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) 2013 by Delphix. All rights reserved.
25 * Copyright 2013 Nexenta Systems, Inc. All rights reserved.
29 * This file contains all the routines used when modifying on-disk SPA state.
30 * This includes opening, importing, destroying, exporting a pool, and syncing a
34 #include <sys/zfs_context.h>
35 #include <sys/fm/fs/zfs.h>
36 #include <sys/spa_impl.h>
38 #include <sys/zio_checksum.h>
40 #include <sys/dmu_tx.h>
44 #include <sys/vdev_impl.h>
45 #include <sys/vdev_disk.h>
46 #include <sys/metaslab.h>
47 #include <sys/metaslab_impl.h>
48 #include <sys/uberblock_impl.h>
51 #include <sys/dmu_traverse.h>
52 #include <sys/dmu_objset.h>
53 #include <sys/unique.h>
54 #include <sys/dsl_pool.h>
55 #include <sys/dsl_dataset.h>
56 #include <sys/dsl_dir.h>
57 #include <sys/dsl_prop.h>
58 #include <sys/dsl_synctask.h>
59 #include <sys/fs/zfs.h>
61 #include <sys/callb.h>
62 #include <sys/systeminfo.h>
63 #include <sys/spa_boot.h>
64 #include <sys/zfs_ioctl.h>
65 #include <sys/dsl_scan.h>
66 #include <sys/zfeature.h>
67 #include <sys/dsl_destroy.h>
71 #include <sys/bootprops.h>
72 #include <sys/callb.h>
73 #include <sys/cpupart.h>
75 #include <sys/sysdc.h>
80 #include "zfs_comutil.h"
82 typedef enum zti_modes {
83 ZTI_MODE_FIXED, /* value is # of threads (min 1) */
84 ZTI_MODE_ONLINE_PERCENT, /* value is % of online CPUs */
85 ZTI_MODE_BATCH, /* cpu-intensive; value is ignored */
86 ZTI_MODE_NULL, /* don't create a taskq */
90 #define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) }
91 #define ZTI_PCT(n) { ZTI_MODE_ONLINE_PERCENT, (n), 1 }
92 #define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 }
93 #define ZTI_NULL { ZTI_MODE_NULL, 0, 0 }
95 #define ZTI_N(n) ZTI_P(n, 1)
96 #define ZTI_ONE ZTI_N(1)
98 typedef struct zio_taskq_info {
104 static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
105 "iss", "iss_h", "int", "int_h"
109 * This table defines the taskq settings for each ZFS I/O type. When
110 * initializing a pool, we use this table to create an appropriately sized
111 * taskq. Some operations are low volume and therefore have a small, static
112 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE
113 * macros. Other operations process a large amount of data; the ZTI_BATCH
114 * macro causes us to create a taskq oriented for throughput. Some operations
115 * are so high frequency and short-lived that the taskq itself can become a a
116 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an
117 * additional degree of parallelism specified by the number of threads per-
118 * taskq and the number of taskqs; when dispatching an event in this case, the
119 * particular taskq is chosen at random.
121 * The different taskq priorities are to handle the different contexts (issue
122 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that
123 * need to be handled with minimum delay.
125 const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
126 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
127 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* NULL */
128 { ZTI_N(8), ZTI_NULL, ZTI_BATCH, ZTI_NULL }, /* READ */
129 { ZTI_BATCH, ZTI_N(5), ZTI_N(16), ZTI_N(5) }, /* WRITE */
130 { ZTI_P(4, 8), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FREE */
131 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* CLAIM */
132 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* IOCTL */
135 static void spa_sync_version(void *arg, dmu_tx_t *tx);
136 static void spa_sync_props(void *arg, dmu_tx_t *tx);
137 static boolean_t spa_has_active_shared_spare(spa_t *spa);
138 static inline int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
139 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
141 static void spa_vdev_resilver_done(spa_t *spa);
143 uint_t zio_taskq_batch_pct = 100; /* 1 thread per cpu in pset */
144 id_t zio_taskq_psrset_bind = PS_NONE;
145 boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */
146 uint_t zio_taskq_basedc = 80; /* base duty cycle */
148 boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */
151 * This (illegal) pool name is used when temporarily importing a spa_t in order
152 * to get the vdev stats associated with the imported devices.
154 #define TRYIMPORT_NAME "$import"
157 * ==========================================================================
158 * SPA properties routines
159 * ==========================================================================
163 * Add a (source=src, propname=propval) list to an nvlist.
166 spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
167 uint64_t intval, zprop_source_t src)
169 const char *propname = zpool_prop_to_name(prop);
172 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_PUSHPAGE) == 0);
173 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
176 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
178 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
180 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
181 nvlist_free(propval);
185 * Get property values from the spa configuration.
188 spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
190 vdev_t *rvd = spa->spa_root_vdev;
191 dsl_pool_t *pool = spa->spa_dsl_pool;
195 uint64_t cap, version;
196 zprop_source_t src = ZPROP_SRC_NONE;
197 spa_config_dirent_t *dp;
200 ASSERT(MUTEX_HELD(&spa->spa_props_lock));
203 alloc = metaslab_class_get_alloc(spa_normal_class(spa));
204 size = metaslab_class_get_space(spa_normal_class(spa));
205 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
206 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
207 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
208 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
212 for (c = 0; c < rvd->vdev_children; c++) {
213 vdev_t *tvd = rvd->vdev_child[c];
214 space += tvd->vdev_max_asize - tvd->vdev_asize;
216 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, space,
219 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
220 (spa_mode(spa) == FREAD), src);
222 cap = (size == 0) ? 0 : (alloc * 100 / size);
223 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
225 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
226 ddt_get_pool_dedup_ratio(spa), src);
228 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
229 rvd->vdev_state, src);
231 version = spa_version(spa);
232 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
233 src = ZPROP_SRC_DEFAULT;
235 src = ZPROP_SRC_LOCAL;
236 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
240 dsl_dir_t *freedir = pool->dp_free_dir;
243 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
244 * when opening pools before this version freedir will be NULL.
246 if (freedir != NULL) {
247 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
248 freedir->dd_phys->dd_used_bytes, src);
250 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
255 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
257 if (spa->spa_comment != NULL) {
258 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
262 if (spa->spa_root != NULL)
263 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
266 if ((dp = list_head(&spa->spa_config_list)) != NULL) {
267 if (dp->scd_path == NULL) {
268 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
269 "none", 0, ZPROP_SRC_LOCAL);
270 } else if (strcmp(dp->scd_path, spa_config_path) != 0) {
271 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
272 dp->scd_path, 0, ZPROP_SRC_LOCAL);
278 * Get zpool property values.
281 spa_prop_get(spa_t *spa, nvlist_t **nvp)
283 objset_t *mos = spa->spa_meta_objset;
288 err = nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_PUSHPAGE);
292 mutex_enter(&spa->spa_props_lock);
295 * Get properties from the spa config.
297 spa_prop_get_config(spa, nvp);
299 /* If no pool property object, no more prop to get. */
300 if (mos == NULL || spa->spa_pool_props_object == 0) {
301 mutex_exit(&spa->spa_props_lock);
306 * Get properties from the MOS pool property object.
308 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
309 (err = zap_cursor_retrieve(&zc, &za)) == 0;
310 zap_cursor_advance(&zc)) {
313 zprop_source_t src = ZPROP_SRC_DEFAULT;
316 if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
319 switch (za.za_integer_length) {
321 /* integer property */
322 if (za.za_first_integer !=
323 zpool_prop_default_numeric(prop))
324 src = ZPROP_SRC_LOCAL;
326 if (prop == ZPOOL_PROP_BOOTFS) {
328 dsl_dataset_t *ds = NULL;
330 dp = spa_get_dsl(spa);
331 dsl_pool_config_enter(dp, FTAG);
332 if ((err = dsl_dataset_hold_obj(dp,
333 za.za_first_integer, FTAG, &ds))) {
334 dsl_pool_config_exit(dp, FTAG);
339 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
341 dsl_dataset_name(ds, strval);
342 dsl_dataset_rele(ds, FTAG);
343 dsl_pool_config_exit(dp, FTAG);
346 intval = za.za_first_integer;
349 spa_prop_add_list(*nvp, prop, strval, intval, src);
353 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
358 /* string property */
359 strval = kmem_alloc(za.za_num_integers, KM_PUSHPAGE);
360 err = zap_lookup(mos, spa->spa_pool_props_object,
361 za.za_name, 1, za.za_num_integers, strval);
363 kmem_free(strval, za.za_num_integers);
366 spa_prop_add_list(*nvp, prop, strval, 0, src);
367 kmem_free(strval, za.za_num_integers);
374 zap_cursor_fini(&zc);
375 mutex_exit(&spa->spa_props_lock);
377 if (err && err != ENOENT) {
387 * Validate the given pool properties nvlist and modify the list
388 * for the property values to be set.
391 spa_prop_validate(spa_t *spa, nvlist_t *props)
394 int error = 0, reset_bootfs = 0;
396 boolean_t has_feature = B_FALSE;
399 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
401 char *strval, *slash, *check, *fname;
402 const char *propname = nvpair_name(elem);
403 zpool_prop_t prop = zpool_name_to_prop(propname);
407 if (!zpool_prop_feature(propname)) {
408 error = SET_ERROR(EINVAL);
413 * Sanitize the input.
415 if (nvpair_type(elem) != DATA_TYPE_UINT64) {
416 error = SET_ERROR(EINVAL);
420 if (nvpair_value_uint64(elem, &intval) != 0) {
421 error = SET_ERROR(EINVAL);
426 error = SET_ERROR(EINVAL);
430 fname = strchr(propname, '@') + 1;
431 if (zfeature_lookup_name(fname, NULL) != 0) {
432 error = SET_ERROR(EINVAL);
436 has_feature = B_TRUE;
439 case ZPOOL_PROP_VERSION:
440 error = nvpair_value_uint64(elem, &intval);
442 (intval < spa_version(spa) ||
443 intval > SPA_VERSION_BEFORE_FEATURES ||
445 error = SET_ERROR(EINVAL);
448 case ZPOOL_PROP_DELEGATION:
449 case ZPOOL_PROP_AUTOREPLACE:
450 case ZPOOL_PROP_LISTSNAPS:
451 case ZPOOL_PROP_AUTOEXPAND:
452 error = nvpair_value_uint64(elem, &intval);
453 if (!error && intval > 1)
454 error = SET_ERROR(EINVAL);
457 case ZPOOL_PROP_BOOTFS:
459 * If the pool version is less than SPA_VERSION_BOOTFS,
460 * or the pool is still being created (version == 0),
461 * the bootfs property cannot be set.
463 if (spa_version(spa) < SPA_VERSION_BOOTFS) {
464 error = SET_ERROR(ENOTSUP);
469 * Make sure the vdev config is bootable
471 if (!vdev_is_bootable(spa->spa_root_vdev)) {
472 error = SET_ERROR(ENOTSUP);
478 error = nvpair_value_string(elem, &strval);
484 if (strval == NULL || strval[0] == '\0') {
485 objnum = zpool_prop_default_numeric(
490 if ((error = dmu_objset_hold(strval,FTAG,&os)))
493 /* Must be ZPL and not gzip compressed. */
495 if (dmu_objset_type(os) != DMU_OST_ZFS) {
496 error = SET_ERROR(ENOTSUP);
498 dsl_prop_get_int_ds(dmu_objset_ds(os),
499 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
501 !BOOTFS_COMPRESS_VALID(compress)) {
502 error = SET_ERROR(ENOTSUP);
504 objnum = dmu_objset_id(os);
506 dmu_objset_rele(os, FTAG);
510 case ZPOOL_PROP_FAILUREMODE:
511 error = nvpair_value_uint64(elem, &intval);
512 if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
513 intval > ZIO_FAILURE_MODE_PANIC))
514 error = SET_ERROR(EINVAL);
517 * This is a special case which only occurs when
518 * the pool has completely failed. This allows
519 * the user to change the in-core failmode property
520 * without syncing it out to disk (I/Os might
521 * currently be blocked). We do this by returning
522 * EIO to the caller (spa_prop_set) to trick it
523 * into thinking we encountered a property validation
526 if (!error && spa_suspended(spa)) {
527 spa->spa_failmode = intval;
528 error = SET_ERROR(EIO);
532 case ZPOOL_PROP_CACHEFILE:
533 if ((error = nvpair_value_string(elem, &strval)) != 0)
536 if (strval[0] == '\0')
539 if (strcmp(strval, "none") == 0)
542 if (strval[0] != '/') {
543 error = SET_ERROR(EINVAL);
547 slash = strrchr(strval, '/');
548 ASSERT(slash != NULL);
550 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
551 strcmp(slash, "/..") == 0)
552 error = SET_ERROR(EINVAL);
555 case ZPOOL_PROP_COMMENT:
556 if ((error = nvpair_value_string(elem, &strval)) != 0)
558 for (check = strval; *check != '\0'; check++) {
559 if (!isprint(*check)) {
560 error = SET_ERROR(EINVAL);
565 if (strlen(strval) > ZPROP_MAX_COMMENT)
566 error = SET_ERROR(E2BIG);
569 case ZPOOL_PROP_DEDUPDITTO:
570 if (spa_version(spa) < SPA_VERSION_DEDUP)
571 error = SET_ERROR(ENOTSUP);
573 error = nvpair_value_uint64(elem, &intval);
575 intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
576 error = SET_ERROR(EINVAL);
587 if (!error && reset_bootfs) {
588 error = nvlist_remove(props,
589 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
592 error = nvlist_add_uint64(props,
593 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
601 spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
604 spa_config_dirent_t *dp;
606 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
610 dp = kmem_alloc(sizeof (spa_config_dirent_t),
613 if (cachefile[0] == '\0')
614 dp->scd_path = spa_strdup(spa_config_path);
615 else if (strcmp(cachefile, "none") == 0)
618 dp->scd_path = spa_strdup(cachefile);
620 list_insert_head(&spa->spa_config_list, dp);
622 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
626 spa_prop_set(spa_t *spa, nvlist_t *nvp)
629 nvpair_t *elem = NULL;
630 boolean_t need_sync = B_FALSE;
632 if ((error = spa_prop_validate(spa, nvp)) != 0)
635 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
636 zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem));
638 if (prop == ZPOOL_PROP_CACHEFILE ||
639 prop == ZPOOL_PROP_ALTROOT ||
640 prop == ZPOOL_PROP_READONLY)
643 if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) {
646 if (prop == ZPOOL_PROP_VERSION) {
647 VERIFY(nvpair_value_uint64(elem, &ver) == 0);
649 ASSERT(zpool_prop_feature(nvpair_name(elem)));
650 ver = SPA_VERSION_FEATURES;
654 /* Save time if the version is already set. */
655 if (ver == spa_version(spa))
659 * In addition to the pool directory object, we might
660 * create the pool properties object, the features for
661 * read object, the features for write object, or the
662 * feature descriptions object.
664 error = dsl_sync_task(spa->spa_name, NULL,
665 spa_sync_version, &ver, 6);
676 return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props,
684 * If the bootfs property value is dsobj, clear it.
687 spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
689 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
690 VERIFY(zap_remove(spa->spa_meta_objset,
691 spa->spa_pool_props_object,
692 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
699 spa_change_guid_check(void *arg, dmu_tx_t *tx)
701 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
702 vdev_t *rvd = spa->spa_root_vdev;
704 ASSERTV(uint64_t *newguid = arg);
706 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
707 vdev_state = rvd->vdev_state;
708 spa_config_exit(spa, SCL_STATE, FTAG);
710 if (vdev_state != VDEV_STATE_HEALTHY)
711 return (SET_ERROR(ENXIO));
713 ASSERT3U(spa_guid(spa), !=, *newguid);
719 spa_change_guid_sync(void *arg, dmu_tx_t *tx)
721 uint64_t *newguid = arg;
722 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
724 vdev_t *rvd = spa->spa_root_vdev;
726 oldguid = spa_guid(spa);
728 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
729 rvd->vdev_guid = *newguid;
730 rvd->vdev_guid_sum += (*newguid - oldguid);
731 vdev_config_dirty(rvd);
732 spa_config_exit(spa, SCL_STATE, FTAG);
734 spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu",
739 * Change the GUID for the pool. This is done so that we can later
740 * re-import a pool built from a clone of our own vdevs. We will modify
741 * the root vdev's guid, our own pool guid, and then mark all of our
742 * vdevs dirty. Note that we must make sure that all our vdevs are
743 * online when we do this, or else any vdevs that weren't present
744 * would be orphaned from our pool. We are also going to issue a
745 * sysevent to update any watchers.
748 spa_change_guid(spa_t *spa)
753 mutex_enter(&spa_namespace_lock);
754 guid = spa_generate_guid(NULL);
756 error = dsl_sync_task(spa->spa_name, spa_change_guid_check,
757 spa_change_guid_sync, &guid, 5);
760 spa_config_sync(spa, B_FALSE, B_TRUE);
761 spa_event_notify(spa, NULL, FM_EREPORT_ZFS_POOL_REGUID);
764 mutex_exit(&spa_namespace_lock);
770 * ==========================================================================
771 * SPA state manipulation (open/create/destroy/import/export)
772 * ==========================================================================
776 spa_error_entry_compare(const void *a, const void *b)
778 spa_error_entry_t *sa = (spa_error_entry_t *)a;
779 spa_error_entry_t *sb = (spa_error_entry_t *)b;
782 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
783 sizeof (zbookmark_t));
794 * Utility function which retrieves copies of the current logs and
795 * re-initializes them in the process.
798 spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
800 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
802 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
803 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
805 avl_create(&spa->spa_errlist_scrub,
806 spa_error_entry_compare, sizeof (spa_error_entry_t),
807 offsetof(spa_error_entry_t, se_avl));
808 avl_create(&spa->spa_errlist_last,
809 spa_error_entry_compare, sizeof (spa_error_entry_t),
810 offsetof(spa_error_entry_t, se_avl));
814 spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
816 const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
817 enum zti_modes mode = ztip->zti_mode;
818 uint_t value = ztip->zti_value;
819 uint_t count = ztip->zti_count;
820 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
823 boolean_t batch = B_FALSE;
825 if (mode == ZTI_MODE_NULL) {
827 tqs->stqs_taskq = NULL;
831 ASSERT3U(count, >, 0);
833 tqs->stqs_count = count;
834 tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP);
836 for (i = 0; i < count; i++) {
841 ASSERT3U(value, >=, 1);
842 value = MAX(value, 1);
847 flags |= TASKQ_THREADS_CPU_PCT;
848 value = zio_taskq_batch_pct;
851 case ZTI_MODE_ONLINE_PERCENT:
852 flags |= TASKQ_THREADS_CPU_PCT;
856 panic("unrecognized mode for %s_%s taskq (%u:%u) in "
858 zio_type_name[t], zio_taskq_types[q], mode, value);
863 (void) snprintf(name, sizeof (name), "%s_%s_%u",
864 zio_type_name[t], zio_taskq_types[q], i);
866 (void) snprintf(name, sizeof (name), "%s_%s",
867 zio_type_name[t], zio_taskq_types[q]);
870 if (zio_taskq_sysdc && spa->spa_proc != &p0) {
872 flags |= TASKQ_DC_BATCH;
874 tq = taskq_create_sysdc(name, value, 50, INT_MAX,
875 spa->spa_proc, zio_taskq_basedc, flags);
877 tq = taskq_create_proc(name, value, maxclsyspri, 50,
878 INT_MAX, spa->spa_proc, flags);
881 tqs->stqs_taskq[i] = tq;
886 spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
888 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
891 if (tqs->stqs_taskq == NULL) {
892 ASSERT3U(tqs->stqs_count, ==, 0);
896 for (i = 0; i < tqs->stqs_count; i++) {
897 ASSERT3P(tqs->stqs_taskq[i], !=, NULL);
898 taskq_destroy(tqs->stqs_taskq[i]);
901 kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *));
902 tqs->stqs_taskq = NULL;
906 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
907 * Note that a type may have multiple discrete taskqs to avoid lock contention
908 * on the taskq itself. In that case we choose which taskq at random by using
909 * the low bits of gethrtime().
