4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
25 * Copyright (c) 2012 by Delphix. All rights reserved.
29 * This file contains all the routines used when modifying on-disk SPA state.
30 * This includes opening, importing, destroying, exporting a pool, and syncing a
34 #include <sys/zfs_context.h>
35 #include <sys/fm/fs/zfs.h>
36 #include <sys/spa_impl.h>
38 #include <sys/zio_checksum.h>
40 #include <sys/dmu_tx.h>
44 #include <sys/vdev_impl.h>
45 #include <sys/vdev_disk.h>
46 #include <sys/metaslab.h>
47 #include <sys/metaslab_impl.h>
48 #include <sys/uberblock_impl.h>
51 #include <sys/dmu_traverse.h>
52 #include <sys/dmu_objset.h>
53 #include <sys/unique.h>
54 #include <sys/dsl_pool.h>
55 #include <sys/dsl_dataset.h>
56 #include <sys/dsl_dir.h>
57 #include <sys/dsl_prop.h>
58 #include <sys/dsl_synctask.h>
59 #include <sys/fs/zfs.h>
61 #include <sys/callb.h>
62 #include <sys/systeminfo.h>
63 #include <sys/spa_boot.h>
64 #include <sys/zfs_ioctl.h>
65 #include <sys/dsl_scan.h>
66 #include <sys/zfeature.h>
70 #include <sys/bootprops.h>
71 #include <sys/callb.h>
72 #include <sys/cpupart.h>
74 #include <sys/sysdc.h>
79 #include "zfs_comutil.h"
81 typedef enum zti_modes {
82 ZTI_MODE_FIXED, /* value is # of threads (min 1) */
83 ZTI_MODE_ONLINE_PERCENT, /* value is % of online CPUs */
84 ZTI_MODE_BATCH, /* cpu-intensive; value is ignored */
85 ZTI_MODE_NULL, /* don't create a taskq */
89 #define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) }
90 #define ZTI_PCT(n) { ZTI_MODE_ONLINE_PERCENT, (n), 1 }
91 #define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 }
92 #define ZTI_NULL { ZTI_MODE_NULL, 0, 0 }
94 #define ZTI_N(n) ZTI_P(n, 1)
95 #define ZTI_ONE ZTI_N(1)
97 typedef struct zio_taskq_info {
103 static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
104 "iss", "iss_h", "int", "int_h"
108 * This table defines the taskq settings for each ZFS I/O type. When
109 * initializing a pool, we use this table to create an appropriately sized
110 * taskq. Some operations are low volume and therefore have a small, static
111 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE
112 * macros. Other operations process a large amount of data; the ZTI_BATCH
113 * macro causes us to create a taskq oriented for throughput. Some operations
114 * are so high frequency and short-lived that the taskq itself can become a a
115 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an
116 * additional degree of parallelism specified by the number of threads per-
117 * taskq and the number of taskqs; when dispatching an event in this case, the
118 * particular taskq is chosen at random.
120 * The different taskq priorities are to handle the different contexts (issue
121 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that
122 * need to be handled with minimum delay.
124 const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
125 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
126 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* NULL */
127 { ZTI_N(8), ZTI_NULL, ZTI_BATCH, ZTI_NULL }, /* READ */
128 { ZTI_BATCH, ZTI_N(5), ZTI_N(16), ZTI_N(5) }, /* WRITE */
129 { ZTI_P(4, 8), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FREE */
130 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* CLAIM */
131 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* IOCTL */
134 static dsl_syncfunc_t spa_sync_version;
135 static dsl_syncfunc_t spa_sync_props;
136 static dsl_checkfunc_t spa_change_guid_check;
137 static dsl_syncfunc_t spa_change_guid_sync;
138 static boolean_t spa_has_active_shared_spare(spa_t *spa);
139 static inline int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
140 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
142 static void spa_vdev_resilver_done(spa_t *spa);
144 uint_t zio_taskq_batch_pct = 100; /* 1 thread per cpu in pset */
145 id_t zio_taskq_psrset_bind = PS_NONE;
146 boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */
147 uint_t zio_taskq_basedc = 80; /* base duty cycle */
149 boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */
152 * This (illegal) pool name is used when temporarily importing a spa_t in order
153 * to get the vdev stats associated with the imported devices.
155 #define TRYIMPORT_NAME "$import"
158 * ==========================================================================
159 * SPA properties routines
160 * ==========================================================================
164 * Add a (source=src, propname=propval) list to an nvlist.
167 spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
168 uint64_t intval, zprop_source_t src)
170 const char *propname = zpool_prop_to_name(prop);
173 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_PUSHPAGE) == 0);
174 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
177 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
179 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
181 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
182 nvlist_free(propval);
186 * Get property values from the spa configuration.
189 spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
191 vdev_t *rvd = spa->spa_root_vdev;
192 dsl_pool_t *pool = spa->spa_dsl_pool;
196 uint64_t cap, version;
197 zprop_source_t src = ZPROP_SRC_NONE;
198 spa_config_dirent_t *dp;
201 ASSERT(MUTEX_HELD(&spa->spa_props_lock));
204 alloc = metaslab_class_get_alloc(spa_normal_class(spa));
205 size = metaslab_class_get_space(spa_normal_class(spa));
206 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
207 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
208 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
209 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
213 for (c = 0; c < rvd->vdev_children; c++) {
214 vdev_t *tvd = rvd->vdev_child[c];
215 space += tvd->vdev_max_asize - tvd->vdev_asize;
217 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, space,
220 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
221 (spa_mode(spa) == FREAD), src);
223 cap = (size == 0) ? 0 : (alloc * 100 / size);
224 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
226 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
227 ddt_get_pool_dedup_ratio(spa), src);
229 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
230 rvd->vdev_state, src);
232 version = spa_version(spa);
233 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
234 src = ZPROP_SRC_DEFAULT;
236 src = ZPROP_SRC_LOCAL;
237 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
241 dsl_dir_t *freedir = pool->dp_free_dir;
244 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
245 * when opening pools before this version freedir will be NULL.
