(spa, config);
3540 spa_add_l2cache(spa, config);
3541 spa_config_exit(spa, SCL_CONFIG, FTAG);
3545 spa_deactivate(spa);
3547 mutex_exit(&spa_namespace_lock);
3553 * Pool export/destroy
3555 * The act of destroying or exporting a pool is very simple. We make sure there
3556 * is no more pending I/O and any references to the pool are gone. Then, we
3557 * update the pool state and sync all the labels to disk, removing the
3558 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
3559 * we don't sync the labels or remove the configuration cache.
3562 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
3563 boolean_t force, boolean_t hardforce)
3570 if (!(spa_mode_global & FWRITE))
3573 mutex_enter(&spa_namespace_lock);
3574 if ((spa = spa_lookup(pool)) == NULL) {
3575 mutex_exit(&spa_namespace_lock);
3580 * Put a hold on the pool, drop the namespace lock, stop async tasks,
3581 * reacquire the namespace lock, and see if we can export.
3583 spa_open_ref(spa, FTAG);
3584 mutex_exit(&spa_namespace_lock);
3585 spa_async_suspend(spa);
3586 mutex_enter(&spa_namespace_lock);
3587 spa_close(spa, FTAG);
3590 * The pool will be in core if it's openable,
3591 * in which case we can modify its state.
3593 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
3595 * Objsets may be open only because they're dirty, so we
3596 * have to force it to sync before checking spa_refcnt.
3598 txg_wait_synced(spa->spa_dsl_pool, 0);
3601 * A pool cannot be exported or destroyed if there are active
3602 * references. If we are resetting a pool, allow references by
3603 * fault injection handlers.
3605 if (!spa_refcount_zero(spa) ||
3606 (spa->spa_inject_ref != 0 &&
3607 new_state != POOL_STATE_UNINITIALIZED)) {
3608 spa_async_resume(spa);
3609 mutex_exit(&spa_namespace_lock);
3614 * A pool cannot be exported if it has an active shared spare.
3615 * This is to prevent other pools stealing the active spare
3616 * from an exported pool. At user's own will, such pool can
3617 * be forcedly exported.
3619 if (!force && new_state == POOL_STATE_EXPORTED &&
3620 spa_has_active_shared_spare(spa)) {
3621 spa_async_resume(spa);
3622 mutex_exit(&spa_namespace_lock);
3627 * We want this to be reflected on every label,
3628 * so mark them all dirty. spa_unload() will do the
3629 * final sync that pushes these changes out.
3631 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
3632 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3633 spa->spa_state = new_state;
3634 spa->spa_final_txg = spa_last_synced_txg(spa) +
3636 vdev_config_dirty(spa->spa_root_vdev);
3637 spa_config_exit(spa, SCL_ALL, FTAG);
3641 spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
3643 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
3645 spa_deactivate(spa);
3648 if (oldconfig && spa->spa_config)
3649 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
3651 if (new_state != POOL_STATE_UNINITIALIZED) {
3653 spa_config_sync(spa, B_TRUE, B_TRUE);
3656 mutex_exit(&spa_namespace_lock);
3662 * Destroy a storage pool.
3665 spa_destroy(char *pool)
3667 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
3672 * Export a storage pool.
3675 spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
3676 boolean_t hardforce)
3678 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
3683 * Similar to spa_export(), this unloads the spa_t without actually removing it
3684 * from the namespace in any way.
3687 spa_reset(char *pool)
3689 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
3694 * ==========================================================================
3695 * Device manipulation
3696 * ==========================================================================
3700 * Add a device to a storage pool.
3703 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
3707 vdev_t *rvd = spa->spa_root_vdev;
3709 nvlist_t **spares, **l2cache;
3710 uint_t nspares, nl2cache;
3713 ASSERT(spa_writeable(spa));
3715 txg = spa_vdev_enter(spa);
3717 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
3718 VDEV_ALLOC_ADD)) != 0)
3719 return (spa_vdev_exit(spa, NULL, txg, error));
3721 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */
3723 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
3727 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
3731 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
3732 return (spa_vdev_exit(spa, vd, txg, EINVAL));
3734 if (vd->vdev_children != 0 &&
3735 (error = vdev_create(vd, txg, B_FALSE)) != 0)
3736 return (spa_vdev_exit(spa, vd, txg, error));
3739 * We must validate the spares and l2cache devices after checking the
3740 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
3742 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
3743 return (spa_vdev_exit(spa, vd, txg, error));
3746 * Transfer each new top-level vdev from vd to rvd.
3748 for (c = 0; c < vd->vdev_children; c++) {
3751 * Set the vdev id to the first hole, if one exists.
3753 for (id = 0; id < rvd->vdev_children; id++) {
3754 if (rvd->vdev_child[id]->vdev_ishole) {
3755 vdev_free(rvd->vdev_child[id]);
3759 tvd = vd->vdev_child[c];
3760 vdev_remove_child(vd, tvd);
3762 vdev_add_child(rvd, tvd);
3763 vdev_config_dirty(tvd);
3767 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
3768 ZPOOL_CONFIG_SPARES);
3769 spa_load_spares(spa);
3770 spa->spa_spares.sav_sync = B_TRUE;
3773 if (nl2cache != 0) {
3774 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
3775 ZPOOL_CONFIG_L2CACHE);
3776 spa_load_l2cache(spa);
3777 spa->spa_l2cache.sav_sync = B_TRUE;
3781 * We have to be careful when adding new vdevs to an existing pool.
3782 * If other threads start allocating from these vdevs before we
3783 * sync the config cache, and we lose power, then upon reboot we may
3784 * fail to open the pool because there are DVAs that the config cache
3785 * can't translate. Therefore, we first add the vdevs without
3786 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
3787 * and then let spa_config_update() initialize the new metaslabs.
3789 * spa_load() checks for added-but-not-initialized vdevs, so that
3790 * if we lose power at any point in this sequence, the remaining
3791 * steps will be completed the next time we load the pool.
3793 (void) spa_vdev_exit(spa, vd, txg, 0);
3795 mutex_enter(&spa_namespace_lock);
3796 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3797 mutex_exit(&spa_namespace_lock);
3803 * Attach a device to a mirror. The arguments are the path to any device
3804 * in the mirror, and the nvroot for the new device. If the path specifies
3805 * a device that is not mirrored, we automatically insert the mirror vdev.
3807 * If 'replacing' is specified, the new device is intended to replace the
3808 * existing device; in this case the two devices are made into their own
3809 * mirror using the 'replacing' vdev, which is functionally identical to
3810 * the mirror vdev (it actually reuses all the same ops) but has a few
3811 * extra rules: you can't attach to it after it's been created, and upon
3812 * completion of resilvering, the first disk (the one being replaced)
3813 * is automatically detached.
