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 inline 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_ereport_post(FM_EREPORT_RESOURCE_AUTOREPLACE,
1297 vd->vdev_spa, vd, NULL, 0, 0);
1298 spa_event_notify(vd->vdev_spa, vd, FM_EREPORT_ZFS_DEVICE_CHECK);
1303 * Validate the current config against the MOS config
1306 spa_config_valid(spa_t *spa, nvlist_t *config)
1308 vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1312 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1314 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1315 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1317 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1320 * If we're doing a normal import, then build up any additional
1321 * diagnostic information about missing devices in this config.
1322 * We'll pass this up to the user for further processing.
1324 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1325 nvlist_t **child, *nv;
1328 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1330 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1332 for (c = 0; c < rvd->vdev_children; c++) {
1333 vdev_t *tvd = rvd->vdev_child[c];
1334 vdev_t *mtvd = mrvd->vdev_child[c];
1336 if (tvd->vdev_ops == &vdev_missing_ops &&
1337 mtvd->vdev_ops != &vdev_missing_ops &&
1339 child[idx++] = vdev_config_generate(spa, mtvd,
1344 VERIFY(nvlist_add_nvlist_array(nv,
1345 ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1346 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1347 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1349 for (i = 0; i < idx; i++)
1350 nvlist_free(child[i]);
1353 kmem_free(child, rvd->vdev_children * sizeof (char **));
1357 * Compare the root vdev tree with the information we have
1358 * from the MOS config (mrvd). Check each top-level vdev
1359 * with the corresponding MOS config top-level (mtvd).
1361 for (c = 0; c < rvd->vdev_children; c++) {
1362 vdev_t *tvd = rvd->vdev_child[c];
1363 vdev_t *mtvd = mrvd->vdev_child[c];
1366 * Resolve any "missing" vdevs in the current configuration.
1367 * If we find that the MOS config has more accurate information
1368 * about the top-level vdev then use that vdev instead.
1370 if (tvd->vdev_ops == &vdev_missing_ops &&
1371 mtvd->vdev_ops != &vdev_missing_ops) {
1373 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1377 * Device specific actions.
1379 if (mtvd->vdev_islog) {
1380 spa_set_log_state(spa, SPA_LOG_CLEAR);
1383 * XXX - once we have 'readonly' pool
1384 * support we should be able to handle
1385 * missing data devices by transitioning
1386 * the pool to readonly.
1392 * Swap the missing vdev with the data we were
1393 * able to obtain from the MOS config.
1395 vdev_remove_child(rvd, tvd);
1396 vdev_remove_child(mrvd, mtvd);
1398 vdev_add_child(rvd, mtvd);
1399 vdev_add_child(mrvd, tvd);
1401 spa_config_exit(spa, SCL_ALL, FTAG);
1403 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1406 } else if (mtvd->vdev_islog) {
1408 * Load the slog device's state from the MOS config
1409 * since it's possible that the label does not
1410 * contain the most up-to-date information.
1412 vdev_load_log_state(tvd, mtvd);
1417 spa_config_exit(spa, SCL_ALL, FTAG);
1420 * Ensure we were able to validate the config.
1422 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1426 * Check for missing log devices
1429 spa_check_logs(spa_t *spa)
1431 switch (spa->spa_log_state) {
1434 case SPA_LOG_MISSING:
1435 /* need to recheck in case slog has been restored */
1436 case SPA_LOG_UNKNOWN:
1437 if (dmu_objset_find(spa->spa_name, zil_check_log_chain, NULL,
1438 DS_FIND_CHILDREN)) {
1439 spa_set_log_state(spa, SPA_LOG_MISSING);
1448 spa_passivate_log(spa_t *spa)
1450 vdev_t *rvd = spa->spa_root_vdev;
1451 boolean_t slog_found = B_FALSE;
1454 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1456 if (!spa_has_slogs(spa))
1459 for (c = 0; c < rvd->vdev_children; c++) {
1460 vdev_t *tvd = rvd->vdev_child[c];
1461 metaslab_group_t *mg = tvd->vdev_mg;
1463 if (tvd->vdev_islog) {
1464 metaslab_group_passivate(mg);
1465 slog_found = B_TRUE;
1469 return (slog_found);
1473 spa_activate_log(spa_t *spa)
1475 vdev_t *rvd = spa->spa_root_vdev;
1478 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1480 for (c = 0; c < rvd->vdev_children; c++) {
1481 vdev_t *tvd = rvd->vdev_child[c];
1482 metaslab_group_t *mg = tvd->vdev_mg;
1484 if (tvd->vdev_islog)
1485 metaslab_group_activate(mg);
1490 spa_offline_log(spa_t *spa)
1494 if ((error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1495 NULL, DS_FIND_CHILDREN)) == 0) {
1498 * We successfully offlined the log device, sync out the
1499 * current txg so that the "stubby" block can be removed
1502 txg_wait_synced(spa->spa_dsl_pool, 0);
1508 spa_aux_check_removed(spa_aux_vdev_t *sav)
1512 for (i = 0; i < sav->sav_count; i++)
1513 spa_check_removed(sav->sav_vdevs[i]);
1517 spa_claim_notify(zio_t *zio)
1519 spa_t *spa = zio->io_spa;
1524 mutex_enter(&spa->spa_props_lock); /* any mutex will do */
1525 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1526 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1527 mutex_exit(&spa->spa_props_lock);
1530 typedef struct spa_load_error {
1531 uint64_t sle_meta_count;
1532 uint64_t sle_data_count;
1536 spa_load_verify_done(zio_t *zio)
1538 blkptr_t *bp = zio->io_bp;
1539 spa_load_error_t *sle = zio->io_private;
1540 dmu_object_type_t type = BP_GET_TYPE(bp);
1541 int error = zio->io_error;
1544 if ((BP_GET_LEVEL(bp) != 0 || dmu_ot[type].ot_metadata) &&
1545 type != DMU_OT_INTENT_LOG)
1546 atomic_add_64(&sle->sle_meta_count, 1);
1548 atomic_add_64(&sle->sle_data_count, 1);
1550 zio_data_buf_free(zio->io_data, zio->io_size);
1555 spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1556 arc_buf_t *pbuf, const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1560 size_t size = BP_GET_PSIZE(bp);
1561 void *data = zio_data_buf_alloc(size);
1563 zio_nowait(zio_read(rio, spa, bp, data, size,
1564 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1565 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1566 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1572 spa_load_verify(spa_t *spa)
1575 spa_load_error_t sle = { 0 };
1576 zpool_rewind_policy_t policy;
1577 boolean_t verify_ok = B_FALSE;
1580 zpool_get_rewind_policy(spa->spa_config, &policy);
1582 if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1585 rio = zio_root(spa, NULL, &sle,
1586 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1588 error = traverse_pool(spa, spa->spa_verify_min_txg,
1589 TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio);
1591 (void) zio_wait(rio);
1593 spa->spa_load_meta_errors = sle.sle_meta_count;
1594 spa->spa_load_data_errors = sle.sle_data_count;
1596 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1597 sle.sle_data_count <= policy.zrp_maxdata) {
1601 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1602 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1604 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1605 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1606 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1607 VERIFY(nvlist_add_int64(spa->spa_load_info,
1608 ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1609 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1610 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1612 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1616 if (error != ENXIO && error != EIO)
1621 return (verify_ok ? 0 : EIO);
1625 * Find a value in the pool props object.
1628 spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1630 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1631 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1635 * Find a value in the pool directory object.
1638 spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1640 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1641 name, sizeof (uint64_t), 1, val));
1645 spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1647 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1652 * Fix up config after a partly-completed split. This is done with the
1653 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
1654 * pool have that entry in their config, but only the splitting one contains
1655 * a list of all the guids of the vdevs that are being split off.
1657 * This function determines what to do with that list: either rejoin
1658 * all the disks to the pool, or complete the splitting process. To attempt
1659 * the rejoin, each disk that is offlined is marked online again, and
1660 * we do a reopen() call. If the vdev label for every disk that was
1661 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1662 * then we call vdev_split() on each disk, and complete the split.
1664 * Otherwise we leave the config alone, with all the vdevs in place in
1665 * the original pool.
1668 spa_try_repair(spa_t *spa, nvlist_t *config)
1675 boolean_t attempt_reopen;
1677 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
1680 /* check that the config is complete */
1681 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
1682 &glist, &gcount) != 0)
1685 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
1687 /* attempt to online all the vdevs & validate */
1688 attempt_reopen = B_TRUE;
1689 for (i = 0; i < gcount; i++) {
1690 if (glist[i] == 0) /* vdev is hole */
1693 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
1694 if (vd[i] == NULL) {
1696 * Don't bother attempting to reopen the disks;
1697 * just do the split.
1699 attempt_reopen = B_FALSE;
1701 /* attempt to re-online it */
1702 vd[i]->vdev_offline = B_FALSE;
1706 if (attempt_reopen) {
1707 vdev_reopen(spa->spa_root_vdev);
1709 /* check each device to see what state it's in */
1710 for (extracted = 0, i = 0; i < gcount; i++) {
1711 if (vd[i] != NULL &&
1712 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
1719 * If every disk has been moved to the new pool, or if we never
1720 * even attempted to look at them, then we split them off for
1723 if (!attempt_reopen || gcount == extracted) {
1724 for (i = 0; i < gcount; i++)
1727 vdev_reopen(spa->spa_root_vdev);
1730 kmem_free(vd, gcount * sizeof (vdev_t *));
1734 spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
1735 boolean_t mosconfig)
1737 nvlist_t *config = spa->spa_config;
1738 char *ereport = FM_EREPORT_ZFS_POOL;
1743 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
1747 * Versioning wasn't explicitly added to the label until later, so if
1748 * it's not present treat it as the initial version.
1750 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
1751 &spa->spa_ubsync.ub_version) != 0)
1752 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
1754 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
1755 &spa->spa_config_txg);
1757 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
1758 spa_guid_exists(pool_guid, 0)) {
1761 spa->spa_load_guid = pool_guid;
1763 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
1765 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
1769 gethrestime(&spa->spa_loaded_ts);
1770 error = spa_load_impl(spa, pool_guid, config, state, type,
1771 mosconfig, &ereport);
1774 spa->spa_minref = refcount_count(&spa->spa_refcount);
1776 if (error != EEXIST) {
1777 spa->spa_loaded_ts.tv_sec = 0;
1778 spa->spa_loaded_ts.tv_nsec = 0;
1780 if (error != EBADF) {
1781 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
1784 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
1791 * Load an existing storage pool, using the pool's builtin spa_config as a
1792 * source of configuration information.
1794 __attribute__((always_inline))
1796 spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
1797 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
1801 nvlist_t *nvroot = NULL;
1803 uberblock_t *ub = &spa->spa_uberblock;
1804 uint64_t children, config_cache_txg = spa->spa_config_txg;
1805 int orig_mode = spa->spa_mode;
1810 * If this is an untrusted config, access the pool in read-only mode.
1811 * This prevents things like resilvering recently removed devices.
1814 spa->spa_mode = FREAD;
1816 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1818 spa->spa_load_state = state;
1820 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
1823 parse = (type == SPA_IMPORT_EXISTING ?
1824 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
1827 * Create "The Godfather" zio to hold all async IOs
1829 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
1830 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
1833 * Parse the configuration into a vdev tree. We explicitly set the
1834 * value that will be returned by spa_version() since parsing the
1835 * configuration requires knowing the version number.
1837 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1838 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
1839 spa_config_exit(spa, SCL_ALL, FTAG);
1844 ASSERT(spa->spa_root_vdev == rvd);
1846 if (type != SPA_IMPORT_ASSEMBLE) {
1847 ASSERT(spa_guid(spa) == pool_guid);
1851 * Try to open all vdevs, loading each label in the process.
1853 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1854 error = vdev_open(rvd);
1855 spa_config_exit(spa, SCL_ALL, FTAG);
1860 * We need to validate the vdev labels against the configuration that
1861 * we have in hand, which is dependent on the setting of mosconfig. If
1862 * mosconfig is true then we're validating the vdev labels based on
1863 * that config. Otherwise, we're validating against the cached config
1864 * (zpool.cache) that was read when we loaded the zfs module, and then
1865 * later we will recursively call spa_load() and validate against
1868 * If we're assembling a new pool that's been split off from an
1869 * existing pool, the labels haven't yet been updated so we skip
1870 * validation for now.
1872 if (type != SPA_IMPORT_ASSEMBLE) {
1873 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1874 error = vdev_validate(rvd);
1875 spa_config_exit(spa, SCL_ALL, FTAG);
1880 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
1885 * Find the best uberblock.
1887 vdev_uberblock_load(NULL, rvd, ub);
1890 * If we weren't able to find a single valid uberblock, return failure.
1892 if (ub->ub_txg == 0)
1893 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
1896 * If the pool is newer than the code, we can't open it.
1898 if (ub->ub_version > SPA_VERSION)
1899 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
1902 * If the vdev guid sum doesn't match the uberblock, we have an
1903 * incomplete configuration. We first check to see if the pool
1904 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
1905 * If it is, defer the vdev_guid_sum check till later so we
1906 * can handle missing vdevs.
