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 (c) 2012 by Delphix. All rights reserved.
28 * This file contains all the routines used when modifying on-disk SPA state.
29 * This includes opening, importing, destroying, exporting a pool, and syncing a
33 #include <sys/zfs_context.h>
34 #include <sys/fm/fs/zfs.h>
35 #include <sys/spa_impl.h>
37 #include <sys/zio_checksum.h>
39 #include <sys/dmu_tx.h>
43 #include <sys/vdev_impl.h>
44 #include <sys/metaslab.h>
45 #include <sys/metaslab_impl.h>
46 #include <sys/uberblock_impl.h>
49 #include <sys/dmu_traverse.h>
50 #include <sys/dmu_objset.h>
51 #include <sys/unique.h>
52 #include <sys/dsl_pool.h>
53 #include <sys/dsl_dataset.h>
54 #include <sys/dsl_dir.h>
55 #include <sys/dsl_prop.h>
56 #include <sys/dsl_synctask.h>
57 #include <sys/fs/zfs.h>
59 #include <sys/callb.h>
60 #include <sys/spa_boot.h>
61 #include <sys/zfs_ioctl.h>
62 #include <sys/dsl_scan.h>
63 #include <sys/zfeature.h>
67 #include <sys/callb.h>
68 #include <sys/cpupart.h>
73 #include "zfs_comutil.h"
75 /* Check hostid on import? */
76 static int check_hostid = 1;
78 SYSCTL_DECL(_vfs_zfs);
79 TUNABLE_INT("vfs.zfs.check_hostid", &check_hostid);
80 SYSCTL_INT(_vfs_zfs, OID_AUTO, check_hostid, CTLFLAG_RW, &check_hostid, 0,
81 "Check hostid on import?");
83 typedef enum zti_modes {
84 zti_mode_fixed, /* value is # of threads (min 1) */
85 zti_mode_online_percent, /* value is % of online CPUs */
86 zti_mode_batch, /* cpu-intensive; value is ignored */
87 zti_mode_null, /* don't create a taskq */
91 #define ZTI_FIX(n) { zti_mode_fixed, (n) }
92 #define ZTI_PCT(n) { zti_mode_online_percent, (n) }
93 #define ZTI_BATCH { zti_mode_batch, 0 }
94 #define ZTI_NULL { zti_mode_null, 0 }
96 #define ZTI_ONE ZTI_FIX(1)
98 typedef struct zio_taskq_info {
99 enum zti_modes zti_mode;
103 static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
104 "issue", "issue_high", "intr", "intr_high"
108 * Define the taskq threads for the following I/O types:
109 * NULL, READ, WRITE, FREE, CLAIM, and IOCTL
111 const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
112 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
113 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL },
114 { ZTI_FIX(8), ZTI_NULL, ZTI_BATCH, ZTI_NULL },
115 { ZTI_BATCH, ZTI_FIX(5), ZTI_FIX(8), ZTI_FIX(5) },
116 { ZTI_FIX(100), ZTI_NULL, ZTI_ONE, ZTI_NULL },
117 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL },
118 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL },
121 static dsl_syncfunc_t spa_sync_version;
122 static dsl_syncfunc_t spa_sync_props;
123 static dsl_checkfunc_t spa_change_guid_check;
124 static dsl_syncfunc_t spa_change_guid_sync;
125 static boolean_t spa_has_active_shared_spare(spa_t *spa);
126 static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
127 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
129 static void spa_vdev_resilver_done(spa_t *spa);
131 uint_t zio_taskq_batch_pct = 100; /* 1 thread per cpu in pset */
133 id_t zio_taskq_psrset_bind = PS_NONE;
136 boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */
138 uint_t zio_taskq_basedc = 80; /* base duty cycle */
140 boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */
141 extern int zfs_sync_pass_deferred_free;
144 * This (illegal) pool name is used when temporarily importing a spa_t in order
145 * to get the vdev stats associated with the imported devices.
147 #define TRYIMPORT_NAME "$import"
150 * ==========================================================================
151 * SPA properties routines
152 * ==========================================================================
156 * Add a (source=src, propname=propval) list to an nvlist.
159 spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
160 uint64_t intval, zprop_source_t src)
162 const char *propname = zpool_prop_to_name(prop);
165 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
166 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
169 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
171 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
173 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
174 nvlist_free(propval);
178 * Get property values from the spa configuration.
181 spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
183 vdev_t *rvd = spa->spa_root_vdev;
184 dsl_pool_t *pool = spa->spa_dsl_pool;
188 uint64_t cap, version;
189 zprop_source_t src = ZPROP_SRC_NONE;
190 spa_config_dirent_t *dp;
192 ASSERT(MUTEX_HELD(&spa->spa_props_lock));
195 alloc = metaslab_class_get_alloc(spa_normal_class(spa));
196 size = metaslab_class_get_space(spa_normal_class(spa));
197 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
198 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
199 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
200 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
204 for (int c = 0; c < rvd->vdev_children; c++) {
205 vdev_t *tvd = rvd->vdev_child[c];
206 space += tvd->vdev_max_asize - tvd->vdev_asize;
208 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, space,
211 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
212 (spa_mode(spa) == FREAD), src);
214 cap = (size == 0) ? 0 : (alloc * 100 / size);
215 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
217 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
218 ddt_get_pool_dedup_ratio(spa), src);
220 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
221 rvd->vdev_state, src);
223 version = spa_version(spa);
224 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
225 src = ZPROP_SRC_DEFAULT;
227 src = ZPROP_SRC_LOCAL;
228 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
232 dsl_dir_t *freedir = pool->dp_free_dir;
235 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
236 * when opening pools before this version freedir will be NULL.
238 if (freedir != NULL) {
239 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
240 freedir->dd_phys->dd_used_bytes, src);
242 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
247 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
249 if (spa->spa_comment != NULL) {
250 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
254 if (spa->spa_root != NULL)
255 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
258 if ((dp = list_head(&spa->spa_config_list)) != NULL) {
259 if (dp->scd_path == NULL) {
260 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
261 "none", 0, ZPROP_SRC_LOCAL);
262 } else if (strcmp(dp->scd_path, spa_config_path) != 0) {
263 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
264 dp->scd_path, 0, ZPROP_SRC_LOCAL);
270 * Get zpool property values.
273 spa_prop_get(spa_t *spa, nvlist_t **nvp)
275 objset_t *mos = spa->spa_meta_objset;
280 VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
282 mutex_enter(&spa->spa_props_lock);
285 * Get properties from the spa config.
287 spa_prop_get_config(spa, nvp);
289 /* If no pool property object, no more prop to get. */
290 if (mos == NULL || spa->spa_pool_props_object == 0) {
291 mutex_exit(&spa->spa_props_lock);
296 * Get properties from the MOS pool property object.
298 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
299 (err = zap_cursor_retrieve(&zc, &za)) == 0;
300 zap_cursor_advance(&zc)) {
303 zprop_source_t src = ZPROP_SRC_DEFAULT;
306 if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
309 switch (za.za_integer_length) {
311 /* integer property */
312 if (za.za_first_integer !=
313 zpool_prop_default_numeric(prop))
314 src = ZPROP_SRC_LOCAL;
316 if (prop == ZPOOL_PROP_BOOTFS) {
318 dsl_dataset_t *ds = NULL;
320 dp = spa_get_dsl(spa);
321 rw_enter(&dp->dp_config_rwlock, RW_READER);
322 if (err = dsl_dataset_hold_obj(dp,
323 za.za_first_integer, FTAG, &ds)) {
324 rw_exit(&dp->dp_config_rwlock);
329 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
331 dsl_dataset_name(ds, strval);
332 dsl_dataset_rele(ds, FTAG);
333 rw_exit(&dp->dp_config_rwlock);
336 intval = za.za_first_integer;
339 spa_prop_add_list(*nvp, prop, strval, intval, src);
343 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
348 /* string property */
349 strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
350 err = zap_lookup(mos, spa->spa_pool_props_object,
351 za.za_name, 1, za.za_num_integers, strval);
353 kmem_free(strval, za.za_num_integers);
356 spa_prop_add_list(*nvp, prop, strval, 0, src);
357 kmem_free(strval, za.za_num_integers);
364 zap_cursor_fini(&zc);
365 mutex_exit(&spa->spa_props_lock);
367 if (err && err != ENOENT) {
377 * Validate the given pool properties nvlist and modify the list
378 * for the property values to be set.
381 spa_prop_validate(spa_t *spa, nvlist_t *props)
384 int error = 0, reset_bootfs = 0;
386 boolean_t has_feature = B_FALSE;
389 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
391 char *strval, *slash, *check, *fname;
392 const char *propname = nvpair_name(elem);
393 zpool_prop_t prop = zpool_name_to_prop(propname);
397 if (!zpool_prop_feature(propname)) {
403 * Sanitize the input.
405 if (nvpair_type(elem) != DATA_TYPE_UINT64) {
410 if (nvpair_value_uint64(elem, &intval) != 0) {
420 fname = strchr(propname, '@') + 1;
421 if (zfeature_lookup_name(fname, NULL) != 0) {
426 has_feature = B_TRUE;
429 case ZPOOL_PROP_VERSION:
430 error = nvpair_value_uint64(elem, &intval);
432 (intval < spa_version(spa) ||
433 intval > SPA_VERSION_BEFORE_FEATURES ||
438 case ZPOOL_PROP_DELEGATION:
439 case ZPOOL_PROP_AUTOREPLACE:
440 case ZPOOL_PROP_LISTSNAPS:
441 case ZPOOL_PROP_AUTOEXPAND:
442 error = nvpair_value_uint64(elem, &intval);
443 if (!error && intval > 1)
447 case ZPOOL_PROP_BOOTFS:
449 * If the pool version is less than SPA_VERSION_BOOTFS,
450 * or the pool is still being created (version == 0),
451 * the bootfs property cannot be set.
453 if (spa_version(spa) < SPA_VERSION_BOOTFS) {
459 * Make sure the vdev config is bootable
461 if (!vdev_is_bootable(spa->spa_root_vdev)) {
468 error = nvpair_value_string(elem, &strval);
474 if (strval == NULL || strval[0] == '\0') {
475 objnum = zpool_prop_default_numeric(
480 if (error = dmu_objset_hold(strval, FTAG, &os))
483 /* Must be ZPL and not gzip compressed. */
485 if (dmu_objset_type(os) != DMU_OST_ZFS) {
487 } else if ((error = dsl_prop_get_integer(strval,
488 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
489 &compress, NULL)) == 0 &&
490 !BOOTFS_COMPRESS_VALID(compress)) {
493 objnum = dmu_objset_id(os);
495 dmu_objset_rele(os, FTAG);
499 case ZPOOL_PROP_FAILUREMODE:
500 error = nvpair_value_uint64(elem, &intval);
501 if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
502 intval > ZIO_FAILURE_MODE_PANIC))
506 * This is a special case which only occurs when
507 * the pool has completely failed. This allows
508 * the user to change the in-core failmode property
509 * without syncing it out to disk (I/Os might
510 * currently be blocked). We do this by returning
511 * EIO to the caller (spa_prop_set) to trick it
512 * into thinking we encountered a property validation
515 if (!error && spa_suspended(spa)) {
516 spa->spa_failmode = intval;
521 case ZPOOL_PROP_CACHEFILE:
522 if ((error = nvpair_value_string(elem, &strval)) != 0)
525 if (strval[0] == '\0')
528 if (strcmp(strval, "none") == 0)
531 if (strval[0] != '/') {
536 slash = strrchr(strval, '/');
537 ASSERT(slash != NULL);
539 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
540 strcmp(slash, "/..") == 0)
544 case ZPOOL_PROP_COMMENT:
545 if ((error = nvpair_value_string(elem, &strval)) != 0)
547 for (check = strval; *check != '\0'; check++) {
549 * The kernel doesn't have an easy isprint()
550 * check. For this kernel check, we merely
551 * check ASCII apart from DEL. Fix this if
552 * there is an easy-to-use kernel isprint().
554 if (*check >= 0x7f) {
560 if (strlen(strval) > ZPROP_MAX_COMMENT)
564 case ZPOOL_PROP_DEDUPDITTO:
565 if (spa_version(spa) < SPA_VERSION_DEDUP)
568 error = nvpair_value_uint64(elem, &intval);
570 intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
579 if (!error && reset_bootfs) {
580 error = nvlist_remove(props,
581 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
584 error = nvlist_add_uint64(props,
585 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
593 spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
596 spa_config_dirent_t *dp;
598 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
602 dp = kmem_alloc(sizeof (spa_config_dirent_t),
605 if (cachefile[0] == '\0')
606 dp->scd_path = spa_strdup(spa_config_path);
607 else if (strcmp(cachefile, "none") == 0)
610 dp->scd_path = spa_strdup(cachefile);
612 list_insert_head(&spa->spa_config_list, dp);
614 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
618 spa_prop_set(spa_t *spa, nvlist_t *nvp)
621 nvpair_t *elem = NULL;
622 boolean_t need_sync = B_FALSE;
624 if ((error = spa_prop_validate(spa, nvp)) != 0)
627 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
628 zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem));
630 if (prop == ZPOOL_PROP_CACHEFILE ||
631 prop == ZPOOL_PROP_ALTROOT ||
632 prop == ZPOOL_PROP_READONLY)
635 if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) {
638 if (prop == ZPOOL_PROP_VERSION) {
639 VERIFY(nvpair_value_uint64(elem, &ver) == 0);
641 ASSERT(zpool_prop_feature(nvpair_name(elem)));
642 ver = SPA_VERSION_FEATURES;
646 /* Save time if the version is already set. */
647 if (ver == spa_version(spa))
651 * In addition to the pool directory object, we might
652 * create the pool properties object, the features for
653 * read object, the features for write object, or the
654 * feature descriptions object.
656 error = dsl_sync_task_do(spa_get_dsl(spa), NULL,
657 spa_sync_version, spa, &ver, 6);
668 return (dsl_sync_task_do(spa_get_dsl(spa), NULL, spa_sync_props,
676 * If the bootfs property value is dsobj, clear it.
679 spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
681 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
682 VERIFY(zap_remove(spa->spa_meta_objset,
683 spa->spa_pool_props_object,
684 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
691 spa_change_guid_check(void *arg1, void *arg2, dmu_tx_t *tx)
694 uint64_t *newguid = arg2;
695 vdev_t *rvd = spa->spa_root_vdev;
698 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
699 vdev_state = rvd->vdev_state;
700 spa_config_exit(spa, SCL_STATE, FTAG);
702 if (vdev_state != VDEV_STATE_HEALTHY)
705 ASSERT3U(spa_guid(spa), !=, *newguid);
711 spa_change_guid_sync(void *arg1, void *arg2, dmu_tx_t *tx)
714 uint64_t *newguid = arg2;
716 vdev_t *rvd = spa->spa_root_vdev;
718 oldguid = spa_guid(spa);
720 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
721 rvd->vdev_guid = *newguid;
722 rvd->vdev_guid_sum += (*newguid - oldguid);
723 vdev_config_dirty(rvd);
724 spa_config_exit(spa, SCL_STATE, FTAG);
728 * TODO: until recent illumos logging changes are merged
729 * log reguid as pool property change
731 spa_history_log_internal(LOG_POOL_PROPSET, spa, tx,
732 "guid change old=%llu new=%llu", oldguid, *newguid);
734 spa_history_log_internal(spa, "guid change", tx, "old=%lld new=%lld",
740 * Change the GUID for the pool. This is done so that we can later
741 * re-import a pool built from a clone of our own vdevs. We will modify
742 * the root vdev's guid, our own pool guid, and then mark all of our
743 * vdevs dirty. Note that we must make sure that all our vdevs are
744 * online when we do this, or else any vdevs that weren't present
745 * would be orphaned from our pool. We are also going to issue a
746 * sysevent to update any watchers.
749 spa_change_guid(spa_t *spa)
754 mutex_enter(&spa_namespace_lock);
755 guid = spa_generate_guid(NULL);
757 error = dsl_sync_task_do(spa_get_dsl(spa), spa_change_guid_check,
758 spa_change_guid_sync, spa, &guid, 5);
761 spa_config_sync(spa, B_FALSE, B_TRUE);
762 spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID);
765 mutex_exit(&spa_namespace_lock);
771 * ==========================================================================
772 * SPA state manipulation (open/create/destroy/import/export)
773 * ==========================================================================
777 spa_error_entry_compare(const void *a, const void *b)
779 spa_error_entry_t *sa = (spa_error_entry_t *)a;
780 spa_error_entry_t *sb = (spa_error_entry_t *)b;
783 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
784 sizeof (zbookmark_t));
795 * Utility function which retrieves copies of the current logs and
796 * re-initializes them in the process.
799 spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
801 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
803 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
804 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
806 avl_create(&spa->spa_errlist_scrub,
807 spa_error_entry_compare, sizeof (spa_error_entry_t),
808 offsetof(spa_error_entry_t, se_avl));
809 avl_create(&spa->spa_errlist_last,
810 spa_error_entry_compare, sizeof (spa_error_entry_t),
811 offsetof(spa_error_entry_t, se_avl));
815 spa_taskq_create(spa_t *spa, const char *name, enum zti_modes mode,
818 uint_t flags = TASKQ_PREPOPULATE;
819 boolean_t batch = B_FALSE;
823 return (NULL); /* no taskq needed */
826 ASSERT3U(value, >=, 1);
827 value = MAX(value, 1);
832 flags |= TASKQ_THREADS_CPU_PCT;
833 value = zio_taskq_batch_pct;
836 case zti_mode_online_percent:
837 flags |= TASKQ_THREADS_CPU_PCT;
841 panic("unrecognized mode for %s taskq (%u:%u) in "
848 if (zio_taskq_sysdc && spa->spa_proc != &p0) {
850 flags |= TASKQ_DC_BATCH;
852 return (taskq_create_sysdc(name, value, 50, INT_MAX,
853 spa->spa_proc, zio_taskq_basedc, flags));
856 return (taskq_create_proc(name, value, maxclsyspri, 50, INT_MAX,
857 spa->spa_proc, flags));
861 spa_create_zio_taskqs(spa_t *spa)
863 for (int t = 0; t < ZIO_TYPES; t++) {
864 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
865 const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
866 enum zti_modes mode = ztip->zti_mode;
867 uint_t value = ztip->zti_value;
870 (void) snprintf(name, sizeof (name),
871 "%s_%s", zio_type_name[t], zio_taskq_types[q]);
873 spa->spa_zio_taskq[t][q] =
874 spa_taskq_create(spa, name, mode, value);
882 spa_thread(void *arg)
887 user_t *pu = PTOU(curproc);
889 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
892 ASSERT(curproc != &p0);
893 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
894 "zpool-%s", spa->spa_name);
895 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
898 /* bind this thread to the requested psrset */
899 if (zio_taskq_psrset_bind != PS_NONE) {
901 mutex_enter(&cpu_lock);
902 mutex_enter(&pidlock);
903 mutex_enter(&curproc->p_lock);
905 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
906 0, NULL, NULL) == 0) {
907 curthread->t_bind_pset = zio_taskq_psrset_bind;
910 "Couldn't bind process for zfs pool \"%s\" to "
911 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
914 mutex_exit(&curproc->p_lock);
915 mutex_exit(&pidlock);
916 mutex_exit(&cpu_lock);
922 if (zio_taskq_sysdc) {
923 sysdc_thread_enter(curthread, 100, 0);
927 spa->spa_proc = curproc;
928 spa->spa_did = curthread->t_did;
930 spa_create_zio_taskqs(spa);
932 mutex_enter(&spa->spa_proc_lock);
933 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
935 spa->spa_proc_state = SPA_PROC_ACTIVE;
936 cv_broadcast(&spa->spa_proc_cv);
938 CALLB_CPR_SAFE_BEGIN(&cprinfo);
939 while (spa->spa_proc_state == SPA_PROC_ACTIVE)
940 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
941 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
943 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
944 spa->spa_proc_state = SPA_PROC_GONE;
946 cv_broadcast(&spa->spa_proc_cv);
947 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */
949 mutex_enter(&curproc->p_lock);
952 #endif /* SPA_PROCESS */
956 * Activate an uninitialized pool.