3816 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
3818 uint64_t txg, dtl_max_txg;
3819 ASSERTV(vdev_t *rvd = spa->spa_root_vdev;)
3820 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
3822 char *oldvdpath, *newvdpath;
3826 ASSERT(spa_writeable(spa));
3828 txg = spa_vdev_enter(spa);
3830 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
3833 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
3835 if (!oldvd->vdev_ops->vdev_op_leaf)
3836 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3838 pvd = oldvd->vdev_parent;
3840 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
3841 VDEV_ALLOC_ADD)) != 0)
3842 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
3844 if (newrootvd->vdev_children != 1)
3845 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
3847 newvd = newrootvd->vdev_child[0];
3849 if (!newvd->vdev_ops->vdev_op_leaf)
3850 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
3852 if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
3853 return (spa_vdev_exit(spa, newrootvd, txg, error));
3856 * Spares can't replace logs
3858 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
3859 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3863 * For attach, the only allowable parent is a mirror or the root
3866 if (pvd->vdev_ops != &vdev_mirror_ops &&
3867 pvd->vdev_ops != &vdev_root_ops)
3868 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3870 pvops = &vdev_mirror_ops;
3873 * Active hot spares can only be replaced by inactive hot
3876 if (pvd->vdev_ops == &vdev_spare_ops &&
3877 oldvd->vdev_isspare &&
3878 !spa_has_spare(spa, newvd->vdev_guid))
3879 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3882 * If the source is a hot spare, and the parent isn't already a
3883 * spare, then we want to create a new hot spare. Otherwise, we
3884 * want to create a replacing vdev. The user is not allowed to
3885 * attach to a spared vdev child unless the 'isspare' state is
3886 * the same (spare replaces spare, non-spare replaces
3889 if (pvd->vdev_ops == &vdev_replacing_ops &&
3890 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
3891 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3892 } else if (pvd->vdev_ops == &vdev_spare_ops &&
3893 newvd->vdev_isspare != oldvd->vdev_isspare) {
3894 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3897 if (newvd->vdev_isspare)
3898 pvops = &vdev_spare_ops;
3900 pvops = &vdev_replacing_ops;
3904 * Make sure the new device is big enough.
3906 if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
3907 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
3910 * The new device cannot have a higher alignment requirement
3911 * than the top-level vdev.
3913 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
3914 return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
3917 * If this is an in-place replacement, update oldvd's path and devid
3918 * to make it distinguishable from newvd, and unopenable from now on.
3920 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
3921 spa_strfree(oldvd->vdev_path);
3922 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
3924 (void) sprintf(oldvd->vdev_path, "%s/%s",
3925 newvd->vdev_path, "old");
3926 if (oldvd->vdev_devid != NULL) {
3927 spa_strfree(oldvd->vdev_devid);
3928 oldvd->vdev_devid = NULL;
3932 /* mark the device being resilvered */
3933 newvd->vdev_resilvering = B_TRUE;
3936 * If the parent is not a mirror, or if we're replacing, insert the new
3937 * mirror/replacing/spare vdev above oldvd.
3939 if (pvd->vdev_ops != pvops)
3940 pvd = vdev_add_parent(oldvd, pvops);
3942 ASSERT(pvd->vdev_top->vdev_parent == rvd);
3943 ASSERT(pvd->vdev_ops == pvops);
3944 ASSERT(oldvd->vdev_parent == pvd);
3947 * Extract the new device from its root and add it to pvd.
3949 vdev_remove_child(newrootvd, newvd);
3950 newvd->vdev_id = pvd->vdev_children;
3951 newvd->vdev_crtxg = oldvd->vdev_crtxg;
3952 vdev_add_child(pvd, newvd);
3954 tvd = newvd->vdev_top;
3955 ASSERT(pvd->vdev_top == tvd);
3956 ASSERT(tvd->vdev_parent == rvd);
3958 vdev_config_dirty(tvd);
3961 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
3962 * for any dmu_sync-ed blocks. It will propagate upward when
3963 * spa_vdev_exit() calls vdev_dtl_reassess().
3965 dtl_max_txg = txg + TXG_CONCURRENT_STATES;
3967 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
3968 dtl_max_txg - TXG_INITIAL);
3970 if (newvd->vdev_isspare) {
3971 spa_spare_activate(newvd);
3972 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
3975 oldvdpath = spa_strdup(oldvd->vdev_path);
3976 newvdpath = spa_strdup(newvd->vdev_path);
3977 newvd_isspare = newvd->vdev_isspare;
3980 * Mark newvd's DTL dirty in this txg.
3982 vdev_dirty(tvd, VDD_DTL, newvd, txg);
3985 * Restart the resilver
3987 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
3992 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
3994 spa_history_log_internal(LOG_POOL_VDEV_ATTACH, spa, NULL,
3995 "%s vdev=%s %s vdev=%s",
3996 replacing && newvd_isspare ? "spare in" :
3997 replacing ? "replace" : "attach", newvdpath,
3998 replacing ? "for" : "to", oldvdpath);
4000 spa_strfree(oldvdpath);
4001 spa_strfree(newvdpath);
4003 if (spa->spa_bootfs)
4004 spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
4010 * Detach a device from a mirror or replacing vdev.
4011 * If 'replace_done' is specified, only detach if the parent
4012 * is a replacing vdev.
4015 spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4019 ASSERTV(vdev_t *rvd = spa->spa_root_vdev;)
4020 vdev_t *vd, *pvd, *cvd, *tvd;
4021 boolean_t unspare = B_FALSE;
4022 uint64_t unspare_guid = 0;
4026 ASSERT(spa_writeable(spa));
4028 txg = spa_vdev_enter(spa);
4030 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4033 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4035 if (!vd->vdev_ops->vdev_op_leaf)
4036 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4038 pvd = vd->vdev_parent;
4041 * If the parent/child relationship is not as expected, don't do it.
4042 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4043 * vdev that's replacing B with C. The user's intent in replacing
4044 * is to go from M(A,B) to M(A,C). If the user decides to cancel
4045 * the replace by detaching C, the expected behavior is to end up
4046 * M(A,B). But suppose that right after deciding to detach C,
4047 * the replacement of B completes. We would have M(A,C), and then
4048 * ask to detach C, which would leave us with just A -- not what
4049 * the user wanted. To prevent this, we make sure that the
4050 * parent/child relationship hasn't changed -- in this example,
4051 * that C's parent is still the replacing vdev R.
4053 if (pvd->vdev_guid != pguid && pguid != 0)
4054 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4057 * Only 'replacing' or 'spare' vdevs can be replaced.
4059 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4060 pvd->vdev_ops != &vdev_spare_ops)
4061 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4063 ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4064 spa_version(spa) >= SPA_VERSION_SPARES);
4067 * Only mirror, replacing, and spare vdevs support detach.
4069 if (pvd->vdev_ops != &vdev_replacing_ops &&
4070 pvd->vdev_ops != &vdev_mirror_ops &&
4071 pvd->vdev_ops != &vdev_spare_ops)
4072 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4075 * If this device has the only valid copy of some data,
4076 * we cannot safely detach it.
4078 if (vdev_dtl_required(vd))
4079 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4081 ASSERT(pvd->vdev_children >= 2);
4084 * If we are detaching the second disk from a replacing vdev, then
4085 * check to see if we changed the original vdev's path to have "/old"
4086 * at the end in spa_vdev_attach(). If so, undo that change now.
4088 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4089 vd->vdev_path != NULL) {
4090 size_t len = strlen(vd->vdev_path);
4092 for (c = 0; c < pvd->vdev_children; c++) {
4093 cvd = pvd->vdev_child[c];
4095 if (cvd == vd || cvd->vdev_path == NULL)
4098 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4099 strcmp(cvd->vdev_path + len, "/old") == 0) {
4100 spa_strfree(cvd->vdev_path);
4101 cvd->vdev_path = spa_strdup(vd->vdev_path);
4108 * If we are detaching the original disk from a spare, then it implies
4109 * that the spare should become a real disk, and be removed from the
4110 * active spare list for the pool.