1908 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
1909 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
1910 rvd->vdev_guid_sum != ub->ub_guid_sum)
1911 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
1913 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
1914 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1915 spa_try_repair(spa, config);
1916 spa_config_exit(spa, SCL_ALL, FTAG);
1917 nvlist_free(spa->spa_config_splitting);
1918 spa->spa_config_splitting = NULL;
1922 * Initialize internal SPA structures.
1924 spa->spa_state = POOL_STATE_ACTIVE;
1925 spa->spa_ubsync = spa->spa_uberblock;
1926 spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
1927 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
1928 spa->spa_first_txg = spa->spa_last_ubsync_txg ?
1929 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
1930 spa->spa_claim_max_txg = spa->spa_first_txg;
1931 spa->spa_prev_software_version = ub->ub_software_version;
1933 error = dsl_pool_open(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
1935 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1936 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
1938 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
1939 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1943 nvlist_t *policy = NULL, *nvconfig;
1945 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
1946 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1948 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
1949 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
1951 unsigned long myhostid = 0;
1953 VERIFY(nvlist_lookup_string(nvconfig,
1954 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
1957 myhostid = zone_get_hostid(NULL);
1960 * We're emulating the system's hostid in userland, so
1961 * we can't use zone_get_hostid().
1963 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
1964 #endif /* _KERNEL */
1965 if (hostid != 0 && myhostid != 0 &&
1966 hostid != myhostid) {
1967 nvlist_free(nvconfig);
1968 cmn_err(CE_WARN, "pool '%s' could not be "
1969 "loaded as it was last accessed by "
1970 "another system (host: %s hostid: 0x%lx). "
1971 "See: http://www.sun.com/msg/ZFS-8000-EY",
1972 spa_name(spa), hostname,
1973 (unsigned long)hostid);
1977 if (nvlist_lookup_nvlist(spa->spa_config,
1978 ZPOOL_REWIND_POLICY, &policy) == 0)
1979 VERIFY(nvlist_add_nvlist(nvconfig,
1980 ZPOOL_REWIND_POLICY, policy) == 0);
1982 spa_config_set(spa, nvconfig);
1984 spa_deactivate(spa);
1985 spa_activate(spa, orig_mode);
1987 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
1990 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
1991 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1992 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
1994 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1997 * Load the bit that tells us to use the new accounting function
1998 * (raid-z deflation). If we have an older pool, this will not
2001 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2002 if (error != 0 && error != ENOENT)
2003 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2005 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2006 &spa->spa_creation_version);
2007 if (error != 0 && error != ENOENT)
2008 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2011 * Load the persistent error log. If we have an older pool, this will
2014 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2015 if (error != 0 && error != ENOENT)
2016 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2018 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2019 &spa->spa_errlog_scrub);
2020 if (error != 0 && error != ENOENT)
2021 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2024 * Load the history object. If we have an older pool, this
2025 * will not be present.
2027 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2028 if (error != 0 && error != ENOENT)
2029 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2032 * If we're assembling the pool from the split-off vdevs of
2033 * an existing pool, we don't want to attach the spares & cache
2038 * Load any hot spares for this pool.
2040 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2041 if (error != 0 && error != ENOENT)
2042 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2043 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2044 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2045 if (load_nvlist(spa, spa->spa_spares.sav_object,
2046 &spa->spa_spares.sav_config) != 0)
2047 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2049 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2050 spa_load_spares(spa);
2051 spa_config_exit(spa, SCL_ALL, FTAG);
2052 } else if (error == 0) {
2053 spa->spa_spares.sav_sync = B_TRUE;
2057 * Load any level 2 ARC devices for this pool.
2059 error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2060 &spa->spa_l2cache.sav_object);
2061 if (error != 0 && error != ENOENT)
2062 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2063 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2064 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2065 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2066 &spa->spa_l2cache.sav_config) != 0)
2067 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2069 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2070 spa_load_l2cache(spa);
2071 spa_config_exit(spa, SCL_ALL, FTAG);
2072 } else if (error == 0) {
2073 spa->spa_l2cache.sav_sync = B_TRUE;
2076 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2078 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2079 if (error && error != ENOENT)
2080 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2083 uint64_t autoreplace;
2085 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2086 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2087 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2088 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2089 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2090 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2091 &spa->spa_dedup_ditto);
2093 spa->spa_autoreplace = (autoreplace != 0);
2097 * If the 'autoreplace' property is set, then post a resource notifying
2098 * the ZFS DE that it should not issue any faults for unopenable
2099 * devices. We also iterate over the vdevs, and post a sysevent for any
2100 * unopenable vdevs so that the normal autoreplace handler can take
2103 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2104 spa_check_removed(spa->spa_root_vdev);
2106 * For the import case, this is done in spa_import(), because
2107 * at this point we're using the spare definitions from
2108 * the MOS config, not necessarily from the userland config.
2110 if (state != SPA_LOAD_IMPORT) {
2111 spa_aux_check_removed(&spa->spa_spares);
2112 spa_aux_check_removed(&spa->spa_l2cache);
2117 * Load the vdev state for all toplevel vdevs.
2122 * Propagate the leaf DTLs we just loaded all the way up the tree.
2124 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2125 vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2126 spa_config_exit(spa, SCL_ALL, FTAG);
2129 * Load the DDTs (dedup tables).
2131 error = ddt_load(spa);
2133 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2135 spa_update_dspace(spa);
2138 * Validate the config, using the MOS config to fill in any
2139 * information which might be missing. If we fail to validate
2140 * the config then declare the pool unfit for use. If we're
2141 * assembling a pool from a split, the log is not transferred
2144 if (type != SPA_IMPORT_ASSEMBLE) {
2147 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2148 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2150 if (!spa_config_valid(spa, nvconfig)) {
2151 nvlist_free(nvconfig);
2152 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2155 nvlist_free(nvconfig);
2158 * Now that we've validate the config, check the state of the
2159 * root vdev. If it can't be opened, it indicates one or
2160 * more toplevel vdevs are faulted.
2162 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2165 if (spa_check_logs(spa)) {
2166 *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2167 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2172 * We've successfully opened the pool, verify that we're ready
2173 * to start pushing transactions.
2175 if (state != SPA_LOAD_TRYIMPORT) {
2176 if ((error = spa_load_verify(spa)))
2177 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2181 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2182 spa->spa_load_max_txg == UINT64_MAX)) {
2184 int need_update = B_FALSE;
2187 ASSERT(state != SPA_LOAD_TRYIMPORT);
2190 * Claim log blocks that haven't been committed yet.
2191 * This must all happen in a single txg.
2192 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2193 * invoked from zil_claim_log_block()'s i/o done callback.
2194 * Price of rollback is that we abandon the log.
2196 spa->spa_claiming = B_TRUE;
2198 tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2199 spa_first_txg(spa));
2200 (void) dmu_objset_find(spa_name(spa),
2201 zil_claim, tx, DS_FIND_CHILDREN);
2204 spa->spa_claiming = B_FALSE;
2206 spa_set_log_state(spa, SPA_LOG_GOOD);
2207 spa->spa_sync_on = B_TRUE;
2208 txg_sync_start(spa->spa_dsl_pool);
2211 * Wait for all claims to sync. We sync up to the highest
2212 * claimed log block birth time so that claimed log blocks
2213 * don't appear to be from the future. spa_claim_max_txg
2214 * will have been set for us by either zil_check_log_chain()
2215 * (invoked from spa_check_logs()) or zil_claim() above.
2217 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2220 * If the config cache is stale, or we have uninitialized
2221 * metaslabs (see spa_vdev_add()), then update the config.
2223 * If this is a verbatim import, trust the current
2224 * in-core spa_config and update the disk labels.
2226 if (config_cache_txg != spa->spa_config_txg ||
2227 state == SPA_LOAD_IMPORT ||
2228 state == SPA_LOAD_RECOVER ||
2229 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2230 need_update = B_TRUE;
2232 for (c = 0; c < rvd->vdev_children; c++)
2233 if (rvd->vdev_child[c]->vdev_ms_array == 0)
2234 need_update = B_TRUE;
2237 * Update the config cache asychronously in case we're the
2238 * root pool, in which case the config cache isn't writable yet.
2241 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2244 * Check all DTLs to see if anything needs resilvering.
2246 if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2247 vdev_resilver_needed(rvd, NULL, NULL))
2248 spa_async_request(spa, SPA_ASYNC_RESILVER);
2251 * Delete any inconsistent datasets.
2253 (void) dmu_objset_find(spa_name(spa),
2254 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2257 * Clean up any stale temporary dataset userrefs.
2259 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2266 spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2268 int mode = spa->spa_mode;
2271 spa_deactivate(spa);
2273 spa->spa_load_max_txg--;
2275 spa_activate(spa, mode);
2276 spa_async_suspend(spa);
2278 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2282 spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2283 uint64_t max_request, int rewind_flags)
2285 nvlist_t *config = NULL;
2286 int load_error, rewind_error;
2287 uint64_t safe_rewind_txg;
2290 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2291 spa->spa_load_max_txg = spa->spa_load_txg;
2292 spa_set_log_state(spa, SPA_LOG_CLEAR);
2294 spa->spa_load_max_txg = max_request;
2297 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2299 if (load_error == 0)
2302 if (spa->spa_root_vdev != NULL)
2303 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2305 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2306 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2308 if (rewind_flags & ZPOOL_NEVER_REWIND) {
2309 nvlist_free(config);
2310 return (load_error);
2313 /* Price of rolling back is discarding txgs, including log */
2314 if (state == SPA_LOAD_RECOVER)
2315 spa_set_log_state(spa, SPA_LOG_CLEAR);
2317 spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2318 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2319 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2320 TXG_INITIAL : safe_rewind_txg;
2323 * Continue as long as we're finding errors, we're still within
2324 * the acceptable rewind range, and we're still finding uberblocks
2326 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2327 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2328 if (spa->spa_load_max_txg < safe_rewind_txg)
2329 spa->spa_extreme_rewind = B_TRUE;
2330 rewind_error = spa_load_retry(spa, state, mosconfig);
2333 spa->spa_extreme_rewind = B_FALSE;
2334 spa->spa_load_max_txg = UINT64_MAX;
2336 if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2337 spa_config_set(spa, config);
2339 return (state == SPA_LOAD_RECOVER ? rewind_error : load_error);
2345 * The import case is identical to an open except that the configuration is sent
2346 * down from userland, instead of grabbed from the configuration cache. For the
2347 * case of an open, the pool configuration will exist in the
2348 * POOL_STATE_UNINITIALIZED state.
2350 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2351 * the same time open the pool, without having to keep around the spa_t in some
2355 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2359 spa_load_state_t state = SPA_LOAD_OPEN;
2361 int locked = B_FALSE;
2366 * As disgusting as this is, we need to support recursive calls to this
2367 * function because dsl_dir_open() is called during spa_load(), and ends
2368 * up calling spa_open() again. The real fix is to figure out how to
2369 * avoid dsl_dir_open() calling this in the first place.
2371 if (mutex_owner(&spa_namespace_lock) != curthread) {
2372 mutex_enter(&spa_namespace_lock);
2376 if ((spa = spa_lookup(pool)) == NULL) {
2378 mutex_exit(&spa_namespace_lock);
2382 if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2383 zpool_rewind_policy_t policy;
2385 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2387 if (policy.zrp_request & ZPOOL_DO_REWIND)
2388 state = SPA_LOAD_RECOVER;
2390 spa_activate(spa, spa_mode_global);
2392 if (state != SPA_LOAD_RECOVER)
2393 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2395 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2396 policy.zrp_request);
2398 if (error == EBADF) {
2400 * If vdev_validate() returns failure (indicated by
2401 * EBADF), it indicates that one of the vdevs indicates
2402 * that the pool has been exported or destroyed. If
2403 * this is the case, the config cache is out of sync and
2404 * we should remove the pool from the namespace.
2407 spa_deactivate(spa);
2408 spa_config_sync(spa, B_TRUE, B_TRUE);
2411 mutex_exit(&spa_namespace_lock);
2417 * We can't open the pool, but we still have useful
2418 * information: the state of each vdev after the
2419 * attempted vdev_open(). Return this to the user.
2421 if (config != NULL && spa->spa_config) {
2422 VERIFY(nvlist_dup(spa->spa_config, config,
2424 VERIFY(nvlist_add_nvlist(*config,
2425 ZPOOL_CONFIG_LOAD_INFO,
2426 spa->spa_load_info) == 0);
2429 spa_deactivate(spa);
2430 spa->spa_last_open_failed = error;
2432 mutex_exit(&spa_namespace_lock);
2438 spa_open_ref(spa, tag);
2441 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2444 * If we've recovered the pool, pass back any information we
2445 * gathered while doing the load.
2447 if (state == SPA_LOAD_RECOVER) {
2448 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
2449 spa->spa_load_info) == 0);
2453 spa->spa_last_open_failed = 0;
2454 spa->spa_last_ubsync_txg = 0;
2455 spa->spa_load_txg = 0;
2456 mutex_exit(&spa_namespace_lock);
2465 spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
2468 return (spa_open_common(name, spapp, tag, policy, config));
2472 spa_open(const char *name, spa_t **spapp, void *tag)
2474 return (spa_open_common(name, spapp, tag, NULL, NULL));
2478 * Lookup the given spa_t, incrementing the inject count in the process,
2479 * preventing it from being exported or destroyed.