247 if (freedir != NULL) {
248 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
249 freedir->dd_phys->dd_used_bytes, src);
251 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
256 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
258 if (spa->spa_comment != NULL) {
259 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
263 if (spa->spa_root != NULL)
264 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
267 if ((dp = list_head(&spa->spa_config_list)) != NULL) {
268 if (dp->scd_path == NULL) {
269 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
270 "none", 0, ZPROP_SRC_LOCAL);
271 } else if (strcmp(dp->scd_path, spa_config_path) != 0) {
272 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
273 dp->scd_path, 0, ZPROP_SRC_LOCAL);
279 * Get zpool property values.
282 spa_prop_get(spa_t *spa, nvlist_t **nvp)
284 objset_t *mos = spa->spa_meta_objset;
289 err = nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_PUSHPAGE);
293 mutex_enter(&spa->spa_props_lock);
296 * Get properties from the spa config.
298 spa_prop_get_config(spa, nvp);
300 /* If no pool property object, no more prop to get. */
301 if (mos == NULL || spa->spa_pool_props_object == 0) {
302 mutex_exit(&spa->spa_props_lock);
307 * Get properties from the MOS pool property object.
309 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
310 (err = zap_cursor_retrieve(&zc, &za)) == 0;
311 zap_cursor_advance(&zc)) {
314 zprop_source_t src = ZPROP_SRC_DEFAULT;
317 if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
320 switch (za.za_integer_length) {
322 /* integer property */
323 if (za.za_first_integer !=
324 zpool_prop_default_numeric(prop))
325 src = ZPROP_SRC_LOCAL;
327 if (prop == ZPOOL_PROP_BOOTFS) {
329 dsl_dataset_t *ds = NULL;
331 dp = spa_get_dsl(spa);
332 rw_enter(&dp->dp_config_rwlock, RW_READER);
333 if ((err = dsl_dataset_hold_obj(dp,
334 za.za_first_integer, FTAG, &ds))) {
335 rw_exit(&dp->dp_config_rwlock);
340 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
342 dsl_dataset_name(ds, strval);
343 dsl_dataset_rele(ds, FTAG);
344 rw_exit(&dp->dp_config_rwlock);
347 intval = za.za_first_integer;
350 spa_prop_add_list(*nvp, prop, strval, intval, src);
354 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
359 /* string property */
360 strval = kmem_alloc(za.za_num_integers, KM_PUSHPAGE);
361 err = zap_lookup(mos, spa->spa_pool_props_object,
362 za.za_name, 1, za.za_num_integers, strval);
364 kmem_free(strval, za.za_num_integers);
367 spa_prop_add_list(*nvp, prop, strval, 0, src);
368 kmem_free(strval, za.za_num_integers);
375 zap_cursor_fini(&zc);
376 mutex_exit(&spa->spa_props_lock);
378 if (err && err != ENOENT) {
388 * Validate the given pool properties nvlist and modify the list
389 * for the property values to be set.
392 spa_prop_validate(spa_t *spa, nvlist_t *props)
395 int error = 0, reset_bootfs = 0;
397 boolean_t has_feature = B_FALSE;
400 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
402 char *strval, *slash, *check, *fname;
403 const char *propname = nvpair_name(elem);
404 zpool_prop_t prop = zpool_name_to_prop(propname);
408 if (!zpool_prop_feature(propname)) {
414 * Sanitize the input.
416 if (nvpair_type(elem) != DATA_TYPE_UINT64) {
421 if (nvpair_value_uint64(elem, &intval) != 0) {
431 fname = strchr(propname, '@') + 1;
432 if (zfeature_lookup_name(fname, NULL) != 0) {
437 has_feature = B_TRUE;
440 case ZPOOL_PROP_VERSION:
441 error = nvpair_value_uint64(elem, &intval);
443 (intval < spa_version(spa) ||
444 intval > SPA_VERSION_BEFORE_FEATURES ||
449 case ZPOOL_PROP_DELEGATION:
450 case ZPOOL_PROP_AUTOREPLACE:
451 case ZPOOL_PROP_LISTSNAPS:
452 case ZPOOL_PROP_AUTOEXPAND:
453 error = nvpair_value_uint64(elem, &intval);
454 if (!error && intval > 1)
458 case ZPOOL_PROP_BOOTFS:
460 * If the pool version is less than SPA_VERSION_BOOTFS,
461 * or the pool is still being created (version == 0),
462 * the bootfs property cannot be set.
464 if (spa_version(spa) < SPA_VERSION_BOOTFS) {
470 * Make sure the vdev config is bootable
472 if (!vdev_is_bootable(spa->spa_root_vdev)) {
479 error = nvpair_value_string(elem, &strval);
485 if (strval == NULL || strval[0] == '\0') {
486 objnum = zpool_prop_default_numeric(
491 if ((error = dmu_objset_hold(strval,FTAG,&os)))
494 /* Must be ZPL and not gzip compressed. */
496 if (dmu_objset_type(os) != DMU_OST_ZFS) {
498 } else if ((error = dsl_prop_get_integer(strval,
499 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
500 &compress, NULL)) == 0 &&
501 !BOOTFS_COMPRESS_VALID(compress)) {
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))
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;
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] != '/') {
547 slash = strrchr(strval, '/');
548 ASSERT(slash != NULL);
550 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
551 strcmp(slash, "/..") == 0)
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)) {
565 if (strlen(strval) > ZPROP_MAX_COMMENT)
569 case ZPOOL_PROP_DEDUPDITTO:
570 if (spa_version(spa) < SPA_VERSION_DEDUP)
573 error = nvpair_value_uint64(elem, &intval);
575 intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
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_do(spa_get_dsl(spa), NULL,
665 spa_sync_version, spa, &ver, 6);
676 return (dsl_sync_task_do(spa_get_dsl(spa), 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 *arg1, void *arg2, dmu_tx_t *tx)
702 vdev_t *rvd = spa->spa_root_vdev;
704 ASSERTV(uint64_t *newguid = arg2);
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)
713 ASSERT3U(spa_guid(spa), !=, *newguid);
719 spa_change_guid_sync(void *arg1, void *arg2, dmu_tx_t *tx)
722 uint64_t *newguid = arg2;
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(LOG_POOL_GUID_CHANGE, spa, tx,
735 "old=%lld new=%lld", oldguid, *newguid);
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_do(spa_get_dsl(spa), spa_change_guid_check,
757 spa_change_guid_sync, spa, &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)
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 = NULL;
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);
1459 oldvdevs = sav->sav_vdevs;
1460 oldnvdevs = sav->sav_count;
1461 sav->sav_vdevs = NULL;
1465 * Process new nvlist of vdevs.
