4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
25 * Copyright (c) 2012 by Delphix. All rights reserved.
29 * This file contains all the routines used when modifying on-disk SPA state.
30 * This includes opening, importing, destroying, exporting a pool, and syncing a
34 #include <sys/zfs_context.h>
35 #include <sys/fm/fs/zfs.h>
36 #include <sys/spa_impl.h>
38 #include <sys/zio_checksum.h>
40 #include <sys/dmu_tx.h>
44 #include <sys/vdev_impl.h>
45 #include <sys/vdev_disk.h>
46 #include <sys/metaslab.h>
47 #include <sys/metaslab_impl.h>
48 #include <sys/uberblock_impl.h>
51 #include <sys/dmu_traverse.h>
52 #include <sys/dmu_objset.h>
53 #include <sys/unique.h>
54 #include <sys/dsl_pool.h>
55 #include <sys/dsl_dataset.h>
56 #include <sys/dsl_dir.h>
57 #include <sys/dsl_prop.h>
58 #include <sys/dsl_synctask.h>
59 #include <sys/fs/zfs.h>
61 #include <sys/callb.h>
62 #include <sys/systeminfo.h>
63 #include <sys/spa_boot.h>
64 #include <sys/zfs_ioctl.h>
65 #include <sys/dsl_scan.h>
66 #include <sys/zfeature.h>
69 #include <sys/bootprops.h>
70 #include <sys/callb.h>
71 #include <sys/cpupart.h>
73 #include <sys/sysdc.h>
78 #include "zfs_comutil.h"
80 typedef enum zti_modes {
81 zti_mode_fixed, /* value is # of threads (min 1) */
82 zti_mode_online_percent, /* value is % of online CPUs */
83 zti_mode_batch, /* cpu-intensive; value is ignored */
84 zti_mode_null, /* don't create a taskq */
88 #define ZTI_FIX(n) { zti_mode_fixed, (n) }
89 #define ZTI_PCT(n) { zti_mode_online_percent, (n) }
90 #define ZTI_BATCH { zti_mode_batch, 0 }
91 #define ZTI_NULL { zti_mode_null, 0 }
93 #define ZTI_ONE ZTI_FIX(1)
95 typedef struct zio_taskq_info {
96 enum zti_modes zti_mode;
100 static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
101 "iss", "iss_h", "int", "int_h"
105 * Define the taskq threads for the following I/O types:
106 * NULL, READ, WRITE, FREE, CLAIM, and IOCTL
108 const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
109 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
110 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL },
111 { ZTI_FIX(8), ZTI_NULL, ZTI_BATCH, ZTI_NULL },
112 { ZTI_BATCH, ZTI_FIX(5), ZTI_FIX(16), ZTI_FIX(5) },
113 { ZTI_FIX(8), ZTI_NULL, ZTI_ONE, ZTI_NULL },
114 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL },
115 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL },
118 static dsl_syncfunc_t spa_sync_version;
119 static dsl_syncfunc_t spa_sync_props;
120 static dsl_checkfunc_t spa_change_guid_check;
121 static dsl_syncfunc_t spa_change_guid_sync;
122 static boolean_t spa_has_active_shared_spare(spa_t *spa);
123 static inline int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
124 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
126 static void spa_vdev_resilver_done(spa_t *spa);
128 uint_t zio_taskq_batch_pct = 100; /* 1 thread per cpu in pset */
129 id_t zio_taskq_psrset_bind = PS_NONE;
130 boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */
131 uint_t zio_taskq_basedc = 80; /* base duty cycle */
133 boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */
136 * This (illegal) pool name is used when temporarily importing a spa_t in order
137 * to get the vdev stats associated with the imported devices.
139 #define TRYIMPORT_NAME "$import"
142 * ==========================================================================
143 * SPA properties routines
144 * ==========================================================================
148 * Add a (source=src, propname=propval) list to an nvlist.
151 spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
152 uint64_t intval, zprop_source_t src)
154 const char *propname = zpool_prop_to_name(prop);
157 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_PUSHPAGE) == 0);
158 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
161 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
163 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
165 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
166 nvlist_free(propval);
170 * Get property values from the spa configuration.
173 spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
175 vdev_t *rvd = spa->spa_root_vdev;
176 dsl_pool_t *pool = spa->spa_dsl_pool;
180 uint64_t cap, version;
181 zprop_source_t src = ZPROP_SRC_NONE;
182 spa_config_dirent_t *dp;
185 ASSERT(MUTEX_HELD(&spa->spa_props_lock));
188 alloc = metaslab_class_get_alloc(spa_normal_class(spa));
189 size = metaslab_class_get_space(spa_normal_class(spa));
190 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
191 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
192 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
193 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
197 for (c = 0; c < rvd->vdev_children; c++) {
198 vdev_t *tvd = rvd->vdev_child[c];
199 space += tvd->vdev_max_asize - tvd->vdev_asize;
201 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, space,
204 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
205 (spa_mode(spa) == FREAD), src);
207 cap = (size == 0) ? 0 : (alloc * 100 / size);
208 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
210 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
211 ddt_get_pool_dedup_ratio(spa), src);
213 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
214 rvd->vdev_state, src);
216 version = spa_version(spa);
217 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
218 src = ZPROP_SRC_DEFAULT;
220 src = ZPROP_SRC_LOCAL;
221 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
225 dsl_dir_t *freedir = pool->dp_free_dir;
228 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
229 * when opening pools before this version freedir will be NULL.
231 if (freedir != NULL) {
232 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
233 freedir->dd_phys->dd_used_bytes, src);
235 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
240 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
242 if (spa->spa_comment != NULL) {
243 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
247 if (spa->spa_root != NULL)
248 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
251 if ((dp = list_head(&spa->spa_config_list)) != NULL) {
252 if (dp->scd_path == NULL) {
253 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
254 "none", 0, ZPROP_SRC_LOCAL);
255 } else if (strcmp(dp->scd_path, spa_config_path) != 0) {
256 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
257 dp->scd_path, 0, ZPROP_SRC_LOCAL);
263 * Get zpool property values.
266 spa_prop_get(spa_t *spa, nvlist_t **nvp)
268 objset_t *mos = spa->spa_meta_objset;
273 err = nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_PUSHPAGE);
277 mutex_enter(&spa->spa_props_lock);
280 * Get properties from the spa config.
282 spa_prop_get_config(spa, nvp);
284 /* If no pool property object, no more prop to get. */
285 if (mos == NULL || spa->spa_pool_props_object == 0) {
286 mutex_exit(&spa->spa_props_lock);
291 * Get properties from the MOS pool property object.
293 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
294 (err = zap_cursor_retrieve(&zc, &za)) == 0;
295 zap_cursor_advance(&zc)) {
298 zprop_source_t src = ZPROP_SRC_DEFAULT;
301 if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
304 switch (za.za_integer_length) {
306 /* integer property */
307 if (za.za_first_integer !=
308 zpool_prop_default_numeric(prop))
309 src = ZPROP_SRC_LOCAL;
311 if (prop == ZPOOL_PROP_BOOTFS) {
313 dsl_dataset_t *ds = NULL;
315 dp = spa_get_dsl(spa);
316 rw_enter(&dp->dp_config_rwlock, RW_READER);
317 if ((err = dsl_dataset_hold_obj(dp,
318 za.za_first_integer, FTAG, &ds))) {
319 rw_exit(&dp->dp_config_rwlock);
324 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
326 dsl_dataset_name(ds, strval);
327 dsl_dataset_rele(ds, FTAG);
328 rw_exit(&dp->dp_config_rwlock);
331 intval = za.za_first_integer;
334 spa_prop_add_list(*nvp, prop, strval, intval, src);
338 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
343 /* string property */
344 strval = kmem_alloc(za.za_num_integers, KM_PUSHPAGE);
345 err = zap_lookup(mos, spa->spa_pool_props_object,
346 za.za_name, 1, za.za_num_integers, strval);
348 kmem_free(strval, za.za_num_integers);
351 spa_prop_add_list(*nvp, prop, strval, 0, src);
352 kmem_free(strval, za.za_num_integers);
359 zap_cursor_fini(&zc);
360 mutex_exit(&spa->spa_props_lock);
362 if (err && err != ENOENT) {
372 * Validate the given pool properties nvlist and modify the list
373 * for the property values to be set.
376 spa_prop_validate(spa_t *spa, nvlist_t *props)
379 int error = 0, reset_bootfs = 0;
381 boolean_t has_feature = B_FALSE;
384 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
386 char *strval, *slash, *check, *fname;
387 const char *propname = nvpair_name(elem);
388 zpool_prop_t prop = zpool_name_to_prop(propname);
392 if (!zpool_prop_feature(propname)) {
398 * Sanitize the input.
400 if (nvpair_type(elem) != DATA_TYPE_UINT64) {
405 if (nvpair_value_uint64(elem, &intval) != 0) {
415 fname = strchr(propname, '@') + 1;
416 if (zfeature_lookup_name(fname, NULL) != 0) {
421 has_feature = B_TRUE;
424 case ZPOOL_PROP_VERSION:
425 error = nvpair_value_uint64(elem, &intval);
427 (intval < spa_version(spa) ||
428 intval > SPA_VERSION_BEFORE_FEATURES ||
433 case ZPOOL_PROP_DELEGATION:
434 case ZPOOL_PROP_AUTOREPLACE:
435 case ZPOOL_PROP_LISTSNAPS:
436 case ZPOOL_PROP_AUTOEXPAND:
437 error = nvpair_value_uint64(elem, &intval);
438 if (!error && intval > 1)
442 case ZPOOL_PROP_BOOTFS:
444 * If the pool version is less than SPA_VERSION_BOOTFS,
445 * or the pool is still being created (version == 0),
446 * the bootfs property cannot be set.
448 if (spa_version(spa) < SPA_VERSION_BOOTFS) {
454 * Make sure the vdev config is bootable
456 if (!vdev_is_bootable(spa->spa_root_vdev)) {
463 error = nvpair_value_string(elem, &strval);
469 if (strval == NULL || strval[0] == '\0') {
470 objnum = zpool_prop_default_numeric(
475 if ((error = dmu_objset_hold(strval,FTAG,&os)))
478 /* Must be ZPL and not gzip compressed. */
480 if (dmu_objset_type(os) != DMU_OST_ZFS) {
482 } else if ((error = dsl_prop_get_integer(strval,
483 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
484 &compress, NULL)) == 0 &&
485 !BOOTFS_COMPRESS_VALID(compress)) {
488 objnum = dmu_objset_id(os);
490 dmu_objset_rele(os, FTAG);
494 case ZPOOL_PROP_FAILUREMODE:
495 error = nvpair_value_uint64(elem, &intval);
496 if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
497 intval > ZIO_FAILURE_MODE_PANIC))
501 * This is a special case which only occurs when
502 * the pool has completely failed. This allows
503 * the user to change the in-core failmode property
504 * without syncing it out to disk (I/Os might
505 * currently be blocked). We do this by returning
506 * EIO to the caller (spa_prop_set) to trick it
507 * into thinking we encountered a property validation
510 if (!error && spa_suspended(spa)) {
511 spa->spa_failmode = intval;
516 case ZPOOL_PROP_CACHEFILE:
517 if ((error = nvpair_value_string(elem, &strval)) != 0)
520 if (strval[0] == '\0')
523 if (strcmp(strval, "none") == 0)
526 if (strval[0] != '/') {
531 slash = strrchr(strval, '/');
532 ASSERT(slash != NULL);
534 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
535 strcmp(slash, "/..") == 0)
539 case ZPOOL_PROP_COMMENT:
540 if ((error = nvpair_value_string(elem, &strval)) != 0)
542 for (check = strval; *check != '\0'; check++) {
543 if (!isprint(*check)) {
549 if (strlen(strval) > ZPROP_MAX_COMMENT)
553 case ZPOOL_PROP_DEDUPDITTO:
554 if (spa_version(spa) < SPA_VERSION_DEDUP)
557 error = nvpair_value_uint64(elem, &intval);
559 intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
571 if (!error && reset_bootfs) {
572 error = nvlist_remove(props,
573 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
576 error = nvlist_add_uint64(props,
577 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
585 spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
588 spa_config_dirent_t *dp;
590 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
594 dp = kmem_alloc(sizeof (spa_config_dirent_t),
597 if (cachefile[0] == '\0')
598 dp->scd_path = spa_strdup(spa_config_path);
599 else if (strcmp(cachefile, "none") == 0)
602 dp->scd_path = spa_strdup(cachefile);
604 list_insert_head(&spa->spa_config_list, dp);
606 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
610 spa_prop_set(spa_t *spa, nvlist_t *nvp)
613 nvpair_t *elem = NULL;
614 boolean_t need_sync = B_FALSE;
616 if ((error = spa_prop_validate(spa, nvp)) != 0)
619 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
620 zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem));
622 if (prop == ZPOOL_PROP_CACHEFILE ||
623 prop == ZPOOL_PROP_ALTROOT ||
624 prop == ZPOOL_PROP_READONLY)
627 if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) {
630 if (prop == ZPOOL_PROP_VERSION) {
631 VERIFY(nvpair_value_uint64(elem, &ver) == 0);
633 ASSERT(zpool_prop_feature(nvpair_name(elem)));
634 ver = SPA_VERSION_FEATURES;
638 /* Save time if the version is already set. */
639 if (ver == spa_version(spa))
643 * In addition to the pool directory object, we might
644 * create the pool properties object, the features for
645 * read object, the features for write object, or the
646 * feature descriptions object.
648 error = dsl_sync_task_do(spa_get_dsl(spa), NULL,
649 spa_sync_version, spa, &ver, 6);
660 return (dsl_sync_task_do(spa_get_dsl(spa), NULL, spa_sync_props,
668 * If the bootfs property value is dsobj, clear it.
671 spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
673 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
674 VERIFY(zap_remove(spa->spa_meta_objset,
675 spa->spa_pool_props_object,
676 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
683 spa_change_guid_check(void *arg1, void *arg2, dmu_tx_t *tx)
686 vdev_t *rvd = spa->spa_root_vdev;
688 ASSERTV(uint64_t *newguid = arg2);
690 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
691 vdev_state = rvd->vdev_state;
692 spa_config_exit(spa, SCL_STATE, FTAG);
694 if (vdev_state != VDEV_STATE_HEALTHY)
697 ASSERT3U(spa_guid(spa), !=, *newguid);
703 spa_change_guid_sync(void *arg1, void *arg2, dmu_tx_t *tx)
706 uint64_t *newguid = arg2;
708 vdev_t *rvd = spa->spa_root_vdev;
710 oldguid = spa_guid(spa);
712 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
713 rvd->vdev_guid = *newguid;
714 rvd->vdev_guid_sum += (*newguid - oldguid);
715 vdev_config_dirty(rvd);
716 spa_config_exit(spa, SCL_STATE, FTAG);
718 spa_history_log_internal(LOG_POOL_GUID_CHANGE, spa, tx,
719 "old=%lld new=%lld", oldguid, *newguid);
723 * Change the GUID for the pool. This is done so that we can later
724 * re-import a pool built from a clone of our own vdevs. We will modify
725 * the root vdev's guid, our own pool guid, and then mark all of our
726 * vdevs dirty. Note that we must make sure that all our vdevs are
727 * online when we do this, or else any vdevs that weren't present
728 * would be orphaned from our pool. We are also going to issue a
729 * sysevent to update any watchers.
732 spa_change_guid(spa_t *spa)
737 mutex_enter(&spa_namespace_lock);
738 guid = spa_generate_guid(NULL);
740 error = dsl_sync_task_do(spa_get_dsl(spa), spa_change_guid_check,
741 spa_change_guid_sync, spa, &guid, 5);
744 spa_config_sync(spa, B_FALSE, B_TRUE);
745 spa_event_notify(spa, NULL, FM_EREPORT_ZFS_POOL_REGUID);
748 mutex_exit(&spa_namespace_lock);
754 * ==========================================================================
755 * SPA state manipulation (open/create/destroy/import/export)
756 * ==========================================================================
760 spa_error_entry_compare(const void *a, const void *b)
762 spa_error_entry_t *sa = (spa_error_entry_t *)a;
763 spa_error_entry_t *sb = (spa_error_entry_t *)b;
766 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
767 sizeof (zbookmark_t));
778 * Utility function which retrieves copies of the current logs and
779 * re-initializes them in the process.
782 spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
784 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
786 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
787 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
789 avl_create(&spa->spa_errlist_scrub,
790 spa_error_entry_compare, sizeof (spa_error_entry_t),
791 offsetof(spa_error_entry_t, se_avl));
792 avl_create(&spa->spa_errlist_last,
793 spa_error_entry_compare, sizeof (spa_error_entry_t),
794 offsetof(spa_error_entry_t, se_avl));
798 spa_taskq_create(spa_t *spa, const char *name, enum zti_modes mode,
801 uint_t flags = TASKQ_PREPOPULATE;
802 boolean_t batch = B_FALSE;
806 return (NULL); /* no taskq needed */
809 ASSERT3U(value, >=, 1);
810 value = MAX(value, 1);
815 flags |= TASKQ_THREADS_CPU_PCT;
816 value = zio_taskq_batch_pct;
819 case zti_mode_online_percent:
820 flags |= TASKQ_THREADS_CPU_PCT;
824 panic("unrecognized mode for %s taskq (%u:%u) in "
830 if (zio_taskq_sysdc && spa->spa_proc != &p0) {
832 flags |= TASKQ_DC_BATCH;
834 return (taskq_create_sysdc(name, value, 50, INT_MAX,
835 spa->spa_proc, zio_taskq_basedc, flags));
837 return (taskq_create_proc(name, value, maxclsyspri, 50, INT_MAX,
838 spa->spa_proc, flags));
842 spa_create_zio_taskqs(spa_t *spa)
846 for (t = 0; t < ZIO_TYPES; t++) {
847 for (q = 0; q < ZIO_TASKQ_TYPES; q++) {
848 const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
849 enum zti_modes mode = ztip->zti_mode;
850 uint_t value = ztip->zti_value;
853 (void) snprintf(name, sizeof (name),
854 "%s_%s", zio_type_name[t], zio_taskq_types[q]);
856 spa->spa_zio_taskq[t][q] =
857 spa_taskq_create(spa, name, mode, value);
862 #if defined(_KERNEL) && defined(HAVE_SPA_THREAD)
864 spa_thread(void *arg)
869 user_t *pu = PTOU(curproc);
871 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
874 ASSERT(curproc != &p0);
875 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
876 "zpool-%s", spa->spa_name);
877 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
879 /* bind this thread to the requested psrset */
880 if (zio_taskq_psrset_bind != PS_NONE) {
882 mutex_enter(&cpu_lock);
883 mutex_enter(&pidlock);
884 mutex_enter(&curproc->p_lock);
886 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
887 0, NULL, NULL) == 0) {
888 curthread->t_bind_pset = zio_taskq_psrset_bind;
891 "Couldn't bind process for zfs pool \"%s\" to "
892 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
895 mutex_exit(&curproc->p_lock);
896 mutex_exit(&pidlock);
897 mutex_exit(&cpu_lock);
901 if (zio_taskq_sysdc) {
902 sysdc_thread_enter(curthread, 100, 0);
905 spa->spa_proc = curproc;
906 spa->spa_did = curthread->t_did;
908 spa_create_zio_taskqs(spa);
910 mutex_enter(&spa->spa_proc_lock);
911 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
913 spa->spa_proc_state = SPA_PROC_ACTIVE;
914 cv_broadcast(&spa->spa_proc_cv);
916 CALLB_CPR_SAFE_BEGIN(&cprinfo);
917 while (spa->spa_proc_state == SPA_PROC_ACTIVE)
918 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
919 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
921 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
922 spa->spa_proc_state = SPA_PROC_GONE;
924 cv_broadcast(&spa->spa_proc_cv);
925 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */
927 mutex_enter(&curproc->p_lock);
933 * Activate an uninitialized pool.