2482 spa_inject_addref(char *name)
2486 mutex_enter(&spa_namespace_lock);
2487 if ((spa = spa_lookup(name)) == NULL) {
2488 mutex_exit(&spa_namespace_lock);
2491 spa->spa_inject_ref++;
2492 mutex_exit(&spa_namespace_lock);
2498 spa_inject_delref(spa_t *spa)
2500 mutex_enter(&spa_namespace_lock);
2501 spa->spa_inject_ref--;
2502 mutex_exit(&spa_namespace_lock);
2506 * Add spares device information to the nvlist.
2509 spa_add_spares(spa_t *spa, nvlist_t *config)
2519 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2521 if (spa->spa_spares.sav_count == 0)
2524 VERIFY(nvlist_lookup_nvlist(config,
2525 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2526 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
2527 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2529 VERIFY(nvlist_add_nvlist_array(nvroot,
2530 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2531 VERIFY(nvlist_lookup_nvlist_array(nvroot,
2532 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2535 * Go through and find any spares which have since been
2536 * repurposed as an active spare. If this is the case, update
2537 * their status appropriately.
2539 for (i = 0; i < nspares; i++) {
2540 VERIFY(nvlist_lookup_uint64(spares[i],
2541 ZPOOL_CONFIG_GUID, &guid) == 0);
2542 if (spa_spare_exists(guid, &pool, NULL) &&
2544 VERIFY(nvlist_lookup_uint64_array(
2545 spares[i], ZPOOL_CONFIG_VDEV_STATS,
2546 (uint64_t **)&vs, &vsc) == 0);
2547 vs->vs_state = VDEV_STATE_CANT_OPEN;
2548 vs->vs_aux = VDEV_AUX_SPARED;
2555 * Add l2cache device information to the nvlist, including vdev stats.
2558 spa_add_l2cache(spa_t *spa, nvlist_t *config)
2561 uint_t i, j, nl2cache;
2568 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2570 if (spa->spa_l2cache.sav_count == 0)
2573 VERIFY(nvlist_lookup_nvlist(config,
2574 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2575 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
2576 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2577 if (nl2cache != 0) {
2578 VERIFY(nvlist_add_nvlist_array(nvroot,
2579 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2580 VERIFY(nvlist_lookup_nvlist_array(nvroot,
2581 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2584 * Update level 2 cache device stats.
2587 for (i = 0; i < nl2cache; i++) {
2588 VERIFY(nvlist_lookup_uint64(l2cache[i],
2589 ZPOOL_CONFIG_GUID, &guid) == 0);
2592 for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
2594 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
2595 vd = spa->spa_l2cache.sav_vdevs[j];
2601 VERIFY(nvlist_lookup_uint64_array(l2cache[i],
2602 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
2604 vdev_get_stats(vd, vs);
2610 spa_get_stats(const char *name, nvlist_t **config, char *altroot, size_t buflen)
2616 error = spa_open_common(name, &spa, FTAG, NULL, config);
2620 * This still leaves a window of inconsistency where the spares
2621 * or l2cache devices could change and the config would be
2622 * self-inconsistent.
2624 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
2626 if (*config != NULL) {
2627 uint64_t loadtimes[2];
2629 loadtimes[0] = spa->spa_loaded_ts.tv_sec;
2630 loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
2631 VERIFY(nvlist_add_uint64_array(*config,
2632 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
2634 VERIFY(nvlist_add_uint64(*config,
2635 ZPOOL_CONFIG_ERRCOUNT,
2636 spa_get_errlog_size(spa)) == 0);
2638 if (spa_suspended(spa))
2639 VERIFY(nvlist_add_uint64(*config,
2640 ZPOOL_CONFIG_SUSPENDED,
2641 spa->spa_failmode) == 0);
2643 spa_add_spares(spa, *config);
2644 spa_add_l2cache(spa, *config);
2649 * We want to get the alternate root even for faulted pools, so we cheat
2650 * and call spa_lookup() directly.
2654 mutex_enter(&spa_namespace_lock);
2655 spa = spa_lookup(name);
2657 spa_altroot(spa, altroot, buflen);
2661 mutex_exit(&spa_namespace_lock);
2663 spa_altroot(spa, altroot, buflen);
2668 spa_config_exit(spa, SCL_CONFIG, FTAG);
2669 spa_close(spa, FTAG);
2676 * Validate that the auxiliary device array is well formed. We must have an
2677 * array of nvlists, each which describes a valid leaf vdev. If this is an
2678 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
2679 * specified, as long as they are well-formed.
2682 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
2683 spa_aux_vdev_t *sav, const char *config, uint64_t version,
2684 vdev_labeltype_t label)
2691 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
2694 * It's acceptable to have no devs specified.
2696 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
2703 * Make sure the pool is formatted with a version that supports this
2706 if (spa_version(spa) < version)
2710 * Set the pending device list so we correctly handle device in-use
2713 sav->sav_pending = dev;
2714 sav->sav_npending = ndev;
2716 for (i = 0; i < ndev; i++) {
2717 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
2721 if (!vd->vdev_ops->vdev_op_leaf) {
2728 * The L2ARC currently only supports disk devices in
2729 * kernel context. For user-level testing, we allow it.
2732 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
2733 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
2740 if ((error = vdev_open(vd)) == 0 &&
2741 (error = vdev_label_init(vd, crtxg, label)) == 0) {
2742 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
2743 vd->vdev_guid) == 0);
2749 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
2756 sav->sav_pending = NULL;
2757 sav->sav_npending = 0;
2762 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
2766 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
2768 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
2769 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
2770 VDEV_LABEL_SPARE)) != 0) {
2774 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
2775 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
2776 VDEV_LABEL_L2CACHE));
2780 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
2785 if (sav->sav_config != NULL) {
2791 * Generate new dev list by concatentating with the
2794 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
2795 &olddevs, &oldndevs) == 0);
2797 newdevs = kmem_alloc(sizeof (void *) *
2798 (ndevs + oldndevs), KM_SLEEP);
2799 for (i = 0; i < oldndevs; i++)
2800 VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
2802 for (i = 0; i < ndevs; i++)
2803 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
2806 VERIFY(nvlist_remove(sav->sav_config, config,
2807 DATA_TYPE_NVLIST_ARRAY) == 0);
2809 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
2810 config, newdevs, ndevs + oldndevs) == 0);
2811 for (i = 0; i < oldndevs + ndevs; i++)
2812 nvlist_free(newdevs[i]);
2813 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
2816 * Generate a new dev list.
2818 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
2820 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
2826 * Stop and drop level 2 ARC devices
2829 spa_l2cache_drop(spa_t *spa)
2833 spa_aux_vdev_t *sav = &spa->spa_l2cache;
2835 for (i = 0; i < sav->sav_count; i++) {
2838 vd = sav->sav_vdevs[i];
2841 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
2842 pool != 0ULL && l2arc_vdev_present(vd))
2843 l2arc_remove_vdev(vd);
2844 if (vd->vdev_isl2cache)
2845 spa_l2cache_remove(vd);
2846 vdev_clear_stats(vd);
2847 (void) vdev_close(vd);
2855 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
2856 const char *history_str, nvlist_t *zplprops)
2859 char *altroot = NULL;
2864 uint64_t txg = TXG_INITIAL;
2865 nvlist_t **spares, **l2cache;
2866 uint_t nspares, nl2cache;
2867 uint64_t version, obj;
2871 * If this pool already exists, return failure.
2873 mutex_enter(&spa_namespace_lock);
2874 if (spa_lookup(pool) != NULL) {
2875 mutex_exit(&spa_namespace_lock);
2880 * Allocate a new spa_t structure.
2882 (void) nvlist_lookup_string(props,
2883 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
2884 spa = spa_add(pool, NULL, altroot);
2885 spa_activate(spa, spa_mode_global);
2887 if (props && (error = spa_prop_validate(spa, props))) {
2888 spa_deactivate(spa);
2890 mutex_exit(&spa_namespace_lock);
2894 if (nvlist_lookup_uint64(props, zpool_prop_to_name(ZPOOL_PROP_VERSION),
2896 version = SPA_VERSION;
2897 ASSERT(version <= SPA_VERSION);
2899 spa->spa_first_txg = txg;
2900 spa->spa_uberblock.ub_txg = txg - 1;
2901 spa->spa_uberblock.ub_version = version;
2902 spa->spa_ubsync = spa->spa_uberblock;
2905 * Create "The Godfather" zio to hold all async IOs
2907 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2908 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2911 * Create the root vdev.
2913 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2915 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
2917 ASSERT(error != 0 || rvd != NULL);
2918 ASSERT(error != 0 || spa->spa_root_vdev == rvd);
2920 if (error == 0 && !zfs_allocatable_devs(nvroot))
2924 (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
2925 (error = spa_validate_aux(spa, nvroot, txg,
2926 VDEV_ALLOC_ADD)) == 0) {
2927 for (c = 0; c < rvd->vdev_children; c++) {
2928 vdev_metaslab_set_size(rvd->vdev_child[c]);
2929 vdev_expand(rvd->vdev_child[c], txg);
2933 spa_config_exit(spa, SCL_ALL, FTAG);
2937 spa_deactivate(spa);
2939 mutex_exit(&spa_namespace_lock);
2944 * Get the list of spares, if specified.
2946 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
2947 &spares, &nspares) == 0) {
2948 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
2950 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
2951 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2952 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2953 spa_load_spares(spa);
2954 spa_config_exit(spa, SCL_ALL, FTAG);
2955 spa->spa_spares.sav_sync = B_TRUE;
2959 * Get the list of level 2 cache devices, if specified.
2961 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
2962 &l2cache, &nl2cache) == 0) {
2963 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
2964 NV_UNIQUE_NAME, KM_SLEEP) == 0);
2965 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
2966 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2967 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2968 spa_load_l2cache(spa);
2969 spa_config_exit(spa, SCL_ALL, FTAG);
2970 spa->spa_l2cache.sav_sync = B_TRUE;
2973 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
2974 spa->spa_meta_objset = dp->dp_meta_objset;
2977 * Create DDTs (dedup tables).
2981 spa_update_dspace(spa);
2983 tx = dmu_tx_create_assigned(dp, txg);
2986 * Create the pool config object.
2988 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
2989 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
2990 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
2992 if (zap_add(spa->spa_meta_objset,
2993 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
2994 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
2995 cmn_err(CE_PANIC, "failed to add pool config");
2998 if (zap_add(spa->spa_meta_objset,
2999 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3000 sizeof (uint64_t), 1, &version, tx) != 0) {
3001 cmn_err(CE_PANIC, "failed to add pool version");
3004 /* Newly created pools with the right version are always deflated. */
3005 if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3006 spa->spa_deflate = TRUE;
3007 if (zap_add(spa->spa_meta_objset,
3008 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3009 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3010 cmn_err(CE_PANIC, "failed to add deflate");
3015 * Create the deferred-free bpobj. Turn off compression
3016 * because sync-to-convergence takes longer if the blocksize
3019 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3020 dmu_object_set_compress(spa->spa_meta_objset, obj,
3021 ZIO_COMPRESS_OFF, tx);
3022 if (zap_add(spa->spa_meta_objset,
3023 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3024 sizeof (uint64_t), 1, &obj, tx) != 0) {
3025 cmn_err(CE_PANIC, "failed to add bpobj");
3027 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3028 spa->spa_meta_objset, obj));
3031 * Create the pool's history object.
3033 if (version >= SPA_VERSION_ZPOOL_HISTORY)
3034 spa_history_create_obj(spa, tx);
3037 * Set pool properties.
3039 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3040 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3041 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3042 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3044 if (props != NULL) {
3045 spa_configfile_set(spa, props, B_FALSE);
3046 spa_sync_props(spa, props, tx);
3051 spa->spa_sync_on = B_TRUE;
3052 txg_sync_start(spa->spa_dsl_pool);
3055 * We explicitly wait for the first transaction to complete so that our
3056 * bean counters are appropriately updated.
3058 txg_wait_synced(spa->spa_dsl_pool, txg);
3060 spa_config_sync(spa, B_FALSE, B_TRUE);
3062 if (version >= SPA_VERSION_ZPOOL_HISTORY && history_str != NULL)
3063 (void) spa_history_log(spa, history_str, LOG_CMD_POOL_CREATE);
3064 spa_history_log_version(spa, LOG_POOL_CREATE);
3066 spa->spa_minref = refcount_count(&spa->spa_refcount);
3068 mutex_exit(&spa_namespace_lock);
3075 * Get the root pool information from the root disk, then import the root pool
3076 * during the system boot up time.
3078 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3081 spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3084 nvlist_t *nvtop, *nvroot;
3087 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3091 * Add this top-level vdev to the child array.
3093 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3095 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3097 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3100 * Put this pool's top-level vdevs into a root vdev.
3102 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3103 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3104 VDEV_TYPE_ROOT) == 0);
3105 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3106 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3107 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3111 * Replace the existing vdev_tree with the new root vdev in
3112 * this pool's configuration (remove the old, add the new).
3114 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3115 nvlist_free(nvroot);
3120 * Walk the vdev tree and see if we can find a device with "better"
3121 * configuration. A configuration is "better" if the label on that
3122 * device has a more recent txg.