4112 if (pvd->vdev_ops == &vdev_spare_ops &&
4114 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4118 * Erase the disk labels so the disk can be used for other things.
4119 * This must be done after all other error cases are handled,
4120 * but before we disembowel vd (so we can still do I/O to it).
4121 * But if we can't do it, don't treat the error as fatal --
4122 * it may be that the unwritability of the disk is the reason
4123 * it's being detached!
4125 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4128 * Remove vd from its parent and compact the parent's children.
4130 vdev_remove_child(pvd, vd);
4131 vdev_compact_children(pvd);
4134 * Remember one of the remaining children so we can get tvd below.
4136 cvd = pvd->vdev_child[pvd->vdev_children - 1];
4139 * If we need to remove the remaining child from the list of hot spares,
4140 * do it now, marking the vdev as no longer a spare in the process.
4141 * We must do this before vdev_remove_parent(), because that can
4142 * change the GUID if it creates a new toplevel GUID. For a similar
4143 * reason, we must remove the spare now, in the same txg as the detach;
4144 * otherwise someone could attach a new sibling, change the GUID, and
4145 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4148 ASSERT(cvd->vdev_isspare);
4149 spa_spare_remove(cvd);
4150 unspare_guid = cvd->vdev_guid;
4151 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4152 cvd->vdev_unspare = B_TRUE;
4156 * If the parent mirror/replacing vdev only has one child,
4157 * the parent is no longer needed. Remove it from the tree.
4159 if (pvd->vdev_children == 1) {
4160 if (pvd->vdev_ops == &vdev_spare_ops)
4161 cvd->vdev_unspare = B_FALSE;
4162 vdev_remove_parent(cvd);
4163 cvd->vdev_resilvering = B_FALSE;
4168 * We don't set tvd until now because the parent we just removed
4169 * may have been the previous top-level vdev.
4171 tvd = cvd->vdev_top;
4172 ASSERT(tvd->vdev_parent == rvd);
4175 * Reevaluate the parent vdev state.
4177 vdev_propagate_state(cvd);
4180 * If the 'autoexpand' property is set on the pool then automatically
4181 * try to expand the size of the pool. For example if the device we
4182 * just detached was smaller than the others, it may be possible to
4183 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4184 * first so that we can obtain the updated sizes of the leaf vdevs.
4186 if (spa->spa_autoexpand) {
4188 vdev_expand(tvd, txg);
4191 vdev_config_dirty(tvd);
4194 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
4195 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4196 * But first make sure we're not on any *other* txg's DTL list, to
4197 * prevent vd from being accessed after it's freed.
4199 vdpath = spa_strdup(vd->vdev_path);
4200 for (t = 0; t < TXG_SIZE; t++)
4201 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4202 vd->vdev_detached = B_TRUE;
4203 vdev_dirty(tvd, VDD_DTL, vd, txg);
4205 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
4207 /* hang on to the spa before we release the lock */
4208 spa_open_ref(spa, FTAG);
4210 error = spa_vdev_exit(spa, vd, txg, 0);
4212 spa_history_log_internal(LOG_POOL_VDEV_DETACH, spa, NULL,
4214 spa_strfree(vdpath);
4217 * If this was the removal of the original device in a hot spare vdev,
4218 * then we want to go through and remove the device from the hot spare
4219 * list of every other pool.
4222 spa_t *altspa = NULL;
4224 mutex_enter(&spa_namespace_lock);
4225 while ((altspa = spa_next(altspa)) != NULL) {
4226 if (altspa->spa_state != POOL_STATE_ACTIVE ||
4230 spa_open_ref(altspa, FTAG);
4231 mutex_exit(&spa_namespace_lock);
4232 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
4233 mutex_enter(&spa_namespace_lock);
4234 spa_close(altspa, FTAG);
4236 mutex_exit(&spa_namespace_lock);
4238 /* search the rest of the vdevs for spares to remove */
4239 spa_vdev_resilver_done(spa);
4242 /* all done with the spa; OK to release */
4243 mutex_enter(&spa_namespace_lock);
4244 spa_close(spa, FTAG);
4245 mutex_exit(&spa_namespace_lock);
4251 * Split a set of devices from their mirrors, and create a new pool from them.
4254 spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
4255 nvlist_t *props, boolean_t exp)
4258 uint64_t txg, *glist;
4260 uint_t c, children, lastlog;
4261 nvlist_t **child, *nvl, *tmp;
4263 char *altroot = NULL;
4264 vdev_t *rvd, **vml = NULL; /* vdev modify list */
4265 boolean_t activate_slog;
4267 ASSERT(spa_writeable(spa));
4269 txg = spa_vdev_enter(spa);
4271 /* clear the log and flush everything up to now */
4272 activate_slog = spa_passivate_log(spa);
4273 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4274 error = spa_offline_log(spa);
4275 txg = spa_vdev_config_enter(spa);
4278 spa_activate_log(spa);
4281 return (spa_vdev_exit(spa, NULL, txg, error));
4283 /* check new spa name before going any further */
4284 if (spa_lookup(newname) != NULL)
4285 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
4288 * scan through all the children to ensure they're all mirrors
4290 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
4291 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
4293 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4295 /* first, check to ensure we've got the right child count */
4296 rvd = spa->spa_root_vdev;
4298 for (c = 0; c < rvd->vdev_children; c++) {
4299 vdev_t *vd = rvd->vdev_child[c];
4301 /* don't count the holes & logs as children */
4302 if (vd->vdev_islog || vd->vdev_ishole) {
4310 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
4311 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4313 /* next, ensure no spare or cache devices are part of the split */
4314 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
4315 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
4316 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4318 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
4319 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
4321 /* then, loop over each vdev and validate it */
4322 for (c = 0; c < children; c++) {
4323 uint64_t is_hole = 0;
4325 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
4329 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
4330 spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
4338 /* which disk is going to be split? */
4339 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
4345 /* look it up in the spa */
4346 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
4347 if (vml[c] == NULL) {
4352 /* make sure there's nothing stopping the split */
4353 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
4354 vml[c]->vdev_islog ||
4355 vml[c]->vdev_ishole ||
4356 vml[c]->vdev_isspare ||
4357 vml[c]->vdev_isl2cache ||
4358 !vdev_writeable(vml[c]) ||
4359 vml[c]->vdev_children != 0 ||
4360 vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
4361 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
4366 if (vdev_dtl_required(vml[c])) {
4371 /* we need certain info from the top level */
4372 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
4373 vml[c]->vdev_top->vdev_ms_array) == 0);
4374 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
4375 vml[c]->vdev_top->vdev_ms_shift) == 0);
4376 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
4377 vml[c]->vdev_top->vdev_asize) == 0);
4378 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
4379 vml[c]->vdev_top->vdev_ashift) == 0);
4383 kmem_free(vml, children * sizeof (vdev_t *));
4384 kmem_free(glist, children * sizeof (uint64_t));
4385 return (spa_vdev_exit(spa, NULL, txg, error));
4388 /* stop writers from using the disks */
4389 for (c = 0; c < children; c++) {
4391 vml[c]->vdev_offline = B_TRUE;
4393 vdev_reopen(spa->spa_root_vdev);
4396 * Temporarily record the splitting vdevs in the spa config. This
4397 * will disappear once the config is regenerated.