912 spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q,
913 task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent)
915 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
918 ASSERT3P(tqs->stqs_taskq, !=, NULL);
919 ASSERT3U(tqs->stqs_count, !=, 0);
921 if (tqs->stqs_count == 1) {
922 tq = tqs->stqs_taskq[0];
924 tq = tqs->stqs_taskq[((uint64_t)gethrtime()) % tqs->stqs_count];
927 taskq_dispatch_ent(tq, func, arg, flags, ent);
931 * Same as spa_taskq_dispatch_ent() but block on the task until completion.
934 spa_taskq_dispatch_sync(spa_t *spa, zio_type_t t, zio_taskq_type_t q,
935 task_func_t *func, void *arg, uint_t flags)
937 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
941 ASSERT3P(tqs->stqs_taskq, !=, NULL);
942 ASSERT3U(tqs->stqs_count, !=, 0);
944 if (tqs->stqs_count == 1) {
945 tq = tqs->stqs_taskq[0];
947 tq = tqs->stqs_taskq[((uint64_t)gethrtime()) % tqs->stqs_count];
950 id = taskq_dispatch(tq, func, arg, flags);
952 taskq_wait_id(tq, id);
956 spa_create_zio_taskqs(spa_t *spa)
960 for (t = 0; t < ZIO_TYPES; t++) {
961 for (q = 0; q < ZIO_TASKQ_TYPES; q++) {
962 spa_taskqs_init(spa, t, q);
967 #if defined(_KERNEL) && defined(HAVE_SPA_THREAD)
969 spa_thread(void *arg)
974 user_t *pu = PTOU(curproc);
976 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
979 ASSERT(curproc != &p0);
980 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
981 "zpool-%s", spa->spa_name);
982 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
984 /* bind this thread to the requested psrset */
985 if (zio_taskq_psrset_bind != PS_NONE) {
987 mutex_enter(&cpu_lock);
988 mutex_enter(&pidlock);
989 mutex_enter(&curproc->p_lock);
991 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
992 0, NULL, NULL) == 0) {
993 curthread->t_bind_pset = zio_taskq_psrset_bind;
996 "Couldn't bind process for zfs pool \"%s\" to "
997 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
1000 mutex_exit(&curproc->p_lock);
1001 mutex_exit(&pidlock);
1002 mutex_exit(&cpu_lock);
1006 if (zio_taskq_sysdc) {
1007 sysdc_thread_enter(curthread, 100, 0);
1010 spa->spa_proc = curproc;
1011 spa->spa_did = curthread->t_did;
1013 spa_create_zio_taskqs(spa);
1015 mutex_enter(&spa->spa_proc_lock);
1016 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
1018 spa->spa_proc_state = SPA_PROC_ACTIVE;
1019 cv_broadcast(&spa->spa_proc_cv);
1021 CALLB_CPR_SAFE_BEGIN(&cprinfo);
1022 while (spa->spa_proc_state == SPA_PROC_ACTIVE)
1023 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1024 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
1026 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
1027 spa->spa_proc_state = SPA_PROC_GONE;
1028 spa->spa_proc = &p0;
1029 cv_broadcast(&spa->spa_proc_cv);
1030 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */
1032 mutex_enter(&curproc->p_lock);
1038 * Activate an uninitialized pool.
1041 spa_activate(spa_t *spa, int mode)
1043 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
1045 spa->spa_state = POOL_STATE_ACTIVE;
1046 spa->spa_mode = mode;
1048 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
1049 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
1051 /* Try to create a covering process */
1052 mutex_enter(&spa->spa_proc_lock);
1053 ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
1054 ASSERT(spa->spa_proc == &p0);
1057 #ifdef HAVE_SPA_THREAD
1058 /* Only create a process if we're going to be around a while. */
1059 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
1060 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
1062 spa->spa_proc_state = SPA_PROC_CREATED;
1063 while (spa->spa_proc_state == SPA_PROC_CREATED) {
1064 cv_wait(&spa->spa_proc_cv,
1065 &spa->spa_proc_lock);
1067 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1068 ASSERT(spa->spa_proc != &p0);
1069 ASSERT(spa->spa_did != 0);
1073 "Couldn't create process for zfs pool \"%s\"\n",
1078 #endif /* HAVE_SPA_THREAD */
1079 mutex_exit(&spa->spa_proc_lock);
1081 /* If we didn't create a process, we need to create our taskqs. */
1082 if (spa->spa_proc == &p0) {
1083 spa_create_zio_taskqs(spa);
1086 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
1087 offsetof(vdev_t, vdev_config_dirty_node));
1088 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
1089 offsetof(vdev_t, vdev_state_dirty_node));
1091 txg_list_create(&spa->spa_vdev_txg_list,
1092 offsetof(struct vdev, vdev_txg_node));
1094 avl_create(&spa->spa_errlist_scrub,
1095 spa_error_entry_compare, sizeof (spa_error_entry_t),
1096 offsetof(spa_error_entry_t, se_avl));
1097 avl_create(&spa->spa_errlist_last,
1098 spa_error_entry_compare, sizeof (spa_error_entry_t),
1099 offsetof(spa_error_entry_t, se_avl));
1103 * Opposite of spa_activate().
1106 spa_deactivate(spa_t *spa)
1110 ASSERT(spa->spa_sync_on == B_FALSE);
1111 ASSERT(spa->spa_dsl_pool == NULL);
1112 ASSERT(spa->spa_root_vdev == NULL);
1113 ASSERT(spa->spa_async_zio_root == NULL);
1114 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
1116 txg_list_destroy(&spa->spa_vdev_txg_list);
1118 list_destroy(&spa->spa_config_dirty_list);
1119 list_destroy(&spa->spa_state_dirty_list);
1121 taskq_cancel_id(system_taskq, spa->spa_deadman_tqid);
1123 for (t = 0; t < ZIO_TYPES; t++) {
1124 for (q = 0; q < ZIO_TASKQ_TYPES; q++) {
1125 spa_taskqs_fini(spa, t, q);
1129 metaslab_class_destroy(spa->spa_normal_class);
1130 spa->spa_normal_class = NULL;
1132 metaslab_class_destroy(spa->spa_log_class);
1133 spa->spa_log_class = NULL;
1136 * If this was part of an import or the open otherwise failed, we may
1137 * still have errors left in the queues. Empty them just in case.
1139 spa_errlog_drain(spa);
1141 avl_destroy(&spa->spa_errlist_scrub);
1142 avl_destroy(&spa->spa_errlist_last);
1144 spa->spa_state = POOL_STATE_UNINITIALIZED;
1146 mutex_enter(&spa->spa_proc_lock);
1147 if (spa->spa_proc_state != SPA_PROC_NONE) {
1148 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1149 spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1150 cv_broadcast(&spa->spa_proc_cv);
1151 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1152 ASSERT(spa->spa_proc != &p0);
1153 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1155 ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1156 spa->spa_proc_state = SPA_PROC_NONE;
1158 ASSERT(spa->spa_proc == &p0);
1159 mutex_exit(&spa->spa_proc_lock);
1162 * We want to make sure spa_thread() has actually exited the ZFS
1163 * module, so that the module can't be unloaded out from underneath
1166 if (spa->spa_did != 0) {
1167 thread_join(spa->spa_did);
1173 * Verify a pool configuration, and construct the vdev tree appropriately. This
1174 * will create all the necessary vdevs in the appropriate layout, with each vdev
1175 * in the CLOSED state. This will prep the pool before open/creation/import.
1176 * All vdev validation is done by the vdev_alloc() routine.
1179 spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1180 uint_t id, int atype)
1187 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1190 if ((*vdp)->vdev_ops->vdev_op_leaf)
1193 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1196 if (error == ENOENT)
1202 return (SET_ERROR(EINVAL));
1205 for (c = 0; c < children; c++) {
1207 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1215 ASSERT(*vdp != NULL);
1221 * Opposite of spa_load().
1224 spa_unload(spa_t *spa)
1228 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1233 spa_async_suspend(spa);
1238 if (spa->spa_sync_on) {
1239 txg_sync_stop(spa->spa_dsl_pool);
1240 spa->spa_sync_on = B_FALSE;
1244 * Wait for any outstanding async I/O to complete.
1246 if (spa->spa_async_zio_root != NULL) {
1247 (void) zio_wait(spa->spa_async_zio_root);
1248 spa->spa_async_zio_root = NULL;
1251 bpobj_close(&spa->spa_deferred_bpobj);
1254 * Close the dsl pool.
1256 if (spa->spa_dsl_pool) {
1257 dsl_pool_close(spa->spa_dsl_pool);
1258 spa->spa_dsl_pool = NULL;
1259 spa->spa_meta_objset = NULL;
1264 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1267 * Drop and purge level 2 cache
1269 spa_l2cache_drop(spa);
1274 if (spa->spa_root_vdev)
1275 vdev_free(spa->spa_root_vdev);
1276 ASSERT(spa->spa_root_vdev == NULL);
1278 for (i = 0; i < spa->spa_spares.sav_count; i++)
1279 vdev_free(spa->spa_spares.sav_vdevs[i]);
1280 if (spa->spa_spares.sav_vdevs) {
1281 kmem_free(spa->spa_spares.sav_vdevs,
1282 spa->spa_spares.sav_count * sizeof (void *));
1283 spa->spa_spares.sav_vdevs = NULL;
1285 if (spa->spa_spares.sav_config) {
1286 nvlist_free(spa->spa_spares.sav_config);
1287 spa->spa_spares.sav_config = NULL;
1289 spa->spa_spares.sav_count = 0;
1291 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1292 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1293 vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1295 if (spa->spa_l2cache.sav_vdevs) {
1296 kmem_free(spa->spa_l2cache.sav_vdevs,
1297 spa->spa_l2cache.sav_count * sizeof (void *));
1298 spa->spa_l2cache.sav_vdevs = NULL;
1300 if (spa->spa_l2cache.sav_config) {
1301 nvlist_free(spa->spa_l2cache.sav_config);
1302 spa->spa_l2cache.sav_config = NULL;
1304 spa->spa_l2cache.sav_count = 0;
1306 spa->spa_async_suspended = 0;
1308 if (spa->spa_comment != NULL) {
1309 spa_strfree(spa->spa_comment);
1310 spa->spa_comment = NULL;
1313 spa_config_exit(spa, SCL_ALL, FTAG);
1317 * Load (or re-load) the current list of vdevs describing the active spares for
1318 * this pool. When this is called, we have some form of basic information in
1319 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1320 * then re-generate a more complete list including status information.
1323 spa_load_spares(spa_t *spa)
1330 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1333 * First, close and free any existing spare vdevs.
1335 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1336 vd = spa->spa_spares.sav_vdevs[i];
1338 /* Undo the call to spa_activate() below */
1339 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1340 B_FALSE)) != NULL && tvd->vdev_isspare)
1341 spa_spare_remove(tvd);
1346 if (spa->spa_spares.sav_vdevs)
1347 kmem_free(spa->spa_spares.sav_vdevs,
1348 spa->spa_spares.sav_count * sizeof (void *));
1350 if (spa->spa_spares.sav_config == NULL)
1353 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1354 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1356 spa->spa_spares.sav_count = (int)nspares;
1357 spa->spa_spares.sav_vdevs = NULL;
1363 * Construct the array of vdevs, opening them to get status in the
1364 * process. For each spare, there is potentially two different vdev_t
1365 * structures associated with it: one in the list of spares (used only
1366 * for basic validation purposes) and one in the active vdev
1367 * configuration (if it's spared in). During this phase we open and
1368 * validate each vdev on the spare list. If the vdev also exists in the
1369 * active configuration, then we also mark this vdev as an active spare.
1371 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1373 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1374 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1375 VDEV_ALLOC_SPARE) == 0);
1378 spa->spa_spares.sav_vdevs[i] = vd;
1380 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1381 B_FALSE)) != NULL) {
1382 if (!tvd->vdev_isspare)
1386 * We only mark the spare active if we were successfully
1387 * able to load the vdev. Otherwise, importing a pool
1388 * with a bad active spare would result in strange
1389 * behavior, because multiple pool would think the spare
1390 * is actively in use.
1392 * There is a vulnerability here to an equally bizarre
1393 * circumstance, where a dead active spare is later
1394 * brought back to life (onlined or otherwise). Given
1395 * the rarity of this scenario, and the extra complexity
1396 * it adds, we ignore the possibility.
1398 if (!vdev_is_dead(tvd))
1399 spa_spare_activate(tvd);
1403 vd->vdev_aux = &spa->spa_spares;
1405 if (vdev_open(vd) != 0)
1408 if (vdev_validate_aux(vd) == 0)
1413 * Recompute the stashed list of spares, with status information
1416 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1417 DATA_TYPE_NVLIST_ARRAY) == 0);
1419 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1421 for (i = 0; i < spa->spa_spares.sav_count; i++)
1422 spares[i] = vdev_config_generate(spa,
1423 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1424 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1425 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1426 for (i = 0; i < spa->spa_spares.sav_count; i++)
1427 nvlist_free(spares[i]);
1428 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1432 * Load (or re-load) the current list of vdevs describing the active l2cache for
1433 * this pool. When this is called, we have some form of basic information in
1434 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1435 * then re-generate a more complete list including status information.
1436 * Devices which are already active have their details maintained, and are
1440 spa_load_l2cache(spa_t *spa)
1444 int i, j, oldnvdevs;
1446 vdev_t *vd, **oldvdevs, **newvdevs;
1447 spa_aux_vdev_t *sav = &spa->spa_l2cache;
1449 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1451 if (sav->sav_config != NULL) {
1452 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1453 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1454 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_PUSHPAGE);
1460 oldvdevs = sav->sav_vdevs;
1461 oldnvdevs = sav->sav_count;
1462 sav->sav_vdevs = NULL;
1466 * Process new nvlist of vdevs.
1468 for (i = 0; i < nl2cache; i++) {
1469 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1473 for (j = 0; j < oldnvdevs; j++) {
1475 if (vd != NULL && guid == vd->vdev_guid) {
1477 * Retain previous vdev for add/remove ops.
1485 if (newvdevs[i] == NULL) {
1489 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1490 VDEV_ALLOC_L2CACHE) == 0);
1495 * Commit this vdev as an l2cache device,
1496 * even if it fails to open.
1498 spa_l2cache_add(vd);
1503 spa_l2cache_activate(vd);
1505 if (vdev_open(vd) != 0)
1508 (void) vdev_validate_aux(vd);
1510 if (!vdev_is_dead(vd))
1511 l2arc_add_vdev(spa, vd);
1516 * Purge vdevs that were dropped
1518 for (i = 0; i < oldnvdevs; i++) {
1523 ASSERT(vd->vdev_isl2cache);
1525 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1526 pool != 0ULL && l2arc_vdev_present(vd))
1527 l2arc_remove_vdev(vd);
1528 vdev_clear_stats(vd);
1534 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1536 if (sav->sav_config == NULL)
1539 sav->sav_vdevs = newvdevs;
1540 sav->sav_count = (int)nl2cache;
1543 * Recompute the stashed list of l2cache devices, with status
1544 * information this time.
1546 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1547 DATA_TYPE_NVLIST_ARRAY) == 0);
1549 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_PUSHPAGE);
1550 for (i = 0; i < sav->sav_count; i++)
1551 l2cache[i] = vdev_config_generate(spa,
1552 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1553 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1554 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1556 for (i = 0; i < sav->sav_count; i++)
1557 nvlist_free(l2cache[i]);
1559 kmem_free(l2cache, sav->sav_count * sizeof (void *));
1563 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1566 char *packed = NULL;
1571 error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db);
1575 nvsize = *(uint64_t *)db->db_data;
1576 dmu_buf_rele(db, FTAG);
1578 packed = kmem_alloc(nvsize, KM_PUSHPAGE | KM_NODEBUG);
1579 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1582 error = nvlist_unpack(packed, nvsize, value, 0);
1583 kmem_free(packed, nvsize);
1589 * Checks to see if the given vdev could not be opened, in which case we post a
1590 * sysevent to notify the autoreplace code that the device has been removed.
1593 spa_check_removed(vdev_t *vd)
1597 for (c = 0; c < vd->vdev_children; c++)
1598 spa_check_removed(vd->vdev_child[c]);
1600 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) &&
1602 zfs_ereport_post(FM_EREPORT_RESOURCE_AUTOREPLACE,
1603 vd->vdev_spa, vd, NULL, 0, 0);
1604 spa_event_notify(vd->vdev_spa, vd, FM_EREPORT_ZFS_DEVICE_CHECK);
1609 * Validate the current config against the MOS config
1612 spa_config_valid(spa_t *spa, nvlist_t *config)
1614 vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1618 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1620 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1621 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1623 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1626 * If we're doing a normal import, then build up any additional
1627 * diagnostic information about missing devices in this config.
1628 * We'll pass this up to the user for further processing.
1630 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1631 nvlist_t **child, *nv;
1634 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1636 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_PUSHPAGE) == 0);
1638 for (c = 0; c < rvd->vdev_children; c++) {
1639 vdev_t *tvd = rvd->vdev_child[c];
1640 vdev_t *mtvd = mrvd->vdev_child[c];
1642 if (tvd->vdev_ops == &vdev_missing_ops &&
1643 mtvd->vdev_ops != &vdev_missing_ops &&
1645 child[idx++] = vdev_config_generate(spa, mtvd,
1650 VERIFY(nvlist_add_nvlist_array(nv,
1651 ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1652 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1653 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1655 for (i = 0; i < idx; i++)
1656 nvlist_free(child[i]);
1659 kmem_free(child, rvd->vdev_children * sizeof (char **));
1663 * Compare the root vdev tree with the information we have
1664 * from the MOS config (mrvd). Check each top-level vdev
1665 * with the corresponding MOS config top-level (mtvd).
1667 for (c = 0; c < rvd->vdev_children; c++) {
1668 vdev_t *tvd = rvd->vdev_child[c];
1669 vdev_t *mtvd = mrvd->vdev_child[c];
1672 * Resolve any "missing" vdevs in the current configuration.
1673 * If we find that the MOS config has more accurate information
1674 * about the top-level vdev then use that vdev instead.
1676 if (tvd->vdev_ops == &vdev_missing_ops &&
1677 mtvd->vdev_ops != &vdev_missing_ops) {
1679 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1683 * Device specific actions.