959 spa_activate(spa_t *spa, int mode)
961 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
963 spa->spa_state = POOL_STATE_ACTIVE;
964 spa->spa_mode = mode;
966 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
967 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
969 /* Try to create a covering process */
970 mutex_enter(&spa->spa_proc_lock);
971 ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
972 ASSERT(spa->spa_proc == &p0);
976 /* Only create a process if we're going to be around a while. */
977 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
978 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
980 spa->spa_proc_state = SPA_PROC_CREATED;
981 while (spa->spa_proc_state == SPA_PROC_CREATED) {
982 cv_wait(&spa->spa_proc_cv,
983 &spa->spa_proc_lock);
985 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
986 ASSERT(spa->spa_proc != &p0);
987 ASSERT(spa->spa_did != 0);
991 "Couldn't create process for zfs pool \"%s\"\n",
996 #endif /* SPA_PROCESS */
997 mutex_exit(&spa->spa_proc_lock);
999 /* If we didn't create a process, we need to create our taskqs. */
1000 ASSERT(spa->spa_proc == &p0);
1001 if (spa->spa_proc == &p0) {
1002 spa_create_zio_taskqs(spa);
1005 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
1006 offsetof(vdev_t, vdev_config_dirty_node));
1007 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
1008 offsetof(vdev_t, vdev_state_dirty_node));
1010 txg_list_create(&spa->spa_vdev_txg_list,
1011 offsetof(struct vdev, vdev_txg_node));
1013 avl_create(&spa->spa_errlist_scrub,
1014 spa_error_entry_compare, sizeof (spa_error_entry_t),
1015 offsetof(spa_error_entry_t, se_avl));
1016 avl_create(&spa->spa_errlist_last,
1017 spa_error_entry_compare, sizeof (spa_error_entry_t),
1018 offsetof(spa_error_entry_t, se_avl));
1022 * Opposite of spa_activate().
1025 spa_deactivate(spa_t *spa)
1027 ASSERT(spa->spa_sync_on == B_FALSE);
1028 ASSERT(spa->spa_dsl_pool == NULL);
1029 ASSERT(spa->spa_root_vdev == NULL);
1030 ASSERT(spa->spa_async_zio_root == NULL);
1031 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
1033 txg_list_destroy(&spa->spa_vdev_txg_list);
1035 list_destroy(&spa->spa_config_dirty_list);
1036 list_destroy(&spa->spa_state_dirty_list);
1038 for (int t = 0; t < ZIO_TYPES; t++) {
1039 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1040 if (spa->spa_zio_taskq[t][q] != NULL)
1041 taskq_destroy(spa->spa_zio_taskq[t][q]);
1042 spa->spa_zio_taskq[t][q] = NULL;
1046 metaslab_class_destroy(spa->spa_normal_class);
1047 spa->spa_normal_class = NULL;
1049 metaslab_class_destroy(spa->spa_log_class);
1050 spa->spa_log_class = NULL;
1053 * If this was part of an import or the open otherwise failed, we may
1054 * still have errors left in the queues. Empty them just in case.
1056 spa_errlog_drain(spa);
1058 avl_destroy(&spa->spa_errlist_scrub);
1059 avl_destroy(&spa->spa_errlist_last);
1061 spa->spa_state = POOL_STATE_UNINITIALIZED;
1063 mutex_enter(&spa->spa_proc_lock);
1064 if (spa->spa_proc_state != SPA_PROC_NONE) {
1065 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1066 spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1067 cv_broadcast(&spa->spa_proc_cv);
1068 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1069 ASSERT(spa->spa_proc != &p0);
1070 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1072 ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1073 spa->spa_proc_state = SPA_PROC_NONE;
1075 ASSERT(spa->spa_proc == &p0);
1076 mutex_exit(&spa->spa_proc_lock);
1080 * We want to make sure spa_thread() has actually exited the ZFS
1081 * module, so that the module can't be unloaded out from underneath
1084 if (spa->spa_did != 0) {
1085 thread_join(spa->spa_did);
1088 #endif /* SPA_PROCESS */
1092 * Verify a pool configuration, and construct the vdev tree appropriately. This
1093 * will create all the necessary vdevs in the appropriate layout, with each vdev
1094 * in the CLOSED state. This will prep the pool before open/creation/import.
1095 * All vdev validation is done by the vdev_alloc() routine.
1098 spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1099 uint_t id, int atype)
1105 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1108 if ((*vdp)->vdev_ops->vdev_op_leaf)
1111 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1114 if (error == ENOENT)
1123 for (int c = 0; c < children; c++) {
1125 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1133 ASSERT(*vdp != NULL);
1139 * Opposite of spa_load().
1142 spa_unload(spa_t *spa)
1146 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1151 spa_async_suspend(spa);
1156 if (spa->spa_sync_on) {
1157 txg_sync_stop(spa->spa_dsl_pool);
1158 spa->spa_sync_on = B_FALSE;
1162 * Wait for any outstanding async I/O to complete.
1164 if (spa->spa_async_zio_root != NULL) {
1165 (void) zio_wait(spa->spa_async_zio_root);
1166 spa->spa_async_zio_root = NULL;
1169 bpobj_close(&spa->spa_deferred_bpobj);
1172 * Close the dsl pool.
1174 if (spa->spa_dsl_pool) {
1175 dsl_pool_close(spa->spa_dsl_pool);
1176 spa->spa_dsl_pool = NULL;
1177 spa->spa_meta_objset = NULL;
1182 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1185 * Drop and purge level 2 cache
1187 spa_l2cache_drop(spa);
1192 if (spa->spa_root_vdev)
1193 vdev_free(spa->spa_root_vdev);
1194 ASSERT(spa->spa_root_vdev == NULL);
1196 for (i = 0; i < spa->spa_spares.sav_count; i++)
1197 vdev_free(spa->spa_spares.sav_vdevs[i]);
1198 if (spa->spa_spares.sav_vdevs) {
1199 kmem_free(spa->spa_spares.sav_vdevs,
1200 spa->spa_spares.sav_count * sizeof (void *));
1201 spa->spa_spares.sav_vdevs = NULL;
1203 if (spa->spa_spares.sav_config) {
1204 nvlist_free(spa->spa_spares.sav_config);
1205 spa->spa_spares.sav_config = NULL;
1207 spa->spa_spares.sav_count = 0;
1209 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1210 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1211 vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1213 if (spa->spa_l2cache.sav_vdevs) {
1214 kmem_free(spa->spa_l2cache.sav_vdevs,
1215 spa->spa_l2cache.sav_count * sizeof (void *));
1216 spa->spa_l2cache.sav_vdevs = NULL;
1218 if (spa->spa_l2cache.sav_config) {
1219 nvlist_free(spa->spa_l2cache.sav_config);
1220 spa->spa_l2cache.sav_config = NULL;
1222 spa->spa_l2cache.sav_count = 0;
1224 spa->spa_async_suspended = 0;
1226 if (spa->spa_comment != NULL) {
1227 spa_strfree(spa->spa_comment);
1228 spa->spa_comment = NULL;
1231 spa_config_exit(spa, SCL_ALL, FTAG);
1235 * Load (or re-load) the current list of vdevs describing the active spares for
1236 * this pool. When this is called, we have some form of basic information in
1237 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1238 * then re-generate a more complete list including status information.
1241 spa_load_spares(spa_t *spa)
1248 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1251 * First, close and free any existing spare vdevs.
1253 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1254 vd = spa->spa_spares.sav_vdevs[i];
1256 /* Undo the call to spa_activate() below */
1257 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1258 B_FALSE)) != NULL && tvd->vdev_isspare)
1259 spa_spare_remove(tvd);
1264 if (spa->spa_spares.sav_vdevs)
1265 kmem_free(spa->spa_spares.sav_vdevs,
1266 spa->spa_spares.sav_count * sizeof (void *));
1268 if (spa->spa_spares.sav_config == NULL)
1271 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1272 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1274 spa->spa_spares.sav_count = (int)nspares;
1275 spa->spa_spares.sav_vdevs = NULL;
1281 * Construct the array of vdevs, opening them to get status in the
1282 * process. For each spare, there is potentially two different vdev_t
1283 * structures associated with it: one in the list of spares (used only
1284 * for basic validation purposes) and one in the active vdev
1285 * configuration (if it's spared in). During this phase we open and
1286 * validate each vdev on the spare list. If the vdev also exists in the
1287 * active configuration, then we also mark this vdev as an active spare.
1289 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1291 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1292 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1293 VDEV_ALLOC_SPARE) == 0);
1296 spa->spa_spares.sav_vdevs[i] = vd;
1298 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1299 B_FALSE)) != NULL) {
1300 if (!tvd->vdev_isspare)
1304 * We only mark the spare active if we were successfully
1305 * able to load the vdev. Otherwise, importing a pool
1306 * with a bad active spare would result in strange
1307 * behavior, because multiple pool would think the spare
1308 * is actively in use.
1310 * There is a vulnerability here to an equally bizarre
1311 * circumstance, where a dead active spare is later
1312 * brought back to life (onlined or otherwise). Given
1313 * the rarity of this scenario, and the extra complexity
1314 * it adds, we ignore the possibility.
1316 if (!vdev_is_dead(tvd))
1317 spa_spare_activate(tvd);
1321 vd->vdev_aux = &spa->spa_spares;
1323 if (vdev_open(vd) != 0)
1326 if (vdev_validate_aux(vd) == 0)
1331 * Recompute the stashed list of spares, with status information
1334 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1335 DATA_TYPE_NVLIST_ARRAY) == 0);
1337 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1339 for (i = 0; i < spa->spa_spares.sav_count; i++)
1340 spares[i] = vdev_config_generate(spa,
1341 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1342 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1343 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1344 for (i = 0; i < spa->spa_spares.sav_count; i++)
1345 nvlist_free(spares[i]);
1346 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1350 * Load (or re-load) the current list of vdevs describing the active l2cache for
1351 * this pool. When this is called, we have some form of basic information in
1352 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1353 * then re-generate a more complete list including status information.
1354 * Devices which are already active have their details maintained, and are
1358 spa_load_l2cache(spa_t *spa)
1362 int i, j, oldnvdevs;
1364 vdev_t *vd, **oldvdevs, **newvdevs;
1365 spa_aux_vdev_t *sav = &spa->spa_l2cache;
1367 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1369 if (sav->sav_config != NULL) {
1370 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1371 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1372 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1377 oldvdevs = sav->sav_vdevs;
1378 oldnvdevs = sav->sav_count;
1379 sav->sav_vdevs = NULL;
1383 * Process new nvlist of vdevs.
1385 for (i = 0; i < nl2cache; i++) {
1386 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1390 for (j = 0; j < oldnvdevs; j++) {
1392 if (vd != NULL && guid == vd->vdev_guid) {
1394 * Retain previous vdev for add/remove ops.
1402 if (newvdevs[i] == NULL) {
1406 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1407 VDEV_ALLOC_L2CACHE) == 0);
1412 * Commit this vdev as an l2cache device,
1413 * even if it fails to open.
1415 spa_l2cache_add(vd);
1420 spa_l2cache_activate(vd);
1422 if (vdev_open(vd) != 0)
1425 (void) vdev_validate_aux(vd);
1427 if (!vdev_is_dead(vd))
1428 l2arc_add_vdev(spa, vd);
1433 * Purge vdevs that were dropped
1435 for (i = 0; i < oldnvdevs; i++) {
1440 ASSERT(vd->vdev_isl2cache);
1442 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1443 pool != 0ULL && l2arc_vdev_present(vd))
1444 l2arc_remove_vdev(vd);
1445 vdev_clear_stats(vd);
1451 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1453 if (sav->sav_config == NULL)
1456 sav->sav_vdevs = newvdevs;
1457 sav->sav_count = (int)nl2cache;
1460 * Recompute the stashed list of l2cache devices, with status
1461 * information this time.
1463 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1464 DATA_TYPE_NVLIST_ARRAY) == 0);
1466 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1467 for (i = 0; i < sav->sav_count; i++)
1468 l2cache[i] = vdev_config_generate(spa,
1469 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1470 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1471 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1473 for (i = 0; i < sav->sav_count; i++)
1474 nvlist_free(l2cache[i]);
1476 kmem_free(l2cache, sav->sav_count * sizeof (void *));
1480 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1483 char *packed = NULL;
1488 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
1489 nvsize = *(uint64_t *)db->db_data;
1490 dmu_buf_rele(db, FTAG);
1492 packed = kmem_alloc(nvsize, KM_SLEEP);
1493 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1496 error = nvlist_unpack(packed, nvsize, value, 0);
1497 kmem_free(packed, nvsize);
1503 * Checks to see if the given vdev could not be opened, in which case we post a
1504 * sysevent to notify the autoreplace code that the device has been removed.
1507 spa_check_removed(vdev_t *vd)
1509 for (int c = 0; c < vd->vdev_children; c++)
1510 spa_check_removed(vd->vdev_child[c]);
1512 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd)) {
1513 zfs_post_autoreplace(vd->vdev_spa, vd);
1514 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1519 * Validate the current config against the MOS config
1522 spa_config_valid(spa_t *spa, nvlist_t *config)
1524 vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1527 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1529 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1530 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1532 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1535 * If we're doing a normal import, then build up any additional
1536 * diagnostic information about missing devices in this config.
1537 * We'll pass this up to the user for further processing.
1539 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1540 nvlist_t **child, *nv;
1543 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1545 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1547 for (int c = 0; c < rvd->vdev_children; c++) {
1548 vdev_t *tvd = rvd->vdev_child[c];
1549 vdev_t *mtvd = mrvd->vdev_child[c];
1551 if (tvd->vdev_ops == &vdev_missing_ops &&
1552 mtvd->vdev_ops != &vdev_missing_ops &&
1554 child[idx++] = vdev_config_generate(spa, mtvd,
1559 VERIFY(nvlist_add_nvlist_array(nv,
1560 ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1561 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1562 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1564 for (int i = 0; i < idx; i++)
1565 nvlist_free(child[i]);
1568 kmem_free(child, rvd->vdev_children * sizeof (char **));
1572 * Compare the root vdev tree with the information we have
1573 * from the MOS config (mrvd). Check each top-level vdev
1574 * with the corresponding MOS config top-level (mtvd).
1576 for (int c = 0; c < rvd->vdev_children; c++) {
1577 vdev_t *tvd = rvd->vdev_child[c];
1578 vdev_t *mtvd = mrvd->vdev_child[c];
1581 * Resolve any "missing" vdevs in the current configuration.
1582 * If we find that the MOS config has more accurate information
1583 * about the top-level vdev then use that vdev instead.
1585 if (tvd->vdev_ops == &vdev_missing_ops &&
1586 mtvd->vdev_ops != &vdev_missing_ops) {
1588 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1592 * Device specific actions.
1594 if (mtvd->vdev_islog) {
1595 spa_set_log_state(spa, SPA_LOG_CLEAR);
1598 * XXX - once we have 'readonly' pool
1599 * support we should be able to handle
1600 * missing data devices by transitioning
1601 * the pool to readonly.
1607 * Swap the missing vdev with the data we were
1608 * able to obtain from the MOS config.
1610 vdev_remove_child(rvd, tvd);
1611 vdev_remove_child(mrvd, mtvd);
1613 vdev_add_child(rvd, mtvd);
1614 vdev_add_child(mrvd, tvd);
1616 spa_config_exit(spa, SCL_ALL, FTAG);
1618 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1621 } else if (mtvd->vdev_islog) {
1623 * Load the slog device's state from the MOS config
1624 * since it's possible that the label does not
1625 * contain the most up-to-date information.
1627 vdev_load_log_state(tvd, mtvd);
1632 spa_config_exit(spa, SCL_ALL, FTAG);
1635 * Ensure we were able to validate the config.
