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>
65 #include <sys/trim_map.h>
68 #include <sys/callb.h>
69 #include <sys/cpupart.h>
74 #include "zfs_comutil.h"
76 /* Check hostid on import? */
77 static int check_hostid = 1;
79 SYSCTL_DECL(_vfs_zfs);
80 TUNABLE_INT("vfs.zfs.check_hostid", &check_hostid);
81 SYSCTL_INT(_vfs_zfs, OID_AUTO, check_hostid, CTLFLAG_RW, &check_hostid, 0,
82 "Check hostid on import?");
84 typedef enum zti_modes {
85 zti_mode_fixed, /* value is # of threads (min 1) */
86 zti_mode_online_percent, /* value is % of online CPUs */
87 zti_mode_batch, /* cpu-intensive; value is ignored */
88 zti_mode_null, /* don't create a taskq */
92 #define ZTI_FIX(n) { zti_mode_fixed, (n) }
93 #define ZTI_PCT(n) { zti_mode_online_percent, (n) }
94 #define ZTI_BATCH { zti_mode_batch, 0 }
95 #define ZTI_NULL { zti_mode_null, 0 }
97 #define ZTI_ONE ZTI_FIX(1)
99 typedef struct zio_taskq_info {
100 enum zti_modes zti_mode;
104 static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
105 "issue", "issue_high", "intr", "intr_high"
109 * Define the taskq threads for the following I/O types:
110 * NULL, READ, WRITE, FREE, CLAIM, and IOCTL
112 const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
113 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
114 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL },
115 { ZTI_FIX(8), ZTI_NULL, ZTI_BATCH, ZTI_NULL },
116 { ZTI_BATCH, ZTI_FIX(5), ZTI_FIX(8), ZTI_FIX(5) },
117 { ZTI_FIX(100), ZTI_NULL, ZTI_ONE, ZTI_NULL },
118 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL },
119 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL },
122 static dsl_syncfunc_t spa_sync_version;
123 static dsl_syncfunc_t spa_sync_props;
124 static dsl_checkfunc_t spa_change_guid_check;
125 static dsl_syncfunc_t spa_change_guid_sync;
126 static boolean_t spa_has_active_shared_spare(spa_t *spa);
127 static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
128 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
130 static void spa_vdev_resilver_done(spa_t *spa);
132 uint_t zio_taskq_batch_pct = 100; /* 1 thread per cpu in pset */
134 id_t zio_taskq_psrset_bind = PS_NONE;
137 boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */
139 uint_t zio_taskq_basedc = 80; /* base duty cycle */
141 boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */
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);
1006 * Start TRIM thread.
1008 trim_thread_create(spa);
1010 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
1011 offsetof(vdev_t, vdev_config_dirty_node));
1012 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
1013 offsetof(vdev_t, vdev_state_dirty_node));
1015 txg_list_create(&spa->spa_vdev_txg_list,
1016 offsetof(struct vdev, vdev_txg_node));
1018 avl_create(&spa->spa_errlist_scrub,
1019 spa_error_entry_compare, sizeof (spa_error_entry_t),
1020 offsetof(spa_error_entry_t, se_avl));
1021 avl_create(&spa->spa_errlist_last,
1022 spa_error_entry_compare, sizeof (spa_error_entry_t),
1023 offsetof(spa_error_entry_t, se_avl));
1027 * Opposite of spa_activate().
1030 spa_deactivate(spa_t *spa)
1032 ASSERT(spa->spa_sync_on == B_FALSE);
1033 ASSERT(spa->spa_dsl_pool == NULL);
1034 ASSERT(spa->spa_root_vdev == NULL);
1035 ASSERT(spa->spa_async_zio_root == NULL);
1036 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
1039 * Stop TRIM thread in case spa_unload() wasn't called directly
1040 * before spa_deactivate().
1042 trim_thread_destroy(spa);
1044 txg_list_destroy(&spa->spa_vdev_txg_list);
1046 list_destroy(&spa->spa_config_dirty_list);
1047 list_destroy(&spa->spa_state_dirty_list);
1049 for (int t = 0; t < ZIO_TYPES; t++) {
1050 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1051 if (spa->spa_zio_taskq[t][q] != NULL)
1052 taskq_destroy(spa->spa_zio_taskq[t][q]);
1053 spa->spa_zio_taskq[t][q] = NULL;
1057 metaslab_class_destroy(spa->spa_normal_class);
1058 spa->spa_normal_class = NULL;
1060 metaslab_class_destroy(spa->spa_log_class);
1061 spa->spa_log_class = NULL;
1064 * If this was part of an import or the open otherwise failed, we may
1065 * still have errors left in the queues. Empty them just in case.
1067 spa_errlog_drain(spa);
1069 avl_destroy(&spa->spa_errlist_scrub);
1070 avl_destroy(&spa->spa_errlist_last);
1072 spa->spa_state = POOL_STATE_UNINITIALIZED;
1074 mutex_enter(&spa->spa_proc_lock);
1075 if (spa->spa_proc_state != SPA_PROC_NONE) {
1076 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1077 spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1078 cv_broadcast(&spa->spa_proc_cv);
1079 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1080 ASSERT(spa->spa_proc != &p0);
1081 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1083 ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1084 spa->spa_proc_state = SPA_PROC_NONE;
1086 ASSERT(spa->spa_proc == &p0);
1087 mutex_exit(&spa->spa_proc_lock);
1091 * We want to make sure spa_thread() has actually exited the ZFS
1092 * module, so that the module can't be unloaded out from underneath
1095 if (spa->spa_did != 0) {
1096 thread_join(spa->spa_did);
1099 #endif /* SPA_PROCESS */
1103 * Verify a pool configuration, and construct the vdev tree appropriately. This
1104 * will create all the necessary vdevs in the appropriate layout, with each vdev
1105 * in the CLOSED state. This will prep the pool before open/creation/import.
1106 * All vdev validation is done by the vdev_alloc() routine.
1109 spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1110 uint_t id, int atype)
1116 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1119 if ((*vdp)->vdev_ops->vdev_op_leaf)
1122 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1125 if (error == ENOENT)
1134 for (int c = 0; c < children; c++) {
1136 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1144 ASSERT(*vdp != NULL);
1150 * Opposite of spa_load().
1153 spa_unload(spa_t *spa)
1157 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1162 trim_thread_destroy(spa);
1167 spa_async_suspend(spa);
1172 if (spa->spa_sync_on) {
1173 txg_sync_stop(spa->spa_dsl_pool);
1174 spa->spa_sync_on = B_FALSE;
1178 * Wait for any outstanding async I/O to complete.
1180 if (spa->spa_async_zio_root != NULL) {
1181 (void) zio_wait(spa->spa_async_zio_root);
1182 spa->spa_async_zio_root = NULL;
1185 bpobj_close(&spa->spa_deferred_bpobj);
1188 * Close the dsl pool.
1190 if (spa->spa_dsl_pool) {
1191 dsl_pool_close(spa->spa_dsl_pool);
1192 spa->spa_dsl_pool = NULL;
1193 spa->spa_meta_objset = NULL;
1198 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1201 * Drop and purge level 2 cache
1203 spa_l2cache_drop(spa);
1208 if (spa->spa_root_vdev)
1209 vdev_free(spa->spa_root_vdev);
1210 ASSERT(spa->spa_root_vdev == NULL);
1212 for (i = 0; i < spa->spa_spares.sav_count; i++)
1213 vdev_free(spa->spa_spares.sav_vdevs[i]);
1214 if (spa->spa_spares.sav_vdevs) {
1215 kmem_free(spa->spa_spares.sav_vdevs,
1216 spa->spa_spares.sav_count * sizeof (void *));
1217 spa->spa_spares.sav_vdevs = NULL;
1219 if (spa->spa_spares.sav_config) {
1220 nvlist_free(spa->spa_spares.sav_config);
1221 spa->spa_spares.sav_config = NULL;
1223 spa->spa_spares.sav_count = 0;
1225 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1226 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1227 vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1229 if (spa->spa_l2cache.sav_vdevs) {
1230 kmem_free(spa->spa_l2cache.sav_vdevs,
1231 spa->spa_l2cache.sav_count * sizeof (void *));
1232 spa->spa_l2cache.sav_vdevs = NULL;
1234 if (spa->spa_l2cache.sav_config) {
1235 nvlist_free(spa->spa_l2cache.sav_config);
1236 spa->spa_l2cache.sav_config = NULL;
1238 spa->spa_l2cache.sav_count = 0;
1240 spa->spa_async_suspended = 0;
1242 if (spa->spa_comment != NULL) {
1243 spa_strfree(spa->spa_comment);
1244 spa->spa_comment = NULL;
1247 spa_config_exit(spa, SCL_ALL, FTAG);
1251 * Load (or re-load) the current list of vdevs describing the active spares for
1252 * this pool. When this is called, we have some form of basic information in
1253 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1254 * then re-generate a more complete list including status information.
1257 spa_load_spares(spa_t *spa)
1264 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1267 * First, close and free any existing spare vdevs.
1269 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1270 vd = spa->spa_spares.sav_vdevs[i];
1272 /* Undo the call to spa_activate() below */
1273 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1274 B_FALSE)) != NULL && tvd->vdev_isspare)
1275 spa_spare_remove(tvd);
1280 if (spa->spa_spares.sav_vdevs)
1281 kmem_free(spa->spa_spares.sav_vdevs,
1282 spa->spa_spares.sav_count * sizeof (void *));
1284 if (spa->spa_spares.sav_config == NULL)
1287 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1288 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1290 spa->spa_spares.sav_count = (int)nspares;
1291 spa->spa_spares.sav_vdevs = NULL;
1297 * Construct the array of vdevs, opening them to get status in the
1298 * process. For each spare, there is potentially two different vdev_t
1299 * structures associated with it: one in the list of spares (used only
1300 * for basic validation purposes) and one in the active vdev
1301 * configuration (if it's spared in). During this phase we open and
1302 * validate each vdev on the spare list. If the vdev also exists in the
1303 * active configuration, then we also mark this vdev as an active spare.
1305 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1307 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1308 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1309 VDEV_ALLOC_SPARE) == 0);
1312 spa->spa_spares.sav_vdevs[i] = vd;
1314 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1315 B_FALSE)) != NULL) {
1316 if (!tvd->vdev_isspare)
1320 * We only mark the spare active if we were successfully
1321 * able to load the vdev. Otherwise, importing a pool
1322 * with a bad active spare would result in strange
1323 * behavior, because multiple pool would think the spare
1324 * is actively in use.
1326 * There is a vulnerability here to an equally bizarre
1327 * circumstance, where a dead active spare is later
1328 * brought back to life (onlined or otherwise). Given
1329 * the rarity of this scenario, and the extra complexity
1330 * it adds, we ignore the possibility.
1332 if (!vdev_is_dead(tvd))
1333 spa_spare_activate(tvd);
1337 vd->vdev_aux = &spa->spa_spares;
1339 if (vdev_open(vd) != 0)
1342 if (vdev_validate_aux(vd) == 0)
1347 * Recompute the stashed list of spares, with status information
1350 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1351 DATA_TYPE_NVLIST_ARRAY) == 0);
1353 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1355 for (i = 0; i < spa->spa_spares.sav_count; i++)
1356 spares[i] = vdev_config_generate(spa,
1357 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1358 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1359 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1360 for (i = 0; i < spa->spa_spares.sav_count; i++)
1361 nvlist_free(spares[i]);
1362 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1366 * Load (or re-load) the current list of vdevs describing the active l2cache for
1367 * this pool. When this is called, we have some form of basic information in
1368 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1369 * then re-generate a more complete list including status information.
1370 * Devices which are already active have their details maintained, and are
1374 spa_load_l2cache(spa_t *spa)
1378 int i, j, oldnvdevs;
1380 vdev_t *vd, **oldvdevs, **newvdevs;
1381 spa_aux_vdev_t *sav = &spa->spa_l2cache;
1383 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1385 if (sav->sav_config != NULL) {
1386 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1387 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1388 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1393 oldvdevs = sav->sav_vdevs;
1394 oldnvdevs = sav->sav_count;
1395 sav->sav_vdevs = NULL;
1399 * Process new nvlist of vdevs.
1401 for (i = 0; i < nl2cache; i++) {
1402 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1406 for (j = 0; j < oldnvdevs; j++) {
1408 if (vd != NULL && guid == vd->vdev_guid) {
1410 * Retain previous vdev for add/remove ops.
1418 if (newvdevs[i] == NULL) {
1422 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1423 VDEV_ALLOC_L2CACHE) == 0);
1428 * Commit this vdev as an l2cache device,
1429 * even if it fails to open.
1431 spa_l2cache_add(vd);
1436 spa_l2cache_activate(vd);
1438 if (vdev_open(vd) != 0)
1441 (void) vdev_validate_aux(vd);
1443 if (!vdev_is_dead(vd))
1444 l2arc_add_vdev(spa, vd);
1449 * Purge vdevs that were dropped
1451 for (i = 0; i < oldnvdevs; i++) {
1456 ASSERT(vd->vdev_isl2cache);
1458 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1459 pool != 0ULL && l2arc_vdev_present(vd))
1460 l2arc_remove_vdev(vd);
1461 vdev_clear_stats(vd);
1467 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1469 if (sav->sav_config == NULL)
1472 sav->sav_vdevs = newvdevs;
1473 sav->sav_count = (int)nl2cache;
1476 * Recompute the stashed list of l2cache devices, with status
1477 * information this time.
1479 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1480 DATA_TYPE_NVLIST_ARRAY) == 0);
1482 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1483 for (i = 0; i < sav->sav_count; i++)
1484 l2cache[i] = vdev_config_generate(spa,
1485 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1486 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1487 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1489 for (i = 0; i < sav->sav_count; i++)
1490 nvlist_free(l2cache[i]);
1492 kmem_free(l2cache, sav->sav_count * sizeof (void *));
1496 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1499 char *packed = NULL;
1504 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
1505 nvsize = *(uint64_t *)db->db_data;
1506 dmu_buf_rele(db, FTAG);
1508 packed = kmem_alloc(nvsize, KM_SLEEP);
1509 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1512 error = nvlist_unpack(packed, nvsize, value, 0);
1513 kmem_free(packed, nvsize);
1519 * Checks to see if the given vdev could not be opened, in which case we post a
1520 * sysevent to notify the autoreplace code that the device has been removed.
1523 spa_check_removed(vdev_t *vd)
1525 for (int c = 0; c < vd->vdev_children; c++)
1526 spa_check_removed(vd->vdev_child[c]);
1528 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd)) {
1529 zfs_post_autoreplace(vd->vdev_spa, vd);
1530 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1535 * Validate the current config against the MOS config
1538 spa_config_valid(spa_t *spa, nvlist_t *config)
1540 vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1543 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1545 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1546 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1548 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1551 * If we're doing a normal import, then build up any additional
1552 * diagnostic information about missing devices in this config.
1553 * We'll pass this up to the user for further processing.
1555 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1556 nvlist_t **child, *nv;
1559 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1561 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1563 for (int c = 0; c < rvd->vdev_children; c++) {
1564 vdev_t *tvd = rvd->vdev_child[c];
1565 vdev_t *mtvd = mrvd->vdev_child[c];
1567 if (tvd->vdev_ops == &vdev_missing_ops &&
1568 mtvd->vdev_ops != &vdev_missing_ops &&
1570 child[idx++] = vdev_config_generate(spa, mtvd,
1575 VERIFY(nvlist_add_nvlist_array(nv,
1576 ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1577 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1578 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1580 for (int i = 0; i < idx; i++)
1581 nvlist_free(child[i]);
1584 kmem_free(child, rvd->vdev_children * sizeof (char **));
1588 * Compare the root vdev tree with the information we have
1589 * from the MOS config (mrvd). Check each top-level vdev
1590 * with the corresponding MOS config top-level (mtvd).
1592 for (int c = 0; c < rvd->vdev_children; c++) {
1593 vdev_t *tvd = rvd->vdev_child[c];
1594 vdev_t *mtvd = mrvd->vdev_child[c];
1597 * Resolve any "missing" vdevs in the current configuration.