1685 if (mtvd->vdev_islog) {
1686 spa_set_log_state(spa, SPA_LOG_CLEAR);
1689 * XXX - once we have 'readonly' pool
1690 * support we should be able to handle
1691 * missing data devices by transitioning
1692 * the pool to readonly.
1698 * Swap the missing vdev with the data we were
1699 * able to obtain from the MOS config.
1701 vdev_remove_child(rvd, tvd);
1702 vdev_remove_child(mrvd, mtvd);
1704 vdev_add_child(rvd, mtvd);
1705 vdev_add_child(mrvd, tvd);
1707 spa_config_exit(spa, SCL_ALL, FTAG);
1709 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1712 } else if (mtvd->vdev_islog) {
1714 * Load the slog device's state from the MOS config
1715 * since it's possible that the label does not
1716 * contain the most up-to-date information.
1718 vdev_load_log_state(tvd, mtvd);
1723 spa_config_exit(spa, SCL_ALL, FTAG);
1726 * Ensure we were able to validate the config.
1728 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1732 * Check for missing log devices
1735 spa_check_logs(spa_t *spa)
1737 boolean_t rv = B_FALSE;
1739 switch (spa->spa_log_state) {
1742 case SPA_LOG_MISSING:
1743 /* need to recheck in case slog has been restored */
1744 case SPA_LOG_UNKNOWN:
1745 rv = (dmu_objset_find(spa->spa_name, zil_check_log_chain,
1746 NULL, DS_FIND_CHILDREN) != 0);
1748 spa_set_log_state(spa, SPA_LOG_MISSING);
1755 spa_passivate_log(spa_t *spa)
1757 vdev_t *rvd = spa->spa_root_vdev;
1758 boolean_t slog_found = B_FALSE;
1761 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1763 if (!spa_has_slogs(spa))
1766 for (c = 0; c < rvd->vdev_children; c++) {
1767 vdev_t *tvd = rvd->vdev_child[c];
1768 metaslab_group_t *mg = tvd->vdev_mg;
1770 if (tvd->vdev_islog) {
1771 metaslab_group_passivate(mg);
1772 slog_found = B_TRUE;
1776 return (slog_found);
1780 spa_activate_log(spa_t *spa)
1782 vdev_t *rvd = spa->spa_root_vdev;
1785 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1787 for (c = 0; c < rvd->vdev_children; c++) {
1788 vdev_t *tvd = rvd->vdev_child[c];
1789 metaslab_group_t *mg = tvd->vdev_mg;
1791 if (tvd->vdev_islog)
1792 metaslab_group_activate(mg);
1797 spa_offline_log(spa_t *spa)
1801 error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1802 NULL, DS_FIND_CHILDREN);
1805 * We successfully offlined the log device, sync out the
1806 * current txg so that the "stubby" block can be removed
1809 txg_wait_synced(spa->spa_dsl_pool, 0);
1815 spa_aux_check_removed(spa_aux_vdev_t *sav)
1819 for (i = 0; i < sav->sav_count; i++)
1820 spa_check_removed(sav->sav_vdevs[i]);
1824 spa_claim_notify(zio_t *zio)
1826 spa_t *spa = zio->io_spa;
1831 mutex_enter(&spa->spa_props_lock); /* any mutex will do */
1832 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1833 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1834 mutex_exit(&spa->spa_props_lock);
1837 typedef struct spa_load_error {
1838 uint64_t sle_meta_count;
1839 uint64_t sle_data_count;
1843 spa_load_verify_done(zio_t *zio)
1845 blkptr_t *bp = zio->io_bp;
1846 spa_load_error_t *sle = zio->io_private;
1847 dmu_object_type_t type = BP_GET_TYPE(bp);
1848 int error = zio->io_error;
1851 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1852 type != DMU_OT_INTENT_LOG)
1853 atomic_add_64(&sle->sle_meta_count, 1);
1855 atomic_add_64(&sle->sle_data_count, 1);
1857 zio_data_buf_free(zio->io_data, zio->io_size);
1862 spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1863 const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1867 size_t size = BP_GET_PSIZE(bp);
1868 void *data = zio_data_buf_alloc(size);
1870 zio_nowait(zio_read(rio, spa, bp, data, size,
1871 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1872 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1873 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1879 spa_load_verify(spa_t *spa)
1882 spa_load_error_t sle = { 0 };
1883 zpool_rewind_policy_t policy;
1884 boolean_t verify_ok = B_FALSE;
1887 zpool_get_rewind_policy(spa->spa_config, &policy);
1889 if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1892 rio = zio_root(spa, NULL, &sle,
1893 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1895 error = traverse_pool(spa, spa->spa_verify_min_txg,
1896 TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio);
1898 (void) zio_wait(rio);
1900 spa->spa_load_meta_errors = sle.sle_meta_count;
1901 spa->spa_load_data_errors = sle.sle_data_count;
1903 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1904 sle.sle_data_count <= policy.zrp_maxdata) {
1908 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1909 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1911 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1912 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1913 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1914 VERIFY(nvlist_add_int64(spa->spa_load_info,
1915 ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1916 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1917 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1919 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1923 if (error != ENXIO && error != EIO)
1924 error = SET_ERROR(EIO);
1928 return (verify_ok ? 0 : EIO);
1932 * Find a value in the pool props object.
1935 spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1937 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1938 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1942 * Find a value in the pool directory object.
1945 spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1947 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1948 name, sizeof (uint64_t), 1, val));
1952 spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1954 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1959 * Fix up config after a partly-completed split. This is done with the
1960 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
1961 * pool have that entry in their config, but only the splitting one contains
1962 * a list of all the guids of the vdevs that are being split off.
1964 * This function determines what to do with that list: either rejoin
1965 * all the disks to the pool, or complete the splitting process. To attempt
1966 * the rejoin, each disk that is offlined is marked online again, and
1967 * we do a reopen() call. If the vdev label for every disk that was
1968 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1969 * then we call vdev_split() on each disk, and complete the split.
1971 * Otherwise we leave the config alone, with all the vdevs in place in
1972 * the original pool.
1975 spa_try_repair(spa_t *spa, nvlist_t *config)
1982 boolean_t attempt_reopen;
1984 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
1987 /* check that the config is complete */
1988 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
1989 &glist, &gcount) != 0)
1992 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_PUSHPAGE);
1994 /* attempt to online all the vdevs & validate */
1995 attempt_reopen = B_TRUE;
1996 for (i = 0; i < gcount; i++) {
1997 if (glist[i] == 0) /* vdev is hole */
2000 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
2001 if (vd[i] == NULL) {
2003 * Don't bother attempting to reopen the disks;
2004 * just do the split.
2006 attempt_reopen = B_FALSE;
2008 /* attempt to re-online it */
2009 vd[i]->vdev_offline = B_FALSE;
2013 if (attempt_reopen) {
2014 vdev_reopen(spa->spa_root_vdev);
2016 /* check each device to see what state it's in */
2017 for (extracted = 0, i = 0; i < gcount; i++) {
2018 if (vd[i] != NULL &&
2019 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
2026 * If every disk has been moved to the new pool, or if we never
2027 * even attempted to look at them, then we split them off for
2030 if (!attempt_reopen || gcount == extracted) {
2031 for (i = 0; i < gcount; i++)
2034 vdev_reopen(spa->spa_root_vdev);
2037 kmem_free(vd, gcount * sizeof (vdev_t *));
2041 spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
2042 boolean_t mosconfig)
2044 nvlist_t *config = spa->spa_config;
2045 char *ereport = FM_EREPORT_ZFS_POOL;
2051 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
2052 return (SET_ERROR(EINVAL));
2054 ASSERT(spa->spa_comment == NULL);
2055 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
2056 spa->spa_comment = spa_strdup(comment);
2059 * Versioning wasn't explicitly added to the label until later, so if
2060 * it's not present treat it as the initial version.
2062 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
2063 &spa->spa_ubsync.ub_version) != 0)
2064 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
2066 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
2067 &spa->spa_config_txg);
2069 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
2070 spa_guid_exists(pool_guid, 0)) {
2071 error = SET_ERROR(EEXIST);
2073 spa->spa_config_guid = pool_guid;
2075 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
2077 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
2081 nvlist_free(spa->spa_load_info);
2082 spa->spa_load_info = fnvlist_alloc();
2084 gethrestime(&spa->spa_loaded_ts);
2085 error = spa_load_impl(spa, pool_guid, config, state, type,
2086 mosconfig, &ereport);
2089 spa->spa_minref = refcount_count(&spa->spa_refcount);
2091 if (error != EEXIST) {
2092 spa->spa_loaded_ts.tv_sec = 0;
2093 spa->spa_loaded_ts.tv_nsec = 0;
2095 if (error != EBADF) {
2096 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
2099 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2106 * Load an existing storage pool, using the pool's builtin spa_config as a
2107 * source of configuration information.
2109 __attribute__((always_inline))
2111 spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
2112 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
2116 nvlist_t *nvroot = NULL;
2119 uberblock_t *ub = &spa->spa_uberblock;
2120 uint64_t children, config_cache_txg = spa->spa_config_txg;
2121 int orig_mode = spa->spa_mode;
2124 boolean_t missing_feat_write = B_FALSE;
2127 * If this is an untrusted config, access the pool in read-only mode.
2128 * This prevents things like resilvering recently removed devices.
2131 spa->spa_mode = FREAD;
2133 ASSERT(MUTEX_HELD(&spa_namespace_lock));
2135 spa->spa_load_state = state;
2137 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
2138 return (SET_ERROR(EINVAL));
2140 parse = (type == SPA_IMPORT_EXISTING ?
2141 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2144 * Create "The Godfather" zio to hold all async IOs
2146 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2147 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2150 * Parse the configuration into a vdev tree. We explicitly set the
2151 * value that will be returned by spa_version() since parsing the
2152 * configuration requires knowing the version number.
2154 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2155 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
2156 spa_config_exit(spa, SCL_ALL, FTAG);
2161 ASSERT(spa->spa_root_vdev == rvd);
2163 if (type != SPA_IMPORT_ASSEMBLE) {
2164 ASSERT(spa_guid(spa) == pool_guid);
2168 * Try to open all vdevs, loading each label in the process.
2170 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2171 error = vdev_open(rvd);
2172 spa_config_exit(spa, SCL_ALL, FTAG);
2177 * We need to validate the vdev labels against the configuration that
2178 * we have in hand, which is dependent on the setting of mosconfig. If
2179 * mosconfig is true then we're validating the vdev labels based on
2180 * that config. Otherwise, we're validating against the cached config
2181 * (zpool.cache) that was read when we loaded the zfs module, and then
2182 * later we will recursively call spa_load() and validate against
2185 * If we're assembling a new pool that's been split off from an
2186 * existing pool, the labels haven't yet been updated so we skip
2187 * validation for now.
2189 if (type != SPA_IMPORT_ASSEMBLE) {
2190 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2191 error = vdev_validate(rvd, mosconfig);
2192 spa_config_exit(spa, SCL_ALL, FTAG);
2197 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2198 return (SET_ERROR(ENXIO));
2202 * Find the best uberblock.
2204 vdev_uberblock_load(rvd, ub, &label);
2207 * If we weren't able to find a single valid uberblock, return failure.
2209 if (ub->ub_txg == 0) {
2211 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2215 * If the pool has an unsupported version we can't open it.
2217 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2219 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2222 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2226 * If we weren't able to find what's necessary for reading the
2227 * MOS in the label, return failure.
2229 if (label == NULL || nvlist_lookup_nvlist(label,
2230 ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) {
2232 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2237 * Update our in-core representation with the definitive values
2240 nvlist_free(spa->spa_label_features);
2241 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2247 * Look through entries in the label nvlist's features_for_read. If
2248 * there is a feature listed there which we don't understand then we
2249 * cannot open a pool.
2251 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2252 nvlist_t *unsup_feat;
2255 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2258 for (nvp = nvlist_next_nvpair(spa->spa_label_features, NULL);
2260 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2261 if (!zfeature_is_supported(nvpair_name(nvp))) {
2262 VERIFY(nvlist_add_string(unsup_feat,
2263 nvpair_name(nvp), "") == 0);
2267 if (!nvlist_empty(unsup_feat)) {
2268 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2269 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2270 nvlist_free(unsup_feat);
2271 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2275 nvlist_free(unsup_feat);
2279 * If the vdev guid sum doesn't match the uberblock, we have an
2280 * incomplete configuration. We first check to see if the pool
2281 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2282 * If it is, defer the vdev_guid_sum check till later so we
2283 * can handle missing vdevs.
2285 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
2286 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
2287 rvd->vdev_guid_sum != ub->ub_guid_sum)
2288 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2290 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2291 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2292 spa_try_repair(spa, config);
2293 spa_config_exit(spa, SCL_ALL, FTAG);
2294 nvlist_free(spa->spa_config_splitting);
2295 spa->spa_config_splitting = NULL;
2299 * Initialize internal SPA structures.
2301 spa->spa_state = POOL_STATE_ACTIVE;
2302 spa->spa_ubsync = spa->spa_uberblock;
2303 spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2304 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2305 spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2306 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2307 spa->spa_claim_max_txg = spa->spa_first_txg;
2308 spa->spa_prev_software_version = ub->ub_software_version;
2310 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2312 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2313 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2315 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2316 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2318 if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2319 boolean_t missing_feat_read = B_FALSE;
2320 nvlist_t *unsup_feat, *enabled_feat;
2322 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2323 &spa->spa_feat_for_read_obj) != 0) {
2324 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2327 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2328 &spa->spa_feat_for_write_obj) != 0) {
2329 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2332 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2333 &spa->spa_feat_desc_obj) != 0) {
2334 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2337 enabled_feat = fnvlist_alloc();
2338 unsup_feat = fnvlist_alloc();
2340 if (!feature_is_supported(spa->spa_meta_objset,
2341 spa->spa_feat_for_read_obj, spa->spa_feat_desc_obj,
2342 unsup_feat, enabled_feat))
2343 missing_feat_read = B_TRUE;
2345 if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) {
2346 if (!feature_is_supported(spa->spa_meta_objset,
2347 spa->spa_feat_for_write_obj, spa->spa_feat_desc_obj,
2348 unsup_feat, enabled_feat)) {
2349 missing_feat_write = B_TRUE;
2353 fnvlist_add_nvlist(spa->spa_load_info,
2354 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
2356 if (!nvlist_empty(unsup_feat)) {
2357 fnvlist_add_nvlist(spa->spa_load_info,
2358 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
2361 fnvlist_free(enabled_feat);
2362 fnvlist_free(unsup_feat);
2364 if (!missing_feat_read) {
2365 fnvlist_add_boolean(spa->spa_load_info,
2366 ZPOOL_CONFIG_CAN_RDONLY);
2370 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2371 * twofold: to determine whether the pool is available for
2372 * import in read-write mode and (if it is not) whether the
2373 * pool is available for import in read-only mode. If the pool
2374 * is available for import in read-write mode, it is displayed
2375 * as available in userland; if it is not available for import
2376 * in read-only mode, it is displayed as unavailable in
2377 * userland. If the pool is available for import in read-only
2378 * mode but not read-write mode, it is displayed as unavailable
2379 * in userland with a special note that the pool is actually
2380 * available for open in read-only mode.
2382 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2383 * missing a feature for write, we must first determine whether
2384 * the pool can be opened read-only before returning to
2385 * userland in order to know whether to display the
2386 * abovementioned note.
2388 if (missing_feat_read || (missing_feat_write &&
2389 spa_writeable(spa))) {
2390 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2395 spa->spa_is_initializing = B_TRUE;
2396 error = dsl_pool_open(spa->spa_dsl_pool);
2397 spa->spa_is_initializing = B_FALSE;
2399 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2403 nvlist_t *policy = NULL, *nvconfig;
2405 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2406 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2408 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2409 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2411 unsigned long myhostid = 0;
2413 VERIFY(nvlist_lookup_string(nvconfig,
2414 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2417 myhostid = zone_get_hostid(NULL);
2420 * We're emulating the system's hostid in userland, so
2421 * we can't use zone_get_hostid().
2423 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2424 #endif /* _KERNEL */
2425 if (hostid != 0 && myhostid != 0 &&
2426 hostid != myhostid) {
2427 nvlist_free(nvconfig);
2428 cmn_err(CE_WARN, "pool '%s' could not be "
2429 "loaded as it was last accessed by "
2430 "another system (host: %s hostid: 0x%lx). "
2431 "See: http://zfsonlinux.org/msg/ZFS-8000-EY",
2432 spa_name(spa), hostname,
2433 (unsigned long)hostid);
2434 return (SET_ERROR(EBADF));
2437 if (nvlist_lookup_nvlist(spa->spa_config,
2438 ZPOOL_REWIND_POLICY, &policy) == 0)
2439 VERIFY(nvlist_add_nvlist(nvconfig,
2440 ZPOOL_REWIND_POLICY, policy) == 0);
2442 spa_config_set(spa, nvconfig);
2444 spa_deactivate(spa);
2445 spa_activate(spa, orig_mode);
2447 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2450 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2451 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2452 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2454 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2457 * Load the bit that tells us to use the new accounting function
2458 * (raid-z deflation). If we have an older pool, this will not
2461 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2462 if (error != 0 && error != ENOENT)
2463 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2465 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2466 &spa->spa_creation_version);
2467 if (error != 0 && error != ENOENT)
2468 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2471 * Load the persistent error log. If we have an older pool, this will
2474 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2475 if (error != 0 && error != ENOENT)
2476 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2478 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2479 &spa->spa_errlog_scrub);
2480 if (error != 0 && error != ENOENT)
2481 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2484 * Load the history object. If we have an older pool, this
2485 * will not be present.
2487 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2488 if (error != 0 && error != ENOENT)
2489 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2492 * If we're assembling the pool from the split-off vdevs of
2493 * an existing pool, we don't want to attach the spares & cache
2498 * Load any hot spares for this pool.
2500 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2501 if (error != 0 && error != ENOENT)
2502 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2503 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2504 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2505 if (load_nvlist(spa, spa->spa_spares.sav_object,
2506 &spa->spa_spares.sav_config) != 0)
2507 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2509 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2510 spa_load_spares(spa);
2511 spa_config_exit(spa, SCL_ALL, FTAG);
2512 } else if (error == 0) {
2513 spa->spa_spares.sav_sync = B_TRUE;
2517 * Load any level 2 ARC devices for this pool.
2519 error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2520 &spa->spa_l2cache.sav_object);
2521 if (error != 0 && error != ENOENT)
2522 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2523 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2524 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2525 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2526 &spa->spa_l2cache.sav_config) != 0)
2527 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2529 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2530 spa_load_l2cache(spa);
2531 spa_config_exit(spa, SCL_ALL, FTAG);
2532 } else if (error == 0) {
2533 spa->spa_l2cache.sav_sync = B_TRUE;
2536 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2538 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2539 if (error && error != ENOENT)
2540 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2543 uint64_t autoreplace;
2545 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2546 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2547 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2548 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2549 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2550 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2551 &spa->spa_dedup_ditto);
2553 spa->spa_autoreplace = (autoreplace != 0);
2557 * If the 'autoreplace' property is set, then post a resource notifying
2558 * the ZFS DE that it should not issue any faults for unopenable
2559 * devices. We also iterate over the vdevs, and post a sysevent for any
2560 * unopenable vdevs so that the normal autoreplace handler can take
2563 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2564 spa_check_removed(spa->spa_root_vdev);
2566 * For the import case, this is done in spa_import(), because
2567 * at this point we're using the spare definitions from
2568 * the MOS config, not necessarily from the userland config.