1637 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1641 * Check for missing log devices
1644 spa_check_logs(spa_t *spa)
1646 switch (spa->spa_log_state) {
1647 case SPA_LOG_MISSING:
1648 /* need to recheck in case slog has been restored */
1649 case SPA_LOG_UNKNOWN:
1650 if (dmu_objset_find(spa->spa_name, zil_check_log_chain, NULL,
1651 DS_FIND_CHILDREN)) {
1652 spa_set_log_state(spa, SPA_LOG_MISSING);
1661 spa_passivate_log(spa_t *spa)
1663 vdev_t *rvd = spa->spa_root_vdev;
1664 boolean_t slog_found = B_FALSE;
1666 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1668 if (!spa_has_slogs(spa))
1671 for (int c = 0; c < rvd->vdev_children; c++) {
1672 vdev_t *tvd = rvd->vdev_child[c];
1673 metaslab_group_t *mg = tvd->vdev_mg;
1675 if (tvd->vdev_islog) {
1676 metaslab_group_passivate(mg);
1677 slog_found = B_TRUE;
1681 return (slog_found);
1685 spa_activate_log(spa_t *spa)
1687 vdev_t *rvd = spa->spa_root_vdev;
1689 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1691 for (int c = 0; c < rvd->vdev_children; c++) {
1692 vdev_t *tvd = rvd->vdev_child[c];
1693 metaslab_group_t *mg = tvd->vdev_mg;
1695 if (tvd->vdev_islog)
1696 metaslab_group_activate(mg);
1701 spa_offline_log(spa_t *spa)
1705 if ((error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1706 NULL, DS_FIND_CHILDREN)) == 0) {
1709 * We successfully offlined the log device, sync out the
1710 * current txg so that the "stubby" block can be removed
1713 txg_wait_synced(spa->spa_dsl_pool, 0);
1719 spa_aux_check_removed(spa_aux_vdev_t *sav)
1723 for (i = 0; i < sav->sav_count; i++)
1724 spa_check_removed(sav->sav_vdevs[i]);
1728 spa_claim_notify(zio_t *zio)
1730 spa_t *spa = zio->io_spa;
1735 mutex_enter(&spa->spa_props_lock); /* any mutex will do */
1736 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1737 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1738 mutex_exit(&spa->spa_props_lock);
1741 typedef struct spa_load_error {
1742 uint64_t sle_meta_count;
1743 uint64_t sle_data_count;
1747 spa_load_verify_done(zio_t *zio)
1749 blkptr_t *bp = zio->io_bp;
1750 spa_load_error_t *sle = zio->io_private;
1751 dmu_object_type_t type = BP_GET_TYPE(bp);
1752 int error = zio->io_error;
1755 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1756 type != DMU_OT_INTENT_LOG)
1757 atomic_add_64(&sle->sle_meta_count, 1);
1759 atomic_add_64(&sle->sle_data_count, 1);
1761 zio_data_buf_free(zio->io_data, zio->io_size);
1766 spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1767 arc_buf_t *pbuf, const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1771 size_t size = BP_GET_PSIZE(bp);
1772 void *data = zio_data_buf_alloc(size);
1774 zio_nowait(zio_read(rio, spa, bp, data, size,
1775 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1776 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1777 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1783 spa_load_verify(spa_t *spa)
1786 spa_load_error_t sle = { 0 };
1787 zpool_rewind_policy_t policy;
1788 boolean_t verify_ok = B_FALSE;
1791 zpool_get_rewind_policy(spa->spa_config, &policy);
1793 if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1796 rio = zio_root(spa, NULL, &sle,
1797 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1799 error = traverse_pool(spa, spa->spa_verify_min_txg,
1800 TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio);
1802 (void) zio_wait(rio);
1804 spa->spa_load_meta_errors = sle.sle_meta_count;
1805 spa->spa_load_data_errors = sle.sle_data_count;
1807 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1808 sle.sle_data_count <= policy.zrp_maxdata) {
1812 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1813 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1815 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1816 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1817 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1818 VERIFY(nvlist_add_int64(spa->spa_load_info,
1819 ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1820 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1821 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1823 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1827 if (error != ENXIO && error != EIO)
1832 return (verify_ok ? 0 : EIO);
1836 * Find a value in the pool props object.
1839 spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1841 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1842 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1846 * Find a value in the pool directory object.
1849 spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1851 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1852 name, sizeof (uint64_t), 1, val));
1856 spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1858 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1863 * Fix up config after a partly-completed split. This is done with the
1864 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
1865 * pool have that entry in their config, but only the splitting one contains
1866 * a list of all the guids of the vdevs that are being split off.
1868 * This function determines what to do with that list: either rejoin
1869 * all the disks to the pool, or complete the splitting process. To attempt
1870 * the rejoin, each disk that is offlined is marked online again, and
1871 * we do a reopen() call. If the vdev label for every disk that was
1872 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1873 * then we call vdev_split() on each disk, and complete the split.
1875 * Otherwise we leave the config alone, with all the vdevs in place in
1876 * the original pool.
1879 spa_try_repair(spa_t *spa, nvlist_t *config)
1886 boolean_t attempt_reopen;
1888 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
1891 /* check that the config is complete */
1892 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
1893 &glist, &gcount) != 0)
1896 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
1898 /* attempt to online all the vdevs & validate */
1899 attempt_reopen = B_TRUE;
1900 for (i = 0; i < gcount; i++) {
1901 if (glist[i] == 0) /* vdev is hole */
1904 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
1905 if (vd[i] == NULL) {
1907 * Don't bother attempting to reopen the disks;
1908 * just do the split.
1910 attempt_reopen = B_FALSE;
1912 /* attempt to re-online it */
1913 vd[i]->vdev_offline = B_FALSE;
1917 if (attempt_reopen) {
1918 vdev_reopen(spa->spa_root_vdev);
1920 /* check each device to see what state it's in */
1921 for (extracted = 0, i = 0; i < gcount; i++) {
1922 if (vd[i] != NULL &&
1923 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
1930 * If every disk has been moved to the new pool, or if we never
1931 * even attempted to look at them, then we split them off for
1934 if (!attempt_reopen || gcount == extracted) {
1935 for (i = 0; i < gcount; i++)
1938 vdev_reopen(spa->spa_root_vdev);
1941 kmem_free(vd, gcount * sizeof (vdev_t *));
1945 spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
1946 boolean_t mosconfig)
1948 nvlist_t *config = spa->spa_config;
1949 char *ereport = FM_EREPORT_ZFS_POOL;
1955 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
1958 ASSERT(spa->spa_comment == NULL);
1959 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
1960 spa->spa_comment = spa_strdup(comment);
1963 * Versioning wasn't explicitly added to the label until later, so if
1964 * it's not present treat it as the initial version.
1966 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
1967 &spa->spa_ubsync.ub_version) != 0)
1968 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
1970 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
1971 &spa->spa_config_txg);
1973 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
1974 spa_guid_exists(pool_guid, 0)) {
1977 spa->spa_config_guid = pool_guid;
1979 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
1981 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
1985 nvlist_free(spa->spa_load_info);
1986 spa->spa_load_info = fnvlist_alloc();
1988 gethrestime(&spa->spa_loaded_ts);
1989 error = spa_load_impl(spa, pool_guid, config, state, type,
1990 mosconfig, &ereport);
1993 spa->spa_minref = refcount_count(&spa->spa_refcount);
1995 if (error != EEXIST) {
1996 spa->spa_loaded_ts.tv_sec = 0;
1997 spa->spa_loaded_ts.tv_nsec = 0;
1999 if (error != EBADF) {
2000 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
2003 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2010 * Load an existing storage pool, using the pool's builtin spa_config as a
2011 * source of configuration information.
2014 spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
2015 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
2019 nvlist_t *nvroot = NULL;
2022 uberblock_t *ub = &spa->spa_uberblock;
2023 uint64_t children, config_cache_txg = spa->spa_config_txg;
2024 int orig_mode = spa->spa_mode;
2027 boolean_t missing_feat_write = B_FALSE;
2030 * If this is an untrusted config, access the pool in read-only mode.
2031 * This prevents things like resilvering recently removed devices.
2034 spa->spa_mode = FREAD;
2036 ASSERT(MUTEX_HELD(&spa_namespace_lock));
2038 spa->spa_load_state = state;
2040 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
2043 parse = (type == SPA_IMPORT_EXISTING ?
2044 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2047 * Create "The Godfather" zio to hold all async IOs
2049 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2050 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2053 * Parse the configuration into a vdev tree. We explicitly set the
2054 * value that will be returned by spa_version() since parsing the
2055 * configuration requires knowing the version number.
2057 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2058 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
2059 spa_config_exit(spa, SCL_ALL, FTAG);
2064 ASSERT(spa->spa_root_vdev == rvd);
2066 if (type != SPA_IMPORT_ASSEMBLE) {
2067 ASSERT(spa_guid(spa) == pool_guid);
2071 * Try to open all vdevs, loading each label in the process.
2073 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2074 error = vdev_open(rvd);
2075 spa_config_exit(spa, SCL_ALL, FTAG);
2080 * We need to validate the vdev labels against the configuration that
2081 * we have in hand, which is dependent on the setting of mosconfig. If
2082 * mosconfig is true then we're validating the vdev labels based on
2083 * that config. Otherwise, we're validating against the cached config
2084 * (zpool.cache) that was read when we loaded the zfs module, and then
2085 * later we will recursively call spa_load() and validate against
2088 * If we're assembling a new pool that's been split off from an
2089 * existing pool, the labels haven't yet been updated so we skip
2090 * validation for now.
2092 if (type != SPA_IMPORT_ASSEMBLE) {
2093 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2094 error = vdev_validate(rvd, mosconfig);
2095 spa_config_exit(spa, SCL_ALL, FTAG);
2100 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2105 * Find the best uberblock.
2107 vdev_uberblock_load(rvd, ub, &label);
2110 * If we weren't able to find a single valid uberblock, return failure.
2112 if (ub->ub_txg == 0) {
2114 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2118 * If the pool has an unsupported version we can't open it.
2120 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2122 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2125 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2129 * If we weren't able to find what's necessary for reading the
2130 * MOS in the label, return failure.
2132 if (label == NULL || nvlist_lookup_nvlist(label,
2133 ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) {
2135 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2140 * Update our in-core representation with the definitive values
2143 nvlist_free(spa->spa_label_features);
2144 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2150 * Look through entries in the label nvlist's features_for_read. If
2151 * there is a feature listed there which we don't understand then we
2152 * cannot open a pool.
2154 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2155 nvlist_t *unsup_feat;
2157 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2160 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2162 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2163 if (!zfeature_is_supported(nvpair_name(nvp))) {
2164 VERIFY(nvlist_add_string(unsup_feat,
2165 nvpair_name(nvp), "") == 0);
2169 if (!nvlist_empty(unsup_feat)) {
2170 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2171 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2172 nvlist_free(unsup_feat);
2173 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2177 nvlist_free(unsup_feat);
2181 * If the vdev guid sum doesn't match the uberblock, we have an
2182 * incomplete configuration. We first check to see if the pool
2183 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2184 * If it is, defer the vdev_guid_sum check till later so we
2185 * can handle missing vdevs.
2187 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
2188 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
2189 rvd->vdev_guid_sum != ub->ub_guid_sum)
2190 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2192 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2193 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2194 spa_try_repair(spa, config);
2195 spa_config_exit(spa, SCL_ALL, FTAG);
2196 nvlist_free(spa->spa_config_splitting);
2197 spa->spa_config_splitting = NULL;
2201 * Initialize internal SPA structures.
2203 spa->spa_state = POOL_STATE_ACTIVE;
2204 spa->spa_ubsync = spa->spa_uberblock;
2205 spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2206 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2207 spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2208 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2209 spa->spa_claim_max_txg = spa->spa_first_txg;
2210 spa->spa_prev_software_version = ub->ub_software_version;
2212 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2214 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2215 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2217 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2218 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2220 if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2221 boolean_t missing_feat_read = B_FALSE;
2222 nvlist_t *unsup_feat, *enabled_feat;
2224 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2225 &spa->spa_feat_for_read_obj) != 0) {
2226 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2229 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2230 &spa->spa_feat_for_write_obj) != 0) {
2231 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2234 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2235 &spa->spa_feat_desc_obj) != 0) {
2236 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2239 enabled_feat = fnvlist_alloc();
2240 unsup_feat = fnvlist_alloc();
2242 if (!feature_is_supported(spa->spa_meta_objset,
2243 spa->spa_feat_for_read_obj, spa->spa_feat_desc_obj,
2244 unsup_feat, enabled_feat))
2245 missing_feat_read = B_TRUE;
2247 if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) {
2248 if (!feature_is_supported(spa->spa_meta_objset,
2249 spa->spa_feat_for_write_obj, spa->spa_feat_desc_obj,
2250 unsup_feat, enabled_feat)) {
2251 missing_feat_write = B_TRUE;
2255 fnvlist_add_nvlist(spa->spa_load_info,
2256 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
2258 if (!nvlist_empty(unsup_feat)) {
2259 fnvlist_add_nvlist(spa->spa_load_info,
2260 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
2263 fnvlist_free(enabled_feat);
2264 fnvlist_free(unsup_feat);
2266 if (!missing_feat_read) {
2267 fnvlist_add_boolean(spa->spa_load_info,
2268 ZPOOL_CONFIG_CAN_RDONLY);
2272 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2273 * twofold: to determine whether the pool is available for
2274 * import in read-write mode and (if it is not) whether the
2275 * pool is available for import in read-only mode. If the pool
2276 * is available for import in read-write mode, it is displayed
2277 * as available in userland; if it is not available for import
2278 * in read-only mode, it is displayed as unavailable in
2279 * userland. If the pool is available for import in read-only
2280 * mode but not read-write mode, it is displayed as unavailable
2281 * in userland with a special note that the pool is actually
2282 * available for open in read-only mode.
2284 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2285 * missing a feature for write, we must first determine whether
2286 * the pool can be opened read-only before returning to
2287 * userland in order to know whether to display the
2288 * abovementioned note.
2290 if (missing_feat_read || (missing_feat_write &&
2291 spa_writeable(spa))) {
2292 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2297 spa->spa_is_initializing = B_TRUE;
2298 error = dsl_pool_open(spa->spa_dsl_pool);
2299 spa->spa_is_initializing = B_FALSE;
2301 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2305 nvlist_t *policy = NULL, *nvconfig;
2307 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2308 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2310 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2311 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2313 unsigned long myhostid = 0;
2315 VERIFY(nvlist_lookup_string(nvconfig,
2316 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2319 myhostid = zone_get_hostid(NULL);
2322 * We're emulating the system's hostid in userland, so
2323 * we can't use zone_get_hostid().
2325 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2326 #endif /* _KERNEL */
2327 if (check_hostid && hostid != 0 && myhostid != 0 &&
2328 hostid != myhostid) {
2329 nvlist_free(nvconfig);
2330 cmn_err(CE_WARN, "pool '%s' could not be "
2331 "loaded as it was last accessed by "
2332 "another system (host: %s hostid: 0x%lx). "
2333 "See: http://illumos.org/msg/ZFS-8000-EY",
2334 spa_name(spa), hostname,
2335 (unsigned long)hostid);
2339 if (nvlist_lookup_nvlist(spa->spa_config,
2340 ZPOOL_REWIND_POLICY, &policy) == 0)
2341 VERIFY(nvlist_add_nvlist(nvconfig,
2342 ZPOOL_REWIND_POLICY, policy) == 0);
2344 spa_config_set(spa, nvconfig);
2346 spa_deactivate(spa);
2347 spa_activate(spa, orig_mode);
2349 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2352 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2353 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2354 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2356 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2359 * Load the bit that tells us to use the new accounting function
2360 * (raid-z deflation). If we have an older pool, this will not
2363 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2364 if (error != 0 && error != ENOENT)
2365 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2367 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2368 &spa->spa_creation_version);
2369 if (error != 0 && error != ENOENT)
2370 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2373 * Load the persistent error log. If we have an older pool, this will
2376 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2377 if (error != 0 && error != ENOENT)
2378 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2380 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2381 &spa->spa_errlog_scrub);
2382 if (error != 0 && error != ENOENT)
2383 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2386 * Load the history object. If we have an older pool, this
2387 * will not be present.
2389 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2390 if (error != 0 && error != ENOENT)
2391 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2394 * If we're assembling the pool from the split-off vdevs of
2395 * an existing pool, we don't want to attach the spares & cache
2400 * Load any hot spares for this pool.
2402 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2403 if (error != 0 && error != ENOENT)
2404 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2405 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2406 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2407 if (load_nvlist(spa, spa->spa_spares.sav_object,
2408 &spa->spa_spares.sav_config) != 0)
2409 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2411 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2412 spa_load_spares(spa);
2413 spa_config_exit(spa, SCL_ALL, FTAG);
2414 } else if (error == 0) {
2415 spa->spa_spares.sav_sync = B_TRUE;
2419 * Load any level 2 ARC devices for this pool.
2421 error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2422 &spa->spa_l2cache.sav_object);
2423 if (error != 0 && error != ENOENT)
2424 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2425 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2426 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2427 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2428 &spa->spa_l2cache.sav_config) != 0)
2429 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2431 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2432 spa_load_l2cache(spa);
2433 spa_config_exit(spa, SCL_ALL, FTAG);
2434 } else if (error == 0) {
2435 spa->spa_l2cache.sav_sync = B_TRUE;
2438 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2440 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2441 if (error && error != ENOENT)
2442 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2445 uint64_t autoreplace;
2447 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2448 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2449 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2450 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2451 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2452 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2453 &spa->spa_dedup_ditto);
2455 spa->spa_autoreplace = (autoreplace != 0);
2459 * If the 'autoreplace' property is set, then post a resource notifying
2460 * the ZFS DE that it should not issue any faults for unopenable
2461 * devices. We also iterate over the vdevs, and post a sysevent for any
2462 * unopenable vdevs so that the normal autoreplace handler can take
2465 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2466 spa_check_removed(spa->spa_root_vdev);
2468 * For the import case, this is done in spa_import(), because
2469 * at this point we're using the spare definitions from
2470 * the MOS config, not necessarily from the userland config.
2472 if (state != SPA_LOAD_IMPORT) {
2473 spa_aux_check_removed(&spa->spa_spares);
2474 spa_aux_check_removed(&spa->spa_l2cache);
2479 * Load the vdev state for all toplevel vdevs.
2484 * Propagate the leaf DTLs we just loaded all the way up the tree.
2486 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2487 vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2488 spa_config_exit(spa, SCL_ALL, FTAG);
2491 * Load the DDTs (dedup tables).
2493 error = ddt_load(spa);
2495 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2497 spa_update_dspace(spa);
2500 * Validate the config, using the MOS config to fill in any
2501 * information which might be missing. If we fail to validate
2502 * the config then declare the pool unfit for use. If we're
2503 * assembling a pool from a split, the log is not transferred
2506 if (type != SPA_IMPORT_ASSEMBLE) {
2509 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2510 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2512 if (!spa_config_valid(spa, nvconfig)) {
2513 nvlist_free(nvconfig);
2514 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2517 nvlist_free(nvconfig);
2520 * Now that we've validated the config, check the state of the
2521 * root vdev. If it can't be opened, it indicates one or
2522 * more toplevel vdevs are faulted.
2524 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2527 if (spa_check_logs(spa)) {
2528 *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2529 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2533 if (missing_feat_write) {
2534 ASSERT(state == SPA_LOAD_TRYIMPORT);
2537 * At this point, we know that we can open the pool in
2538 * read-only mode but not read-write mode. We now have enough
2539 * information and can return to userland.
2541 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
2545 * We've successfully opened the pool, verify that we're ready
2546 * to start pushing transactions.
2548 if (state != SPA_LOAD_TRYIMPORT) {
2549 if (error = spa_load_verify(spa))
2550 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2554 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2555 spa->spa_load_max_txg == UINT64_MAX)) {
2557 int need_update = B_FALSE;
2559 ASSERT(state != SPA_LOAD_TRYIMPORT);
2562 * Claim log blocks that haven't been committed yet.
2563 * This must all happen in a single txg.
2564 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2565 * invoked from zil_claim_log_block()'s i/o done callback.
2566 * Price of rollback is that we abandon the log.