1598 * If we find that the MOS config has more accurate information
1599 * about the top-level vdev then use that vdev instead.
1601 if (tvd->vdev_ops == &vdev_missing_ops &&
1602 mtvd->vdev_ops != &vdev_missing_ops) {
1604 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1608 * Device specific actions.
1610 if (mtvd->vdev_islog) {
1611 spa_set_log_state(spa, SPA_LOG_CLEAR);
1614 * XXX - once we have 'readonly' pool
1615 * support we should be able to handle
1616 * missing data devices by transitioning
1617 * the pool to readonly.
1623 * Swap the missing vdev with the data we were
1624 * able to obtain from the MOS config.
1626 vdev_remove_child(rvd, tvd);
1627 vdev_remove_child(mrvd, mtvd);
1629 vdev_add_child(rvd, mtvd);
1630 vdev_add_child(mrvd, tvd);
1632 spa_config_exit(spa, SCL_ALL, FTAG);
1634 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1637 } else if (mtvd->vdev_islog) {
1639 * Load the slog device's state from the MOS config
1640 * since it's possible that the label does not
1641 * contain the most up-to-date information.
1643 vdev_load_log_state(tvd, mtvd);
1648 spa_config_exit(spa, SCL_ALL, FTAG);
1651 * Ensure we were able to validate the config.
1653 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1657 * Check for missing log devices
1660 spa_check_logs(spa_t *spa)
1662 switch (spa->spa_log_state) {
1663 case SPA_LOG_MISSING:
1664 /* need to recheck in case slog has been restored */
1665 case SPA_LOG_UNKNOWN:
1666 if (dmu_objset_find(spa->spa_name, zil_check_log_chain, NULL,
1667 DS_FIND_CHILDREN)) {
1668 spa_set_log_state(spa, SPA_LOG_MISSING);
1677 spa_passivate_log(spa_t *spa)
1679 vdev_t *rvd = spa->spa_root_vdev;
1680 boolean_t slog_found = B_FALSE;
1682 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1684 if (!spa_has_slogs(spa))
1687 for (int c = 0; c < rvd->vdev_children; c++) {
1688 vdev_t *tvd = rvd->vdev_child[c];
1689 metaslab_group_t *mg = tvd->vdev_mg;
1691 if (tvd->vdev_islog) {
1692 metaslab_group_passivate(mg);
1693 slog_found = B_TRUE;
1697 return (slog_found);
1701 spa_activate_log(spa_t *spa)
1703 vdev_t *rvd = spa->spa_root_vdev;
1705 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1707 for (int c = 0; c < rvd->vdev_children; c++) {
1708 vdev_t *tvd = rvd->vdev_child[c];
1709 metaslab_group_t *mg = tvd->vdev_mg;
1711 if (tvd->vdev_islog)
1712 metaslab_group_activate(mg);
1717 spa_offline_log(spa_t *spa)
1721 if ((error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1722 NULL, DS_FIND_CHILDREN)) == 0) {
1725 * We successfully offlined the log device, sync out the
1726 * current txg so that the "stubby" block can be removed
1729 txg_wait_synced(spa->spa_dsl_pool, 0);
1735 spa_aux_check_removed(spa_aux_vdev_t *sav)
1739 for (i = 0; i < sav->sav_count; i++)
1740 spa_check_removed(sav->sav_vdevs[i]);
1744 spa_claim_notify(zio_t *zio)
1746 spa_t *spa = zio->io_spa;
1751 mutex_enter(&spa->spa_props_lock); /* any mutex will do */
1752 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1753 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1754 mutex_exit(&spa->spa_props_lock);
1757 typedef struct spa_load_error {
1758 uint64_t sle_meta_count;
1759 uint64_t sle_data_count;
1763 spa_load_verify_done(zio_t *zio)
1765 blkptr_t *bp = zio->io_bp;
1766 spa_load_error_t *sle = zio->io_private;
1767 dmu_object_type_t type = BP_GET_TYPE(bp);
1768 int error = zio->io_error;
1771 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1772 type != DMU_OT_INTENT_LOG)
1773 atomic_add_64(&sle->sle_meta_count, 1);
1775 atomic_add_64(&sle->sle_data_count, 1);
1777 zio_data_buf_free(zio->io_data, zio->io_size);
1782 spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1783 arc_buf_t *pbuf, const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1787 size_t size = BP_GET_PSIZE(bp);
1788 void *data = zio_data_buf_alloc(size);
1790 zio_nowait(zio_read(rio, spa, bp, data, size,
1791 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1792 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1793 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1799 spa_load_verify(spa_t *spa)
1802 spa_load_error_t sle = { 0 };
1803 zpool_rewind_policy_t policy;
1804 boolean_t verify_ok = B_FALSE;
1807 zpool_get_rewind_policy(spa->spa_config, &policy);
1809 if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1812 rio = zio_root(spa, NULL, &sle,
1813 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1815 error = traverse_pool(spa, spa->spa_verify_min_txg,
1816 TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio);
1818 (void) zio_wait(rio);
1820 spa->spa_load_meta_errors = sle.sle_meta_count;
1821 spa->spa_load_data_errors = sle.sle_data_count;
1823 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1824 sle.sle_data_count <= policy.zrp_maxdata) {
1828 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1829 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1831 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1832 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1833 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1834 VERIFY(nvlist_add_int64(spa->spa_load_info,
1835 ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1836 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1837 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1839 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1843 if (error != ENXIO && error != EIO)
1848 return (verify_ok ? 0 : EIO);
1852 * Find a value in the pool props object.
1855 spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1857 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1858 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1862 * Find a value in the pool directory object.
1865 spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1867 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1868 name, sizeof (uint64_t), 1, val));
1872 spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1874 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1879 * Fix up config after a partly-completed split. This is done with the
1880 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
1881 * pool have that entry in their config, but only the splitting one contains
1882 * a list of all the guids of the vdevs that are being split off.
1884 * This function determines what to do with that list: either rejoin
1885 * all the disks to the pool, or complete the splitting process. To attempt
1886 * the rejoin, each disk that is offlined is marked online again, and
1887 * we do a reopen() call. If the vdev label for every disk that was
1888 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1889 * then we call vdev_split() on each disk, and complete the split.
1891 * Otherwise we leave the config alone, with all the vdevs in place in
1892 * the original pool.
1895 spa_try_repair(spa_t *spa, nvlist_t *config)
1902 boolean_t attempt_reopen;
1904 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
1907 /* check that the config is complete */
1908 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
1909 &glist, &gcount) != 0)
1912 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
1914 /* attempt to online all the vdevs & validate */
1915 attempt_reopen = B_TRUE;
1916 for (i = 0; i < gcount; i++) {
1917 if (glist[i] == 0) /* vdev is hole */
1920 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
1921 if (vd[i] == NULL) {
1923 * Don't bother attempting to reopen the disks;
1924 * just do the split.
1926 attempt_reopen = B_FALSE;
1928 /* attempt to re-online it */
1929 vd[i]->vdev_offline = B_FALSE;
1933 if (attempt_reopen) {
1934 vdev_reopen(spa->spa_root_vdev);
1936 /* check each device to see what state it's in */
1937 for (extracted = 0, i = 0; i < gcount; i++) {
1938 if (vd[i] != NULL &&
1939 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
1946 * If every disk has been moved to the new pool, or if we never
1947 * even attempted to look at them, then we split them off for
1950 if (!attempt_reopen || gcount == extracted) {
1951 for (i = 0; i < gcount; i++)
1954 vdev_reopen(spa->spa_root_vdev);
1957 kmem_free(vd, gcount * sizeof (vdev_t *));
1961 spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
1962 boolean_t mosconfig)
1964 nvlist_t *config = spa->spa_config;
1965 char *ereport = FM_EREPORT_ZFS_POOL;
1971 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
1974 ASSERT(spa->spa_comment == NULL);
1975 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
1976 spa->spa_comment = spa_strdup(comment);
1979 * Versioning wasn't explicitly added to the label until later, so if
1980 * it's not present treat it as the initial version.
1982 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
1983 &spa->spa_ubsync.ub_version) != 0)
1984 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
1986 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
1987 &spa->spa_config_txg);
1989 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
1990 spa_guid_exists(pool_guid, 0)) {
1993 spa->spa_config_guid = pool_guid;
1995 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
1997 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
2001 nvlist_free(spa->spa_load_info);
2002 spa->spa_load_info = fnvlist_alloc();
2004 gethrestime(&spa->spa_loaded_ts);
2005 error = spa_load_impl(spa, pool_guid, config, state, type,
2006 mosconfig, &ereport);
2009 spa->spa_minref = refcount_count(&spa->spa_refcount);
2011 if (error != EEXIST) {
2012 spa->spa_loaded_ts.tv_sec = 0;
2013 spa->spa_loaded_ts.tv_nsec = 0;
2015 if (error != EBADF) {
2016 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
2019 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2026 * Load an existing storage pool, using the pool's builtin spa_config as a
2027 * source of configuration information.
2030 spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
2031 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
2035 nvlist_t *nvroot = NULL;
2038 uberblock_t *ub = &spa->spa_uberblock;
2039 uint64_t children, config_cache_txg = spa->spa_config_txg;
2040 int orig_mode = spa->spa_mode;
2043 boolean_t missing_feat_write = B_FALSE;
2046 * If this is an untrusted config, access the pool in read-only mode.
2047 * This prevents things like resilvering recently removed devices.
2050 spa->spa_mode = FREAD;
2052 ASSERT(MUTEX_HELD(&spa_namespace_lock));
2054 spa->spa_load_state = state;
2056 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
2059 parse = (type == SPA_IMPORT_EXISTING ?
2060 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2063 * Create "The Godfather" zio to hold all async IOs
2065 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2066 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2069 * Parse the configuration into a vdev tree. We explicitly set the
2070 * value that will be returned by spa_version() since parsing the
2071 * configuration requires knowing the version number.
2073 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2074 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
2075 spa_config_exit(spa, SCL_ALL, FTAG);
2080 ASSERT(spa->spa_root_vdev == rvd);
2082 if (type != SPA_IMPORT_ASSEMBLE) {
2083 ASSERT(spa_guid(spa) == pool_guid);
2087 * Try to open all vdevs, loading each label in the process.
2089 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2090 error = vdev_open(rvd);
2091 spa_config_exit(spa, SCL_ALL, FTAG);
2096 * We need to validate the vdev labels against the configuration that
2097 * we have in hand, which is dependent on the setting of mosconfig. If
2098 * mosconfig is true then we're validating the vdev labels based on
2099 * that config. Otherwise, we're validating against the cached config
2100 * (zpool.cache) that was read when we loaded the zfs module, and then
2101 * later we will recursively call spa_load() and validate against
2104 * If we're assembling a new pool that's been split off from an
2105 * existing pool, the labels haven't yet been updated so we skip
2106 * validation for now.
2108 if (type != SPA_IMPORT_ASSEMBLE) {
2109 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2110 error = vdev_validate(rvd, mosconfig);
2111 spa_config_exit(spa, SCL_ALL, FTAG);
2116 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2121 * Find the best uberblock.
2123 vdev_uberblock_load(rvd, ub, &label);
2126 * If we weren't able to find a single valid uberblock, return failure.
2128 if (ub->ub_txg == 0) {
2130 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2134 * If the pool has an unsupported version we can't open it.
2136 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2138 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2141 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2145 * If we weren't able to find what's necessary for reading the
2146 * MOS in the label, return failure.
2148 if (label == NULL || nvlist_lookup_nvlist(label,
2149 ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) {
2151 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2156 * Update our in-core representation with the definitive values
2159 nvlist_free(spa->spa_label_features);
2160 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2166 * Look through entries in the label nvlist's features_for_read. If
2167 * there is a feature listed there which we don't understand then we
2168 * cannot open a pool.
2170 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2171 nvlist_t *unsup_feat;
2173 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2176 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2178 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2179 if (!zfeature_is_supported(nvpair_name(nvp))) {
2180 VERIFY(nvlist_add_string(unsup_feat,
2181 nvpair_name(nvp), "") == 0);
2185 if (!nvlist_empty(unsup_feat)) {
2186 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2187 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2188 nvlist_free(unsup_feat);
2189 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2193 nvlist_free(unsup_feat);
2197 * If the vdev guid sum doesn't match the uberblock, we have an
2198 * incomplete configuration. We first check to see if the pool
2199 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2200 * If it is, defer the vdev_guid_sum check till later so we
2201 * can handle missing vdevs.
2203 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
2204 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
2205 rvd->vdev_guid_sum != ub->ub_guid_sum)
2206 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2208 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2209 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2210 spa_try_repair(spa, config);
2211 spa_config_exit(spa, SCL_ALL, FTAG);
2212 nvlist_free(spa->spa_config_splitting);
2213 spa->spa_config_splitting = NULL;
2217 * Initialize internal SPA structures.
2219 spa->spa_state = POOL_STATE_ACTIVE;
2220 spa->spa_ubsync = spa->spa_uberblock;
2221 spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2222 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2223 spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2224 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2225 spa->spa_claim_max_txg = spa->spa_first_txg;
2226 spa->spa_prev_software_version = ub->ub_software_version;
2228 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2230 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2231 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2233 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2234 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2236 if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2237 boolean_t missing_feat_read = B_FALSE;
2238 nvlist_t *unsup_feat, *enabled_feat;
2240 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2241 &spa->spa_feat_for_read_obj) != 0) {
2242 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2245 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2246 &spa->spa_feat_for_write_obj) != 0) {
2247 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2250 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2251 &spa->spa_feat_desc_obj) != 0) {
2252 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2255 enabled_feat = fnvlist_alloc();
2256 unsup_feat = fnvlist_alloc();
2258 if (!feature_is_supported(spa->spa_meta_objset,
2259 spa->spa_feat_for_read_obj, spa->spa_feat_desc_obj,
2260 unsup_feat, enabled_feat))
2261 missing_feat_read = B_TRUE;
2263 if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) {
2264 if (!feature_is_supported(spa->spa_meta_objset,
2265 spa->spa_feat_for_write_obj, spa->spa_feat_desc_obj,
2266 unsup_feat, enabled_feat)) {
2267 missing_feat_write = B_TRUE;
2271 fnvlist_add_nvlist(spa->spa_load_info,
2272 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
2274 if (!nvlist_empty(unsup_feat)) {
2275 fnvlist_add_nvlist(spa->spa_load_info,
2276 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
2279 fnvlist_free(enabled_feat);
2280 fnvlist_free(unsup_feat);
2282 if (!missing_feat_read) {
2283 fnvlist_add_boolean(spa->spa_load_info,
2284 ZPOOL_CONFIG_CAN_RDONLY);
2288 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2289 * twofold: to determine whether the pool is available for
2290 * import in read-write mode and (if it is not) whether the
2291 * pool is available for import in read-only mode. If the pool
2292 * is available for import in read-write mode, it is displayed
2293 * as available in userland; if it is not available for import
2294 * in read-only mode, it is displayed as unavailable in
2295 * userland. If the pool is available for import in read-only
2296 * mode but not read-write mode, it is displayed as unavailable
2297 * in userland with a special note that the pool is actually
2298 * available for open in read-only mode.
2300 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2301 * missing a feature for write, we must first determine whether
2302 * the pool can be opened read-only before returning to
2303 * userland in order to know whether to display the
2304 * abovementioned note.
2306 if (missing_feat_read || (missing_feat_write &&
2307 spa_writeable(spa))) {
2308 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2313 spa->spa_is_initializing = B_TRUE;
2314 error = dsl_pool_open(spa->spa_dsl_pool);
2315 spa->spa_is_initializing = B_FALSE;
2317 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2321 nvlist_t *policy = NULL, *nvconfig;
2323 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2324 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2326 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2327 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2329 unsigned long myhostid = 0;
2331 VERIFY(nvlist_lookup_string(nvconfig,
2332 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2335 myhostid = zone_get_hostid(NULL);
2338 * We're emulating the system's hostid in userland, so
2339 * we can't use zone_get_hostid().