2570 if (state != SPA_LOAD_IMPORT) {
2571 spa_aux_check_removed(&spa->spa_spares);
2572 spa_aux_check_removed(&spa->spa_l2cache);
2577 * Load the vdev state for all toplevel vdevs.
2582 * Propagate the leaf DTLs we just loaded all the way up the tree.
2584 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2585 vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2586 spa_config_exit(spa, SCL_ALL, FTAG);
2589 * Load the DDTs (dedup tables).
2591 error = ddt_load(spa);
2593 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2595 spa_update_dspace(spa);
2598 * Validate the config, using the MOS config to fill in any
2599 * information which might be missing. If we fail to validate
2600 * the config then declare the pool unfit for use. If we're
2601 * assembling a pool from a split, the log is not transferred
2604 if (type != SPA_IMPORT_ASSEMBLE) {
2607 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2608 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2610 if (!spa_config_valid(spa, nvconfig)) {
2611 nvlist_free(nvconfig);
2612 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2615 nvlist_free(nvconfig);
2618 * Now that we've validated the config, check the state of the
2619 * root vdev. If it can't be opened, it indicates one or
2620 * more toplevel vdevs are faulted.
2622 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2623 return (SET_ERROR(ENXIO));
2625 if (spa_check_logs(spa)) {
2626 *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2627 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2631 if (missing_feat_write) {
2632 ASSERT(state == SPA_LOAD_TRYIMPORT);
2635 * At this point, we know that we can open the pool in
2636 * read-only mode but not read-write mode. We now have enough
2637 * information and can return to userland.
2639 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
2643 * We've successfully opened the pool, verify that we're ready
2644 * to start pushing transactions.
2646 if (state != SPA_LOAD_TRYIMPORT) {
2647 if ((error = spa_load_verify(spa)))
2648 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2652 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2653 spa->spa_load_max_txg == UINT64_MAX)) {
2655 int need_update = B_FALSE;
2658 ASSERT(state != SPA_LOAD_TRYIMPORT);
2661 * Claim log blocks that haven't been committed yet.
2662 * This must all happen in a single txg.
2663 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2664 * invoked from zil_claim_log_block()'s i/o done callback.
2665 * Price of rollback is that we abandon the log.
2667 spa->spa_claiming = B_TRUE;
2669 tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2670 spa_first_txg(spa));
2671 (void) dmu_objset_find(spa_name(spa),
2672 zil_claim, tx, DS_FIND_CHILDREN);
2675 spa->spa_claiming = B_FALSE;
2677 spa_set_log_state(spa, SPA_LOG_GOOD);
2678 spa->spa_sync_on = B_TRUE;
2679 txg_sync_start(spa->spa_dsl_pool);
2682 * Wait for all claims to sync. We sync up to the highest
2683 * claimed log block birth time so that claimed log blocks
2684 * don't appear to be from the future. spa_claim_max_txg
2685 * will have been set for us by either zil_check_log_chain()
2686 * (invoked from spa_check_logs()) or zil_claim() above.
2688 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2691 * If the config cache is stale, or we have uninitialized
2692 * metaslabs (see spa_vdev_add()), then update the config.
2694 * If this is a verbatim import, trust the current
2695 * in-core spa_config and update the disk labels.
2697 if (config_cache_txg != spa->spa_config_txg ||
2698 state == SPA_LOAD_IMPORT ||
2699 state == SPA_LOAD_RECOVER ||
2700 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2701 need_update = B_TRUE;
2703 for (c = 0; c < rvd->vdev_children; c++)
2704 if (rvd->vdev_child[c]->vdev_ms_array == 0)
2705 need_update = B_TRUE;
2708 * Update the config cache asychronously in case we're the
2709 * root pool, in which case the config cache isn't writable yet.
2712 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2715 * Check all DTLs to see if anything needs resilvering.
2717 if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2718 vdev_resilver_needed(rvd, NULL, NULL))
2719 spa_async_request(spa, SPA_ASYNC_RESILVER);
2722 * Log the fact that we booted up (so that we can detect if
2723 * we rebooted in the middle of an operation).
2725 spa_history_log_version(spa, "open");
2728 * Delete any inconsistent datasets.
2730 (void) dmu_objset_find(spa_name(spa),
2731 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2734 * Clean up any stale temporary dataset userrefs.
2736 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2743 spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2745 int mode = spa->spa_mode;
2748 spa_deactivate(spa);
2750 spa->spa_load_max_txg--;
2752 spa_activate(spa, mode);
2753 spa_async_suspend(spa);
2755 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2759 * If spa_load() fails this function will try loading prior txg's. If
2760 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2761 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2762 * function will not rewind the pool and will return the same error as
2766 spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2767 uint64_t max_request, int rewind_flags)
2769 nvlist_t *loadinfo = NULL;
2770 nvlist_t *config = NULL;
2771 int load_error, rewind_error;
2772 uint64_t safe_rewind_txg;
2775 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2776 spa->spa_load_max_txg = spa->spa_load_txg;
2777 spa_set_log_state(spa, SPA_LOG_CLEAR);
2779 spa->spa_load_max_txg = max_request;
2782 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2784 if (load_error == 0)
2787 if (spa->spa_root_vdev != NULL)
2788 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2790 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2791 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2793 if (rewind_flags & ZPOOL_NEVER_REWIND) {
2794 nvlist_free(config);
2795 return (load_error);
2798 if (state == SPA_LOAD_RECOVER) {
2799 /* Price of rolling back is discarding txgs, including log */
2800 spa_set_log_state(spa, SPA_LOG_CLEAR);
2803 * If we aren't rolling back save the load info from our first
2804 * import attempt so that we can restore it after attempting
2807 loadinfo = spa->spa_load_info;
2808 spa->spa_load_info = fnvlist_alloc();
2811 spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2812 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2813 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2814 TXG_INITIAL : safe_rewind_txg;
2817 * Continue as long as we're finding errors, we're still within
2818 * the acceptable rewind range, and we're still finding uberblocks
2820 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2821 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2822 if (spa->spa_load_max_txg < safe_rewind_txg)
2823 spa->spa_extreme_rewind = B_TRUE;
2824 rewind_error = spa_load_retry(spa, state, mosconfig);
2827 spa->spa_extreme_rewind = B_FALSE;
2828 spa->spa_load_max_txg = UINT64_MAX;
2830 if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2831 spa_config_set(spa, config);
2833 if (state == SPA_LOAD_RECOVER) {
2834 ASSERT3P(loadinfo, ==, NULL);
2835 return (rewind_error);
2837 /* Store the rewind info as part of the initial load info */
2838 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
2839 spa->spa_load_info);
2841 /* Restore the initial load info */
2842 fnvlist_free(spa->spa_load_info);
2843 spa->spa_load_info = loadinfo;
2845 return (load_error);
2852 * The import case is identical to an open except that the configuration is sent
2853 * down from userland, instead of grabbed from the configuration cache. For the
2854 * case of an open, the pool configuration will exist in the
2855 * POOL_STATE_UNINITIALIZED state.
2857 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2858 * the same time open the pool, without having to keep around the spa_t in some
2862 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2866 spa_load_state_t state = SPA_LOAD_OPEN;
2868 int locked = B_FALSE;
2869 int firstopen = B_FALSE;
2874 * As disgusting as this is, we need to support recursive calls to this
2875 * function because dsl_dir_open() is called during spa_load(), and ends
2876 * up calling spa_open() again. The real fix is to figure out how to
2877 * avoid dsl_dir_open() calling this in the first place.
2879 if (mutex_owner(&spa_namespace_lock) != curthread) {
2880 mutex_enter(&spa_namespace_lock);
2884 if ((spa = spa_lookup(pool)) == NULL) {
2886 mutex_exit(&spa_namespace_lock);
2887 return (SET_ERROR(ENOENT));
2890 if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2891 zpool_rewind_policy_t policy;
2895 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2897 if (policy.zrp_request & ZPOOL_DO_REWIND)
2898 state = SPA_LOAD_RECOVER;
2900 spa_activate(spa, spa_mode_global);
2902 if (state != SPA_LOAD_RECOVER)
2903 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2905 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2906 policy.zrp_request);
2908 if (error == EBADF) {
2910 * If vdev_validate() returns failure (indicated by
2911 * EBADF), it indicates that one of the vdevs indicates
2912 * that the pool has been exported or destroyed. If
2913 * this is the case, the config cache is out of sync and
2914 * we should remove the pool from the namespace.
2917 spa_deactivate(spa);
2918 spa_config_sync(spa, B_TRUE, B_TRUE);
2921 mutex_exit(&spa_namespace_lock);
2922 return (SET_ERROR(ENOENT));
2927 * We can't open the pool, but we still have useful
2928 * information: the state of each vdev after the
2929 * attempted vdev_open(). Return this to the user.
2931 if (config != NULL && spa->spa_config) {
2932 VERIFY(nvlist_dup(spa->spa_config, config,
2934 VERIFY(nvlist_add_nvlist(*config,
2935 ZPOOL_CONFIG_LOAD_INFO,
2936 spa->spa_load_info) == 0);
2939 spa_deactivate(spa);
2940 spa->spa_last_open_failed = error;
2942 mutex_exit(&spa_namespace_lock);
2948 spa_open_ref(spa, tag);
2951 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2954 * If we've recovered the pool, pass back any information we
2955 * gathered while doing the load.
2957 if (state == SPA_LOAD_RECOVER) {
2958 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
2959 spa->spa_load_info) == 0);
2963 spa->spa_last_open_failed = 0;
2964 spa->spa_last_ubsync_txg = 0;
2965 spa->spa_load_txg = 0;
2966 mutex_exit(&spa_namespace_lock);
2971 zvol_create_minors(spa->spa_name);
2980 spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
2983 return (spa_open_common(name, spapp, tag, policy, config));
2987 spa_open(const char *name, spa_t **spapp, void *tag)
2989 return (spa_open_common(name, spapp, tag, NULL, NULL));
2993 * Lookup the given spa_t, incrementing the inject count in the process,
2994 * preventing it from being exported or destroyed.
2997 spa_inject_addref(char *name)
3001 mutex_enter(&spa_namespace_lock);
3002 if ((spa = spa_lookup(name)) == NULL) {
3003 mutex_exit(&spa_namespace_lock);
3006 spa->spa_inject_ref++;
3007 mutex_exit(&spa_namespace_lock);
3013 spa_inject_delref(spa_t *spa)
3015 mutex_enter(&spa_namespace_lock);
3016 spa->spa_inject_ref--;
3017 mutex_exit(&spa_namespace_lock);
3021 * Add spares device information to the nvlist.
3024 spa_add_spares(spa_t *spa, nvlist_t *config)
3034 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3036 if (spa->spa_spares.sav_count == 0)
3039 VERIFY(nvlist_lookup_nvlist(config,
3040 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3041 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
3042 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3044 VERIFY(nvlist_add_nvlist_array(nvroot,
3045 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3046 VERIFY(nvlist_lookup_nvlist_array(nvroot,
3047 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3050 * Go through and find any spares which have since been
3051 * repurposed as an active spare. If this is the case, update
3052 * their status appropriately.
3054 for (i = 0; i < nspares; i++) {
3055 VERIFY(nvlist_lookup_uint64(spares[i],
3056 ZPOOL_CONFIG_GUID, &guid) == 0);
3057 if (spa_spare_exists(guid, &pool, NULL) &&
3059 VERIFY(nvlist_lookup_uint64_array(
3060 spares[i], ZPOOL_CONFIG_VDEV_STATS,
3061 (uint64_t **)&vs, &vsc) == 0);
3062 vs->vs_state = VDEV_STATE_CANT_OPEN;
3063 vs->vs_aux = VDEV_AUX_SPARED;
3070 * Add l2cache device information to the nvlist, including vdev stats.
3073 spa_add_l2cache(spa_t *spa, nvlist_t *config)
3076 uint_t i, j, nl2cache;
3083 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3085 if (spa->spa_l2cache.sav_count == 0)
3088 VERIFY(nvlist_lookup_nvlist(config,
3089 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3090 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
3091 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3092 if (nl2cache != 0) {
3093 VERIFY(nvlist_add_nvlist_array(nvroot,
3094 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3095 VERIFY(nvlist_lookup_nvlist_array(nvroot,
3096 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3099 * Update level 2 cache device stats.
3102 for (i = 0; i < nl2cache; i++) {
3103 VERIFY(nvlist_lookup_uint64(l2cache[i],
3104 ZPOOL_CONFIG_GUID, &guid) == 0);
3107 for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
3109 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
3110 vd = spa->spa_l2cache.sav_vdevs[j];
3116 VERIFY(nvlist_lookup_uint64_array(l2cache[i],
3117 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
3119 vdev_get_stats(vd, vs);
3125 spa_add_feature_stats(spa_t *spa, nvlist_t *config)
3131 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3132 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3134 if (spa->spa_feat_for_read_obj != 0) {
3135 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3136 spa->spa_feat_for_read_obj);
3137 zap_cursor_retrieve(&zc, &za) == 0;
3138 zap_cursor_advance(&zc)) {
3139 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3140 za.za_num_integers == 1);
3141 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3142 za.za_first_integer));
3144 zap_cursor_fini(&zc);
3147 if (spa->spa_feat_for_write_obj != 0) {
3148 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3149 spa->spa_feat_for_write_obj);
3150 zap_cursor_retrieve(&zc, &za) == 0;
3151 zap_cursor_advance(&zc)) {
3152 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3153 za.za_num_integers == 1);
3154 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3155 za.za_first_integer));
3157 zap_cursor_fini(&zc);
3160 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
3162 nvlist_free(features);
3166 spa_get_stats(const char *name, nvlist_t **config,
3167 char *altroot, size_t buflen)
3173 error = spa_open_common(name, &spa, FTAG, NULL, config);
3177 * This still leaves a window of inconsistency where the spares
3178 * or l2cache devices could change and the config would be
3179 * self-inconsistent.
3181 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3183 if (*config != NULL) {
3184 uint64_t loadtimes[2];
3186 loadtimes[0] = spa->spa_loaded_ts.tv_sec;
3187 loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
3188 VERIFY(nvlist_add_uint64_array(*config,
3189 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
3191 VERIFY(nvlist_add_uint64(*config,
3192 ZPOOL_CONFIG_ERRCOUNT,
3193 spa_get_errlog_size(spa)) == 0);
3195 if (spa_suspended(spa))
3196 VERIFY(nvlist_add_uint64(*config,
3197 ZPOOL_CONFIG_SUSPENDED,
3198 spa->spa_failmode) == 0);
3200 spa_add_spares(spa, *config);
3201 spa_add_l2cache(spa, *config);
3202 spa_add_feature_stats(spa, *config);
3207 * We want to get the alternate root even for faulted pools, so we cheat
3208 * and call spa_lookup() directly.
3212 mutex_enter(&spa_namespace_lock);
3213 spa = spa_lookup(name);
3215 spa_altroot(spa, altroot, buflen);
3219 mutex_exit(&spa_namespace_lock);
3221 spa_altroot(spa, altroot, buflen);
3226 spa_config_exit(spa, SCL_CONFIG, FTAG);
3227 spa_close(spa, FTAG);
3234 * Validate that the auxiliary device array is well formed. We must have an
3235 * array of nvlists, each which describes a valid leaf vdev. If this is an
3236 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3237 * specified, as long as they are well-formed.
3240 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
3241 spa_aux_vdev_t *sav, const char *config, uint64_t version,
3242 vdev_labeltype_t label)
3249 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3252 * It's acceptable to have no devs specified.
3254 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
3258 return (SET_ERROR(EINVAL));
3261 * Make sure the pool is formatted with a version that supports this
3264 if (spa_version(spa) < version)
3265 return (SET_ERROR(ENOTSUP));
3268 * Set the pending device list so we correctly handle device in-use
3271 sav->sav_pending = dev;
3272 sav->sav_npending = ndev;
3274 for (i = 0; i < ndev; i++) {
3275 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
3279 if (!vd->vdev_ops->vdev_op_leaf) {
3281 error = SET_ERROR(EINVAL);
3286 * The L2ARC currently only supports disk devices in
3287 * kernel context. For user-level testing, we allow it.
3290 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
3291 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
3292 error = SET_ERROR(ENOTBLK);
3299 if ((error = vdev_open(vd)) == 0 &&
3300 (error = vdev_label_init(vd, crtxg, label)) == 0) {
3301 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
3302 vd->vdev_guid) == 0);
3308 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
3315 sav->sav_pending = NULL;
3316 sav->sav_npending = 0;
3321 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
3325 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3327 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3328 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
3329 VDEV_LABEL_SPARE)) != 0) {
3333 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3334 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
3335 VDEV_LABEL_L2CACHE));
3339 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
3344 if (sav->sav_config != NULL) {
3350 * Generate new dev list by concatentating with the
3353 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
3354 &olddevs, &oldndevs) == 0);
3356 newdevs = kmem_alloc(sizeof (void *) *
3357 (ndevs + oldndevs), KM_PUSHPAGE);
3358 for (i = 0; i < oldndevs; i++)
3359 VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
3361 for (i = 0; i < ndevs; i++)
3362 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
3365 VERIFY(nvlist_remove(sav->sav_config, config,
3366 DATA_TYPE_NVLIST_ARRAY) == 0);
3368 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3369 config, newdevs, ndevs + oldndevs) == 0);
3370 for (i = 0; i < oldndevs + ndevs; i++)
3371 nvlist_free(newdevs[i]);
3372 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
3375 * Generate a new dev list.
3377 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
3379 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
3385 * Stop and drop level 2 ARC devices
3388 spa_l2cache_drop(spa_t *spa)
3392 spa_aux_vdev_t *sav = &spa->spa_l2cache;
3394 for (i = 0; i < sav->sav_count; i++) {
3397 vd = sav->sav_vdevs[i];
3400 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
3401 pool != 0ULL && l2arc_vdev_present(vd))
3402 l2arc_remove_vdev(vd);
3410 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
3414 char *altroot = NULL;
3419 uint64_t txg = TXG_INITIAL;
3420 nvlist_t **spares, **l2cache;
3421 uint_t nspares, nl2cache;
3422 uint64_t version, obj;
3423 boolean_t has_features;
3428 * If this pool already exists, return failure.
3430 mutex_enter(&spa_namespace_lock);
3431 if (spa_lookup(pool) != NULL) {
3432 mutex_exit(&spa_namespace_lock);
3433 return (SET_ERROR(EEXIST));
3437 * Allocate a new spa_t structure.
3439 (void) nvlist_lookup_string(props,
3440 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3441 spa = spa_add(pool, NULL, altroot);
3442 spa_activate(spa, spa_mode_global);
3444 if (props && (error = spa_prop_validate(spa, props))) {
3445 spa_deactivate(spa);
3447 mutex_exit(&spa_namespace_lock);
3451 has_features = B_FALSE;
3452 for (elem = nvlist_next_nvpair(props, NULL);
3453 elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
3454 if (zpool_prop_feature(nvpair_name(elem)))
3455 has_features = B_TRUE;
3458 if (has_features || nvlist_lookup_uint64(props,
3459 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
3460 version = SPA_VERSION;
3462 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
3464 spa->spa_first_txg = txg;
3465 spa->spa_uberblock.ub_txg = txg - 1;
3466 spa->spa_uberblock.ub_version = version;
3467 spa->spa_ubsync = spa->spa_uberblock;
3470 * Create "The Godfather" zio to hold all async IOs
3472 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
3473 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
3476 * Create the root vdev.