1467 for (i = 0; i < nl2cache; i++) {
1468 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1472 for (j = 0; j < oldnvdevs; j++) {
1474 if (vd != NULL && guid == vd->vdev_guid) {
1476 * Retain previous vdev for add/remove ops.
1484 if (newvdevs[i] == NULL) {
1488 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1489 VDEV_ALLOC_L2CACHE) == 0);
1494 * Commit this vdev as an l2cache device,
1495 * even if it fails to open.
1497 spa_l2cache_add(vd);
1502 spa_l2cache_activate(vd);
1504 if (vdev_open(vd) != 0)
1507 (void) vdev_validate_aux(vd);
1509 if (!vdev_is_dead(vd))
1510 l2arc_add_vdev(spa, vd);
1515 * Purge vdevs that were dropped
1517 for (i = 0; i < oldnvdevs; i++) {
1522 ASSERT(vd->vdev_isl2cache);
1524 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1525 pool != 0ULL && l2arc_vdev_present(vd))
1526 l2arc_remove_vdev(vd);
1527 vdev_clear_stats(vd);
1533 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1535 if (sav->sav_config == NULL)
1538 sav->sav_vdevs = newvdevs;
1539 sav->sav_count = (int)nl2cache;
1542 * Recompute the stashed list of l2cache devices, with status
1543 * information this time.
1545 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1546 DATA_TYPE_NVLIST_ARRAY) == 0);
1548 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_PUSHPAGE);
1549 for (i = 0; i < sav->sav_count; i++)
1550 l2cache[i] = vdev_config_generate(spa,
1551 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1552 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1553 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1555 for (i = 0; i < sav->sav_count; i++)
1556 nvlist_free(l2cache[i]);
1558 kmem_free(l2cache, sav->sav_count * sizeof (void *));
1562 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1565 char *packed = NULL;
1570 error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db);
1574 nvsize = *(uint64_t *)db->db_data;
1575 dmu_buf_rele(db, FTAG);
1577 packed = kmem_alloc(nvsize, KM_PUSHPAGE | KM_NODEBUG);
1578 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1581 error = nvlist_unpack(packed, nvsize, value, 0);
1582 kmem_free(packed, nvsize);
1588 * Checks to see if the given vdev could not be opened, in which case we post a
1589 * sysevent to notify the autoreplace code that the device has been removed.
1592 spa_check_removed(vdev_t *vd)
1596 for (c = 0; c < vd->vdev_children; c++)
1597 spa_check_removed(vd->vdev_child[c]);
1599 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd)) {
1600 zfs_ereport_post(FM_EREPORT_RESOURCE_AUTOREPLACE,
1601 vd->vdev_spa, vd, NULL, 0, 0);
1602 spa_event_notify(vd->vdev_spa, vd, FM_EREPORT_ZFS_DEVICE_CHECK);
1607 * Validate the current config against the MOS config
1610 spa_config_valid(spa_t *spa, nvlist_t *config)
1612 vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1616 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1618 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1619 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1621 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1624 * If we're doing a normal import, then build up any additional
1625 * diagnostic information about missing devices in this config.
1626 * We'll pass this up to the user for further processing.
1628 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1629 nvlist_t **child, *nv;
1632 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1634 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_PUSHPAGE) == 0);
1636 for (c = 0; c < rvd->vdev_children; c++) {
1637 vdev_t *tvd = rvd->vdev_child[c];
1638 vdev_t *mtvd = mrvd->vdev_child[c];
1640 if (tvd->vdev_ops == &vdev_missing_ops &&
1641 mtvd->vdev_ops != &vdev_missing_ops &&
1643 child[idx++] = vdev_config_generate(spa, mtvd,
1648 VERIFY(nvlist_add_nvlist_array(nv,
1649 ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1650 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1651 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1653 for (i = 0; i < idx; i++)
1654 nvlist_free(child[i]);
1657 kmem_free(child, rvd->vdev_children * sizeof (char **));
1661 * Compare the root vdev tree with the information we have
1662 * from the MOS config (mrvd). Check each top-level vdev
1663 * with the corresponding MOS config top-level (mtvd).
1665 for (c = 0; c < rvd->vdev_children; c++) {
1666 vdev_t *tvd = rvd->vdev_child[c];
1667 vdev_t *mtvd = mrvd->vdev_child[c];
1670 * Resolve any "missing" vdevs in the current configuration.
1671 * If we find that the MOS config has more accurate information
1672 * about the top-level vdev then use that vdev instead.
1674 if (tvd->vdev_ops == &vdev_missing_ops &&
1675 mtvd->vdev_ops != &vdev_missing_ops) {
1677 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1681 * Device specific actions.
1683 if (mtvd->vdev_islog) {
1684 spa_set_log_state(spa, SPA_LOG_CLEAR);
1687 * XXX - once we have 'readonly' pool
1688 * support we should be able to handle
1689 * missing data devices by transitioning
1690 * the pool to readonly.