936 spa_activate(spa_t *spa, int mode)
938 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
940 spa->spa_state = POOL_STATE_ACTIVE;
941 spa->spa_mode = mode;
943 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
944 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
946 /* Try to create a covering process */
947 mutex_enter(&spa->spa_proc_lock);
948 ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
949 ASSERT(spa->spa_proc == &p0);
952 #ifdef HAVE_SPA_THREAD
953 /* Only create a process if we're going to be around a while. */
954 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
955 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
957 spa->spa_proc_state = SPA_PROC_CREATED;
958 while (spa->spa_proc_state == SPA_PROC_CREATED) {
959 cv_wait(&spa->spa_proc_cv,
960 &spa->spa_proc_lock);
962 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
963 ASSERT(spa->spa_proc != &p0);
964 ASSERT(spa->spa_did != 0);
968 "Couldn't create process for zfs pool \"%s\"\n",
973 #endif /* HAVE_SPA_THREAD */
974 mutex_exit(&spa->spa_proc_lock);
976 /* If we didn't create a process, we need to create our taskqs. */
977 if (spa->spa_proc == &p0) {
978 spa_create_zio_taskqs(spa);
981 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
982 offsetof(vdev_t, vdev_config_dirty_node));
983 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
984 offsetof(vdev_t, vdev_state_dirty_node));
986 txg_list_create(&spa->spa_vdev_txg_list,
987 offsetof(struct vdev, vdev_txg_node));
989 avl_create(&spa->spa_errlist_scrub,
990 spa_error_entry_compare, sizeof (spa_error_entry_t),
991 offsetof(spa_error_entry_t, se_avl));
992 avl_create(&spa->spa_errlist_last,
993 spa_error_entry_compare, sizeof (spa_error_entry_t),
994 offsetof(spa_error_entry_t, se_avl));
998 * Opposite of spa_activate().
1001 spa_deactivate(spa_t *spa)
1005 ASSERT(spa->spa_sync_on == B_FALSE);
1006 ASSERT(spa->spa_dsl_pool == NULL);
1007 ASSERT(spa->spa_root_vdev == NULL);
1008 ASSERT(spa->spa_async_zio_root == NULL);
1009 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
1011 txg_list_destroy(&spa->spa_vdev_txg_list);
1013 list_destroy(&spa->spa_config_dirty_list);
1014 list_destroy(&spa->spa_state_dirty_list);
1016 taskq_cancel_id(system_taskq, spa->spa_deadman_tqid);
1018 for (t = 0; t < ZIO_TYPES; t++) {
1019 for (q = 0; q < ZIO_TASKQ_TYPES; q++) {
1020 if (spa->spa_zio_taskq[t][q] != NULL)
1021 taskq_destroy(spa->spa_zio_taskq[t][q]);
1022 spa->spa_zio_taskq[t][q] = NULL;
1026 metaslab_class_destroy(spa->spa_normal_class);
1027 spa->spa_normal_class = NULL;
1029 metaslab_class_destroy(spa->spa_log_class);
1030 spa->spa_log_class = NULL;
1033 * If this was part of an import or the open otherwise failed, we may
1034 * still have errors left in the queues. Empty them just in case.
1036 spa_errlog_drain(spa);
1038 avl_destroy(&spa->spa_errlist_scrub);
1039 avl_destroy(&spa->spa_errlist_last);
1041 spa->spa_state = POOL_STATE_UNINITIALIZED;
1043 mutex_enter(&spa->spa_proc_lock);
1044 if (spa->spa_proc_state != SPA_PROC_NONE) {
1045 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1046 spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1047 cv_broadcast(&spa->spa_proc_cv);
1048 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1049 ASSERT(spa->spa_proc != &p0);
1050 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1052 ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1053 spa->spa_proc_state = SPA_PROC_NONE;
1055 ASSERT(spa->spa_proc == &p0);
1056 mutex_exit(&spa->spa_proc_lock);
1059 * We want to make sure spa_thread() has actually exited the ZFS
1060 * module, so that the module can't be unloaded out from underneath
1063 if (spa->spa_did != 0) {
1064 thread_join(spa->spa_did);
1070 * Verify a pool configuration, and construct the vdev tree appropriately. This
1071 * will create all the necessary vdevs in the appropriate layout, with each vdev
1072 * in the CLOSED state. This will prep the pool before open/creation/import.
1073 * All vdev validation is done by the vdev_alloc() routine.
1076 spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1077 uint_t id, int atype)
1084 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1087 if ((*vdp)->vdev_ops->vdev_op_leaf)
1090 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1093 if (error == ENOENT)
1102 for (c = 0; c < children; c++) {
1104 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1112 ASSERT(*vdp != NULL);
1118 * Opposite of spa_load().
1121 spa_unload(spa_t *spa)
1125 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1130 spa_async_suspend(spa);
1135 if (spa->spa_sync_on) {
1136 txg_sync_stop(spa->spa_dsl_pool);
1137 spa->spa_sync_on = B_FALSE;
1141 * Wait for any outstanding async I/O to complete.
1143 if (spa->spa_async_zio_root != NULL) {
1144 (void) zio_wait(spa->spa_async_zio_root);
1145 spa->spa_async_zio_root = NULL;
1148 bpobj_close(&spa->spa_deferred_bpobj);
1151 * Close the dsl pool.
1153 if (spa->spa_dsl_pool) {
1154 dsl_pool_close(spa->spa_dsl_pool);
1155 spa->spa_dsl_pool = NULL;
1156 spa->spa_meta_objset = NULL;
1161 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1164 * Drop and purge level 2 cache
1166 spa_l2cache_drop(spa);
1171 if (spa->spa_root_vdev)
1172 vdev_free(spa->spa_root_vdev);
1173 ASSERT(spa->spa_root_vdev == NULL);
1175 for (i = 0; i < spa->spa_spares.sav_count; i++)
1176 vdev_free(spa->spa_spares.sav_vdevs[i]);
1177 if (spa->spa_spares.sav_vdevs) {
1178 kmem_free(spa->spa_spares.sav_vdevs,
1179 spa->spa_spares.sav_count * sizeof (void *));
1180 spa->spa_spares.sav_vdevs = NULL;
1182 if (spa->spa_spares.sav_config) {
1183 nvlist_free(spa->spa_spares.sav_config);
1184 spa->spa_spares.sav_config = NULL;
1186 spa->spa_spares.sav_count = 0;
1188 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1189 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1190 vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1192 if (spa->spa_l2cache.sav_vdevs) {
1193 kmem_free(spa->spa_l2cache.sav_vdevs,
1194 spa->spa_l2cache.sav_count * sizeof (void *));
1195 spa->spa_l2cache.sav_vdevs = NULL;
1197 if (spa->spa_l2cache.sav_config) {
1198 nvlist_free(spa->spa_l2cache.sav_config);
1199 spa->spa_l2cache.sav_config = NULL;
1201 spa->spa_l2cache.sav_count = 0;
1203 spa->spa_async_suspended = 0;
1205 if (spa->spa_comment != NULL) {
1206 spa_strfree(spa->spa_comment);
1207 spa->spa_comment = NULL;
1210 spa_config_exit(spa, SCL_ALL, FTAG);
1214 * Load (or re-load) the current list of vdevs describing the active spares for
1215 * this pool. When this is called, we have some form of basic information in
1216 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1217 * then re-generate a more complete list including status information.
1220 spa_load_spares(spa_t *spa)
1227 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1230 * First, close and free any existing spare vdevs.
1232 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1233 vd = spa->spa_spares.sav_vdevs[i];
1235 /* Undo the call to spa_activate() below */
1236 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1237 B_FALSE)) != NULL && tvd->vdev_isspare)
1238 spa_spare_remove(tvd);
1243 if (spa->spa_spares.sav_vdevs)
1244 kmem_free(spa->spa_spares.sav_vdevs,
1245 spa->spa_spares.sav_count * sizeof (void *));
1247 if (spa->spa_spares.sav_config == NULL)
1250 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1251 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1253 spa->spa_spares.sav_count = (int)nspares;
1254 spa->spa_spares.sav_vdevs = NULL;
1260 * Construct the array of vdevs, opening them to get status in the
1261 * process. For each spare, there is potentially two different vdev_t
1262 * structures associated with it: one in the list of spares (used only
1263 * for basic validation purposes) and one in the active vdev
1264 * configuration (if it's spared in). During this phase we open and
1265 * validate each vdev on the spare list. If the vdev also exists in the
1266 * active configuration, then we also mark this vdev as an active spare.
1268 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1270 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1271 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1272 VDEV_ALLOC_SPARE) == 0);
1275 spa->spa_spares.sav_vdevs[i] = vd;
1277 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1278 B_FALSE)) != NULL) {
1279 if (!tvd->vdev_isspare)
1283 * We only mark the spare active if we were successfully
1284 * able to load the vdev. Otherwise, importing a pool
1285 * with a bad active spare would result in strange
1286 * behavior, because multiple pool would think the spare
1287 * is actively in use.
1289 * There is a vulnerability here to an equally bizarre
1290 * circumstance, where a dead active spare is later
1291 * brought back to life (onlined or otherwise). Given
1292 * the rarity of this scenario, and the extra complexity
1293 * it adds, we ignore the possibility.
1295 if (!vdev_is_dead(tvd))
1296 spa_spare_activate(tvd);
1300 vd->vdev_aux = &spa->spa_spares;
1302 if (vdev_open(vd) != 0)
1305 if (vdev_validate_aux(vd) == 0)
1310 * Recompute the stashed list of spares, with status information
1313 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1314 DATA_TYPE_NVLIST_ARRAY) == 0);
1316 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1318 for (i = 0; i < spa->spa_spares.sav_count; i++)
1319 spares[i] = vdev_config_generate(spa,
1320 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1321 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1322 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1323 for (i = 0; i < spa->spa_spares.sav_count; i++)
1324 nvlist_free(spares[i]);
1325 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1329 * Load (or re-load) the current list of vdevs describing the active l2cache for
1330 * this pool. When this is called, we have some form of basic information in
1331 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1332 * then re-generate a more complete list including status information.
1333 * Devices which are already active have their details maintained, and are
1337 spa_load_l2cache(spa_t *spa)
1341 int i, j, oldnvdevs;
1343 vdev_t *vd, **oldvdevs, **newvdevs = NULL;
1344 spa_aux_vdev_t *sav = &spa->spa_l2cache;
1346 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1348 if (sav->sav_config != NULL) {
1349 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1350 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1351 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_PUSHPAGE);
1356 oldvdevs = sav->sav_vdevs;
1357 oldnvdevs = sav->sav_count;
1358 sav->sav_vdevs = NULL;
1362 * Process new nvlist of vdevs.
1364 for (i = 0; i < nl2cache; i++) {
1365 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1369 for (j = 0; j < oldnvdevs; j++) {
1371 if (vd != NULL && guid == vd->vdev_guid) {
1373 * Retain previous vdev for add/remove ops.
1381 if (newvdevs[i] == NULL) {
1385 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1386 VDEV_ALLOC_L2CACHE) == 0);
1391 * Commit this vdev as an l2cache device,
1392 * even if it fails to open.
1394 spa_l2cache_add(vd);
1399 spa_l2cache_activate(vd);
1401 if (vdev_open(vd) != 0)
1404 (void) vdev_validate_aux(vd);
1406 if (!vdev_is_dead(vd))
1407 l2arc_add_vdev(spa, vd);
1412 * Purge vdevs that were dropped
1414 for (i = 0; i < oldnvdevs; i++) {
1419 ASSERT(vd->vdev_isl2cache);
1421 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1422 pool != 0ULL && l2arc_vdev_present(vd))
1423 l2arc_remove_vdev(vd);
1424 vdev_clear_stats(vd);
1430 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1432 if (sav->sav_config == NULL)
1435 sav->sav_vdevs = newvdevs;
1436 sav->sav_count = (int)nl2cache;
1439 * Recompute the stashed list of l2cache devices, with status
1440 * information this time.
1442 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1443 DATA_TYPE_NVLIST_ARRAY) == 0);
1445 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_PUSHPAGE);
1446 for (i = 0; i < sav->sav_count; i++)
1447 l2cache[i] = vdev_config_generate(spa,
1448 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1449 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1450 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1452 for (i = 0; i < sav->sav_count; i++)
1453 nvlist_free(l2cache[i]);
1455 kmem_free(l2cache, sav->sav_count * sizeof (void *));
1459 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1462 char *packed = NULL;
1467 error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db);
1471 nvsize = *(uint64_t *)db->db_data;
1472 dmu_buf_rele(db, FTAG);
1474 packed = kmem_alloc(nvsize, KM_PUSHPAGE | KM_NODEBUG);
1475 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1478 error = nvlist_unpack(packed, nvsize, value, 0);
1479 kmem_free(packed, nvsize);
1485 * Checks to see if the given vdev could not be opened, in which case we post a
1486 * sysevent to notify the autoreplace code that the device has been removed.
1489 spa_check_removed(vdev_t *vd)
1493 for (c = 0; c < vd->vdev_children; c++)
1494 spa_check_removed(vd->vdev_child[c]);
1496 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd)) {
1497 zfs_ereport_post(FM_EREPORT_RESOURCE_AUTOREPLACE,
1498 vd->vdev_spa, vd, NULL, 0, 0);
1499 spa_event_notify(vd->vdev_spa, vd, FM_EREPORT_ZFS_DEVICE_CHECK);
1504 * Validate the current config against the MOS config
1507 spa_config_valid(spa_t *spa, nvlist_t *config)
1509 vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1513 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1515 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1516 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1518 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1521 * If we're doing a normal import, then build up any additional
1522 * diagnostic information about missing devices in this config.
1523 * We'll pass this up to the user for further processing.
1525 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1526 nvlist_t **child, *nv;
1529 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1531 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_PUSHPAGE) == 0);
1533 for (c = 0; c < rvd->vdev_children; c++) {
1534 vdev_t *tvd = rvd->vdev_child[c];
1535 vdev_t *mtvd = mrvd->vdev_child[c];
1537 if (tvd->vdev_ops == &vdev_missing_ops &&
1538 mtvd->vdev_ops != &vdev_missing_ops &&
1540 child[idx++] = vdev_config_generate(spa, mtvd,
1545 VERIFY(nvlist_add_nvlist_array(nv,
1546 ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1547 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1548 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1550 for (i = 0; i < idx; i++)
1551 nvlist_free(child[i]);
1554 kmem_free(child, rvd->vdev_children * sizeof (char **));
1558 * Compare the root vdev tree with the information we have
1559 * from the MOS config (mrvd). Check each top-level vdev
1560 * with the corresponding MOS config top-level (mtvd).
1562 for (c = 0; c < rvd->vdev_children; c++) {
1563 vdev_t *tvd = rvd->vdev_child[c];
1564 vdev_t *mtvd = mrvd->vdev_child[c];
1567 * Resolve any "missing" vdevs in the current configuration.
1568 * If we find that the MOS config has more accurate information
1569 * about the top-level vdev then use that vdev instead.
1571 if (tvd->vdev_ops == &vdev_missing_ops &&
1572 mtvd->vdev_ops != &vdev_missing_ops) {
1574 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1578 * Device specific actions.
1580 if (mtvd->vdev_islog) {
1581 spa_set_log_state(spa, SPA_LOG_CLEAR);
1584 * XXX - once we have 'readonly' pool
1585 * support we should be able to handle
1586 * missing data devices by transitioning
1587 * the pool to readonly.
1593 * Swap the missing vdev with the data we were
1594 * able to obtain from the MOS config.
1596 vdev_remove_child(rvd, tvd);
1597 vdev_remove_child(mrvd, mtvd);
1599 vdev_add_child(rvd, mtvd);
1600 vdev_add_child(mrvd, tvd);
1602 spa_config_exit(spa, SCL_ALL, FTAG);
1604 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1607 } else if (mtvd->vdev_islog) {
1609 * Load the slog device's state from the MOS config
1610 * since it's possible that the label does not
1611 * contain the most up-to-date information.
1613 vdev_load_log_state(tvd, mtvd);
1618 spa_config_exit(spa, SCL_ALL, FTAG);
1621 * Ensure we were able to validate the config.