3478 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3480 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3482 ASSERT(error != 0 || rvd != NULL);
3483 ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3485 if (error == 0 && !zfs_allocatable_devs(nvroot))
3486 error = SET_ERROR(EINVAL);
3489 (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3490 (error = spa_validate_aux(spa, nvroot, txg,
3491 VDEV_ALLOC_ADD)) == 0) {
3492 for (c = 0; c < rvd->vdev_children; c++) {
3493 vdev_metaslab_set_size(rvd->vdev_child[c]);
3494 vdev_expand(rvd->vdev_child[c], txg);
3498 spa_config_exit(spa, SCL_ALL, FTAG);
3502 spa_deactivate(spa);
3504 mutex_exit(&spa_namespace_lock);
3509 * Get the list of spares, if specified.
3511 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3512 &spares, &nspares) == 0) {
3513 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3515 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3516 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3517 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3518 spa_load_spares(spa);
3519 spa_config_exit(spa, SCL_ALL, FTAG);
3520 spa->spa_spares.sav_sync = B_TRUE;
3524 * Get the list of level 2 cache devices, if specified.
3526 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3527 &l2cache, &nl2cache) == 0) {
3528 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3529 NV_UNIQUE_NAME, KM_PUSHPAGE) == 0);
3530 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3531 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3532 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3533 spa_load_l2cache(spa);
3534 spa_config_exit(spa, SCL_ALL, FTAG);
3535 spa->spa_l2cache.sav_sync = B_TRUE;
3538 spa->spa_is_initializing = B_TRUE;
3539 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3540 spa->spa_meta_objset = dp->dp_meta_objset;
3541 spa->spa_is_initializing = B_FALSE;
3544 * Create DDTs (dedup tables).
3548 spa_update_dspace(spa);
3550 tx = dmu_tx_create_assigned(dp, txg);
3553 * Create the pool config object.
3555 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3556 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3557 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3559 if (zap_add(spa->spa_meta_objset,
3560 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3561 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3562 cmn_err(CE_PANIC, "failed to add pool config");
3565 if (spa_version(spa) >= SPA_VERSION_FEATURES)
3566 spa_feature_create_zap_objects(spa, tx);
3568 if (zap_add(spa->spa_meta_objset,
3569 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3570 sizeof (uint64_t), 1, &version, tx) != 0) {
3571 cmn_err(CE_PANIC, "failed to add pool version");
3574 /* Newly created pools with the right version are always deflated. */
3575 if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3576 spa->spa_deflate = TRUE;
3577 if (zap_add(spa->spa_meta_objset,
3578 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3579 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3580 cmn_err(CE_PANIC, "failed to add deflate");
3585 * Create the deferred-free bpobj. Turn off compression
3586 * because sync-to-convergence takes longer if the blocksize
3589 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3590 dmu_object_set_compress(spa->spa_meta_objset, obj,
3591 ZIO_COMPRESS_OFF, tx);
3592 if (zap_add(spa->spa_meta_objset,
3593 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3594 sizeof (uint64_t), 1, &obj, tx) != 0) {
3595 cmn_err(CE_PANIC, "failed to add bpobj");
3597 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3598 spa->spa_meta_objset, obj));
3601 * Create the pool's history object.
3603 if (version >= SPA_VERSION_ZPOOL_HISTORY)
3604 spa_history_create_obj(spa, tx);
3607 * Set pool properties.
3609 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3610 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3611 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3612 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3614 if (props != NULL) {
3615 spa_configfile_set(spa, props, B_FALSE);
3616 spa_sync_props(props, tx);
3621 spa->spa_sync_on = B_TRUE;
3622 txg_sync_start(spa->spa_dsl_pool);
3625 * We explicitly wait for the first transaction to complete so that our
3626 * bean counters are appropriately updated.
3628 txg_wait_synced(spa->spa_dsl_pool, txg);
3630 spa_config_sync(spa, B_FALSE, B_TRUE);
3632 spa_history_log_version(spa, "create");
3634 spa->spa_minref = refcount_count(&spa->spa_refcount);
3636 mutex_exit(&spa_namespace_lock);
3643 * Get the root pool information from the root disk, then import the root pool
3644 * during the system boot up time.
3646 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3649 spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3652 nvlist_t *nvtop, *nvroot;
3655 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3659 * Add this top-level vdev to the child array.
3661 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3663 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3665 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3668 * Put this pool's top-level vdevs into a root vdev.
3670 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_PUSHPAGE) == 0);
3671 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3672 VDEV_TYPE_ROOT) == 0);
3673 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3674 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3675 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3679 * Replace the existing vdev_tree with the new root vdev in
3680 * this pool's configuration (remove the old, add the new).
3682 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3683 nvlist_free(nvroot);
3688 * Walk the vdev tree and see if we can find a device with "better"
3689 * configuration. A configuration is "better" if the label on that
3690 * device has a more recent txg.
3693 spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3697 for (c = 0; c < vd->vdev_children; c++)
3698 spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3700 if (vd->vdev_ops->vdev_op_leaf) {
3704 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3708 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3712 * Do we have a better boot device?
3714 if (label_txg > *txg) {
3723 * Import a root pool.
3725 * For x86. devpath_list will consist of devid and/or physpath name of
3726 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3727 * The GRUB "findroot" command will return the vdev we should boot.
3729 * For Sparc, devpath_list consists the physpath name of the booting device
3730 * no matter the rootpool is a single device pool or a mirrored pool.
3732 * "/pci@1f,0/ide@d/disk@0,0:a"
3735 spa_import_rootpool(char *devpath, char *devid)
3738 vdev_t *rvd, *bvd, *avd = NULL;
3739 nvlist_t *config, *nvtop;
3745 * Read the label from the boot device and generate a configuration.
3747 config = spa_generate_rootconf(devpath, devid, &guid);
3748 #if defined(_OBP) && defined(_KERNEL)
3749 if (config == NULL) {
3750 if (strstr(devpath, "/iscsi/ssd") != NULL) {
3752 get_iscsi_bootpath_phy(devpath);
3753 config = spa_generate_rootconf(devpath, devid, &guid);
3757 if (config == NULL) {
3758 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
3760 return (SET_ERROR(EIO));
3763 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3765 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3767 mutex_enter(&spa_namespace_lock);
3768 if ((spa = spa_lookup(pname)) != NULL) {
3770 * Remove the existing root pool from the namespace so that we
3771 * can replace it with the correct config we just read in.
3776 spa = spa_add(pname, config, NULL);
3777 spa->spa_is_root = B_TRUE;
3778 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3781 * Build up a vdev tree based on the boot device's label config.
3783 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3785 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3786 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3787 VDEV_ALLOC_ROOTPOOL);
3788 spa_config_exit(spa, SCL_ALL, FTAG);
3790 mutex_exit(&spa_namespace_lock);
3791 nvlist_free(config);
3792 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3798 * Get the boot vdev.
3800 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3801 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3802 (u_longlong_t)guid);
3803 error = SET_ERROR(ENOENT);
3808 * Determine if there is a better boot device.
3811 spa_alt_rootvdev(rvd, &avd, &txg);
3813 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3814 "try booting from '%s'", avd->vdev_path);
3815 error = SET_ERROR(EINVAL);
3820 * If the boot device is part of a spare vdev then ensure that
3821 * we're booting off the active spare.
3823 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3824 !bvd->vdev_isspare) {
3825 cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3826 "try booting from '%s'",
3828 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3829 error = SET_ERROR(EINVAL);
3835 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3837 spa_config_exit(spa, SCL_ALL, FTAG);
3838 mutex_exit(&spa_namespace_lock);
3840 nvlist_free(config);
3847 * Import a non-root pool into the system.
3850 spa_import(char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
3853 char *altroot = NULL;
3854 spa_load_state_t state = SPA_LOAD_IMPORT;
3855 zpool_rewind_policy_t policy;
3856 uint64_t mode = spa_mode_global;
3857 uint64_t readonly = B_FALSE;
3860 nvlist_t **spares, **l2cache;
3861 uint_t nspares, nl2cache;
3864 * If a pool with this name exists, return failure.
3866 mutex_enter(&spa_namespace_lock);
3867 if (spa_lookup(pool) != NULL) {
3868 mutex_exit(&spa_namespace_lock);
3869 return (SET_ERROR(EEXIST));
3873 * Create and initialize the spa structure.
3875 (void) nvlist_lookup_string(props,
3876 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3877 (void) nvlist_lookup_uint64(props,
3878 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
3881 spa = spa_add(pool, config, altroot);
3882 spa->spa_import_flags = flags;
3885 * Verbatim import - Take a pool and insert it into the namespace
3886 * as if it had been loaded at boot.
3888 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
3890 spa_configfile_set(spa, props, B_FALSE);
3892 spa_config_sync(spa, B_FALSE, B_TRUE);
3894 mutex_exit(&spa_namespace_lock);
3895 spa_history_log_version(spa, "import");
3900 spa_activate(spa, mode);
3903 * Don't start async tasks until we know everything is healthy.
3905 spa_async_suspend(spa);
3907 zpool_get_rewind_policy(config, &policy);
3908 if (policy.zrp_request & ZPOOL_DO_REWIND)
3909 state = SPA_LOAD_RECOVER;
3912 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
3913 * because the user-supplied config is actually the one to trust when
3916 if (state != SPA_LOAD_RECOVER)
3917 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3919 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
3920 policy.zrp_request);
3923 * Propagate anything learned while loading the pool and pass it
3924 * back to caller (i.e. rewind info, missing devices, etc).
3926 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
3927 spa->spa_load_info) == 0);
3929 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3931 * Toss any existing sparelist, as it doesn't have any validity
3932 * anymore, and conflicts with spa_has_spare().
3934 if (spa->spa_spares.sav_config) {
3935 nvlist_free(spa->spa_spares.sav_config);
3936 spa->spa_spares.sav_config = NULL;
3937 spa_load_spares(spa);
3939 if (spa->spa_l2cache.sav_config) {
3940 nvlist_free(spa->spa_l2cache.sav_config);
3941 spa->spa_l2cache.sav_config = NULL;
3942 spa_load_l2cache(spa);
3945 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3948 error = spa_validate_aux(spa, nvroot, -1ULL,
3951 error = spa_validate_aux(spa, nvroot, -1ULL,
3952 VDEV_ALLOC_L2CACHE);
3953 spa_config_exit(spa, SCL_ALL, FTAG);
3956 spa_configfile_set(spa, props, B_FALSE);
3958 if (error != 0 || (props && spa_writeable(spa) &&
3959 (error = spa_prop_set(spa, props)))) {
3961 spa_deactivate(spa);
3963 mutex_exit(&spa_namespace_lock);
3967 spa_async_resume(spa);
3970 * Override any spares and level 2 cache devices as specified by
3971 * the user, as these may have correct device names/devids, etc.
3973 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3974 &spares, &nspares) == 0) {
3975 if (spa->spa_spares.sav_config)
3976 VERIFY(nvlist_remove(spa->spa_spares.sav_config,
3977 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
3979 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
3980 NV_UNIQUE_NAME, KM_PUSHPAGE) == 0);
3981 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3982 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3983 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3984 spa_load_spares(spa);
3985 spa_config_exit(spa, SCL_ALL, FTAG);
3986 spa->spa_spares.sav_sync = B_TRUE;
3988 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3989 &l2cache, &nl2cache) == 0) {
3990 if (spa->spa_l2cache.sav_config)
3991 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
3992 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
3994 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3995 NV_UNIQUE_NAME, KM_PUSHPAGE) == 0);
3996 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3997 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3998 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3999 spa_load_l2cache(spa);
4000 spa_config_exit(spa, SCL_ALL, FTAG);
4001 spa->spa_l2cache.sav_sync = B_TRUE;
4005 * Check for any removed devices.
4007 if (spa->spa_autoreplace) {
4008 spa_aux_check_removed(&spa->spa_spares);
4009 spa_aux_check_removed(&spa->spa_l2cache);
4012 if (spa_writeable(spa)) {
4014 * Update the config cache to include the newly-imported pool.
4016 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4020 * It's possible that the pool was expanded while it was exported.
4021 * We kick off an async task to handle this for us.
4023 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
4025 mutex_exit(&spa_namespace_lock);
4026 spa_history_log_version(spa, "import");
4029 zvol_create_minors(pool);
4036 spa_tryimport(nvlist_t *tryconfig)
4038 nvlist_t *config = NULL;
4044 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
4047 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
4051 * Create and initialize the spa structure.
4053 mutex_enter(&spa_namespace_lock);
4054 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
4055 spa_activate(spa, FREAD);
4058 * Pass off the heavy lifting to spa_load().
4059 * Pass TRUE for mosconfig because the user-supplied config
4060 * is actually the one to trust when doing an import.
4062 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
4065 * If 'tryconfig' was at least parsable, return the current config.
4067 if (spa->spa_root_vdev != NULL) {
4068 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4069 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
4071 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4073 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
4074 spa->spa_uberblock.ub_timestamp) == 0);
4075 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4076 spa->spa_load_info) == 0);
4079 * If the bootfs property exists on this pool then we
4080 * copy it out so that external consumers can tell which
4081 * pools are bootable.
4083 if ((!error || error == EEXIST) && spa->spa_bootfs) {
4084 char *tmpname = kmem_alloc(MAXPATHLEN, KM_PUSHPAGE);
4087 * We have to play games with the name since the
4088 * pool was opened as TRYIMPORT_NAME.
4090 if (dsl_dsobj_to_dsname(spa_name(spa),
4091 spa->spa_bootfs, tmpname) == 0) {
4093 char *dsname = kmem_alloc(MAXPATHLEN, KM_PUSHPAGE);
4095 cp = strchr(tmpname, '/');
4097 (void) strlcpy(dsname, tmpname,
4100 (void) snprintf(dsname, MAXPATHLEN,
4101 "%s/%s", poolname, ++cp);
4103 VERIFY(nvlist_add_string(config,
4104 ZPOOL_CONFIG_BOOTFS, dsname) == 0);
4105 kmem_free(dsname, MAXPATHLEN);
4107 kmem_free(tmpname, MAXPATHLEN);
4111 * Add the list of hot spares and level 2 cache devices.
4113 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4114 spa_add_spares(spa, config);
4115 spa_add_l2cache(spa, config);
4116 spa_config_exit(spa, SCL_CONFIG, FTAG);
4120 spa_deactivate(spa);
4122 mutex_exit(&spa_namespace_lock);
4128 * Pool export/destroy
4130 * The act of destroying or exporting a pool is very simple. We make sure there
4131 * is no more pending I/O and any references to the pool are gone. Then, we
4132 * update the pool state and sync all the labels to disk, removing the
4133 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4134 * we don't sync the labels or remove the configuration cache.
4137 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
4138 boolean_t force, boolean_t hardforce)
4145 if (!(spa_mode_global & FWRITE))
4146 return (SET_ERROR(EROFS));
4148 mutex_enter(&spa_namespace_lock);
4149 if ((spa = spa_lookup(pool)) == NULL) {
4150 mutex_exit(&spa_namespace_lock);
4151 return (SET_ERROR(ENOENT));
4155 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4156 * reacquire the namespace lock, and see if we can export.
4158 spa_open_ref(spa, FTAG);
4159 mutex_exit(&spa_namespace_lock);
4160 spa_async_suspend(spa);
4161 mutex_enter(&spa_namespace_lock);
4162 spa_close(spa, FTAG);
4165 * The pool will be in core if it's openable,
4166 * in which case we can modify its state.
4168 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
4170 * Objsets may be open only because they're dirty, so we
4171 * have to force it to sync before checking spa_refcnt.
4173 txg_wait_synced(spa->spa_dsl_pool, 0);
4176 * A pool cannot be exported or destroyed if there are active
4177 * references. If we are resetting a pool, allow references by
4178 * fault injection handlers.
4180 if (!spa_refcount_zero(spa) ||
4181 (spa->spa_inject_ref != 0 &&
4182 new_state != POOL_STATE_UNINITIALIZED)) {
4183 spa_async_resume(spa);
4184 mutex_exit(&spa_namespace_lock);
4185 return (SET_ERROR(EBUSY));
4189 * A pool cannot be exported if it has an active shared spare.
4190 * This is to prevent other pools stealing the active spare
4191 * from an exported pool. At user's own will, such pool can
4192 * be forcedly exported.
4194 if (!force && new_state == POOL_STATE_EXPORTED &&
4195 spa_has_active_shared_spare(spa)) {
4196 spa_async_resume(spa);
4197 mutex_exit(&spa_namespace_lock);
4198 return (SET_ERROR(EXDEV));
4202 * We want this to be reflected on every label,
4203 * so mark them all dirty. spa_unload() will do the
4204 * final sync that pushes these changes out.
4206 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
4207 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4208 spa->spa_state = new_state;
4209 spa->spa_final_txg = spa_last_synced_txg(spa) +
4211 vdev_config_dirty(spa->spa_root_vdev);
4212 spa_config_exit(spa, SCL_ALL, FTAG);
4216 spa_event_notify(spa, NULL, FM_EREPORT_ZFS_POOL_DESTROY);
4218 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4220 spa_deactivate(spa);
4223 if (oldconfig && spa->spa_config)
4224 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
4226 if (new_state != POOL_STATE_UNINITIALIZED) {
4228 spa_config_sync(spa, B_TRUE, B_TRUE);
4231 mutex_exit(&spa_namespace_lock);
4237 * Destroy a storage pool.
4240 spa_destroy(char *pool)
4242 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
4247 * Export a storage pool.
4250 spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
4251 boolean_t hardforce)
4253 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
4258 * Similar to spa_export(), this unloads the spa_t without actually removing it
4259 * from the namespace in any way.
4262 spa_reset(char *pool)
4264 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
4269 * ==========================================================================
4270 * Device manipulation
4271 * ==========================================================================
4275 * Add a device to a storage pool.
4278 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
4282 vdev_t *rvd = spa->spa_root_vdev;
4284 nvlist_t **spares, **l2cache;
4285 uint_t nspares, nl2cache;
4288 ASSERT(spa_writeable(spa));
4290 txg = spa_vdev_enter(spa);
4292 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
4293 VDEV_ALLOC_ADD)) != 0)
4294 return (spa_vdev_exit(spa, NULL, txg, error));
4296 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */
4298 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
4302 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
4306 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
4307 return (spa_vdev_exit(spa, vd, txg, EINVAL));
4309 if (vd->vdev_children != 0 &&
4310 (error = vdev_create(vd, txg, B_FALSE)) != 0)
4311 return (spa_vdev_exit(spa, vd, txg, error));
4314 * We must validate the spares and l2cache devices after checking the
4315 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
4317 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
4318 return (spa_vdev_exit(spa, vd, txg, error));
4321 * Transfer each new top-level vdev from vd to rvd.