2341 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2342 #endif /* _KERNEL */
2343 if (check_hostid && hostid != 0 && myhostid != 0 &&
2344 hostid != myhostid) {
2345 nvlist_free(nvconfig);
2346 cmn_err(CE_WARN, "pool '%s' could not be "
2347 "loaded as it was last accessed by "
2348 "another system (host: %s hostid: 0x%lx). "
2349 "See: http://illumos.org/msg/ZFS-8000-EY",
2350 spa_name(spa), hostname,
2351 (unsigned long)hostid);
2355 if (nvlist_lookup_nvlist(spa->spa_config,
2356 ZPOOL_REWIND_POLICY, &policy) == 0)
2357 VERIFY(nvlist_add_nvlist(nvconfig,
2358 ZPOOL_REWIND_POLICY, policy) == 0);
2360 spa_config_set(spa, nvconfig);
2362 spa_deactivate(spa);
2363 spa_activate(spa, orig_mode);
2365 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2368 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2369 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2370 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2372 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2375 * Load the bit that tells us to use the new accounting function
2376 * (raid-z deflation). If we have an older pool, this will not
2379 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2380 if (error != 0 && error != ENOENT)
2381 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2383 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2384 &spa->spa_creation_version);
2385 if (error != 0 && error != ENOENT)
2386 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2389 * Load the persistent error log. If we have an older pool, this will
2392 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2393 if (error != 0 && error != ENOENT)
2394 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2396 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2397 &spa->spa_errlog_scrub);
2398 if (error != 0 && error != ENOENT)
2399 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2402 * Load the history object. If we have an older pool, this
2403 * will not be present.
2405 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2406 if (error != 0 && error != ENOENT)
2407 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2410 * If we're assembling the pool from the split-off vdevs of
2411 * an existing pool, we don't want to attach the spares & cache
2416 * Load any hot spares for this pool.
2418 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2419 if (error != 0 && error != ENOENT)
2420 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2421 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2422 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2423 if (load_nvlist(spa, spa->spa_spares.sav_object,
2424 &spa->spa_spares.sav_config) != 0)
2425 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2427 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2428 spa_load_spares(spa);
2429 spa_config_exit(spa, SCL_ALL, FTAG);
2430 } else if (error == 0) {
2431 spa->spa_spares.sav_sync = B_TRUE;
2435 * Load any level 2 ARC devices for this pool.
2437 error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2438 &spa->spa_l2cache.sav_object);
2439 if (error != 0 && error != ENOENT)
2440 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2441 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2442 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2443 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2444 &spa->spa_l2cache.sav_config) != 0)
2445 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2447 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2448 spa_load_l2cache(spa);
2449 spa_config_exit(spa, SCL_ALL, FTAG);
2450 } else if (error == 0) {
2451 spa->spa_l2cache.sav_sync = B_TRUE;
2454 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2456 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2457 if (error && error != ENOENT)
2458 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2461 uint64_t autoreplace;
2463 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2464 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2465 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2466 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2467 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2468 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2469 &spa->spa_dedup_ditto);
2471 spa->spa_autoreplace = (autoreplace != 0);
2475 * If the 'autoreplace' property is set, then post a resource notifying
2476 * the ZFS DE that it should not issue any faults for unopenable
2477 * devices. We also iterate over the vdevs, and post a sysevent for any
2478 * unopenable vdevs so that the normal autoreplace handler can take
2481 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2482 spa_check_removed(spa->spa_root_vdev);
2484 * For the import case, this is done in spa_import(), because
2485 * at this point we're using the spare definitions from
2486 * the MOS config, not necessarily from the userland config.
2488 if (state != SPA_LOAD_IMPORT) {
2489 spa_aux_check_removed(&spa->spa_spares);
2490 spa_aux_check_removed(&spa->spa_l2cache);
2495 * Load the vdev state for all toplevel vdevs.
2500 * Propagate the leaf DTLs we just loaded all the way up the tree.
2502 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2503 vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2504 spa_config_exit(spa, SCL_ALL, FTAG);
2507 * Load the DDTs (dedup tables).
2509 error = ddt_load(spa);
2511 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2513 spa_update_dspace(spa);
2516 * Validate the config, using the MOS config to fill in any
2517 * information which might be missing. If we fail to validate
2518 * the config then declare the pool unfit for use. If we're
2519 * assembling a pool from a split, the log is not transferred
2522 if (type != SPA_IMPORT_ASSEMBLE) {
2525 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2526 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2528 if (!spa_config_valid(spa, nvconfig)) {
2529 nvlist_free(nvconfig);
2530 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2533 nvlist_free(nvconfig);
2536 * Now that we've validated the config, check the state of the
2537 * root vdev. If it can't be opened, it indicates one or
2538 * more toplevel vdevs are faulted.
2540 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2543 if (spa_check_logs(spa)) {
2544 *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2545 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2549 if (missing_feat_write) {
2550 ASSERT(state == SPA_LOAD_TRYIMPORT);
2553 * At this point, we know that we can open the pool in
2554 * read-only mode but not read-write mode. We now have enough
2555 * information and can return to userland.
2557 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
2561 * We've successfully opened the pool, verify that we're ready
2562 * to start pushing transactions.
2564 if (state != SPA_LOAD_TRYIMPORT) {
2565 if (error = spa_load_verify(spa))
2566 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2570 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2571 spa->spa_load_max_txg == UINT64_MAX)) {
2573 int need_update = B_FALSE;
2575 ASSERT(state != SPA_LOAD_TRYIMPORT);
2578 * Claim log blocks that haven't been committed yet.
2579 * This must all happen in a single txg.
2580 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2581 * invoked from zil_claim_log_block()'s i/o done callback.
2582 * Price of rollback is that we abandon the log.
2584 spa->spa_claiming = B_TRUE;
2586 tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2587 spa_first_txg(spa));
2588 (void) dmu_objset_find(spa_name(spa),
2589 zil_claim, tx, DS_FIND_CHILDREN);
2592 spa->spa_claiming = B_FALSE;
2594 spa_set_log_state(spa, SPA_LOG_GOOD);
2595 spa->spa_sync_on = B_TRUE;
2596 txg_sync_start(spa->spa_dsl_pool);
2599 * Wait for all claims to sync. We sync up to the highest
2600 * claimed log block birth time so that claimed log blocks
2601 * don't appear to be from the future. spa_claim_max_txg
2602 * will have been set for us by either zil_check_log_chain()
2603 * (invoked from spa_check_logs()) or zil_claim() above.
2605 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2608 * If the config cache is stale, or we have uninitialized
2609 * metaslabs (see spa_vdev_add()), then update the config.
2611 * If this is a verbatim import, trust the current
2612 * in-core spa_config and update the disk labels.
2614 if (config_cache_txg != spa->spa_config_txg ||
2615 state == SPA_LOAD_IMPORT ||
2616 state == SPA_LOAD_RECOVER ||
2617 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2618 need_update = B_TRUE;
2620 for (int c = 0; c < rvd->vdev_children; c++)
2621 if (rvd->vdev_child[c]->vdev_ms_array == 0)
2622 need_update = B_TRUE;
2625 * Update the config cache asychronously in case we're the
2626 * root pool, in which case the config cache isn't writable yet.
2629 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2632 * Check all DTLs to see if anything needs resilvering.
2634 if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2635 vdev_resilver_needed(rvd, NULL, NULL))
2636 spa_async_request(spa, SPA_ASYNC_RESILVER);
2639 * Delete any inconsistent datasets.
2641 (void) dmu_objset_find(spa_name(spa),
2642 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2645 * Clean up any stale temporary dataset userrefs.
2647 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2654 spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2656 int mode = spa->spa_mode;
2659 spa_deactivate(spa);
2661 spa->spa_load_max_txg--;
2663 spa_activate(spa, mode);
2664 spa_async_suspend(spa);
2666 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2670 * If spa_load() fails this function will try loading prior txg's. If
2671 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2672 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2673 * function will not rewind the pool and will return the same error as
2677 spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2678 uint64_t max_request, int rewind_flags)
2680 nvlist_t *loadinfo = NULL;
2681 nvlist_t *config = NULL;
2682 int load_error, rewind_error;
2683 uint64_t safe_rewind_txg;
2686 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2687 spa->spa_load_max_txg = spa->spa_load_txg;
2688 spa_set_log_state(spa, SPA_LOG_CLEAR);
2690 spa->spa_load_max_txg = max_request;
2693 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2695 if (load_error == 0)
2698 if (spa->spa_root_vdev != NULL)
2699 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2701 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2702 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2704 if (rewind_flags & ZPOOL_NEVER_REWIND) {
2705 nvlist_free(config);
2706 return (load_error);
2709 if (state == SPA_LOAD_RECOVER) {
2710 /* Price of rolling back is discarding txgs, including log */
2711 spa_set_log_state(spa, SPA_LOG_CLEAR);
2714 * If we aren't rolling back save the load info from our first
2715 * import attempt so that we can restore it after attempting
2718 loadinfo = spa->spa_load_info;
2719 spa->spa_load_info = fnvlist_alloc();
2722 spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2723 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2724 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2725 TXG_INITIAL : safe_rewind_txg;
2728 * Continue as long as we're finding errors, we're still within
2729 * the acceptable rewind range, and we're still finding uberblocks
2731 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2732 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2733 if (spa->spa_load_max_txg < safe_rewind_txg)
2734 spa->spa_extreme_rewind = B_TRUE;
2735 rewind_error = spa_load_retry(spa, state, mosconfig);
2738 spa->spa_extreme_rewind = B_FALSE;
2739 spa->spa_load_max_txg = UINT64_MAX;
2741 if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2742 spa_config_set(spa, config);
2744 if (state == SPA_LOAD_RECOVER) {
2745 ASSERT3P(loadinfo, ==, NULL);
2746 return (rewind_error);
2748 /* Store the rewind info as part of the initial load info */
2749 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
2750 spa->spa_load_info);
2752 /* Restore the initial load info */
2753 fnvlist_free(spa->spa_load_info);
2754 spa->spa_load_info = loadinfo;
2756 return (load_error);
2763 * The import case is identical to an open except that the configuration is sent
2764 * down from userland, instead of grabbed from the configuration cache. For the
2765 * case of an open, the pool configuration will exist in the
2766 * POOL_STATE_UNINITIALIZED state.
2768 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2769 * the same time open the pool, without having to keep around the spa_t in some
2773 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2777 spa_load_state_t state = SPA_LOAD_OPEN;
2779 int locked = B_FALSE;
2780 int firstopen = B_FALSE;
2785 * As disgusting as this is, we need to support recursive calls to this
2786 * function because dsl_dir_open() is called during spa_load(), and ends
2787 * up calling spa_open() again. The real fix is to figure out how to
2788 * avoid dsl_dir_open() calling this in the first place.
2790 if (mutex_owner(&spa_namespace_lock) != curthread) {
2791 mutex_enter(&spa_namespace_lock);
2795 if ((spa = spa_lookup(pool)) == NULL) {
2797 mutex_exit(&spa_namespace_lock);
2801 if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2802 zpool_rewind_policy_t policy;
2806 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2808 if (policy.zrp_request & ZPOOL_DO_REWIND)
2809 state = SPA_LOAD_RECOVER;
2811 spa_activate(spa, spa_mode_global);
2813 if (state != SPA_LOAD_RECOVER)
2814 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2816 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2817 policy.zrp_request);
2819 if (error == EBADF) {
2821 * If vdev_validate() returns failure (indicated by
2822 * EBADF), it indicates that one of the vdevs indicates
2823 * that the pool has been exported or destroyed. If
2824 * this is the case, the config cache is out of sync and
2825 * we should remove the pool from the namespace.
2828 spa_deactivate(spa);
2829 spa_config_sync(spa, B_TRUE, B_TRUE);
2832 mutex_exit(&spa_namespace_lock);
2838 * We can't open the pool, but we still have useful
2839 * information: the state of each vdev after the
2840 * attempted vdev_open(). Return this to the user.
2842 if (config != NULL && spa->spa_config) {
2843 VERIFY(nvlist_dup(spa->spa_config, config,
2845 VERIFY(nvlist_add_nvlist(*config,
2846 ZPOOL_CONFIG_LOAD_INFO,
2847 spa->spa_load_info) == 0);
2850 spa_deactivate(spa);
2851 spa->spa_last_open_failed = error;
2853 mutex_exit(&spa_namespace_lock);
2859 spa_open_ref(spa, tag);
2862 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2865 * If we've recovered the pool, pass back any information we
2866 * gathered while doing the load.
2868 if (state == SPA_LOAD_RECOVER) {
2869 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
2870 spa->spa_load_info) == 0);
2874 spa->spa_last_open_failed = 0;
2875 spa->spa_last_ubsync_txg = 0;
2876 spa->spa_load_txg = 0;
2877 mutex_exit(&spa_namespace_lock);
2881 zvol_create_minors(pool);
2892 spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
2895 return (spa_open_common(name, spapp, tag, policy, config));
2899 spa_open(const char *name, spa_t **spapp, void *tag)
2901 return (spa_open_common(name, spapp, tag, NULL, NULL));
2905 * Lookup the given spa_t, incrementing the inject count in the process,
2906 * preventing it from being exported or destroyed.
2909 spa_inject_addref(char *name)
2913 mutex_enter(&spa_namespace_lock);
2914 if ((spa = spa_lookup(name)) == NULL) {
2915 mutex_exit(&spa_namespace_lock);
2918 spa->spa_inject_ref++;
2919 mutex_exit(&spa_namespace_lock);
2925 spa_inject_delref(spa_t *spa)
2927 mutex_enter(&spa_namespace_lock);
2928 spa->spa_inject_ref--;
2929 mutex_exit(&spa_namespace_lock);
2933 * Add spares device information to the nvlist.
2936 spa_add_spares(spa_t *spa, nvlist_t *config)
2946 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2948 if (spa->spa_spares.sav_count == 0)
2951 VERIFY(nvlist_lookup_nvlist(config,
2952 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2953 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
2954 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2956 VERIFY(nvlist_add_nvlist_array(nvroot,
2957 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2958 VERIFY(nvlist_lookup_nvlist_array(nvroot,
2959 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2962 * Go through and find any spares which have since been
2963 * repurposed as an active spare. If this is the case, update
2964 * their status appropriately.
2966 for (i = 0; i < nspares; i++) {
2967 VERIFY(nvlist_lookup_uint64(spares[i],
2968 ZPOOL_CONFIG_GUID, &guid) == 0);
2969 if (spa_spare_exists(guid, &pool, NULL) &&
2971 VERIFY(nvlist_lookup_uint64_array(
2972 spares[i], ZPOOL_CONFIG_VDEV_STATS,
2973 (uint64_t **)&vs, &vsc) == 0);
2974 vs->vs_state = VDEV_STATE_CANT_OPEN;
2975 vs->vs_aux = VDEV_AUX_SPARED;
2982 * Add l2cache device information to the nvlist, including vdev stats.
2985 spa_add_l2cache(spa_t *spa, nvlist_t *config)
2988 uint_t i, j, nl2cache;
2995 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2997 if (spa->spa_l2cache.sav_count == 0)
3000 VERIFY(nvlist_lookup_nvlist(config,
3001 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3002 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
3003 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3004 if (nl2cache != 0) {
3005 VERIFY(nvlist_add_nvlist_array(nvroot,
3006 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3007 VERIFY(nvlist_lookup_nvlist_array(nvroot,
3008 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3011 * Update level 2 cache device stats.