1696 * Swap the missing vdev with the data we were
1697 * able to obtain from the MOS config.
1699 vdev_remove_child(rvd, tvd);
1700 vdev_remove_child(mrvd, mtvd);
1702 vdev_add_child(rvd, mtvd);
1703 vdev_add_child(mrvd, tvd);
1705 spa_config_exit(spa, SCL_ALL, FTAG);
1707 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1710 } else if (mtvd->vdev_islog) {
1712 * Load the slog device's state from the MOS config
1713 * since it's possible that the label does not
1714 * contain the most up-to-date information.
1716 vdev_load_log_state(tvd, mtvd);
1721 spa_config_exit(spa, SCL_ALL, FTAG);
1724 * Ensure we were able to validate the config.
1726 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1730 * Check for missing log devices
1733 spa_check_logs(spa_t *spa)
1735 switch (spa->spa_log_state) {
1738 case SPA_LOG_MISSING:
1739 /* need to recheck in case slog has been restored */
1740 case SPA_LOG_UNKNOWN:
1741 if (dmu_objset_find(spa->spa_name, zil_check_log_chain, NULL,
1742 DS_FIND_CHILDREN)) {
1743 spa_set_log_state(spa, SPA_LOG_MISSING);
1752 spa_passivate_log(spa_t *spa)
1754 vdev_t *rvd = spa->spa_root_vdev;
1755 boolean_t slog_found = B_FALSE;
1758 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1760 if (!spa_has_slogs(spa))
1763 for (c = 0; c < rvd->vdev_children; c++) {
1764 vdev_t *tvd = rvd->vdev_child[c];
1765 metaslab_group_t *mg = tvd->vdev_mg;
1767 if (tvd->vdev_islog) {
1768 metaslab_group_passivate(mg);
1769 slog_found = B_TRUE;
1773 return (slog_found);
1777 spa_activate_log(spa_t *spa)
1779 vdev_t *rvd = spa->spa_root_vdev;
1782 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1784 for (c = 0; c < rvd->vdev_children; c++) {
1785 vdev_t *tvd = rvd->vdev_child[c];
1786 metaslab_group_t *mg = tvd->vdev_mg;
1788 if (tvd->vdev_islog)
1789 metaslab_group_activate(mg);
1794 spa_offline_log(spa_t *spa)
1798 if ((error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1799 NULL, DS_FIND_CHILDREN)) == 0) {
1802 * We successfully offlined the log device, sync out the
1803 * current txg so that the "stubby" block can be removed
1806 txg_wait_synced(spa->spa_dsl_pool, 0);
1812 spa_aux_check_removed(spa_aux_vdev_t *sav)
1816 for (i = 0; i < sav->sav_count; i++)
1817 spa_check_removed(sav->sav_vdevs[i]);
1821 spa_claim_notify(zio_t *zio)
1823 spa_t *spa = zio->io_spa;
1828 mutex_enter(&spa->spa_props_lock); /* any mutex will do */
1829 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1830 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1831 mutex_exit(&spa->spa_props_lock);
1834 typedef struct spa_load_error {
1835 uint64_t sle_meta_count;
1836 uint64_t sle_data_count;
1840 spa_load_verify_done(zio_t *zio)
1842 blkptr_t *bp = zio->io_bp;
1843 spa_load_error_t *sle = zio->io_private;
1844 dmu_object_type_t type = BP_GET_TYPE(bp);
1845 int error = zio->io_error;
1848 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1849 type != DMU_OT_INTENT_LOG)
1850 atomic_add_64(&sle->sle_meta_count, 1);
1852 atomic_add_64(&sle->sle_data_count, 1);
1854 zio_data_buf_free(zio->io_data, zio->io_size);
1859 spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1860 const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1864 size_t size = BP_GET_PSIZE(bp);
1865 void *data = zio_data_buf_alloc(size);
1867 zio_nowait(zio_read(rio, spa, bp, data, size,
1868 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1869 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1870 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1876 spa_load_verify(spa_t *spa)
1879 spa_load_error_t sle = { 0 };
1880 zpool_rewind_policy_t policy;
1881 boolean_t verify_ok = B_FALSE;
1884 zpool_get_rewind_policy(spa->spa_config, &policy);
1886 if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1889 rio = zio_root(spa, NULL, &sle,
1890 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1892 error = traverse_pool(spa, spa->spa_verify_min_txg,
1893 TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio);
1895 (void) zio_wait(rio);
1897 spa->spa_load_meta_errors = sle.sle_meta_count;
1898 spa->spa_load_data_errors = sle.sle_data_count;
1900 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1901 sle.sle_data_count <= policy.zrp_maxdata) {
1905 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1906 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1908 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1909 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1910 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1911 VERIFY(nvlist_add_int64(spa->spa_load_info,
1912 ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1913 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1914 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1916 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1920 if (error != ENXIO && error != EIO)
1925 return (verify_ok ? 0 : EIO);
1929 * Find a value in the pool props object.
1932 spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1934 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1935 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1939 * Find a value in the pool directory object.
1942 spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1944 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1945 name, sizeof (uint64_t), 1, val));
1949 spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1951 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1956 * Fix up config after a partly-completed split. This is done with the
1957 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
1958 * pool have that entry in their config, but only the splitting one contains
1959 * a list of all the guids of the vdevs that are being split off.
1961 * This function determines what to do with that list: either rejoin
1962 * all the disks to the pool, or complete the splitting process. To attempt
1963 * the rejoin, each disk that is offlined is marked online again, and
1964 * we do a reopen() call. If the vdev label for every disk that was
1965 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1966 * then we call vdev_split() on each disk, and complete the split.
1968 * Otherwise we leave the config alone, with all the vdevs in place in
1969 * the original pool.