4399 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4400 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
4401 glist, children) == 0);
4402 kmem_free(glist, children * sizeof (uint64_t));
4404 mutex_enter(&spa->spa_props_lock);
4405 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
4407 mutex_exit(&spa->spa_props_lock);
4408 spa->spa_config_splitting = nvl;
4409 vdev_config_dirty(spa->spa_root_vdev);
4411 /* configure and create the new pool */
4412 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
4413 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4414 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
4415 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
4416 spa_version(spa)) == 0);
4417 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
4418 spa->spa_config_txg) == 0);
4419 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4420 spa_generate_guid(NULL)) == 0);
4421 (void) nvlist_lookup_string(props,
4422 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4424 /* add the new pool to the namespace */
4425 newspa = spa_add(newname, config, altroot);
4426 newspa->spa_config_txg = spa->spa_config_txg;
4427 spa_set_log_state(newspa, SPA_LOG_CLEAR);
4429 /* release the spa config lock, retaining the namespace lock */
4430 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4432 if (zio_injection_enabled)
4433 zio_handle_panic_injection(spa, FTAG, 1);
4435 spa_activate(newspa, spa_mode_global);
4436 spa_async_suspend(newspa);
4438 /* create the new pool from the disks of the original pool */
4439 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
4443 /* if that worked, generate a real config for the new pool */
4444 if (newspa->spa_root_vdev != NULL) {
4445 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
4446 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4447 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
4448 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
4449 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
4454 if (props != NULL) {
4455 spa_configfile_set(newspa, props, B_FALSE);
4456 error = spa_prop_set(newspa, props);
4461 /* flush everything */
4462 txg = spa_vdev_config_enter(newspa);
4463 vdev_config_dirty(newspa->spa_root_vdev);
4464 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
4466 if (zio_injection_enabled)
4467 zio_handle_panic_injection(spa, FTAG, 2);
4469 spa_async_resume(newspa);
4471 /* finally, update the original pool's config */
4472 txg = spa_vdev_config_enter(spa);
4473 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
4474 error = dmu_tx_assign(tx, TXG_WAIT);
4477 for (c = 0; c < children; c++) {
4478 if (vml[c] != NULL) {
4481 spa_history_log_internal(LOG_POOL_VDEV_DETACH,
4487 vdev_config_dirty(spa->spa_root_vdev);
4488 spa->spa_config_splitting = NULL;
4492 (void) spa_vdev_exit(spa, NULL, txg, 0);
4494 if (zio_injection_enabled)
4495 zio_handle_panic_injection(spa, FTAG, 3);
4497 /* split is complete; log a history record */
4498 spa_history_log_internal(LOG_POOL_SPLIT, newspa, NULL,
4499 "split new pool %s from pool %s", newname, spa_name(spa));
4501 kmem_free(vml, children * sizeof (vdev_t *));
4503 /* if we're not going to mount the filesystems in userland, export */
4505 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
4512 spa_deactivate(newspa);
4515 txg = spa_vdev_config_enter(spa);
4517 /* re-online all offlined disks */
4518 for (c = 0; c < children; c++) {
4520 vml[c]->vdev_offline = B_FALSE;
4522 vdev_reopen(spa->spa_root_vdev);
4524 nvlist_free(spa->spa_config_splitting);
4525 spa->spa_config_splitting = NULL;
4526 (void) spa_vdev_exit(spa, NULL, txg, error);
4528 kmem_free(vml, children * sizeof (vdev_t *));
4533 spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
4537 for (i = 0; i < count; i++) {
4540 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
4543 if (guid == target_guid)
4551 spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
4552 nvlist_t *dev_to_remove)
4554 nvlist_t **newdev = NULL;
4558 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
4560 for (i = 0, j = 0; i < count; i++) {
4561 if (dev[i] == dev_to_remove)
4563 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
4566 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
4567 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
4569 for (i = 0; i < count - 1; i++)
4570 nvlist_free(newdev[i]);
4573 kmem_free(newdev, (count - 1) * sizeof (void *));
4577 * Evacuate the device.
4580 spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
4585 ASSERT(MUTEX_HELD(&spa_namespace_lock));
4586 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
4587 ASSERT(vd == vd->vdev_top);
4590 * Evacuate the device. We don't hold the config lock as writer
4591 * since we need to do I/O but we do keep the
4592 * spa_namespace_lock held. Once this completes the device
4593 * should no longer have any blocks allocated on it.
4595 if (vd->vdev_islog) {
4596 if (vd->vdev_stat.vs_alloc != 0)
4597 error = spa_offline_log(spa);
4606 * The evacuation succeeded. Remove any remaining MOS metadata
4607 * associated with this vdev, and wait for these changes to sync.
4609 ASSERT3U(vd->vdev_stat.vs_alloc, ==, 0);
4610 txg = spa_vdev_config_enter(spa);
4611 vd->vdev_removing = B_TRUE;
4612 vdev_dirty(vd, 0, NULL, txg);
4613 vdev_config_dirty(vd);
4614 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4620 * Complete the removal by cleaning up the namespace.
4623 spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
4625 vdev_t *rvd = spa->spa_root_vdev;
4626 uint64_t id = vd->vdev_id;
4627 boolean_t last_vdev = (id == (rvd->vdev_children - 1));
4629 ASSERT(MUTEX_HELD(&spa_namespace_lock));
4630 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
4631 ASSERT(vd == vd->vdev_top);
4634 * Only remove any devices which are empty.
4636 if (vd->vdev_stat.vs_alloc != 0)
4639 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4641 if (list_link_active(&vd->vdev_state_dirty_node))
4642 vdev_state_clean(vd);
4643 if (list_link_active(&vd->vdev_config_dirty_node))
4644 vdev_config_clean(vd);
4649 vdev_compact_children(rvd);
4651 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
4652 vdev_add_child(rvd, vd);
4654 vdev_config_dirty(rvd);
4657 * Reassess the health of our root vdev.
4663 * Remove a device from the pool -
4665 * Removing a device from the vdev namespace requires several steps
4666 * and can take a significant amount of time. As a result we use
4667 * the spa_vdev_config_[enter/exit] functions which allow us to
4668 * grab and release the spa_config_lock while still holding the namespace
4669 * lock. During each step the configuration is synced out.
4673 * Remove a device from the pool. Currently, this supports removing only hot
4674 * spares, slogs, and level 2 ARC devices.
4677 spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
4680 metaslab_group_t *mg;
4681 nvlist_t **spares, **l2cache, *nv;
4683 uint_t nspares, nl2cache;
4685 boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
4687 ASSERT(spa_writeable(spa));
4690 txg = spa_vdev_enter(spa);
4692 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4694 if (spa->spa_spares.sav_vdevs != NULL &&
4695 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
4696 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
4697 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
4699 * Only remove the hot spare if it's not currently in use
4702 if (vd == NULL || unspare) {
4703 spa_vdev_remove_aux(spa->spa_spares.sav_config,
4704 ZPOOL_CONFIG_SPARES, spares, nspares, nv);
4705 spa_load_spares(spa);
4706 spa->spa_spares.sav_sync = B_TRUE;
4710 } else if (spa->spa_l2cache.sav_vdevs != NULL &&
4711 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
4712 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
4713 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
4715 * Cache devices can always be removed.