1623 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1627 * Check for missing log devices
1630 spa_check_logs(spa_t *spa)
1632 switch (spa->spa_log_state) {
1635 case SPA_LOG_MISSING:
1636 /* need to recheck in case slog has been restored */
1637 case SPA_LOG_UNKNOWN:
1638 if (dmu_objset_find(spa->spa_name, zil_check_log_chain, NULL,
1639 DS_FIND_CHILDREN)) {
1640 spa_set_log_state(spa, SPA_LOG_MISSING);
1649 spa_passivate_log(spa_t *spa)
1651 vdev_t *rvd = spa->spa_root_vdev;
1652 boolean_t slog_found = B_FALSE;
1655 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1657 if (!spa_has_slogs(spa))
1660 for (c = 0; c < rvd->vdev_children; c++) {
1661 vdev_t *tvd = rvd->vdev_child[c];
1662 metaslab_group_t *mg = tvd->vdev_mg;
1664 if (tvd->vdev_islog) {
1665 metaslab_group_passivate(mg);
1666 slog_found = B_TRUE;
1670 return (slog_found);
1674 spa_activate_log(spa_t *spa)
1676 vdev_t *rvd = spa->spa_root_vdev;
1679 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1681 for (c = 0; c < rvd->vdev_children; c++) {
1682 vdev_t *tvd = rvd->vdev_child[c];
1683 metaslab_group_t *mg = tvd->vdev_mg;
1685 if (tvd->vdev_islog)
1686 metaslab_group_activate(mg);
1691 spa_offline_log(spa_t *spa)
1695 if ((error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1696 NULL, DS_FIND_CHILDREN)) == 0) {
1699 * We successfully offlined the log device, sync out the
1700 * current txg so that the "stubby" block can be removed
1703 txg_wait_synced(spa->spa_dsl_pool, 0);
1709 spa_aux_check_removed(spa_aux_vdev_t *sav)
1713 for (i = 0; i < sav->sav_count; i++)
1714 spa_check_removed(sav->sav_vdevs[i]);
1718 spa_claim_notify(zio_t *zio)
1720 spa_t *spa = zio->io_spa;
1725 mutex_enter(&spa->spa_props_lock); /* any mutex will do */
1726 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1727 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1728 mutex_exit(&spa->spa_props_lock);
1731 typedef struct spa_load_error {
1732 uint64_t sle_meta_count;
1733 uint64_t sle_data_count;
1737 spa_load_verify_done(zio_t *zio)
1739 blkptr_t *bp = zio->io_bp;
1740 spa_load_error_t *sle = zio->io_private;
1741 dmu_object_type_t type = BP_GET_TYPE(bp);
1742 int error = zio->io_error;
1745 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1746 type != DMU_OT_INTENT_LOG)
1747 atomic_add_64(&sle->sle_meta_count, 1);
1749 atomic_add_64(&sle->sle_data_count, 1);
1751 zio_data_buf_free(zio->io_data, zio->io_size);
1756 spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1757 arc_buf_t *pbuf, const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1761 size_t size = BP_GET_PSIZE(bp);
1762 void *data = zio_data_buf_alloc(size);
1764 zio_nowait(zio_read(rio, spa, bp, data, size,
1765 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1766 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1767 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1773 spa_load_verify(spa_t *spa)
1776 spa_load_error_t sle = { 0 };
1777 zpool_rewind_policy_t policy;
1778 boolean_t verify_ok = B_FALSE;
1781 zpool_get_rewind_policy(spa->spa_config, &policy);
1783 if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1786 rio = zio_root(spa, NULL, &sle,
1787 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1789 error = traverse_pool(spa, spa->spa_verify_min_txg,
1790 TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio);
1792 (void) zio_wait(rio);
1794 spa->spa_load_meta_errors = sle.sle_meta_count;
1795 spa->spa_load_data_errors = sle.sle_data_count;
1797 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1798 sle.sle_data_count <= policy.zrp_maxdata) {
1802 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1803 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1805 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1806 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1807 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1808 VERIFY(nvlist_add_int64(spa->spa_load_info,
1809 ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1810 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1811 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1813 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1817 if (error != ENXIO && error != EIO)
1822 return (verify_ok ? 0 : EIO);
1826 * Find a value in the pool props object.
1829 spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1831 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1832 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1836 * Find a value in the pool directory object.
1839 spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1841 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1842 name, sizeof (uint64_t), 1, val));
1846 spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1848 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1853 * Fix up config after a partly-completed split. This is done with the
1854 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
1855 * pool have that entry in their config, but only the splitting one contains
1856 * a list of all the guids of the vdevs that are being split off.
1858 * This function determines what to do with that list: either rejoin
1859 * all the disks to the pool, or complete the splitting process. To attempt
1860 * the rejoin, each disk that is offlined is marked online again, and
1861 * we do a reopen() call. If the vdev label for every disk that was
1862 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1863 * then we call vdev_split() on each disk, and complete the split.
1865 * Otherwise we leave the config alone, with all the vdevs in place in
1866 * the original pool.
1869 spa_try_repair(spa_t *spa, nvlist_t *config)
1876 boolean_t attempt_reopen;
1878 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
1881 /* check that the config is complete */
1882 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
1883 &glist, &gcount) != 0)
1886 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_PUSHPAGE);
1888 /* attempt to online all the vdevs & validate */
1889 attempt_reopen = B_TRUE;
1890 for (i = 0; i < gcount; i++) {
1891 if (glist[i] == 0) /* vdev is hole */
1894 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
1895 if (vd[i] == NULL) {
1897 * Don't bother attempting to reopen the disks;
1898 * just do the split.
1900 attempt_reopen = B_FALSE;
1902 /* attempt to re-online it */
1903 vd[i]->vdev_offline = B_FALSE;
1907 if (attempt_reopen) {
1908 vdev_reopen(spa->spa_root_vdev);
1910 /* check each device to see what state it's in */
1911 for (extracted = 0, i = 0; i < gcount; i++) {
1912 if (vd[i] != NULL &&
1913 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
1920 * If every disk has been moved to the new pool, or if we never
1921 * even attempted to look at them, then we split them off for
1924 if (!attempt_reopen || gcount == extracted) {
1925 for (i = 0; i < gcount; i++)
1928 vdev_reopen(spa->spa_root_vdev);
1931 kmem_free(vd, gcount * sizeof (vdev_t *));
1935 spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
1936 boolean_t mosconfig)
1938 nvlist_t *config = spa->spa_config;
1939 char *ereport = FM_EREPORT_ZFS_POOL;
1945 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
1948 ASSERT(spa->spa_comment == NULL);
1949 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
1950 spa->spa_comment = spa_strdup(comment);
1953 * Versioning wasn't explicitly added to the label until later, so if
1954 * it's not present treat it as the initial version.
1956 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
1957 &spa->spa_ubsync.ub_version) != 0)
1958 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
1960 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
1961 &spa->spa_config_txg);
1963 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
1964 spa_guid_exists(pool_guid, 0)) {
1967 spa->spa_config_guid = pool_guid;
1969 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
1971 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
1975 nvlist_free(spa->spa_load_info);
1976 spa->spa_load_info = fnvlist_alloc();
1978 gethrestime(&spa->spa_loaded_ts);
1979 error = spa_load_impl(spa, pool_guid, config, state, type,
1980 mosconfig, &ereport);
1983 spa->spa_minref = refcount_count(&spa->spa_refcount);
1985 if (error != EEXIST) {
1986 spa->spa_loaded_ts.tv_sec = 0;
1987 spa->spa_loaded_ts.tv_nsec = 0;
1989 if (error != EBADF) {
1990 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
1993 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2000 * Load an existing storage pool, using the pool's builtin spa_config as a
2001 * source of configuration information.
2003 __attribute__((always_inline))
2005 spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
2006 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
2010 nvlist_t *nvroot = NULL;
2013 uberblock_t *ub = &spa->spa_uberblock;
2014 uint64_t children, config_cache_txg = spa->spa_config_txg;
2015 int orig_mode = spa->spa_mode;
2018 boolean_t missing_feat_write = B_FALSE;
2021 * If this is an untrusted config, access the pool in read-only mode.
2022 * This prevents things like resilvering recently removed devices.
2025 spa->spa_mode = FREAD;
2027 ASSERT(MUTEX_HELD(&spa_namespace_lock));
2029 spa->spa_load_state = state;
2031 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
2034 parse = (type == SPA_IMPORT_EXISTING ?
2035 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2038 * Create "The Godfather" zio to hold all async IOs
2040 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2041 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2044 * Parse the configuration into a vdev tree. We explicitly set the
2045 * value that will be returned by spa_version() since parsing the
2046 * configuration requires knowing the version number.
2048 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2049 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
2050 spa_config_exit(spa, SCL_ALL, FTAG);
2055 ASSERT(spa->spa_root_vdev == rvd);
2057 if (type != SPA_IMPORT_ASSEMBLE) {
2058 ASSERT(spa_guid(spa) == pool_guid);
2062 * Try to open all vdevs, loading each label in the process.
2064 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2065 error = vdev_open(rvd);
2066 spa_config_exit(spa, SCL_ALL, FTAG);
2071 * We need to validate the vdev labels against the configuration that
2072 * we have in hand, which is dependent on the setting of mosconfig. If
2073 * mosconfig is true then we're validating the vdev labels based on
2074 * that config. Otherwise, we're validating against the cached config
2075 * (zpool.cache) that was read when we loaded the zfs module, and then
2076 * later we will recursively call spa_load() and validate against
2079 * If we're assembling a new pool that's been split off from an
2080 * existing pool, the labels haven't yet been updated so we skip
2081 * validation for now.
2083 if (type != SPA_IMPORT_ASSEMBLE) {
2084 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2085 error = vdev_validate(rvd, mosconfig);
2086 spa_config_exit(spa, SCL_ALL, FTAG);
2091 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2096 * Find the best uberblock.
2098 vdev_uberblock_load(rvd, ub, &label);
2101 * If we weren't able to find a single valid uberblock, return failure.
2103 if (ub->ub_txg == 0) {
2105 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2109 * If the pool has an unsupported version we can't open it.
2111 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2113 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2116 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2120 * If we weren't able to find what's necessary for reading the
2121 * MOS in the label, return failure.
2123 if (label == NULL || nvlist_lookup_nvlist(label,
2124 ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) {
2126 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2131 * Update our in-core representation with the definitive values
2134 nvlist_free(spa->spa_label_features);
2135 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2141 * Look through entries in the label nvlist's features_for_read. If
2142 * there is a feature listed there which we don't understand then we
2143 * cannot open a pool.
2145 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2146 nvlist_t *unsup_feat;
2149 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2152 for (nvp = nvlist_next_nvpair(spa->spa_label_features, NULL);
2154 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2155 if (!zfeature_is_supported(nvpair_name(nvp))) {
2156 VERIFY(nvlist_add_string(unsup_feat,
2157 nvpair_name(nvp), "") == 0);
2161 if (!nvlist_empty(unsup_feat)) {
2162 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2163 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2164 nvlist_free(unsup_feat);
2165 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2169 nvlist_free(unsup_feat);
2173 * If the vdev guid sum doesn't match the uberblock, we have an
2174 * incomplete configuration. We first check to see if the pool
2175 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2176 * If it is, defer the vdev_guid_sum check till later so we
2177 * can handle missing vdevs.
2179 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
2180 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
2181 rvd->vdev_guid_sum != ub->ub_guid_sum)
2182 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2184 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2185 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2186 spa_try_repair(spa, config);
2187 spa_config_exit(spa, SCL_ALL, FTAG);
2188 nvlist_free(spa->spa_config_splitting);
2189 spa->spa_config_splitting = NULL;
2193 * Initialize internal SPA structures.
2195 spa->spa_state = POOL_STATE_ACTIVE;
2196 spa->spa_ubsync = spa->spa_uberblock;
2197 spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2198 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2199 spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2200 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2201 spa->spa_claim_max_txg = spa->spa_first_txg;
2202 spa->spa_prev_software_version = ub->ub_software_version;
2204 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2206 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2207 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2209 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2210 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2212 if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2213 boolean_t missing_feat_read = B_FALSE;
2214 nvlist_t *unsup_feat, *enabled_feat;
2216 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2217 &spa->spa_feat_for_read_obj) != 0) {
2218 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2221 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2222 &spa->spa_feat_for_write_obj) != 0) {
2223 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2226 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2227 &spa->spa_feat_desc_obj) != 0) {
2228 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2231 enabled_feat = fnvlist_alloc();
2232 unsup_feat = fnvlist_alloc();
2234 if (!feature_is_supported(spa->spa_meta_objset,
2235 spa->spa_feat_for_read_obj, spa->spa_feat_desc_obj,
2236 unsup_feat, enabled_feat))
2237 missing_feat_read = B_TRUE;
2239 if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) {
2240 if (!feature_is_supported(spa->spa_meta_objset,
2241 spa->spa_feat_for_write_obj, spa->spa_feat_desc_obj,
2242 unsup_feat, enabled_feat)) {
2243 missing_feat_write = B_TRUE;
2247 fnvlist_add_nvlist(spa->spa_load_info,
2248 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
2250 if (!nvlist_empty(unsup_feat)) {
2251 fnvlist_add_nvlist(spa->spa_load_info,
2252 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
2255 fnvlist_free(enabled_feat);
2256 fnvlist_free(unsup_feat);
2258 if (!missing_feat_read) {
2259 fnvlist_add_boolean(spa->spa_load_info,
2260 ZPOOL_CONFIG_CAN_RDONLY);
2264 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2265 * twofold: to determine whether the pool is available for
2266 * import in read-write mode and (if it is not) whether the
2267 * pool is available for import in read-only mode. If the pool
2268 * is available for import in read-write mode, it is displayed
2269 * as available in userland; if it is not available for import
2270 * in read-only mode, it is displayed as unavailable in
2271 * userland. If the pool is available for import in read-only
2272 * mode but not read-write mode, it is displayed as unavailable
2273 * in userland with a special note that the pool is actually
2274 * available for open in read-only mode.
2276 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2277 * missing a feature for write, we must first determine whether
2278 * the pool can be opened read-only before returning to
2279 * userland in order to know whether to display the
2280 * abovementioned note.
2282 if (missing_feat_read || (missing_feat_write &&
2283 spa_writeable(spa))) {
2284 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2289 spa->spa_is_initializing = B_TRUE;
2290 error = dsl_pool_open(spa->spa_dsl_pool);
2291 spa->spa_is_initializing = B_FALSE;
2293 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2297 nvlist_t *policy = NULL, *nvconfig;
2299 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2300 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2302 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2303 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2305 unsigned long myhostid = 0;
2307 VERIFY(nvlist_lookup_string(nvconfig,
2308 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2311 myhostid = zone_get_hostid(NULL);
2314 * We're emulating the system's hostid in userland, so
2315 * we can't use zone_get_hostid().
2317 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2318 #endif /* _KERNEL */
2319 if (hostid != 0 && myhostid != 0 &&
2320 hostid != myhostid) {
2321 nvlist_free(nvconfig);
2322 cmn_err(CE_WARN, "pool '%s' could not be "
2323 "loaded as it was last accessed by "
2324 "another system (host: %s hostid: 0x%lx). "
2325 "See: http://zfsonlinux.org/msg/ZFS-8000-EY",
2326 spa_name(spa), hostname,
2327 (unsigned long)hostid);
2331 if (nvlist_lookup_nvlist(spa->spa_config,
2332 ZPOOL_REWIND_POLICY, &policy) == 0)
2333 VERIFY(nvlist_add_nvlist(nvconfig,
2334 ZPOOL_REWIND_POLICY, policy) == 0);
2336 spa_config_set(spa, nvconfig);
2338 spa_deactivate(spa);
2339 spa_activate(spa, orig_mode);
2341 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2344 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2345 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2346 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2348 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2351 * Load the bit that tells us to use the new accounting function
2352 * (raid-z deflation). If we have an older pool, this will not
2355 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2356 if (error != 0 && error != ENOENT)
2357 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2359 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2360 &spa->spa_creation_version);
2361 if (error != 0 && error != ENOENT)
2362 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2365 * Load the persistent error log. If we have an older pool, this will
2368 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2369 if (error != 0 && error != ENOENT)
2370 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2372 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2373 &spa->spa_errlog_scrub);
2374 if (error != 0 && error != ENOENT)
2375 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2378 * Load the history object. If we have an older pool, this
2379 * will not be present.
2381 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2382 if (error != 0 && error != ENOENT)
2383 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2386 * If we're assembling the pool from the split-off vdevs of
2387 * an existing pool, we don't want to attach the spares & cache
2392 * Load any hot spares for this pool.
2394 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2395 if (error != 0 && error != ENOENT)
2396 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2397 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2398 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2399 if (load_nvlist(spa, spa->spa_spares.sav_object,
2400 &spa->spa_spares.sav_config) != 0)
2401 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2403 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2404 spa_load_spares(spa);
2405 spa_config_exit(spa, SCL_ALL, FTAG);
2406 } else if (error == 0) {
2407 spa->spa_spares.sav_sync = B_TRUE;
2411 * Load any level 2 ARC devices for this pool.
2413 error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2414 &spa->spa_l2cache.sav_object);
2415 if (error != 0 && error != ENOENT)
2416 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2417 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2418 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2419 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2420 &spa->spa_l2cache.sav_config) != 0)
2421 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2423 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2424 spa_load_l2cache(spa);
2425 spa_config_exit(spa, SCL_ALL, FTAG);
2426 } else if (error == 0) {
2427 spa->spa_l2cache.sav_sync = B_TRUE;
2430 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2432 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2433 if (error && error != ENOENT)
2434 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2437 uint64_t autoreplace;
2439 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2440 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2441 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2442 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2443 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2444 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2445 &spa->spa_dedup_ditto);
2447 spa->spa_autoreplace = (autoreplace != 0);
2451 * If the 'autoreplace' property is set, then post a resource notifying
2452 * the ZFS DE that it should not issue any faults for unopenable
2453 * devices. We also iterate over the vdevs, and post a sysevent for any
2454 * unopenable vdevs so that the normal autoreplace handler can take
2457 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2458 spa_check_removed(spa->spa_root_vdev);
2460 * For the import case, this is done in spa_import(), because
2461 * at this point we're using the spare definitions from
2462 * the MOS config, not necessarily from the userland config.
2464 if (state != SPA_LOAD_IMPORT) {
2465 spa_aux_check_removed(&spa->spa_spares);
2466 spa_aux_check_removed(&spa->spa_l2cache);
2471 * Load the vdev state for all toplevel vdevs.
2476 * Propagate the leaf DTLs we just loaded all the way up the tree.
2478 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2479 vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2480 spa_config_exit(spa, SCL_ALL, FTAG);
2483 * Load the DDTs (dedup tables).
2485 error = ddt_load(spa);
2487 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2489 spa_update_dspace(spa);
2492 * Validate the config, using the MOS config to fill in any
2493 * information which might be missing. If we fail to validate
2494 * the config then declare the pool unfit for use. If we're
2495 * assembling a pool from a split, the log is not transferred
2498 if (type != SPA_IMPORT_ASSEMBLE) {
2501 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2502 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2504 if (!spa_config_valid(spa, nvconfig)) {
2505 nvlist_free(nvconfig);
2506 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2509 nvlist_free(nvconfig);
2512 * Now that we've validated the config, check the state of the
2513 * root vdev. If it can't be opened, it indicates one or
2514 * more toplevel vdevs are faulted.
2516 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2519 if (spa_check_logs(spa)) {
2520 *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2521 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2525 if (missing_feat_write) {
2526 ASSERT(state == SPA_LOAD_TRYIMPORT);
2529 * At this point, we know that we can open the pool in
2530 * read-only mode but not read-write mode. We now have enough
2531 * information and can return to userland.