3014 for (i = 0; i < nl2cache; i++) {
3015 VERIFY(nvlist_lookup_uint64(l2cache[i],
3016 ZPOOL_CONFIG_GUID, &guid) == 0);
3019 for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
3021 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
3022 vd = spa->spa_l2cache.sav_vdevs[j];
3028 VERIFY(nvlist_lookup_uint64_array(l2cache[i],
3029 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
3031 vdev_get_stats(vd, vs);
3037 spa_add_feature_stats(spa_t *spa, nvlist_t *config)
3043 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3044 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3046 if (spa->spa_feat_for_read_obj != 0) {
3047 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3048 spa->spa_feat_for_read_obj);
3049 zap_cursor_retrieve(&zc, &za) == 0;
3050 zap_cursor_advance(&zc)) {
3051 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3052 za.za_num_integers == 1);
3053 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3054 za.za_first_integer));
3056 zap_cursor_fini(&zc);
3059 if (spa->spa_feat_for_write_obj != 0) {
3060 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3061 spa->spa_feat_for_write_obj);
3062 zap_cursor_retrieve(&zc, &za) == 0;
3063 zap_cursor_advance(&zc)) {
3064 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3065 za.za_num_integers == 1);
3066 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3067 za.za_first_integer));
3069 zap_cursor_fini(&zc);
3072 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
3074 nvlist_free(features);
3078 spa_get_stats(const char *name, nvlist_t **config,
3079 char *altroot, size_t buflen)
3085 error = spa_open_common(name, &spa, FTAG, NULL, config);
3089 * This still leaves a window of inconsistency where the spares
3090 * or l2cache devices could change and the config would be
3091 * self-inconsistent.
3093 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3095 if (*config != NULL) {
3096 uint64_t loadtimes[2];
3098 loadtimes[0] = spa->spa_loaded_ts.tv_sec;
3099 loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
3100 VERIFY(nvlist_add_uint64_array(*config,
3101 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
3103 VERIFY(nvlist_add_uint64(*config,
3104 ZPOOL_CONFIG_ERRCOUNT,
3105 spa_get_errlog_size(spa)) == 0);
3107 if (spa_suspended(spa))
3108 VERIFY(nvlist_add_uint64(*config,
3109 ZPOOL_CONFIG_SUSPENDED,
3110 spa->spa_failmode) == 0);
3112 spa_add_spares(spa, *config);
3113 spa_add_l2cache(spa, *config);
3114 spa_add_feature_stats(spa, *config);
3119 * We want to get the alternate root even for faulted pools, so we cheat
3120 * and call spa_lookup() directly.
3124 mutex_enter(&spa_namespace_lock);
3125 spa = spa_lookup(name);
3127 spa_altroot(spa, altroot, buflen);
3131 mutex_exit(&spa_namespace_lock);
3133 spa_altroot(spa, altroot, buflen);
3138 spa_config_exit(spa, SCL_CONFIG, FTAG);
3139 spa_close(spa, FTAG);
3146 * Validate that the auxiliary device array is well formed. We must have an
3147 * array of nvlists, each which describes a valid leaf vdev. If this is an
3148 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3149 * specified, as long as they are well-formed.
3152 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
3153 spa_aux_vdev_t *sav, const char *config, uint64_t version,
3154 vdev_labeltype_t label)
3161 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3164 * It's acceptable to have no devs specified.
3166 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
3173 * Make sure the pool is formatted with a version that supports this
3176 if (spa_version(spa) < version)
3180 * Set the pending device list so we correctly handle device in-use
3183 sav->sav_pending = dev;
3184 sav->sav_npending = ndev;
3186 for (i = 0; i < ndev; i++) {
3187 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
3191 if (!vd->vdev_ops->vdev_op_leaf) {
3198 * The L2ARC currently only supports disk devices in
3199 * kernel context. For user-level testing, we allow it.
3202 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
3203 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
3211 if ((error = vdev_open(vd)) == 0 &&
3212 (error = vdev_label_init(vd, crtxg, label)) == 0) {
3213 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
3214 vd->vdev_guid) == 0);
3220 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
3227 sav->sav_pending = NULL;
3228 sav->sav_npending = 0;
3233 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
3237 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3239 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3240 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
3241 VDEV_LABEL_SPARE)) != 0) {
3245 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3246 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
3247 VDEV_LABEL_L2CACHE));
3251 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
3256 if (sav->sav_config != NULL) {
3262 * Generate new dev list by concatentating with the
3265 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
3266 &olddevs, &oldndevs) == 0);
3268 newdevs = kmem_alloc(sizeof (void *) *
3269 (ndevs + oldndevs), KM_SLEEP);
3270 for (i = 0; i < oldndevs; i++)
3271 VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
3273 for (i = 0; i < ndevs; i++)
3274 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
3277 VERIFY(nvlist_remove(sav->sav_config, config,
3278 DATA_TYPE_NVLIST_ARRAY) == 0);
3280 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3281 config, newdevs, ndevs + oldndevs) == 0);
3282 for (i = 0; i < oldndevs + ndevs; i++)
3283 nvlist_free(newdevs[i]);
3284 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
3287 * Generate a new dev list.
3289 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
3291 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
3297 * Stop and drop level 2 ARC devices
3300 spa_l2cache_drop(spa_t *spa)
3304 spa_aux_vdev_t *sav = &spa->spa_l2cache;
3306 for (i = 0; i < sav->sav_count; i++) {
3309 vd = sav->sav_vdevs[i];
3312 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
3313 pool != 0ULL && l2arc_vdev_present(vd))
3314 l2arc_remove_vdev(vd);
3322 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
3323 const char *history_str, nvlist_t *zplprops)
3326 char *altroot = NULL;
3331 uint64_t txg = TXG_INITIAL;
3332 nvlist_t **spares, **l2cache;
3333 uint_t nspares, nl2cache;
3334 uint64_t version, obj;
3335 boolean_t has_features;
3338 * If this pool already exists, return failure.
3340 mutex_enter(&spa_namespace_lock);
3341 if (spa_lookup(pool) != NULL) {
3342 mutex_exit(&spa_namespace_lock);
3347 * Allocate a new spa_t structure.
3349 (void) nvlist_lookup_string(props,
3350 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3351 spa = spa_add(pool, NULL, altroot);
3352 spa_activate(spa, spa_mode_global);
3354 if (props && (error = spa_prop_validate(spa, props))) {
3355 spa_deactivate(spa);
3357 mutex_exit(&spa_namespace_lock);
3361 has_features = B_FALSE;
3362 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
3363 elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
3364 if (zpool_prop_feature(nvpair_name(elem)))
3365 has_features = B_TRUE;
3368 if (has_features || nvlist_lookup_uint64(props,
3369 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
3370 version = SPA_VERSION;
3372 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
3374 spa->spa_first_txg = txg;
3375 spa->spa_uberblock.ub_txg = txg - 1;
3376 spa->spa_uberblock.ub_version = version;
3377 spa->spa_ubsync = spa->spa_uberblock;
3380 * Create "The Godfather" zio to hold all async IOs
3382 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
3383 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
3386 * Create the root vdev.
3388 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3390 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3392 ASSERT(error != 0 || rvd != NULL);
3393 ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3395 if (error == 0 && !zfs_allocatable_devs(nvroot))
3399 (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3400 (error = spa_validate_aux(spa, nvroot, txg,
3401 VDEV_ALLOC_ADD)) == 0) {
3402 for (int c = 0; c < rvd->vdev_children; c++) {
3403 vdev_metaslab_set_size(rvd->vdev_child[c]);
3404 vdev_expand(rvd->vdev_child[c], txg);
3408 spa_config_exit(spa, SCL_ALL, FTAG);
3412 spa_deactivate(spa);
3414 mutex_exit(&spa_namespace_lock);
3419 * Get the list of spares, if specified.
3421 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3422 &spares, &nspares) == 0) {
3423 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3425 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3426 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3427 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3428 spa_load_spares(spa);
3429 spa_config_exit(spa, SCL_ALL, FTAG);
3430 spa->spa_spares.sav_sync = B_TRUE;
3434 * Get the list of level 2 cache devices, if specified.
3436 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3437 &l2cache, &nl2cache) == 0) {
3438 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3439 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3440 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3441 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3442 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3443 spa_load_l2cache(spa);
3444 spa_config_exit(spa, SCL_ALL, FTAG);
3445 spa->spa_l2cache.sav_sync = B_TRUE;
3448 spa->spa_is_initializing = B_TRUE;
3449 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3450 spa->spa_meta_objset = dp->dp_meta_objset;
3451 spa->spa_is_initializing = B_FALSE;
3454 * Create DDTs (dedup tables).
3458 spa_update_dspace(spa);
3460 tx = dmu_tx_create_assigned(dp, txg);
3463 * Create the pool config object.
3465 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3466 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3467 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3469 if (zap_add(spa->spa_meta_objset,
3470 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3471 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3472 cmn_err(CE_PANIC, "failed to add pool config");
3475 if (spa_version(spa) >= SPA_VERSION_FEATURES)
3476 spa_feature_create_zap_objects(spa, tx);
3478 if (zap_add(spa->spa_meta_objset,
3479 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3480 sizeof (uint64_t), 1, &version, tx) != 0) {
3481 cmn_err(CE_PANIC, "failed to add pool version");
3484 /* Newly created pools with the right version are always deflated. */
3485 if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3486 spa->spa_deflate = TRUE;
3487 if (zap_add(spa->spa_meta_objset,
3488 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3489 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3490 cmn_err(CE_PANIC, "failed to add deflate");
3495 * Create the deferred-free bpobj. Turn off compression
3496 * because sync-to-convergence takes longer if the blocksize
3499 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3500 dmu_object_set_compress(spa->spa_meta_objset, obj,
3501 ZIO_COMPRESS_OFF, tx);
3502 if (zap_add(spa->spa_meta_objset,
3503 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3504 sizeof (uint64_t), 1, &obj, tx) != 0) {
3505 cmn_err(CE_PANIC, "failed to add bpobj");
3507 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3508 spa->spa_meta_objset, obj));
3511 * Create the pool's history object.
3513 if (version >= SPA_VERSION_ZPOOL_HISTORY)
3514 spa_history_create_obj(spa, tx);
3517 * Set pool properties.
3519 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3520 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3521 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3522 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3524 if (props != NULL) {
3525 spa_configfile_set(spa, props, B_FALSE);
3526 spa_sync_props(spa, props, tx);
3531 spa->spa_sync_on = B_TRUE;
3532 txg_sync_start(spa->spa_dsl_pool);
3535 * We explicitly wait for the first transaction to complete so that our
3536 * bean counters are appropriately updated.
3538 txg_wait_synced(spa->spa_dsl_pool, txg);
3540 spa_config_sync(spa, B_FALSE, B_TRUE);
3542 if (version >= SPA_VERSION_ZPOOL_HISTORY && history_str != NULL)
3543 (void) spa_history_log(spa, history_str, LOG_CMD_POOL_CREATE);
3544 spa_history_log_version(spa, LOG_POOL_CREATE);
3546 spa->spa_minref = refcount_count(&spa->spa_refcount);
3548 mutex_exit(&spa_namespace_lock);
3556 * Get the root pool information from the root disk, then import the root pool
3557 * during the system boot up time.
3559 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3562 spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3565 nvlist_t *nvtop, *nvroot;
3568 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3572 * Add this top-level vdev to the child array.
3574 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3576 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3578 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3581 * Put this pool's top-level vdevs into a root vdev.
3583 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3584 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3585 VDEV_TYPE_ROOT) == 0);
3586 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3587 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3588 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3592 * Replace the existing vdev_tree with the new root vdev in
3593 * this pool's configuration (remove the old, add the new).
3595 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3596 nvlist_free(nvroot);
3601 * Walk the vdev tree and see if we can find a device with "better"
3602 * configuration. A configuration is "better" if the label on that
3603 * device has a more recent txg.
3606 spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3608 for (int c = 0; c < vd->vdev_children; c++)
3609 spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3611 if (vd->vdev_ops->vdev_op_leaf) {
3615 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3619 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3623 * Do we have a better boot device?
3625 if (label_txg > *txg) {
3634 * Import a root pool.
3636 * For x86. devpath_list will consist of devid and/or physpath name of
3637 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3638 * The GRUB "findroot" command will return the vdev we should boot.
3640 * For Sparc, devpath_list consists the physpath name of the booting device
3641 * no matter the rootpool is a single device pool or a mirrored pool.
3643 * "/pci@1f,0/ide@d/disk@0,0:a"
3646 spa_import_rootpool(char *devpath, char *devid)
3649 vdev_t *rvd, *bvd, *avd = NULL;
3650 nvlist_t *config, *nvtop;
3656 * Read the label from the boot device and generate a configuration.
3658 config = spa_generate_rootconf(devpath, devid, &guid);
3659 #if defined(_OBP) && defined(_KERNEL)
3660 if (config == NULL) {
3661 if (strstr(devpath, "/iscsi/ssd") != NULL) {
3663 get_iscsi_bootpath_phy(devpath);
3664 config = spa_generate_rootconf(devpath, devid, &guid);
3668 if (config == NULL) {
3669 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
3674 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3676 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3678 mutex_enter(&spa_namespace_lock);
3679 if ((spa = spa_lookup(pname)) != NULL) {
3681 * Remove the existing root pool from the namespace so that we
3682 * can replace it with the correct config we just read in.
3687 spa = spa_add(pname, config, NULL);
3688 spa->spa_is_root = B_TRUE;
3689 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3692 * Build up a vdev tree based on the boot device's label config.
3694 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3696 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3697 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3698 VDEV_ALLOC_ROOTPOOL);
3699 spa_config_exit(spa, SCL_ALL, FTAG);
3701 mutex_exit(&spa_namespace_lock);
3702 nvlist_free(config);
3703 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3709 * Get the boot vdev.
3711 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3712 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3713 (u_longlong_t)guid);
3719 * Determine if there is a better boot device.
3722 spa_alt_rootvdev(rvd, &avd, &txg);
3724 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3725 "try booting from '%s'", avd->vdev_path);
3731 * If the boot device is part of a spare vdev then ensure that
3732 * we're booting off the active spare.
3734 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3735 !bvd->vdev_isspare) {
3736 cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3737 "try booting from '%s'",
3739 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3745 spa_history_log_version(spa, LOG_POOL_IMPORT);
3747 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3749 spa_config_exit(spa, SCL_ALL, FTAG);
3750 mutex_exit(&spa_namespace_lock);
3752 nvlist_free(config);
3759 vdev_geom_read_pool_label(const char *name, nvlist_t **config);
3762 spa_generate_rootconf(const char *name)
3765 nvlist_t *nvtop, *nvroot;
3768 if (vdev_geom_read_pool_label(name, &config) != 0)
3772 * Add this top-level vdev to the child array.
3774 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3776 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3780 * Put this pool's top-level vdevs into a root vdev.
3782 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3783 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3784 VDEV_TYPE_ROOT) == 0);
3785 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3786 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3787 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3791 * Replace the existing vdev_tree with the new root vdev in
3792 * this pool's configuration (remove the old, add the new).
3794 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3795 nvlist_free(nvroot);
3800 spa_import_rootpool(const char *name)
3803 vdev_t *rvd, *bvd, *avd = NULL;
3804 nvlist_t *config, *nvtop;
3810 * Read the label from the boot device and generate a configuration.
3812 config = spa_generate_rootconf(name);
3813 if (config == NULL) {
3814 cmn_err(CE_NOTE, "Cannot find the pool label for '%s'",
3819 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3820 &pname) == 0 && strcmp(name, pname) == 0);
3821 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3823 mutex_enter(&spa_namespace_lock);
3824 if ((spa = spa_lookup(pname)) != NULL) {
3826 * Remove the existing root pool from the namespace so that we
3827 * can replace it with the correct config we just read in.
3831 spa = spa_add(pname, config, NULL);
3832 spa->spa_is_root = B_TRUE;
3833 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3836 * Build up a vdev tree based on the boot device's label config.
3838 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3840 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3841 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3842 VDEV_ALLOC_ROOTPOOL);
3843 spa_config_exit(spa, SCL_ALL, FTAG);
3845 mutex_exit(&spa_namespace_lock);
3846 nvlist_free(config);
3847 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3853 spa_history_log_version(spa, LOG_POOL_IMPORT);
3855 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3857 spa_config_exit(spa, SCL_ALL, FTAG);
3858 mutex_exit(&spa_namespace_lock);
3867 * Import a non-root pool into the system.
3870 spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
3873 char *altroot = NULL;
3874 spa_load_state_t state = SPA_LOAD_IMPORT;
3875 zpool_rewind_policy_t policy;
3876 uint64_t mode = spa_mode_global;
3877 uint64_t readonly = B_FALSE;
3880 nvlist_t **spares, **l2cache;
3881 uint_t nspares, nl2cache;
3884 * If a pool with this name exists, return failure.