4717 spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
4718 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
4719 spa_load_l2cache(spa);
4720 spa->spa_l2cache.sav_sync = B_TRUE;
4721 } else if (vd != NULL && vd->vdev_islog) {
4723 ASSERT(vd == vd->vdev_top);
4726 * XXX - Once we have bp-rewrite this should
4727 * become the common case.
4733 * Stop allocating from this vdev.
4735 metaslab_group_passivate(mg);
4738 * Wait for the youngest allocations and frees to sync,
4739 * and then wait for the deferral of those frees to finish.
4741 spa_vdev_config_exit(spa, NULL,
4742 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
4745 * Attempt to evacuate the vdev.
4747 error = spa_vdev_remove_evacuate(spa, vd);
4749 txg = spa_vdev_config_enter(spa);
4752 * If we couldn't evacuate the vdev, unwind.
4755 metaslab_group_activate(mg);
4756 return (spa_vdev_exit(spa, NULL, txg, error));
4760 * Clean up the vdev namespace.
4762 spa_vdev_remove_from_namespace(spa, vd);
4764 } else if (vd != NULL) {
4766 * Normal vdevs cannot be removed (yet).
4771 * There is no vdev of any kind with the specified guid.
4777 return (spa_vdev_exit(spa, NULL, txg, error));
4783 * Find any device that's done replacing, or a vdev marked 'unspare' that's
4784 * current spared, so we can detach it.
4787 spa_vdev_resilver_done_hunt(vdev_t *vd)
4789 vdev_t *newvd, *oldvd;
4792 for (c = 0; c < vd->vdev_children; c++) {
4793 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
4799 * Check for a completed replacement. We always consider the first
4800 * vdev in the list to be the oldest vdev, and the last one to be
4801 * the newest (see spa_vdev_attach() for how that works). In
4802 * the case where the newest vdev is faulted, we will not automatically
4803 * remove it after a resilver completes. This is OK as it will require
4804 * user intervention to determine which disk the admin wishes to keep.
4806 if (vd->vdev_ops == &vdev_replacing_ops) {
4807 ASSERT(vd->vdev_children > 1);
4809 newvd = vd->vdev_child[vd->vdev_children - 1];
4810 oldvd = vd->vdev_child[0];
4812 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
4813 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
4814 !vdev_dtl_required(oldvd))
4819 * Check for a completed resilver with the 'unspare' flag set.
4821 if (vd->vdev_ops == &vdev_spare_ops) {
4822 vdev_t *first = vd->vdev_child[0];
4823 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
4825 if (last->vdev_unspare) {
4828 } else if (first->vdev_unspare) {
4835 if (oldvd != NULL &&
4836 vdev_dtl_empty(newvd, DTL_MISSING) &&
4837 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
4838 !vdev_dtl_required(oldvd))
4842 * If there are more than two spares attached to a disk,
4843 * and those spares are not required, then we want to
4844 * attempt to free them up now so that they can be used
4845 * by other pools. Once we're back down to a single
4846 * disk+spare, we stop removing them.
4848 if (vd->vdev_children > 2) {
4849 newvd = vd->vdev_child[1];
4851 if (newvd->vdev_isspare && last->vdev_isspare &&
4852 vdev_dtl_empty(last, DTL_MISSING) &&
4853 vdev_dtl_empty(last, DTL_OUTAGE) &&
4854 !vdev_dtl_required(newvd))
4863 spa_vdev_resilver_done(spa_t *spa)
4865 vdev_t *vd, *pvd, *ppvd;
4866 uint64_t guid, sguid, pguid, ppguid;
4868 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4870 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
4871 pvd = vd->vdev_parent;
4872 ppvd = pvd->vdev_parent;
4873 guid = vd->vdev_guid;
4874 pguid = pvd->vdev_guid;
4875 ppguid = ppvd->vdev_guid;
4878 * If we have just finished replacing a hot spared device, then
4879 * we need to detach the parent's first child (the original hot
4882 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
4883 ppvd->vdev_children == 2) {
4884 ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
4885 sguid = ppvd->vdev_child[1]->vdev_guid;
4887 spa_config_exit(spa, SCL_ALL, FTAG);
4888 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
4890 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
4892 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4895 spa_config_exit(spa, SCL_ALL, FTAG);
4899 * Update the stored path or FRU for this vdev.
4902 spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
4906 boolean_t sync = B_FALSE;
4908 ASSERT(spa_writeable(spa));
4910 spa_vdev_state_enter(spa, SCL_ALL);
4912 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
4913 return (spa_vdev_state_exit(spa, NULL, ENOENT));
4915 if (!vd->vdev_ops->vdev_op_leaf)
4916 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
4919 if (strcmp(value, vd->vdev_path) != 0) {
4920 spa_strfree(vd->vdev_path);
4921 vd->vdev_path = spa_strdup(value);
4925 if (vd->vdev_fru == NULL) {
4926 vd->vdev_fru = spa_strdup(value);
4928 } else if (strcmp(value, vd->vdev_fru) != 0) {
4929 spa_strfree(vd->vdev_fru);
4930 vd->vdev_fru = spa_strdup(value);
4935 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
4939 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
4941 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
4945 spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
4947 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
4951 * ==========================================================================
4953 * ==========================================================================
4957 spa_scan_stop(spa_t *spa)
4959 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
4960 if (dsl_scan_resilvering(spa->spa_dsl_pool))
4962 return (dsl_scan_cancel(spa->spa_dsl_pool));
4966 spa_scan(spa_t *spa, pool_scan_func_t func)
4968 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
4970 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
4974 * If a resilver was requested, but there is no DTL on a
4975 * writeable leaf device, we have nothing to do.
4977 if (func == POOL_SCAN_RESILVER &&
4978 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
4979 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
4983 return (dsl_scan(spa->spa_dsl_pool, func));
4987 * ==========================================================================
4988 * SPA async task processing
4989 * ==========================================================================
4993 spa_async_remove(spa_t *spa, vdev_t *vd)
4997 if (vd->vdev_remove_wanted) {
4998 vd->vdev_remove_wanted = B_FALSE;
4999 vd->vdev_delayed_close = B_FALSE;
5000 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5003 * We want to clear the stats, but we don't want to do a full
5004 * vdev_clear() as that will cause us to throw away
5005 * degraded/faulted state as well as attempt to reopen the
5006 * device, all of which is a waste.
5008 vd->vdev_stat.vs_read_errors = 0;
5009 vd->vdev_stat.vs_write_errors = 0;
5010 vd->vdev_stat.vs_checksum_errors = 0;
5012 vdev_state_dirty(vd->vdev_top);
5015 for (c = 0; c < vd->vdev_children; c++)
5016 spa_async_remove(spa, vd->vdev_child[c]);
5020 spa_async_probe(spa_t *spa, vdev_t *vd)
5024 if (vd->vdev_probe_wanted) {
5025 vd->vdev_probe_wanted = B_FALSE;
5026 vdev_reopen(vd); /* vdev_open() does the actual probe */
5029 for (c = 0; c < vd->vdev_children; c++)
5030 spa_async_probe(spa, vd->vdev_child[c]);
5034 spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5041 if (!spa->spa_autoexpand)
5044 for (c = 0; c < vd->vdev_children; c++) {
5045 vdev_t *cvd = vd->vdev_child[c];
5046 spa_async_autoexpand(spa, cvd);
5049 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5052 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
5053 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
5055 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5056 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
5058 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
5059 ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
5062 kmem_free(physpath, MAXPATHLEN);
5066 spa_async_thread(spa_t *spa)
5070 ASSERT(spa->spa_sync_on);
5072 mutex_enter(&spa->spa_async_lock);
5073 tasks = spa->spa_async_tasks;
5074 spa->spa_async_tasks = 0;
5075 mutex_exit(&spa->spa_async_lock);
5078 * See if the config needs to be updated.