2533 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
2537 * We've successfully opened the pool, verify that we're ready
2538 * to start pushing transactions.
2540 if (state != SPA_LOAD_TRYIMPORT) {
2541 if ((error = spa_load_verify(spa)))
2542 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2546 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2547 spa->spa_load_max_txg == UINT64_MAX)) {
2549 int need_update = B_FALSE;
2552 ASSERT(state != SPA_LOAD_TRYIMPORT);
2555 * Claim log blocks that haven't been committed yet.
2556 * This must all happen in a single txg.
2557 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2558 * invoked from zil_claim_log_block()'s i/o done callback.
2559 * Price of rollback is that we abandon the log.
2561 spa->spa_claiming = B_TRUE;
2563 tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2564 spa_first_txg(spa));
2565 (void) dmu_objset_find(spa_name(spa),
2566 zil_claim, tx, DS_FIND_CHILDREN);
2569 spa->spa_claiming = B_FALSE;
2571 spa_set_log_state(spa, SPA_LOG_GOOD);
2572 spa->spa_sync_on = B_TRUE;
2573 txg_sync_start(spa->spa_dsl_pool);
2576 * Wait for all claims to sync. We sync up to the highest
2577 * claimed log block birth time so that claimed log blocks
2578 * don't appear to be from the future. spa_claim_max_txg
2579 * will have been set for us by either zil_check_log_chain()
2580 * (invoked from spa_check_logs()) or zil_claim() above.
2582 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2585 * If the config cache is stale, or we have uninitialized
2586 * metaslabs (see spa_vdev_add()), then update the config.
2588 * If this is a verbatim import, trust the current
2589 * in-core spa_config and update the disk labels.
2591 if (config_cache_txg != spa->spa_config_txg ||
2592 state == SPA_LOAD_IMPORT ||
2593 state == SPA_LOAD_RECOVER ||
2594 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2595 need_update = B_TRUE;
2597 for (c = 0; c < rvd->vdev_children; c++)
2598 if (rvd->vdev_child[c]->vdev_ms_array == 0)
2599 need_update = B_TRUE;
2602 * Update the config cache asychronously in case we're the
2603 * root pool, in which case the config cache isn't writable yet.
2606 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2609 * Check all DTLs to see if anything needs resilvering.
2611 if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2612 vdev_resilver_needed(rvd, NULL, NULL))
2613 spa_async_request(spa, SPA_ASYNC_RESILVER);
2616 * Delete any inconsistent datasets.
2618 (void) dmu_objset_find(spa_name(spa),
2619 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2622 * Clean up any stale temporary dataset userrefs.
2624 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2631 spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2633 int mode = spa->spa_mode;
2636 spa_deactivate(spa);
2638 spa->spa_load_max_txg--;
2640 spa_activate(spa, mode);
2641 spa_async_suspend(spa);
2643 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2647 * If spa_load() fails this function will try loading prior txg's. If
2648 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2649 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2650 * function will not rewind the pool and will return the same error as
2654 spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2655 uint64_t max_request, int rewind_flags)
2657 nvlist_t *loadinfo = NULL;
2658 nvlist_t *config = NULL;
2659 int load_error, rewind_error;
2660 uint64_t safe_rewind_txg;
2663 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2664 spa->spa_load_max_txg = spa->spa_load_txg;
2665 spa_set_log_state(spa, SPA_LOG_CLEAR);
2667 spa->spa_load_max_txg = max_request;
2670 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2672 if (load_error == 0)
2675 if (spa->spa_root_vdev != NULL)
2676 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2678 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2679 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2681 if (rewind_flags & ZPOOL_NEVER_REWIND) {
2682 nvlist_free(config);
2683 return (load_error);
2686 if (state == SPA_LOAD_RECOVER) {
2687 /* Price of rolling back is discarding txgs, including log */
2688 spa_set_log_state(spa, SPA_LOG_CLEAR);
2691 * If we aren't rolling back save the load info from our first
2692 * import attempt so that we can restore it after attempting
2695 loadinfo = spa->spa_load_info;
2696 spa->spa_load_info = fnvlist_alloc();
2699 spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2700 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2701 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2702 TXG_INITIAL : safe_rewind_txg;
2705 * Continue as long as we're finding errors, we're still within
2706 * the acceptable rewind range, and we're still finding uberblocks
2708 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2709 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2710 if (spa->spa_load_max_txg < safe_rewind_txg)
2711 spa->spa_extreme_rewind = B_TRUE;
2712 rewind_error = spa_load_retry(spa, state, mosconfig);
2715 spa->spa_extreme_rewind = B_FALSE;
2716 spa->spa_load_max_txg = UINT64_MAX;
2718 if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2719 spa_config_set(spa, config);
2721 if (state == SPA_LOAD_RECOVER) {
2722 ASSERT3P(loadinfo, ==, NULL);
2723 return (rewind_error);
2725 /* Store the rewind info as part of the initial load info */
2726 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
2727 spa->spa_load_info);
2729 /* Restore the initial load info */
2730 fnvlist_free(spa->spa_load_info);
2731 spa->spa_load_info = loadinfo;
2733 return (load_error);
2740 * The import case is identical to an open except that the configuration is sent
2741 * down from userland, instead of grabbed from the configuration cache. For the
2742 * case of an open, the pool configuration will exist in the
2743 * POOL_STATE_UNINITIALIZED state.
2745 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2746 * the same time open the pool, without having to keep around the spa_t in some
2750 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2754 spa_load_state_t state = SPA_LOAD_OPEN;
2756 int locked = B_FALSE;
2761 * As disgusting as this is, we need to support recursive calls to this
2762 * function because dsl_dir_open() is called during spa_load(), and ends
2763 * up calling spa_open() again. The real fix is to figure out how to
2764 * avoid dsl_dir_open() calling this in the first place.
2766 if (mutex_owner(&spa_namespace_lock) != curthread) {
2767 mutex_enter(&spa_namespace_lock);
2771 if ((spa = spa_lookup(pool)) == NULL) {
2773 mutex_exit(&spa_namespace_lock);
2777 if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2778 zpool_rewind_policy_t policy;
2780 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2782 if (policy.zrp_request & ZPOOL_DO_REWIND)
2783 state = SPA_LOAD_RECOVER;
2785 spa_activate(spa, spa_mode_global);
2787 if (state != SPA_LOAD_RECOVER)
2788 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2790 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2791 policy.zrp_request);
2793 if (error == EBADF) {
2795 * If vdev_validate() returns failure (indicated by
2796 * EBADF), it indicates that one of the vdevs indicates
2797 * that the pool has been exported or destroyed. If
2798 * this is the case, the config cache is out of sync and
2799 * we should remove the pool from the namespace.
2802 spa_deactivate(spa);
2803 spa_config_sync(spa, B_TRUE, B_TRUE);
2806 mutex_exit(&spa_namespace_lock);
2812 * We can't open the pool, but we still have useful
2813 * information: the state of each vdev after the
2814 * attempted vdev_open(). Return this to the user.
2816 if (config != NULL && spa->spa_config) {
2817 VERIFY(nvlist_dup(spa->spa_config, config,
2819 VERIFY(nvlist_add_nvlist(*config,
2820 ZPOOL_CONFIG_LOAD_INFO,
2821 spa->spa_load_info) == 0);
2824 spa_deactivate(spa);
2825 spa->spa_last_open_failed = error;
2827 mutex_exit(&spa_namespace_lock);
2833 spa_open_ref(spa, tag);
2836 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2839 * If we've recovered the pool, pass back any information we
2840 * gathered while doing the load.
2842 if (state == SPA_LOAD_RECOVER) {
2843 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
2844 spa->spa_load_info) == 0);
2848 spa->spa_last_open_failed = 0;
2849 spa->spa_last_ubsync_txg = 0;
2850 spa->spa_load_txg = 0;
2851 mutex_exit(&spa_namespace_lock);
2860 spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
2863 return (spa_open_common(name, spapp, tag, policy, config));
2867 spa_open(const char *name, spa_t **spapp, void *tag)
2869 return (spa_open_common(name, spapp, tag, NULL, NULL));
2873 * Lookup the given spa_t, incrementing the inject count in the process,
2874 * preventing it from being exported or destroyed.
2877 spa_inject_addref(char *name)
2881 mutex_enter(&spa_namespace_lock);
2882 if ((spa = spa_lookup(name)) == NULL) {
2883 mutex_exit(&spa_namespace_lock);
2886 spa->spa_inject_ref++;
2887 mutex_exit(&spa_namespace_lock);
2893 spa_inject_delref(spa_t *spa)
2895 mutex_enter(&spa_namespace_lock);
2896 spa->spa_inject_ref--;
2897 mutex_exit(&spa_namespace_lock);
2901 * Add spares device information to the nvlist.
2904 spa_add_spares(spa_t *spa, nvlist_t *config)
2914 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2916 if (spa->spa_spares.sav_count == 0)
2919 VERIFY(nvlist_lookup_nvlist(config,
2920 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2921 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
2922 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2924 VERIFY(nvlist_add_nvlist_array(nvroot,
2925 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2926 VERIFY(nvlist_lookup_nvlist_array(nvroot,
2927 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2930 * Go through and find any spares which have since been
2931 * repurposed as an active spare. If this is the case, update
2932 * their status appropriately.
2934 for (i = 0; i < nspares; i++) {
2935 VERIFY(nvlist_lookup_uint64(spares[i],
2936 ZPOOL_CONFIG_GUID, &guid) == 0);
2937 if (spa_spare_exists(guid, &pool, NULL) &&
2939 VERIFY(nvlist_lookup_uint64_array(
2940 spares[i], ZPOOL_CONFIG_VDEV_STATS,
2941 (uint64_t **)&vs, &vsc) == 0);
2942 vs->vs_state = VDEV_STATE_CANT_OPEN;
2943 vs->vs_aux = VDEV_AUX_SPARED;
2950 * Add l2cache device information to the nvlist, including vdev stats.
2953 spa_add_l2cache(spa_t *spa, nvlist_t *config)
2956 uint_t i, j, nl2cache;
2963 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2965 if (spa->spa_l2cache.sav_count == 0)
2968 VERIFY(nvlist_lookup_nvlist(config,
2969 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2970 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
2971 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2972 if (nl2cache != 0) {
2973 VERIFY(nvlist_add_nvlist_array(nvroot,
2974 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2975 VERIFY(nvlist_lookup_nvlist_array(nvroot,
2976 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2979 * Update level 2 cache device stats.
2982 for (i = 0; i < nl2cache; i++) {
2983 VERIFY(nvlist_lookup_uint64(l2cache[i],
2984 ZPOOL_CONFIG_GUID, &guid) == 0);
2987 for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
2989 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
2990 vd = spa->spa_l2cache.sav_vdevs[j];
2996 VERIFY(nvlist_lookup_uint64_array(l2cache[i],
2997 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
2999 vdev_get_stats(vd, vs);
3005 spa_add_feature_stats(spa_t *spa, nvlist_t *config)
3011 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3012 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3014 if (spa->spa_feat_for_read_obj != 0) {
3015 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3016 spa->spa_feat_for_read_obj);
3017 zap_cursor_retrieve(&zc, &za) == 0;
3018 zap_cursor_advance(&zc)) {
3019 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3020 za.za_num_integers == 1);
3021 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3022 za.za_first_integer));
3024 zap_cursor_fini(&zc);
3027 if (spa->spa_feat_for_write_obj != 0) {
3028 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3029 spa->spa_feat_for_write_obj);
3030 zap_cursor_retrieve(&zc, &za) == 0;
3031 zap_cursor_advance(&zc)) {
3032 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3033 za.za_num_integers == 1);
3034 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3035 za.za_first_integer));
3037 zap_cursor_fini(&zc);
3040 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
3042 nvlist_free(features);
3046 spa_get_stats(const char *name, nvlist_t **config,
3047 char *altroot, size_t buflen)
3053 error = spa_open_common(name, &spa, FTAG, NULL, config);
3057 * This still leaves a window of inconsistency where the spares
3058 * or l2cache devices could change and the config would be
3059 * self-inconsistent.
3061 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3063 if (*config != NULL) {
3064 uint64_t loadtimes[2];
3066 loadtimes[0] = spa->spa_loaded_ts.tv_sec;
3067 loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
3068 VERIFY(nvlist_add_uint64_array(*config,
3069 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
3071 VERIFY(nvlist_add_uint64(*config,
3072 ZPOOL_CONFIG_ERRCOUNT,
3073 spa_get_errlog_size(spa)) == 0);
3075 if (spa_suspended(spa))
3076 VERIFY(nvlist_add_uint64(*config,
3077 ZPOOL_CONFIG_SUSPENDED,
3078 spa->spa_failmode) == 0);
3080 spa_add_spares(spa, *config);
3081 spa_add_l2cache(spa, *config);
3082 spa_add_feature_stats(spa, *config);
3087 * We want to get the alternate root even for faulted pools, so we cheat
3088 * and call spa_lookup() directly.
3092 mutex_enter(&spa_namespace_lock);
3093 spa = spa_lookup(name);
3095 spa_altroot(spa, altroot, buflen);
3099 mutex_exit(&spa_namespace_lock);
3101 spa_altroot(spa, altroot, buflen);
3106 spa_config_exit(spa, SCL_CONFIG, FTAG);
3107 spa_close(spa, FTAG);
3114 * Validate that the auxiliary device array is well formed. We must have an
3115 * array of nvlists, each which describes a valid leaf vdev. If this is an
3116 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3117 * specified, as long as they are well-formed.
3120 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
3121 spa_aux_vdev_t *sav, const char *config, uint64_t version,
3122 vdev_labeltype_t label)
3129 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3132 * It's acceptable to have no devs specified.
3134 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
3141 * Make sure the pool is formatted with a version that supports this
3144 if (spa_version(spa) < version)
3148 * Set the pending device list so we correctly handle device in-use
3151 sav->sav_pending = dev;
3152 sav->sav_npending = ndev;
3154 for (i = 0; i < ndev; i++) {
3155 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
3159 if (!vd->vdev_ops->vdev_op_leaf) {
3166 * The L2ARC currently only supports disk devices in
3167 * kernel context. For user-level testing, we allow it.
3170 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
3171 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
3179 if ((error = vdev_open(vd)) == 0 &&
3180 (error = vdev_label_init(vd, crtxg, label)) == 0) {
3181 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
3182 vd->vdev_guid) == 0);
3188 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
3195 sav->sav_pending = NULL;
3196 sav->sav_npending = 0;
3201 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
3205 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3207 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3208 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
3209 VDEV_LABEL_SPARE)) != 0) {
3213 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3214 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
3215 VDEV_LABEL_L2CACHE));
3219 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
3224 if (sav->sav_config != NULL) {
3230 * Generate new dev list by concatentating with the
3233 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
3234 &olddevs, &oldndevs) == 0);
3236 newdevs = kmem_alloc(sizeof (void *) *
3237 (ndevs + oldndevs), KM_PUSHPAGE);
3238 for (i = 0; i < oldndevs; i++)
3239 VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
3241 for (i = 0; i < ndevs; i++)
3242 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
3245 VERIFY(nvlist_remove(sav->sav_config, config,
3246 DATA_TYPE_NVLIST_ARRAY) == 0);
3248 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3249 config, newdevs, ndevs + oldndevs) == 0);
3250 for (i = 0; i < oldndevs + ndevs; i++)
3251 nvlist_free(newdevs[i]);
3252 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
3255 * Generate a new dev list.
3257 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
3259 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
3265 * Stop and drop level 2 ARC devices
3268 spa_l2cache_drop(spa_t *spa)
3272 spa_aux_vdev_t *sav = &spa->spa_l2cache;
3274 for (i = 0; i < sav->sav_count; i++) {
3277 vd = sav->sav_vdevs[i];
3280 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
3281 pool != 0ULL && l2arc_vdev_present(vd))
3282 l2arc_remove_vdev(vd);
3290 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
3291 const char *history_str, nvlist_t *zplprops)
3294 char *altroot = NULL;
3299 uint64_t txg = TXG_INITIAL;
3300 nvlist_t **spares, **l2cache;
3301 uint_t nspares, nl2cache;
3302 uint64_t version, obj;
3303 boolean_t has_features;
3308 * If this pool already exists, return failure.
3310 mutex_enter(&spa_namespace_lock);
3311 if (spa_lookup(pool) != NULL) {
3312 mutex_exit(&spa_namespace_lock);
3317 * Allocate a new spa_t structure.
3319 (void) nvlist_lookup_string(props,
3320 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3321 spa = spa_add(pool, NULL, altroot);
3322 spa_activate(spa, spa_mode_global);
3324 if (props && (error = spa_prop_validate(spa, props))) {
3325 spa_deactivate(spa);
3327 mutex_exit(&spa_namespace_lock);
3331 has_features = B_FALSE;
3332 for (elem = nvlist_next_nvpair(props, NULL);
3333 elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
3334 if (zpool_prop_feature(nvpair_name(elem)))
3335 has_features = B_TRUE;
3338 if (has_features || nvlist_lookup_uint64(props,
3339 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
3340 version = SPA_VERSION;
3342 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
3344 spa->spa_first_txg = txg;
3345 spa->spa_uberblock.ub_txg = txg - 1;
3346 spa->spa_uberblock.ub_version = version;
3347 spa->spa_ubsync = spa->spa_uberblock;
3350 * Create "The Godfather" zio to hold all async IOs
3352 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
3353 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
3356 * Create the root vdev.
3358 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3360 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3362 ASSERT(error != 0 || rvd != NULL);
3363 ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3365 if (error == 0 && !zfs_allocatable_devs(nvroot))
3369 (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3370 (error = spa_validate_aux(spa, nvroot, txg,
3371 VDEV_ALLOC_ADD)) == 0) {
3372 for (c = 0; c < rvd->vdev_children; c++) {
3373 vdev_metaslab_set_size(rvd->vdev_child[c]);
3374 vdev_expand(rvd->vdev_child[c], txg);
3378 spa_config_exit(spa, SCL_ALL, FTAG);
3382 spa_deactivate(spa);
3384 mutex_exit(&spa_namespace_lock);
3389 * Get the list of spares, if specified.
3391 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3392 &spares, &nspares) == 0) {
3393 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3395 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3396 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3397 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3398 spa_load_spares(spa);
3399 spa_config_exit(spa, SCL_ALL, FTAG);
3400 spa->spa_spares.sav_sync = B_TRUE;
3404 * Get the list of level 2 cache devices, if specified.