2568 spa->spa_claiming = B_TRUE;
2570 tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2571 spa_first_txg(spa));
2572 (void) dmu_objset_find(spa_name(spa),
2573 zil_claim, tx, DS_FIND_CHILDREN);
2576 spa->spa_claiming = B_FALSE;
2578 spa_set_log_state(spa, SPA_LOG_GOOD);
2579 spa->spa_sync_on = B_TRUE;
2580 txg_sync_start(spa->spa_dsl_pool);
2583 * Wait for all claims to sync. We sync up to the highest
2584 * claimed log block birth time so that claimed log blocks
2585 * don't appear to be from the future. spa_claim_max_txg
2586 * will have been set for us by either zil_check_log_chain()
2587 * (invoked from spa_check_logs()) or zil_claim() above.
2589 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2592 * If the config cache is stale, or we have uninitialized
2593 * metaslabs (see spa_vdev_add()), then update the config.
2595 * If this is a verbatim import, trust the current
2596 * in-core spa_config and update the disk labels.
2598 if (config_cache_txg != spa->spa_config_txg ||
2599 state == SPA_LOAD_IMPORT ||
2600 state == SPA_LOAD_RECOVER ||
2601 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2602 need_update = B_TRUE;
2604 for (int c = 0; c < rvd->vdev_children; c++)
2605 if (rvd->vdev_child[c]->vdev_ms_array == 0)
2606 need_update = B_TRUE;
2609 * Update the config cache asychronously in case we're the
2610 * root pool, in which case the config cache isn't writable yet.
2613 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2616 * Check all DTLs to see if anything needs resilvering.
2618 if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2619 vdev_resilver_needed(rvd, NULL, NULL))
2620 spa_async_request(spa, SPA_ASYNC_RESILVER);
2623 * Delete any inconsistent datasets.
2625 (void) dmu_objset_find(spa_name(spa),
2626 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2629 * Clean up any stale temporary dataset userrefs.
2631 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2638 spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2640 int mode = spa->spa_mode;
2643 spa_deactivate(spa);
2645 spa->spa_load_max_txg--;
2647 spa_activate(spa, mode);
2648 spa_async_suspend(spa);
2650 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2654 * If spa_load() fails this function will try loading prior txg's. If
2655 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2656 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2657 * function will not rewind the pool and will return the same error as
2661 spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2662 uint64_t max_request, int rewind_flags)
2664 nvlist_t *loadinfo = NULL;
2665 nvlist_t *config = NULL;
2666 int load_error, rewind_error;
2667 uint64_t safe_rewind_txg;
2670 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2671 spa->spa_load_max_txg = spa->spa_load_txg;
2672 spa_set_log_state(spa, SPA_LOG_CLEAR);
2674 spa->spa_load_max_txg = max_request;
2677 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2679 if (load_error == 0)
2682 if (spa->spa_root_vdev != NULL)
2683 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2685 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2686 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2688 if (rewind_flags & ZPOOL_NEVER_REWIND) {
2689 nvlist_free(config);
2690 return (load_error);
2693 if (state == SPA_LOAD_RECOVER) {
2694 /* Price of rolling back is discarding txgs, including log */
2695 spa_set_log_state(spa, SPA_LOG_CLEAR);
2698 * If we aren't rolling back save the load info from our first
2699 * import attempt so that we can restore it after attempting
2702 loadinfo = spa->spa_load_info;
2703 spa->spa_load_info = fnvlist_alloc();
2706 spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2707 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2708 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2709 TXG_INITIAL : safe_rewind_txg;
2712 * Continue as long as we're finding errors, we're still within
2713 * the acceptable rewind range, and we're still finding uberblocks
2715 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2716 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2717 if (spa->spa_load_max_txg < safe_rewind_txg)
2718 spa->spa_extreme_rewind = B_TRUE;
2719 rewind_error = spa_load_retry(spa, state, mosconfig);
2722 spa->spa_extreme_rewind = B_FALSE;
2723 spa->spa_load_max_txg = UINT64_MAX;
2725 if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2726 spa_config_set(spa, config);
2728 if (state == SPA_LOAD_RECOVER) {
2729 ASSERT3P(loadinfo, ==, NULL);
2730 return (rewind_error);
2732 /* Store the rewind info as part of the initial load info */
2733 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
2734 spa->spa_load_info);
2736 /* Restore the initial load info */
2737 fnvlist_free(spa->spa_load_info);
2738 spa->spa_load_info = loadinfo;
2740 return (load_error);
2747 * The import case is identical to an open except that the configuration is sent
2748 * down from userland, instead of grabbed from the configuration cache. For the
2749 * case of an open, the pool configuration will exist in the
2750 * POOL_STATE_UNINITIALIZED state.
2752 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2753 * the same time open the pool, without having to keep around the spa_t in some
2757 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2761 spa_load_state_t state = SPA_LOAD_OPEN;
2763 int locked = B_FALSE;
2764 int firstopen = B_FALSE;
2769 * As disgusting as this is, we need to support recursive calls to this
2770 * function because dsl_dir_open() is called during spa_load(), and ends
2771 * up calling spa_open() again. The real fix is to figure out how to
2772 * avoid dsl_dir_open() calling this in the first place.
2774 if (mutex_owner(&spa_namespace_lock) != curthread) {
2775 mutex_enter(&spa_namespace_lock);
2779 if ((spa = spa_lookup(pool)) == NULL) {
2781 mutex_exit(&spa_namespace_lock);
2785 if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2786 zpool_rewind_policy_t policy;
2790 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2792 if (policy.zrp_request & ZPOOL_DO_REWIND)
2793 state = SPA_LOAD_RECOVER;
2795 spa_activate(spa, spa_mode_global);
2797 if (state != SPA_LOAD_RECOVER)
2798 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2800 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2801 policy.zrp_request);
2803 if (error == EBADF) {
2805 * If vdev_validate() returns failure (indicated by
2806 * EBADF), it indicates that one of the vdevs indicates
2807 * that the pool has been exported or destroyed. If
2808 * this is the case, the config cache is out of sync and
2809 * we should remove the pool from the namespace.
2812 spa_deactivate(spa);
2813 spa_config_sync(spa, B_TRUE, B_TRUE);
2816 mutex_exit(&spa_namespace_lock);
2822 * We can't open the pool, but we still have useful
2823 * information: the state of each vdev after the
2824 * attempted vdev_open(). Return this to the user.
2826 if (config != NULL && spa->spa_config) {
2827 VERIFY(nvlist_dup(spa->spa_config, config,
2829 VERIFY(nvlist_add_nvlist(*config,
2830 ZPOOL_CONFIG_LOAD_INFO,
2831 spa->spa_load_info) == 0);
2834 spa_deactivate(spa);
2835 spa->spa_last_open_failed = error;
2837 mutex_exit(&spa_namespace_lock);
2843 spa_open_ref(spa, tag);
2846 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2849 * If we've recovered the pool, pass back any information we
2850 * gathered while doing the load.
2852 if (state == SPA_LOAD_RECOVER) {
2853 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
2854 spa->spa_load_info) == 0);
2858 spa->spa_last_open_failed = 0;
2859 spa->spa_last_ubsync_txg = 0;
2860 spa->spa_load_txg = 0;
2861 mutex_exit(&spa_namespace_lock);
2865 zvol_create_minors(pool);
2876 spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
2879 return (spa_open_common(name, spapp, tag, policy, config));
2883 spa_open(const char *name, spa_t **spapp, void *tag)
2885 return (spa_open_common(name, spapp, tag, NULL, NULL));
2889 * Lookup the given spa_t, incrementing the inject count in the process,
2890 * preventing it from being exported or destroyed.
2893 spa_inject_addref(char *name)
2897 mutex_enter(&spa_namespace_lock);
2898 if ((spa = spa_lookup(name)) == NULL) {
2899 mutex_exit(&spa_namespace_lock);
2902 spa->spa_inject_ref++;
2903 mutex_exit(&spa_namespace_lock);
2909 spa_inject_delref(spa_t *spa)
2911 mutex_enter(&spa_namespace_lock);
2912 spa->spa_inject_ref--;
2913 mutex_exit(&spa_namespace_lock);
2917 * Add spares device information to the nvlist.
2920 spa_add_spares(spa_t *spa, nvlist_t *config)
2930 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2932 if (spa->spa_spares.sav_count == 0)
2935 VERIFY(nvlist_lookup_nvlist(config,
2936 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2937 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
2938 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2940 VERIFY(nvlist_add_nvlist_array(nvroot,
2941 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2942 VERIFY(nvlist_lookup_nvlist_array(nvroot,
2943 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2946 * Go through and find any spares which have since been
2947 * repurposed as an active spare. If this is the case, update
2948 * their status appropriately.
2950 for (i = 0; i < nspares; i++) {
2951 VERIFY(nvlist_lookup_uint64(spares[i],
2952 ZPOOL_CONFIG_GUID, &guid) == 0);
2953 if (spa_spare_exists(guid, &pool, NULL) &&
2955 VERIFY(nvlist_lookup_uint64_array(
2956 spares[i], ZPOOL_CONFIG_VDEV_STATS,
2957 (uint64_t **)&vs, &vsc) == 0);
2958 vs->vs_state = VDEV_STATE_CANT_OPEN;
2959 vs->vs_aux = VDEV_AUX_SPARED;
2966 * Add l2cache device information to the nvlist, including vdev stats.
2969 spa_add_l2cache(spa_t *spa, nvlist_t *config)
2972 uint_t i, j, nl2cache;
2979 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2981 if (spa->spa_l2cache.sav_count == 0)
2984 VERIFY(nvlist_lookup_nvlist(config,
2985 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2986 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
2987 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2988 if (nl2cache != 0) {
2989 VERIFY(nvlist_add_nvlist_array(nvroot,
2990 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2991 VERIFY(nvlist_lookup_nvlist_array(nvroot,
2992 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2995 * Update level 2 cache device stats.
2998 for (i = 0; i < nl2cache; i++) {
2999 VERIFY(nvlist_lookup_uint64(l2cache[i],
3000 ZPOOL_CONFIG_GUID, &guid) == 0);
3003 for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
3005 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
3006 vd = spa->spa_l2cache.sav_vdevs[j];
3012 VERIFY(nvlist_lookup_uint64_array(l2cache[i],
3013 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
3015 vdev_get_stats(vd, vs);
3021 spa_add_feature_stats(spa_t *spa, nvlist_t *config)
3027 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3028 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3030 if (spa->spa_feat_for_read_obj != 0) {
3031 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3032 spa->spa_feat_for_read_obj);
3033 zap_cursor_retrieve(&zc, &za) == 0;
3034 zap_cursor_advance(&zc)) {
3035 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3036 za.za_num_integers == 1);
3037 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3038 za.za_first_integer));
3040 zap_cursor_fini(&zc);
3043 if (spa->spa_feat_for_write_obj != 0) {
3044 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3045 spa->spa_feat_for_write_obj);
3046 zap_cursor_retrieve(&zc, &za) == 0;
3047 zap_cursor_advance(&zc)) {
3048 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3049 za.za_num_integers == 1);
3050 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3051 za.za_first_integer));
3053 zap_cursor_fini(&zc);
3056 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
3058 nvlist_free(features);
3062 spa_get_stats(const char *name, nvlist_t **config,
3063 char *altroot, size_t buflen)
3069 error = spa_open_common(name, &spa, FTAG, NULL, config);
3073 * This still leaves a window of inconsistency where the spares
3074 * or l2cache devices could change and the config would be
3075 * self-inconsistent.
3077 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3079 if (*config != NULL) {
3080 uint64_t loadtimes[2];
3082 loadtimes[0] = spa->spa_loaded_ts.tv_sec;
3083 loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
3084 VERIFY(nvlist_add_uint64_array(*config,
3085 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
3087 VERIFY(nvlist_add_uint64(*config,
3088 ZPOOL_CONFIG_ERRCOUNT,
3089 spa_get_errlog_size(spa)) == 0);
3091 if (spa_suspended(spa))
3092 VERIFY(nvlist_add_uint64(*config,
3093 ZPOOL_CONFIG_SUSPENDED,
3094 spa->spa_failmode) == 0);
3096 spa_add_spares(spa, *config);
3097 spa_add_l2cache(spa, *config);
3098 spa_add_feature_stats(spa, *config);
3103 * We want to get the alternate root even for faulted pools, so we cheat
3104 * and call spa_lookup() directly.
3108 mutex_enter(&spa_namespace_lock);
3109 spa = spa_lookup(name);
3111 spa_altroot(spa, altroot, buflen);
3115 mutex_exit(&spa_namespace_lock);
3117 spa_altroot(spa, altroot, buflen);
3122 spa_config_exit(spa, SCL_CONFIG, FTAG);
3123 spa_close(spa, FTAG);
3130 * Validate that the auxiliary device array is well formed. We must have an
3131 * array of nvlists, each which describes a valid leaf vdev. If this is an
3132 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3133 * specified, as long as they are well-formed.
3136 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
3137 spa_aux_vdev_t *sav, const char *config, uint64_t version,
3138 vdev_labeltype_t label)
3145 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3148 * It's acceptable to have no devs specified.
3150 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
3157 * Make sure the pool is formatted with a version that supports this
3160 if (spa_version(spa) < version)
3164 * Set the pending device list so we correctly handle device in-use
3167 sav->sav_pending = dev;
3168 sav->sav_npending = ndev;
3170 for (i = 0; i < ndev; i++) {
3171 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
3175 if (!vd->vdev_ops->vdev_op_leaf) {
3182 * The L2ARC currently only supports disk devices in
3183 * kernel context. For user-level testing, we allow it.
3186 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
3187 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
3195 if ((error = vdev_open(vd)) == 0 &&
3196 (error = vdev_label_init(vd, crtxg, label)) == 0) {
3197 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
3198 vd->vdev_guid) == 0);
3204 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
3211 sav->sav_pending = NULL;
3212 sav->sav_npending = 0;
3217 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
3221 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3223 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3224 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
3225 VDEV_LABEL_SPARE)) != 0) {
3229 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3230 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
3231 VDEV_LABEL_L2CACHE));
3235 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
3240 if (sav->sav_config != NULL) {
3246 * Generate new dev list by concatentating with the
3249 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
3250 &olddevs, &oldndevs) == 0);
3252 newdevs = kmem_alloc(sizeof (void *) *
3253 (ndevs + oldndevs), KM_SLEEP);
3254 for (i = 0; i < oldndevs; i++)
3255 VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
3257 for (i = 0; i < ndevs; i++)
3258 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
3261 VERIFY(nvlist_remove(sav->sav_config, config,
3262 DATA_TYPE_NVLIST_ARRAY) == 0);
3264 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3265 config, newdevs, ndevs + oldndevs) == 0);
3266 for (i = 0; i < oldndevs + ndevs; i++)
3267 nvlist_free(newdevs[i]);
3268 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
3271 * Generate a new dev list.
3273 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
3275 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
3281 * Stop and drop level 2 ARC devices
3284 spa_l2cache_drop(spa_t *spa)
3288 spa_aux_vdev_t *sav = &spa->spa_l2cache;
3290 for (i = 0; i < sav->sav_count; i++) {
3293 vd = sav->sav_vdevs[i];
3296 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
3297 pool != 0ULL && l2arc_vdev_present(vd))
3298 l2arc_remove_vdev(vd);
3306 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
3307 const char *history_str, nvlist_t *zplprops)
3310 char *altroot = NULL;
3315 uint64_t txg = TXG_INITIAL;
3316 nvlist_t **spares, **l2cache;
3317 uint_t nspares, nl2cache;
3318 uint64_t version, obj;
3319 boolean_t has_features;
3322 * If this pool already exists, return failure.
3324 mutex_enter(&spa_namespace_lock);
3325 if (spa_lookup(pool) != NULL) {
3326 mutex_exit(&spa_namespace_lock);
3331 * Allocate a new spa_t structure.
3333 (void) nvlist_lookup_string(props,
3334 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3335 spa = spa_add(pool, NULL, altroot);
3336 spa_activate(spa, spa_mode_global);
3338 if (props && (error = spa_prop_validate(spa, props))) {
3339 spa_deactivate(spa);
3341 mutex_exit(&spa_namespace_lock);
3345 has_features = B_FALSE;
3346 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
3347 elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
3348 if (zpool_prop_feature(nvpair_name(elem)))
3349 has_features = B_TRUE;
3352 if (has_features || nvlist_lookup_uint64(props,
3353 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
3354 version = SPA_VERSION;
3356 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
3358 spa->spa_first_txg = txg;
3359 spa->spa_uberblock.ub_txg = txg - 1;
3360 spa->spa_uberblock.ub_version = version;
3361 spa->spa_ubsync = spa->spa_uberblock;
3364 * Create "The Godfather" zio to hold all async IOs
3366 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
3367 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
3370 * Create the root vdev.
3372 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3374 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3376 ASSERT(error != 0 || rvd != NULL);
3377 ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3379 if (error == 0 && !zfs_allocatable_devs(nvroot))
3383 (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3384 (error = spa_validate_aux(spa, nvroot, txg,
3385 VDEV_ALLOC_ADD)) == 0) {
3386 for (int c = 0; c < rvd->vdev_children; c++) {
3387 vdev_metaslab_set_size(rvd->vdev_child[c]);
3388 vdev_expand(rvd->vdev_child[c], txg);
3392 spa_config_exit(spa, SCL_ALL, FTAG);
3396 spa_deactivate(spa);
3398 mutex_exit(&spa_namespace_lock);
3403 * Get the list of spares, if specified.
3405 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3406 &spares, &nspares) == 0) {
3407 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3409 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3410 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3411 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3412 spa_load_spares(spa);
3413 spa_config_exit(spa, SCL_ALL, FTAG);
3414 spa->spa_spares.sav_sync = B_TRUE;
3418 * Get the list of level 2 cache devices, if specified.
3420 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3421 &l2cache, &nl2cache) == 0) {
3422 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3423 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3424 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3425 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3426 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3427 spa_load_l2cache(spa);
3428 spa_config_exit(spa, SCL_ALL, FTAG);
3429 spa->spa_l2cache.sav_sync = B_TRUE;
3432 spa->spa_is_initializing = B_TRUE;
3433 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3434 spa->spa_meta_objset = dp->dp_meta_objset;
3435 spa->spa_is_initializing = B_FALSE;
3438 * Create DDTs (dedup tables).
3442 spa_update_dspace(spa);
3444 tx = dmu_tx_create_assigned(dp, txg);
3447 * Create the pool config object.