5080 if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5081 uint64_t old_space, new_space;
5083 mutex_enter(&spa_namespace_lock);
5084 old_space = metaslab_class_get_space(spa_normal_class(spa));
5085 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5086 new_space = metaslab_class_get_space(spa_normal_class(spa));
5087 mutex_exit(&spa_namespace_lock);
5090 * If the pool grew as a result of the config update,
5091 * then log an internal history event.
5093 if (new_space != old_space) {
5094 spa_history_log_internal(LOG_POOL_VDEV_ONLINE,
5096 "pool '%s' size: %llu(+%llu)",
5097 spa_name(spa), new_space, new_space - old_space);
5102 * See if any devices need to be marked REMOVED.
5104 if (tasks & SPA_ASYNC_REMOVE) {
5105 spa_vdev_state_enter(spa, SCL_NONE);
5106 spa_async_remove(spa, spa->spa_root_vdev);
5107 for (i = 0; i < spa->spa_l2cache.sav_count; i++)
5108 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5109 for (i = 0; i < spa->spa_spares.sav_count; i++)
5110 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5111 (void) spa_vdev_state_exit(spa, NULL, 0);
5114 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5115 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5116 spa_async_autoexpand(spa, spa->spa_root_vdev);
5117 spa_config_exit(spa, SCL_CONFIG, FTAG);
5121 * See if any devices need to be probed.
5123 if (tasks & SPA_ASYNC_PROBE) {
5124 spa_vdev_state_enter(spa, SCL_NONE);
5125 spa_async_probe(spa, spa->spa_root_vdev);
5126 (void) spa_vdev_state_exit(spa, NULL, 0);
5130 * If any devices are done replacing, detach them.
5132 if (tasks & SPA_ASYNC_RESILVER_DONE)
5133 spa_vdev_resilver_done(spa);
5136 * Kick off a resilver.
5138 if (tasks & SPA_ASYNC_RESILVER)
5139 dsl_resilver_restart(spa->spa_dsl_pool, 0);
5142 * Let the world know that we're done.
5144 mutex_enter(&spa->spa_async_lock);
5145 spa->spa_async_thread = NULL;
5146 cv_broadcast(&spa->spa_async_cv);
5147 mutex_exit(&spa->spa_async_lock);
5152 spa_async_suspend(spa_t *spa)
5154 mutex_enter(&spa->spa_async_lock);
5155 spa->spa_async_suspended++;
5156 while (spa->spa_async_thread != NULL)
5157 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5158 mutex_exit(&spa->spa_async_lock);
5162 spa_async_resume(spa_t *spa)
5164 mutex_enter(&spa->spa_async_lock);
5165 ASSERT(spa->spa_async_suspended != 0);
5166 spa->spa_async_suspended--;
5167 mutex_exit(&spa->spa_async_lock);
5171 spa_async_dispatch(spa_t *spa)
5173 mutex_enter(&spa->spa_async_lock);
5174 if (spa->spa_async_tasks && !spa->spa_async_suspended &&
5175 spa->spa_async_thread == NULL &&
5176 rootdir != NULL && !vn_is_readonly(rootdir))
5177 spa->spa_async_thread = thread_create(NULL, 0,
5178 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
5179 mutex_exit(&spa->spa_async_lock);
5183 spa_async_request(spa_t *spa, int task)
5185 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
5186 mutex_enter(&spa->spa_async_lock);
5187 spa->spa_async_tasks |= task;
5188 mutex_exit(&spa->spa_async_lock);
5192 * ==========================================================================
5193 * SPA syncing routines
5194 * ==========================================================================
5198 bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5201 bpobj_enqueue(bpo, bp, tx);
5206 spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5210 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
5216 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
5218 char *packed = NULL;
5223 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
5226 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5227 * information. This avoids the dbuf_will_dirty() path and
5228 * saves us a pre-read to get data we don't actually care about.
5230 bufsize = P2ROUNDUP(nvsize, SPA_CONFIG_BLOCKSIZE);
5231 packed = kmem_alloc(bufsize, KM_SLEEP);
5233 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
5235 bzero(packed + nvsize, bufsize - nvsize);
5237 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
5239 kmem_free(packed, bufsize);
5241 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
5242 dmu_buf_will_dirty(db, tx);
5243 *(uint64_t *)db->db_data = nvsize;
5244 dmu_buf_rele(db, FTAG);
5248 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
5249 const char *config, const char *entry)
5259 * Update the MOS nvlist describing the list of available devices.
5260 * spa_validate_aux() will have already made sure this nvlist is
5261 * valid and the vdevs are labeled appropriately.
5263 if (sav->sav_object == 0) {
5264 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
5265 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
5266 sizeof (uint64_t), tx);
5267 VERIFY(zap_update(spa->spa_meta_objset,
5268 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
5269 &sav->sav_object, tx) == 0);
5272 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5273 if (sav->sav_count == 0) {
5274 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
5276 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
5277 for (i = 0; i < sav->sav_count; i++)
5278 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
5279 B_FALSE, VDEV_CONFIG_L2CACHE);
5280 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
5281 sav->sav_count) == 0);
5282 for (i = 0; i < sav->sav_count; i++)
5283 nvlist_free(list[i]);
5284 kmem_free(list, sav->sav_count * sizeof (void *));
5287 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
5288 nvlist_free(nvroot);
5290 sav->sav_sync = B_FALSE;
5294 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
5298 if (list_is_empty(&spa->spa_config_dirty_list))
5301 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5303 config = spa_config_generate(spa, spa->spa_root_vdev,
5304 dmu_tx_get_txg(tx), B_FALSE);
5306 spa_config_exit(spa, SCL_STATE, FTAG);
5308 if (spa->spa_config_syncing)
5309 nvlist_free(spa->spa_config_syncing);
5310 spa->spa_config_syncing = config;
5312 spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
5316 * Set zpool properties.
5319 spa_sync_props(void *arg1, void *arg2, dmu_tx_t *tx)
5322 objset_t *mos = spa->spa_meta_objset;
5323 nvlist_t *nvp = arg2;
5328 const char *propname;
5329 zprop_type_t proptype;
5331 mutex_enter(&spa->spa_props_lock);
5334 while ((elem = nvlist_next_nvpair(nvp, elem))) {
5335 switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
5336 case ZPOOL_PROP_VERSION:
5338 * Only set version for non-zpool-creation cases
5339 * (set/import). spa_create() needs special care
5340 * for version setting.