3406 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3407 &l2cache, &nl2cache) == 0) {
3408 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3409 NV_UNIQUE_NAME, KM_PUSHPAGE) == 0);
3410 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3411 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3412 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3413 spa_load_l2cache(spa);
3414 spa_config_exit(spa, SCL_ALL, FTAG);
3415 spa->spa_l2cache.sav_sync = B_TRUE;
3418 spa->spa_is_initializing = B_TRUE;
3419 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3420 spa->spa_meta_objset = dp->dp_meta_objset;
3421 spa->spa_is_initializing = B_FALSE;
3424 * Create DDTs (dedup tables).
3428 spa_update_dspace(spa);
3430 tx = dmu_tx_create_assigned(dp, txg);
3433 * Create the pool config object.
3435 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3436 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3437 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3439 if (zap_add(spa->spa_meta_objset,
3440 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3441 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3442 cmn_err(CE_PANIC, "failed to add pool config");
3445 if (spa_version(spa) >= SPA_VERSION_FEATURES)
3446 spa_feature_create_zap_objects(spa, tx);
3448 if (zap_add(spa->spa_meta_objset,
3449 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3450 sizeof (uint64_t), 1, &version, tx) != 0) {
3451 cmn_err(CE_PANIC, "failed to add pool version");
3454 /* Newly created pools with the right version are always deflated. */
3455 if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3456 spa->spa_deflate = TRUE;
3457 if (zap_add(spa->spa_meta_objset,
3458 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3459 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3460 cmn_err(CE_PANIC, "failed to add deflate");
3465 * Create the deferred-free bpobj. Turn off compression
3466 * because sync-to-convergence takes longer if the blocksize
3469 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3470 dmu_object_set_compress(spa->spa_meta_objset, obj,
3471 ZIO_COMPRESS_OFF, tx);
3472 if (zap_add(spa->spa_meta_objset,
3473 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3474 sizeof (uint64_t), 1, &obj, tx) != 0) {
3475 cmn_err(CE_PANIC, "failed to add bpobj");
3477 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3478 spa->spa_meta_objset, obj));
3481 * Create the pool's history object.
3483 if (version >= SPA_VERSION_ZPOOL_HISTORY)
3484 spa_history_create_obj(spa, tx);
3487 * Set pool properties.
3489 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3490 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3491 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3492 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3494 if (props != NULL) {
3495 spa_configfile_set(spa, props, B_FALSE);
3496 spa_sync_props(spa, props, tx);
3501 spa->spa_sync_on = B_TRUE;
3502 txg_sync_start(spa->spa_dsl_pool);
3505 * We explicitly wait for the first transaction to complete so that our
3506 * bean counters are appropriately updated.
3508 txg_wait_synced(spa->spa_dsl_pool, txg);
3510 spa_config_sync(spa, B_FALSE, B_TRUE);
3512 if (version >= SPA_VERSION_ZPOOL_HISTORY && history_str != NULL)
3513 (void) spa_history_log(spa, history_str, LOG_CMD_POOL_CREATE);
3514 spa_history_log_version(spa, LOG_POOL_CREATE);
3516 spa->spa_minref = refcount_count(&spa->spa_refcount);
3518 mutex_exit(&spa_namespace_lock);
3525 * Get the root pool information from the root disk, then import the root pool
3526 * during the system boot up time.
3528 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3531 spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3534 nvlist_t *nvtop, *nvroot;
3537 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3541 * Add this top-level vdev to the child array.
3543 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3545 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3547 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3550 * Put this pool's top-level vdevs into a root vdev.
3552 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_PUSHPAGE) == 0);
3553 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3554 VDEV_TYPE_ROOT) == 0);
3555 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3556 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3557 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3561 * Replace the existing vdev_tree with the new root vdev in
3562 * this pool's configuration (remove the old, add the new).
3564 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3565 nvlist_free(nvroot);
3570 * Walk the vdev tree and see if we can find a device with "better"
3571 * configuration. A configuration is "better" if the label on that
3572 * device has a more recent txg.
3575 spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3579 for (c = 0; c < vd->vdev_children; c++)
3580 spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3582 if (vd->vdev_ops->vdev_op_leaf) {
3586 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3590 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3594 * Do we have a better boot device?
3596 if (label_txg > *txg) {
3605 * Import a root pool.
3607 * For x86. devpath_list will consist of devid and/or physpath name of
3608 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3609 * The GRUB "findroot" command will return the vdev we should boot.
3611 * For Sparc, devpath_list consists the physpath name of the booting device
3612 * no matter the rootpool is a single device pool or a mirrored pool.
3614 * "/pci@1f,0/ide@d/disk@0,0:a"
3617 spa_import_rootpool(char *devpath, char *devid)
3620 vdev_t *rvd, *bvd, *avd = NULL;
3621 nvlist_t *config, *nvtop;
3627 * Read the label from the boot device and generate a configuration.
3629 config = spa_generate_rootconf(devpath, devid, &guid);
3630 #if defined(_OBP) && defined(_KERNEL)
3631 if (config == NULL) {
3632 if (strstr(devpath, "/iscsi/ssd") != NULL) {
3634 get_iscsi_bootpath_phy(devpath);
3635 config = spa_generate_rootconf(devpath, devid, &guid);
3639 if (config == NULL) {
3640 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
3645 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3647 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3649 mutex_enter(&spa_namespace_lock);
3650 if ((spa = spa_lookup(pname)) != NULL) {
3652 * Remove the existing root pool from the namespace so that we
3653 * can replace it with the correct config we just read in.
3658 spa = spa_add(pname, config, NULL);
3659 spa->spa_is_root = B_TRUE;
3660 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3663 * Build up a vdev tree based on the boot device's label config.
3665 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3667 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3668 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3669 VDEV_ALLOC_ROOTPOOL);
3670 spa_config_exit(spa, SCL_ALL, FTAG);
3672 mutex_exit(&spa_namespace_lock);
3673 nvlist_free(config);
3674 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3680 * Get the boot vdev.
3682 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3683 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3684 (u_longlong_t)guid);
3690 * Determine if there is a better boot device.
3693 spa_alt_rootvdev(rvd, &avd, &txg);
3695 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3696 "try booting from '%s'", avd->vdev_path);
3702 * If the boot device is part of a spare vdev then ensure that
3703 * we're booting off the active spare.
3705 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3706 !bvd->vdev_isspare) {
3707 cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3708 "try booting from '%s'",
3710 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3716 spa_history_log_version(spa, LOG_POOL_IMPORT);
3718 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3720 spa_config_exit(spa, SCL_ALL, FTAG);
3721 mutex_exit(&spa_namespace_lock);
3723 nvlist_free(config);
3730 * Import a non-root pool into the system.
3733 spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
3736 char *altroot = NULL;
3737 spa_load_state_t state = SPA_LOAD_IMPORT;
3738 zpool_rewind_policy_t policy;
3739 uint64_t mode = spa_mode_global;
3740 uint64_t readonly = B_FALSE;
3743 nvlist_t **spares, **l2cache;
3744 uint_t nspares, nl2cache;
3747 * If a pool with this name exists, return failure.
3749 mutex_enter(&spa_namespace_lock);
3750 if (spa_lookup(pool) != NULL) {
3751 mutex_exit(&spa_namespace_lock);
3756 * Create and initialize the spa structure.
3758 (void) nvlist_lookup_string(props,
3759 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3760 (void) nvlist_lookup_uint64(props,
3761 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
3764 spa = spa_add(pool, config, altroot);
3765 spa->spa_import_flags = flags;
3768 * Verbatim import - Take a pool and insert it into the namespace
3769 * as if it had been loaded at boot.
3771 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
3773 spa_configfile_set(spa, props, B_FALSE);
3775 spa_config_sync(spa, B_FALSE, B_TRUE);
3777 mutex_exit(&spa_namespace_lock);
3778 spa_history_log_version(spa, LOG_POOL_IMPORT);
3783 spa_activate(spa, mode);
3786 * Don't start async tasks until we know everything is healthy.
3788 spa_async_suspend(spa);
3790 zpool_get_rewind_policy(config, &policy);
3791 if (policy.zrp_request & ZPOOL_DO_REWIND)
3792 state = SPA_LOAD_RECOVER;
3795 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
3796 * because the user-supplied config is actually the one to trust when
3799 if (state != SPA_LOAD_RECOVER)
3800 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3802 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
3803 policy.zrp_request);
3806 * Propagate anything learned while loading the pool and pass it
3807 * back to caller (i.e. rewind info, missing devices, etc).
3809 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
3810 spa->spa_load_info) == 0);
3812 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3814 * Toss any existing sparelist, as it doesn't have any validity
3815 * anymore, and conflicts with spa_has_spare().
3817 if (spa->spa_spares.sav_config) {
3818 nvlist_free(spa->spa_spares.sav_config);
3819 spa->spa_spares.sav_config = NULL;
3820 spa_load_spares(spa);
3822 if (spa->spa_l2cache.sav_config) {
3823 nvlist_free(spa->spa_l2cache.sav_config);
3824 spa->spa_l2cache.sav_config = NULL;
3825 spa_load_l2cache(spa);
3828 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3831 error = spa_validate_aux(spa, nvroot, -1ULL,
3834 error = spa_validate_aux(spa, nvroot, -1ULL,
3835 VDEV_ALLOC_L2CACHE);
3836 spa_config_exit(spa, SCL_ALL, FTAG);
3839 spa_configfile_set(spa, props, B_FALSE);
3841 if (error != 0 || (props && spa_writeable(spa) &&
3842 (error = spa_prop_set(spa, props)))) {
3844 spa_deactivate(spa);
3846 mutex_exit(&spa_namespace_lock);
3850 spa_async_resume(spa);
3853 * Override any spares and level 2 cache devices as specified by
3854 * the user, as these may have correct device names/devids, etc.
3856 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3857 &spares, &nspares) == 0) {
3858 if (spa->spa_spares.sav_config)
3859 VERIFY(nvlist_remove(spa->spa_spares.sav_config,
3860 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
3862 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
3863 NV_UNIQUE_NAME, KM_PUSHPAGE) == 0);
3864 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3865 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3866 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3867 spa_load_spares(spa);
3868 spa_config_exit(spa, SCL_ALL, FTAG);
3869 spa->spa_spares.sav_sync = B_TRUE;
3871 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3872 &l2cache, &nl2cache) == 0) {
3873 if (spa->spa_l2cache.sav_config)
3874 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
3875 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
3877 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3878 NV_UNIQUE_NAME, KM_PUSHPAGE) == 0);
3879 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3880 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3881 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3882 spa_load_l2cache(spa);
3883 spa_config_exit(spa, SCL_ALL, FTAG);
3884 spa->spa_l2cache.sav_sync = B_TRUE;
3888 * Check for any removed devices.
3890 if (spa->spa_autoreplace) {
3891 spa_aux_check_removed(&spa->spa_spares);
3892 spa_aux_check_removed(&spa->spa_l2cache);
3895 if (spa_writeable(spa)) {
3897 * Update the config cache to include the newly-imported pool.
3899 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3903 * It's possible that the pool was expanded while it was exported.
3904 * We kick off an async task to handle this for us.
3906 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
3908 mutex_exit(&spa_namespace_lock);
3909 spa_history_log_version(spa, LOG_POOL_IMPORT);
3915 spa_tryimport(nvlist_t *tryconfig)
3917 nvlist_t *config = NULL;
3923 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
3926 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
3930 * Create and initialize the spa structure.
3932 mutex_enter(&spa_namespace_lock);
3933 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
3934 spa_activate(spa, FREAD);
3937 * Pass off the heavy lifting to spa_load().
3938 * Pass TRUE for mosconfig because the user-supplied config
3939 * is actually the one to trust when doing an import.
3941 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
3944 * If 'tryconfig' was at least parsable, return the current config.
3946 if (spa->spa_root_vdev != NULL) {
3947 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
3948 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
3950 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
3952 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
3953 spa->spa_uberblock.ub_timestamp) == 0);
3954 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
3955 spa->spa_load_info) == 0);
3958 * If the bootfs property exists on this pool then we
3959 * copy it out so that external consumers can tell which
3960 * pools are bootable.
3962 if ((!error || error == EEXIST) && spa->spa_bootfs) {
3963 char *tmpname = kmem_alloc(MAXPATHLEN, KM_PUSHPAGE);
3966 * We have to play games with the name since the
3967 * pool was opened as TRYIMPORT_NAME.
3969 if (dsl_dsobj_to_dsname(spa_name(spa),
3970 spa->spa_bootfs, tmpname) == 0) {
3972 char *dsname = kmem_alloc(MAXPATHLEN, KM_PUSHPAGE);
3974 cp = strchr(tmpname, '/');
3976 (void) strlcpy(dsname, tmpname,
3979 (void) snprintf(dsname, MAXPATHLEN,
3980 "%s/%s", poolname, ++cp);
3982 VERIFY(nvlist_add_string(config,
3983 ZPOOL_CONFIG_BOOTFS, dsname) == 0);
3984 kmem_free(dsname, MAXPATHLEN);
3986 kmem_free(tmpname, MAXPATHLEN);
3990 * Add the list of hot spares and level 2 cache devices.
3992 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3993 spa_add_spares(spa, config);
3994 spa_add_l2cache(spa, config);
3995 spa_config_exit(spa, SCL_CONFIG, FTAG);
3999 spa_deactivate(spa);
4001 mutex_exit(&spa_namespace_lock);
4007 * Pool export/destroy
4009 * The act of destroying or exporting a pool is very simple. We make sure there
4010 * is no more pending I/O and any references to the pool are gone. Then, we
4011 * update the pool state and sync all the labels to disk, removing the
4012 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4013 * we don't sync the labels or remove the configuration cache.
4016 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
4017 boolean_t force, boolean_t hardforce)
4024 if (!(spa_mode_global & FWRITE))
4027 mutex_enter(&spa_namespace_lock);
4028 if ((spa = spa_lookup(pool)) == NULL) {
4029 mutex_exit(&spa_namespace_lock);
4034 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4035 * reacquire the namespace lock, and see if we can export.
4037 spa_open_ref(spa, FTAG);
4038 mutex_exit(&spa_namespace_lock);
4039 spa_async_suspend(spa);
4040 mutex_enter(&spa_namespace_lock);
4041 spa_close(spa, FTAG);
4044 * The pool will be in core if it's openable,
4045 * in which case we can modify its state.
4047 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
4049 * Objsets may be open only because they're dirty, so we
4050 * have to force it to sync before checking spa_refcnt.
4052 txg_wait_synced(spa->spa_dsl_pool, 0);
4055 * A pool cannot be exported or destroyed if there are active
4056 * references. If we are resetting a pool, allow references by
4057 * fault injection handlers.
4059 if (!spa_refcount_zero(spa) ||
4060 (spa->spa_inject_ref != 0 &&
4061 new_state != POOL_STATE_UNINITIALIZED)) {
4062 spa_async_resume(spa);
4063 mutex_exit(&spa_namespace_lock);
4068 * A pool cannot be exported if it has an active shared spare.
4069 * This is to prevent other pools stealing the active spare
4070 * from an exported pool. At user's own will, such pool can
4071 * be forcedly exported.
4073 if (!force && new_state == POOL_STATE_EXPORTED &&
4074 spa_has_active_shared_spare(spa)) {
4075 spa_async_resume(spa);
4076 mutex_exit(&spa_namespace_lock);
4081 * We want this to be reflected on every label,
4082 * so mark them all dirty. spa_unload() will do the
4083 * final sync that pushes these changes out.
4085 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
4086 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4087 spa->spa_state = new_state;
4088 spa->spa_final_txg = spa_last_synced_txg(spa) +
4090 vdev_config_dirty(spa->spa_root_vdev);
4091 spa_config_exit(spa, SCL_ALL, FTAG);
4095 spa_event_notify(spa, NULL, FM_EREPORT_ZFS_POOL_DESTROY);
4097 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4099 spa_deactivate(spa);
4102 if (oldconfig && spa->spa_config)
4103 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
4105 if (new_state != POOL_STATE_UNINITIALIZED) {
4107 spa_config_sync(spa, B_TRUE, B_TRUE);
4110 mutex_exit(&spa_namespace_lock);
4116 * Destroy a storage pool.
4119 spa_destroy(char *pool)
4121 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
4126 * Export a storage pool.
4129 spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
4130 boolean_t hardforce)
4132 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
4137 * Similar to spa_export(), this unloads the spa_t without actually removing it
4138 * from the namespace in any way.
4141 spa_reset(char *pool)
4143 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
4148 * ==========================================================================
4149 * Device manipulation
4150 * ==========================================================================
4154 * Add a device to a storage pool.
4157 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
4161 vdev_t *rvd = spa->spa_root_vdev;
4163 nvlist_t **spares, **l2cache;
4164 uint_t nspares, nl2cache;
4167 ASSERT(spa_writeable(spa));
4169 txg = spa_vdev_enter(spa);
4171 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
4172 VDEV_ALLOC_ADD)) != 0)
4173 return (spa_vdev_exit(spa, NULL, txg, error));
4175 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */
4177 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
4181 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
4185 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
4186 return (spa_vdev_exit(spa, vd, txg, EINVAL));
4188 if (vd->vdev_children != 0 &&
4189 (error = vdev_create(vd, txg, B_FALSE)) != 0)
4190 return (spa_vdev_exit(spa, vd, txg, error));
4193 * We must validate the spares and l2cache devices after checking the
4194 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
4196 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
4197 return (spa_vdev_exit(spa, vd, txg, error));
4200 * Transfer each new top-level vdev from vd to rvd.
4202 for (c = 0; c < vd->vdev_children; c++) {
4205 * Set the vdev id to the first hole, if one exists.
4207 for (id = 0; id < rvd->vdev_children; id++) {
4208 if (rvd->vdev_child[id]->vdev_ishole) {
4209 vdev_free(rvd->vdev_child[id]);
4213 tvd = vd->vdev_child[c];
4214 vdev_remove_child(vd, tvd);
4216 vdev_add_child(rvd, tvd);
4217 vdev_config_dirty(tvd);
4221 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
4222 ZPOOL_CONFIG_SPARES);
4223 spa_load_spares(spa);
4224 spa->spa_spares.sav_sync = B_TRUE;
4227 if (nl2cache != 0) {
4228 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
4229 ZPOOL_CONFIG_L2CACHE);
4230 spa_load_l2cache(spa);
4231 spa->spa_l2cache.sav_sync = B_TRUE;
4235 * We have to be careful when adding new vdevs to an existing pool.
4236 * If other threads start allocating from these vdevs before we
4237 * sync the config cache, and we lose power, then upon reboot we may
4238 * fail to open the pool because there are DVAs that the config cache
4239 * can't translate. Therefore, we first add the vdevs without
4240 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4241 * and then let spa_config_update() initialize the new metaslabs.