3886 mutex_enter(&spa_namespace_lock);
3887 if (spa_lookup(pool) != NULL) {
3888 mutex_exit(&spa_namespace_lock);
3893 * Create and initialize the spa structure.
3895 (void) nvlist_lookup_string(props,
3896 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3897 (void) nvlist_lookup_uint64(props,
3898 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
3901 spa = spa_add(pool, config, altroot);
3902 spa->spa_import_flags = flags;
3905 * Verbatim import - Take a pool and insert it into the namespace
3906 * as if it had been loaded at boot.
3908 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
3910 spa_configfile_set(spa, props, B_FALSE);
3912 spa_config_sync(spa, B_FALSE, B_TRUE);
3914 mutex_exit(&spa_namespace_lock);
3915 spa_history_log_version(spa, LOG_POOL_IMPORT);
3920 spa_activate(spa, mode);
3923 * Don't start async tasks until we know everything is healthy.
3925 spa_async_suspend(spa);
3927 zpool_get_rewind_policy(config, &policy);
3928 if (policy.zrp_request & ZPOOL_DO_REWIND)
3929 state = SPA_LOAD_RECOVER;
3932 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
3933 * because the user-supplied config is actually the one to trust when
3936 if (state != SPA_LOAD_RECOVER)
3937 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3939 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
3940 policy.zrp_request);
3943 * Propagate anything learned while loading the pool and pass it
3944 * back to caller (i.e. rewind info, missing devices, etc).
3946 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
3947 spa->spa_load_info) == 0);
3949 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3951 * Toss any existing sparelist, as it doesn't have any validity
3952 * anymore, and conflicts with spa_has_spare().
3954 if (spa->spa_spares.sav_config) {
3955 nvlist_free(spa->spa_spares.sav_config);
3956 spa->spa_spares.sav_config = NULL;
3957 spa_load_spares(spa);
3959 if (spa->spa_l2cache.sav_config) {
3960 nvlist_free(spa->spa_l2cache.sav_config);
3961 spa->spa_l2cache.sav_config = NULL;
3962 spa_load_l2cache(spa);
3965 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3968 error = spa_validate_aux(spa, nvroot, -1ULL,
3971 error = spa_validate_aux(spa, nvroot, -1ULL,
3972 VDEV_ALLOC_L2CACHE);
3973 spa_config_exit(spa, SCL_ALL, FTAG);
3976 spa_configfile_set(spa, props, B_FALSE);
3978 if (error != 0 || (props && spa_writeable(spa) &&
3979 (error = spa_prop_set(spa, props)))) {
3981 spa_deactivate(spa);
3983 mutex_exit(&spa_namespace_lock);
3987 spa_async_resume(spa);
3990 * Override any spares and level 2 cache devices as specified by
3991 * the user, as these may have correct device names/devids, etc.
3993 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3994 &spares, &nspares) == 0) {
3995 if (spa->spa_spares.sav_config)
3996 VERIFY(nvlist_remove(spa->spa_spares.sav_config,
3997 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
3999 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
4000 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4001 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
4002 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
4003 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4004 spa_load_spares(spa);
4005 spa_config_exit(spa, SCL_ALL, FTAG);
4006 spa->spa_spares.sav_sync = B_TRUE;
4008 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
4009 &l2cache, &nl2cache) == 0) {
4010 if (spa->spa_l2cache.sav_config)
4011 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
4012 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
4014 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
4015 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4016 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
4017 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
4018 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4019 spa_load_l2cache(spa);
4020 spa_config_exit(spa, SCL_ALL, FTAG);
4021 spa->spa_l2cache.sav_sync = B_TRUE;
4025 * Check for any removed devices.
4027 if (spa->spa_autoreplace) {
4028 spa_aux_check_removed(&spa->spa_spares);
4029 spa_aux_check_removed(&spa->spa_l2cache);
4032 if (spa_writeable(spa)) {
4034 * Update the config cache to include the newly-imported pool.
4036 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4040 * It's possible that the pool was expanded while it was exported.
4041 * We kick off an async task to handle this for us.
4043 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
4045 mutex_exit(&spa_namespace_lock);
4046 spa_history_log_version(spa, LOG_POOL_IMPORT);
4050 zvol_create_minors(pool);
4057 spa_tryimport(nvlist_t *tryconfig)
4059 nvlist_t *config = NULL;
4065 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
4068 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
4072 * Create and initialize the spa structure.
4074 mutex_enter(&spa_namespace_lock);
4075 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
4076 spa_activate(spa, FREAD);
4079 * Pass off the heavy lifting to spa_load().
4080 * Pass TRUE for mosconfig because the user-supplied config
4081 * is actually the one to trust when doing an import.
4083 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
4086 * If 'tryconfig' was at least parsable, return the current config.
4088 if (spa->spa_root_vdev != NULL) {
4089 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4090 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
4092 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4094 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
4095 spa->spa_uberblock.ub_timestamp) == 0);
4096 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4097 spa->spa_load_info) == 0);
4100 * If the bootfs property exists on this pool then we
4101 * copy it out so that external consumers can tell which
4102 * pools are bootable.
4104 if ((!error || error == EEXIST) && spa->spa_bootfs) {
4105 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4108 * We have to play games with the name since the
4109 * pool was opened as TRYIMPORT_NAME.
4111 if (dsl_dsobj_to_dsname(spa_name(spa),
4112 spa->spa_bootfs, tmpname) == 0) {
4114 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4116 cp = strchr(tmpname, '/');
4118 (void) strlcpy(dsname, tmpname,
4121 (void) snprintf(dsname, MAXPATHLEN,
4122 "%s/%s", poolname, ++cp);
4124 VERIFY(nvlist_add_string(config,
4125 ZPOOL_CONFIG_BOOTFS, dsname) == 0);
4126 kmem_free(dsname, MAXPATHLEN);
4128 kmem_free(tmpname, MAXPATHLEN);
4132 * Add the list of hot spares and level 2 cache devices.
4134 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4135 spa_add_spares(spa, config);
4136 spa_add_l2cache(spa, config);
4137 spa_config_exit(spa, SCL_CONFIG, FTAG);
4141 spa_deactivate(spa);
4143 mutex_exit(&spa_namespace_lock);
4149 * Pool export/destroy
4151 * The act of destroying or exporting a pool is very simple. We make sure there
4152 * is no more pending I/O and any references to the pool are gone. Then, we
4153 * update the pool state and sync all the labels to disk, removing the
4154 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4155 * we don't sync the labels or remove the configuration cache.
4158 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
4159 boolean_t force, boolean_t hardforce)
4166 if (!(spa_mode_global & FWRITE))
4169 mutex_enter(&spa_namespace_lock);
4170 if ((spa = spa_lookup(pool)) == NULL) {
4171 mutex_exit(&spa_namespace_lock);
4176 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4177 * reacquire the namespace lock, and see if we can export.
4179 spa_open_ref(spa, FTAG);
4180 mutex_exit(&spa_namespace_lock);
4181 spa_async_suspend(spa);
4182 mutex_enter(&spa_namespace_lock);
4183 spa_close(spa, FTAG);
4186 * The pool will be in core if it's openable,
4187 * in which case we can modify its state.
4189 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
4191 * Objsets may be open only because they're dirty, so we
4192 * have to force it to sync before checking spa_refcnt.
4194 txg_wait_synced(spa->spa_dsl_pool, 0);
4197 * A pool cannot be exported or destroyed if there are active
4198 * references. If we are resetting a pool, allow references by
4199 * fault injection handlers.
4201 if (!spa_refcount_zero(spa) ||
4202 (spa->spa_inject_ref != 0 &&
4203 new_state != POOL_STATE_UNINITIALIZED)) {
4204 spa_async_resume(spa);
4205 mutex_exit(&spa_namespace_lock);
4210 * A pool cannot be exported if it has an active shared spare.
4211 * This is to prevent other pools stealing the active spare
4212 * from an exported pool. At user's own will, such pool can
4213 * be forcedly exported.
4215 if (!force && new_state == POOL_STATE_EXPORTED &&
4216 spa_has_active_shared_spare(spa)) {
4217 spa_async_resume(spa);
4218 mutex_exit(&spa_namespace_lock);
4223 * We want this to be reflected on every label,
4224 * so mark them all dirty. spa_unload() will do the
4225 * final sync that pushes these changes out.
4227 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
4228 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4229 spa->spa_state = new_state;
4230 spa->spa_final_txg = spa_last_synced_txg(spa) +
4232 vdev_config_dirty(spa->spa_root_vdev);
4233 spa_config_exit(spa, SCL_ALL, FTAG);
4237 spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
4239 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4241 spa_deactivate(spa);
4244 if (oldconfig && spa->spa_config)
4245 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
4247 if (new_state != POOL_STATE_UNINITIALIZED) {
4249 spa_config_sync(spa, B_TRUE, B_TRUE);
4252 mutex_exit(&spa_namespace_lock);
4258 * Destroy a storage pool.
4261 spa_destroy(char *pool)
4263 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
4268 * Export a storage pool.
4271 spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
4272 boolean_t hardforce)
4274 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
4279 * Similar to spa_export(), this unloads the spa_t without actually removing it
4280 * from the namespace in any way.
4283 spa_reset(char *pool)
4285 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
4290 * ==========================================================================
4291 * Device manipulation
4292 * ==========================================================================
4296 * Add a device to a storage pool.
4299 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
4303 vdev_t *rvd = spa->spa_root_vdev;
4305 nvlist_t **spares, **l2cache;
4306 uint_t nspares, nl2cache;
4308 ASSERT(spa_writeable(spa));
4310 txg = spa_vdev_enter(spa);
4312 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
4313 VDEV_ALLOC_ADD)) != 0)
4314 return (spa_vdev_exit(spa, NULL, txg, error));
4316 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */
4318 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
4322 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
4326 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
4327 return (spa_vdev_exit(spa, vd, txg, EINVAL));
4329 if (vd->vdev_children != 0 &&
4330 (error = vdev_create(vd, txg, B_FALSE)) != 0)
4331 return (spa_vdev_exit(spa, vd, txg, error));
4334 * We must validate the spares and l2cache devices after checking the
4335 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
4337 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
4338 return (spa_vdev_exit(spa, vd, txg, error));
4341 * Transfer each new top-level vdev from vd to rvd.
4343 for (int c = 0; c < vd->vdev_children; c++) {
4346 * Set the vdev id to the first hole, if one exists.
4348 for (id = 0; id < rvd->vdev_children; id++) {
4349 if (rvd->vdev_child[id]->vdev_ishole) {
4350 vdev_free(rvd->vdev_child[id]);
4354 tvd = vd->vdev_child[c];
4355 vdev_remove_child(vd, tvd);
4357 vdev_add_child(rvd, tvd);
4358 vdev_config_dirty(tvd);
4362 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
4363 ZPOOL_CONFIG_SPARES);
4364 spa_load_spares(spa);
4365 spa->spa_spares.sav_sync = B_TRUE;
4368 if (nl2cache != 0) {
4369 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
4370 ZPOOL_CONFIG_L2CACHE);
4371 spa_load_l2cache(spa);
4372 spa->spa_l2cache.sav_sync = B_TRUE;
4376 * We have to be careful when adding new vdevs to an existing pool.
4377 * If other threads start allocating from these vdevs before we
4378 * sync the config cache, and we lose power, then upon reboot we may
4379 * fail to open the pool because there are DVAs that the config cache
4380 * can't translate. Therefore, we first add the vdevs without
4381 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4382 * and then let spa_config_update() initialize the new metaslabs.
4384 * spa_load() checks for added-but-not-initialized vdevs, so that
4385 * if we lose power at any point in this sequence, the remaining
4386 * steps will be completed the next time we load the pool.
4388 (void) spa_vdev_exit(spa, vd, txg, 0);
4390 mutex_enter(&spa_namespace_lock);
4391 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4392 mutex_exit(&spa_namespace_lock);
4398 * Attach a device to a mirror. The arguments are the path to any device
4399 * in the mirror, and the nvroot for the new device. If the path specifies
4400 * a device that is not mirrored, we automatically insert the mirror vdev.
4402 * If 'replacing' is specified, the new device is intended to replace the
4403 * existing device; in this case the two devices are made into their own
4404 * mirror using the 'replacing' vdev, which is functionally identical to
4405 * the mirror vdev (it actually reuses all the same ops) but has a few
4406 * extra rules: you can't attach to it after it's been created, and upon
4407 * completion of resilvering, the first disk (the one being replaced)
4408 * is automatically detached.
4411 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
4413 uint64_t txg, dtl_max_txg;
4414 vdev_t *rvd = spa->spa_root_vdev;
4415 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
4417 char *oldvdpath, *newvdpath;
4421 ASSERT(spa_writeable(spa));
4423 txg = spa_vdev_enter(spa);
4425 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
4428 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4430 if (!oldvd->vdev_ops->vdev_op_leaf)
4431 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4433 pvd = oldvd->vdev_parent;
4435 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
4436 VDEV_ALLOC_ATTACH)) != 0)
4437 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4439 if (newrootvd->vdev_children != 1)
4440 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4442 newvd = newrootvd->vdev_child[0];
4444 if (!newvd->vdev_ops->vdev_op_leaf)
4445 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4447 if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
4448 return (spa_vdev_exit(spa, newrootvd, txg, error));
4451 * Spares can't replace logs
4453 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
4454 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4458 * For attach, the only allowable parent is a mirror or the root
4461 if (pvd->vdev_ops != &vdev_mirror_ops &&
4462 pvd->vdev_ops != &vdev_root_ops)
4463 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4465 pvops = &vdev_mirror_ops;
4468 * Active hot spares can only be replaced by inactive hot
4471 if (pvd->vdev_ops == &vdev_spare_ops &&
4472 oldvd->vdev_isspare &&
4473 !spa_has_spare(spa, newvd->vdev_guid))
4474 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4477 * If the source is a hot spare, and the parent isn't already a
4478 * spare, then we want to create a new hot spare. Otherwise, we
4479 * want to create a replacing vdev. The user is not allowed to
4480 * attach to a spared vdev child unless the 'isspare' state is
4481 * the same (spare replaces spare, non-spare replaces
4484 if (pvd->vdev_ops == &vdev_replacing_ops &&
4485 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
4486 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4487 } else if (pvd->vdev_ops == &vdev_spare_ops &&
4488 newvd->vdev_isspare != oldvd->vdev_isspare) {
4489 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4492 if (newvd->vdev_isspare)
4493 pvops = &vdev_spare_ops;
4495 pvops = &vdev_replacing_ops;
4499 * Make sure the new device is big enough.
4501 if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
4502 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
4505 * The new device cannot have a higher alignment requirement
4506 * than the top-level vdev.
4508 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4509 return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4512 * If this is an in-place replacement, update oldvd's path and devid
4513 * to make it distinguishable from newvd, and unopenable from now on.
4515 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4516 spa_strfree(oldvd->vdev_path);
4517 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4519 (void) sprintf(oldvd->vdev_path, "%s/%s",
4520 newvd->vdev_path, "old");
4521 if (oldvd->vdev_devid != NULL) {
4522 spa_strfree(oldvd->vdev_devid);
4523 oldvd->vdev_devid = NULL;
4527 /* mark the device being resilvered */
4528 newvd->vdev_resilvering = B_TRUE;
4531 * If the parent is not a mirror, or if we're replacing, insert the new
4532 * mirror/replacing/spare vdev above oldvd.
4534 if (pvd->vdev_ops != pvops)
4535 pvd = vdev_add_parent(oldvd, pvops);
4537 ASSERT(pvd->vdev_top->vdev_parent == rvd);
4538 ASSERT(pvd->vdev_ops == pvops);
4539 ASSERT(oldvd->vdev_parent == pvd);
4542 * Extract the new device from its root and add it to pvd.
4544 vdev_remove_child(newrootvd, newvd);
4545 newvd->vdev_id = pvd->vdev_children;
4546 newvd->vdev_crtxg = oldvd->vdev_crtxg;
4547 vdev_add_child(pvd, newvd);
4549 tvd = newvd->vdev_top;
4550 ASSERT(pvd->vdev_top == tvd);
4551 ASSERT(tvd->vdev_parent == rvd);
4553 vdev_config_dirty(tvd);
4556 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4557 * for any dmu_sync-ed blocks. It will propagate upward when
4558 * spa_vdev_exit() calls vdev_dtl_reassess().