1972 spa_try_repair(spa_t *spa, nvlist_t *config)
1979 boolean_t attempt_reopen;
1981 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
1984 /* check that the config is complete */
1985 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
1986 &glist, &gcount) != 0)
1989 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_PUSHPAGE);
1991 /* attempt to online all the vdevs & validate */
1992 attempt_reopen = B_TRUE;
1993 for (i = 0; i < gcount; i++) {
1994 if (glist[i] == 0) /* vdev is hole */
1997 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
1998 if (vd[i] == NULL) {
2000 * Don't bother attempting to reopen the disks;
2001 * just do the split.
2003 attempt_reopen = B_FALSE;
2005 /* attempt to re-online it */
2006 vd[i]->vdev_offline = B_FALSE;
2010 if (attempt_reopen) {
2011 vdev_reopen(spa->spa_root_vdev);
2013 /* check each device to see what state it's in */
2014 for (extracted = 0, i = 0; i < gcount; i++) {
2015 if (vd[i] != NULL &&
2016 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
2023 * If every disk has been moved to the new pool, or if we never
2024 * even attempted to look at them, then we split them off for
2027 if (!attempt_reopen || gcount == extracted) {
2028 for (i = 0; i < gcount; i++)
2031 vdev_reopen(spa->spa_root_vdev);
2034 kmem_free(vd, gcount * sizeof (vdev_t *));
2038 spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
2039 boolean_t mosconfig)
2041 nvlist_t *config = spa->spa_config;
2042 char *ereport = FM_EREPORT_ZFS_POOL;
2048 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
2051 ASSERT(spa->spa_comment == NULL);
2052 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
2053 spa->spa_comment = spa_strdup(comment);
2056 * Versioning wasn't explicitly added to the label until later, so if
2057 * it's not present treat it as the initial version.
2059 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
2060 &spa->spa_ubsync.ub_version) != 0)
2061 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
2063 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
2064 &spa->spa_config_txg);
2066 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
2067 spa_guid_exists(pool_guid, 0)) {
2070 spa->spa_config_guid = pool_guid;
2072 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
2074 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
2078 nvlist_free(spa->spa_load_info);
2079 spa->spa_load_info = fnvlist_alloc();
2081 gethrestime(&spa->spa_loaded_ts);
2082 error = spa_load_impl(spa, pool_guid, config, state, type,
2083 mosconfig, &ereport);
2086 spa->spa_minref = refcount_count(&spa->spa_refcount);
2088 if (error != EEXIST) {
2089 spa->spa_loaded_ts.tv_sec = 0;
2090 spa->spa_loaded_ts.tv_nsec = 0;
2092 if (error != EBADF) {
2093 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
2096 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2103 * Load an existing storage pool, using the pool's builtin spa_config as a
2104 * source of configuration information.
2106 __attribute__((always_inline))
2108 spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
2109 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
2113 nvlist_t *nvroot = NULL;
2116 uberblock_t *ub = &spa->spa_uberblock;
2117 uint64_t children, config_cache_txg = spa->spa_config_txg;
2118 int orig_mode = spa->spa_mode;
2121 boolean_t missing_feat_write = B_FALSE;
2124 * If this is an untrusted config, access the pool in read-only mode.
2125 * This prevents things like resilvering recently removed devices.
2128 spa->spa_mode = FREAD;
2130 ASSERT(MUTEX_HELD(&spa_namespace_lock));
2132 spa->spa_load_state = state;
2134 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
2137 parse = (type == SPA_IMPORT_EXISTING ?
2138 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2141 * Create "The Godfather" zio to hold all async IOs
2143 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2144 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2147 * Parse the configuration into a vdev tree. We explicitly set the
2148 * value that will be returned by spa_version() since parsing the
2149 * configuration requires knowing the version number.
2151 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2152 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
2153 spa_config_exit(spa, SCL_ALL, FTAG);
2158 ASSERT(spa->spa_root_vdev == rvd);
2160 if (type != SPA_IMPORT_ASSEMBLE) {
2161 ASSERT(spa_guid(spa) == pool_guid);
2165 * Try to open all vdevs, loading each label in the process.
2167 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2168 error = vdev_open(rvd);
2169 spa_config_exit(spa, SCL_ALL, FTAG);
2174 * We need to validate the vdev labels against the configuration that
2175 * we have in hand, which is dependent on the setting of mosconfig. If
2176 * mosconfig is true then we're validating the vdev labels based on
2177 * that config. Otherwise, we're validating against the cached config
2178 * (zpool.cache) that was read when we loaded the zfs module, and then
2179 * later we will recursively call spa_load() and validate against
2182 * If we're assembling a new pool that's been split off from an
2183 * existing pool, the labels haven't yet been updated so we skip
2184 * validation for now.
2186 if (type != SPA_IMPORT_ASSEMBLE) {
2187 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2188 error = vdev_validate(rvd, mosconfig);
2189 spa_config_exit(spa, SCL_ALL, FTAG);
2194 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2199 * Find the best uberblock.
2201 vdev_uberblock_load(rvd, ub, &label);
2204 * If we weren't able to find a single valid uberblock, return failure.
2206 if (ub->ub_txg == 0) {
2208 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2212 * If the pool has an unsupported version we can't open it.
2214 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2216 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2219 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2223 * If we weren't able to find what's necessary for reading the
2224 * MOS in the label, return failure.
2226 if (label == NULL || nvlist_lookup_nvlist(label,
2227 ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) {
2229 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2234 * Update our in-core representation with the definitive values
2237 nvlist_free(spa->spa_label_features);
2238 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2244 * Look through entries in the label nvlist's features_for_read. If
2245 * there is a feature listed there which we don't understand then we
2246 * cannot open a pool.