3125 spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3129 for (c = 0; c < vd->vdev_children; c++)
3130 spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3132 if (vd->vdev_ops->vdev_op_leaf) {
3136 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3140 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3144 * Do we have a better boot device?
3146 if (label_txg > *txg) {
3155 * Import a root pool.
3157 * For x86. devpath_list will consist of devid and/or physpath name of
3158 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3159 * The GRUB "findroot" command will return the vdev we should boot.
3161 * For Sparc, devpath_list consists the physpath name of the booting device
3162 * no matter the rootpool is a single device pool or a mirrored pool.
3164 * "/pci@1f,0/ide@d/disk@0,0:a"
3167 spa_import_rootpool(char *devpath, char *devid)
3170 vdev_t *rvd, *bvd, *avd = NULL;
3171 nvlist_t *config, *nvtop;
3177 * Read the label from the boot device and generate a configuration.
3179 config = spa_generate_rootconf(devpath, devid, &guid);
3180 #if defined(_OBP) && defined(_KERNEL)
3181 if (config == NULL) {
3182 if (strstr(devpath, "/iscsi/ssd") != NULL) {
3184 get_iscsi_bootpath_phy(devpath);
3185 config = spa_generate_rootconf(devpath, devid, &guid);
3189 if (config == NULL) {
3190 cmn_err(CE_NOTE, "Can not read the pool label from '%s'",
3195 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3197 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3199 mutex_enter(&spa_namespace_lock);
3200 if ((spa = spa_lookup(pname)) != NULL) {
3202 * Remove the existing root pool from the namespace so that we
3203 * can replace it with the correct config we just read in.
3208 spa = spa_add(pname, config, NULL);
3209 spa->spa_is_root = B_TRUE;
3210 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3213 * Build up a vdev tree based on the boot device's label config.
3215 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3217 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3218 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3219 VDEV_ALLOC_ROOTPOOL);
3220 spa_config_exit(spa, SCL_ALL, FTAG);
3222 mutex_exit(&spa_namespace_lock);
3223 nvlist_free(config);
3224 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3230 * Get the boot vdev.
3232 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3233 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3234 (u_longlong_t)guid);
3240 * Determine if there is a better boot device.
3243 spa_alt_rootvdev(rvd, &avd, &txg);
3245 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3246 "try booting from '%s'", avd->vdev_path);
3252 * If the boot device is part of a spare vdev then ensure that
3253 * we're booting off the active spare.
3255 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3256 !bvd->vdev_isspare) {
3257 cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3258 "try booting from '%s'",
3260 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3266 spa_history_log_version(spa, LOG_POOL_IMPORT);
3268 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3270 spa_config_exit(spa, SCL_ALL, FTAG);
3271 mutex_exit(&spa_namespace_lock);
3273 nvlist_free(config);
3280 * Import a non-root pool into the system.
3283 spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
3286 char *altroot = NULL;
3287 spa_load_state_t state = SPA_LOAD_IMPORT;
3288 zpool_rewind_policy_t policy;
3289 uint64_t mode = spa_mode_global;
3290 uint64_t readonly = B_FALSE;
3293 nvlist_t **spares, **l2cache;
3294 uint_t nspares, nl2cache;
3297 * If a pool with this name exists, return failure.
3299 mutex_enter(&spa_namespace_lock);
3300 if (spa_lookup(pool) != NULL) {
3301 mutex_exit(&spa_namespace_lock);
3306 * Create and initialize the spa structure.
3308 (void) nvlist_lookup_string(props,
3309 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3310 (void) nvlist_lookup_uint64(props,
3311 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
3314 spa = spa_add(pool, config, altroot);
3315 spa->spa_import_flags = flags;
3318 * Verbatim import - Take a pool and insert it into the namespace
3319 * as if it had been loaded at boot.
3321 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
3323 spa_configfile_set(spa, props, B_FALSE);
3325 spa_config_sync(spa, B_FALSE, B_TRUE);
3327 mutex_exit(&spa_namespace_lock);
3328 spa_history_log_version(spa, LOG_POOL_IMPORT);
3333 spa_activate(spa, mode);
3336 * Don't start async tasks until we know everything is healthy.
3338 spa_async_suspend(spa);
3340 zpool_get_rewind_policy(config, &policy);
3341 if (policy.zrp_request & ZPOOL_DO_REWIND)
3342 state = SPA_LOAD_RECOVER;
3345 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
3346 * because the user-supplied config is actually the one to trust when
3349 if (state != SPA_LOAD_RECOVER)
3350 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3352 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
3353 policy.zrp_request);
3356 * Propagate anything learned while loading the pool and pass it
3357 * back to caller (i.e. rewind info, missing devices, etc).
3359 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
3360 spa->spa_load_info) == 0);
3362 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3364 * Toss any existing sparelist, as it doesn't have any validity
3365 * anymore, and conflicts with spa_has_spare().
3367 if (spa->spa_spares.sav_config) {
3368 nvlist_free(spa->spa_spares.sav_config);
3369 spa->spa_spares.sav_config = NULL;
3370 spa_load_spares(spa);
3372 if (spa->spa_l2cache.sav_config) {
3373 nvlist_free(spa->spa_l2cache.sav_config);
3374 spa->spa_l2cache.sav_config = NULL;
3375 spa_load_l2cache(spa);
3378 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3381 error = spa_validate_aux(spa, nvroot, -1ULL,
3384 error = spa_validate_aux(spa, nvroot, -1ULL,
3385 VDEV_ALLOC_L2CACHE);
3386 spa_config_exit(spa, SCL_ALL, FTAG);
3389 spa_configfile_set(spa, props, B_FALSE);
3391 if (error != 0 || (props && spa_writeable(spa) &&
3392 (error = spa_prop_set(spa, props)))) {
3394 spa_deactivate(spa);
3396 mutex_exit(&spa_namespace_lock);
3400 spa_async_resume(spa);
3403 * Override any spares and level 2 cache devices as specified by
3404 * the user, as these may have correct device names/devids, etc.
3406 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3407 &spares, &nspares) == 0) {
3408 if (spa->spa_spares.sav_config)
3409 VERIFY(nvlist_remove(spa->spa_spares.sav_config,
3410 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
3412 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
3413 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3414 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3415 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3416 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3417 spa_load_spares(spa);
3418 spa_config_exit(spa, SCL_ALL, FTAG);
3419 spa->spa_spares.sav_sync = B_TRUE;
3421 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3422 &l2cache, &nl2cache) == 0) {
3423 if (spa->spa_l2cache.sav_config)
3424 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
3425 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
3427 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3428 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3429 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3430 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3431 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3432 spa_load_l2cache(spa);
3433 spa_config_exit(spa, SCL_ALL, FTAG);
3434 spa->spa_l2cache.sav_sync = B_TRUE;
3438 * Check for any removed devices.
3440 if (spa->spa_autoreplace) {
3441 spa_aux_check_removed(&spa->spa_spares);
3442 spa_aux_check_removed(&spa->spa_l2cache);
3445 if (spa_writeable(spa)) {
3447 * Update the config cache to include the newly-imported pool.
3449 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3453 * It's possible that the pool was expanded while it was exported.
3454 * We kick off an async task to handle this for us.
3456 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
3458 mutex_exit(&spa_namespace_lock);
3459 spa_history_log_version(spa, LOG_POOL_IMPORT);
3465 spa_tryimport(nvlist_t *tryconfig)
3467 nvlist_t *config = NULL;
3473 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
3476 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
3480 * Create and initialize the spa structure.
3482 mutex_enter(&spa_namespace_lock);
3483 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
3484 spa_activate(spa, FREAD);
3487 * Pass off the heavy lifting to spa_load().
3488 * Pass TRUE for mosconfig because the user-supplied config
3489 * is actually the one to trust when doing an import.
3491 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
3494 * If 'tryconfig' was at least parsable, return the current config.
3496 if (spa->spa_root_vdev != NULL) {
3497 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
3498 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
3500 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
3502 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
3503 spa->spa_uberblock.ub_timestamp) == 0);
3506 * If the bootfs property exists on this pool then we
3507 * copy it out so that external consumers can tell which
3508 * pools are bootable.
3510 if ((!error || error == EEXIST) && spa->spa_bootfs) {
3511 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3514 * We have to play games with the name since the
3515 * pool was opened as TRYIMPORT_NAME.
3517 if (dsl_dsobj_to_dsname(spa_name(spa),
3518 spa->spa_bootfs, tmpname) == 0) {
3520 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3522 cp = strchr(tmpname, '/');
3524 (void) strlcpy(dsname, tmpname,
3527 (void) snprintf(dsname, MAXPATHLEN,
3528 "%s/%s", poolname, ++cp);
3530 VERIFY(nvlist_add_string(config,
3531 ZPOOL_CONFIG_BOOTFS, dsname) == 0);
3532 kmem_free(dsname, MAXPATHLEN);
3534 kmem_free(tmpname, MAXPATHLEN);
3538 * Add the list of hot spares and level 2 cache devices.
3540 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3541 spa_add_spares(spa, config);
3542 spa_add_l2cache(spa, config);
3543 spa_config_exit(spa, SCL_CONFIG, FTAG);
3547 spa_deactivate(spa);
3549 mutex_exit(&spa_namespace_lock);
3555 * Pool export/destroy
3557 * The act of destroying or exporting a pool is very simple. We make sure there
3558 * is no more pending I/O and any references to the pool are gone. Then, we
3559 * update the pool state and sync all the labels to disk, removing the
3560 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
3561 * we don't sync the labels or remove the configuration cache.
3564 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
3565 boolean_t force, boolean_t hardforce)
3572 if (!(spa_mode_global & FWRITE))
3575 mutex_enter(&spa_namespace_lock);
3576 if ((spa = spa_lookup(pool)) == NULL) {
3577 mutex_exit(&spa_namespace_lock);
3582 * Put a hold on the pool, drop the namespace lock, stop async tasks,
3583 * reacquire the namespace lock, and see if we can export.
3585 spa_open_ref(spa, FTAG);
3586 mutex_exit(&spa_namespace_lock);
3587 spa_async_suspend(spa);
3588 mutex_enter(&spa_namespace_lock);
3589 spa_close(spa, FTAG);
3592 * The pool will be in core if it's openable,
3593 * in which case we can modify its state.
3595 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
3597 * Objsets may be open only because they're dirty, so we
3598 * have to force it to sync before checking spa_refcnt.
3600 txg_wait_synced(spa->spa_dsl_pool, 0);
3603 * A pool cannot be exported or destroyed if there are active
3604 * references. If we are resetting a pool, allow references by
3605 * fault injection handlers.
3607 if (!spa_refcount_zero(spa) ||
3608 (spa->spa_inject_ref != 0 &&
3609 new_state != POOL_STATE_UNINITIALIZED)) {
3610 spa_async_resume(spa);
3611 mutex_exit(&spa_namespace_lock);
3616 * A pool cannot be exported if it has an active shared spare.
3617 * This is to prevent other pools stealing the active spare
3618 * from an exported pool. At user's own will, such pool can
3619 * be forcedly exported.
3621 if (!force && new_state == POOL_STATE_EXPORTED &&
3622 spa_has_active_shared_spare(spa)) {
3623 spa_async_resume(spa);
3624 mutex_exit(&spa_namespace_lock);
3629 * We want this to be reflected on every label,
3630 * so mark them all dirty. spa_unload() will do the
3631 * final sync that pushes these changes out.
3633 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
3634 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3635 spa->spa_state = new_state;
3636 spa->spa_final_txg = spa_last_synced_txg(spa) +
3638 vdev_config_dirty(spa->spa_root_vdev);
3639 spa_config_exit(spa, SCL_ALL, FTAG);
3643 spa_event_notify(spa, NULL, FM_EREPORT_ZFS_POOL_DESTROY);
3645 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
3647 spa_deactivate(spa);
3650 if (oldconfig && spa->spa_config)
3651 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
3653 if (new_state != POOL_STATE_UNINITIALIZED) {
3655 spa_config_sync(spa, B_TRUE, B_TRUE);
3658 mutex_exit(&spa_namespace_lock);
3664 * Destroy a storage pool.
3667 spa_destroy(char *pool)
3669 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
3674 * Export a storage pool.
3677 spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
3678 boolean_t hardforce)
3680 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
3685 * Similar to spa_export(), this unloads the spa_t without actually removing it
3686 * from the namespace in any way.
3689 spa_reset(char *pool)
3691 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
3696 * ==========================================================================
3697 * Device manipulation
3698 * ==========================================================================
3702 * Add a device to a storage pool.
3705 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
3709 vdev_t *rvd = spa->spa_root_vdev;
3711 nvlist_t **spares, **l2cache;
3712 uint_t nspares, nl2cache;
3715 ASSERT(spa_writeable(spa));
3717 txg = spa_vdev_enter(spa);
3719 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
3720 VDEV_ALLOC_ADD)) != 0)
3721 return (spa_vdev_exit(spa, NULL, txg, error));
3723 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */
3725 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
3729 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
3733 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
3734 return (spa_vdev_exit(spa, vd, txg, EINVAL));
3736 if (vd->vdev_children != 0 &&
3737 (error = vdev_create(vd, txg, B_FALSE)) != 0)
3738 return (spa_vdev_exit(spa, vd, txg, error));
3741 * We must validate the spares and l2cache devices after checking the
3742 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
3744 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
3745 return (spa_vdev_exit(spa, vd, txg, error));
3748 * Transfer each new top-level vdev from vd to rvd.