5342 if (tx->tx_txg != TXG_INITIAL) {
5343 VERIFY(nvpair_value_uint64(elem,
5345 ASSERT(intval <= SPA_VERSION);
5346 ASSERT(intval >= spa_version(spa));
5347 spa->spa_uberblock.ub_version = intval;
5348 vdev_config_dirty(spa->spa_root_vdev);
5352 case ZPOOL_PROP_ALTROOT:
5354 * 'altroot' is a non-persistent property. It should
5355 * have been set temporarily at creation or import time.
5357 ASSERT(spa->spa_root != NULL);
5360 case ZPOOL_PROP_READONLY:
5361 case ZPOOL_PROP_CACHEFILE:
5363 * 'readonly' and 'cachefile' are also non-persisitent
5369 * Set pool property values in the poolprops mos object.
5371 if (spa->spa_pool_props_object == 0) {
5372 VERIFY((spa->spa_pool_props_object =
5373 zap_create(mos, DMU_OT_POOL_PROPS,
5374 DMU_OT_NONE, 0, tx)) > 0);
5376 VERIFY(zap_update(mos,
5377 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
5378 8, 1, &spa->spa_pool_props_object, tx)
5382 /* normalize the property name */
5383 propname = zpool_prop_to_name(prop);
5384 proptype = zpool_prop_get_type(prop);
5386 if (nvpair_type(elem) == DATA_TYPE_STRING) {
5387 ASSERT(proptype == PROP_TYPE_STRING);
5388 VERIFY(nvpair_value_string(elem, &strval) == 0);
5389 VERIFY(zap_update(mos,
5390 spa->spa_pool_props_object, propname,
5391 1, strlen(strval) + 1, strval, tx) == 0);
5393 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
5394 VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5396 if (proptype == PROP_TYPE_INDEX) {
5398 VERIFY(zpool_prop_index_to_string(
5399 prop, intval, &unused) == 0);
5401 VERIFY(zap_update(mos,
5402 spa->spa_pool_props_object, propname,
5403 8, 1, &intval, tx) == 0);
5405 ASSERT(0); /* not allowed */
5409 case ZPOOL_PROP_DELEGATION:
5410 spa->spa_delegation = intval;
5412 case ZPOOL_PROP_BOOTFS:
5413 spa->spa_bootfs = intval;
5415 case ZPOOL_PROP_FAILUREMODE:
5416 spa->spa_failmode = intval;
5418 case ZPOOL_PROP_AUTOEXPAND:
5419 spa->spa_autoexpand = intval;
5420 if (tx->tx_txg != TXG_INITIAL)
5421 spa_async_request(spa,
5422 SPA_ASYNC_AUTOEXPAND);
5424 case ZPOOL_PROP_DEDUPDITTO:
5425 spa->spa_dedup_ditto = intval;
5432 /* log internal history if this is not a zpool create */
5433 if (spa_version(spa) >= SPA_VERSION_ZPOOL_HISTORY &&
5434 tx->tx_txg != TXG_INITIAL) {
5435 spa_history_log_internal(LOG_POOL_PROPSET,
5436 spa, tx, "%s %lld %s",
5437 nvpair_name(elem), intval, spa_name(spa));
5441 mutex_exit(&spa->spa_props_lock);
5445 * Perform one-time upgrade on-disk changes. spa_version() does not
5446 * reflect the new version this txg, so there must be no changes this
5447 * txg to anything that the upgrade code depends on after it executes.
5448 * Therefore this must be called after dsl_pool_sync() does the sync
5452 spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
5454 dsl_pool_t *dp = spa->spa_dsl_pool;
5456 ASSERT(spa->spa_sync_pass == 1);
5458 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
5459 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
5460 dsl_pool_create_origin(dp, tx);
5462 /* Keeping the origin open increases spa_minref */
5463 spa->spa_minref += 3;
5466 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
5467 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
5468 dsl_pool_upgrade_clones(dp, tx);
5471 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
5472 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
5473 dsl_pool_upgrade_dir_clones(dp, tx);
5475 /* Keeping the freedir open increases spa_minref */
5476 spa->spa_minref += 3;
5481 * Sync the specified transaction group. New blocks may be dirtied as
5482 * part of the process, so we iterate until it converges.
5485 spa_sync(spa_t *spa, uint64_t txg)
5487 dsl_pool_t *dp = spa->spa_dsl_pool;
5488 objset_t *mos = spa->spa_meta_objset;
5489 bpobj_t *defer_bpo = &spa->spa_deferred_bpobj;
5490 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
5491 vdev_t *rvd = spa->spa_root_vdev;
5497 VERIFY(spa_writeable(spa));
5500 * Lock out configuration changes.
5502 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5504 spa->spa_syncing_txg = txg;
5505 spa->spa_sync_pass = 0;
5508 * If there are any pending vdev state changes, convert them
5509 * into config changes that go out with this transaction group.
5511 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5512 while (list_head(&spa->spa_state_dirty_list) != NULL) {
5514 * We need the write lock here because, for aux vdevs,
5515 * calling vdev_config_dirty() modifies sav_config.
5516 * This is ugly and will become unnecessary when we
5517 * eliminate the aux vdev wart by integrating all vdevs
5518 * into the root vdev tree.
5520 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
5521 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
5522 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
5523 vdev_state_clean(vd);
5524 vdev_config_dirty(vd);
5526 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
5527 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
5529 spa_config_exit(spa, SCL_STATE, FTAG);
5531 tx = dmu_tx_create_assigned(dp, txg);
5534 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
5535 * set spa_deflate if we have no raid-z vdevs.
5537 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
5538 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
5541 for (i = 0; i < rvd->vdev_children; i++) {
5542 vd = rvd->vdev_child[i];
5543 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
5546 if (i == rvd->vdev_children) {
5547 spa->spa_deflate = TRUE;
5548 VERIFY(0 == zap_add(spa->spa_meta_objset,
5549 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
5550 sizeof (uint64_t), 1, &spa->spa_deflate, tx));
5555 * If anything has changed in this txg, or if someone is waiting
5556 * for this txg to sync (eg, spa_vdev_remove()), push the
5557 * deferred frees from the previous txg. If not, leave them
5558 * alone so that we don't generate work on an otherwise idle
5561 if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
5562 !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
5563 !txg_list_empty(&dp->dp_sync_tasks, txg) ||
5564 ((dsl_scan_active(dp->dp_scan) ||
5565 txg_sync_waiting(dp)) && !spa_shutting_down(spa))) {
5566 zio_t *zio = zio_root(spa, NULL, NULL, 0);
5567 VERIFY3U(bpobj_iterate(defer_bpo,
5568 spa_free_sync_cb, zio, tx), ==, 0);
5569 VERIFY3U(zio_wait(zio), ==, 0);
5573 * Iterate to convergence.
5576 int pass = ++spa->spa_sync_pass;
5578 spa_sync_config_object(spa, tx);
5579 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
5580 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
5581 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
5582 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
5583 spa_errlog_sync(spa, txg);
5584 dsl_pool_sync(dp, txg);
5586 if (pass <= SYNC_PASS_DEFERRED_FREE) {
5587 zio_t *zio = zio_root(spa, NULL, NULL, 0);
5588 bplist_iterate(free_bpl, spa_free_sync_cb,
5590 VERIFY(zio_wait(zio) == 0);
5592 bplist_iterate(free_bpl, bpobj_enqueue_cb,
5597 dsl_scan_sync(dp, tx);
5599 while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)))
5603 spa_sync_upgrades(spa, tx);
5605 } while (dmu_objset_is_dirty(mos, txg));
5608 * Rewrite the vdev configuration (which includes the uberblock)
5609 * to commit the transaction group.