2248 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2249 nvlist_t *unsup_feat;
2252 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2255 for (nvp = nvlist_next_nvpair(spa->spa_label_features, NULL);
2257 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2258 if (!zfeature_is_supported(nvpair_name(nvp))) {
2259 VERIFY(nvlist_add_string(unsup_feat,
2260 nvpair_name(nvp), "") == 0);
2264 if (!nvlist_empty(unsup_feat)) {
2265 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2266 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2267 nvlist_free(unsup_feat);
2268 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2272 nvlist_free(unsup_feat);
2276 * If the vdev guid sum doesn't match the uberblock, we have an
2277 * incomplete configuration. We first check to see if the pool
2278 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2279 * If it is, defer the vdev_guid_sum check till later so we
2280 * can handle missing vdevs.
2282 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
2283 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
2284 rvd->vdev_guid_sum != ub->ub_guid_sum)
2285 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2287 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2288 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2289 spa_try_repair(spa, config);
2290 spa_config_exit(spa, SCL_ALL, FTAG);
2291 nvlist_free(spa->spa_config_splitting);
2292 spa->spa_config_splitting = NULL;
2296 * Initialize internal SPA structures.
2298 spa->spa_state = POOL_STATE_ACTIVE;
2299 spa->spa_ubsync = spa->spa_uberblock;
2300 spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2301 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2302 spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2303 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2304 spa->spa_claim_max_txg = spa->spa_first_txg;
2305 spa->spa_prev_software_version = ub->ub_software_version;
2307 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2309 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2310 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2312 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2313 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2315 if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2316 boolean_t missing_feat_read = B_FALSE;
2317 nvlist_t *unsup_feat, *enabled_feat;
2319 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2320 &spa->spa_feat_for_read_obj) != 0) {
2321 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2324 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2325 &spa->spa_feat_for_write_obj) != 0) {
2326 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2329 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2330 &spa->spa_feat_desc_obj) != 0) {
2331 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2334 enabled_feat = fnvlist_alloc();
2335 unsup_feat = fnvlist_alloc();
2337 if (!feature_is_supported(spa->spa_meta_objset,
2338 spa->spa_feat_for_read_obj, spa->spa_feat_desc_obj,
2339 unsup_feat, enabled_feat))
2340 missing_feat_read = B_TRUE;
2342 if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) {
2343 if (!feature_is_supported(spa->spa_meta_objset,
2344 spa->spa_feat_for_write_obj, spa->spa_feat_desc_obj,
2345 unsup_feat, enabled_feat)) {
2346 missing_feat_write = B_TRUE;
2350 fnvlist_add_nvlist(spa->spa_load_info,
2351 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
2353 if (!nvlist_empty(unsup_feat)) {
2354 fnvlist_add_nvlist(spa->spa_load_info,
2355 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
2358 fnvlist_free(enabled_feat);
2359 fnvlist_free(unsup_feat);
2361 if (!missing_feat_read) {
2362 fnvlist_add_boolean(spa->spa_load_info,
2363 ZPOOL_CONFIG_CAN_RDONLY);
2367 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2368 * twofold: to determine whether the pool is available for
2369 * import in read-write mode and (if it is not) whether the
2370 * pool is available for import in read-only mode. If the pool
2371 * is available for import in read-write mode, it is displayed
2372 * as available in userland; if it is not available for import
2373 * in read-only mode, it is displayed as unavailable in
2374 * userland. If the pool is available for import in read-only
2375 * mode but not read-write mode, it is displayed as unavailable
2376 * in userland with a special note that the pool is actually
2377 * available for open in read-only mode.
2379 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2380 * missing a feature for write, we must first determine whether
2381 * the pool can be opened read-only before returning to
2382 * userland in order to know whether to display the
2383 * abovementioned note.
2385 if (missing_feat_read || (missing_feat_write &&
2386 spa_writeable(spa))) {
2387 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2392 spa->spa_is_initializing = B_TRUE;
2393 error = dsl_pool_open(spa->spa_dsl_pool);
2394 spa->spa_is_initializing = B_FALSE;
2396 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2400 nvlist_t *policy = NULL, *nvconfig;
2402 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2403 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2405 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2406 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2408 unsigned long myhostid = 0;
2410 VERIFY(nvlist_lookup_string(nvconfig,
2411 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2414 myhostid = zone_get_hostid(NULL);
2417 * We're emulating the system's hostid in userland, so
2418 * we can't use zone_get_hostid().
2420 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2421 #endif /* _KERNEL */
2422 if (hostid != 0 && myhostid != 0 &&
2423 hostid != myhostid) {
2424 nvlist_free(nvconfig);
2425 cmn_err(CE_WARN, "pool '%s' could not be "
2426 "loaded as it was last accessed by "
2427 "another system (host: %s hostid: 0x%lx). "
2428 "See: http://zfsonlinux.org/msg/ZFS-8000-EY",
2429 spa_name(spa), hostname,
2430 (unsigned long)hostid);
2434 if (nvlist_lookup_nvlist(spa->spa_config,
2435 ZPOOL_REWIND_POLICY, &policy) == 0)
2436 VERIFY(nvlist_add_nvlist(nvconfig,
2437 ZPOOL_REWIND_POLICY, policy) == 0);
2439 spa_config_set(spa, nvconfig);
2441 spa_deactivate(spa);
2442 spa_activate(spa, orig_mode);
2444 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2447 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2448 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2449 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2451 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2454 * Load the bit that tells us to use the new accounting function
2455 * (raid-z deflation). If we have an older pool, this will not
2458 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2459 if (error != 0 && error != ENOENT)
2460 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2462 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2463 &spa->spa_creation_version);
2464 if (error != 0 && error != ENOENT)
2465 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2468 * Load the persistent error log. If we have an older pool, this will
2471 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2472 if (error != 0 && error != ENOENT)
2473 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2475 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2476 &spa->spa_errlog_scrub);
2477 if (error != 0 && error != ENOENT)
2478 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2481 * Load the history object. If we have an older pool, this
2482 * will not be present.