4243 * spa_load() checks for added-but-not-initialized vdevs, so that
4244 * if we lose power at any point in this sequence, the remaining
4245 * steps will be completed the next time we load the pool.
4247 (void) spa_vdev_exit(spa, vd, txg, 0);
4249 mutex_enter(&spa_namespace_lock);
4250 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4251 mutex_exit(&spa_namespace_lock);
4257 * Attach a device to a mirror. The arguments are the path to any device
4258 * in the mirror, and the nvroot for the new device. If the path specifies
4259 * a device that is not mirrored, we automatically insert the mirror vdev.
4261 * If 'replacing' is specified, the new device is intended to replace the
4262 * existing device; in this case the two devices are made into their own
4263 * mirror using the 'replacing' vdev, which is functionally identical to
4264 * the mirror vdev (it actually reuses all the same ops) but has a few
4265 * extra rules: you can't attach to it after it's been created, and upon
4266 * completion of resilvering, the first disk (the one being replaced)
4267 * is automatically detached.
4270 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
4272 uint64_t txg, dtl_max_txg;
4273 ASSERTV(vdev_t *rvd = spa->spa_root_vdev;)
4274 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
4276 char *oldvdpath, *newvdpath;
4280 ASSERT(spa_writeable(spa));
4282 txg = spa_vdev_enter(spa);
4284 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
4287 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4289 if (!oldvd->vdev_ops->vdev_op_leaf)
4290 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4292 pvd = oldvd->vdev_parent;
4294 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
4295 VDEV_ALLOC_ATTACH)) != 0)
4296 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4298 if (newrootvd->vdev_children != 1)
4299 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4301 newvd = newrootvd->vdev_child[0];
4303 if (!newvd->vdev_ops->vdev_op_leaf)
4304 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4306 if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
4307 return (spa_vdev_exit(spa, newrootvd, txg, error));
4310 * Spares can't replace logs
4312 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
4313 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4317 * For attach, the only allowable parent is a mirror or the root
4320 if (pvd->vdev_ops != &vdev_mirror_ops &&
4321 pvd->vdev_ops != &vdev_root_ops)
4322 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4324 pvops = &vdev_mirror_ops;
4327 * Active hot spares can only be replaced by inactive hot
4330 if (pvd->vdev_ops == &vdev_spare_ops &&
4331 oldvd->vdev_isspare &&
4332 !spa_has_spare(spa, newvd->vdev_guid))
4333 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4336 * If the source is a hot spare, and the parent isn't already a
4337 * spare, then we want to create a new hot spare. Otherwise, we
4338 * want to create a replacing vdev. The user is not allowed to
4339 * attach to a spared vdev child unless the 'isspare' state is
4340 * the same (spare replaces spare, non-spare replaces
4343 if (pvd->vdev_ops == &vdev_replacing_ops &&
4344 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
4345 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4346 } else if (pvd->vdev_ops == &vdev_spare_ops &&
4347 newvd->vdev_isspare != oldvd->vdev_isspare) {
4348 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4351 if (newvd->vdev_isspare)
4352 pvops = &vdev_spare_ops;
4354 pvops = &vdev_replacing_ops;
4358 * Make sure the new device is big enough.
4360 if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
4361 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
4364 * The new device cannot have a higher alignment requirement
4365 * than the top-level vdev.
4367 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4368 return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4371 * If this is an in-place replacement, update oldvd's path and devid
4372 * to make it distinguishable from newvd, and unopenable from now on.
4374 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4375 spa_strfree(oldvd->vdev_path);
4376 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4378 (void) sprintf(oldvd->vdev_path, "%s/%s",
4379 newvd->vdev_path, "old");
4380 if (oldvd->vdev_devid != NULL) {
4381 spa_strfree(oldvd->vdev_devid);
4382 oldvd->vdev_devid = NULL;
4386 /* mark the device being resilvered */
4387 newvd->vdev_resilvering = B_TRUE;
4390 * If the parent is not a mirror, or if we're replacing, insert the new
4391 * mirror/replacing/spare vdev above oldvd.
4393 if (pvd->vdev_ops != pvops)
4394 pvd = vdev_add_parent(oldvd, pvops);
4396 ASSERT(pvd->vdev_top->vdev_parent == rvd);
4397 ASSERT(pvd->vdev_ops == pvops);
4398 ASSERT(oldvd->vdev_parent == pvd);
4401 * Extract the new device from its root and add it to pvd.
4403 vdev_remove_child(newrootvd, newvd);
4404 newvd->vdev_id = pvd->vdev_children;
4405 newvd->vdev_crtxg = oldvd->vdev_crtxg;
4406 vdev_add_child(pvd, newvd);
4408 tvd = newvd->vdev_top;
4409 ASSERT(pvd->vdev_top == tvd);
4410 ASSERT(tvd->vdev_parent == rvd);
4412 vdev_config_dirty(tvd);
4415 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4416 * for any dmu_sync-ed blocks. It will propagate upward when
4417 * spa_vdev_exit() calls vdev_dtl_reassess().
4419 dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4421 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4422 dtl_max_txg - TXG_INITIAL);
4424 if (newvd->vdev_isspare) {
4425 spa_spare_activate(newvd);
4426 spa_event_notify(spa, newvd, FM_EREPORT_ZFS_DEVICE_SPARE);
4429 oldvdpath = spa_strdup(oldvd->vdev_path);
4430 newvdpath = spa_strdup(newvd->vdev_path);
4431 newvd_isspare = newvd->vdev_isspare;
4434 * Mark newvd's DTL dirty in this txg.
4436 vdev_dirty(tvd, VDD_DTL, newvd, txg);
4439 * Restart the resilver
4441 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4446 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4448 spa_history_log_internal(LOG_POOL_VDEV_ATTACH, spa, NULL,
4449 "%s vdev=%s %s vdev=%s",
4450 replacing && newvd_isspare ? "spare in" :
4451 replacing ? "replace" : "attach", newvdpath,
4452 replacing ? "for" : "to", oldvdpath);
4454 spa_strfree(oldvdpath);
4455 spa_strfree(newvdpath);
4457 if (spa->spa_bootfs)
4458 spa_event_notify(spa, newvd, FM_EREPORT_ZFS_BOOTFS_VDEV_ATTACH);
4464 * Detach a device from a mirror or replacing vdev.
4465 * If 'replace_done' is specified, only detach if the parent
4466 * is a replacing vdev.
4469 spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4473 ASSERTV(vdev_t *rvd = spa->spa_root_vdev;)
4474 vdev_t *vd, *pvd, *cvd, *tvd;
4475 boolean_t unspare = B_FALSE;
4476 uint64_t unspare_guid = 0;
4480 ASSERT(spa_writeable(spa));
4482 txg = spa_vdev_enter(spa);
4484 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4487 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4489 if (!vd->vdev_ops->vdev_op_leaf)
4490 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4492 pvd = vd->vdev_parent;
4495 * If the parent/child relationship is not as expected, don't do it.
4496 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4497 * vdev that's replacing B with C. The user's intent in replacing
4498 * is to go from M(A,B) to M(A,C). If the user decides to cancel
4499 * the replace by detaching C, the expected behavior is to end up
4500 * M(A,B). But suppose that right after deciding to detach C,
4501 * the replacement of B completes. We would have M(A,C), and then
4502 * ask to detach C, which would leave us with just A -- not what
4503 * the user wanted. To prevent this, we make sure that the
4504 * parent/child relationship hasn't changed -- in this example,
4505 * that C's parent is still the replacing vdev R.
4507 if (pvd->vdev_guid != pguid && pguid != 0)
4508 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4511 * Only 'replacing' or 'spare' vdevs can be replaced.
4513 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4514 pvd->vdev_ops != &vdev_spare_ops)
4515 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4517 ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4518 spa_version(spa) >= SPA_VERSION_SPARES);
4521 * Only mirror, replacing, and spare vdevs support detach.
4523 if (pvd->vdev_ops != &vdev_replacing_ops &&
4524 pvd->vdev_ops != &vdev_mirror_ops &&
4525 pvd->vdev_ops != &vdev_spare_ops)
4526 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4529 * If this device has the only valid copy of some data,
4530 * we cannot safely detach it.
4532 if (vdev_dtl_required(vd))
4533 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4535 ASSERT(pvd->vdev_children >= 2);
4538 * If we are detaching the second disk from a replacing vdev, then
4539 * check to see if we changed the original vdev's path to have "/old"
4540 * at the end in spa_vdev_attach(). If so, undo that change now.
4542 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4543 vd->vdev_path != NULL) {
4544 size_t len = strlen(vd->vdev_path);
4546 for (c = 0; c < pvd->vdev_children; c++) {
4547 cvd = pvd->vdev_child[c];
4549 if (cvd == vd || cvd->vdev_path == NULL)
4552 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4553 strcmp(cvd->vdev_path + len, "/old") == 0) {
4554 spa_strfree(cvd->vdev_path);
4555 cvd->vdev_path = spa_strdup(vd->vdev_path);
4562 * If we are detaching the original disk from a spare, then it implies
4563 * that the spare should become a real disk, and be removed from the
4564 * active spare list for the pool.
4566 if (pvd->vdev_ops == &vdev_spare_ops &&
4568 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4572 * Erase the disk labels so the disk can be used for other things.
4573 * This must be done after all other error cases are handled,
4574 * but before we disembowel vd (so we can still do I/O to it).
4575 * But if we can't do it, don't treat the error as fatal --
4576 * it may be that the unwritability of the disk is the reason
4577 * it's being detached!
4579 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4582 * Remove vd from its parent and compact the parent's children.
4584 vdev_remove_child(pvd, vd);
4585 vdev_compact_children(pvd);
4588 * Remember one of the remaining children so we can get tvd below.
4590 cvd = pvd->vdev_child[pvd->vdev_children - 1];
4593 * If we need to remove the remaining child from the list of hot spares,
4594 * do it now, marking the vdev as no longer a spare in the process.
4595 * We must do this before vdev_remove_parent(), because that can
4596 * change the GUID if it creates a new toplevel GUID. For a similar
4597 * reason, we must remove the spare now, in the same txg as the detach;
4598 * otherwise someone could attach a new sibling, change the GUID, and
4599 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4602 ASSERT(cvd->vdev_isspare);
4603 spa_spare_remove(cvd);
4604 unspare_guid = cvd->vdev_guid;
4605 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4606 cvd->vdev_unspare = B_TRUE;
4610 * If the parent mirror/replacing vdev only has one child,
4611 * the parent is no longer needed. Remove it from the tree.
4613 if (pvd->vdev_children == 1) {
4614 if (pvd->vdev_ops == &vdev_spare_ops)
4615 cvd->vdev_unspare = B_FALSE;
4616 vdev_remove_parent(cvd);
4617 cvd->vdev_resilvering = B_FALSE;
4622 * We don't set tvd until now because the parent we just removed
4623 * may have been the previous top-level vdev.
4625 tvd = cvd->vdev_top;
4626 ASSERT(tvd->vdev_parent == rvd);
4629 * Reevaluate the parent vdev state.
4631 vdev_propagate_state(cvd);
4634 * If the 'autoexpand' property is set on the pool then automatically
4635 * try to expand the size of the pool. For example if the device we
4636 * just detached was smaller than the others, it may be possible to
4637 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4638 * first so that we can obtain the updated sizes of the leaf vdevs.
4640 if (spa->spa_autoexpand) {
4642 vdev_expand(tvd, txg);
4645 vdev_config_dirty(tvd);
4648 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
4649 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4650 * But first make sure we're not on any *other* txg's DTL list, to
4651 * prevent vd from being accessed after it's freed.
4653 vdpath = spa_strdup(vd->vdev_path);
4654 for (t = 0; t < TXG_SIZE; t++)
4655 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4656 vd->vdev_detached = B_TRUE;
4657 vdev_dirty(tvd, VDD_DTL, vd, txg);
4659 spa_event_notify(spa, vd, FM_EREPORT_ZFS_DEVICE_REMOVE);
4661 /* hang on to the spa before we release the lock */
4662 spa_open_ref(spa, FTAG);
4664 error = spa_vdev_exit(spa, vd, txg, 0);
4666 spa_history_log_internal(LOG_POOL_VDEV_DETACH, spa, NULL,
4668 spa_strfree(vdpath);
4671 * If this was the removal of the original device in a hot spare vdev,
4672 * then we want to go through and remove the device from the hot spare
4673 * list of every other pool.
4676 spa_t *altspa = NULL;
4678 mutex_enter(&spa_namespace_lock);
4679 while ((altspa = spa_next(altspa)) != NULL) {
4680 if (altspa->spa_state != POOL_STATE_ACTIVE ||
4684 spa_open_ref(altspa, FTAG);
4685 mutex_exit(&spa_namespace_lock);
4686 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
4687 mutex_enter(&spa_namespace_lock);
4688 spa_close(altspa, FTAG);
4690 mutex_exit(&spa_namespace_lock);
4692 /* search the rest of the vdevs for spares to remove */
4693 spa_vdev_resilver_done(spa);
4696 /* all done with the spa; OK to release */
4697 mutex_enter(&spa_namespace_lock);
4698 spa_close(spa, FTAG);
4699 mutex_exit(&spa_namespace_lock);
4705 * Split a set of devices from their mirrors, and create a new pool from them.
4708 spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
4709 nvlist_t *props, boolean_t exp)
4712 uint64_t txg, *glist;
4714 uint_t c, children, lastlog;
4715 nvlist_t **child, *nvl, *tmp;
4717 char *altroot = NULL;
4718 vdev_t *rvd, **vml = NULL; /* vdev modify list */
4719 boolean_t activate_slog;
4721 ASSERT(spa_writeable(spa));
4723 txg = spa_vdev_enter(spa);
4725 /* clear the log and flush everything up to now */
4726 activate_slog = spa_passivate_log(spa);
4727 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4728 error = spa_offline_log(spa);
4729 txg = spa_vdev_config_enter(spa);
4732 spa_activate_log(spa);
4735 return (spa_vdev_exit(spa, NULL, txg, error));
4737 /* check new spa name before going any further */
4738 if (spa_lookup(newname) != NULL)
4739 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
4742 * scan through all the children to ensure they're all mirrors
4744 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
4745 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
4747 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4749 /* first, check to ensure we've got the right child count */
4750 rvd = spa->spa_root_vdev;
4752 for (c = 0; c < rvd->vdev_children; c++) {
4753 vdev_t *vd = rvd->vdev_child[c];
4755 /* don't count the holes & logs as children */
4756 if (vd->vdev_islog || vd->vdev_ishole) {
4764 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
4765 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4767 /* next, ensure no spare or cache devices are part of the split */
4768 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
4769 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
4770 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4772 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_PUSHPAGE);
4773 glist = kmem_zalloc(children * sizeof (uint64_t), KM_PUSHPAGE);
4775 /* then, loop over each vdev and validate it */
4776 for (c = 0; c < children; c++) {
4777 uint64_t is_hole = 0;
4779 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
4783 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
4784 spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
4792 /* which disk is going to be split? */
4793 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
4799 /* look it up in the spa */
4800 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
4801 if (vml[c] == NULL) {
4806 /* make sure there's nothing stopping the split */
4807 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
4808 vml[c]->vdev_islog ||
4809 vml[c]->vdev_ishole ||
4810 vml[c]->vdev_isspare ||
4811 vml[c]->vdev_isl2cache ||
4812 !vdev_writeable(vml[c]) ||
4813 vml[c]->vdev_children != 0 ||
4814 vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
4815 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
4820 if (vdev_dtl_required(vml[c])) {
4825 /* we need certain info from the top level */
4826 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
4827 vml[c]->vdev_top->vdev_ms_array) == 0);
4828 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
4829 vml[c]->vdev_top->vdev_ms_shift) == 0);
4830 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
4831 vml[c]->vdev_top->vdev_asize) == 0);
4832 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
4833 vml[c]->vdev_top->vdev_ashift) == 0);
4837 kmem_free(vml, children * sizeof (vdev_t *));
4838 kmem_free(glist, children * sizeof (uint64_t));
4839 return (spa_vdev_exit(spa, NULL, txg, error));
4842 /* stop writers from using the disks */
4843 for (c = 0; c < children; c++) {
4845 vml[c]->vdev_offline = B_TRUE;
4847 vdev_reopen(spa->spa_root_vdev);
4850 * Temporarily record the splitting vdevs in the spa config. This
4851 * will disappear once the config is regenerated.
4853 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_PUSHPAGE) == 0);
4854 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
4855 glist, children) == 0);
4856 kmem_free(glist, children * sizeof (uint64_t));
4858 mutex_enter(&spa->spa_props_lock);
4859 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
4861 mutex_exit(&spa->spa_props_lock);
4862 spa->spa_config_splitting = nvl;
4863 vdev_config_dirty(spa->spa_root_vdev);
4865 /* configure and create the new pool */
4866 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
4867 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4868 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
4869 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
4870 spa_version(spa)) == 0);
4871 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
4872 spa->spa_config_txg) == 0);
4873 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4874 spa_generate_guid(NULL)) == 0);
4875 (void) nvlist_lookup_string(props,
4876 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4878 /* add the new pool to the namespace */
4879 newspa = spa_add(newname, config, altroot);
4880 newspa->spa_config_txg = spa->spa_config_txg;
4881 spa_set_log_state(newspa, SPA_LOG_CLEAR);
4883 /* release the spa config lock, retaining the namespace lock */
4884 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4886 if (zio_injection_enabled)
4887 zio_handle_panic_injection(spa, FTAG, 1);
4889 spa_activate(newspa, spa_mode_global);
4890 spa_async_suspend(newspa);
4892 /* create the new pool from the disks of the original pool */
4893 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
4897 /* if that worked, generate a real config for the new pool */
4898 if (newspa->spa_root_vdev != NULL) {
4899 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
4900 NV_UNIQUE_NAME, KM_PUSHPAGE) == 0);
4901 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
4902 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
4903 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
4908 if (props != NULL) {
4909 spa_configfile_set(newspa, props, B_FALSE);
4910 error = spa_prop_set(newspa, props);
4915 /* flush everything */
4916 txg = spa_vdev_config_enter(newspa);
4917 vdev_config_dirty(newspa->spa_root_vdev);
4918 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
4920 if (zio_injection_enabled)
4921 zio_handle_panic_injection(spa, FTAG, 2);
4923 spa_async_resume(newspa);
4925 /* finally, update the original pool's config */
4926 txg = spa_vdev_config_enter(spa);
4927 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
4928 error = dmu_tx_assign(tx, TXG_WAIT);
4931 for (c = 0; c < children; c++) {
4932 if (vml[c] != NULL) {
4935 spa_history_log_internal(LOG_POOL_VDEV_DETACH,
4941 vdev_config_dirty(spa->spa_root_vdev);
4942 spa->spa_config_splitting = NULL;
4946 (void) spa_vdev_exit(spa, NULL, txg, 0);
4948 if (zio_injection_enabled)
4949 zio_handle_panic_injection(spa, FTAG, 3);
4951 /* split is complete; log a history record */
4952 spa_history_log_internal(LOG_POOL_SPLIT, newspa, NULL,
4953 "split new pool %s from pool %s", newname, spa_name(spa));
4955 kmem_free(vml, children * sizeof (vdev_t *));
4957 /* if we're not going to mount the filesystems in userland, export */
4959 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
4966 spa_deactivate(newspa);
4969 txg = spa_vdev_config_enter(spa);
4971 /* re-online all offlined disks */
4972 for (c = 0; c < children; c++) {
4974 vml[c]->vdev_offline = B_FALSE;
4976 vdev_reopen(spa->spa_root_vdev);
4978 nvlist_free(spa->spa_config_splitting);
4979 spa->spa_config_splitting = NULL;
4980 (void) spa_vdev_exit(spa, NULL, txg, error);
4982 kmem_free(vml, children * sizeof (vdev_t *));
4987 spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
4991 for (i = 0; i < count; i++) {
4994 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
4997 if (guid == target_guid)
5005 spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
5006 nvlist_t *dev_to_remove)
5008 nvlist_t **newdev = NULL;
5012 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_PUSHPAGE);
5014 for (i = 0, j = 0; i < count; i++) {
5015 if (dev[i] == dev_to_remove)
5017 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_PUSHPAGE) == 0);
5020 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
5021 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
5023 for (i = 0; i < count - 1; i++)
5024 nvlist_free(newdev[i]);
5027 kmem_free(newdev, (count - 1) * sizeof (void *));
5031 * Evacuate the device.