3750 for (c = 0; c < vd->vdev_children; c++) {
3753 * Set the vdev id to the first hole, if one exists.
3755 for (id = 0; id < rvd->vdev_children; id++) {
3756 if (rvd->vdev_child[id]->vdev_ishole) {
3757 vdev_free(rvd->vdev_child[id]);
3761 tvd = vd->vdev_child[c];
3762 vdev_remove_child(vd, tvd);
3764 vdev_add_child(rvd, tvd);
3765 vdev_config_dirty(tvd);
3769 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
3770 ZPOOL_CONFIG_SPARES);
3771 spa_load_spares(spa);
3772 spa->spa_spares.sav_sync = B_TRUE;
3775 if (nl2cache != 0) {
3776 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
3777 ZPOOL_CONFIG_L2CACHE);
3778 spa_load_l2cache(spa);
3779 spa->spa_l2cache.sav_sync = B_TRUE;
3783 * We have to be careful when adding new vdevs to an existing pool.
3784 * If other threads start allocating from these vdevs before we
3785 * sync the config cache, and we lose power, then upon reboot we may
3786 * fail to open the pool because there are DVAs that the config cache
3787 * can't translate. Therefore, we first add the vdevs without
3788 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
3789 * and then let spa_config_update() initialize the new metaslabs.
3791 * spa_load() checks for added-but-not-initialized vdevs, so that
3792 * if we lose power at any point in this sequence, the remaining
3793 * steps will be completed the next time we load the pool.
3795 (void) spa_vdev_exit(spa, vd, txg, 0);
3797 mutex_enter(&spa_namespace_lock);
3798 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3799 mutex_exit(&spa_namespace_lock);
3805 * Attach a device to a mirror. The arguments are the path to any device
3806 * in the mirror, and the nvroot for the new device. If the path specifies
3807 * a device that is not mirrored, we automatically insert the mirror vdev.
3809 * If 'replacing' is specified, the new device is intended to replace the
3810 * existing device; in this case the two devices are made into their own
3811 * mirror using the 'replacing' vdev, which is functionally identical to
3812 * the mirror vdev (it actually reuses all the same ops) but has a few
3813 * extra rules: you can't attach to it after it's been created, and upon
3814 * completion of resilvering, the first disk (the one being replaced)
3815 * is automatically detached.
3818 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
3820 uint64_t txg, dtl_max_txg;
3821 ASSERTV(vdev_t *rvd = spa->spa_root_vdev;)
3822 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
3824 char *oldvdpath, *newvdpath;
3828 ASSERT(spa_writeable(spa));
3830 txg = spa_vdev_enter(spa);
3832 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
3835 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
3837 if (!oldvd->vdev_ops->vdev_op_leaf)
3838 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3840 pvd = oldvd->vdev_parent;
3842 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
3843 VDEV_ALLOC_ADD)) != 0)
3844 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
3846 if (newrootvd->vdev_children != 1)
3847 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
3849 newvd = newrootvd->vdev_child[0];
3851 if (!newvd->vdev_ops->vdev_op_leaf)
3852 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
3854 if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
3855 return (spa_vdev_exit(spa, newrootvd, txg, error));
3858 * Spares can't replace logs
3860 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
3861 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3865 * For attach, the only allowable parent is a mirror or the root
3868 if (pvd->vdev_ops != &vdev_mirror_ops &&
3869 pvd->vdev_ops != &vdev_root_ops)
3870 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3872 pvops = &vdev_mirror_ops;
3875 * Active hot spares can only be replaced by inactive hot
3878 if (pvd->vdev_ops == &vdev_spare_ops &&
3879 oldvd->vdev_isspare &&
3880 !spa_has_spare(spa, newvd->vdev_guid))
3881 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3884 * If the source is a hot spare, and the parent isn't already a
3885 * spare, then we want to create a new hot spare. Otherwise, we
3886 * want to create a replacing vdev. The user is not allowed to
3887 * attach to a spared vdev child unless the 'isspare' state is
3888 * the same (spare replaces spare, non-spare replaces
3891 if (pvd->vdev_ops == &vdev_replacing_ops &&
3892 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
3893 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3894 } else if (pvd->vdev_ops == &vdev_spare_ops &&
3895 newvd->vdev_isspare != oldvd->vdev_isspare) {
3896 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3899 if (newvd->vdev_isspare)
3900 pvops = &vdev_spare_ops;
3902 pvops = &vdev_replacing_ops;
3906 * Make sure the new device is big enough.
3908 if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
3909 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
3912 * The new device cannot have a higher alignment requirement
3913 * than the top-level vdev.
3915 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
3916 return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
3919 * If this is an in-place replacement, update oldvd's path and devid
3920 * to make it distinguishable from newvd, and unopenable from now on.
3922 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
3923 spa_strfree(oldvd->vdev_path);
3924 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
3926 (void) sprintf(oldvd->vdev_path, "%s/%s",
3927 newvd->vdev_path, "old");
3928 if (oldvd->vdev_devid != NULL) {
3929 spa_strfree(oldvd->vdev_devid);
3930 oldvd->vdev_devid = NULL;
3934 /* mark the device being resilvered */
3935 newvd->vdev_resilvering = B_TRUE;
3938 * If the parent is not a mirror, or if we're replacing, insert the new
3939 * mirror/replacing/spare vdev above oldvd.
3941 if (pvd->vdev_ops != pvops)
3942 pvd = vdev_add_parent(oldvd, pvops);
3944 ASSERT(pvd->vdev_top->vdev_parent == rvd);
3945 ASSERT(pvd->vdev_ops == pvops);
3946 ASSERT(oldvd->vdev_parent == pvd);
3949 * Extract the new device from its root and add it to pvd.
3951 vdev_remove_child(newrootvd, newvd);
3952 newvd->vdev_id = pvd->vdev_children;
3953 newvd->vdev_crtxg = oldvd->vdev_crtxg;
3954 vdev_add_child(pvd, newvd);
3956 tvd = newvd->vdev_top;
3957 ASSERT(pvd->vdev_top == tvd);
3958 ASSERT(tvd->vdev_parent == rvd);
3960 vdev_config_dirty(tvd);
3963 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
3964 * for any dmu_sync-ed blocks. It will propagate upward when
3965 * spa_vdev_exit() calls vdev_dtl_reassess().
3967 dtl_max_txg = txg + TXG_CONCURRENT_STATES;
3969 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
3970 dtl_max_txg - TXG_INITIAL);
3972 if (newvd->vdev_isspare) {
3973 spa_spare_activate(newvd);
3974 spa_event_notify(spa, newvd, FM_EREPORT_ZFS_DEVICE_SPARE);
3977 oldvdpath = spa_strdup(oldvd->vdev_path);
3978 newvdpath = spa_strdup(newvd->vdev_path);
3979 newvd_isspare = newvd->vdev_isspare;
3982 * Mark newvd's DTL dirty in this txg.
3984 vdev_dirty(tvd, VDD_DTL, newvd, txg);
3987 * Restart the resilver
3989 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
3994 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
3996 spa_history_log_internal(LOG_POOL_VDEV_ATTACH, spa, NULL,
3997 "%s vdev=%s %s vdev=%s",
3998 replacing && newvd_isspare ? "spare in" :
3999 replacing ? "replace" : "attach", newvdpath,
4000 replacing ? "for" : "to", oldvdpath);
4002 spa_strfree(oldvdpath);
4003 spa_strfree(newvdpath);
4005 if (spa->spa_bootfs)
4006 spa_event_notify(spa, newvd, FM_EREPORT_ZFS_BOOTFS_VDEV_ATTACH);
4012 * Detach a device from a mirror or replacing vdev.
4013 * If 'replace_done' is specified, only detach if the parent
4014 * is a replacing vdev.
4017 spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4021 ASSERTV(vdev_t *rvd = spa->spa_root_vdev;)
4022 vdev_t *vd, *pvd, *cvd, *tvd;
4023 boolean_t unspare = B_FALSE;
4024 uint64_t unspare_guid = 0;
4028 ASSERT(spa_writeable(spa));
4030 txg = spa_vdev_enter(spa);
4032 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4035 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4037 if (!vd->vdev_ops->vdev_op_leaf)
4038 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4040 pvd = vd->vdev_parent;
4043 * If the parent/child relationship is not as expected, don't do it.
4044 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4045 * vdev that's replacing B with C. The user's intent in replacing
4046 * is to go from M(A,B) to M(A,C). If the user decides to cancel
4047 * the replace by detaching C, the expected behavior is to end up
4048 * M(A,B). But suppose that right after deciding to detach C,
4049 * the replacement of B completes. We would have M(A,C), and then
4050 * ask to detach C, which would leave us with just A -- not what
4051 * the user wanted. To prevent this, we make sure that the
4052 * parent/child relationship hasn't changed -- in this example,
4053 * that C's parent is still the replacing vdev R.
4055 if (pvd->vdev_guid != pguid && pguid != 0)
4056 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4059 * Only 'replacing' or 'spare' vdevs can be replaced.
4061 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4062 pvd->vdev_ops != &vdev_spare_ops)
4063 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4065 ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4066 spa_version(spa) >= SPA_VERSION_SPARES);
4069 * Only mirror, replacing, and spare vdevs support detach.
4071 if (pvd->vdev_ops != &vdev_replacing_ops &&
4072 pvd->vdev_ops != &vdev_mirror_ops &&
4073 pvd->vdev_ops != &vdev_spare_ops)
4074 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4077 * If this device has the only valid copy of some data,
4078 * we cannot safely detach it.
4080 if (vdev_dtl_required(vd))
4081 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4083 ASSERT(pvd->vdev_children >= 2);
4086 * If we are detaching the second disk from a replacing vdev, then
4087 * check to see if we changed the original vdev's path to have "/old"
4088 * at the end in spa_vdev_attach(). If so, undo that change now.
4090 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4091 vd->vdev_path != NULL) {
4092 size_t len = strlen(vd->vdev_path);
4094 for (c = 0; c < pvd->vdev_children; c++) {
4095 cvd = pvd->vdev_child[c];
4097 if (cvd == vd || cvd->vdev_path == NULL)
4100 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4101 strcmp(cvd->vdev_path + len, "/old") == 0) {
4102 spa_strfree(cvd->vdev_path);
4103 cvd->vdev_path = spa_strdup(vd->vdev_path);
4110 * If we are detaching the original disk from a spare, then it implies
4111 * that the spare should become a real disk, and be removed from the
4112 * active spare list for the pool.
4114 if (pvd->vdev_ops == &vdev_spare_ops &&
4116 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4120 * Erase the disk labels so the disk can be used for other things.
4121 * This must be done after all other error cases are handled,
4122 * but before we disembowel vd (so we can still do I/O to it).
4123 * But if we can't do it, don't treat the error as fatal --
4124 * it may be that the unwritability of the disk is the reason
4125 * it's being detached!
4127 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4130 * Remove vd from its parent and compact the parent's children.
4132 vdev_remove_child(pvd, vd);
4133 vdev_compact_children(pvd);
4136 * Remember one of the remaining children so we can get tvd below.
4138 cvd = pvd->vdev_child[pvd->vdev_children - 1];
4141 * If we need to remove the remaining child from the list of hot spares,
4142 * do it now, marking the vdev as no longer a spare in the process.
4143 * We must do this before vdev_remove_parent(), because that can
4144 * change the GUID if it creates a new toplevel GUID. For a similar
4145 * reason, we must remove the spare now, in the same txg as the detach;
4146 * otherwise someone could attach a new sibling, change the GUID, and
4147 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4150 ASSERT(cvd->vdev_isspare);
4151 spa_spare_remove(cvd);
4152 unspare_guid = cvd->vdev_guid;
4153 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4154 cvd->vdev_unspare = B_TRUE;
4158 * If the parent mirror/replacing vdev only has one child,
4159 * the parent is no longer needed. Remove it from the tree.
4161 if (pvd->vdev_children == 1) {
4162 if (pvd->vdev_ops == &vdev_spare_ops)
4163 cvd->vdev_unspare = B_FALSE;
4164 vdev_remove_parent(cvd);
4165 cvd->vdev_resilvering = B_FALSE;
4170 * We don't set tvd until now because the parent we just removed
4171 * may have been the previous top-level vdev.
4173 tvd = cvd->vdev_top;
4174 ASSERT(tvd->vdev_parent == rvd);
4177 * Reevaluate the parent vdev state.
4179 vdev_propagate_state(cvd);
4182 * If the 'autoexpand' property is set on the pool then automatically
4183 * try to expand the size of the pool. For example if the device we
4184 * just detached was smaller than the others, it may be possible to
4185 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4186 * first so that we can obtain the updated sizes of the leaf vdevs.
4188 if (spa->spa_autoexpand) {
4190 vdev_expand(tvd, txg);
4193 vdev_config_dirty(tvd);
4196 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
4197 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4198 * But first make sure we're not on any *other* txg's DTL list, to
4199 * prevent vd from being accessed after it's freed.