4323 for (c = 0; c < vd->vdev_children; c++) {
4326 * Set the vdev id to the first hole, if one exists.
4328 for (id = 0; id < rvd->vdev_children; id++) {
4329 if (rvd->vdev_child[id]->vdev_ishole) {
4330 vdev_free(rvd->vdev_child[id]);
4334 tvd = vd->vdev_child[c];
4335 vdev_remove_child(vd, tvd);
4337 vdev_add_child(rvd, tvd);
4338 vdev_config_dirty(tvd);
4342 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
4343 ZPOOL_CONFIG_SPARES);
4344 spa_load_spares(spa);
4345 spa->spa_spares.sav_sync = B_TRUE;
4348 if (nl2cache != 0) {
4349 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
4350 ZPOOL_CONFIG_L2CACHE);
4351 spa_load_l2cache(spa);
4352 spa->spa_l2cache.sav_sync = B_TRUE;
4356 * We have to be careful when adding new vdevs to an existing pool.
4357 * If other threads start allocating from these vdevs before we
4358 * sync the config cache, and we lose power, then upon reboot we may
4359 * fail to open the pool because there are DVAs that the config cache
4360 * can't translate. Therefore, we first add the vdevs without
4361 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4362 * and then let spa_config_update() initialize the new metaslabs.
4364 * spa_load() checks for added-but-not-initialized vdevs, so that
4365 * if we lose power at any point in this sequence, the remaining
4366 * steps will be completed the next time we load the pool.
4368 (void) spa_vdev_exit(spa, vd, txg, 0);
4370 mutex_enter(&spa_namespace_lock);
4371 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4372 mutex_exit(&spa_namespace_lock);
4378 * Attach a device to a mirror. The arguments are the path to any device
4379 * in the mirror, and the nvroot for the new device. If the path specifies
4380 * a device that is not mirrored, we automatically insert the mirror vdev.
4382 * If 'replacing' is specified, the new device is intended to replace the
4383 * existing device; in this case the two devices are made into their own
4384 * mirror using the 'replacing' vdev, which is functionally identical to
4385 * the mirror vdev (it actually reuses all the same ops) but has a few
4386 * extra rules: you can't attach to it after it's been created, and upon
4387 * completion of resilvering, the first disk (the one being replaced)
4388 * is automatically detached.
4391 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
4393 uint64_t txg, dtl_max_txg;
4394 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
4396 char *oldvdpath, *newvdpath;
4399 ASSERTV(vdev_t *rvd = spa->spa_root_vdev);
4401 ASSERT(spa_writeable(spa));
4403 txg = spa_vdev_enter(spa);
4405 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
4408 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4410 if (!oldvd->vdev_ops->vdev_op_leaf)
4411 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4413 pvd = oldvd->vdev_parent;
4415 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
4416 VDEV_ALLOC_ATTACH)) != 0)
4417 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4419 if (newrootvd->vdev_children != 1)
4420 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4422 newvd = newrootvd->vdev_child[0];
4424 if (!newvd->vdev_ops->vdev_op_leaf)
4425 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4427 if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
4428 return (spa_vdev_exit(spa, newrootvd, txg, error));
4431 * Spares can't replace logs
4433 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
4434 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4438 * For attach, the only allowable parent is a mirror or the root
4441 if (pvd->vdev_ops != &vdev_mirror_ops &&
4442 pvd->vdev_ops != &vdev_root_ops)
4443 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4445 pvops = &vdev_mirror_ops;
4448 * Active hot spares can only be replaced by inactive hot
4451 if (pvd->vdev_ops == &vdev_spare_ops &&
4452 oldvd->vdev_isspare &&
4453 !spa_has_spare(spa, newvd->vdev_guid))
4454 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4457 * If the source is a hot spare, and the parent isn't already a
4458 * spare, then we want to create a new hot spare. Otherwise, we
4459 * want to create a replacing vdev. The user is not allowed to
4460 * attach to a spared vdev child unless the 'isspare' state is
4461 * the same (spare replaces spare, non-spare replaces
4464 if (pvd->vdev_ops == &vdev_replacing_ops &&
4465 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
4466 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4467 } else if (pvd->vdev_ops == &vdev_spare_ops &&
4468 newvd->vdev_isspare != oldvd->vdev_isspare) {
4469 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4472 if (newvd->vdev_isspare)
4473 pvops = &vdev_spare_ops;
4475 pvops = &vdev_replacing_ops;
4479 * Make sure the new device is big enough.
4481 if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
4482 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
4485 * The new device cannot have a higher alignment requirement
4486 * than the top-level vdev.
4488 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4489 return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4492 * If this is an in-place replacement, update oldvd's path and devid
4493 * to make it distinguishable from newvd, and unopenable from now on.
4495 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4496 spa_strfree(oldvd->vdev_path);
4497 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4499 (void) sprintf(oldvd->vdev_path, "%s/%s",
4500 newvd->vdev_path, "old");
4501 if (oldvd->vdev_devid != NULL) {
4502 spa_strfree(oldvd->vdev_devid);
4503 oldvd->vdev_devid = NULL;
4507 /* mark the device being resilvered */
4508 newvd->vdev_resilvering = B_TRUE;
4511 * If the parent is not a mirror, or if we're replacing, insert the new
4512 * mirror/replacing/spare vdev above oldvd.
4514 if (pvd->vdev_ops != pvops)
4515 pvd = vdev_add_parent(oldvd, pvops);
4517 ASSERT(pvd->vdev_top->vdev_parent == rvd);
4518 ASSERT(pvd->vdev_ops == pvops);
4519 ASSERT(oldvd->vdev_parent == pvd);
4522 * Extract the new device from its root and add it to pvd.
4524 vdev_remove_child(newrootvd, newvd);
4525 newvd->vdev_id = pvd->vdev_children;
4526 newvd->vdev_crtxg = oldvd->vdev_crtxg;
4527 vdev_add_child(pvd, newvd);
4529 tvd = newvd->vdev_top;
4530 ASSERT(pvd->vdev_top == tvd);
4531 ASSERT(tvd->vdev_parent == rvd);
4533 vdev_config_dirty(tvd);
4536 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4537 * for any dmu_sync-ed blocks. It will propagate upward when
4538 * spa_vdev_exit() calls vdev_dtl_reassess().
4540 dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4542 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4543 dtl_max_txg - TXG_INITIAL);
4545 if (newvd->vdev_isspare) {
4546 spa_spare_activate(newvd);
4547 spa_event_notify(spa, newvd, FM_EREPORT_ZFS_DEVICE_SPARE);
4550 oldvdpath = spa_strdup(oldvd->vdev_path);
4551 newvdpath = spa_strdup(newvd->vdev_path);
4552 newvd_isspare = newvd->vdev_isspare;
4555 * Mark newvd's DTL dirty in this txg.
4557 vdev_dirty(tvd, VDD_DTL, newvd, txg);
4560 * Restart the resilver
4562 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4567 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4569 spa_history_log_internal(spa, "vdev attach", NULL,
4570 "%s vdev=%s %s vdev=%s",
4571 replacing && newvd_isspare ? "spare in" :
4572 replacing ? "replace" : "attach", newvdpath,
4573 replacing ? "for" : "to", oldvdpath);
4575 spa_strfree(oldvdpath);
4576 spa_strfree(newvdpath);
4578 if (spa->spa_bootfs)
4579 spa_event_notify(spa, newvd, FM_EREPORT_ZFS_BOOTFS_VDEV_ATTACH);
4585 * Detach a device from a mirror or replacing vdev.
4586 * If 'replace_done' is specified, only detach if the parent
4587 * is a replacing vdev.
4590 spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4594 vdev_t *vd, *pvd, *cvd, *tvd;
4595 boolean_t unspare = B_FALSE;
4596 uint64_t unspare_guid = 0;
4599 ASSERTV(vdev_t *rvd = spa->spa_root_vdev);
4600 ASSERT(spa_writeable(spa));
4602 txg = spa_vdev_enter(spa);
4604 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4607 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4609 if (!vd->vdev_ops->vdev_op_leaf)
4610 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4612 pvd = vd->vdev_parent;
4615 * If the parent/child relationship is not as expected, don't do it.
4616 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4617 * vdev that's replacing B with C. The user's intent in replacing
4618 * is to go from M(A,B) to M(A,C). If the user decides to cancel
4619 * the replace by detaching C, the expected behavior is to end up
4620 * M(A,B). But suppose that right after deciding to detach C,
4621 * the replacement of B completes. We would have M(A,C), and then
4622 * ask to detach C, which would leave us with just A -- not what
4623 * the user wanted. To prevent this, we make sure that the
4624 * parent/child relationship hasn't changed -- in this example,
4625 * that C's parent is still the replacing vdev R.
4627 if (pvd->vdev_guid != pguid && pguid != 0)
4628 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4631 * Only 'replacing' or 'spare' vdevs can be replaced.
4633 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4634 pvd->vdev_ops != &vdev_spare_ops)
4635 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4637 ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4638 spa_version(spa) >= SPA_VERSION_SPARES);
4641 * Only mirror, replacing, and spare vdevs support detach.
4643 if (pvd->vdev_ops != &vdev_replacing_ops &&
4644 pvd->vdev_ops != &vdev_mirror_ops &&
4645 pvd->vdev_ops != &vdev_spare_ops)
4646 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4649 * If this device has the only valid copy of some data,
4650 * we cannot safely detach it.
4652 if (vdev_dtl_required(vd))
4653 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4655 ASSERT(pvd->vdev_children >= 2);
4658 * If we are detaching the second disk from a replacing vdev, then
4659 * check to see if we changed the original vdev's path to have "/old"
4660 * at the end in spa_vdev_attach(). If so, undo that change now.
4662 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4663 vd->vdev_path != NULL) {
4664 size_t len = strlen(vd->vdev_path);
4666 for (c = 0; c < pvd->vdev_children; c++) {
4667 cvd = pvd->vdev_child[c];
4669 if (cvd == vd || cvd->vdev_path == NULL)
4672 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4673 strcmp(cvd->vdev_path + len, "/old") == 0) {
4674 spa_strfree(cvd->vdev_path);
4675 cvd->vdev_path = spa_strdup(vd->vdev_path);
4682 * If we are detaching the original disk from a spare, then it implies
4683 * that the spare should become a real disk, and be removed from the
4684 * active spare list for the pool.
4686 if (pvd->vdev_ops == &vdev_spare_ops &&
4688 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4692 * Erase the disk labels so the disk can be used for other things.
4693 * This must be done after all other error cases are handled,
4694 * but before we disembowel vd (so we can still do I/O to it).
4695 * But if we can't do it, don't treat the error as fatal --
4696 * it may be that the unwritability of the disk is the reason
4697 * it's being detached!
4699 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4702 * Remove vd from its parent and compact the parent's children.
4704 vdev_remove_child(pvd, vd);
4705 vdev_compact_children(pvd);
4708 * Remember one of the remaining children so we can get tvd below.
4710 cvd = pvd->vdev_child[pvd->vdev_children - 1];
4713 * If we need to remove the remaining child from the list of hot spares,
4714 * do it now, marking the vdev as no longer a spare in the process.
4715 * We must do this before vdev_remove_parent(), because that can
4716 * change the GUID if it creates a new toplevel GUID. For a similar
4717 * reason, we must remove the spare now, in the same txg as the detach;
4718 * otherwise someone could attach a new sibling, change the GUID, and
4719 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4722 ASSERT(cvd->vdev_isspare);
4723 spa_spare_remove(cvd);
4724 unspare_guid = cvd->vdev_guid;
4725 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4726 cvd->vdev_unspare = B_TRUE;
4730 * If the parent mirror/replacing vdev only has one child,
4731 * the parent is no longer needed. Remove it from the tree.
4733 if (pvd->vdev_children == 1) {
4734 if (pvd->vdev_ops == &vdev_spare_ops)
4735 cvd->vdev_unspare = B_FALSE;
4736 vdev_remove_parent(cvd);
4737 cvd->vdev_resilvering = B_FALSE;
4742 * We don't set tvd until now because the parent we just removed
4743 * may have been the previous top-level vdev.
4745 tvd = cvd->vdev_top;
4746 ASSERT(tvd->vdev_parent == rvd);
4749 * Reevaluate the parent vdev state.
4751 vdev_propagate_state(cvd);
4754 * If the 'autoexpand' property is set on the pool then automatically
4755 * try to expand the size of the pool. For example if the device we
4756 * just detached was smaller than the others, it may be possible to
4757 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4758 * first so that we can obtain the updated sizes of the leaf vdevs.
4760 if (spa->spa_autoexpand) {
4762 vdev_expand(tvd, txg);
4765 vdev_config_dirty(tvd);
4768 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
4769 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4770 * But first make sure we're not on any *other* txg's DTL list, to
4771 * prevent vd from being accessed after it's freed.
4773 vdpath = spa_strdup(vd->vdev_path);
4774 for (t = 0; t < TXG_SIZE; t++)
4775 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4776 vd->vdev_detached = B_TRUE;
4777 vdev_dirty(tvd, VDD_DTL, vd, txg);
4779 spa_event_notify(spa, vd, FM_EREPORT_ZFS_DEVICE_REMOVE);
4781 /* hang on to the spa before we release the lock */
4782 spa_open_ref(spa, FTAG);
4784 error = spa_vdev_exit(spa, vd, txg, 0);
4786 spa_history_log_internal(spa, "detach", NULL,
4788 spa_strfree(vdpath);
4791 * If this was the removal of the original device in a hot spare vdev,
4792 * then we want to go through and remove the device from the hot spare
4793 * list of every other pool.
4796 spa_t *altspa = NULL;
4798 mutex_enter(&spa_namespace_lock);
4799 while ((altspa = spa_next(altspa)) != NULL) {
4800 if (altspa->spa_state != POOL_STATE_ACTIVE ||
4804 spa_open_ref(altspa, FTAG);
4805 mutex_exit(&spa_namespace_lock);
4806 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
4807 mutex_enter(&spa_namespace_lock);
4808 spa_close(altspa, FTAG);
4810 mutex_exit(&spa_namespace_lock);
4812 /* search the rest of the vdevs for spares to remove */
4813 spa_vdev_resilver_done(spa);
4816 /* all done with the spa; OK to release */
4817 mutex_enter(&spa_namespace_lock);
4818 spa_close(spa, FTAG);
4819 mutex_exit(&spa_namespace_lock);
4825 * Split a set of devices from their mirrors, and create a new pool from them.
4828 spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
4829 nvlist_t *props, boolean_t exp)
4832 uint64_t txg, *glist;
4834 uint_t c, children, lastlog;
4835 nvlist_t **child, *nvl, *tmp;
4837 char *altroot = NULL;
4838 vdev_t *rvd, **vml = NULL; /* vdev modify list */
4839 boolean_t activate_slog;
4841 ASSERT(spa_writeable(spa));
4843 txg = spa_vdev_enter(spa);
4845 /* clear the log and flush everything up to now */
4846 activate_slog = spa_passivate_log(spa);
4847 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4848 error = spa_offline_log(spa);
4849 txg = spa_vdev_config_enter(spa);
4852 spa_activate_log(spa);
4855 return (spa_vdev_exit(spa, NULL, txg, error));
4857 /* check new spa name before going any further */
4858 if (spa_lookup(newname) != NULL)
4859 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
4862 * scan through all the children to ensure they're all mirrors
4864 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
4865 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
4867 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4869 /* first, check to ensure we've got the right child count */
4870 rvd = spa->spa_root_vdev;
4872 for (c = 0; c < rvd->vdev_children; c++) {
4873 vdev_t *vd = rvd->vdev_child[c];
4875 /* don't count the holes & logs as children */
4876 if (vd->vdev_islog || vd->vdev_ishole) {
4884 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
4885 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4887 /* next, ensure no spare or cache devices are part of the split */
4888 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
4889 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
4890 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4892 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_PUSHPAGE);
4893 glist = kmem_zalloc(children * sizeof (uint64_t), KM_PUSHPAGE);
4895 /* then, loop over each vdev and validate it */
4896 for (c = 0; c < children; c++) {
4897 uint64_t is_hole = 0;
4899 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
4903 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
4904 spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
4907 error = SET_ERROR(EINVAL);
4912 /* which disk is going to be split? */
4913 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
4915 error = SET_ERROR(EINVAL);
4919 /* look it up in the spa */
4920 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
4921 if (vml[c] == NULL) {
4922 error = SET_ERROR(ENODEV);
4926 /* make sure there's nothing stopping the split */
4927 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
4928 vml[c]->vdev_islog ||
4929 vml[c]->vdev_ishole ||
4930 vml[c]->vdev_isspare ||
4931 vml[c]->vdev_isl2cache ||
4932 !vdev_writeable(vml[c]) ||
4933 vml[c]->vdev_children != 0 ||
4934 vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
4935 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
4936 error = SET_ERROR(EINVAL);
4940 if (vdev_dtl_required(vml[c])) {
4941 error = SET_ERROR(EBUSY);
4945 /* we need certain info from the top level */
4946 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
4947 vml[c]->vdev_top->vdev_ms_array) == 0);
4948 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
4949 vml[c]->vdev_top->vdev_ms_shift) == 0);
4950 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
4951 vml[c]->vdev_top->vdev_asize) == 0);
4952 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
4953 vml[c]->vdev_top->vdev_ashift) == 0);
4957 kmem_free(vml, children * sizeof (vdev_t *));
4958 kmem_free(glist, children * sizeof (uint64_t));
4959 return (spa_vdev_exit(spa, NULL, txg, error));
4962 /* stop writers from using the disks */
4963 for (c = 0; c < children; c++) {
4965 vml[c]->vdev_offline = B_TRUE;
4967 vdev_reopen(spa->spa_root_vdev);
4970 * Temporarily record the splitting vdevs in the spa config. This
4971 * will disappear once the config is regenerated.