5034 spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
5039 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5040 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5041 ASSERT(vd == vd->vdev_top);
5044 * Evacuate the device. We don't hold the config lock as writer
5045 * since we need to do I/O but we do keep the
5046 * spa_namespace_lock held. Once this completes the device
5047 * should no longer have any blocks allocated on it.
5049 if (vd->vdev_islog) {
5050 if (vd->vdev_stat.vs_alloc != 0)
5051 error = spa_offline_log(spa);
5060 * The evacuation succeeded. Remove any remaining MOS metadata
5061 * associated with this vdev, and wait for these changes to sync.
5063 ASSERT3U(vd->vdev_stat.vs_alloc, ==, 0);
5064 txg = spa_vdev_config_enter(spa);
5065 vd->vdev_removing = B_TRUE;
5066 vdev_dirty(vd, 0, NULL, txg);
5067 vdev_config_dirty(vd);
5068 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5074 * Complete the removal by cleaning up the namespace.
5077 spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
5079 vdev_t *rvd = spa->spa_root_vdev;
5080 uint64_t id = vd->vdev_id;
5081 boolean_t last_vdev = (id == (rvd->vdev_children - 1));
5083 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5084 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5085 ASSERT(vd == vd->vdev_top);
5088 * Only remove any devices which are empty.
5090 if (vd->vdev_stat.vs_alloc != 0)
5093 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5095 if (list_link_active(&vd->vdev_state_dirty_node))
5096 vdev_state_clean(vd);
5097 if (list_link_active(&vd->vdev_config_dirty_node))
5098 vdev_config_clean(vd);
5103 vdev_compact_children(rvd);
5105 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
5106 vdev_add_child(rvd, vd);
5108 vdev_config_dirty(rvd);
5111 * Reassess the health of our root vdev.
5117 * Remove a device from the pool -
5119 * Removing a device from the vdev namespace requires several steps
5120 * and can take a significant amount of time. As a result we use
5121 * the spa_vdev_config_[enter/exit] functions which allow us to
5122 * grab and release the spa_config_lock while still holding the namespace
5123 * lock. During each step the configuration is synced out.
5127 * Remove a device from the pool. Currently, this supports removing only hot
5128 * spares, slogs, and level 2 ARC devices.
5131 spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
5134 metaslab_group_t *mg;
5135 nvlist_t **spares, **l2cache, *nv;
5137 uint_t nspares, nl2cache;
5139 boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
5141 ASSERT(spa_writeable(spa));
5144 txg = spa_vdev_enter(spa);
5146 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5148 if (spa->spa_spares.sav_vdevs != NULL &&
5149 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5150 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
5151 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
5153 * Only remove the hot spare if it's not currently in use
5156 if (vd == NULL || unspare) {
5157 spa_vdev_remove_aux(spa->spa_spares.sav_config,
5158 ZPOOL_CONFIG_SPARES, spares, nspares, nv);
5159 spa_load_spares(spa);
5160 spa->spa_spares.sav_sync = B_TRUE;
5164 } else if (spa->spa_l2cache.sav_vdevs != NULL &&
5165 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5166 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
5167 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
5169 * Cache devices can always be removed.
5171 spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
5172 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
5173 spa_load_l2cache(spa);
5174 spa->spa_l2cache.sav_sync = B_TRUE;
5175 } else if (vd != NULL && vd->vdev_islog) {
5177 ASSERT(vd == vd->vdev_top);
5180 * XXX - Once we have bp-rewrite this should
5181 * become the common case.
5187 * Stop allocating from this vdev.
5189 metaslab_group_passivate(mg);
5192 * Wait for the youngest allocations and frees to sync,
5193 * and then wait for the deferral of those frees to finish.
5195 spa_vdev_config_exit(spa, NULL,
5196 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
5199 * Attempt to evacuate the vdev.
5201 error = spa_vdev_remove_evacuate(spa, vd);
5203 txg = spa_vdev_config_enter(spa);
5206 * If we couldn't evacuate the vdev, unwind.
5209 metaslab_group_activate(mg);
5210 return (spa_vdev_exit(spa, NULL, txg, error));
5214 * Clean up the vdev namespace.
5216 spa_vdev_remove_from_namespace(spa, vd);
5218 } else if (vd != NULL) {
5220 * Normal vdevs cannot be removed (yet).
5225 * There is no vdev of any kind with the specified guid.
5231 return (spa_vdev_exit(spa, NULL, txg, error));
5237 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5238 * current spared, so we can detach it.
5241 spa_vdev_resilver_done_hunt(vdev_t *vd)
5243 vdev_t *newvd, *oldvd;
5246 for (c = 0; c < vd->vdev_children; c++) {
5247 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
5253 * Check for a completed replacement. We always consider the first
5254 * vdev in the list to be the oldest vdev, and the last one to be
5255 * the newest (see spa_vdev_attach() for how that works). In
5256 * the case where the newest vdev is faulted, we will not automatically
5257 * remove it after a resilver completes. This is OK as it will require
5258 * user intervention to determine which disk the admin wishes to keep.
5260 if (vd->vdev_ops == &vdev_replacing_ops) {
5261 ASSERT(vd->vdev_children > 1);
5263 newvd = vd->vdev_child[vd->vdev_children - 1];
5264 oldvd = vd->vdev_child[0];
5266 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
5267 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5268 !vdev_dtl_required(oldvd))
5273 * Check for a completed resilver with the 'unspare' flag set.
5275 if (vd->vdev_ops == &vdev_spare_ops) {
5276 vdev_t *first = vd->vdev_child[0];
5277 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
5279 if (last->vdev_unspare) {
5282 } else if (first->vdev_unspare) {
5289 if (oldvd != NULL &&
5290 vdev_dtl_empty(newvd, DTL_MISSING) &&
5291 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5292 !vdev_dtl_required(oldvd))
5296 * If there are more than two spares attached to a disk,
5297 * and those spares are not required, then we want to
5298 * attempt to free them up now so that they can be used
5299 * by other pools. Once we're back down to a single
5300 * disk+spare, we stop removing them.
5302 if (vd->vdev_children > 2) {
5303 newvd = vd->vdev_child[1];
5305 if (newvd->vdev_isspare && last->vdev_isspare &&
5306 vdev_dtl_empty(last, DTL_MISSING) &&
5307 vdev_dtl_empty(last, DTL_OUTAGE) &&
5308 !vdev_dtl_required(newvd))
5317 spa_vdev_resilver_done(spa_t *spa)
5319 vdev_t *vd, *pvd, *ppvd;
5320 uint64_t guid, sguid, pguid, ppguid;
5322 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5324 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
5325 pvd = vd->vdev_parent;
5326 ppvd = pvd->vdev_parent;
5327 guid = vd->vdev_guid;
5328 pguid = pvd->vdev_guid;
5329 ppguid = ppvd->vdev_guid;
5332 * If we have just finished replacing a hot spared device, then
5333 * we need to detach the parent's first child (the original hot
5336 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
5337 ppvd->vdev_children == 2) {
5338 ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
5339 sguid = ppvd->vdev_child[1]->vdev_guid;
5341 spa_config_exit(spa, SCL_ALL, FTAG);
5342 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
5344 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
5346 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5349 spa_config_exit(spa, SCL_ALL, FTAG);
5353 * Update the stored path or FRU for this vdev.
5356 spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
5360 boolean_t sync = B_FALSE;
5362 ASSERT(spa_writeable(spa));
5364 spa_vdev_state_enter(spa, SCL_ALL);
5366 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
5367 return (spa_vdev_state_exit(spa, NULL, ENOENT));
5369 if (!vd->vdev_ops->vdev_op_leaf)
5370 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
5373 if (strcmp(value, vd->vdev_path) != 0) {
5374 spa_strfree(vd->vdev_path);
5375 vd->vdev_path = spa_strdup(value);
5379 if (vd->vdev_fru == NULL) {
5380 vd->vdev_fru = spa_strdup(value);
5382 } else if (strcmp(value, vd->vdev_fru) != 0) {
5383 spa_strfree(vd->vdev_fru);
5384 vd->vdev_fru = spa_strdup(value);
5389 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
5393 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
5395 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5399 spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5401 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5405 * ==========================================================================
5407 * ==========================================================================
5411 spa_scan_stop(spa_t *spa)
5413 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5414 if (dsl_scan_resilvering(spa->spa_dsl_pool))
5416 return (dsl_scan_cancel(spa->spa_dsl_pool));
5420 spa_scan(spa_t *spa, pool_scan_func_t func)
5422 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5424 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5428 * If a resilver was requested, but there is no DTL on a
5429 * writeable leaf device, we have nothing to do.
5431 if (func == POOL_SCAN_RESILVER &&
5432 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5433 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5437 return (dsl_scan(spa->spa_dsl_pool, func));
5441 * ==========================================================================
5442 * SPA async task processing
5443 * ==========================================================================
5447 spa_async_remove(spa_t *spa, vdev_t *vd)
5451 if (vd->vdev_remove_wanted) {
5452 vd->vdev_remove_wanted = B_FALSE;
5453 vd->vdev_delayed_close = B_FALSE;
5454 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5457 * We want to clear the stats, but we don't want to do a full
5458 * vdev_clear() as that will cause us to throw away
5459 * degraded/faulted state as well as attempt to reopen the
5460 * device, all of which is a waste.
5462 vd->vdev_stat.vs_read_errors = 0;
5463 vd->vdev_stat.vs_write_errors = 0;
5464 vd->vdev_stat.vs_checksum_errors = 0;
5466 vdev_state_dirty(vd->vdev_top);
5469 for (c = 0; c < vd->vdev_children; c++)
5470 spa_async_remove(spa, vd->vdev_child[c]);
5474 spa_async_probe(spa_t *spa, vdev_t *vd)
5478 if (vd->vdev_probe_wanted) {
5479 vd->vdev_probe_wanted = B_FALSE;
5480 vdev_reopen(vd); /* vdev_open() does the actual probe */
5483 for (c = 0; c < vd->vdev_children; c++)
5484 spa_async_probe(spa, vd->vdev_child[c]);
5488 spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5492 if (!spa->spa_autoexpand)
5495 for (c = 0; c < vd->vdev_children; c++) {
5496 vdev_t *cvd = vd->vdev_child[c];
5497 spa_async_autoexpand(spa, cvd);
5500 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5503 spa_event_notify(vd->vdev_spa, vd, FM_EREPORT_ZFS_DEVICE_AUTOEXPAND);
5507 spa_async_thread(spa_t *spa)
5511 ASSERT(spa->spa_sync_on);
5513 mutex_enter(&spa->spa_async_lock);
5514 tasks = spa->spa_async_tasks;
5515 spa->spa_async_tasks = 0;
5516 mutex_exit(&spa->spa_async_lock);
5519 * See if the config needs to be updated.
5521 if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5522 uint64_t old_space, new_space;
5524 mutex_enter(&spa_namespace_lock);
5525 old_space = metaslab_class_get_space(spa_normal_class(spa));
5526 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5527 new_space = metaslab_class_get_space(spa_normal_class(spa));
5528 mutex_exit(&spa_namespace_lock);
5531 * If the pool grew as a result of the config update,
5532 * then log an internal history event.
5534 if (new_space != old_space) {
5535 spa_history_log_internal(LOG_POOL_VDEV_ONLINE,
5537 "pool '%s' size: %llu(+%llu)",
5538 spa_name(spa), new_space, new_space - old_space);
5543 * See if any devices need to be marked REMOVED.
5545 if (tasks & SPA_ASYNC_REMOVE) {
5546 spa_vdev_state_enter(spa, SCL_NONE);
5547 spa_async_remove(spa, spa->spa_root_vdev);
5548 for (i = 0; i < spa->spa_l2cache.sav_count; i++)
5549 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5550 for (i = 0; i < spa->spa_spares.sav_count; i++)
5551 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5552 (void) spa_vdev_state_exit(spa, NULL, 0);
5555 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5556 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5557 spa_async_autoexpand(spa, spa->spa_root_vdev);
5558 spa_config_exit(spa, SCL_CONFIG, FTAG);
5562 * See if any devices need to be probed.
5564 if (tasks & SPA_ASYNC_PROBE) {
5565 spa_vdev_state_enter(spa, SCL_NONE);
5566 spa_async_probe(spa, spa->spa_root_vdev);
5567 (void) spa_vdev_state_exit(spa, NULL, 0);
5571 * If any devices are done replacing, detach them.
5573 if (tasks & SPA_ASYNC_RESILVER_DONE)
5574 spa_vdev_resilver_done(spa);
5577 * Kick off a resilver.
5579 if (tasks & SPA_ASYNC_RESILVER)
5580 dsl_resilver_restart(spa->spa_dsl_pool, 0);
5583 * Let the world know that we're done.
5585 mutex_enter(&spa->spa_async_lock);
5586 spa->spa_async_thread = NULL;
5587 cv_broadcast(&spa->spa_async_cv);
5588 mutex_exit(&spa->spa_async_lock);
5593 spa_async_suspend(spa_t *spa)
5595 mutex_enter(&spa->spa_async_lock);
5596 spa->spa_async_suspended++;
5597 while (spa->spa_async_thread != NULL)
5598 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5599 mutex_exit(&spa->spa_async_lock);
5603 spa_async_resume(spa_t *spa)
5605 mutex_enter(&spa->spa_async_lock);
5606 ASSERT(spa->spa_async_suspended != 0);
5607 spa->spa_async_suspended--;
5608 mutex_exit(&spa->spa_async_lock);
5612 spa_async_dispatch(spa_t *spa)
5614 mutex_enter(&spa->spa_async_lock);
5615 if (spa->spa_async_tasks && !spa->spa_async_suspended &&
5616 spa->spa_async_thread == NULL &&
5617 rootdir != NULL && !vn_is_readonly(rootdir))
5618 spa->spa_async_thread = thread_create(NULL, 0,
5619 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
5620 mutex_exit(&spa->spa_async_lock);
5624 spa_async_request(spa_t *spa, int task)
5626 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
5627 mutex_enter(&spa->spa_async_lock);
5628 spa->spa_async_tasks |= task;
5629 mutex_exit(&spa->spa_async_lock);
5633 * ==========================================================================
5634 * SPA syncing routines
5635 * ==========================================================================
5639 bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5642 bpobj_enqueue(bpo, bp, tx);
5647 spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5651 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
5657 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
5659 char *packed = NULL;
5664 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
5667 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5668 * information. This avoids the dbuf_will_dirty() path and
5669 * saves us a pre-read to get data we don't actually care about.
5671 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
5672 packed = vmem_alloc(bufsize, KM_PUSHPAGE);
5674 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
5676 bzero(packed + nvsize, bufsize - nvsize);
5678 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
5680 vmem_free(packed, bufsize);
5682 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
5683 dmu_buf_will_dirty(db, tx);
5684 *(uint64_t *)db->db_data = nvsize;
5685 dmu_buf_rele(db, FTAG);
5689 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
5690 const char *config, const char *entry)
5700 * Update the MOS nvlist describing the list of available devices.
5701 * spa_validate_aux() will have already made sure this nvlist is
5702 * valid and the vdevs are labeled appropriately.
5704 if (sav->sav_object == 0) {
5705 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
5706 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
5707 sizeof (uint64_t), tx);
5708 VERIFY(zap_update(spa->spa_meta_objset,
5709 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
5710 &sav->sav_object, tx) == 0);
5713 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_PUSHPAGE) == 0);
5714 if (sav->sav_count == 0) {
5715 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
5717 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_PUSHPAGE);
5718 for (i = 0; i < sav->sav_count; i++)
5719 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
5720 B_FALSE, VDEV_CONFIG_L2CACHE);
5721 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
5722 sav->sav_count) == 0);
5723 for (i = 0; i < sav->sav_count; i++)
5724 nvlist_free(list[i]);
5725 kmem_free(list, sav->sav_count * sizeof (void *));
5728 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
5729 nvlist_free(nvroot);
5731 sav->sav_sync = B_FALSE;
5735 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
5739 if (list_is_empty(&spa->spa_config_dirty_list))
5742 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5744 config = spa_config_generate(spa, spa->spa_root_vdev,
5745 dmu_tx_get_txg(tx), B_FALSE);
5748 * If we're upgrading the spa version then make sure that
5749 * the config object gets updated with the correct version.
5751 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
5752 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5753 spa->spa_uberblock.ub_version);
5755 spa_config_exit(spa, SCL_STATE, FTAG);
5757 if (spa->spa_config_syncing)
5758 nvlist_free(spa->spa_config_syncing);
5759 spa->spa_config_syncing = config;
5761 spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
5765 spa_sync_version(void *arg1, void *arg2, dmu_tx_t *tx)
5768 uint64_t version = *(uint64_t *)arg2;
5771 * Setting the version is special cased when first creating the pool.