4201 vdpath = spa_strdup(vd->vdev_path);
4202 for (t = 0; t < TXG_SIZE; t++)
4203 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4204 vd->vdev_detached = B_TRUE;
4205 vdev_dirty(tvd, VDD_DTL, vd, txg);
4207 spa_event_notify(spa, vd, FM_EREPORT_ZFS_DEVICE_REMOVE);
4209 /* hang on to the spa before we release the lock */
4210 spa_open_ref(spa, FTAG);
4212 error = spa_vdev_exit(spa, vd, txg, 0);
4214 spa_history_log_internal(LOG_POOL_VDEV_DETACH, spa, NULL,
4216 spa_strfree(vdpath);
4219 * If this was the removal of the original device in a hot spare vdev,
4220 * then we want to go through and remove the device from the hot spare
4221 * list of every other pool.
4224 spa_t *altspa = NULL;
4226 mutex_enter(&spa_namespace_lock);
4227 while ((altspa = spa_next(altspa)) != NULL) {
4228 if (altspa->spa_state != POOL_STATE_ACTIVE ||
4232 spa_open_ref(altspa, FTAG);
4233 mutex_exit(&spa_namespace_lock);
4234 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
4235 mutex_enter(&spa_namespace_lock);
4236 spa_close(altspa, FTAG);
4238 mutex_exit(&spa_namespace_lock);
4240 /* search the rest of the vdevs for spares to remove */
4241 spa_vdev_resilver_done(spa);
4244 /* all done with the spa; OK to release */
4245 mutex_enter(&spa_namespace_lock);
4246 spa_close(spa, FTAG);
4247 mutex_exit(&spa_namespace_lock);
4253 * Split a set of devices from their mirrors, and create a new pool from them.
4256 spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
4257 nvlist_t *props, boolean_t exp)
4260 uint64_t txg, *glist;
4262 uint_t c, children, lastlog;
4263 nvlist_t **child, *nvl, *tmp;
4265 char *altroot = NULL;
4266 vdev_t *rvd, **vml = NULL; /* vdev modify list */
4267 boolean_t activate_slog;
4269 ASSERT(spa_writeable(spa));
4271 txg = spa_vdev_enter(spa);
4273 /* clear the log and flush everything up to now */
4274 activate_slog = spa_passivate_log(spa);
4275 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4276 error = spa_offline_log(spa);
4277 txg = spa_vdev_config_enter(spa);
4280 spa_activate_log(spa);
4283 return (spa_vdev_exit(spa, NULL, txg, error));
4285 /* check new spa name before going any further */
4286 if (spa_lookup(newname) != NULL)
4287 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
4290 * scan through all the children to ensure they're all mirrors
4292 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
4293 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
4295 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4297 /* first, check to ensure we've got the right child count */
4298 rvd = spa->spa_root_vdev;
4300 for (c = 0; c < rvd->vdev_children; c++) {
4301 vdev_t *vd = rvd->vdev_child[c];
4303 /* don't count the holes & logs as children */
4304 if (vd->vdev_islog || vd->vdev_ishole) {
4312 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
4313 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4315 /* next, ensure no spare or cache devices are part of the split */
4316 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
4317 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
4318 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4320 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
4321 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
4323 /* then, loop over each vdev and validate it */
4324 for (c = 0; c < children; c++) {
4325 uint64_t is_hole = 0;
4327 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
4331 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
4332 spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
4340 /* which disk is going to be split? */
4341 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
4347 /* look it up in the spa */
4348 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
4349 if (vml[c] == NULL) {
4354 /* make sure there's nothing stopping the split */
4355 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
4356 vml[c]->vdev_islog ||
4357 vml[c]->vdev_ishole ||
4358 vml[c]->vdev_isspare ||
4359 vml[c]->vdev_isl2cache ||
4360 !vdev_writeable(vml[c]) ||
4361 vml[c]->vdev_children != 0 ||
4362 vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
4363 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
4368 if (vdev_dtl_required(vml[c])) {
4373 /* we need certain info from the top level */
4374 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
4375 vml[c]->vdev_top->vdev_ms_array) == 0);
4376 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
4377 vml[c]->vdev_top->vdev_ms_shift) == 0);
4378 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
4379 vml[c]->vdev_top->vdev_asize) == 0);
4380 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
4381 vml[c]->vdev_top->vdev_ashift) == 0);
4385 kmem_free(vml, children * sizeof (vdev_t *));
4386 kmem_free(glist, children * sizeof (uint64_t));
4387 return (spa_vdev_exit(spa, NULL, txg, error));
4390 /* stop writers from using the disks */
4391 for (c = 0; c < children; c++) {
4393 vml[c]->vdev_offline = B_TRUE;
4395 vdev_reopen(spa->spa_root_vdev);
4398 * Temporarily record the splitting vdevs in the spa config. This
4399 * will disappear once the config is regenerated.
4401 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4402 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
4403 glist, children) == 0);
4404 kmem_free(glist, children * sizeof (uint64_t));
4406 mutex_enter(&spa->spa_props_lock);
4407 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
4409 mutex_exit(&spa->spa_props_lock);
4410 spa->spa_config_splitting = nvl;
4411 vdev_config_dirty(spa->spa_root_vdev);
4413 /* configure and create the new pool */
4414 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
4415 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4416 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
4417 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
4418 spa_version(spa)) == 0);
4419 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
4420 spa->spa_config_txg) == 0);
4421 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4422 spa_generate_guid(NULL)) == 0);
4423 (void) nvlist_lookup_string(props,
4424 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4426 /* add the new pool to the namespace */
4427 newspa = spa_add(newname, config, altroot);
4428 newspa->spa_config_txg = spa->spa_config_txg;
4429 spa_set_log_state(newspa, SPA_LOG_CLEAR);
4431 /* release the spa config lock, retaining the namespace lock */
4432 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4434 if (zio_injection_enabled)
4435 zio_handle_panic_injection(spa, FTAG, 1);
4437 spa_activate(newspa, spa_mode_global);
4438 spa_async_suspend(newspa);
4440 /* create the new pool from the disks of the original pool */
4441 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
4445 /* if that worked, generate a real config for the new pool */
4446 if (newspa->spa_root_vdev != NULL) {
4447 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
4448 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4449 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
4450 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
4451 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
4456 if (props != NULL) {
4457 spa_configfile_set(newspa, props, B_FALSE);
4458 error = spa_prop_set(newspa, props);
4463 /* flush everything */
4464 txg = spa_vdev_config_enter(newspa);
4465 vdev_config_dirty(newspa->spa_root_vdev);
4466 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
4468 if (zio_injection_enabled)
4469 zio_handle_panic_injection(spa, FTAG, 2);
4471 spa_async_resume(newspa);
4473 /* finally, update the original pool's config */
4474 txg = spa_vdev_config_enter(spa);
4475 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
4476 error = dmu_tx_assign(tx, TXG_WAIT);
4479 for (c = 0; c < children; c++) {
4480 if (vml[c] != NULL) {
4483 spa_history_log_internal(LOG_POOL_VDEV_DETACH,
4489 vdev_config_dirty(spa->spa_root_vdev);
4490 spa->spa_config_splitting = NULL;
4494 (void) spa_vdev_exit(spa, NULL, txg, 0);
4496 if (zio_injection_enabled)
4497 zio_handle_panic_injection(spa, FTAG, 3);
4499 /* split is complete; log a history record */
4500 spa_history_log_internal(LOG_POOL_SPLIT, newspa, NULL,
4501 "split new pool %s from pool %s", newname, spa_name(spa));
4503 kmem_free(vml, children * sizeof (vdev_t *));
4505 /* if we're not going to mount the filesystems in userland, export */
4507 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
4514 spa_deactivate(newspa);
4517 txg = spa_vdev_config_enter(spa);
4519 /* re-online all offlined disks */
4520 for (c = 0; c < children; c++) {
4522 vml[c]->vdev_offline = B_FALSE;
4524 vdev_reopen(spa->spa_root_vdev);
4526 nvlist_free(spa->spa_config_splitting);
4527 spa->spa_config_splitting = NULL;
4528 (void) spa_vdev_exit(spa, NULL, txg, error);
4530 kmem_free(vml, children * sizeof (vdev_t *));
4535 spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
4539 for (i = 0; i < count; i++) {
4542 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
4545 if (guid == target_guid)
4553 spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
4554 nvlist_t *dev_to_remove)
4556 nvlist_t **newdev = NULL;
4560 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
4562 for (i = 0, j = 0; i < count; i++) {
4563 if (dev[i] == dev_to_remove)
4565 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
4568 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
4569 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
4571 for (i = 0; i < count - 1; i++)
4572 nvlist_free(newdev[i]);
4575 kmem_free(newdev, (count - 1) * sizeof (void *));
4579 * Evacuate the device.
4582 spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
4587 ASSERT(MUTEX_HELD(&spa_namespace_lock));
4588 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
4589 ASSERT(vd == vd->vdev_top);
4592 * Evacuate the device. We don't hold the config lock as writer
4593 * since we need to do I/O but we do keep the
4594 * spa_namespace_lock held. Once this completes the device
4595 * should no longer have any blocks allocated on it.
4597 if (vd->vdev_islog) {
4598 if (vd->vdev_stat.vs_alloc != 0)
4599 error = spa_offline_log(spa);
4608 * The evacuation succeeded. Remove any remaining MOS metadata
4609 * associated with this vdev, and wait for these changes to sync.
4611 ASSERT3U(vd->vdev_stat.vs_alloc, ==, 0);
4612 txg = spa_vdev_config_enter(spa);
4613 vd->vdev_removing = B_TRUE;
4614 vdev_dirty(vd, 0, NULL, txg);
4615 vdev_config_dirty(vd);
4616 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4622 * Complete the removal by cleaning up the namespace.
4625 spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
4627 vdev_t *rvd = spa->spa_root_vdev;
4628 uint64_t id = vd->vdev_id;
4629 boolean_t last_vdev = (id == (rvd->vdev_children - 1));
4631 ASSERT(MUTEX_HELD(&spa_namespace_lock));
4632 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
4633 ASSERT(vd == vd->vdev_top);
4636 * Only remove any devices which are empty.
4638 if (vd->vdev_stat.vs_alloc != 0)
4641 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4643 if (list_link_active(&vd->vdev_state_dirty_node))
4644 vdev_state_clean(vd);
4645 if (list_link_active(&vd->vdev_config_dirty_node))
4646 vdev_config_clean(vd);
4651 vdev_compact_children(rvd);
4653 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
4654 vdev_add_child(rvd, vd);
4656 vdev_config_dirty(rvd);
4659 * Reassess the health of our root vdev.
4665 * Remove a device from the pool -
4667 * Removing a device from the vdev namespace requires several steps
4668 * and can take a significant amount of time. As a result we use
4669 * the spa_vdev_config_[enter/exit] functions which allow us to
4670 * grab and release the spa_config_lock while still holding the namespace
4671 * lock. During each step the configuration is synced out.
4675 * Remove a device from the pool. Currently, this supports removing only hot
4676 * spares, slogs, and level 2 ARC devices.
4679 spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
4682 metaslab_group_t *mg;
4683 nvlist_t **spares, **l2cache, *nv;
4685 uint_t nspares, nl2cache;
4687 boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
4689 ASSERT(spa_writeable(spa));
4692 txg = spa_vdev_enter(spa);
4694 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4696 if (spa->spa_spares.sav_vdevs != NULL &&
4697 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
4698 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
4699 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
4701 * Only remove the hot spare if it's not currently in use
4704 if (vd == NULL || unspare) {
4705 spa_vdev_remove_aux(spa->spa_spares.sav_config,
4706 ZPOOL_CONFIG_SPARES, spares, nspares, nv);
4707 spa_load_spares(spa);
4708 spa->spa_spares.sav_sync = B_TRUE;
4712 } else if (spa->spa_l2cache.sav_vdevs != NULL &&
4713 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
4714 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
4715 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
4717 * Cache devices can always be removed.
4719 spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
4720 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
4721 spa_load_l2cache(spa);
4722 spa->spa_l2cache.sav_sync = B_TRUE;
4723 } else if (vd != NULL && vd->vdev_islog) {
4725 ASSERT(vd == vd->vdev_top);
4728 * XXX - Once we have bp-rewrite this should
4729 * become the common case.
4735 * Stop allocating from this vdev.
4737 metaslab_group_passivate(mg);
4740 * Wait for the youngest allocations and frees to sync,
4741 * and then wait for the deferral of those frees to finish.
4743 spa_vdev_config_exit(spa, NULL,
4744 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
4747 * Attempt to evacuate the vdev.
4749 error = spa_vdev_remove_evacuate(spa, vd);
4751 txg = spa_vdev_config_enter(spa);
4754 * If we couldn't evacuate the vdev, unwind.
4757 metaslab_group_activate(mg);
4758 return (spa_vdev_exit(spa, NULL, txg, error));
4762 * Clean up the vdev namespace.
4764 spa_vdev_remove_from_namespace(spa, vd);
4766 } else if (vd != NULL) {
4768 * Normal vdevs cannot be removed (yet).
4773 * There is no vdev of any kind with the specified guid.
4779 return (spa_vdev_exit(spa, NULL, txg, error));
4785 * Find any device that's done replacing, or a vdev marked 'unspare' that's
4786 * current spared, so we can detach it.