4560 dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4562 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4563 dtl_max_txg - TXG_INITIAL);
4565 if (newvd->vdev_isspare) {
4566 spa_spare_activate(newvd);
4567 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
4570 oldvdpath = spa_strdup(oldvd->vdev_path);
4571 newvdpath = spa_strdup(newvd->vdev_path);
4572 newvd_isspare = newvd->vdev_isspare;
4575 * Mark newvd's DTL dirty in this txg.
4577 vdev_dirty(tvd, VDD_DTL, newvd, txg);
4580 * Restart the resilver
4582 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4587 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4589 spa_history_log_internal(LOG_POOL_VDEV_ATTACH, spa, NULL,
4590 "%s vdev=%s %s vdev=%s",
4591 replacing && newvd_isspare ? "spare in" :
4592 replacing ? "replace" : "attach", newvdpath,
4593 replacing ? "for" : "to", oldvdpath);
4595 spa_strfree(oldvdpath);
4596 spa_strfree(newvdpath);
4598 if (spa->spa_bootfs)
4599 spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
4605 * Detach a device from a mirror or replacing vdev.
4606 * If 'replace_done' is specified, only detach if the parent
4607 * is a replacing vdev.
4610 spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4614 vdev_t *rvd = spa->spa_root_vdev;
4615 vdev_t *vd, *pvd, *cvd, *tvd;
4616 boolean_t unspare = B_FALSE;
4617 uint64_t unspare_guid;
4620 ASSERT(spa_writeable(spa));
4622 txg = spa_vdev_enter(spa);
4624 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4627 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4629 if (!vd->vdev_ops->vdev_op_leaf)
4630 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4632 pvd = vd->vdev_parent;
4635 * If the parent/child relationship is not as expected, don't do it.
4636 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4637 * vdev that's replacing B with C. The user's intent in replacing
4638 * is to go from M(A,B) to M(A,C). If the user decides to cancel
4639 * the replace by detaching C, the expected behavior is to end up
4640 * M(A,B). But suppose that right after deciding to detach C,
4641 * the replacement of B completes. We would have M(A,C), and then
4642 * ask to detach C, which would leave us with just A -- not what
4643 * the user wanted. To prevent this, we make sure that the
4644 * parent/child relationship hasn't changed -- in this example,
4645 * that C's parent is still the replacing vdev R.
4647 if (pvd->vdev_guid != pguid && pguid != 0)
4648 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4651 * Only 'replacing' or 'spare' vdevs can be replaced.
4653 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4654 pvd->vdev_ops != &vdev_spare_ops)
4655 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4657 ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4658 spa_version(spa) >= SPA_VERSION_SPARES);
4661 * Only mirror, replacing, and spare vdevs support detach.
4663 if (pvd->vdev_ops != &vdev_replacing_ops &&
4664 pvd->vdev_ops != &vdev_mirror_ops &&
4665 pvd->vdev_ops != &vdev_spare_ops)
4666 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4669 * If this device has the only valid copy of some data,
4670 * we cannot safely detach it.
4672 if (vdev_dtl_required(vd))
4673 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4675 ASSERT(pvd->vdev_children >= 2);
4678 * If we are detaching the second disk from a replacing vdev, then
4679 * check to see if we changed the original vdev's path to have "/old"
4680 * at the end in spa_vdev_attach(). If so, undo that change now.
4682 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4683 vd->vdev_path != NULL) {
4684 size_t len = strlen(vd->vdev_path);
4686 for (int c = 0; c < pvd->vdev_children; c++) {
4687 cvd = pvd->vdev_child[c];
4689 if (cvd == vd || cvd->vdev_path == NULL)
4692 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4693 strcmp(cvd->vdev_path + len, "/old") == 0) {
4694 spa_strfree(cvd->vdev_path);
4695 cvd->vdev_path = spa_strdup(vd->vdev_path);
4702 * If we are detaching the original disk from a spare, then it implies
4703 * that the spare should become a real disk, and be removed from the
4704 * active spare list for the pool.
4706 if (pvd->vdev_ops == &vdev_spare_ops &&
4708 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4712 * Erase the disk labels so the disk can be used for other things.
4713 * This must be done after all other error cases are handled,
4714 * but before we disembowel vd (so we can still do I/O to it).
4715 * But if we can't do it, don't treat the error as fatal --
4716 * it may be that the unwritability of the disk is the reason
4717 * it's being detached!
4719 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4722 * Remove vd from its parent and compact the parent's children.
4724 vdev_remove_child(pvd, vd);
4725 vdev_compact_children(pvd);
4728 * Remember one of the remaining children so we can get tvd below.
4730 cvd = pvd->vdev_child[pvd->vdev_children - 1];
4733 * If we need to remove the remaining child from the list of hot spares,
4734 * do it now, marking the vdev as no longer a spare in the process.
4735 * We must do this before vdev_remove_parent(), because that can
4736 * change the GUID if it creates a new toplevel GUID. For a similar
4737 * reason, we must remove the spare now, in the same txg as the detach;
4738 * otherwise someone could attach a new sibling, change the GUID, and
4739 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4742 ASSERT(cvd->vdev_isspare);
4743 spa_spare_remove(cvd);
4744 unspare_guid = cvd->vdev_guid;
4745 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4746 cvd->vdev_unspare = B_TRUE;
4750 * If the parent mirror/replacing vdev only has one child,
4751 * the parent is no longer needed. Remove it from the tree.
4753 if (pvd->vdev_children == 1) {
4754 if (pvd->vdev_ops == &vdev_spare_ops)
4755 cvd->vdev_unspare = B_FALSE;
4756 vdev_remove_parent(cvd);
4757 cvd->vdev_resilvering = B_FALSE;
4762 * We don't set tvd until now because the parent we just removed
4763 * may have been the previous top-level vdev.
4765 tvd = cvd->vdev_top;
4766 ASSERT(tvd->vdev_parent == rvd);
4769 * Reevaluate the parent vdev state.
4771 vdev_propagate_state(cvd);
4774 * If the 'autoexpand' property is set on the pool then automatically
4775 * try to expand the size of the pool. For example if the device we
4776 * just detached was smaller than the others, it may be possible to
4777 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4778 * first so that we can obtain the updated sizes of the leaf vdevs.
4780 if (spa->spa_autoexpand) {
4782 vdev_expand(tvd, txg);
4785 vdev_config_dirty(tvd);
4788 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
4789 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4790 * But first make sure we're not on any *other* txg's DTL list, to
4791 * prevent vd from being accessed after it's freed.
4793 vdpath = spa_strdup(vd->vdev_path);
4794 for (int t = 0; t < TXG_SIZE; t++)
4795 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4796 vd->vdev_detached = B_TRUE;
4797 vdev_dirty(tvd, VDD_DTL, vd, txg);
4799 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
4801 /* hang on to the spa before we release the lock */
4802 spa_open_ref(spa, FTAG);
4804 error = spa_vdev_exit(spa, vd, txg, 0);
4806 spa_history_log_internal(LOG_POOL_VDEV_DETACH, spa, NULL,
4808 spa_strfree(vdpath);
4811 * If this was the removal of the original device in a hot spare vdev,
4812 * then we want to go through and remove the device from the hot spare
4813 * list of every other pool.
4816 spa_t *altspa = NULL;
4818 mutex_enter(&spa_namespace_lock);
4819 while ((altspa = spa_next(altspa)) != NULL) {
4820 if (altspa->spa_state != POOL_STATE_ACTIVE ||
4824 spa_open_ref(altspa, FTAG);
4825 mutex_exit(&spa_namespace_lock);
4826 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
4827 mutex_enter(&spa_namespace_lock);
4828 spa_close(altspa, FTAG);
4830 mutex_exit(&spa_namespace_lock);
4832 /* search the rest of the vdevs for spares to remove */
4833 spa_vdev_resilver_done(spa);
4836 /* all done with the spa; OK to release */
4837 mutex_enter(&spa_namespace_lock);
4838 spa_close(spa, FTAG);
4839 mutex_exit(&spa_namespace_lock);
4845 * Split a set of devices from their mirrors, and create a new pool from them.
4848 spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
4849 nvlist_t *props, boolean_t exp)
4852 uint64_t txg, *glist;
4854 uint_t c, children, lastlog;
4855 nvlist_t **child, *nvl, *tmp;
4857 char *altroot = NULL;
4858 vdev_t *rvd, **vml = NULL; /* vdev modify list */
4859 boolean_t activate_slog;
4861 ASSERT(spa_writeable(spa));
4863 txg = spa_vdev_enter(spa);
4865 /* clear the log and flush everything up to now */
4866 activate_slog = spa_passivate_log(spa);
4867 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4868 error = spa_offline_log(spa);
4869 txg = spa_vdev_config_enter(spa);
4872 spa_activate_log(spa);
4875 return (spa_vdev_exit(spa, NULL, txg, error));
4877 /* check new spa name before going any further */
4878 if (spa_lookup(newname) != NULL)
4879 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
4882 * scan through all the children to ensure they're all mirrors
4884 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
4885 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
4887 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4889 /* first, check to ensure we've got the right child count */
4890 rvd = spa->spa_root_vdev;
4892 for (c = 0; c < rvd->vdev_children; c++) {
4893 vdev_t *vd = rvd->vdev_child[c];
4895 /* don't count the holes & logs as children */
4896 if (vd->vdev_islog || vd->vdev_ishole) {
4904 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
4905 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4907 /* next, ensure no spare or cache devices are part of the split */
4908 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
4909 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
4910 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4912 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
4913 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
4915 /* then, loop over each vdev and validate it */
4916 for (c = 0; c < children; c++) {
4917 uint64_t is_hole = 0;
4919 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
4923 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
4924 spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
4932 /* which disk is going to be split? */
4933 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
4939 /* look it up in the spa */
4940 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
4941 if (vml[c] == NULL) {
4946 /* make sure there's nothing stopping the split */
4947 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
4948 vml[c]->vdev_islog ||
4949 vml[c]->vdev_ishole ||
4950 vml[c]->vdev_isspare ||
4951 vml[c]->vdev_isl2cache ||
4952 !vdev_writeable(vml[c]) ||
4953 vml[c]->vdev_children != 0 ||
4954 vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
4955 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
4960 if (vdev_dtl_required(vml[c])) {
4965 /* we need certain info from the top level */
4966 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
4967 vml[c]->vdev_top->vdev_ms_array) == 0);
4968 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
4969 vml[c]->vdev_top->vdev_ms_shift) == 0);
4970 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
4971 vml[c]->vdev_top->vdev_asize) == 0);
4972 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
4973 vml[c]->vdev_top->vdev_ashift) == 0);
4977 kmem_free(vml, children * sizeof (vdev_t *));
4978 kmem_free(glist, children * sizeof (uint64_t));
4979 return (spa_vdev_exit(spa, NULL, txg, error));
4982 /* stop writers from using the disks */
4983 for (c = 0; c < children; c++) {
4985 vml[c]->vdev_offline = B_TRUE;
4987 vdev_reopen(spa->spa_root_vdev);
4990 * Temporarily record the splitting vdevs in the spa config. This
4991 * will disappear once the config is regenerated.
4993 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4994 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
4995 glist, children) == 0);
4996 kmem_free(glist, children * sizeof (uint64_t));
4998 mutex_enter(&spa->spa_props_lock);
4999 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
5001 mutex_exit(&spa->spa_props_lock);
5002 spa->spa_config_splitting = nvl;
5003 vdev_config_dirty(spa->spa_root_vdev);
5005 /* configure and create the new pool */
5006 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
5007 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
5008 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
5009 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5010 spa_version(spa)) == 0);
5011 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
5012 spa->spa_config_txg) == 0);
5013 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
5014 spa_generate_guid(NULL)) == 0);
5015 (void) nvlist_lookup_string(props,
5016 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
5018 /* add the new pool to the namespace */
5019 newspa = spa_add(newname, config, altroot);
5020 newspa->spa_config_txg = spa->spa_config_txg;
5021 spa_set_log_state(newspa, SPA_LOG_CLEAR);
5023 /* release the spa config lock, retaining the namespace lock */
5024 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5026 if (zio_injection_enabled)
5027 zio_handle_panic_injection(spa, FTAG, 1);
5029 spa_activate(newspa, spa_mode_global);
5030 spa_async_suspend(newspa);
5033 /* mark that we are creating new spa by splitting */
5034 newspa->spa_splitting_newspa = B_TRUE;
5036 /* create the new pool from the disks of the original pool */
5037 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
5039 newspa->spa_splitting_newspa = B_FALSE;
5044 /* if that worked, generate a real config for the new pool */
5045 if (newspa->spa_root_vdev != NULL) {
5046 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
5047 NV_UNIQUE_NAME, KM_SLEEP) == 0);
5048 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
5049 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
5050 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
5055 if (props != NULL) {
5056 spa_configfile_set(newspa, props, B_FALSE);
5057 error = spa_prop_set(newspa, props);
5062 /* flush everything */
5063 txg = spa_vdev_config_enter(newspa);
5064 vdev_config_dirty(newspa->spa_root_vdev);
5065 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
5067 if (zio_injection_enabled)
5068 zio_handle_panic_injection(spa, FTAG, 2);
5070 spa_async_resume(newspa);
5072 /* finally, update the original pool's config */
5073 txg = spa_vdev_config_enter(spa);
5074 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
5075 error = dmu_tx_assign(tx, TXG_WAIT);
5078 for (c = 0; c < children; c++) {
5079 if (vml[c] != NULL) {
5082 spa_history_log_internal(LOG_POOL_VDEV_DETACH,
5088 vdev_config_dirty(spa->spa_root_vdev);
5089 spa->spa_config_splitting = NULL;
5093 (void) spa_vdev_exit(spa, NULL, txg, 0);
5095 if (zio_injection_enabled)
5096 zio_handle_panic_injection(spa, FTAG, 3);
5098 /* split is complete; log a history record */
5099 spa_history_log_internal(LOG_POOL_SPLIT, newspa, NULL,
5100 "split new pool %s from pool %s", newname, spa_name(spa));
5102 kmem_free(vml, children * sizeof (vdev_t *));
5104 /* if we're not going to mount the filesystems in userland, export */
5106 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
5113 spa_deactivate(newspa);
5116 txg = spa_vdev_config_enter(spa);
5118 /* re-online all offlined disks */
5119 for (c = 0; c < children; c++) {
5121 vml[c]->vdev_offline = B_FALSE;
5123 vdev_reopen(spa->spa_root_vdev);
5125 nvlist_free(spa->spa_config_splitting);
5126 spa->spa_config_splitting = NULL;
5127 (void) spa_vdev_exit(spa, NULL, txg, error);
5129 kmem_free(vml, children * sizeof (vdev_t *));
5134 spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
5136 for (int i = 0; i < count; i++) {
5139 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
5142 if (guid == target_guid)
5150 spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
5151 nvlist_t *dev_to_remove)
5153 nvlist_t **newdev = NULL;
5156 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
5158 for (int i = 0, j = 0; i < count; i++) {
5159 if (dev[i] == dev_to_remove)
5161 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
5164 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
5165 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
5167 for (int i = 0; i < count - 1; i++)
5168 nvlist_free(newdev[i]);
5171 kmem_free(newdev, (count - 1) * sizeof (void *));
5175 * Evacuate the device.
5178 spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
5183 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5184 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5185 ASSERT(vd == vd->vdev_top);
5188 * Evacuate the device. We don't hold the config lock as writer
5189 * since we need to do I/O but we do keep the
5190 * spa_namespace_lock held. Once this completes the device
5191 * should no longer have any blocks allocated on it.
5193 if (vd->vdev_islog) {
5194 if (vd->vdev_stat.vs_alloc != 0)
5195 error = spa_offline_log(spa);
5204 * The evacuation succeeded. Remove any remaining MOS metadata
5205 * associated with this vdev, and wait for these changes to sync.