5611 * If there are no dirty vdevs, we sync the uberblock to a few
5612 * random top-level vdevs that are known to be visible in the
5613 * config cache (see spa_vdev_add() for a complete description).
5614 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
5618 * We hold SCL_STATE to prevent vdev open/close/etc.
5619 * while we're attempting to write the vdev labels.
5621 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5623 if (list_is_empty(&spa->spa_config_dirty_list)) {
5624 vdev_t *svd[SPA_DVAS_PER_BP];
5626 int children = rvd->vdev_children;
5627 int c0 = spa_get_random(children);
5629 for (c = 0; c < children; c++) {
5630 vd = rvd->vdev_child[(c0 + c) % children];
5631 if (vd->vdev_ms_array == 0 || vd->vdev_islog)
5633 svd[svdcount++] = vd;
5634 if (svdcount == SPA_DVAS_PER_BP)
5637 error = vdev_config_sync(svd, svdcount, txg, B_FALSE);
5639 error = vdev_config_sync(svd, svdcount, txg,
5642 error = vdev_config_sync(rvd->vdev_child,
5643 rvd->vdev_children, txg, B_FALSE);
5645 error = vdev_config_sync(rvd->vdev_child,
5646 rvd->vdev_children, txg, B_TRUE);
5649 spa_config_exit(spa, SCL_STATE, FTAG);
5653 zio_suspend(spa, NULL);
5654 zio_resume_wait(spa);
5659 * Clear the dirty config list.
5661 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
5662 vdev_config_clean(vd);
5665 * Now that the new config has synced transactionally,
5666 * let it become visible to the config cache.
5668 if (spa->spa_config_syncing != NULL) {
5669 spa_config_set(spa, spa->spa_config_syncing);
5670 spa->spa_config_txg = txg;
5671 spa->spa_config_syncing = NULL;
5674 spa->spa_ubsync = spa->spa_uberblock;
5676 dsl_pool_sync_done(dp, txg);
5679 * Update usable space statistics.
5681 while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg))))
5682 vdev_sync_done(vd, txg);
5684 spa_update_dspace(spa);
5687 * It had better be the case that we didn't dirty anything
5688 * since vdev_config_sync().
5690 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
5691 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
5692 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
5694 spa->spa_sync_pass = 0;
5696 spa_config_exit(spa, SCL_CONFIG, FTAG);
5698 spa_handle_ignored_writes(spa);
5701 * If any async tasks have been requested, kick them off.
5703 spa_async_dispatch(spa);
5707 * Sync all pools. We don't want to hold the namespace lock across these
5708 * operations, so we take a reference on the spa_t and drop the lock during the
5712 spa_sync_allpools(void)
5715 mutex_enter(&spa_namespace_lock);
5716 while ((spa = spa_next(spa)) != NULL) {
5717 if (spa_state(spa) != POOL_STATE_ACTIVE ||
5718 !spa_writeable(spa) || spa_suspended(spa))
5720 spa_open_ref(spa, FTAG);
5721 mutex_exit(&spa_namespace_lock);
5722 txg_wait_synced(spa_get_dsl(spa), 0);
5723 mutex_enter(&spa_namespace_lock);
5724 spa_close(spa, FTAG);
5726 mutex_exit(&spa_namespace_lock);
5730 * ==========================================================================
5731 * Miscellaneous routines
5732 * ==========================================================================
5736 * Remove all pools in the system.
5744 * Remove all cached state. All pools should be closed now,
5745 * so every spa in the AVL tree should be unreferenced.
5747 mutex_enter(&spa_namespace_lock);
5748 while ((spa = spa_next(NULL)) != NULL) {
5750 * Stop async tasks. The async thread may need to detach
5751 * a device that's been replaced, which requires grabbing
5752 * spa_namespace_lock, so we must drop it here.
5754 spa_open_ref(spa, FTAG);
5755 mutex_exit(&spa_namespace_lock);
5756 spa_async_suspend(spa);
5757 mutex_enter(&spa_namespace_lock);
5758 spa_close(spa, FTAG);
5760 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
5762 spa_deactivate(spa);
5766 mutex_exit(&spa_namespace_lock);
5770 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
5775 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
5779 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
5780 vd = spa->spa_l2cache.sav_vdevs[i];
5781 if (vd->vdev_guid == guid)
5785 for (i = 0; i < spa->spa_spares.sav_count; i++) {
5786 vd = spa->spa_spares.sav_vdevs[i];
5787 if (vd->vdev_guid == guid)
5796 spa_upgrade(spa_t *spa, uint64_t version)
5798 ASSERT(spa_writeable(spa));
5800 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5803 * This should only be called for a non-faulted pool, and since a
5804 * future version would result in an unopenable pool, this shouldn't be
5807 ASSERT(spa->spa_uberblock.ub_version <= SPA_VERSION);
5808 ASSERT(version >= spa->spa_uberblock.ub_version);
5810 spa->spa_uberblock.ub_version = version;
5811 vdev_config_dirty(spa->spa_root_vdev);
5813 spa_config_exit(spa, SCL_ALL, FTAG);
5815 txg_wait_synced(spa_get_dsl(spa), 0);
5819 spa_has_spare(spa_t *spa, uint64_t guid)
5823 spa_aux_vdev_t *sav = &spa->spa_spares;
5825 for (i = 0; i < sav->sav_count; i++)
5826 if (sav->sav_vdevs[i]->vdev_guid == guid)
5829 for (i = 0; i < sav->sav_npending; i++) {
5830 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
5831 &spareguid) == 0 && spareguid == guid)
5839 * Check if a pool has an active shared spare device.
5840 * Note: reference count of an active spare is 2, as a spare and as a replace
5843 spa_has_active_shared_spare(spa_t *spa)
5847 spa_aux_vdev_t *sav = &spa->spa_spares;
5849 for (i = 0; i < sav->sav_count; i++) {
5850 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
5851 &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
5860 * Post a sysevent corresponding to the given event. The 'name' must be one of
5861 * the event definitions in sys/sysevent/eventdefs.h. The payload will be
5862 * filled in from the spa and (optionally) the vdev. This doesn't do anything
5863 * in the userland libzpool, as we don't want consumers to misinterpret ztest
5864 * or zdb as real changes.
5867 spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
5871 sysevent_attr_list_t *attr = NULL;
5872 sysevent_value_t value;
5875 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
5878 value.value_type = SE_DATA_TYPE_STRING;
5879 value.value.sv_string = spa_name(spa);
5880 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
5883 value.value_type = SE_DATA_TYPE_UINT64;
5884 value.value.sv_uint64 = spa_guid(spa);
5885 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
5889 value.value_type = SE_DATA_TYPE_UINT64;
5890 value.value.sv_uint64 = vd->vdev_guid;
5891 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
5895 if (vd->vdev_path) {
5896 value.value_type = SE_DATA_TYPE_STRING;
5897 value.value.sv_string = vd->vdev_path;
5898 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
5899 &value, SE_SLEEP) != 0)
5904 if (sysevent_attach_attributes(ev, attr) != 0)
5908 (void) log_sysevent(ev, SE_SLEEP, &eid);
5912 sysevent_free_attr(attr);