4789 spa_vdev_resilver_done_hunt(vdev_t *vd)
4791 vdev_t *newvd, *oldvd;
4794 for (c = 0; c < vd->vdev_children; c++) {
4795 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
4801 * Check for a completed replacement. We always consider the first
4802 * vdev in the list to be the oldest vdev, and the last one to be
4803 * the newest (see spa_vdev_attach() for how that works). In
4804 * the case where the newest vdev is faulted, we will not automatically
4805 * remove it after a resilver completes. This is OK as it will require
4806 * user intervention to determine which disk the admin wishes to keep.
4808 if (vd->vdev_ops == &vdev_replacing_ops) {
4809 ASSERT(vd->vdev_children > 1);
4811 newvd = vd->vdev_child[vd->vdev_children - 1];
4812 oldvd = vd->vdev_child[0];
4814 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
4815 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
4816 !vdev_dtl_required(oldvd))
4821 * Check for a completed resilver with the 'unspare' flag set.
4823 if (vd->vdev_ops == &vdev_spare_ops) {
4824 vdev_t *first = vd->vdev_child[0];
4825 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
4827 if (last->vdev_unspare) {
4830 } else if (first->vdev_unspare) {
4837 if (oldvd != NULL &&
4838 vdev_dtl_empty(newvd, DTL_MISSING) &&
4839 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
4840 !vdev_dtl_required(oldvd))
4844 * If there are more than two spares attached to a disk,
4845 * and those spares are not required, then we want to
4846 * attempt to free them up now so that they can be used
4847 * by other pools. Once we're back down to a single
4848 * disk+spare, we stop removing them.
4850 if (vd->vdev_children > 2) {
4851 newvd = vd->vdev_child[1];
4853 if (newvd->vdev_isspare && last->vdev_isspare &&
4854 vdev_dtl_empty(last, DTL_MISSING) &&
4855 vdev_dtl_empty(last, DTL_OUTAGE) &&
4856 !vdev_dtl_required(newvd))
4865 spa_vdev_resilver_done(spa_t *spa)
4867 vdev_t *vd, *pvd, *ppvd;
4868 uint64_t guid, sguid, pguid, ppguid;
4870 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4872 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
4873 pvd = vd->vdev_parent;
4874 ppvd = pvd->vdev_parent;
4875 guid = vd->vdev_guid;
4876 pguid = pvd->vdev_guid;
4877 ppguid = ppvd->vdev_guid;
4880 * If we have just finished replacing a hot spared device, then
4881 * we need to detach the parent's first child (the original hot
4884 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
4885 ppvd->vdev_children == 2) {
4886 ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
4887 sguid = ppvd->vdev_child[1]->vdev_guid;
4889 spa_config_exit(spa, SCL_ALL, FTAG);
4890 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
4892 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
4894 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4897 spa_config_exit(spa, SCL_ALL, FTAG);
4901 * Update the stored path or FRU for this vdev.
4904 spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
4908 boolean_t sync = B_FALSE;
4910 ASSERT(spa_writeable(spa));
4912 spa_vdev_state_enter(spa, SCL_ALL);
4914 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
4915 return (spa_vdev_state_exit(spa, NULL, ENOENT));
4917 if (!vd->vdev_ops->vdev_op_leaf)
4918 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
4921 if (strcmp(value, vd->vdev_path) != 0) {
4922 spa_strfree(vd->vdev_path);
4923 vd->vdev_path = spa_strdup(value);
4927 if (vd->vdev_fru == NULL) {
4928 vd->vdev_fru = spa_strdup(value);
4930 } else if (strcmp(value, vd->vdev_fru) != 0) {
4931 spa_strfree(vd->vdev_fru);
4932 vd->vdev_fru = spa_strdup(value);
4937 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
4941 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
4943 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
4947 spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
4949 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
4953 * ==========================================================================
4955 * ==========================================================================
4959 spa_scan_stop(spa_t *spa)
4961 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
4962 if (dsl_scan_resilvering(spa->spa_dsl_pool))
4964 return (dsl_scan_cancel(spa->spa_dsl_pool));
4968 spa_scan(spa_t *spa, pool_scan_func_t func)
4970 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
4972 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
4976 * If a resilver was requested, but there is no DTL on a
4977 * writeable leaf device, we have nothing to do.
4979 if (func == POOL_SCAN_RESILVER &&
4980 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
4981 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
4985 return (dsl_scan(spa->spa_dsl_pool, func));
4989 * ==========================================================================
4990 * SPA async task processing
4991 * ==========================================================================
4995 spa_async_remove(spa_t *spa, vdev_t *vd)
4999 if (vd->vdev_remove_wanted) {
5000 vd->vdev_remove_wanted = B_FALSE;
5001 vd->vdev_delayed_close = B_FALSE;
5002 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5005 * We want to clear the stats, but we don't want to do a full
5006 * vdev_clear() as that will cause us to throw away
5007 * degraded/faulted state as well as attempt to reopen the
5008 * device, all of which is a waste.
5010 vd->vdev_stat.vs_read_errors = 0;
5011 vd->vdev_stat.vs_write_errors = 0;
5012 vd->vdev_stat.vs_checksum_errors = 0;
5014 vdev_state_dirty(vd->vdev_top);
5017 for (c = 0; c < vd->vdev_children; c++)
5018 spa_async_remove(spa, vd->vdev_child[c]);
5022 spa_async_probe(spa_t *spa, vdev_t *vd)
5026 if (vd->vdev_probe_wanted) {
5027 vd->vdev_probe_wanted = B_FALSE;
5028 vdev_reopen(vd); /* vdev_open() does the actual probe */
5031 for (c = 0; c < vd->vdev_children; c++)
5032 spa_async_probe(spa, vd->vdev_child[c]);
5036 spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5040 if (!spa->spa_autoexpand)
5043 for (c = 0; c < vd->vdev_children; c++) {
5044 vdev_t *cvd = vd->vdev_child[c];
5045 spa_async_autoexpand(spa, cvd);
5048 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5051 spa_event_notify(vd->vdev_spa, vd, FM_EREPORT_ZFS_DEVICE_AUTOEXPAND);
5055 spa_async_thread(spa_t *spa)
5059 ASSERT(spa->spa_sync_on);
5061 mutex_enter(&spa->spa_async_lock);
5062 tasks = spa->spa_async_tasks;
5063 spa->spa_async_tasks = 0;
5064 mutex_exit(&spa->spa_async_lock);
5067 * See if the config needs to be updated.
5069 if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5070 uint64_t old_space, new_space;
5072 mutex_enter(&spa_namespace_lock);
5073 old_space = metaslab_class_get_space(spa_normal_class(spa));
5074 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5075 new_space = metaslab_class_get_space(spa_normal_class(spa));
5076 mutex_exit(&spa_namespace_lock);
5079 * If the pool grew as a result of the config update,
5080 * then log an internal history event.
5082 if (new_space != old_space) {
5083 spa_history_log_internal(LOG_POOL_VDEV_ONLINE,
5085 "pool '%s' size: %llu(+%llu)",
5086 spa_name(spa), new_space, new_space - old_space);
5091 * See if any devices need to be marked REMOVED.
5093 if (tasks & SPA_ASYNC_REMOVE) {
5094 spa_vdev_state_enter(spa, SCL_NONE);
5095 spa_async_remove(spa, spa->spa_root_vdev);
5096 for (i = 0; i < spa->spa_l2cache.sav_count; i++)
5097 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5098 for (i = 0; i < spa->spa_spares.sav_count; i++)
5099 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5100 (void) spa_vdev_state_exit(spa, NULL, 0);
5103 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5104 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5105 spa_async_autoexpand(spa, spa->spa_root_vdev);
5106 spa_config_exit(spa, SCL_CONFIG, FTAG);
5110 * See if any devices need to be probed.
5112 if (tasks & SPA_ASYNC_PROBE) {
5113 spa_vdev_state_enter(spa, SCL_NONE);
5114 spa_async_probe(spa, spa->spa_root_vdev);
5115 (void) spa_vdev_state_exit(spa, NULL, 0);
5119 * If any devices are done replacing, detach them.
5121 if (tasks & SPA_ASYNC_RESILVER_DONE)
5122 spa_vdev_resilver_done(spa);
5125 * Kick off a resilver.
5127 if (tasks & SPA_ASYNC_RESILVER)
5128 dsl_resilver_restart(spa->spa_dsl_pool, 0);
5131 * Let the world know that we're done.
5133 mutex_enter(&spa->spa_async_lock);
5134 spa->spa_async_thread = NULL;
5135 cv_broadcast(&spa->spa_async_cv);
5136 mutex_exit(&spa->spa_async_lock);
5141 spa_async_suspend(spa_t *spa)
5143 mutex_enter(&spa->spa_async_lock);
5144 spa->spa_async_suspended++;
5145 while (spa->spa_async_thread != NULL)
5146 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5147 mutex_exit(&spa->spa_async_lock);
5151 spa_async_resume(spa_t *spa)
5153 mutex_enter(&spa->spa_async_lock);
5154 ASSERT(spa->spa_async_suspended != 0);
5155 spa->spa_async_suspended--;
5156 mutex_exit(&spa->spa_async_lock);
5160 spa_async_dispatch(spa_t *spa)
5162 mutex_enter(&spa->spa_async_lock);
5163 if (spa->spa_async_tasks && !spa->spa_async_suspended &&
5164 spa->spa_async_thread == NULL &&
5165 rootdir != NULL && !vn_is_readonly(rootdir))
5166 spa->spa_async_thread = thread_create(NULL, 0,
5167 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
5168 mutex_exit(&spa->spa_async_lock);
5172 spa_async_request(spa_t *spa, int task)
5174 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
5175 mutex_enter(&spa->spa_async_lock);
5176 spa->spa_async_tasks |= task;
5177 mutex_exit(&spa->spa_async_lock);
5181 * ==========================================================================
5182 * SPA syncing routines
5183 * ==========================================================================
5187 bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5190 bpobj_enqueue(bpo, bp, tx);
5195 spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5199 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
5205 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
5207 char *packed = NULL;
5212 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
5215 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5216 * information. This avoids the dbuf_will_dirty() path and
5217 * saves us a pre-read to get data we don't actually care about.
5219 bufsize = P2ROUNDUP(nvsize, SPA_CONFIG_BLOCKSIZE);
5220 packed = kmem_alloc(bufsize, KM_SLEEP);
5222 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
5224 bzero(packed + nvsize, bufsize - nvsize);
5226 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
5228 kmem_free(packed, bufsize);
5230 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
5231 dmu_buf_will_dirty(db, tx);
5232 *(uint64_t *)db->db_data = nvsize;
5233 dmu_buf_rele(db, FTAG);
5237 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
5238 const char *config, const char *entry)
5248 * Update the MOS nvlist describing the list of available devices.
5249 * spa_validate_aux() will have already made sure this nvlist is
5250 * valid and the vdevs are labeled appropriately.
5252 if (sav->sav_object == 0) {
5253 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
5254 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
5255 sizeof (uint64_t), tx);
5256 VERIFY(zap_update(spa->spa_meta_objset,
5257 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
5258 &sav->sav_object, tx) == 0);
5261 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5262 if (sav->sav_count == 0) {
5263 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
5265 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
5266 for (i = 0; i < sav->sav_count; i++)
5267 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
5268 B_FALSE, VDEV_CONFIG_L2CACHE);
5269 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
5270 sav->sav_count) == 0);
5271 for (i = 0; i < sav->sav_count; i++)
5272 nvlist_free(list[i]);
5273 kmem_free(list, sav->sav_count * sizeof (void *));
5276 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
5277 nvlist_free(nvroot);
5279 sav->sav_sync = B_FALSE;
5283 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
5287 if (list_is_empty(&spa->spa_config_dirty_list))
5290 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5292 config = spa_config_generate(spa, spa->spa_root_vdev,
5293 dmu_tx_get_txg(tx), B_FALSE);
5295 spa_config_exit(spa, SCL_STATE, FTAG);
5297 if (spa->spa_config_syncing)
5298 nvlist_free(spa->spa_config_syncing);
5299 spa->spa_config_syncing = config;
5301 spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
5305 * Set zpool properties.
5308 spa_sync_props(void *arg1, void *arg2, dmu_tx_t *tx)
5311 objset_t *mos = spa->spa_meta_objset;
5312 nvlist_t *nvp = arg2;
5317 const char *propname;
5318 zprop_type_t proptype;
5320 mutex_enter(&spa->spa_props_lock);
5323 while ((elem = nvlist_next_nvpair(nvp, elem))) {
5324 switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
5325 case ZPOOL_PROP_VERSION:
5327 * Only set version for non-zpool-creation cases
5328 * (set/import). spa_create() needs special care
5329 * for version setting.
5331 if (tx->tx_txg != TXG_INITIAL) {
5332 VERIFY(nvpair_value_uint64(elem,
5334 ASSERT(intval <= SPA_VERSION);
5335 ASSERT(intval >= spa_version(spa));
5336 spa->spa_uberblock.ub_version = intval;
5337 vdev_config_dirty(spa->spa_root_vdev);
5341 case ZPOOL_PROP_ALTROOT:
5343 * 'altroot' is a non-persistent property. It should
5344 * have been set temporarily at creation or import time.