5207 ASSERT0(vd->vdev_stat.vs_alloc);
5208 txg = spa_vdev_config_enter(spa);
5209 vd->vdev_removing = B_TRUE;
5210 vdev_dirty(vd, 0, NULL, txg);
5211 vdev_config_dirty(vd);
5212 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5218 * Complete the removal by cleaning up the namespace.
5221 spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
5223 vdev_t *rvd = spa->spa_root_vdev;
5224 uint64_t id = vd->vdev_id;
5225 boolean_t last_vdev = (id == (rvd->vdev_children - 1));
5227 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5228 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5229 ASSERT(vd == vd->vdev_top);
5232 * Only remove any devices which are empty.
5234 if (vd->vdev_stat.vs_alloc != 0)
5237 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5239 if (list_link_active(&vd->vdev_state_dirty_node))
5240 vdev_state_clean(vd);
5241 if (list_link_active(&vd->vdev_config_dirty_node))
5242 vdev_config_clean(vd);
5247 vdev_compact_children(rvd);
5249 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
5250 vdev_add_child(rvd, vd);
5252 vdev_config_dirty(rvd);
5255 * Reassess the health of our root vdev.
5261 * Remove a device from the pool -
5263 * Removing a device from the vdev namespace requires several steps
5264 * and can take a significant amount of time. As a result we use
5265 * the spa_vdev_config_[enter/exit] functions which allow us to
5266 * grab and release the spa_config_lock while still holding the namespace
5267 * lock. During each step the configuration is synced out.
5271 * Remove a device from the pool. Currently, this supports removing only hot
5272 * spares, slogs, and level 2 ARC devices.
5275 spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
5278 metaslab_group_t *mg;
5279 nvlist_t **spares, **l2cache, *nv;
5281 uint_t nspares, nl2cache;
5283 boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
5285 ASSERT(spa_writeable(spa));
5288 txg = spa_vdev_enter(spa);
5290 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5292 if (spa->spa_spares.sav_vdevs != NULL &&
5293 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5294 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
5295 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
5297 * Only remove the hot spare if it's not currently in use
5300 if (vd == NULL || unspare) {
5301 spa_vdev_remove_aux(spa->spa_spares.sav_config,
5302 ZPOOL_CONFIG_SPARES, spares, nspares, nv);
5303 spa_load_spares(spa);
5304 spa->spa_spares.sav_sync = B_TRUE;
5308 } else if (spa->spa_l2cache.sav_vdevs != NULL &&
5309 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5310 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
5311 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
5313 * Cache devices can always be removed.
5315 spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
5316 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
5317 spa_load_l2cache(spa);
5318 spa->spa_l2cache.sav_sync = B_TRUE;
5319 } else if (vd != NULL && vd->vdev_islog) {
5321 ASSERT(vd == vd->vdev_top);
5324 * XXX - Once we have bp-rewrite this should
5325 * become the common case.
5331 * Stop allocating from this vdev.
5333 metaslab_group_passivate(mg);
5336 * Wait for the youngest allocations and frees to sync,
5337 * and then wait for the deferral of those frees to finish.
5339 spa_vdev_config_exit(spa, NULL,
5340 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
5343 * Attempt to evacuate the vdev.
5345 error = spa_vdev_remove_evacuate(spa, vd);
5347 txg = spa_vdev_config_enter(spa);
5350 * If we couldn't evacuate the vdev, unwind.
5353 metaslab_group_activate(mg);
5354 return (spa_vdev_exit(spa, NULL, txg, error));
5358 * Clean up the vdev namespace.
5360 spa_vdev_remove_from_namespace(spa, vd);
5362 } else if (vd != NULL) {
5364 * Normal vdevs cannot be removed (yet).
5369 * There is no vdev of any kind with the specified guid.
5375 return (spa_vdev_exit(spa, NULL, txg, error));
5381 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5382 * current spared, so we can detach it.
5385 spa_vdev_resilver_done_hunt(vdev_t *vd)
5387 vdev_t *newvd, *oldvd;
5389 for (int c = 0; c < vd->vdev_children; c++) {
5390 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
5396 * Check for a completed replacement. We always consider the first
5397 * vdev in the list to be the oldest vdev, and the last one to be
5398 * the newest (see spa_vdev_attach() for how that works). In
5399 * the case where the newest vdev is faulted, we will not automatically
5400 * remove it after a resilver completes. This is OK as it will require
5401 * user intervention to determine which disk the admin wishes to keep.
5403 if (vd->vdev_ops == &vdev_replacing_ops) {
5404 ASSERT(vd->vdev_children > 1);
5406 newvd = vd->vdev_child[vd->vdev_children - 1];
5407 oldvd = vd->vdev_child[0];
5409 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
5410 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5411 !vdev_dtl_required(oldvd))
5416 * Check for a completed resilver with the 'unspare' flag set.
5418 if (vd->vdev_ops == &vdev_spare_ops) {
5419 vdev_t *first = vd->vdev_child[0];
5420 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
5422 if (last->vdev_unspare) {
5425 } else if (first->vdev_unspare) {
5432 if (oldvd != NULL &&
5433 vdev_dtl_empty(newvd, DTL_MISSING) &&
5434 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5435 !vdev_dtl_required(oldvd))
5439 * If there are more than two spares attached to a disk,
5440 * and those spares are not required, then we want to
5441 * attempt to free them up now so that they can be used
5442 * by other pools. Once we're back down to a single
5443 * disk+spare, we stop removing them.
5445 if (vd->vdev_children > 2) {
5446 newvd = vd->vdev_child[1];
5448 if (newvd->vdev_isspare && last->vdev_isspare &&
5449 vdev_dtl_empty(last, DTL_MISSING) &&
5450 vdev_dtl_empty(last, DTL_OUTAGE) &&
5451 !vdev_dtl_required(newvd))
5460 spa_vdev_resilver_done(spa_t *spa)
5462 vdev_t *vd, *pvd, *ppvd;
5463 uint64_t guid, sguid, pguid, ppguid;
5465 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5467 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
5468 pvd = vd->vdev_parent;
5469 ppvd = pvd->vdev_parent;
5470 guid = vd->vdev_guid;
5471 pguid = pvd->vdev_guid;
5472 ppguid = ppvd->vdev_guid;
5475 * If we have just finished replacing a hot spared device, then
5476 * we need to detach the parent's first child (the original hot
5479 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
5480 ppvd->vdev_children == 2) {
5481 ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
5482 sguid = ppvd->vdev_child[1]->vdev_guid;
5484 spa_config_exit(spa, SCL_ALL, FTAG);
5485 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
5487 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
5489 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5492 spa_config_exit(spa, SCL_ALL, FTAG);
5496 * Update the stored path or FRU for this vdev.
5499 spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
5503 boolean_t sync = B_FALSE;
5505 ASSERT(spa_writeable(spa));
5507 spa_vdev_state_enter(spa, SCL_ALL);
5509 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
5510 return (spa_vdev_state_exit(spa, NULL, ENOENT));
5512 if (!vd->vdev_ops->vdev_op_leaf)
5513 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
5516 if (strcmp(value, vd->vdev_path) != 0) {
5517 spa_strfree(vd->vdev_path);
5518 vd->vdev_path = spa_strdup(value);
5522 if (vd->vdev_fru == NULL) {
5523 vd->vdev_fru = spa_strdup(value);
5525 } else if (strcmp(value, vd->vdev_fru) != 0) {
5526 spa_strfree(vd->vdev_fru);
5527 vd->vdev_fru = spa_strdup(value);
5532 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
5536 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
5538 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5542 spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5544 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5548 * ==========================================================================
5550 * ==========================================================================
5554 spa_scan_stop(spa_t *spa)
5556 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5557 if (dsl_scan_resilvering(spa->spa_dsl_pool))
5559 return (dsl_scan_cancel(spa->spa_dsl_pool));
5563 spa_scan(spa_t *spa, pool_scan_func_t func)
5565 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5567 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5571 * If a resilver was requested, but there is no DTL on a
5572 * writeable leaf device, we have nothing to do.
5574 if (func == POOL_SCAN_RESILVER &&
5575 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5576 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5580 return (dsl_scan(spa->spa_dsl_pool, func));
5584 * ==========================================================================
5585 * SPA async task processing
5586 * ==========================================================================
5590 spa_async_remove(spa_t *spa, vdev_t *vd)
5592 if (vd->vdev_remove_wanted) {
5593 vd->vdev_remove_wanted = B_FALSE;
5594 vd->vdev_delayed_close = B_FALSE;
5595 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5598 * We want to clear the stats, but we don't want to do a full
5599 * vdev_clear() as that will cause us to throw away
5600 * degraded/faulted state as well as attempt to reopen the
5601 * device, all of which is a waste.
5603 vd->vdev_stat.vs_read_errors = 0;
5604 vd->vdev_stat.vs_write_errors = 0;
5605 vd->vdev_stat.vs_checksum_errors = 0;
5607 vdev_state_dirty(vd->vdev_top);
5610 for (int c = 0; c < vd->vdev_children; c++)
5611 spa_async_remove(spa, vd->vdev_child[c]);
5615 spa_async_probe(spa_t *spa, vdev_t *vd)
5617 if (vd->vdev_probe_wanted) {
5618 vd->vdev_probe_wanted = B_FALSE;
5619 vdev_reopen(vd); /* vdev_open() does the actual probe */
5622 for (int c = 0; c < vd->vdev_children; c++)
5623 spa_async_probe(spa, vd->vdev_child[c]);
5627 spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5633 if (!spa->spa_autoexpand)
5636 for (int c = 0; c < vd->vdev_children; c++) {
5637 vdev_t *cvd = vd->vdev_child[c];
5638 spa_async_autoexpand(spa, cvd);
5641 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5644 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
5645 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
5647 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5648 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
5650 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
5651 ESC_ZFS_VDEV_AUTOEXPAND, attr, &eid, DDI_SLEEP);
5654 kmem_free(physpath, MAXPATHLEN);
5658 spa_async_thread(void *arg)
5663 ASSERT(spa->spa_sync_on);
5665 mutex_enter(&spa->spa_async_lock);
5666 tasks = spa->spa_async_tasks;
5667 spa->spa_async_tasks = 0;
5668 mutex_exit(&spa->spa_async_lock);
5671 * See if the config needs to be updated.
5673 if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5674 uint64_t old_space, new_space;
5676 mutex_enter(&spa_namespace_lock);
5677 old_space = metaslab_class_get_space(spa_normal_class(spa));
5678 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5679 new_space = metaslab_class_get_space(spa_normal_class(spa));
5680 mutex_exit(&spa_namespace_lock);
5683 * If the pool grew as a result of the config update,
5684 * then log an internal history event.
5686 if (new_space != old_space) {
5687 spa_history_log_internal(LOG_POOL_VDEV_ONLINE,
5689 "pool '%s' size: %llu(+%llu)",
5690 spa_name(spa), new_space, new_space - old_space);
5695 * See if any devices need to be marked REMOVED.
5697 if (tasks & SPA_ASYNC_REMOVE) {
5698 spa_vdev_state_enter(spa, SCL_NONE);
5699 spa_async_remove(spa, spa->spa_root_vdev);
5700 for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
5701 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5702 for (int i = 0; i < spa->spa_spares.sav_count; i++)
5703 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5704 (void) spa_vdev_state_exit(spa, NULL, 0);
5707 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5708 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5709 spa_async_autoexpand(spa, spa->spa_root_vdev);
5710 spa_config_exit(spa, SCL_CONFIG, FTAG);
5714 * See if any devices need to be probed.
5716 if (tasks & SPA_ASYNC_PROBE) {
5717 spa_vdev_state_enter(spa, SCL_NONE);
5718 spa_async_probe(spa, spa->spa_root_vdev);
5719 (void) spa_vdev_state_exit(spa, NULL, 0);
5723 * If any devices are done replacing, detach them.
5725 if (tasks & SPA_ASYNC_RESILVER_DONE)
5726 spa_vdev_resilver_done(spa);
5729 * Kick off a resilver.
5731 if (tasks & SPA_ASYNC_RESILVER)
5732 dsl_resilver_restart(spa->spa_dsl_pool, 0);
5735 * Let the world know that we're done.
5737 mutex_enter(&spa->spa_async_lock);
5738 spa->spa_async_thread = NULL;
5739 cv_broadcast(&spa->spa_async_cv);
5740 mutex_exit(&spa->spa_async_lock);
5745 spa_async_suspend(spa_t *spa)
5747 mutex_enter(&spa->spa_async_lock);
5748 spa->spa_async_suspended++;
5749 while (spa->spa_async_thread != NULL)
5750 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5751 mutex_exit(&spa->spa_async_lock);
5755 spa_async_resume(spa_t *spa)
5757 mutex_enter(&spa->spa_async_lock);
5758 ASSERT(spa->spa_async_suspended != 0);
5759 spa->spa_async_suspended--;
5760 mutex_exit(&spa->spa_async_lock);
5764 spa_async_dispatch(spa_t *spa)
5766 mutex_enter(&spa->spa_async_lock);
5767 if (spa->spa_async_tasks && !spa->spa_async_suspended &&
5768 spa->spa_async_thread == NULL &&
5769 rootdir != NULL && !vn_is_readonly(rootdir))
5770 spa->spa_async_thread = thread_create(NULL, 0,
5771 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
5772 mutex_exit(&spa->spa_async_lock);
5776 spa_async_request(spa_t *spa, int task)
5778 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
5779 mutex_enter(&spa->spa_async_lock);
5780 spa->spa_async_tasks |= task;
5781 mutex_exit(&spa->spa_async_lock);
5785 * ==========================================================================
5786 * SPA syncing routines
5787 * ==========================================================================
5791 bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5794 bpobj_enqueue(bpo, bp, tx);
5799 spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5803 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
5804 BP_GET_PSIZE(bp), zio->io_flags));
5809 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
5811 char *packed = NULL;
5816 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
5819 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5820 * information. This avoids the dbuf_will_dirty() path and
5821 * saves us a pre-read to get data we don't actually care about.
5823 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
5824 packed = kmem_alloc(bufsize, KM_SLEEP);
5826 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
5828 bzero(packed + nvsize, bufsize - nvsize);
5830 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
5832 kmem_free(packed, bufsize);
5834 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
5835 dmu_buf_will_dirty(db, tx);
5836 *(uint64_t *)db->db_data = nvsize;
5837 dmu_buf_rele(db, FTAG);
5841 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
5842 const char *config, const char *entry)
5852 * Update the MOS nvlist describing the list of available devices.
5853 * spa_validate_aux() will have already made sure this nvlist is
5854 * valid and the vdevs are labeled appropriately.
5856 if (sav->sav_object == 0) {
5857 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
5858 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
5859 sizeof (uint64_t), tx);
5860 VERIFY(zap_update(spa->spa_meta_objset,
5861 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
5862 &sav->sav_object, tx) == 0);
5865 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5866 if (sav->sav_count == 0) {
5867 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
5869 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
5870 for (i = 0; i < sav->sav_count; i++)
5871 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
5872 B_FALSE, VDEV_CONFIG_L2CACHE);
5873 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
5874 sav->sav_count) == 0);
5875 for (i = 0; i < sav->sav_count; i++)
5876 nvlist_free(list[i]);
5877 kmem_free(list, sav->sav_count * sizeof (void *));
5880 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
5881 nvlist_free(nvroot);
5883 sav->sav_sync = B_FALSE;
5887 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
5891 if (list_is_empty(&spa->spa_config_dirty_list))
5894 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5896 config = spa_config_generate(spa, spa->spa_root_vdev,
5897 dmu_tx_get_txg(tx), B_FALSE);
5899 spa_config_exit(spa, SCL_STATE, FTAG);
5901 if (spa->spa_config_syncing)
5902 nvlist_free(spa->spa_config_syncing);
5903 spa->spa_config_syncing = config;
5905 spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
5909 spa_sync_version(void *arg1, void *arg2, dmu_tx_t *tx)
5912 uint64_t version = *(uint64_t *)arg2;
5915 * Setting the version is special cased when first creating the pool.