3449 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3450 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3451 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3453 if (zap_add(spa->spa_meta_objset,
3454 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3455 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3456 cmn_err(CE_PANIC, "failed to add pool config");
3459 if (spa_version(spa) >= SPA_VERSION_FEATURES)
3460 spa_feature_create_zap_objects(spa, tx);
3462 if (zap_add(spa->spa_meta_objset,
3463 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3464 sizeof (uint64_t), 1, &version, tx) != 0) {
3465 cmn_err(CE_PANIC, "failed to add pool version");
3468 /* Newly created pools with the right version are always deflated. */
3469 if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3470 spa->spa_deflate = TRUE;
3471 if (zap_add(spa->spa_meta_objset,
3472 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3473 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3474 cmn_err(CE_PANIC, "failed to add deflate");
3479 * Create the deferred-free bpobj. Turn off compression
3480 * because sync-to-convergence takes longer if the blocksize
3483 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3484 dmu_object_set_compress(spa->spa_meta_objset, obj,
3485 ZIO_COMPRESS_OFF, tx);
3486 if (zap_add(spa->spa_meta_objset,
3487 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3488 sizeof (uint64_t), 1, &obj, tx) != 0) {
3489 cmn_err(CE_PANIC, "failed to add bpobj");
3491 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3492 spa->spa_meta_objset, obj));
3495 * Create the pool's history object.
3497 if (version >= SPA_VERSION_ZPOOL_HISTORY)
3498 spa_history_create_obj(spa, tx);
3501 * Set pool properties.
3503 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3504 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3505 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3506 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3508 if (props != NULL) {
3509 spa_configfile_set(spa, props, B_FALSE);
3510 spa_sync_props(spa, props, tx);
3515 spa->spa_sync_on = B_TRUE;
3516 txg_sync_start(spa->spa_dsl_pool);
3519 * We explicitly wait for the first transaction to complete so that our
3520 * bean counters are appropriately updated.
3522 txg_wait_synced(spa->spa_dsl_pool, txg);
3524 spa_config_sync(spa, B_FALSE, B_TRUE);
3526 if (version >= SPA_VERSION_ZPOOL_HISTORY && history_str != NULL)
3527 (void) spa_history_log(spa, history_str, LOG_CMD_POOL_CREATE);
3528 spa_history_log_version(spa, LOG_POOL_CREATE);
3530 spa->spa_minref = refcount_count(&spa->spa_refcount);
3532 mutex_exit(&spa_namespace_lock);
3540 * Get the root pool information from the root disk, then import the root pool
3541 * during the system boot up time.
3543 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3546 spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3549 nvlist_t *nvtop, *nvroot;
3552 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3556 * Add this top-level vdev to the child array.
3558 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3560 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3562 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3565 * Put this pool's top-level vdevs into a root vdev.
3567 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3568 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3569 VDEV_TYPE_ROOT) == 0);
3570 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3571 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3572 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3576 * Replace the existing vdev_tree with the new root vdev in
3577 * this pool's configuration (remove the old, add the new).
3579 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3580 nvlist_free(nvroot);
3585 * Walk the vdev tree and see if we can find a device with "better"
3586 * configuration. A configuration is "better" if the label on that
3587 * device has a more recent txg.
3590 spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3592 for (int c = 0; c < vd->vdev_children; c++)
3593 spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3595 if (vd->vdev_ops->vdev_op_leaf) {
3599 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3603 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3607 * Do we have a better boot device?
3609 if (label_txg > *txg) {
3618 * Import a root pool.
3620 * For x86. devpath_list will consist of devid and/or physpath name of
3621 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3622 * The GRUB "findroot" command will return the vdev we should boot.
3624 * For Sparc, devpath_list consists the physpath name of the booting device
3625 * no matter the rootpool is a single device pool or a mirrored pool.
3627 * "/pci@1f,0/ide@d/disk@0,0:a"
3630 spa_import_rootpool(char *devpath, char *devid)
3633 vdev_t *rvd, *bvd, *avd = NULL;
3634 nvlist_t *config, *nvtop;
3640 * Read the label from the boot device and generate a configuration.
3642 config = spa_generate_rootconf(devpath, devid, &guid);
3643 #if defined(_OBP) && defined(_KERNEL)
3644 if (config == NULL) {
3645 if (strstr(devpath, "/iscsi/ssd") != NULL) {
3647 get_iscsi_bootpath_phy(devpath);
3648 config = spa_generate_rootconf(devpath, devid, &guid);
3652 if (config == NULL) {
3653 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
3658 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3660 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3662 mutex_enter(&spa_namespace_lock);
3663 if ((spa = spa_lookup(pname)) != NULL) {
3665 * Remove the existing root pool from the namespace so that we
3666 * can replace it with the correct config we just read in.
3671 spa = spa_add(pname, config, NULL);
3672 spa->spa_is_root = B_TRUE;
3673 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3676 * Build up a vdev tree based on the boot device's label config.
3678 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3680 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3681 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3682 VDEV_ALLOC_ROOTPOOL);
3683 spa_config_exit(spa, SCL_ALL, FTAG);
3685 mutex_exit(&spa_namespace_lock);
3686 nvlist_free(config);
3687 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3693 * Get the boot vdev.
3695 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3696 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3697 (u_longlong_t)guid);
3703 * Determine if there is a better boot device.
3706 spa_alt_rootvdev(rvd, &avd, &txg);
3708 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3709 "try booting from '%s'", avd->vdev_path);
3715 * If the boot device is part of a spare vdev then ensure that
3716 * we're booting off the active spare.
3718 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3719 !bvd->vdev_isspare) {
3720 cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3721 "try booting from '%s'",
3723 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3729 spa_history_log_version(spa, LOG_POOL_IMPORT);
3731 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3733 spa_config_exit(spa, SCL_ALL, FTAG);
3734 mutex_exit(&spa_namespace_lock);
3736 nvlist_free(config);
3742 extern int vdev_geom_read_pool_label(const char *name, nvlist_t ***configs,
3746 spa_generate_rootconf(const char *name)
3748 nvlist_t **configs, **tops;
3750 nvlist_t *best_cfg, *nvtop, *nvroot;
3759 if (vdev_geom_read_pool_label(name, &configs, &count) != 0)
3762 ASSERT3U(count, !=, 0);
3764 for (i = 0; i < count; i++) {
3767 VERIFY(nvlist_lookup_uint64(configs[i], ZPOOL_CONFIG_POOL_TXG,
3769 if (txg > best_txg) {
3771 best_cfg = configs[i];
3776 * Multi-vdev root pool configuration discovery is not supported yet.
3779 nvlist_lookup_uint64(best_cfg, ZPOOL_CONFIG_VDEV_CHILDREN, &nchildren);
3781 nvlist_lookup_uint64_array(best_cfg, ZPOOL_CONFIG_HOLE_ARRAY,
3784 tops = kmem_zalloc(nchildren * sizeof(void *), KM_SLEEP);
3785 for (i = 0; i < nchildren; i++) {
3788 if (configs[i] == NULL)
3790 VERIFY(nvlist_lookup_nvlist(configs[i], ZPOOL_CONFIG_VDEV_TREE,
3792 nvlist_dup(nvtop, &tops[i], KM_SLEEP);
3794 for (i = 0; holes != NULL && i < nholes; i++) {
3797 if (tops[holes[i]] != NULL)
3799 nvlist_alloc(&tops[holes[i]], NV_UNIQUE_NAME, KM_SLEEP);
3800 VERIFY(nvlist_add_string(tops[holes[i]], ZPOOL_CONFIG_TYPE,
3801 VDEV_TYPE_HOLE) == 0);
3802 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_ID,
3804 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_GUID,
3807 for (i = 0; i < nchildren; i++) {
3808 if (tops[i] != NULL)
3810 nvlist_alloc(&tops[i], NV_UNIQUE_NAME, KM_SLEEP);
3811 VERIFY(nvlist_add_string(tops[i], ZPOOL_CONFIG_TYPE,
3812 VDEV_TYPE_MISSING) == 0);
3813 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_ID,
3815 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_GUID,
3820 * Create pool config based on the best vdev config.
3822 nvlist_dup(best_cfg, &config, KM_SLEEP);
3825 * Put this pool's top-level vdevs into a root vdev.
3827 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3829 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3830 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3831 VDEV_TYPE_ROOT) == 0);
3832 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3833 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3834 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3835 tops, nchildren) == 0);
3838 * Replace the existing vdev_tree with the new root vdev in
3839 * this pool's configuration (remove the old, add the new).
3841 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3844 * Drop vdev config elements that should not be present at pool level.
3846 nvlist_remove(config, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64);
3847 nvlist_remove(config, ZPOOL_CONFIG_TOP_GUID, DATA_TYPE_UINT64);
3849 for (i = 0; i < count; i++)
3850 nvlist_free(configs[i]);
3851 kmem_free(configs, count * sizeof(void *));
3852 for (i = 0; i < nchildren; i++)
3853 nvlist_free(tops[i]);
3854 kmem_free(tops, nchildren * sizeof(void *));
3855 nvlist_free(nvroot);
3860 spa_import_rootpool(const char *name)
3863 vdev_t *rvd, *bvd, *avd = NULL;
3864 nvlist_t *config, *nvtop;
3870 * Read the label from the boot device and generate a configuration.
3872 config = spa_generate_rootconf(name);
3874 mutex_enter(&spa_namespace_lock);
3875 if (config != NULL) {
3876 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3877 &pname) == 0 && strcmp(name, pname) == 0);
3878 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg)
3881 if ((spa = spa_lookup(pname)) != NULL) {
3883 * Remove the existing root pool from the namespace so
3884 * that we can replace it with the correct config
3889 spa = spa_add(pname, config, NULL);
3892 * Set spa_ubsync.ub_version as it can be used in vdev_alloc()
3893 * via spa_version().
3895 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
3896 &spa->spa_ubsync.ub_version) != 0)
3897 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
3898 } else if ((spa = spa_lookup(name)) == NULL) {
3899 cmn_err(CE_NOTE, "Cannot find the pool label for '%s'",
3903 VERIFY(nvlist_dup(spa->spa_config, &config, KM_SLEEP) == 0);
3905 spa->spa_is_root = B_TRUE;
3906 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3909 * Build up a vdev tree based on the boot device's label config.
3911 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3913 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3914 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3915 VDEV_ALLOC_ROOTPOOL);
3916 spa_config_exit(spa, SCL_ALL, FTAG);
3918 mutex_exit(&spa_namespace_lock);
3919 nvlist_free(config);
3920 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3925 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3927 spa_config_exit(spa, SCL_ALL, FTAG);
3928 mutex_exit(&spa_namespace_lock);
3930 nvlist_free(config);
3938 * Import a non-root pool into the system.
3941 spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
3944 char *altroot = NULL;
3945 spa_load_state_t state = SPA_LOAD_IMPORT;
3946 zpool_rewind_policy_t policy;
3947 uint64_t mode = spa_mode_global;
3948 uint64_t readonly = B_FALSE;
3951 nvlist_t **spares, **l2cache;
3952 uint_t nspares, nl2cache;
3955 * If a pool with this name exists, return failure.
3957 mutex_enter(&spa_namespace_lock);
3958 if (spa_lookup(pool) != NULL) {
3959 mutex_exit(&spa_namespace_lock);
3964 * Create and initialize the spa structure.
3966 (void) nvlist_lookup_string(props,
3967 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3968 (void) nvlist_lookup_uint64(props,
3969 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
3972 spa = spa_add(pool, config, altroot);
3973 spa->spa_import_flags = flags;
3976 * Verbatim import - Take a pool and insert it into the namespace
3977 * as if it had been loaded at boot.
3979 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
3981 spa_configfile_set(spa, props, B_FALSE);
3983 spa_config_sync(spa, B_FALSE, B_TRUE);
3985 mutex_exit(&spa_namespace_lock);
3986 spa_history_log_version(spa, LOG_POOL_IMPORT);
3991 spa_activate(spa, mode);
3994 * Don't start async tasks until we know everything is healthy.
3996 spa_async_suspend(spa);
3998 zpool_get_rewind_policy(config, &policy);
3999 if (policy.zrp_request & ZPOOL_DO_REWIND)
4000 state = SPA_LOAD_RECOVER;
4003 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
4004 * because the user-supplied config is actually the one to trust when
4007 if (state != SPA_LOAD_RECOVER)
4008 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
4010 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
4011 policy.zrp_request);
4014 * Propagate anything learned while loading the pool and pass it
4015 * back to caller (i.e. rewind info, missing devices, etc).
4017 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4018 spa->spa_load_info) == 0);
4020 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4022 * Toss any existing sparelist, as it doesn't have any validity
4023 * anymore, and conflicts with spa_has_spare().
4025 if (spa->spa_spares.sav_config) {
4026 nvlist_free(spa->spa_spares.sav_config);
4027 spa->spa_spares.sav_config = NULL;
4028 spa_load_spares(spa);
4030 if (spa->spa_l2cache.sav_config) {
4031 nvlist_free(spa->spa_l2cache.sav_config);
4032 spa->spa_l2cache.sav_config = NULL;
4033 spa_load_l2cache(spa);
4036 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4039 error = spa_validate_aux(spa, nvroot, -1ULL,
4042 error = spa_validate_aux(spa, nvroot, -1ULL,
4043 VDEV_ALLOC_L2CACHE);
4044 spa_config_exit(spa, SCL_ALL, FTAG);
4047 spa_configfile_set(spa, props, B_FALSE);
4049 if (error != 0 || (props && spa_writeable(spa) &&
4050 (error = spa_prop_set(spa, props)))) {
4052 spa_deactivate(spa);
4054 mutex_exit(&spa_namespace_lock);
4058 spa_async_resume(spa);
4061 * Override any spares and level 2 cache devices as specified by
4062 * the user, as these may have correct device names/devids, etc.
4064 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
4065 &spares, &nspares) == 0) {
4066 if (spa->spa_spares.sav_config)
4067 VERIFY(nvlist_remove(spa->spa_spares.sav_config,
4068 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
4070 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
4071 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4072 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
4073 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
4074 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4075 spa_load_spares(spa);
4076 spa_config_exit(spa, SCL_ALL, FTAG);
4077 spa->spa_spares.sav_sync = B_TRUE;
4079 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
4080 &l2cache, &nl2cache) == 0) {
4081 if (spa->spa_l2cache.sav_config)
4082 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
4083 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
4085 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
4086 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4087 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
4088 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
4089 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4090 spa_load_l2cache(spa);
4091 spa_config_exit(spa, SCL_ALL, FTAG);
4092 spa->spa_l2cache.sav_sync = B_TRUE;
4096 * Check for any removed devices.
4098 if (spa->spa_autoreplace) {
4099 spa_aux_check_removed(&spa->spa_spares);
4100 spa_aux_check_removed(&spa->spa_l2cache);
4103 if (spa_writeable(spa)) {
4105 * Update the config cache to include the newly-imported pool.
4107 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4111 * It's possible that the pool was expanded while it was exported.
4112 * We kick off an async task to handle this for us.
4114 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
4116 mutex_exit(&spa_namespace_lock);
4117 spa_history_log_version(spa, LOG_POOL_IMPORT);
4121 zvol_create_minors(pool);
4128 spa_tryimport(nvlist_t *tryconfig)
4130 nvlist_t *config = NULL;
4136 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
4139 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
4143 * Create and initialize the spa structure.
4145 mutex_enter(&spa_namespace_lock);
4146 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
4147 spa_activate(spa, FREAD);
4150 * Pass off the heavy lifting to spa_load().
4151 * Pass TRUE for mosconfig because the user-supplied config
4152 * is actually the one to trust when doing an import.
4154 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
4157 * If 'tryconfig' was at least parsable, return the current config.
4159 if (spa->spa_root_vdev != NULL) {
4160 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4161 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
4163 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4165 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
4166 spa->spa_uberblock.ub_timestamp) == 0);
4167 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4168 spa->spa_load_info) == 0);
4171 * If the bootfs property exists on this pool then we
4172 * copy it out so that external consumers can tell which
4173 * pools are bootable.
4175 if ((!error || error == EEXIST) && spa->spa_bootfs) {
4176 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4179 * We have to play games with the name since the
4180 * pool was opened as TRYIMPORT_NAME.
4182 if (dsl_dsobj_to_dsname(spa_name(spa),
4183 spa->spa_bootfs, tmpname) == 0) {
4185 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4187 cp = strchr(tmpname, '/');
4189 (void) strlcpy(dsname, tmpname,
4192 (void) snprintf(dsname, MAXPATHLEN,
4193 "%s/%s", poolname, ++cp);
4195 VERIFY(nvlist_add_string(config,
4196 ZPOOL_CONFIG_BOOTFS, dsname) == 0);
4197 kmem_free(dsname, MAXPATHLEN);
4199 kmem_free(tmpname, MAXPATHLEN);
4203 * Add the list of hot spares and level 2 cache devices.
4205 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4206 spa_add_spares(spa, config);
4207 spa_add_l2cache(spa, config);
4208 spa_config_exit(spa, SCL_CONFIG, FTAG);
4212 spa_deactivate(spa);
4214 mutex_exit(&spa_namespace_lock);
4220 * Pool export/destroy
4222 * The act of destroying or exporting a pool is very simple. We make sure there
4223 * is no more pending I/O and any references to the pool are gone. Then, we
4224 * update the pool state and sync all the labels to disk, removing the
4225 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4226 * we don't sync the labels or remove the configuration cache.
4229 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
4230 boolean_t force, boolean_t hardforce)
4237 if (!(spa_mode_global & FWRITE))
4240 mutex_enter(&spa_namespace_lock);
4241 if ((spa = spa_lookup(pool)) == NULL) {
4242 mutex_exit(&spa_namespace_lock);
4247 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4248 * reacquire the namespace lock, and see if we can export.
4250 spa_open_ref(spa, FTAG);
4251 mutex_exit(&spa_namespace_lock);
4252 spa_async_suspend(spa);
4253 mutex_enter(&spa_namespace_lock);
4254 spa_close(spa, FTAG);
4257 * The pool will be in core if it's openable,
4258 * in which case we can modify its state.
4260 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
4262 * Objsets may be open only because they're dirty, so we
4263 * have to force it to sync before checking spa_refcnt.
4265 txg_wait_synced(spa->spa_dsl_pool, 0);
4268 * A pool cannot be exported or destroyed if there are active
4269 * references. If we are resetting a pool, allow references by
4270 * fault injection handlers.
4272 if (!spa_refcount_zero(spa) ||
4273 (spa->spa_inject_ref != 0 &&
4274 new_state != POOL_STATE_UNINITIALIZED)) {
4275 spa_async_resume(spa);
4276 mutex_exit(&spa_namespace_lock);
4281 * A pool cannot be exported if it has an active shared spare.
4282 * This is to prevent other pools stealing the active spare
4283 * from an exported pool. At user's own will, such pool can
4284 * be forcedly exported.
4286 if (!force && new_state == POOL_STATE_EXPORTED &&
4287 spa_has_active_shared_spare(spa)) {
4288 spa_async_resume(spa);
4289 mutex_exit(&spa_namespace_lock);
4294 * We want this to be reflected on every label,
4295 * so mark them all dirty. spa_unload() will do the
4296 * final sync that pushes these changes out.