5773 ASSERT(tx->tx_txg != TXG_INITIAL);
5775 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
5776 ASSERT(version >= spa_version(spa));
5778 spa->spa_uberblock.ub_version = version;
5779 vdev_config_dirty(spa->spa_root_vdev);
5783 * Set zpool properties.
5786 spa_sync_props(void *arg1, void *arg2, dmu_tx_t *tx)
5789 objset_t *mos = spa->spa_meta_objset;
5790 nvlist_t *nvp = arg2;
5791 nvpair_t *elem = NULL;
5793 mutex_enter(&spa->spa_props_lock);
5795 while ((elem = nvlist_next_nvpair(nvp, elem))) {
5797 char *strval, *fname;
5799 const char *propname;
5800 zprop_type_t proptype;
5801 zfeature_info_t *feature;
5803 prop = zpool_name_to_prop(nvpair_name(elem));
5804 switch ((int)prop) {
5807 * We checked this earlier in spa_prop_validate().
5809 ASSERT(zpool_prop_feature(nvpair_name(elem)));
5811 fname = strchr(nvpair_name(elem), '@') + 1;
5812 VERIFY3U(0, ==, zfeature_lookup_name(fname, &feature));
5814 spa_feature_enable(spa, feature, tx);
5817 case ZPOOL_PROP_VERSION:
5818 VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5820 * The version is synced seperatly before other
5821 * properties and should be correct by now.
5823 ASSERT3U(spa_version(spa), >=, intval);
5826 case ZPOOL_PROP_ALTROOT:
5828 * 'altroot' is a non-persistent property. It should
5829 * have been set temporarily at creation or import time.
5831 ASSERT(spa->spa_root != NULL);
5834 case ZPOOL_PROP_READONLY:
5835 case ZPOOL_PROP_CACHEFILE:
5837 * 'readonly' and 'cachefile' are also non-persisitent
5841 case ZPOOL_PROP_COMMENT:
5842 VERIFY(nvpair_value_string(elem, &strval) == 0);
5843 if (spa->spa_comment != NULL)
5844 spa_strfree(spa->spa_comment);
5845 spa->spa_comment = spa_strdup(strval);
5847 * We need to dirty the configuration on all the vdevs
5848 * so that their labels get updated. It's unnecessary
5849 * to do this for pool creation since the vdev's
5850 * configuratoin has already been dirtied.
5852 if (tx->tx_txg != TXG_INITIAL)
5853 vdev_config_dirty(spa->spa_root_vdev);
5857 * Set pool property values in the poolprops mos object.
5859 if (spa->spa_pool_props_object == 0) {
5860 spa->spa_pool_props_object =
5861 zap_create_link(mos, DMU_OT_POOL_PROPS,
5862 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
5866 /* normalize the property name */
5867 propname = zpool_prop_to_name(prop);
5868 proptype = zpool_prop_get_type(prop);
5870 if (nvpair_type(elem) == DATA_TYPE_STRING) {
5871 ASSERT(proptype == PROP_TYPE_STRING);
5872 VERIFY(nvpair_value_string(elem, &strval) == 0);
5873 VERIFY(zap_update(mos,
5874 spa->spa_pool_props_object, propname,
5875 1, strlen(strval) + 1, strval, tx) == 0);
5877 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
5878 VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5880 if (proptype == PROP_TYPE_INDEX) {
5882 VERIFY(zpool_prop_index_to_string(
5883 prop, intval, &unused) == 0);
5885 VERIFY(zap_update(mos,
5886 spa->spa_pool_props_object, propname,
5887 8, 1, &intval, tx) == 0);
5889 ASSERT(0); /* not allowed */
5893 case ZPOOL_PROP_DELEGATION:
5894 spa->spa_delegation = intval;
5896 case ZPOOL_PROP_BOOTFS:
5897 spa->spa_bootfs = intval;
5899 case ZPOOL_PROP_FAILUREMODE:
5900 spa->spa_failmode = intval;
5902 case ZPOOL_PROP_AUTOEXPAND:
5903 spa->spa_autoexpand = intval;
5904 if (tx->tx_txg != TXG_INITIAL)
5905 spa_async_request(spa,
5906 SPA_ASYNC_AUTOEXPAND);
5908 case ZPOOL_PROP_DEDUPDITTO:
5909 spa->spa_dedup_ditto = intval;
5916 /* log internal history if this is not a zpool create */
5917 if (spa_version(spa) >= SPA_VERSION_ZPOOL_HISTORY &&
5918 tx->tx_txg != TXG_INITIAL) {
5919 spa_history_log_internal(LOG_POOL_PROPSET,
5920 spa, tx, "%s %lld %s",
5921 nvpair_name(elem), intval, spa_name(spa));
5925 mutex_exit(&spa->spa_props_lock);
5929 * Perform one-time upgrade on-disk changes. spa_version() does not
5930 * reflect the new version this txg, so there must be no changes this
5931 * txg to anything that the upgrade code depends on after it executes.
5932 * Therefore this must be called after dsl_pool_sync() does the sync
5936 spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
5938 dsl_pool_t *dp = spa->spa_dsl_pool;
5940 ASSERT(spa->spa_sync_pass == 1);
5942 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
5943 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
5944 dsl_pool_create_origin(dp, tx);
5946 /* Keeping the origin open increases spa_minref */
5947 spa->spa_minref += 3;
5950 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
5951 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
5952 dsl_pool_upgrade_clones(dp, tx);
5955 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
5956 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
5957 dsl_pool_upgrade_dir_clones(dp, tx);
5959 /* Keeping the freedir open increases spa_minref */
5960 spa->spa_minref += 3;
5963 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES &&
5964 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
5965 spa_feature_create_zap_objects(spa, tx);
5970 * Sync the specified transaction group. New blocks may be dirtied as
5971 * part of the process, so we iterate until it converges.
5974 spa_sync(spa_t *spa, uint64_t txg)
5976 dsl_pool_t *dp = spa->spa_dsl_pool;
5977 objset_t *mos = spa->spa_meta_objset;
5978 bpobj_t *defer_bpo = &spa->spa_deferred_bpobj;
5979 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
5980 vdev_t *rvd = spa->spa_root_vdev;
5986 VERIFY(spa_writeable(spa));
5989 * Lock out configuration changes.
5991 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5993 spa->spa_syncing_txg = txg;
5994 spa->spa_sync_pass = 0;
5997 * If there are any pending vdev state changes, convert them
5998 * into config changes that go out with this transaction group.
6000 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6001 while (list_head(&spa->spa_state_dirty_list) != NULL) {
6003 * We need the write lock here because, for aux vdevs,
6004 * calling vdev_config_dirty() modifies sav_config.
6005 * This is ugly and will become unnecessary when we
6006 * eliminate the aux vdev wart by integrating all vdevs
6007 * into the root vdev tree.
6009 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6010 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
6011 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
6012 vdev_state_clean(vd);
6013 vdev_config_dirty(vd);
6015 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6016 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
6018 spa_config_exit(spa, SCL_STATE, FTAG);
6020 tx = dmu_tx_create_assigned(dp, txg);
6022 spa->spa_sync_starttime = gethrtime();
6023 taskq_cancel_id(system_taskq, spa->spa_deadman_tqid);
6024 spa->spa_deadman_tqid = taskq_dispatch_delay(system_taskq,
6025 spa_deadman, spa, TQ_SLEEP, ddi_get_lbolt() +
6026 NSEC_TO_TICK(spa->spa_deadman_synctime));
6029 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
6030 * set spa_deflate if we have no raid-z vdevs.
6032 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
6033 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
6036 for (i = 0; i < rvd->vdev_children; i++) {
6037 vd = rvd->vdev_child[i];
6038 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
6041 if (i == rvd->vdev_children) {
6042 spa->spa_deflate = TRUE;
6043 VERIFY(0 == zap_add(spa->spa_meta_objset,
6044 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
6045 sizeof (uint64_t), 1, &spa->spa_deflate, tx));
6050 * If anything has changed in this txg, or if someone is waiting
6051 * for this txg to sync (eg, spa_vdev_remove()), push the
6052 * deferred frees from the previous txg. If not, leave them
6053 * alone so that we don't generate work on an otherwise idle
6056 if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
6057 !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
6058 !txg_list_empty(&dp->dp_sync_tasks, txg) ||
6059 ((dsl_scan_active(dp->dp_scan) ||
6060 txg_sync_waiting(dp)) && !spa_shutting_down(spa))) {
6061 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6062 VERIFY3U(bpobj_iterate(defer_bpo,
6063 spa_free_sync_cb, zio, tx), ==, 0);
6064 VERIFY3U(zio_wait(zio), ==, 0);
6068 * Iterate to convergence.
6071 int pass = ++spa->spa_sync_pass;
6073 spa_sync_config_object(spa, tx);
6074 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
6075 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
6076 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
6077 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
6078 spa_errlog_sync(spa, txg);
6079 dsl_pool_sync(dp, txg);
6081 if (pass < zfs_sync_pass_deferred_free) {
6082 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6083 bplist_iterate(free_bpl, spa_free_sync_cb,
6085 VERIFY(zio_wait(zio) == 0);
6087 bplist_iterate(free_bpl, bpobj_enqueue_cb,
6092 dsl_scan_sync(dp, tx);
6094 while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)))
6098 spa_sync_upgrades(spa, tx);
6100 } while (dmu_objset_is_dirty(mos, txg));
6103 * Rewrite the vdev configuration (which includes the uberblock)
6104 * to commit the transaction group.
6106 * If there are no dirty vdevs, we sync the uberblock to a few
6107 * random top-level vdevs that are known to be visible in the
6108 * config cache (see spa_vdev_add() for a complete description).
6109 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
6113 * We hold SCL_STATE to prevent vdev open/close/etc.
6114 * while we're attempting to write the vdev labels.
6116 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6118 if (list_is_empty(&spa->spa_config_dirty_list)) {
6119 vdev_t *svd[SPA_DVAS_PER_BP];
6121 int children = rvd->vdev_children;
6122 int c0 = spa_get_random(children);
6124 for (c = 0; c < children; c++) {
6125 vd = rvd->vdev_child[(c0 + c) % children];
6126 if (vd->vdev_ms_array == 0 || vd->vdev_islog)
6128 svd[svdcount++] = vd;
6129 if (svdcount == SPA_DVAS_PER_BP)
6132 error = vdev_config_sync(svd, svdcount, txg, B_FALSE);
6134 error = vdev_config_sync(svd, svdcount, txg,
6137 error = vdev_config_sync(rvd->vdev_child,
6138 rvd->vdev_children, txg, B_FALSE);
6140 error = vdev_config_sync(rvd->vdev_child,
6141 rvd->vdev_children, txg, B_TRUE);
6145 spa->spa_last_synced_guid = rvd->vdev_guid;
6147 spa_config_exit(spa, SCL_STATE, FTAG);
6151 zio_suspend(spa, NULL);
6152 zio_resume_wait(spa);
6156 taskq_cancel_id(system_taskq, spa->spa_deadman_tqid);
6157 spa->spa_deadman_tqid = 0;
6160 * Clear the dirty config list.
6162 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
6163 vdev_config_clean(vd);
6166 * Now that the new config has synced transactionally,
6167 * let it become visible to the config cache.
6169 if (spa->spa_config_syncing != NULL) {
6170 spa_config_set(spa, spa->spa_config_syncing);
6171 spa->spa_config_txg = txg;
6172 spa->spa_config_syncing = NULL;
6175 spa->spa_ubsync = spa->spa_uberblock;
6177 dsl_pool_sync_done(dp, txg);
6180 * Update usable space statistics.
6182 while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg))))
6183 vdev_sync_done(vd, txg);
6185 spa_update_dspace(spa);
6188 * It had better be the case that we didn't dirty anything
6189 * since vdev_config_sync().
6191 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
6192 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
6193 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
6195 spa->spa_sync_pass = 0;
6197 spa_config_exit(spa, SCL_CONFIG, FTAG);
6199 spa_handle_ignored_writes(spa);
6202 * If any async tasks have been requested, kick them off.
6204 spa_async_dispatch(spa);
6208 * Sync all pools. We don't want to hold the namespace lock across these
6209 * operations, so we take a reference on the spa_t and drop the lock during the
6213 spa_sync_allpools(void)
6216 mutex_enter(&spa_namespace_lock);
6217 while ((spa = spa_next(spa)) != NULL) {
6218 if (spa_state(spa) != POOL_STATE_ACTIVE ||
6219 !spa_writeable(spa) || spa_suspended(spa))
6221 spa_open_ref(spa, FTAG);
6222 mutex_exit(&spa_namespace_lock);
6223 txg_wait_synced(spa_get_dsl(spa), 0);
6224 mutex_enter(&spa_namespace_lock);
6225 spa_close(spa, FTAG);
6227 mutex_exit(&spa_namespace_lock);
6231 * ==========================================================================
6232 * Miscellaneous routines
6233 * ==========================================================================
6237 * Remove all pools in the system.
6245 * Remove all cached state. All pools should be closed now,
6246 * so every spa in the AVL tree should be unreferenced.
6248 mutex_enter(&spa_namespace_lock);
6249 while ((spa = spa_next(NULL)) != NULL) {
6251 * Stop async tasks. The async thread may need to detach
6252 * a device that's been replaced, which requires grabbing
6253 * spa_namespace_lock, so we must drop it here.
6255 spa_open_ref(spa, FTAG);
6256 mutex_exit(&spa_namespace_lock);
6257 spa_async_suspend(spa);
6258 mutex_enter(&spa_namespace_lock);
6259 spa_close(spa, FTAG);
6261 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
6263 spa_deactivate(spa);
6267 mutex_exit(&spa_namespace_lock);
6271 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
6276 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
6280 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
6281 vd = spa->spa_l2cache.sav_vdevs[i];
6282 if (vd->vdev_guid == guid)
6286 for (i = 0; i < spa->spa_spares.sav_count; i++) {
6287 vd = spa->spa_spares.sav_vdevs[i];
6288 if (vd->vdev_guid == guid)
6297 spa_upgrade(spa_t *spa, uint64_t version)
6299 ASSERT(spa_writeable(spa));
6301 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6304 * This should only be called for a non-faulted pool, and since a
6305 * future version would result in an unopenable pool, this shouldn't be
6308 ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version));
6309 ASSERT(version >= spa->spa_uberblock.ub_version);
6311 spa->spa_uberblock.ub_version = version;
6312 vdev_config_dirty(spa->spa_root_vdev);
6314 spa_config_exit(spa, SCL_ALL, FTAG);
6316 txg_wait_synced(spa_get_dsl(spa), 0);
6320 spa_has_spare(spa_t *spa, uint64_t guid)
6324 spa_aux_vdev_t *sav = &spa->spa_spares;
6326 for (i = 0; i < sav->sav_count; i++)
6327 if (sav->sav_vdevs[i]->vdev_guid == guid)
6330 for (i = 0; i < sav->sav_npending; i++) {
6331 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
6332 &spareguid) == 0 && spareguid == guid)
6340 * Check if a pool has an active shared spare device.
6341 * Note: reference count of an active spare is 2, as a spare and as a replace
6344 spa_has_active_shared_spare(spa_t *spa)
6348 spa_aux_vdev_t *sav = &spa->spa_spares;
6350 for (i = 0; i < sav->sav_count; i++) {
6351 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
6352 &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
6361 * Post a FM_EREPORT_ZFS_* event from sys/fm/fs/zfs.h. The payload will be
6362 * filled in from the spa and (optionally) the vdev. This doesn't do anything
6363 * in the userland libzpool, as we don't want consumers to misinterpret ztest
6364 * or zdb as real changes.
6367 spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
6370 zfs_ereport_post(name, spa, vd, NULL, 0, 0);
6374 #if defined(_KERNEL) && defined(HAVE_SPL)
6375 /* state manipulation functions */
6376 EXPORT_SYMBOL(spa_open);
6377 EXPORT_SYMBOL(spa_open_rewind);
6378 EXPORT_SYMBOL(spa_get_stats);
6379 EXPORT_SYMBOL(spa_create);
6380 EXPORT_SYMBOL(spa_import_rootpool);
6381 EXPORT_SYMBOL(spa_import);
6382 EXPORT_SYMBOL(spa_tryimport);
6383 EXPORT_SYMBOL(spa_destroy);
6384 EXPORT_SYMBOL(spa_export);
6385 EXPORT_SYMBOL(spa_reset);
6386 EXPORT_SYMBOL(spa_async_request);
6387 EXPORT_SYMBOL(spa_async_suspend);
6388 EXPORT_SYMBOL(spa_async_resume);
6389 EXPORT_SYMBOL(spa_inject_addref);
6390 EXPORT_SYMBOL(spa_inject_delref);
6391 EXPORT_SYMBOL(spa_scan_stat_init);
6392 EXPORT_SYMBOL(spa_scan_get_stats);
6394 /* device maniion */
6395 EXPORT_SYMBOL(spa_vdev_add);
6396 EXPORT_SYMBOL(spa_vdev_attach);
6397 EXPORT_SYMBOL(spa_vdev_detach);
6398 EXPORT_SYMBOL(spa_vdev_remove);
6399 EXPORT_SYMBOL(spa_vdev_setpath);
6400 EXPORT_SYMBOL(spa_vdev_setfru);
6401 EXPORT_SYMBOL(spa_vdev_split_mirror);
6403 /* spare statech is global across all pools) */
6404 EXPORT_SYMBOL(spa_spare_add);
6405 EXPORT_SYMBOL(spa_spare_remove);
6406 EXPORT_SYMBOL(spa_spare_exists);
6407 EXPORT_SYMBOL(spa_spare_activate);
6409 /* L2ARC statech is global across all pools) */
6410 EXPORT_SYMBOL(spa_l2cache_add);
6411 EXPORT_SYMBOL(spa_l2cache_remove);
6412 EXPORT_SYMBOL(spa_l2cache_exists);
6413 EXPORT_SYMBOL(spa_l2cache_activate);
6414 EXPORT_SYMBOL(spa_l2cache_drop);
6417 EXPORT_SYMBOL(spa_scan);
6418 EXPORT_SYMBOL(spa_scan_stop);
6421 EXPORT_SYMBOL(spa_sync); /* only for DMU use */
6422 EXPORT_SYMBOL(spa_sync_allpools);
6425 EXPORT_SYMBOL(spa_prop_set);
6426 EXPORT_SYMBOL(spa_prop_get);
6427 EXPORT_SYMBOL(spa_prop_clear_bootfs);
6429 /* asynchronous event notification */
6430 EXPORT_SYMBOL(spa_event_notify);