5917 ASSERT(tx->tx_txg != TXG_INITIAL);
5919 ASSERT(version <= SPA_VERSION);
5920 ASSERT(version >= spa_version(spa));
5922 spa->spa_uberblock.ub_version = version;
5923 vdev_config_dirty(spa->spa_root_vdev);
5927 * Set zpool properties.
5930 spa_sync_props(void *arg1, void *arg2, dmu_tx_t *tx)
5933 objset_t *mos = spa->spa_meta_objset;
5934 nvlist_t *nvp = arg2;
5935 nvpair_t *elem = NULL;
5937 mutex_enter(&spa->spa_props_lock);
5939 while ((elem = nvlist_next_nvpair(nvp, elem))) {
5941 char *strval, *fname;
5943 const char *propname;
5944 zprop_type_t proptype;
5945 zfeature_info_t *feature;
5947 switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
5950 * We checked this earlier in spa_prop_validate().
5952 ASSERT(zpool_prop_feature(nvpair_name(elem)));
5954 fname = strchr(nvpair_name(elem), '@') + 1;
5955 VERIFY3U(0, ==, zfeature_lookup_name(fname, &feature));
5957 spa_feature_enable(spa, feature, tx);
5960 case ZPOOL_PROP_VERSION:
5961 VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5963 * The version is synced seperatly before other
5964 * properties and should be correct by now.
5966 ASSERT3U(spa_version(spa), >=, intval);
5969 case ZPOOL_PROP_ALTROOT:
5971 * 'altroot' is a non-persistent property. It should
5972 * have been set temporarily at creation or import time.
5974 ASSERT(spa->spa_root != NULL);
5977 case ZPOOL_PROP_READONLY:
5978 case ZPOOL_PROP_CACHEFILE:
5980 * 'readonly' and 'cachefile' are also non-persisitent
5984 case ZPOOL_PROP_COMMENT:
5985 VERIFY(nvpair_value_string(elem, &strval) == 0);
5986 if (spa->spa_comment != NULL)
5987 spa_strfree(spa->spa_comment);
5988 spa->spa_comment = spa_strdup(strval);
5990 * We need to dirty the configuration on all the vdevs
5991 * so that their labels get updated. It's unnecessary
5992 * to do this for pool creation since the vdev's
5993 * configuratoin has already been dirtied.
5995 if (tx->tx_txg != TXG_INITIAL)
5996 vdev_config_dirty(spa->spa_root_vdev);
6000 * Set pool property values in the poolprops mos object.
6002 if (spa->spa_pool_props_object == 0) {
6003 spa->spa_pool_props_object =
6004 zap_create_link(mos, DMU_OT_POOL_PROPS,
6005 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
6009 /* normalize the property name */
6010 propname = zpool_prop_to_name(prop);
6011 proptype = zpool_prop_get_type(prop);
6013 if (nvpair_type(elem) == DATA_TYPE_STRING) {
6014 ASSERT(proptype == PROP_TYPE_STRING);
6015 VERIFY(nvpair_value_string(elem, &strval) == 0);
6016 VERIFY(zap_update(mos,
6017 spa->spa_pool_props_object, propname,
6018 1, strlen(strval) + 1, strval, tx) == 0);
6020 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
6021 VERIFY(nvpair_value_uint64(elem, &intval) == 0);
6023 if (proptype == PROP_TYPE_INDEX) {
6025 VERIFY(zpool_prop_index_to_string(
6026 prop, intval, &unused) == 0);
6028 VERIFY(zap_update(mos,
6029 spa->spa_pool_props_object, propname,
6030 8, 1, &intval, tx) == 0);
6032 ASSERT(0); /* not allowed */
6036 case ZPOOL_PROP_DELEGATION:
6037 spa->spa_delegation = intval;
6039 case ZPOOL_PROP_BOOTFS:
6040 spa->spa_bootfs = intval;
6042 case ZPOOL_PROP_FAILUREMODE:
6043 spa->spa_failmode = intval;
6045 case ZPOOL_PROP_AUTOEXPAND:
6046 spa->spa_autoexpand = intval;
6047 if (tx->tx_txg != TXG_INITIAL)
6048 spa_async_request(spa,
6049 SPA_ASYNC_AUTOEXPAND);
6051 case ZPOOL_PROP_DEDUPDITTO:
6052 spa->spa_dedup_ditto = intval;
6059 /* log internal history if this is not a zpool create */
6060 if (spa_version(spa) >= SPA_VERSION_ZPOOL_HISTORY &&
6061 tx->tx_txg != TXG_INITIAL) {
6062 spa_history_log_internal(LOG_POOL_PROPSET,
6063 spa, tx, "%s %lld %s",
6064 nvpair_name(elem), intval, spa_name(spa));
6068 mutex_exit(&spa->spa_props_lock);
6072 * Perform one-time upgrade on-disk changes. spa_version() does not
6073 * reflect the new version this txg, so there must be no changes this
6074 * txg to anything that the upgrade code depends on after it executes.
6075 * Therefore this must be called after dsl_pool_sync() does the sync
6079 spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
6081 dsl_pool_t *dp = spa->spa_dsl_pool;
6083 ASSERT(spa->spa_sync_pass == 1);
6085 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
6086 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
6087 dsl_pool_create_origin(dp, tx);
6089 /* Keeping the origin open increases spa_minref */
6090 spa->spa_minref += 3;
6093 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
6094 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
6095 dsl_pool_upgrade_clones(dp, tx);
6098 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
6099 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
6100 dsl_pool_upgrade_dir_clones(dp, tx);
6102 /* Keeping the freedir open increases spa_minref */
6103 spa->spa_minref += 3;
6106 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES &&
6107 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
6108 spa_feature_create_zap_objects(spa, tx);
6113 * Sync the specified transaction group. New blocks may be dirtied as
6114 * part of the process, so we iterate until it converges.
6117 spa_sync(spa_t *spa, uint64_t txg)
6119 dsl_pool_t *dp = spa->spa_dsl_pool;
6120 objset_t *mos = spa->spa_meta_objset;
6121 bpobj_t *defer_bpo = &spa->spa_deferred_bpobj;
6122 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
6123 vdev_t *rvd = spa->spa_root_vdev;
6128 VERIFY(spa_writeable(spa));
6131 * Lock out configuration changes.
6133 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
6135 spa->spa_syncing_txg = txg;
6136 spa->spa_sync_pass = 0;
6139 * If there are any pending vdev state changes, convert them
6140 * into config changes that go out with this transaction group.
6142 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6143 while (list_head(&spa->spa_state_dirty_list) != NULL) {
6145 * We need the write lock here because, for aux vdevs,
6146 * calling vdev_config_dirty() modifies sav_config.
6147 * This is ugly and will become unnecessary when we
6148 * eliminate the aux vdev wart by integrating all vdevs
6149 * into the root vdev tree.
6151 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6152 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
6153 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
6154 vdev_state_clean(vd);
6155 vdev_config_dirty(vd);
6157 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6158 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
6160 spa_config_exit(spa, SCL_STATE, FTAG);
6162 tx = dmu_tx_create_assigned(dp, txg);
6165 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
6166 * set spa_deflate if we have no raid-z vdevs.
6168 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
6169 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
6172 for (i = 0; i < rvd->vdev_children; i++) {
6173 vd = rvd->vdev_child[i];
6174 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
6177 if (i == rvd->vdev_children) {
6178 spa->spa_deflate = TRUE;
6179 VERIFY(0 == zap_add(spa->spa_meta_objset,
6180 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
6181 sizeof (uint64_t), 1, &spa->spa_deflate, tx));
6186 * If anything has changed in this txg, or if someone is waiting
6187 * for this txg to sync (eg, spa_vdev_remove()), push the
6188 * deferred frees from the previous txg. If not, leave them
6189 * alone so that we don't generate work on an otherwise idle
6192 if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
6193 !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
6194 !txg_list_empty(&dp->dp_sync_tasks, txg) ||
6195 ((dsl_scan_active(dp->dp_scan) ||
6196 txg_sync_waiting(dp)) && !spa_shutting_down(spa))) {
6197 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6198 VERIFY3U(bpobj_iterate(defer_bpo,
6199 spa_free_sync_cb, zio, tx), ==, 0);
6200 VERIFY0(zio_wait(zio));
6204 * Iterate to convergence.
6207 int pass = ++spa->spa_sync_pass;
6209 spa_sync_config_object(spa, tx);
6210 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
6211 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
6212 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
6213 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
6214 spa_errlog_sync(spa, txg);
6215 dsl_pool_sync(dp, txg);
6217 if (pass <= SYNC_PASS_DEFERRED_FREE) {
6218 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6219 bplist_iterate(free_bpl, spa_free_sync_cb,
6221 VERIFY(zio_wait(zio) == 0);
6223 bplist_iterate(free_bpl, bpobj_enqueue_cb,
6228 dsl_scan_sync(dp, tx);
6230 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
6234 spa_sync_upgrades(spa, tx);
6236 } while (dmu_objset_is_dirty(mos, txg));
6239 * Rewrite the vdev configuration (which includes the uberblock)
6240 * to commit the transaction group.
6242 * If there are no dirty vdevs, we sync the uberblock to a few
6243 * random top-level vdevs that are known to be visible in the
6244 * config cache (see spa_vdev_add() for a complete description).
6245 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
6249 * We hold SCL_STATE to prevent vdev open/close/etc.
6250 * while we're attempting to write the vdev labels.
6252 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6254 if (list_is_empty(&spa->spa_config_dirty_list)) {
6255 vdev_t *svd[SPA_DVAS_PER_BP];
6257 int children = rvd->vdev_children;
6258 int c0 = spa_get_random(children);
6260 for (int c = 0; c < children; c++) {
6261 vd = rvd->vdev_child[(c0 + c) % children];
6262 if (vd->vdev_ms_array == 0 || vd->vdev_islog)
6264 svd[svdcount++] = vd;
6265 if (svdcount == SPA_DVAS_PER_BP)
6268 error = vdev_config_sync(svd, svdcount, txg, B_FALSE);
6270 error = vdev_config_sync(svd, svdcount, txg,
6273 error = vdev_config_sync(rvd->vdev_child,
6274 rvd->vdev_children, txg, B_FALSE);
6276 error = vdev_config_sync(rvd->vdev_child,
6277 rvd->vdev_children, txg, B_TRUE);
6281 spa->spa_last_synced_guid = rvd->vdev_guid;
6283 spa_config_exit(spa, SCL_STATE, FTAG);
6287 zio_suspend(spa, NULL);
6288 zio_resume_wait(spa);
6293 * Clear the dirty config list.
6295 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
6296 vdev_config_clean(vd);
6299 * Now that the new config has synced transactionally,
6300 * let it become visible to the config cache.
6302 if (spa->spa_config_syncing != NULL) {
6303 spa_config_set(spa, spa->spa_config_syncing);
6304 spa->spa_config_txg = txg;
6305 spa->spa_config_syncing = NULL;
6308 spa->spa_ubsync = spa->spa_uberblock;
6310 dsl_pool_sync_done(dp, txg);
6313 * Update usable space statistics.
6315 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
6316 vdev_sync_done(vd, txg);
6318 spa_update_dspace(spa);
6321 * It had better be the case that we didn't dirty anything
6322 * since vdev_config_sync().
6324 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
6325 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
6326 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
6328 spa->spa_sync_pass = 0;
6330 spa_config_exit(spa, SCL_CONFIG, FTAG);
6332 spa_handle_ignored_writes(spa);
6335 * If any async tasks have been requested, kick them off.
6337 spa_async_dispatch(spa);
6341 * Sync all pools. We don't want to hold the namespace lock across these
6342 * operations, so we take a reference on the spa_t and drop the lock during the
6346 spa_sync_allpools(void)
6349 mutex_enter(&spa_namespace_lock);
6350 while ((spa = spa_next(spa)) != NULL) {
6351 if (spa_state(spa) != POOL_STATE_ACTIVE ||
6352 !spa_writeable(spa) || spa_suspended(spa))
6354 spa_open_ref(spa, FTAG);
6355 mutex_exit(&spa_namespace_lock);
6356 txg_wait_synced(spa_get_dsl(spa), 0);
6357 mutex_enter(&spa_namespace_lock);
6358 spa_close(spa, FTAG);
6360 mutex_exit(&spa_namespace_lock);
6364 * ==========================================================================
6365 * Miscellaneous routines
6366 * ==========================================================================
6370 * Remove all pools in the system.
6378 * Remove all cached state. All pools should be closed now,
6379 * so every spa in the AVL tree should be unreferenced.
6381 mutex_enter(&spa_namespace_lock);
6382 while ((spa = spa_next(NULL)) != NULL) {
6384 * Stop async tasks. The async thread may need to detach
6385 * a device that's been replaced, which requires grabbing
6386 * spa_namespace_lock, so we must drop it here.
6388 spa_open_ref(spa, FTAG);
6389 mutex_exit(&spa_namespace_lock);
6390 spa_async_suspend(spa);
6391 mutex_enter(&spa_namespace_lock);
6392 spa_close(spa, FTAG);
6394 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
6396 spa_deactivate(spa);
6400 mutex_exit(&spa_namespace_lock);
6404 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
6409 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
6413 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
6414 vd = spa->spa_l2cache.sav_vdevs[i];
6415 if (vd->vdev_guid == guid)
6419 for (i = 0; i < spa->spa_spares.sav_count; i++) {
6420 vd = spa->spa_spares.sav_vdevs[i];
6421 if (vd->vdev_guid == guid)
6430 spa_upgrade(spa_t *spa, uint64_t version)
6432 ASSERT(spa_writeable(spa));
6434 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6437 * This should only be called for a non-faulted pool, and since a
6438 * future version would result in an unopenable pool, this shouldn't be
6441 ASSERT(spa->spa_uberblock.ub_version <= SPA_VERSION);
6442 ASSERT(version >= spa->spa_uberblock.ub_version);
6444 spa->spa_uberblock.ub_version = version;
6445 vdev_config_dirty(spa->spa_root_vdev);
6447 spa_config_exit(spa, SCL_ALL, FTAG);
6449 txg_wait_synced(spa_get_dsl(spa), 0);
6453 spa_has_spare(spa_t *spa, uint64_t guid)
6457 spa_aux_vdev_t *sav = &spa->spa_spares;
6459 for (i = 0; i < sav->sav_count; i++)
6460 if (sav->sav_vdevs[i]->vdev_guid == guid)
6463 for (i = 0; i < sav->sav_npending; i++) {
6464 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
6465 &spareguid) == 0 && spareguid == guid)
6473 * Check if a pool has an active shared spare device.
6474 * Note: reference count of an active spare is 2, as a spare and as a replace
6477 spa_has_active_shared_spare(spa_t *spa)
6481 spa_aux_vdev_t *sav = &spa->spa_spares;
6483 for (i = 0; i < sav->sav_count; i++) {
6484 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
6485 &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
6494 * Post a sysevent corresponding to the given event. The 'name' must be one of
6495 * the event definitions in sys/sysevent/eventdefs.h. The payload will be
6496 * filled in from the spa and (optionally) the vdev. This doesn't do anything
6497 * in the userland libzpool, as we don't want consumers to misinterpret ztest
6498 * or zdb as real changes.
6501 spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
6505 sysevent_attr_list_t *attr = NULL;
6506 sysevent_value_t value;
6509 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
6512 value.value_type = SE_DATA_TYPE_STRING;
6513 value.value.sv_string = spa_name(spa);
6514 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
6517 value.value_type = SE_DATA_TYPE_UINT64;
6518 value.value.sv_uint64 = spa_guid(spa);
6519 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
6523 value.value_type = SE_DATA_TYPE_UINT64;
6524 value.value.sv_uint64 = vd->vdev_guid;
6525 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
6529 if (vd->vdev_path) {
6530 value.value_type = SE_DATA_TYPE_STRING;
6531 value.value.sv_string = vd->vdev_path;
6532 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
6533 &value, SE_SLEEP) != 0)
6538 if (sysevent_attach_attributes(ev, attr) != 0)
6542 (void) log_sysevent(ev, SE_SLEEP, &eid);
6546 sysevent_free_attr(attr);