2484 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2485 if (error != 0 && error != ENOENT)
2486 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2489 * If we're assembling the pool from the split-off vdevs of
2490 * an existing pool, we don't want to attach the spares & cache
2495 * Load any hot spares for this pool.
2497 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2498 if (error != 0 && error != ENOENT)
2499 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2500 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2501 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2502 if (load_nvlist(spa, spa->spa_spares.sav_object,
2503 &spa->spa_spares.sav_config) != 0)
2504 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2506 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2507 spa_load_spares(spa);
2508 spa_config_exit(spa, SCL_ALL, FTAG);
2509 } else if (error == 0) {
2510 spa->spa_spares.sav_sync = B_TRUE;
2514 * Load any level 2 ARC devices for this pool.
2516 error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2517 &spa->spa_l2cache.sav_object);
2518 if (error != 0 && error != ENOENT)
2519 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2520 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2521 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2522 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2523 &spa->spa_l2cache.sav_config) != 0)
2524 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2526 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2527 spa_load_l2cache(spa);
2528 spa_config_exit(spa, SCL_ALL, FTAG);
2529 } else if (error == 0) {
2530 spa->spa_l2cache.sav_sync = B_TRUE;
2533 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2535 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2536 if (error && error != ENOENT)
2537 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2540 uint64_t autoreplace;
2542 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2543 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2544 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2545 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2546 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2547 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2548 &spa->spa_dedup_ditto);
2550 spa->spa_autoreplace = (autoreplace != 0);
2554 * If the 'autoreplace' property is set, then post a resource notifying
2555 * the ZFS DE that it should not issue any faults for unopenable
2556 * devices. We also iterate over the vdevs, and post a sysevent for any
2557 * unopenable vdevs so that the normal autoreplace handler can take
2560 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2561 spa_check_removed(spa->spa_root_vdev);
2563 * For the import case, this is done in spa_import(), because
2564 * at this point we're using the spare definitions from
2565 * the MOS config, not necessarily from the userland config.
2567 if (state != SPA_LOAD_IMPORT) {
2568 spa_aux_check_removed(&spa->spa_spares);
2569 spa_aux_check_removed(&spa->spa_l2cache);
2574 * Load the vdev state for all toplevel vdevs.
2579 * Propagate the leaf DTLs we just loaded all the way up the tree.
2581 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2582 vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2583 spa_config_exit(spa, SCL_ALL, FTAG);
2586 * Load the DDTs (dedup tables).
2588 error = ddt_load(spa);
2590 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2592 spa_update_dspace(spa);
2595 * Validate the config, using the MOS config to fill in any
2596 * information which might be missing. If we fail to validate
2597 * the config then declare the pool unfit for use. If we're
2598 * assembling a pool from a split, the log is not transferred
2601 if (type != SPA_IMPORT_ASSEMBLE) {
2604 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2605 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2607 if (!spa_config_valid(spa, nvconfig)) {
2608 nvlist_free(nvconfig);
2609 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2612 nvlist_free(nvconfig);
2615 * Now that we've validated the config, check the state of the
2616 * root vdev. If it can't be opened, it indicates one or
2617 * more toplevel vdevs are faulted.
2619 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2622 if (spa_check_logs(spa)) {
2623 *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2624 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2628 if (missing_feat_write) {
2629 ASSERT(state == SPA_LOAD_TRYIMPORT);
2632 * At this point, we know that we can open the pool in
2633 * read-only mode but not read-write mode. We now have enough
2634 * information and can return to userland.
2636 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
2640 * We've successfully opened the pool, verify that we're ready
2641 * to start pushing transactions.
2643 if (state != SPA_LOAD_TRYIMPORT) {
2644 if ((error = spa_load_verify(spa)))
2645 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2649 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2650 spa->spa_load_max_txg == UINT64_MAX)) {
2652 int need_update = B_FALSE;
2655 ASSERT(state != SPA_LOAD_TRYIMPORT);
2658 * Claim log blocks that haven't been committed yet.
2659 * This must all happen in a single txg.
2660 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2661 * invoked from zil_claim_log_block()'s i/o done callback.
2662 * Price of rollback is that we abandon the log.
2664 spa->spa_claiming = B_TRUE;
2666 tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2667 spa_first_txg(spa));
2668 (void) dmu_objset_find(spa_name(spa),
2669 zil_claim, tx, DS_FIND_CHILDREN);
2672 spa->spa_claiming = B_FALSE;
2674 spa_set_log_state(spa, SPA_LOG_GOOD);
2675 spa->spa_sync_on = B_TRUE;
2676 txg_sync_start(spa->spa_dsl_pool);
2679 * Wait for all claims to sync. We sync up to the highest
2680 * claimed log block birth time so that claimed log blocks
2681 * don't appear to be from the future. spa_claim_max_txg
2682 * will have been set for us by either zil_check_log_chain()
2683 * (invoked from spa_check_logs()) or zil_claim() above.
2685 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2688 * If the config cache is stale, or we have uninitialized
2689 * metaslabs (see spa_vdev_add()), then update the config.
2691 * If this is a verbatim import, trust the current
2692 * in-core spa_config and update the disk labels.
2694 if (config_cache_txg != spa->spa_config_txg ||
2695 state == SPA_LOAD_IMPORT ||
2696 state == SPA_LOAD_RECOVER ||
2697 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2698 need_update = B_TRUE;
2700 for (c = 0; c < rvd->vdev_children; c++)
2701 if (rvd->vdev_child[c]->vdev_ms_array == 0)
2702 need_update = B_TRUE;
2705 * Update the config cache asychronously in case we're the
2706 * root pool, in which case the config cache isn't writable yet.
2709 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2712 * Check all DTLs to see if anything needs resilvering.
2714 if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2715 vdev_resilver_needed(rvd, NULL, NULL))
2716 spa_async_request(spa, SPA_ASYNC_RESILVER);
2719 * Delete any inconsistent datasets.