4298 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
4299 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4300 spa->spa_state = new_state;
4301 spa->spa_final_txg = spa_last_synced_txg(spa) +
4303 vdev_config_dirty(spa->spa_root_vdev);
4304 spa_config_exit(spa, SCL_ALL, FTAG);
4308 spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
4310 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4312 spa_deactivate(spa);
4315 if (oldconfig && spa->spa_config)
4316 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
4318 if (new_state != POOL_STATE_UNINITIALIZED) {
4320 spa_config_sync(spa, B_TRUE, B_TRUE);
4323 mutex_exit(&spa_namespace_lock);
4329 * Destroy a storage pool.
4332 spa_destroy(char *pool)
4334 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
4339 * Export a storage pool.
4342 spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
4343 boolean_t hardforce)
4345 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
4350 * Similar to spa_export(), this unloads the spa_t without actually removing it
4351 * from the namespace in any way.
4354 spa_reset(char *pool)
4356 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
4361 * ==========================================================================
4362 * Device manipulation
4363 * ==========================================================================
4367 * Add a device to a storage pool.
4370 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
4374 vdev_t *rvd = spa->spa_root_vdev;
4376 nvlist_t **spares, **l2cache;
4377 uint_t nspares, nl2cache;
4379 ASSERT(spa_writeable(spa));
4381 txg = spa_vdev_enter(spa);
4383 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
4384 VDEV_ALLOC_ADD)) != 0)
4385 return (spa_vdev_exit(spa, NULL, txg, error));
4387 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */
4389 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
4393 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
4397 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
4398 return (spa_vdev_exit(spa, vd, txg, EINVAL));
4400 if (vd->vdev_children != 0 &&
4401 (error = vdev_create(vd, txg, B_FALSE)) != 0)
4402 return (spa_vdev_exit(spa, vd, txg, error));
4405 * We must validate the spares and l2cache devices after checking the
4406 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
4408 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
4409 return (spa_vdev_exit(spa, vd, txg, error));
4412 * Transfer each new top-level vdev from vd to rvd.
4414 for (int c = 0; c < vd->vdev_children; c++) {
4417 * Set the vdev id to the first hole, if one exists.
4419 for (id = 0; id < rvd->vdev_children; id++) {
4420 if (rvd->vdev_child[id]->vdev_ishole) {
4421 vdev_free(rvd->vdev_child[id]);
4425 tvd = vd->vdev_child[c];
4426 vdev_remove_child(vd, tvd);
4428 vdev_add_child(rvd, tvd);
4429 vdev_config_dirty(tvd);
4433 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
4434 ZPOOL_CONFIG_SPARES);
4435 spa_load_spares(spa);
4436 spa->spa_spares.sav_sync = B_TRUE;
4439 if (nl2cache != 0) {
4440 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
4441 ZPOOL_CONFIG_L2CACHE);
4442 spa_load_l2cache(spa);
4443 spa->spa_l2cache.sav_sync = B_TRUE;
4447 * We have to be careful when adding new vdevs to an existing pool.
4448 * If other threads start allocating from these vdevs before we
4449 * sync the config cache, and we lose power, then upon reboot we may
4450 * fail to open the pool because there are DVAs that the config cache
4451 * can't translate. Therefore, we first add the vdevs without
4452 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4453 * and then let spa_config_update() initialize the new metaslabs.
4455 * spa_load() checks for added-but-not-initialized vdevs, so that
4456 * if we lose power at any point in this sequence, the remaining
4457 * steps will be completed the next time we load the pool.
4459 (void) spa_vdev_exit(spa, vd, txg, 0);
4461 mutex_enter(&spa_namespace_lock);
4462 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4463 mutex_exit(&spa_namespace_lock);
4469 * Attach a device to a mirror. The arguments are the path to any device
4470 * in the mirror, and the nvroot for the new device. If the path specifies
4471 * a device that is not mirrored, we automatically insert the mirror vdev.
4473 * If 'replacing' is specified, the new device is intended to replace the
4474 * existing device; in this case the two devices are made into their own
4475 * mirror using the 'replacing' vdev, which is functionally identical to
4476 * the mirror vdev (it actually reuses all the same ops) but has a few
4477 * extra rules: you can't attach to it after it's been created, and upon
4478 * completion of resilvering, the first disk (the one being replaced)
4479 * is automatically detached.
4482 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
4484 uint64_t txg, dtl_max_txg;
4485 vdev_t *rvd = spa->spa_root_vdev;
4486 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
4488 char *oldvdpath, *newvdpath;
4492 ASSERT(spa_writeable(spa));
4494 txg = spa_vdev_enter(spa);
4496 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
4499 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4501 if (!oldvd->vdev_ops->vdev_op_leaf)
4502 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4504 pvd = oldvd->vdev_parent;
4506 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
4507 VDEV_ALLOC_ATTACH)) != 0)
4508 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4510 if (newrootvd->vdev_children != 1)
4511 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4513 newvd = newrootvd->vdev_child[0];
4515 if (!newvd->vdev_ops->vdev_op_leaf)
4516 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4518 if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
4519 return (spa_vdev_exit(spa, newrootvd, txg, error));
4522 * Spares can't replace logs
4524 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
4525 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4529 * For attach, the only allowable parent is a mirror or the root
4532 if (pvd->vdev_ops != &vdev_mirror_ops &&
4533 pvd->vdev_ops != &vdev_root_ops)
4534 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4536 pvops = &vdev_mirror_ops;
4539 * Active hot spares can only be replaced by inactive hot
4542 if (pvd->vdev_ops == &vdev_spare_ops &&
4543 oldvd->vdev_isspare &&
4544 !spa_has_spare(spa, newvd->vdev_guid))
4545 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4548 * If the source is a hot spare, and the parent isn't already a
4549 * spare, then we want to create a new hot spare. Otherwise, we
4550 * want to create a replacing vdev. The user is not allowed to
4551 * attach to a spared vdev child unless the 'isspare' state is
4552 * the same (spare replaces spare, non-spare replaces
4555 if (pvd->vdev_ops == &vdev_replacing_ops &&
4556 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
4557 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4558 } else if (pvd->vdev_ops == &vdev_spare_ops &&
4559 newvd->vdev_isspare != oldvd->vdev_isspare) {
4560 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4563 if (newvd->vdev_isspare)
4564 pvops = &vdev_spare_ops;
4566 pvops = &vdev_replacing_ops;
4570 * Make sure the new device is big enough.
4572 if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
4573 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
4576 * The new device cannot have a higher alignment requirement
4577 * than the top-level vdev.
4579 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4580 return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4583 * If this is an in-place replacement, update oldvd's path and devid
4584 * to make it distinguishable from newvd, and unopenable from now on.
4586 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4587 spa_strfree(oldvd->vdev_path);
4588 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4590 (void) sprintf(oldvd->vdev_path, "%s/%s",
4591 newvd->vdev_path, "old");
4592 if (oldvd->vdev_devid != NULL) {
4593 spa_strfree(oldvd->vdev_devid);
4594 oldvd->vdev_devid = NULL;
4598 /* mark the device being resilvered */
4599 newvd->vdev_resilvering = B_TRUE;
4602 * If the parent is not a mirror, or if we're replacing, insert the new
4603 * mirror/replacing/spare vdev above oldvd.
4605 if (pvd->vdev_ops != pvops)
4606 pvd = vdev_add_parent(oldvd, pvops);
4608 ASSERT(pvd->vdev_top->vdev_parent == rvd);
4609 ASSERT(pvd->vdev_ops == pvops);
4610 ASSERT(oldvd->vdev_parent == pvd);
4613 * Extract the new device from its root and add it to pvd.
4615 vdev_remove_child(newrootvd, newvd);
4616 newvd->vdev_id = pvd->vdev_children;
4617 newvd->vdev_crtxg = oldvd->vdev_crtxg;
4618 vdev_add_child(pvd, newvd);
4620 tvd = newvd->vdev_top;
4621 ASSERT(pvd->vdev_top == tvd);
4622 ASSERT(tvd->vdev_parent == rvd);
4624 vdev_config_dirty(tvd);
4627 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4628 * for any dmu_sync-ed blocks. It will propagate upward when
4629 * spa_vdev_exit() calls vdev_dtl_reassess().
4631 dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4633 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4634 dtl_max_txg - TXG_INITIAL);
4636 if (newvd->vdev_isspare) {
4637 spa_spare_activate(newvd);
4638 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
4641 oldvdpath = spa_strdup(oldvd->vdev_path);
4642 newvdpath = spa_strdup(newvd->vdev_path);
4643 newvd_isspare = newvd->vdev_isspare;
4646 * Mark newvd's DTL dirty in this txg.
4648 vdev_dirty(tvd, VDD_DTL, newvd, txg);
4651 * Restart the resilver
4653 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4658 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4660 spa_history_log_internal(LOG_POOL_VDEV_ATTACH, spa, NULL,
4661 "%s vdev=%s %s vdev=%s",
4662 replacing && newvd_isspare ? "spare in" :
4663 replacing ? "replace" : "attach", newvdpath,
4664 replacing ? "for" : "to", oldvdpath);
4666 spa_strfree(oldvdpath);
4667 spa_strfree(newvdpath);
4669 if (spa->spa_bootfs)
4670 spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
4676 * Detach a device from a mirror or replacing vdev.
4677 * If 'replace_done' is specified, only detach if the parent
4678 * is a replacing vdev.
4681 spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4685 vdev_t *rvd = spa->spa_root_vdev;
4686 vdev_t *vd, *pvd, *cvd, *tvd;
4687 boolean_t unspare = B_FALSE;
4688 uint64_t unspare_guid;
4691 ASSERT(spa_writeable(spa));
4693 txg = spa_vdev_enter(spa);
4695 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4698 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4700 if (!vd->vdev_ops->vdev_op_leaf)
4701 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4703 pvd = vd->vdev_parent;
4706 * If the parent/child relationship is not as expected, don't do it.
4707 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4708 * vdev that's replacing B with C. The user's intent in replacing
4709 * is to go from M(A,B) to M(A,C). If the user decides to cancel
4710 * the replace by detaching C, the expected behavior is to end up
4711 * M(A,B). But suppose that right after deciding to detach C,
4712 * the replacement of B completes. We would have M(A,C), and then
4713 * ask to detach C, which would leave us with just A -- not what
4714 * the user wanted. To prevent this, we make sure that the
4715 * parent/child relationship hasn't changed -- in this example,
4716 * that C's parent is still the replacing vdev R.
4718 if (pvd->vdev_guid != pguid && pguid != 0)
4719 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4722 * Only 'replacing' or 'spare' vdevs can be replaced.
4724 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4725 pvd->vdev_ops != &vdev_spare_ops)
4726 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4728 ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4729 spa_version(spa) >= SPA_VERSION_SPARES);
4732 * Only mirror, replacing, and spare vdevs support detach.
4734 if (pvd->vdev_ops != &vdev_replacing_ops &&
4735 pvd->vdev_ops != &vdev_mirror_ops &&
4736 pvd->vdev_ops != &vdev_spare_ops)
4737 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4740 * If this device has the only valid copy of some data,
4741 * we cannot safely detach it.
4743 if (vdev_dtl_required(vd))
4744 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4746 ASSERT(pvd->vdev_children >= 2);
4749 * If we are detaching the second disk from a replacing vdev, then
4750 * check to see if we changed the original vdev's path to have "/old"
4751 * at the end in spa_vdev_attach(). If so, undo that change now.
4753 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4754 vd->vdev_path != NULL) {
4755 size_t len = strlen(vd->vdev_path);
4757 for (int c = 0; c < pvd->vdev_children; c++) {
4758 cvd = pvd->vdev_child[c];
4760 if (cvd == vd || cvd->vdev_path == NULL)
4763 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4764 strcmp(cvd->vdev_path + len, "/old") == 0) {
4765 spa_strfree(cvd->vdev_path);
4766 cvd->vdev_path = spa_strdup(vd->vdev_path);
4773 * If we are detaching the original disk from a spare, then it implies
4774 * that the spare should become a real disk, and be removed from the
4775 * active spare list for the pool.
4777 if (pvd->vdev_ops == &vdev_spare_ops &&
4779 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4783 * Erase the disk labels so the disk can be used for other things.
4784 * This must be done after all other error cases are handled,
4785 * but before we disembowel vd (so we can still do I/O to it).
4786 * But if we can't do it, don't treat the error as fatal --
4787 * it may be that the unwritability of the disk is the reason
4788 * it's being detached!
4790 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4793 * Remove vd from its parent and compact the parent's children.
4795 vdev_remove_child(pvd, vd);
4796 vdev_compact_children(pvd);
4799 * Remember one of the remaining children so we can get tvd below.
4801 cvd = pvd->vdev_child[pvd->vdev_children - 1];
4804 * If we need to remove the remaining child from the list of hot spares,
4805 * do it now, marking the vdev as no longer a spare in the process.
4806 * We must do this before vdev_remove_parent(), because that can
4807 * change the GUID if it creates a new toplevel GUID. For a similar
4808 * reason, we must remove the spare now, in the same txg as the detach;
4809 * otherwise someone could attach a new sibling, change the GUID, and
4810 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4813 ASSERT(cvd->vdev_isspare);
4814 spa_spare_remove(cvd);
4815 unspare_guid = cvd->vdev_guid;
4816 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4817 cvd->vdev_unspare = B_TRUE;
4821 * If the parent mirror/replacing vdev only has one child,
4822 * the parent is no longer needed. Remove it from the tree.
4824 if (pvd->vdev_children == 1) {
4825 if (pvd->vdev_ops == &vdev_spare_ops)
4826 cvd->vdev_unspare = B_FALSE;
4827 vdev_remove_parent(cvd);
4828 cvd->vdev_resilvering = B_FALSE;
4833 * We don't set tvd until now because the parent we just removed
4834 * may have been the previous top-level vdev.
4836 tvd = cvd->vdev_top;
4837 ASSERT(tvd->vdev_parent == rvd);
4840 * Reevaluate the parent vdev state.
4842 vdev_propagate_state(cvd);
4845 * If the 'autoexpand' property is set on the pool then automatically
4846 * try to expand the size of the pool. For example if the device we
4847 * just detached was smaller than the others, it may be possible to
4848 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4849 * first so that we can obtain the updated sizes of the leaf vdevs.
4851 if (spa->spa_autoexpand) {
4853 vdev_expand(tvd, txg);
4856 vdev_config_dirty(tvd);
4859 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
4860 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4861 * But first make sure we're not on any *other* txg's DTL list, to
4862 * prevent vd from being accessed after it's freed.
4864 vdpath = spa_strdup(vd->vdev_path);
4865 for (int t = 0; t < TXG_SIZE; t++)
4866 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4867 vd->vdev_detached = B_TRUE;
4868 vdev_dirty(tvd, VDD_DTL, vd, txg);
4870 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
4872 /* hang on to the spa before we release the lock */
4873 spa_open_ref(spa, FTAG);
4875 error = spa_vdev_exit(spa, vd, txg, 0);
4877 spa_history_log_internal(LOG_POOL_VDEV_DETACH, spa, NULL,
4879 spa_strfree(vdpath);
4882 * If this was the removal of the original device in a hot spare vdev,
4883 * then we want to go through and remove the device from the hot spare
4884 * list of every other pool.
4887 spa_t *altspa = NULL;
4889 mutex_enter(&spa_namespace_lock);
4890 while ((altspa = spa_next(altspa)) != NULL) {
4891 if (altspa->spa_state != POOL_STATE_ACTIVE ||
4895 spa_open_ref(altspa, FTAG);
4896 mutex_exit(&spa_namespace_lock);
4897 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
4898 mutex_enter(&spa_namespace_lock);
4899 spa_close(altspa, FTAG);
4901 mutex_exit(&spa_namespace_lock);
4903 /* search the rest of the vdevs for spares to remove */
4904 spa_vdev_resilver_done(spa);
4907 /* all done with the spa; OK to release */
4908 mutex_enter(&spa_namespace_lock);
4909 spa_close(spa, FTAG);
4910 mutex_exit(&spa_namespace_lock);
4916 * Split a set of devices from their mirrors, and create a new pool from them.
4919 spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
4920 nvlist_t *props, boolean_t exp)
4923 uint64_t txg, *glist;
4925 uint_t c, children, lastlog;
4926 nvlist_t **child, *nvl, *tmp;
4928 char *altroot = NULL;
4929 vdev_t *rvd, **vml = NULL; /* vdev modify list */
4930 boolean_t activate_slog;
4932 ASSERT(spa_writeable(spa));
4934 txg = spa_vdev_enter(spa);
4936 /* clear the log and flush everything up to now */
4937 activate_slog = spa_passivate_log(spa);
4938 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4939 error = spa_offline_log(spa);
4940 txg = spa_vdev_config_enter(spa);
4943 spa_activate_log(spa);
4946 return (spa_vdev_exit(spa, NULL, txg, error));
4948 /* check new spa name before going any further */
4949 if (spa_lookup(newname) != NULL)
4950 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
4953 * scan through all the children to ensure they're all mirrors
4955 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
4956 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
4958 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4960 /* first, check to ensure we've got the right child count */
4961 rvd = spa->spa_root_vdev;
4963 for (c = 0; c < rvd->vdev_children; c++) {
4964 vdev_t *vd = rvd->vdev_child[c];
4966 /* don't count the holes & logs as children */
4967 if (vd->vdev_islog || vd->vdev_ishole) {
4975 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
4976 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4978 /* next, ensure no spare or cache devices are part of the split */
4979 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
4980 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
4981 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4983 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
4984 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
4986 /* then, loop over each vdev and validate it */
4987 for (c = 0; c < children; c++) {
4988 uint64_t is_hole = 0;
4990 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
4994 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
4995 spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
5003 /* which disk is going to be split? */
5004 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
5010 /* look it up in the spa */
5011 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
5012 if (vml[c] == NULL) {
5017 /* make sure there's nothing stopping the split */
5018 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
5019 vml[c]->vdev_islog ||
5020 vml[c]->vdev_ishole ||
5021 vml[c]->vdev_isspare ||
5022 vml[c]->vdev_isl2cache ||
5023 !vdev_writeable(vml[c]) ||
5024 vml[c]->vdev_children != 0 ||
5025 vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
5026 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
5031 if (vdev_dtl_required(vml[c])) {
5036 /* we need certain info from the top level */
5037 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
5038 vml[c]->vdev_top->vdev_ms_array) == 0);
5039 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
5040 vml[c]->vdev_top->vdev_ms_shift) == 0);
5041 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
5042 vml[c]->vdev_top->vdev_asize) == 0);
5043 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
5044 vml[c]->vdev_top->vdev_ashift) == 0);
5048 kmem_free(vml, children * sizeof (vdev_t *));
5049 kmem_free(glist, children * sizeof (uint64_t));
5050 return (spa_vdev_exit(spa, NULL, txg, error));
5053 /* stop writers from using the disks */
5054 for (c = 0; c < children; c++) {
5056 vml[c]->vdev_offline = B_TRUE;
5058 vdev_reopen(spa->spa_root_vdev);
5061 * Temporarily record the splitting vdevs in the spa config. This
5062 * will disappear once the config is regenerated.