4973 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_PUSHPAGE) == 0);
4974 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
4975 glist, children) == 0);
4976 kmem_free(glist, children * sizeof (uint64_t));
4978 mutex_enter(&spa->spa_props_lock);
4979 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
4981 mutex_exit(&spa->spa_props_lock);
4982 spa->spa_config_splitting = nvl;
4983 vdev_config_dirty(spa->spa_root_vdev);
4985 /* configure and create the new pool */
4986 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
4987 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4988 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
4989 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
4990 spa_version(spa)) == 0);
4991 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
4992 spa->spa_config_txg) == 0);
4993 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4994 spa_generate_guid(NULL)) == 0);
4995 (void) nvlist_lookup_string(props,
4996 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4998 /* add the new pool to the namespace */
4999 newspa = spa_add(newname, config, altroot);
5000 newspa->spa_config_txg = spa->spa_config_txg;
5001 spa_set_log_state(newspa, SPA_LOG_CLEAR);
5003 /* release the spa config lock, retaining the namespace lock */
5004 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5006 if (zio_injection_enabled)
5007 zio_handle_panic_injection(spa, FTAG, 1);
5009 spa_activate(newspa, spa_mode_global);
5010 spa_async_suspend(newspa);
5012 /* create the new pool from the disks of the original pool */
5013 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
5017 /* if that worked, generate a real config for the new pool */
5018 if (newspa->spa_root_vdev != NULL) {
5019 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
5020 NV_UNIQUE_NAME, KM_PUSHPAGE) == 0);
5021 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
5022 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
5023 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
5028 if (props != NULL) {
5029 spa_configfile_set(newspa, props, B_FALSE);
5030 error = spa_prop_set(newspa, props);
5035 /* flush everything */
5036 txg = spa_vdev_config_enter(newspa);
5037 vdev_config_dirty(newspa->spa_root_vdev);
5038 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
5040 if (zio_injection_enabled)
5041 zio_handle_panic_injection(spa, FTAG, 2);
5043 spa_async_resume(newspa);
5045 /* finally, update the original pool's config */
5046 txg = spa_vdev_config_enter(spa);
5047 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
5048 error = dmu_tx_assign(tx, TXG_WAIT);
5051 for (c = 0; c < children; c++) {
5052 if (vml[c] != NULL) {
5055 spa_history_log_internal(spa, "detach", tx,
5056 "vdev=%s", vml[c]->vdev_path);
5060 vdev_config_dirty(spa->spa_root_vdev);
5061 spa->spa_config_splitting = NULL;
5065 (void) spa_vdev_exit(spa, NULL, txg, 0);
5067 if (zio_injection_enabled)
5068 zio_handle_panic_injection(spa, FTAG, 3);
5070 /* split is complete; log a history record */
5071 spa_history_log_internal(newspa, "split", NULL,
5072 "from pool %s", spa_name(spa));
5074 kmem_free(vml, children * sizeof (vdev_t *));
5076 /* if we're not going to mount the filesystems in userland, export */
5078 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
5085 spa_deactivate(newspa);
5088 txg = spa_vdev_config_enter(spa);
5090 /* re-online all offlined disks */
5091 for (c = 0; c < children; c++) {
5093 vml[c]->vdev_offline = B_FALSE;
5095 vdev_reopen(spa->spa_root_vdev);
5097 nvlist_free(spa->spa_config_splitting);
5098 spa->spa_config_splitting = NULL;
5099 (void) spa_vdev_exit(spa, NULL, txg, error);
5101 kmem_free(vml, children * sizeof (vdev_t *));
5106 spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
5110 for (i = 0; i < count; i++) {
5113 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
5116 if (guid == target_guid)
5124 spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
5125 nvlist_t *dev_to_remove)
5127 nvlist_t **newdev = NULL;
5131 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_PUSHPAGE);
5133 for (i = 0, j = 0; i < count; i++) {
5134 if (dev[i] == dev_to_remove)
5136 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_PUSHPAGE) == 0);
5139 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
5140 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
5142 for (i = 0; i < count - 1; i++)
5143 nvlist_free(newdev[i]);
5146 kmem_free(newdev, (count - 1) * sizeof (void *));
5150 * Evacuate the device.
5153 spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
5158 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5159 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5160 ASSERT(vd == vd->vdev_top);
5163 * Evacuate the device. We don't hold the config lock as writer
5164 * since we need to do I/O but we do keep the
5165 * spa_namespace_lock held. Once this completes the device
5166 * should no longer have any blocks allocated on it.
5168 if (vd->vdev_islog) {
5169 if (vd->vdev_stat.vs_alloc != 0)
5170 error = spa_offline_log(spa);
5172 error = SET_ERROR(ENOTSUP);
5179 * The evacuation succeeded. Remove any remaining MOS metadata
5180 * associated with this vdev, and wait for these changes to sync.
5182 ASSERT0(vd->vdev_stat.vs_alloc);
5183 txg = spa_vdev_config_enter(spa);
5184 vd->vdev_removing = B_TRUE;
5185 vdev_dirty(vd, 0, NULL, txg);
5186 vdev_config_dirty(vd);
5187 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5193 * Complete the removal by cleaning up the namespace.
5196 spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
5198 vdev_t *rvd = spa->spa_root_vdev;
5199 uint64_t id = vd->vdev_id;
5200 boolean_t last_vdev = (id == (rvd->vdev_children - 1));
5202 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5203 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5204 ASSERT(vd == vd->vdev_top);
5207 * Only remove any devices which are empty.
5209 if (vd->vdev_stat.vs_alloc != 0)
5212 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5214 if (list_link_active(&vd->vdev_state_dirty_node))
5215 vdev_state_clean(vd);
5216 if (list_link_active(&vd->vdev_config_dirty_node))
5217 vdev_config_clean(vd);
5222 vdev_compact_children(rvd);
5224 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
5225 vdev_add_child(rvd, vd);
5227 vdev_config_dirty(rvd);
5230 * Reassess the health of our root vdev.
5236 * Remove a device from the pool -
5238 * Removing a device from the vdev namespace requires several steps
5239 * and can take a significant amount of time. As a result we use
5240 * the spa_vdev_config_[enter/exit] functions which allow us to
5241 * grab and release the spa_config_lock while still holding the namespace
5242 * lock. During each step the configuration is synced out.
5246 * Remove a device from the pool. Currently, this supports removing only hot
5247 * spares, slogs, and level 2 ARC devices.
5250 spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
5253 metaslab_group_t *mg;
5254 nvlist_t **spares, **l2cache, *nv;
5256 uint_t nspares, nl2cache;
5258 boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
5260 ASSERT(spa_writeable(spa));
5263 txg = spa_vdev_enter(spa);
5265 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5267 if (spa->spa_spares.sav_vdevs != NULL &&
5268 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5269 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
5270 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
5272 * Only remove the hot spare if it's not currently in use
5275 if (vd == NULL || unspare) {
5276 spa_vdev_remove_aux(spa->spa_spares.sav_config,
5277 ZPOOL_CONFIG_SPARES, spares, nspares, nv);
5278 spa_load_spares(spa);
5279 spa->spa_spares.sav_sync = B_TRUE;
5281 error = SET_ERROR(EBUSY);
5283 } else if (spa->spa_l2cache.sav_vdevs != NULL &&
5284 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5285 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
5286 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
5288 * Cache devices can always be removed.
5290 spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
5291 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
5292 spa_load_l2cache(spa);
5293 spa->spa_l2cache.sav_sync = B_TRUE;
5294 } else if (vd != NULL && vd->vdev_islog) {
5296 ASSERT(vd == vd->vdev_top);
5299 * XXX - Once we have bp-rewrite this should
5300 * become the common case.
5306 * Stop allocating from this vdev.
5308 metaslab_group_passivate(mg);
5311 * Wait for the youngest allocations and frees to sync,
5312 * and then wait for the deferral of those frees to finish.
5314 spa_vdev_config_exit(spa, NULL,
5315 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
5318 * Attempt to evacuate the vdev.
5320 error = spa_vdev_remove_evacuate(spa, vd);
5322 txg = spa_vdev_config_enter(spa);
5325 * If we couldn't evacuate the vdev, unwind.
5328 metaslab_group_activate(mg);
5329 return (spa_vdev_exit(spa, NULL, txg, error));
5333 * Clean up the vdev namespace.
5335 spa_vdev_remove_from_namespace(spa, vd);
5337 } else if (vd != NULL) {
5339 * Normal vdevs cannot be removed (yet).
5341 error = SET_ERROR(ENOTSUP);
5344 * There is no vdev of any kind with the specified guid.
5346 error = SET_ERROR(ENOENT);
5350 return (spa_vdev_exit(spa, NULL, txg, error));
5356 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5357 * current spared, so we can detach it.
5360 spa_vdev_resilver_done_hunt(vdev_t *vd)
5362 vdev_t *newvd, *oldvd;
5365 for (c = 0; c < vd->vdev_children; c++) {
5366 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
5372 * Check for a completed replacement. We always consider the first
5373 * vdev in the list to be the oldest vdev, and the last one to be
5374 * the newest (see spa_vdev_attach() for how that works). In
5375 * the case where the newest vdev is faulted, we will not automatically
5376 * remove it after a resilver completes. This is OK as it will require
5377 * user intervention to determine which disk the admin wishes to keep.
5379 if (vd->vdev_ops == &vdev_replacing_ops) {
5380 ASSERT(vd->vdev_children > 1);
5382 newvd = vd->vdev_child[vd->vdev_children - 1];
5383 oldvd = vd->vdev_child[0];
5385 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
5386 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5387 !vdev_dtl_required(oldvd))
5392 * Check for a completed resilver with the 'unspare' flag set.
5394 if (vd->vdev_ops == &vdev_spare_ops) {
5395 vdev_t *first = vd->vdev_child[0];
5396 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
5398 if (last->vdev_unspare) {
5401 } else if (first->vdev_unspare) {
5408 if (oldvd != NULL &&
5409 vdev_dtl_empty(newvd, DTL_MISSING) &&
5410 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5411 !vdev_dtl_required(oldvd))
5415 * If there are more than two spares attached to a disk,
5416 * and those spares are not required, then we want to
5417 * attempt to free them up now so that they can be used
5418 * by other pools. Once we're back down to a single
5419 * disk+spare, we stop removing them.
5421 if (vd->vdev_children > 2) {
5422 newvd = vd->vdev_child[1];
5424 if (newvd->vdev_isspare && last->vdev_isspare &&
5425 vdev_dtl_empty(last, DTL_MISSING) &&
5426 vdev_dtl_empty(last, DTL_OUTAGE) &&
5427 !vdev_dtl_required(newvd))
5436 spa_vdev_resilver_done(spa_t *spa)
5438 vdev_t *vd, *pvd, *ppvd;
5439 uint64_t guid, sguid, pguid, ppguid;
5441 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5443 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
5444 pvd = vd->vdev_parent;
5445 ppvd = pvd->vdev_parent;
5446 guid = vd->vdev_guid;
5447 pguid = pvd->vdev_guid;
5448 ppguid = ppvd->vdev_guid;
5451 * If we have just finished replacing a hot spared device, then
5452 * we need to detach the parent's first child (the original hot
5455 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
5456 ppvd->vdev_children == 2) {
5457 ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
5458 sguid = ppvd->vdev_child[1]->vdev_guid;
5460 spa_config_exit(spa, SCL_ALL, FTAG);
5461 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
5463 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
5465 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5468 spa_config_exit(spa, SCL_ALL, FTAG);
5472 * Update the stored path or FRU for this vdev.
5475 spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
5479 boolean_t sync = B_FALSE;
5481 ASSERT(spa_writeable(spa));
5483 spa_vdev_state_enter(spa, SCL_ALL);
5485 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
5486 return (spa_vdev_state_exit(spa, NULL, ENOENT));
5488 if (!vd->vdev_ops->vdev_op_leaf)
5489 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
5492 if (strcmp(value, vd->vdev_path) != 0) {
5493 spa_strfree(vd->vdev_path);
5494 vd->vdev_path = spa_strdup(value);
5498 if (vd->vdev_fru == NULL) {
5499 vd->vdev_fru = spa_strdup(value);
5501 } else if (strcmp(value, vd->vdev_fru) != 0) {
5502 spa_strfree(vd->vdev_fru);
5503 vd->vdev_fru = spa_strdup(value);
5508 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
5512 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
5514 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5518 spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5520 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5524 * ==========================================================================
5526 * ==========================================================================
5530 spa_scan_stop(spa_t *spa)
5532 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5533 if (dsl_scan_resilvering(spa->spa_dsl_pool))
5534 return (SET_ERROR(EBUSY));
5535 return (dsl_scan_cancel(spa->spa_dsl_pool));
5539 spa_scan(spa_t *spa, pool_scan_func_t func)
5541 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5543 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5544 return (SET_ERROR(ENOTSUP));
5547 * If a resilver was requested, but there is no DTL on a
5548 * writeable leaf device, we have nothing to do.
5550 if (func == POOL_SCAN_RESILVER &&
5551 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5552 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5556 return (dsl_scan(spa->spa_dsl_pool, func));
5560 * ==========================================================================
5561 * SPA async task processing
5562 * ==========================================================================
5566 spa_async_remove(spa_t *spa, vdev_t *vd)
5570 if (vd->vdev_remove_wanted) {
5571 vd->vdev_remove_wanted = B_FALSE;
5572 vd->vdev_delayed_close = B_FALSE;
5573 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5576 * We want to clear the stats, but we don't want to do a full
5577 * vdev_clear() as that will cause us to throw away
5578 * degraded/faulted state as well as attempt to reopen the
5579 * device, all of which is a waste.
5581 vd->vdev_stat.vs_read_errors = 0;
5582 vd->vdev_stat.vs_write_errors = 0;
5583 vd->vdev_stat.vs_checksum_errors = 0;
5585 vdev_state_dirty(vd->vdev_top);
5588 for (c = 0; c < vd->vdev_children; c++)
5589 spa_async_remove(spa, vd->vdev_child[c]);
5593 spa_async_probe(spa_t *spa, vdev_t *vd)
5597 if (vd->vdev_probe_wanted) {
5598 vd->vdev_probe_wanted = B_FALSE;
5599 vdev_reopen(vd); /* vdev_open() does the actual probe */
5602 for (c = 0; c < vd->vdev_children; c++)
5603 spa_async_probe(spa, vd->vdev_child[c]);
5607 spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5611 if (!spa->spa_autoexpand)
5614 for (c = 0; c < vd->vdev_children; c++) {
5615 vdev_t *cvd = vd->vdev_child[c];
5616 spa_async_autoexpand(spa, cvd);
5619 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5622 spa_event_notify(vd->vdev_spa, vd, FM_EREPORT_ZFS_DEVICE_AUTOEXPAND);
5626 spa_async_thread(spa_t *spa)
5630 ASSERT(spa->spa_sync_on);
5632 mutex_enter(&spa->spa_async_lock);
5633 tasks = spa->spa_async_tasks;
5634 spa->spa_async_tasks = 0;
5635 mutex_exit(&spa->spa_async_lock);
5638 * See if the config needs to be updated.
5640 if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5641 uint64_t old_space, new_space;
5643 mutex_enter(&spa_namespace_lock);
5644 old_space = metaslab_class_get_space(spa_normal_class(spa));
5645 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5646 new_space = metaslab_class_get_space(spa_normal_class(spa));
5647 mutex_exit(&spa_namespace_lock);
5650 * If the pool grew as a result of the config update,
5651 * then log an internal history event.
5653 if (new_space != old_space) {
5654 spa_history_log_internal(spa, "vdev online", NULL,
5655 "pool '%s' size: %llu(+%llu)",
5656 spa_name(spa), new_space, new_space - old_space);
5661 * See if any devices need to be marked REMOVED.
5663 if (tasks & SPA_ASYNC_REMOVE) {
5664 spa_vdev_state_enter(spa, SCL_NONE);
5665 spa_async_remove(spa, spa->spa_root_vdev);
5666 for (i = 0; i < spa->spa_l2cache.sav_count; i++)
5667 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5668 for (i = 0; i < spa->spa_spares.sav_count; i++)
5669 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5670 (void) spa_vdev_state_exit(spa, NULL, 0);
5673 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5674 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5675 spa_async_autoexpand(spa, spa->spa_root_vdev);
5676 spa_config_exit(spa, SCL_CONFIG, FTAG);
5680 * See if any devices need to be probed.
5682 if (tasks & SPA_ASYNC_PROBE) {
5683 spa_vdev_state_enter(spa, SCL_NONE);
5684 spa_async_probe(spa, spa->spa_root_vdev);
5685 (void) spa_vdev_state_exit(spa, NULL, 0);
5689 * If any devices are done replacing, detach them.
5691 if (tasks & SPA_ASYNC_RESILVER_DONE)
5692 spa_vdev_resilver_done(spa);
5695 * Kick off a resilver.
5697 if (tasks & SPA_ASYNC_RESILVER)
5698 dsl_resilver_restart(spa->spa_dsl_pool, 0);
5701 * Let the world know that we're done.
5703 mutex_enter(&spa->spa_async_lock);
5704 spa->spa_async_thread = NULL;
5705 cv_broadcast(&spa->spa_async_cv);
5706 mutex_exit(&spa->spa_async_lock);
5711 spa_async_suspend(spa_t *spa)
5713 mutex_enter(&spa->spa_async_lock);
5714 spa->spa_async_suspended++;
5715 while (spa->spa_async_thread != NULL)
5716 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5717 mutex_exit(&spa->spa_async_lock);
5721 spa_async_resume(spa_t *spa)
5723 mutex_enter(&spa->spa_async_lock);
5724 ASSERT(spa->spa_async_suspended != 0);
5725 spa->spa_async_suspended--;
5726 mutex_exit(&spa->spa_async_lock);
5730 spa_async_dispatch(spa_t *spa)
5732 mutex_enter(&spa->spa_async_lock);
5733 if (spa->spa_async_tasks && !spa->spa_async_suspended &&
5734 spa->spa_async_thread == NULL &&
5735 rootdir != NULL && !vn_is_readonly(rootdir))
5736 spa->spa_async_thread = thread_create(NULL, 0,
5737 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
5738 mutex_exit(&spa->spa_async_lock);
5742 spa_async_request(spa_t *spa, int task)
5744 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
5745 mutex_enter(&spa->spa_async_lock);
5746 spa->spa_async_tasks |= task;
5747 mutex_exit(&spa->spa_async_lock);
5751 * ==========================================================================
5752 * SPA syncing routines
5753 * ==========================================================================
5757 bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5760 bpobj_enqueue(bpo, bp, tx);
5765 spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5769 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
5775 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
5777 char *packed = NULL;
5782 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
5785 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5786 * information. This avoids the dbuf_will_dirty() path and
5787 * saves us a pre-read to get data we don't actually care about.
5789 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
5790 packed = vmem_alloc(bufsize, KM_PUSHPAGE);
5792 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
5794 bzero(packed + nvsize, bufsize - nvsize);
5796 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
5798 vmem_free(packed, bufsize);
5800 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
5801 dmu_buf_will_dirty(db, tx);
5802 *(uint64_t *)db->db_data = nvsize;
5803 dmu_buf_rele(db, FTAG);
5807 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
5808 const char *config, const char *entry)
5818 * Update the MOS nvlist describing the list of available devices.
5819 * spa_validate_aux() will have already made sure this nvlist is
5820 * valid and the vdevs are labeled appropriately.
5822 if (sav->sav_object == 0) {
5823 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
5824 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
5825 sizeof (uint64_t), tx);
5826 VERIFY(zap_update(spa->spa_meta_objset,
5827 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
5828 &sav->sav_object, tx) == 0);
5831 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_PUSHPAGE) == 0);
5832 if (sav->sav_count == 0) {
5833 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
5835 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_PUSHPAGE);
5836 for (i = 0; i < sav->sav_count; i++)
5837 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
5838 B_FALSE, VDEV_CONFIG_L2CACHE);
5839 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
5840 sav->sav_count) == 0);
5841 for (i = 0; i < sav->sav_count; i++)
5842 nvlist_free(list[i]);
5843 kmem_free(list, sav->sav_count * sizeof (void *));
5846 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
5847 nvlist_free(nvroot);
5849 sav->sav_sync = B_FALSE;
5853 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
5857 if (list_is_empty(&spa->spa_config_dirty_list))
5860 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5862 config = spa_config_generate(spa, spa->spa_root_vdev,
5863 dmu_tx_get_txg(tx), B_FALSE);
5866 * If we're upgrading the spa version then make sure that
5867 * the config object gets updated with the correct version.