5346 ASSERT(spa->spa_root != NULL);
5349 case ZPOOL_PROP_READONLY:
5350 case ZPOOL_PROP_CACHEFILE:
5352 * 'readonly' and 'cachefile' are also non-persisitent
5358 * Set pool property values in the poolprops mos object.
5360 if (spa->spa_pool_props_object == 0) {
5361 VERIFY((spa->spa_pool_props_object =
5362 zap_create(mos, DMU_OT_POOL_PROPS,
5363 DMU_OT_NONE, 0, tx)) > 0);
5365 VERIFY(zap_update(mos,
5366 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
5367 8, 1, &spa->spa_pool_props_object, tx)
5371 /* normalize the property name */
5372 propname = zpool_prop_to_name(prop);
5373 proptype = zpool_prop_get_type(prop);
5375 if (nvpair_type(elem) == DATA_TYPE_STRING) {
5376 ASSERT(proptype == PROP_TYPE_STRING);
5377 VERIFY(nvpair_value_string(elem, &strval) == 0);
5378 VERIFY(zap_update(mos,
5379 spa->spa_pool_props_object, propname,
5380 1, strlen(strval) + 1, strval, tx) == 0);
5382 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
5383 VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5385 if (proptype == PROP_TYPE_INDEX) {
5387 VERIFY(zpool_prop_index_to_string(
5388 prop, intval, &unused) == 0);
5390 VERIFY(zap_update(mos,
5391 spa->spa_pool_props_object, propname,
5392 8, 1, &intval, tx) == 0);
5394 ASSERT(0); /* not allowed */
5398 case ZPOOL_PROP_DELEGATION:
5399 spa->spa_delegation = intval;
5401 case ZPOOL_PROP_BOOTFS:
5402 spa->spa_bootfs = intval;
5404 case ZPOOL_PROP_FAILUREMODE:
5405 spa->spa_failmode = intval;
5407 case ZPOOL_PROP_AUTOEXPAND:
5408 spa->spa_autoexpand = intval;
5409 if (tx->tx_txg != TXG_INITIAL)
5410 spa_async_request(spa,
5411 SPA_ASYNC_AUTOEXPAND);
5413 case ZPOOL_PROP_DEDUPDITTO:
5414 spa->spa_dedup_ditto = intval;
5421 /* log internal history if this is not a zpool create */
5422 if (spa_version(spa) >= SPA_VERSION_ZPOOL_HISTORY &&
5423 tx->tx_txg != TXG_INITIAL) {
5424 spa_history_log_internal(LOG_POOL_PROPSET,
5425 spa, tx, "%s %lld %s",
5426 nvpair_name(elem), intval, spa_name(spa));
5430 mutex_exit(&spa->spa_props_lock);
5434 * Perform one-time upgrade on-disk changes. spa_version() does not
5435 * reflect the new version this txg, so there must be no changes this
5436 * txg to anything that the upgrade code depends on after it executes.
5437 * Therefore this must be called after dsl_pool_sync() does the sync
5441 spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
5443 dsl_pool_t *dp = spa->spa_dsl_pool;
5445 ASSERT(spa->spa_sync_pass == 1);
5447 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
5448 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
5449 dsl_pool_create_origin(dp, tx);
5451 /* Keeping the origin open increases spa_minref */
5452 spa->spa_minref += 3;
5455 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
5456 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
5457 dsl_pool_upgrade_clones(dp, tx);
5460 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
5461 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
5462 dsl_pool_upgrade_dir_clones(dp, tx);
5464 /* Keeping the freedir open increases spa_minref */
5465 spa->spa_minref += 3;
5470 * Sync the specified transaction group. New blocks may be dirtied as
5471 * part of the process, so we iterate until it converges.
5474 spa_sync(spa_t *spa, uint64_t txg)
5476 dsl_pool_t *dp = spa->spa_dsl_pool;
5477 objset_t *mos = spa->spa_meta_objset;
5478 bpobj_t *defer_bpo = &spa->spa_deferred_bpobj;
5479 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
5480 vdev_t *rvd = spa->spa_root_vdev;
5486 VERIFY(spa_writeable(spa));
5489 * Lock out configuration changes.
5491 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5493 spa->spa_syncing_txg = txg;
5494 spa->spa_sync_pass = 0;
5497 * If there are any pending vdev state changes, convert them
5498 * into config changes that go out with this transaction group.
5500 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5501 while (list_head(&spa->spa_state_dirty_list) != NULL) {
5503 * We need the write lock here because, for aux vdevs,
5504 * calling vdev_config_dirty() modifies sav_config.
5505 * This is ugly and will become unnecessary when we
5506 * eliminate the aux vdev wart by integrating all vdevs
5507 * into the root vdev tree.
5509 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
5510 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
5511 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
5512 vdev_state_clean(vd);
5513 vdev_config_dirty(vd);
5515 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
5516 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
5518 spa_config_exit(spa, SCL_STATE, FTAG);
5520 tx = dmu_tx_create_assigned(dp, txg);
5523 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
5524 * set spa_deflate if we have no raid-z vdevs.
5526 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
5527 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
5530 for (i = 0; i < rvd->vdev_children; i++) {
5531 vd = rvd->vdev_child[i];
5532 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
5535 if (i == rvd->vdev_children) {
5536 spa->spa_deflate = TRUE;
5537 VERIFY(0 == zap_add(spa->spa_meta_objset,
5538 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
5539 sizeof (uint64_t), 1, &spa->spa_deflate, tx));
5544 * If anything has changed in this txg, or if someone is waiting
5545 * for this txg to sync (eg, spa_vdev_remove()), push the
5546 * deferred frees from the previous txg. If not, leave them
5547 * alone so that we don't generate work on an otherwise idle
5550 if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
5551 !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
5552 !txg_list_empty(&dp->dp_sync_tasks, txg) ||
5553 ((dsl_scan_active(dp->dp_scan) ||
5554 txg_sync_waiting(dp)) && !spa_shutting_down(spa))) {
5555 zio_t *zio = zio_root(spa, NULL, NULL, 0);
5556 VERIFY3U(bpobj_iterate(defer_bpo,
5557 spa_free_sync_cb, zio, tx), ==, 0);
5558 VERIFY3U(zio_wait(zio), ==, 0);
5562 * Iterate to convergence.
5565 int pass = ++spa->spa_sync_pass;
5567 spa_sync_config_object(spa, tx);
5568 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
5569 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
5570 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
5571 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
5572 spa_errlog_sync(spa, txg);
5573 dsl_pool_sync(dp, txg);
5575 if (pass <= SYNC_PASS_DEFERRED_FREE) {
5576 zio_t *zio = zio_root(spa, NULL, NULL, 0);
5577 bplist_iterate(free_bpl, spa_free_sync_cb,
5579 VERIFY(zio_wait(zio) == 0);
5581 bplist_iterate(free_bpl, bpobj_enqueue_cb,
5586 dsl_scan_sync(dp, tx);
5588 while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)))
5592 spa_sync_upgrades(spa, tx);
5594 } while (dmu_objset_is_dirty(mos, txg));
5597 * Rewrite the vdev configuration (which includes the uberblock)
5598 * to commit the transaction group.
5600 * If there are no dirty vdevs, we sync the uberblock to a few
5601 * random top-level vdevs that are known to be visible in the
5602 * config cache (see spa_vdev_add() for a complete description).
5603 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
5607 * We hold SCL_STATE to prevent vdev open/close/etc.
5608 * while we're attempting to write the vdev labels.
5610 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5612 if (list_is_empty(&spa->spa_config_dirty_list)) {
5613 vdev_t *svd[SPA_DVAS_PER_BP];
5615 int children = rvd->vdev_children;
5616 int c0 = spa_get_random(children);
5618 for (c = 0; c < children; c++) {
5619 vd = rvd->vdev_child[(c0 + c) % children];
5620 if (vd->vdev_ms_array == 0 || vd->vdev_islog)
5622 svd[svdcount++] = vd;
5623 if (svdcount == SPA_DVAS_PER_BP)
5626 error = vdev_config_sync(svd, svdcount, txg, B_FALSE);
5628 error = vdev_config_sync(svd, svdcount, txg,
5631 error = vdev_config_sync(rvd->vdev_child,
5632 rvd->vdev_children, txg, B_FALSE);
5634 error = vdev_config_sync(rvd->vdev_child,
5635 rvd->vdev_children, txg, B_TRUE);
5638 spa_config_exit(spa, SCL_STATE, FTAG);
5642 zio_suspend(spa, NULL);
5643 zio_resume_wait(spa);
5648 * Clear the dirty config list.
5650 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
5651 vdev_config_clean(vd);
5654 * Now that the new config has synced transactionally,
5655 * let it become visible to the config cache.
5657 if (spa->spa_config_syncing != NULL) {
5658 spa_config_set(spa, spa->spa_config_syncing);
5659 spa->spa_config_txg = txg;
5660 spa->spa_config_syncing = NULL;
5663 spa->spa_ubsync = spa->spa_uberblock;
5665 dsl_pool_sync_done(dp, txg);
5668 * Update usable space statistics.
5670 while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg))))
5671 vdev_sync_done(vd, txg);
5673 spa_update_dspace(spa);
5676 * It had better be the case that we didn't dirty anything
5677 * since vdev_config_sync().
5679 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
5680 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
5681 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
5683 spa->spa_sync_pass = 0;
5685 spa_config_exit(spa, SCL_CONFIG, FTAG);
5687 spa_handle_ignored_writes(spa);
5690 * If any async tasks have been requested, kick them off.
5692 spa_async_dispatch(spa);
5696 * Sync all pools. We don't want to hold the namespace lock across these
5697 * operations, so we take a reference on the spa_t and drop the lock during the
5701 spa_sync_allpools(void)
5704 mutex_enter(&spa_namespace_lock);
5705 while ((spa = spa_next(spa)) != NULL) {
5706 if (spa_state(spa) != POOL_STATE_ACTIVE ||
5707 !spa_writeable(spa) || spa_suspended(spa))
5709 spa_open_ref(spa, FTAG);
5710 mutex_exit(&spa_namespace_lock);
5711 txg_wait_synced(spa_get_dsl(spa), 0);
5712 mutex_enter(&spa_namespace_lock);
5713 spa_close(spa, FTAG);
5715 mutex_exit(&spa_namespace_lock);
5719 * ==========================================================================
5720 * Miscellaneous routines
5721 * ==========================================================================
5725 * Remove all pools in the system.
5733 * Remove all cached state. All pools should be closed now,
5734 * so every spa in the AVL tree should be unreferenced.
5736 mutex_enter(&spa_namespace_lock);
5737 while ((spa = spa_next(NULL)) != NULL) {
5739 * Stop async tasks. The async thread may need to detach
5740 * a device that's been replaced, which requires grabbing
5741 * spa_namespace_lock, so we must drop it here.
5743 spa_open_ref(spa, FTAG);
5744 mutex_exit(&spa_namespace_lock);
5745 spa_async_suspend(spa);
5746 mutex_enter(&spa_namespace_lock);
5747 spa_close(spa, FTAG);
5749 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
5751 spa_deactivate(spa);
5755 mutex_exit(&spa_namespace_lock);
5759 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
5764 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
5768 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
5769 vd = spa->spa_l2cache.sav_vdevs[i];
5770 if (vd->vdev_guid == guid)
5774 for (i = 0; i < spa->spa_spares.sav_count; i++) {
5775 vd = spa->spa_spares.sav_vdevs[i];
5776 if (vd->vdev_guid == guid)
5785 spa_upgrade(spa_t *spa, uint64_t version)
5787 ASSERT(spa_writeable(spa));
5789 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5792 * This should only be called for a non-faulted pool, and since a
5793 * future version would result in an unopenable pool, this shouldn't be
5796 ASSERT(spa->spa_uberblock.ub_version <= SPA_VERSION);
5797 ASSERT(version >= spa->spa_uberblock.ub_version);
5799 spa->spa_uberblock.ub_version = version;
5800 vdev_config_dirty(spa->spa_root_vdev);
5802 spa_config_exit(spa, SCL_ALL, FTAG);
5804 txg_wait_synced(spa_get_dsl(spa), 0);
5808 spa_has_spare(spa_t *spa, uint64_t guid)
5812 spa_aux_vdev_t *sav = &spa->spa_spares;
5814 for (i = 0; i < sav->sav_count; i++)
5815 if (sav->sav_vdevs[i]->vdev_guid == guid)
5818 for (i = 0; i < sav->sav_npending; i++) {
5819 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
5820 &spareguid) == 0 && spareguid == guid)
5828 * Check if a pool has an active shared spare device.
5829 * Note: reference count of an active spare is 2, as a spare and as a replace
5832 spa_has_active_shared_spare(spa_t *spa)
5836 spa_aux_vdev_t *sav = &spa->spa_spares;
5838 for (i = 0; i < sav->sav_count; i++) {
5839 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
5840 &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
5849 * Post a FM_EREPORT_ZFS_* event from sys/fm/fs/zfs.h. The payload will be
5850 * filled in from the spa and (optionally) the vdev. This doesn't do anything
5851 * in the userland libzpool, as we don't want consumers to misinterpret ztest
5852 * or zdb as real changes.
5855 spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
5858 zfs_ereport_post(name, spa, vd, NULL, 0, 0);