5064 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5065 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
5066 glist, children) == 0);
5067 kmem_free(glist, children * sizeof (uint64_t));
5069 mutex_enter(&spa->spa_props_lock);
5070 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
5072 mutex_exit(&spa->spa_props_lock);
5073 spa->spa_config_splitting = nvl;
5074 vdev_config_dirty(spa->spa_root_vdev);
5076 /* configure and create the new pool */
5077 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
5078 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
5079 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
5080 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5081 spa_version(spa)) == 0);
5082 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
5083 spa->spa_config_txg) == 0);
5084 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
5085 spa_generate_guid(NULL)) == 0);
5086 (void) nvlist_lookup_string(props,
5087 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
5089 /* add the new pool to the namespace */
5090 newspa = spa_add(newname, config, altroot);
5091 newspa->spa_config_txg = spa->spa_config_txg;
5092 spa_set_log_state(newspa, SPA_LOG_CLEAR);
5094 /* release the spa config lock, retaining the namespace lock */
5095 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5097 if (zio_injection_enabled)
5098 zio_handle_panic_injection(spa, FTAG, 1);
5100 spa_activate(newspa, spa_mode_global);
5101 spa_async_suspend(newspa);
5104 /* mark that we are creating new spa by splitting */
5105 newspa->spa_splitting_newspa = B_TRUE;
5107 /* create the new pool from the disks of the original pool */
5108 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
5110 newspa->spa_splitting_newspa = B_FALSE;
5115 /* if that worked, generate a real config for the new pool */
5116 if (newspa->spa_root_vdev != NULL) {
5117 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
5118 NV_UNIQUE_NAME, KM_SLEEP) == 0);
5119 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
5120 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
5121 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
5126 if (props != NULL) {
5127 spa_configfile_set(newspa, props, B_FALSE);
5128 error = spa_prop_set(newspa, props);
5133 /* flush everything */
5134 txg = spa_vdev_config_enter(newspa);
5135 vdev_config_dirty(newspa->spa_root_vdev);
5136 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
5138 if (zio_injection_enabled)
5139 zio_handle_panic_injection(spa, FTAG, 2);
5141 spa_async_resume(newspa);
5143 /* finally, update the original pool's config */
5144 txg = spa_vdev_config_enter(spa);
5145 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
5146 error = dmu_tx_assign(tx, TXG_WAIT);
5149 for (c = 0; c < children; c++) {
5150 if (vml[c] != NULL) {
5153 spa_history_log_internal(LOG_POOL_VDEV_DETACH,
5159 vdev_config_dirty(spa->spa_root_vdev);
5160 spa->spa_config_splitting = NULL;
5164 (void) spa_vdev_exit(spa, NULL, txg, 0);
5166 if (zio_injection_enabled)
5167 zio_handle_panic_injection(spa, FTAG, 3);
5169 /* split is complete; log a history record */
5170 spa_history_log_internal(LOG_POOL_SPLIT, newspa, NULL,
5171 "split new pool %s from pool %s", newname, spa_name(spa));
5173 kmem_free(vml, children * sizeof (vdev_t *));
5175 /* if we're not going to mount the filesystems in userland, export */
5177 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
5184 spa_deactivate(newspa);
5187 txg = spa_vdev_config_enter(spa);
5189 /* re-online all offlined disks */
5190 for (c = 0; c < children; c++) {
5192 vml[c]->vdev_offline = B_FALSE;
5194 vdev_reopen(spa->spa_root_vdev);
5196 nvlist_free(spa->spa_config_splitting);
5197 spa->spa_config_splitting = NULL;
5198 (void) spa_vdev_exit(spa, NULL, txg, error);
5200 kmem_free(vml, children * sizeof (vdev_t *));
5205 spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
5207 for (int i = 0; i < count; i++) {
5210 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
5213 if (guid == target_guid)
5221 spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
5222 nvlist_t *dev_to_remove)
5224 nvlist_t **newdev = NULL;
5227 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
5229 for (int i = 0, j = 0; i < count; i++) {
5230 if (dev[i] == dev_to_remove)
5232 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
5235 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
5236 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
5238 for (int i = 0; i < count - 1; i++)
5239 nvlist_free(newdev[i]);
5242 kmem_free(newdev, (count - 1) * sizeof (void *));
5246 * Evacuate the device.
5249 spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
5254 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5255 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5256 ASSERT(vd == vd->vdev_top);
5259 * Evacuate the device. We don't hold the config lock as writer
5260 * since we need to do I/O but we do keep the
5261 * spa_namespace_lock held. Once this completes the device
5262 * should no longer have any blocks allocated on it.
5264 if (vd->vdev_islog) {
5265 if (vd->vdev_stat.vs_alloc != 0)
5266 error = spa_offline_log(spa);
5275 * The evacuation succeeded. Remove any remaining MOS metadata
5276 * associated with this vdev, and wait for these changes to sync.
5278 ASSERT0(vd->vdev_stat.vs_alloc);
5279 txg = spa_vdev_config_enter(spa);
5280 vd->vdev_removing = B_TRUE;
5281 vdev_dirty(vd, 0, NULL, txg);
5282 vdev_config_dirty(vd);
5283 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5289 * Complete the removal by cleaning up the namespace.
5292 spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
5294 vdev_t *rvd = spa->spa_root_vdev;
5295 uint64_t id = vd->vdev_id;
5296 boolean_t last_vdev = (id == (rvd->vdev_children - 1));
5298 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5299 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5300 ASSERT(vd == vd->vdev_top);
5303 * Only remove any devices which are empty.
5305 if (vd->vdev_stat.vs_alloc != 0)
5308 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5310 if (list_link_active(&vd->vdev_state_dirty_node))
5311 vdev_state_clean(vd);
5312 if (list_link_active(&vd->vdev_config_dirty_node))
5313 vdev_config_clean(vd);
5318 vdev_compact_children(rvd);
5320 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
5321 vdev_add_child(rvd, vd);
5323 vdev_config_dirty(rvd);
5326 * Reassess the health of our root vdev.
5332 * Remove a device from the pool -
5334 * Removing a device from the vdev namespace requires several steps
5335 * and can take a significant amount of time. As a result we use
5336 * the spa_vdev_config_[enter/exit] functions which allow us to
5337 * grab and release the spa_config_lock while still holding the namespace
5338 * lock. During each step the configuration is synced out.
5342 * Remove a device from the pool. Currently, this supports removing only hot
5343 * spares, slogs, and level 2 ARC devices.
5346 spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
5349 metaslab_group_t *mg;
5350 nvlist_t **spares, **l2cache, *nv;
5352 uint_t nspares, nl2cache;
5354 boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
5356 ASSERT(spa_writeable(spa));
5359 txg = spa_vdev_enter(spa);
5361 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5363 if (spa->spa_spares.sav_vdevs != NULL &&
5364 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5365 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
5366 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
5368 * Only remove the hot spare if it's not currently in use
5371 if (vd == NULL || unspare) {
5372 spa_vdev_remove_aux(spa->spa_spares.sav_config,
5373 ZPOOL_CONFIG_SPARES, spares, nspares, nv);
5374 spa_load_spares(spa);
5375 spa->spa_spares.sav_sync = B_TRUE;
5379 } else if (spa->spa_l2cache.sav_vdevs != NULL &&
5380 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5381 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
5382 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
5384 * Cache devices can always be removed.
5386 spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
5387 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
5388 spa_load_l2cache(spa);
5389 spa->spa_l2cache.sav_sync = B_TRUE;
5390 } else if (vd != NULL && vd->vdev_islog) {
5392 ASSERT(vd == vd->vdev_top);
5395 * XXX - Once we have bp-rewrite this should
5396 * become the common case.
5402 * Stop allocating from this vdev.
5404 metaslab_group_passivate(mg);
5407 * Wait for the youngest allocations and frees to sync,
5408 * and then wait for the deferral of those frees to finish.
5410 spa_vdev_config_exit(spa, NULL,
5411 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
5414 * Attempt to evacuate the vdev.
5416 error = spa_vdev_remove_evacuate(spa, vd);
5418 txg = spa_vdev_config_enter(spa);
5421 * If we couldn't evacuate the vdev, unwind.
5424 metaslab_group_activate(mg);
5425 return (spa_vdev_exit(spa, NULL, txg, error));
5429 * Clean up the vdev namespace.
5431 spa_vdev_remove_from_namespace(spa, vd);
5433 } else if (vd != NULL) {
5435 * Normal vdevs cannot be removed (yet).
5440 * There is no vdev of any kind with the specified guid.
5446 return (spa_vdev_exit(spa, NULL, txg, error));
5452 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5453 * current spared, so we can detach it.
5456 spa_vdev_resilver_done_hunt(vdev_t *vd)
5458 vdev_t *newvd, *oldvd;
5460 for (int c = 0; c < vd->vdev_children; c++) {
5461 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
5467 * Check for a completed replacement. We always consider the first
5468 * vdev in the list to be the oldest vdev, and the last one to be
5469 * the newest (see spa_vdev_attach() for how that works). In
5470 * the case where the newest vdev is faulted, we will not automatically
5471 * remove it after a resilver completes. This is OK as it will require
5472 * user intervention to determine which disk the admin wishes to keep.
5474 if (vd->vdev_ops == &vdev_replacing_ops) {
5475 ASSERT(vd->vdev_children > 1);
5477 newvd = vd->vdev_child[vd->vdev_children - 1];
5478 oldvd = vd->vdev_child[0];
5480 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
5481 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5482 !vdev_dtl_required(oldvd))
5487 * Check for a completed resilver with the 'unspare' flag set.
5489 if (vd->vdev_ops == &vdev_spare_ops) {
5490 vdev_t *first = vd->vdev_child[0];
5491 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
5493 if (last->vdev_unspare) {
5496 } else if (first->vdev_unspare) {
5503 if (oldvd != NULL &&
5504 vdev_dtl_empty(newvd, DTL_MISSING) &&
5505 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5506 !vdev_dtl_required(oldvd))
5510 * If there are more than two spares attached to a disk,
5511 * and those spares are not required, then we want to
5512 * attempt to free them up now so that they can be used
5513 * by other pools. Once we're back down to a single
5514 * disk+spare, we stop removing them.
5516 if (vd->vdev_children > 2) {
5517 newvd = vd->vdev_child[1];
5519 if (newvd->vdev_isspare && last->vdev_isspare &&
5520 vdev_dtl_empty(last, DTL_MISSING) &&
5521 vdev_dtl_empty(last, DTL_OUTAGE) &&
5522 !vdev_dtl_required(newvd))
5531 spa_vdev_resilver_done(spa_t *spa)
5533 vdev_t *vd, *pvd, *ppvd;
5534 uint64_t guid, sguid, pguid, ppguid;
5536 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5538 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
5539 pvd = vd->vdev_parent;
5540 ppvd = pvd->vdev_parent;
5541 guid = vd->vdev_guid;
5542 pguid = pvd->vdev_guid;
5543 ppguid = ppvd->vdev_guid;
5546 * If we have just finished replacing a hot spared device, then
5547 * we need to detach the parent's first child (the original hot
5550 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
5551 ppvd->vdev_children == 2) {
5552 ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
5553 sguid = ppvd->vdev_child[1]->vdev_guid;
5555 spa_config_exit(spa, SCL_ALL, FTAG);
5556 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
5558 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
5560 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5563 spa_config_exit(spa, SCL_ALL, FTAG);
5567 * Update the stored path or FRU for this vdev.
5570 spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
5574 boolean_t sync = B_FALSE;
5576 ASSERT(spa_writeable(spa));
5578 spa_vdev_state_enter(spa, SCL_ALL);
5580 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
5581 return (spa_vdev_state_exit(spa, NULL, ENOENT));
5583 if (!vd->vdev_ops->vdev_op_leaf)
5584 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
5587 if (strcmp(value, vd->vdev_path) != 0) {
5588 spa_strfree(vd->vdev_path);
5589 vd->vdev_path = spa_strdup(value);
5593 if (vd->vdev_fru == NULL) {
5594 vd->vdev_fru = spa_strdup(value);
5596 } else if (strcmp(value, vd->vdev_fru) != 0) {
5597 spa_strfree(vd->vdev_fru);
5598 vd->vdev_fru = spa_strdup(value);
5603 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
5607 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
5609 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5613 spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5615 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5619 * ==========================================================================
5621 * ==========================================================================
5625 spa_scan_stop(spa_t *spa)
5627 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5628 if (dsl_scan_resilvering(spa->spa_dsl_pool))
5630 return (dsl_scan_cancel(spa->spa_dsl_pool));
5634 spa_scan(spa_t *spa, pool_scan_func_t func)
5636 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5638 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5642 * If a resilver was requested, but there is no DTL on a
5643 * writeable leaf device, we have nothing to do.
5645 if (func == POOL_SCAN_RESILVER &&
5646 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5647 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5651 return (dsl_scan(spa->spa_dsl_pool, func));
5655 * ==========================================================================
5656 * SPA async task processing
5657 * ==========================================================================
5661 spa_async_remove(spa_t *spa, vdev_t *vd)
5663 if (vd->vdev_remove_wanted) {
5664 vd->vdev_remove_wanted = B_FALSE;
5665 vd->vdev_delayed_close = B_FALSE;
5666 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5669 * We want to clear the stats, but we don't want to do a full
5670 * vdev_clear() as that will cause us to throw away
5671 * degraded/faulted state as well as attempt to reopen the
5672 * device, all of which is a waste.
5674 vd->vdev_stat.vs_read_errors = 0;
5675 vd->vdev_stat.vs_write_errors = 0;
5676 vd->vdev_stat.vs_checksum_errors = 0;
5678 vdev_state_dirty(vd->vdev_top);
5681 for (int c = 0; c < vd->vdev_children; c++)
5682 spa_async_remove(spa, vd->vdev_child[c]);
5686 spa_async_probe(spa_t *spa, vdev_t *vd)
5688 if (vd->vdev_probe_wanted) {
5689 vd->vdev_probe_wanted = B_FALSE;
5690 vdev_reopen(vd); /* vdev_open() does the actual probe */
5693 for (int c = 0; c < vd->vdev_children; c++)
5694 spa_async_probe(spa, vd->vdev_child[c]);
5698 spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5704 if (!spa->spa_autoexpand)
5707 for (int c = 0; c < vd->vdev_children; c++) {
5708 vdev_t *cvd = vd->vdev_child[c];
5709 spa_async_autoexpand(spa, cvd);
5712 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5715 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
5716 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
5718 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5719 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
5721 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
5722 ESC_ZFS_VDEV_AUTOEXPAND, attr, &eid, DDI_SLEEP);
5725 kmem_free(physpath, MAXPATHLEN);
5729 spa_async_thread(void *arg)
5734 ASSERT(spa->spa_sync_on);
5736 mutex_enter(&spa->spa_async_lock);
5737 tasks = spa->spa_async_tasks;
5738 spa->spa_async_tasks = 0;
5739 mutex_exit(&spa->spa_async_lock);
5742 * See if the config needs to be updated.
5744 if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5745 uint64_t old_space, new_space;
5747 mutex_enter(&spa_namespace_lock);
5748 old_space = metaslab_class_get_space(spa_normal_class(spa));
5749 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5750 new_space = metaslab_class_get_space(spa_normal_class(spa));
5751 mutex_exit(&spa_namespace_lock);
5754 * If the pool grew as a result of the config update,
5755 * then log an internal history event.
5757 if (new_space != old_space) {
5758 spa_history_log_internal(LOG_POOL_VDEV_ONLINE,
5760 "pool '%s' size: %llu(+%llu)",
5761 spa_name(spa), new_space, new_space - old_space);
5766 * See if any devices need to be marked REMOVED.
5768 if (tasks & SPA_ASYNC_REMOVE) {
5769 spa_vdev_state_enter(spa, SCL_NONE);
5770 spa_async_remove(spa, spa->spa_root_vdev);
5771 for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
5772 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5773 for (int i = 0; i < spa->spa_spares.sav_count; i++)
5774 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5775 (void) spa_vdev_state_exit(spa, NULL, 0);
5778 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5779 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5780 spa_async_autoexpand(spa, spa->spa_root_vdev);
5781 spa_config_exit(spa, SCL_CONFIG, FTAG);
5785 * See if any devices need to be probed.
5787 if (tasks & SPA_ASYNC_PROBE) {
5788 spa_vdev_state_enter(spa, SCL_NONE);
5789 spa_async_probe(spa, spa->spa_root_vdev);
5790 (void) spa_vdev_state_exit(spa, NULL, 0);
5794 * If any devices are done replacing, detach them.
5796 if (tasks & SPA_ASYNC_RESILVER_DONE)
5797 spa_vdev_resilver_done(spa);
5800 * Kick off a resilver.
5802 if (tasks & SPA_ASYNC_RESILVER)
5803 dsl_resilver_restart(spa->spa_dsl_pool, 0);
5806 * Let the world know that we're done.
5808 mutex_enter(&spa->spa_async_lock);
5809 spa->spa_async_thread = NULL;
5810 cv_broadcast(&spa->spa_async_cv);
5811 mutex_exit(&spa->spa_async_lock);
5816 spa_async_suspend(spa_t *spa)
5818 mutex_enter(&spa->spa_async_lock);
5819 spa->spa_async_suspended++;
5820 while (spa->spa_async_thread != NULL)
5821 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5822 mutex_exit(&spa->spa_async_lock);
5826 spa_async_resume(spa_t *spa)
5828 mutex_enter(&spa->spa_async_lock);
5829 ASSERT(spa->spa_async_suspended != 0);
5830 spa->spa_async_suspended--;
5831 mutex_exit(&spa->spa_async_lock);
5835 spa_async_dispatch(spa_t *spa)
5837 mutex_enter(&spa->spa_async_lock);
5838 if (spa->spa_async_tasks && !spa->spa_async_suspended &&
5839 spa->spa_async_thread == NULL &&
5840 rootdir != NULL && !vn_is_readonly(rootdir))
5841 spa->spa_async_thread = thread_create(NULL, 0,
5842 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
5843 mutex_exit(&spa->spa_async_lock);
5847 spa_async_request(spa_t *spa, int task)
5849 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
5850 mutex_enter(&spa->spa_async_lock);
5851 spa->spa_async_tasks |= task;
5852 mutex_exit(&spa->spa_async_lock);
5856 * ==========================================================================
5857 * SPA syncing routines
5858 * ==========================================================================
5862 bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5865 bpobj_enqueue(bpo, bp, tx);
5870 spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5874 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
5880 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
5882 char *packed = NULL;
5887 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
5890 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5891 * information. This avoids the dbuf_will_dirty() path and
5892 * saves us a pre-read to get data we don't actually care about.