5869 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
5870 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5871 spa->spa_uberblock.ub_version);
5873 spa_config_exit(spa, SCL_STATE, FTAG);
5875 if (spa->spa_config_syncing)
5876 nvlist_free(spa->spa_config_syncing);
5877 spa->spa_config_syncing = config;
5879 spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
5883 spa_sync_version(void *arg, dmu_tx_t *tx)
5885 uint64_t *versionp = arg;
5886 uint64_t version = *versionp;
5887 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
5890 * Setting the version is special cased when first creating the pool.
5892 ASSERT(tx->tx_txg != TXG_INITIAL);
5894 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
5895 ASSERT(version >= spa_version(spa));
5897 spa->spa_uberblock.ub_version = version;
5898 vdev_config_dirty(spa->spa_root_vdev);
5899 spa_history_log_internal(spa, "set", tx, "version=%lld", version);
5903 * Set zpool properties.
5906 spa_sync_props(void *arg, dmu_tx_t *tx)
5908 nvlist_t *nvp = arg;
5909 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
5910 objset_t *mos = spa->spa_meta_objset;
5911 nvpair_t *elem = NULL;
5913 mutex_enter(&spa->spa_props_lock);
5915 while ((elem = nvlist_next_nvpair(nvp, elem))) {
5917 char *strval, *fname;
5919 const char *propname;
5920 zprop_type_t proptype;
5921 zfeature_info_t *feature;
5923 prop = zpool_name_to_prop(nvpair_name(elem));
5924 switch ((int)prop) {
5927 * We checked this earlier in spa_prop_validate().
5929 ASSERT(zpool_prop_feature(nvpair_name(elem)));
5931 fname = strchr(nvpair_name(elem), '@') + 1;
5932 VERIFY3U(0, ==, zfeature_lookup_name(fname, &feature));
5934 spa_feature_enable(spa, feature, tx);
5935 spa_history_log_internal(spa, "set", tx,
5936 "%s=enabled", nvpair_name(elem));
5939 case ZPOOL_PROP_VERSION:
5940 VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5942 * The version is synced seperatly before other
5943 * properties and should be correct by now.
5945 ASSERT3U(spa_version(spa), >=, intval);
5948 case ZPOOL_PROP_ALTROOT:
5950 * 'altroot' is a non-persistent property. It should
5951 * have been set temporarily at creation or import time.
5953 ASSERT(spa->spa_root != NULL);
5956 case ZPOOL_PROP_READONLY:
5957 case ZPOOL_PROP_CACHEFILE:
5959 * 'readonly' and 'cachefile' are also non-persisitent
5963 case ZPOOL_PROP_COMMENT:
5964 VERIFY(nvpair_value_string(elem, &strval) == 0);
5965 if (spa->spa_comment != NULL)
5966 spa_strfree(spa->spa_comment);
5967 spa->spa_comment = spa_strdup(strval);
5969 * We need to dirty the configuration on all the vdevs
5970 * so that their labels get updated. It's unnecessary
5971 * to do this for pool creation since the vdev's
5972 * configuratoin has already been dirtied.
5974 if (tx->tx_txg != TXG_INITIAL)
5975 vdev_config_dirty(spa->spa_root_vdev);
5976 spa_history_log_internal(spa, "set", tx,
5977 "%s=%s", nvpair_name(elem), strval);
5981 * Set pool property values in the poolprops mos object.
5983 if (spa->spa_pool_props_object == 0) {
5984 spa->spa_pool_props_object =
5985 zap_create_link(mos, DMU_OT_POOL_PROPS,
5986 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
5990 /* normalize the property name */
5991 propname = zpool_prop_to_name(prop);
5992 proptype = zpool_prop_get_type(prop);
5994 if (nvpair_type(elem) == DATA_TYPE_STRING) {
5995 ASSERT(proptype == PROP_TYPE_STRING);
5996 VERIFY(nvpair_value_string(elem, &strval) == 0);
5997 VERIFY(zap_update(mos,
5998 spa->spa_pool_props_object, propname,
5999 1, strlen(strval) + 1, strval, tx) == 0);
6000 spa_history_log_internal(spa, "set", tx,
6001 "%s=%s", nvpair_name(elem), strval);
6002 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
6003 VERIFY(nvpair_value_uint64(elem, &intval) == 0);
6005 if (proptype == PROP_TYPE_INDEX) {
6007 VERIFY(zpool_prop_index_to_string(
6008 prop, intval, &unused) == 0);
6010 VERIFY(zap_update(mos,
6011 spa->spa_pool_props_object, propname,
6012 8, 1, &intval, tx) == 0);
6013 spa_history_log_internal(spa, "set", tx,
6014 "%s=%lld", nvpair_name(elem), intval);
6016 ASSERT(0); /* not allowed */
6020 case ZPOOL_PROP_DELEGATION:
6021 spa->spa_delegation = intval;
6023 case ZPOOL_PROP_BOOTFS:
6024 spa->spa_bootfs = intval;
6026 case ZPOOL_PROP_FAILUREMODE:
6027 spa->spa_failmode = intval;
6029 case ZPOOL_PROP_AUTOEXPAND:
6030 spa->spa_autoexpand = intval;
6031 if (tx->tx_txg != TXG_INITIAL)
6032 spa_async_request(spa,
6033 SPA_ASYNC_AUTOEXPAND);
6035 case ZPOOL_PROP_DEDUPDITTO:
6036 spa->spa_dedup_ditto = intval;
6045 mutex_exit(&spa->spa_props_lock);
6049 * Perform one-time upgrade on-disk changes. spa_version() does not
6050 * reflect the new version this txg, so there must be no changes this
6051 * txg to anything that the upgrade code depends on after it executes.
6052 * Therefore this must be called after dsl_pool_sync() does the sync
6056 spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
6058 dsl_pool_t *dp = spa->spa_dsl_pool;
6060 ASSERT(spa->spa_sync_pass == 1);
6062 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
6064 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
6065 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
6066 dsl_pool_create_origin(dp, tx);
6068 /* Keeping the origin open increases spa_minref */
6069 spa->spa_minref += 3;
6072 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
6073 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
6074 dsl_pool_upgrade_clones(dp, tx);
6077 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
6078 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
6079 dsl_pool_upgrade_dir_clones(dp, tx);
6081 /* Keeping the freedir open increases spa_minref */
6082 spa->spa_minref += 3;
6085 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES &&
6086 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
6087 spa_feature_create_zap_objects(spa, tx);
6089 rrw_exit(&dp->dp_config_rwlock, FTAG);
6093 * Sync the specified transaction group. New blocks may be dirtied as
6094 * part of the process, so we iterate until it converges.
6097 spa_sync(spa_t *spa, uint64_t txg)
6099 dsl_pool_t *dp = spa->spa_dsl_pool;
6100 objset_t *mos = spa->spa_meta_objset;
6101 bpobj_t *defer_bpo = &spa->spa_deferred_bpobj;
6102 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
6103 vdev_t *rvd = spa->spa_root_vdev;
6109 VERIFY(spa_writeable(spa));
6112 * Lock out configuration changes.
6114 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
6116 spa->spa_syncing_txg = txg;
6117 spa->spa_sync_pass = 0;
6120 * If there are any pending vdev state changes, convert them
6121 * into config changes that go out with this transaction group.
6123 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6124 while (list_head(&spa->spa_state_dirty_list) != NULL) {
6126 * We need the write lock here because, for aux vdevs,
6127 * calling vdev_config_dirty() modifies sav_config.
6128 * This is ugly and will become unnecessary when we
6129 * eliminate the aux vdev wart by integrating all vdevs
6130 * into the root vdev tree.
6132 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6133 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
6134 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
6135 vdev_state_clean(vd);
6136 vdev_config_dirty(vd);
6138 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6139 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
6141 spa_config_exit(spa, SCL_STATE, FTAG);
6143 tx = dmu_tx_create_assigned(dp, txg);
6145 spa->spa_sync_starttime = gethrtime();
6146 taskq_cancel_id(system_taskq, spa->spa_deadman_tqid);
6147 spa->spa_deadman_tqid = taskq_dispatch_delay(system_taskq,
6148 spa_deadman, spa, TQ_PUSHPAGE, ddi_get_lbolt() +
6149 NSEC_TO_TICK(spa->spa_deadman_synctime));
6152 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
6153 * set spa_deflate if we have no raid-z vdevs.
6155 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
6156 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
6159 for (i = 0; i < rvd->vdev_children; i++) {
6160 vd = rvd->vdev_child[i];
6161 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
6164 if (i == rvd->vdev_children) {
6165 spa->spa_deflate = TRUE;
6166 VERIFY(0 == zap_add(spa->spa_meta_objset,
6167 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
6168 sizeof (uint64_t), 1, &spa->spa_deflate, tx));
6173 * If anything has changed in this txg, or if someone is waiting
6174 * for this txg to sync (eg, spa_vdev_remove()), push the
6175 * deferred frees from the previous txg. If not, leave them
6176 * alone so that we don't generate work on an otherwise idle
6179 if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
6180 !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
6181 !txg_list_empty(&dp->dp_sync_tasks, txg) ||
6182 ((dsl_scan_active(dp->dp_scan) ||
6183 txg_sync_waiting(dp)) && !spa_shutting_down(spa))) {
6184 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6185 VERIFY3U(bpobj_iterate(defer_bpo,
6186 spa_free_sync_cb, zio, tx), ==, 0);
6187 VERIFY0(zio_wait(zio));
6191 * Iterate to convergence.
6194 int pass = ++spa->spa_sync_pass;
6196 spa_sync_config_object(spa, tx);
6197 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
6198 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
6199 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
6200 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
6201 spa_errlog_sync(spa, txg);
6202 dsl_pool_sync(dp, txg);
6204 if (pass < zfs_sync_pass_deferred_free) {
6205 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6206 bplist_iterate(free_bpl, spa_free_sync_cb,
6208 VERIFY(zio_wait(zio) == 0);
6210 bplist_iterate(free_bpl, bpobj_enqueue_cb,
6215 dsl_scan_sync(dp, tx);
6217 while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)))
6221 spa_sync_upgrades(spa, tx);
6223 } while (dmu_objset_is_dirty(mos, txg));
6226 * Rewrite the vdev configuration (which includes the uberblock)
6227 * to commit the transaction group.
6229 * If there are no dirty vdevs, we sync the uberblock to a few
6230 * random top-level vdevs that are known to be visible in the
6231 * config cache (see spa_vdev_add() for a complete description).
6232 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
6236 * We hold SCL_STATE to prevent vdev open/close/etc.
6237 * while we're attempting to write the vdev labels.
6239 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6241 if (list_is_empty(&spa->spa_config_dirty_list)) {
6242 vdev_t *svd[SPA_DVAS_PER_BP];
6244 int children = rvd->vdev_children;
6245 int c0 = spa_get_random(children);
6247 for (c = 0; c < children; c++) {
6248 vd = rvd->vdev_child[(c0 + c) % children];
6249 if (vd->vdev_ms_array == 0 || vd->vdev_islog)
6251 svd[svdcount++] = vd;
6252 if (svdcount == SPA_DVAS_PER_BP)
6255 error = vdev_config_sync(svd, svdcount, txg, B_FALSE);
6257 error = vdev_config_sync(svd, svdcount, txg,
6260 error = vdev_config_sync(rvd->vdev_child,
6261 rvd->vdev_children, txg, B_FALSE);
6263 error = vdev_config_sync(rvd->vdev_child,
6264 rvd->vdev_children, txg, B_TRUE);
6268 spa->spa_last_synced_guid = rvd->vdev_guid;
6270 spa_config_exit(spa, SCL_STATE, FTAG);
6274 zio_suspend(spa, NULL);
6275 zio_resume_wait(spa);
6279 taskq_cancel_id(system_taskq, spa->spa_deadman_tqid);
6280 spa->spa_deadman_tqid = 0;
6283 * Clear the dirty config list.
6285 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
6286 vdev_config_clean(vd);
6289 * Now that the new config has synced transactionally,
6290 * let it become visible to the config cache.
6292 if (spa->spa_config_syncing != NULL) {
6293 spa_config_set(spa, spa->spa_config_syncing);
6294 spa->spa_config_txg = txg;
6295 spa->spa_config_syncing = NULL;
6298 spa->spa_ubsync = spa->spa_uberblock;
6300 dsl_pool_sync_done(dp, txg);
6303 * Update usable space statistics.
6305 while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg))))
6306 vdev_sync_done(vd, txg);
6308 spa_update_dspace(spa);
6311 * It had better be the case that we didn't dirty anything
6312 * since vdev_config_sync().
6314 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
6315 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
6316 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
6318 spa->spa_sync_pass = 0;
6320 spa_config_exit(spa, SCL_CONFIG, FTAG);
6322 spa_handle_ignored_writes(spa);
6325 * If any async tasks have been requested, kick them off.
6327 spa_async_dispatch(spa);
6331 * Sync all pools. We don't want to hold the namespace lock across these
6332 * operations, so we take a reference on the spa_t and drop the lock during the
6336 spa_sync_allpools(void)
6339 mutex_enter(&spa_namespace_lock);
6340 while ((spa = spa_next(spa)) != NULL) {
6341 if (spa_state(spa) != POOL_STATE_ACTIVE ||
6342 !spa_writeable(spa) || spa_suspended(spa))
6344 spa_open_ref(spa, FTAG);
6345 mutex_exit(&spa_namespace_lock);
6346 txg_wait_synced(spa_get_dsl(spa), 0);
6347 mutex_enter(&spa_namespace_lock);
6348 spa_close(spa, FTAG);
6350 mutex_exit(&spa_namespace_lock);
6354 * ==========================================================================
6355 * Miscellaneous routines
6356 * ==========================================================================
6360 * Remove all pools in the system.
6368 * Remove all cached state. All pools should be closed now,
6369 * so every spa in the AVL tree should be unreferenced.
6371 mutex_enter(&spa_namespace_lock);
6372 while ((spa = spa_next(NULL)) != NULL) {
6374 * Stop async tasks. The async thread may need to detach
6375 * a device that's been replaced, which requires grabbing
6376 * spa_namespace_lock, so we must drop it here.
6378 spa_open_ref(spa, FTAG);
6379 mutex_exit(&spa_namespace_lock);
6380 spa_async_suspend(spa);
6381 mutex_enter(&spa_namespace_lock);
6382 spa_close(spa, FTAG);
6384 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
6386 spa_deactivate(spa);
6390 mutex_exit(&spa_namespace_lock);
6394 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
6399 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
6403 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
6404 vd = spa->spa_l2cache.sav_vdevs[i];
6405 if (vd->vdev_guid == guid)
6409 for (i = 0; i < spa->spa_spares.sav_count; i++) {
6410 vd = spa->spa_spares.sav_vdevs[i];
6411 if (vd->vdev_guid == guid)
6420 spa_upgrade(spa_t *spa, uint64_t version)
6422 ASSERT(spa_writeable(spa));
6424 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6427 * This should only be called for a non-faulted pool, and since a
6428 * future version would result in an unopenable pool, this shouldn't be
6431 ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version));
6432 ASSERT(version >= spa->spa_uberblock.ub_version);
6434 spa->spa_uberblock.ub_version = version;
6435 vdev_config_dirty(spa->spa_root_vdev);
6437 spa_config_exit(spa, SCL_ALL, FTAG);
6439 txg_wait_synced(spa_get_dsl(spa), 0);
6443 spa_has_spare(spa_t *spa, uint64_t guid)
6447 spa_aux_vdev_t *sav = &spa->spa_spares;
6449 for (i = 0; i < sav->sav_count; i++)
6450 if (sav->sav_vdevs[i]->vdev_guid == guid)
6453 for (i = 0; i < sav->sav_npending; i++) {
6454 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
6455 &spareguid) == 0 && spareguid == guid)
6463 * Check if a pool has an active shared spare device.
6464 * Note: reference count of an active spare is 2, as a spare and as a replace
6467 spa_has_active_shared_spare(spa_t *spa)
6471 spa_aux_vdev_t *sav = &spa->spa_spares;
6473 for (i = 0; i < sav->sav_count; i++) {
6474 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
6475 &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
6484 * Post a FM_EREPORT_ZFS_* event from sys/fm/fs/zfs.h. The payload will be
6485 * filled in from the spa and (optionally) the vdev. This doesn't do anything
6486 * in the userland libzpool, as we don't want consumers to misinterpret ztest
6487 * or zdb as real changes.
6490 spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
6493 zfs_ereport_post(name, spa, vd, NULL, 0, 0);
6497 #if defined(_KERNEL) && defined(HAVE_SPL)
6498 /* state manipulation functions */
6499 EXPORT_SYMBOL(spa_open);
6500 EXPORT_SYMBOL(spa_open_rewind);
6501 EXPORT_SYMBOL(spa_get_stats);
6502 EXPORT_SYMBOL(spa_create);
6503 EXPORT_SYMBOL(spa_import_rootpool);
6504 EXPORT_SYMBOL(spa_import);
6505 EXPORT_SYMBOL(spa_tryimport);
6506 EXPORT_SYMBOL(spa_destroy);
6507 EXPORT_SYMBOL(spa_export);
6508 EXPORT_SYMBOL(spa_reset);
6509 EXPORT_SYMBOL(spa_async_request);
6510 EXPORT_SYMBOL(spa_async_suspend);
6511 EXPORT_SYMBOL(spa_async_resume);
6512 EXPORT_SYMBOL(spa_inject_addref);
6513 EXPORT_SYMBOL(spa_inject_delref);
6514 EXPORT_SYMBOL(spa_scan_stat_init);
6515 EXPORT_SYMBOL(spa_scan_get_stats);
6517 /* device maniion */
6518 EXPORT_SYMBOL(spa_vdev_add);
6519 EXPORT_SYMBOL(spa_vdev_attach);
6520 EXPORT_SYMBOL(spa_vdev_detach);
6521 EXPORT_SYMBOL(spa_vdev_remove);
6522 EXPORT_SYMBOL(spa_vdev_setpath);
6523 EXPORT_SYMBOL(spa_vdev_setfru);
6524 EXPORT_SYMBOL(spa_vdev_split_mirror);
6526 /* spare statech is global across all pools) */
6527 EXPORT_SYMBOL(spa_spare_add);
6528 EXPORT_SYMBOL(spa_spare_remove);
6529 EXPORT_SYMBOL(spa_spare_exists);
6530 EXPORT_SYMBOL(spa_spare_activate);
6532 /* L2ARC statech is global across all pools) */
6533 EXPORT_SYMBOL(spa_l2cache_add);
6534 EXPORT_SYMBOL(spa_l2cache_remove);
6535 EXPORT_SYMBOL(spa_l2cache_exists);
6536 EXPORT_SYMBOL(spa_l2cache_activate);
6537 EXPORT_SYMBOL(spa_l2cache_drop);
6540 EXPORT_SYMBOL(spa_scan);
6541 EXPORT_SYMBOL(spa_scan_stop);
6544 EXPORT_SYMBOL(spa_sync); /* only for DMU use */
6545 EXPORT_SYMBOL(spa_sync_allpools);
6548 EXPORT_SYMBOL(spa_prop_set);
6549 EXPORT_SYMBOL(spa_prop_get);
6550 EXPORT_SYMBOL(spa_prop_clear_bootfs);
6552 /* asynchronous event notification */
6553 EXPORT_SYMBOL(spa_event_notify);