5894 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
5895 packed = kmem_alloc(bufsize, KM_SLEEP);
5897 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
5899 bzero(packed + nvsize, bufsize - nvsize);
5901 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
5903 kmem_free(packed, bufsize);
5905 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
5906 dmu_buf_will_dirty(db, tx);
5907 *(uint64_t *)db->db_data = nvsize;
5908 dmu_buf_rele(db, FTAG);
5912 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
5913 const char *config, const char *entry)
5923 * Update the MOS nvlist describing the list of available devices.
5924 * spa_validate_aux() will have already made sure this nvlist is
5925 * valid and the vdevs are labeled appropriately.
5927 if (sav->sav_object == 0) {
5928 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
5929 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
5930 sizeof (uint64_t), tx);
5931 VERIFY(zap_update(spa->spa_meta_objset,
5932 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
5933 &sav->sav_object, tx) == 0);
5936 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5937 if (sav->sav_count == 0) {
5938 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
5940 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
5941 for (i = 0; i < sav->sav_count; i++)
5942 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
5943 B_FALSE, VDEV_CONFIG_L2CACHE);
5944 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
5945 sav->sav_count) == 0);
5946 for (i = 0; i < sav->sav_count; i++)
5947 nvlist_free(list[i]);
5948 kmem_free(list, sav->sav_count * sizeof (void *));
5951 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
5952 nvlist_free(nvroot);
5954 sav->sav_sync = B_FALSE;
5958 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
5962 if (list_is_empty(&spa->spa_config_dirty_list))
5965 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5967 config = spa_config_generate(spa, spa->spa_root_vdev,
5968 dmu_tx_get_txg(tx), B_FALSE);
5971 * If we're upgrading the spa version then make sure that
5972 * the config object gets updated with the correct version.
5974 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
5975 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5976 spa->spa_uberblock.ub_version);
5978 spa_config_exit(spa, SCL_STATE, FTAG);
5980 if (spa->spa_config_syncing)
5981 nvlist_free(spa->spa_config_syncing);
5982 spa->spa_config_syncing = config;
5984 spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
5988 spa_sync_version(void *arg1, void *arg2, dmu_tx_t *tx)
5991 uint64_t version = *(uint64_t *)arg2;
5994 * Setting the version is special cased when first creating the pool.
5996 ASSERT(tx->tx_txg != TXG_INITIAL);
5998 ASSERT(version <= SPA_VERSION);
5999 ASSERT(version >= spa_version(spa));
6001 spa->spa_uberblock.ub_version = version;
6002 vdev_config_dirty(spa->spa_root_vdev);
6006 * Set zpool properties.
6009 spa_sync_props(void *arg1, void *arg2, dmu_tx_t *tx)
6012 objset_t *mos = spa->spa_meta_objset;
6013 nvlist_t *nvp = arg2;
6014 nvpair_t *elem = NULL;
6016 mutex_enter(&spa->spa_props_lock);
6018 while ((elem = nvlist_next_nvpair(nvp, elem))) {
6020 char *strval, *fname;
6022 const char *propname;
6023 zprop_type_t proptype;
6024 zfeature_info_t *feature;
6026 switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
6029 * We checked this earlier in spa_prop_validate().
6031 ASSERT(zpool_prop_feature(nvpair_name(elem)));
6033 fname = strchr(nvpair_name(elem), '@') + 1;
6034 VERIFY3U(0, ==, zfeature_lookup_name(fname, &feature));
6036 spa_feature_enable(spa, feature, tx);
6039 case ZPOOL_PROP_VERSION:
6040 VERIFY(nvpair_value_uint64(elem, &intval) == 0);
6042 * The version is synced seperatly before other
6043 * properties and should be correct by now.
6045 ASSERT3U(spa_version(spa), >=, intval);
6048 case ZPOOL_PROP_ALTROOT:
6050 * 'altroot' is a non-persistent property. It should
6051 * have been set temporarily at creation or import time.
6053 ASSERT(spa->spa_root != NULL);
6056 case ZPOOL_PROP_READONLY:
6057 case ZPOOL_PROP_CACHEFILE:
6059 * 'readonly' and 'cachefile' are also non-persisitent
6063 case ZPOOL_PROP_COMMENT:
6064 VERIFY(nvpair_value_string(elem, &strval) == 0);
6065 if (spa->spa_comment != NULL)
6066 spa_strfree(spa->spa_comment);
6067 spa->spa_comment = spa_strdup(strval);
6069 * We need to dirty the configuration on all the vdevs
6070 * so that their labels get updated. It's unnecessary
6071 * to do this for pool creation since the vdev's
6072 * configuratoin has already been dirtied.
6074 if (tx->tx_txg != TXG_INITIAL)
6075 vdev_config_dirty(spa->spa_root_vdev);
6079 * Set pool property values in the poolprops mos object.
6081 if (spa->spa_pool_props_object == 0) {
6082 spa->spa_pool_props_object =
6083 zap_create_link(mos, DMU_OT_POOL_PROPS,
6084 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
6088 /* normalize the property name */
6089 propname = zpool_prop_to_name(prop);
6090 proptype = zpool_prop_get_type(prop);
6092 if (nvpair_type(elem) == DATA_TYPE_STRING) {
6093 ASSERT(proptype == PROP_TYPE_STRING);
6094 VERIFY(nvpair_value_string(elem, &strval) == 0);
6095 VERIFY(zap_update(mos,
6096 spa->spa_pool_props_object, propname,
6097 1, strlen(strval) + 1, strval, tx) == 0);
6099 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
6100 VERIFY(nvpair_value_uint64(elem, &intval) == 0);
6102 if (proptype == PROP_TYPE_INDEX) {
6104 VERIFY(zpool_prop_index_to_string(
6105 prop, intval, &unused) == 0);
6107 VERIFY(zap_update(mos,
6108 spa->spa_pool_props_object, propname,
6109 8, 1, &intval, tx) == 0);
6111 ASSERT(0); /* not allowed */
6115 case ZPOOL_PROP_DELEGATION:
6116 spa->spa_delegation = intval;
6118 case ZPOOL_PROP_BOOTFS:
6119 spa->spa_bootfs = intval;
6121 case ZPOOL_PROP_FAILUREMODE:
6122 spa->spa_failmode = intval;
6124 case ZPOOL_PROP_AUTOEXPAND:
6125 spa->spa_autoexpand = intval;
6126 if (tx->tx_txg != TXG_INITIAL)
6127 spa_async_request(spa,
6128 SPA_ASYNC_AUTOEXPAND);
6130 case ZPOOL_PROP_DEDUPDITTO:
6131 spa->spa_dedup_ditto = intval;
6138 /* log internal history if this is not a zpool create */
6139 if (spa_version(spa) >= SPA_VERSION_ZPOOL_HISTORY &&
6140 tx->tx_txg != TXG_INITIAL) {
6141 spa_history_log_internal(LOG_POOL_PROPSET,
6142 spa, tx, "%s %lld %s",
6143 nvpair_name(elem), intval, spa_name(spa));
6147 mutex_exit(&spa->spa_props_lock);
6151 * Perform one-time upgrade on-disk changes. spa_version() does not
6152 * reflect the new version this txg, so there must be no changes this
6153 * txg to anything that the upgrade code depends on after it executes.
6154 * Therefore this must be called after dsl_pool_sync() does the sync
6158 spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
6160 dsl_pool_t *dp = spa->spa_dsl_pool;
6162 ASSERT(spa->spa_sync_pass == 1);
6164 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
6165 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
6166 dsl_pool_create_origin(dp, tx);
6168 /* Keeping the origin open increases spa_minref */
6169 spa->spa_minref += 3;
6172 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
6173 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
6174 dsl_pool_upgrade_clones(dp, tx);
6177 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
6178 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
6179 dsl_pool_upgrade_dir_clones(dp, tx);
6181 /* Keeping the freedir open increases spa_minref */
6182 spa->spa_minref += 3;
6185 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES &&
6186 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
6187 spa_feature_create_zap_objects(spa, tx);
6192 * Sync the specified transaction group. New blocks may be dirtied as
6193 * part of the process, so we iterate until it converges.
6196 spa_sync(spa_t *spa, uint64_t txg)
6198 dsl_pool_t *dp = spa->spa_dsl_pool;
6199 objset_t *mos = spa->spa_meta_objset;
6200 bpobj_t *defer_bpo = &spa->spa_deferred_bpobj;
6201 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
6202 vdev_t *rvd = spa->spa_root_vdev;
6207 VERIFY(spa_writeable(spa));
6210 * Lock out configuration changes.
6212 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
6214 spa->spa_syncing_txg = txg;
6215 spa->spa_sync_pass = 0;
6218 * If there are any pending vdev state changes, convert them
6219 * into config changes that go out with this transaction group.
6221 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6222 while (list_head(&spa->spa_state_dirty_list) != NULL) {
6224 * We need the write lock here because, for aux vdevs,
6225 * calling vdev_config_dirty() modifies sav_config.
6226 * This is ugly and will become unnecessary when we
6227 * eliminate the aux vdev wart by integrating all vdevs
6228 * into the root vdev tree.
6230 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6231 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
6232 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
6233 vdev_state_clean(vd);
6234 vdev_config_dirty(vd);
6236 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6237 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
6239 spa_config_exit(spa, SCL_STATE, FTAG);
6241 tx = dmu_tx_create_assigned(dp, txg);
6244 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
6245 * set spa_deflate if we have no raid-z vdevs.
6247 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
6248 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
6251 for (i = 0; i < rvd->vdev_children; i++) {
6252 vd = rvd->vdev_child[i];
6253 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
6256 if (i == rvd->vdev_children) {
6257 spa->spa_deflate = TRUE;
6258 VERIFY(0 == zap_add(spa->spa_meta_objset,
6259 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
6260 sizeof (uint64_t), 1, &spa->spa_deflate, tx));
6265 * If anything has changed in this txg, or if someone is waiting
6266 * for this txg to sync (eg, spa_vdev_remove()), push the
6267 * deferred frees from the previous txg. If not, leave them
6268 * alone so that we don't generate work on an otherwise idle
6271 if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
6272 !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
6273 !txg_list_empty(&dp->dp_sync_tasks, txg) ||
6274 ((dsl_scan_active(dp->dp_scan) ||
6275 txg_sync_waiting(dp)) && !spa_shutting_down(spa))) {
6276 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6277 VERIFY3U(bpobj_iterate(defer_bpo,
6278 spa_free_sync_cb, zio, tx), ==, 0);
6279 VERIFY0(zio_wait(zio));
6283 * Iterate to convergence.
6286 int pass = ++spa->spa_sync_pass;
6288 spa_sync_config_object(spa, tx);
6289 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
6290 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
6291 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
6292 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
6293 spa_errlog_sync(spa, txg);
6294 dsl_pool_sync(dp, txg);
6296 if (pass < zfs_sync_pass_deferred_free) {
6297 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6298 bplist_iterate(free_bpl, spa_free_sync_cb,
6300 VERIFY(zio_wait(zio) == 0);
6302 bplist_iterate(free_bpl, bpobj_enqueue_cb,
6307 dsl_scan_sync(dp, tx);
6309 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
6313 spa_sync_upgrades(spa, tx);
6315 } while (dmu_objset_is_dirty(mos, txg));
6318 * Rewrite the vdev configuration (which includes the uberblock)
6319 * to commit the transaction group.
6321 * If there are no dirty vdevs, we sync the uberblock to a few
6322 * random top-level vdevs that are known to be visible in the
6323 * config cache (see spa_vdev_add() for a complete description).
6324 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
6328 * We hold SCL_STATE to prevent vdev open/close/etc.
6329 * while we're attempting to write the vdev labels.
6331 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6333 if (list_is_empty(&spa->spa_config_dirty_list)) {
6334 vdev_t *svd[SPA_DVAS_PER_BP];
6336 int children = rvd->vdev_children;
6337 int c0 = spa_get_random(children);
6339 for (int c = 0; c < children; c++) {
6340 vd = rvd->vdev_child[(c0 + c) % children];
6341 if (vd->vdev_ms_array == 0 || vd->vdev_islog)
6343 svd[svdcount++] = vd;
6344 if (svdcount == SPA_DVAS_PER_BP)
6347 error = vdev_config_sync(svd, svdcount, txg, B_FALSE);
6349 error = vdev_config_sync(svd, svdcount, txg,
6352 error = vdev_config_sync(rvd->vdev_child,
6353 rvd->vdev_children, txg, B_FALSE);
6355 error = vdev_config_sync(rvd->vdev_child,
6356 rvd->vdev_children, txg, B_TRUE);
6360 spa->spa_last_synced_guid = rvd->vdev_guid;
6362 spa_config_exit(spa, SCL_STATE, FTAG);
6366 zio_suspend(spa, NULL);
6367 zio_resume_wait(spa);
6372 * Clear the dirty config list.
6374 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
6375 vdev_config_clean(vd);
6378 * Now that the new config has synced transactionally,
6379 * let it become visible to the config cache.
6381 if (spa->spa_config_syncing != NULL) {
6382 spa_config_set(spa, spa->spa_config_syncing);
6383 spa->spa_config_txg = txg;
6384 spa->spa_config_syncing = NULL;
6387 spa->spa_ubsync = spa->spa_uberblock;
6389 dsl_pool_sync_done(dp, txg);
6392 * Update usable space statistics.
6394 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
6395 vdev_sync_done(vd, txg);
6397 spa_update_dspace(spa);
6400 * It had better be the case that we didn't dirty anything
6401 * since vdev_config_sync().
6403 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
6404 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
6405 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
6407 spa->spa_sync_pass = 0;
6409 spa_config_exit(spa, SCL_CONFIG, FTAG);
6411 spa_handle_ignored_writes(spa);
6414 * If any async tasks have been requested, kick them off.
6416 spa_async_dispatch(spa);
6420 * Sync all pools. We don't want to hold the namespace lock across these
6421 * operations, so we take a reference on the spa_t and drop the lock during the
6425 spa_sync_allpools(void)
6428 mutex_enter(&spa_namespace_lock);
6429 while ((spa = spa_next(spa)) != NULL) {
6430 if (spa_state(spa) != POOL_STATE_ACTIVE ||
6431 !spa_writeable(spa) || spa_suspended(spa))
6433 spa_open_ref(spa, FTAG);
6434 mutex_exit(&spa_namespace_lock);
6435 txg_wait_synced(spa_get_dsl(spa), 0);
6436 mutex_enter(&spa_namespace_lock);
6437 spa_close(spa, FTAG);
6439 mutex_exit(&spa_namespace_lock);
6443 * ==========================================================================
6444 * Miscellaneous routines
6445 * ==========================================================================
6449 * Remove all pools in the system.
6457 * Remove all cached state. All pools should be closed now,
6458 * so every spa in the AVL tree should be unreferenced.
6460 mutex_enter(&spa_namespace_lock);
6461 while ((spa = spa_next(NULL)) != NULL) {
6463 * Stop async tasks. The async thread may need to detach
6464 * a device that's been replaced, which requires grabbing
6465 * spa_namespace_lock, so we must drop it here.
6467 spa_open_ref(spa, FTAG);
6468 mutex_exit(&spa_namespace_lock);
6469 spa_async_suspend(spa);
6470 mutex_enter(&spa_namespace_lock);
6471 spa_close(spa, FTAG);
6473 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
6475 spa_deactivate(spa);
6479 mutex_exit(&spa_namespace_lock);
6483 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
6488 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
6492 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
6493 vd = spa->spa_l2cache.sav_vdevs[i];
6494 if (vd->vdev_guid == guid)
6498 for (i = 0; i < spa->spa_spares.sav_count; i++) {
6499 vd = spa->spa_spares.sav_vdevs[i];
6500 if (vd->vdev_guid == guid)
6509 spa_upgrade(spa_t *spa, uint64_t version)
6511 ASSERT(spa_writeable(spa));
6513 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6516 * This should only be called for a non-faulted pool, and since a
6517 * future version would result in an unopenable pool, this shouldn't be
6520 ASSERT(spa->spa_uberblock.ub_version <= SPA_VERSION);
6521 ASSERT(version >= spa->spa_uberblock.ub_version);
6523 spa->spa_uberblock.ub_version = version;
6524 vdev_config_dirty(spa->spa_root_vdev);
6526 spa_config_exit(spa, SCL_ALL, FTAG);
6528 txg_wait_synced(spa_get_dsl(spa), 0);
6532 spa_has_spare(spa_t *spa, uint64_t guid)
6536 spa_aux_vdev_t *sav = &spa->spa_spares;
6538 for (i = 0; i < sav->sav_count; i++)
6539 if (sav->sav_vdevs[i]->vdev_guid == guid)
6542 for (i = 0; i < sav->sav_npending; i++) {
6543 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
6544 &spareguid) == 0 && spareguid == guid)
6552 * Check if a pool has an active shared spare device.
6553 * Note: reference count of an active spare is 2, as a spare and as a replace
6556 spa_has_active_shared_spare(spa_t *spa)
6560 spa_aux_vdev_t *sav = &spa->spa_spares;
6562 for (i = 0; i < sav->sav_count; i++) {
6563 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
6564 &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
6573 * Post a sysevent corresponding to the given event. The 'name' must be one of
6574 * the event definitions in sys/sysevent/eventdefs.h. The payload will be
6575 * filled in from the spa and (optionally) the vdev. This doesn't do anything
6576 * in the userland libzpool, as we don't want consumers to misinterpret ztest
6577 * or zdb as real changes.
6580 spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
6584 sysevent_attr_list_t *attr = NULL;
6585 sysevent_value_t value;
6588 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
6591 value.value_type = SE_DATA_TYPE_STRING;
6592 value.value.sv_string = spa_name(spa);
6593 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
6596 value.value_type = SE_DATA_TYPE_UINT64;
6597 value.value.sv_uint64 = spa_guid(spa);
6598 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
6602 value.value_type = SE_DATA_TYPE_UINT64;
6603 value.value.sv_uint64 = vd->vdev_guid;
6604 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
6608 if (vd->vdev_path) {
6609 value.value_type = SE_DATA_TYPE_STRING;
6610 value.value.sv_string = vd->vdev_path;
6611 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
6612 &value, SE_SLEEP) != 0)
6617 if (sysevent_attach_attributes(ev, attr) != 0)
6621 (void) log_sysevent(ev, SE_SLEEP, &eid);
6625 sysevent_free_attr(attr);