2721 (void) dmu_objset_find(spa_name(spa),
2722 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2725 * Clean up any stale temporary dataset userrefs.
2727 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2734 spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2736 int mode = spa->spa_mode;
2739 spa_deactivate(spa);
2741 spa->spa_load_max_txg--;
2743 spa_activate(spa, mode);
2744 spa_async_suspend(spa);
2746 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2750 * If spa_load() fails this function will try loading prior txg's. If
2751 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2752 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2753 * function will not rewind the pool and will return the same error as
2757 spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2758 uint64_t max_request, int rewind_flags)
2760 nvlist_t *loadinfo = NULL;
2761 nvlist_t *config = NULL;
2762 int load_error, rewind_error;
2763 uint64_t safe_rewind_txg;
2766 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2767 spa->spa_load_max_txg = spa->spa_load_txg;
2768 spa_set_log_state(spa, SPA_LOG_CLEAR);
2770 spa->spa_load_max_txg = max_request;
2773 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2775 if (load_error == 0)
2778 if (spa->spa_root_vdev != NULL)
2779 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2781 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2782 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2784 if (rewind_flags & ZPOOL_NEVER_REWIND) {
2785 nvlist_free(config);
2786 return (load_error);
2789 if (state == SPA_LOAD_RECOVER) {
2790 /* Price of rolling back is discarding txgs, including log */
2791 spa_set_log_state(spa, SPA_LOG_CLEAR);
2794 * If we aren't rolling back save the load info from our first
2795 * import attempt so that we can restore it after attempting
2798 loadinfo = spa->spa_load_info;
2799 spa->spa_load_info = fnvlist_alloc();
2802 spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2803 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2804 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2805 TXG_INITIAL : safe_rewind_txg;
2808 * Continue as long as we're finding errors, we're still within
2809 * the acceptable rewind range, and we're still finding uberblocks
2811 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2812 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2813 if (spa->spa_load_max_txg < safe_rewind_txg)
2814 spa->spa_extreme_rewind = B_TRUE;
2815 rewind_error = spa_load_retry(spa, state, mosconfig);
2818 spa->spa_extreme_rewind = B_FALSE;
2819 spa->spa_load_max_txg = UINT64_MAX;
2821 if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2822 spa_config_set(spa, config);
2824 if (state == SPA_LOAD_RECOVER) {
2825 ASSERT3P(loadinfo, ==, NULL);
2826 return (rewind_error);
2828 /* Store the rewind info as part of the initial load info */
2829 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
2830 spa->spa_load_info);
2832 /* Restore the initial load info */
2833 fnvlist_free(spa->spa_load_info);
2834 spa->spa_load_info = loadinfo;
2836 return (load_error);
2843 * The import case is identical to an open except that the configuration is sent
2844 * down from userland, instead of grabbed from the configuration cache. For the
2845 * case of an open, the pool configuration will exist in the
2846 * POOL_STATE_UNINITIALIZED state.
2848 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2849 * the same time open the pool, without having to keep around the spa_t in some
2853 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2857 spa_load_state_t state = SPA_LOAD_OPEN;
2859 int locked = B_FALSE;
2860 int firstopen = B_FALSE;
2865 * As disgusting as this is, we need to support recursive calls to this
2866 * function because dsl_dir_open() is called during spa_load(), and ends
2867 * up calling spa_open() again. The real fix is to figure out how to
2868 * avoid dsl_dir_open() calling this in the first place.
2870 if (mutex_owner(&spa_namespace_lock) != curthread) {
2871 mutex_enter(&spa_namespace_lock);
2875 if ((spa = spa_lookup(pool)) == NULL) {
2877 mutex_exit(&spa_namespace_lock);
2881 if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2882 zpool_rewind_policy_t policy;
2886 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2888 if (policy.zrp_request & ZPOOL_DO_REWIND)
2889 state = SPA_LOAD_RECOVER;
2891 spa_activate(spa, spa_mode_global);
2893 if (state != SPA_LOAD_RECOVER)
2894 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2896 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2897 policy.zrp_request);
2899 if (error == EBADF) {
2901 * If vdev_validate() returns failure (indicated by
2902 * EBADF), it indicates that one of the vdevs indicates
2903 * that the pool has been exported or destroyed. If
2904 * this is the case, the config cache is out of sync and
2905 * we should remove the pool from the namespace.
2908 spa_deactivate(spa);
2909 spa_config_sync(spa, B_TRUE, B_TRUE);
2912 mutex_exit(&spa_namespace_lock);
2918 * We can't open the pool, but we still have useful
2919 * information: the state of each vdev after the
2920 * attempted vdev_open(). Return this to the user.
2922 if (config != NULL && spa->spa_config) {
2923 VERIFY(nvlist_dup(spa->spa_config, config,
2925 VERIFY(nvlist_add_nvlist(*config,
2926 ZPOOL_CONFIG_LOAD_INFO,
2927 spa->spa_load_info) == 0);
2930 spa_deactivate(spa);
2931 spa->spa_last_open_failed = error;
2933 mutex_exit(&spa_namespace_lock);
2939 spa_open_ref(spa, tag);
2942 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2945 * If we've recovered the pool, pass back any information we
2946 * gathered while doing the load.
2948 if (state == SPA_LOAD_RECOVER) {
2949 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
2950 spa->spa_load_info) == 0);
2954 spa->spa_last_open_failed = 0;
2955 spa->spa_last_ubsync_txg = 0;
2956 spa->spa_load_txg = 0;
2957 mutex_exit(&spa_namespace_lock);
2962 zvol_create_minors(spa->spa_name);
2971 spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
2974 return (spa_open_common(name, spapp, tag, policy, config));
2978 spa_open(const char *name, spa_t **spapp, void *tag)
2980 return (spa_open_common(name, spapp, tag, NULL, NULL));
2984 * Lookup the given spa_t, incrementing the inject count in the process,
2985 * preventing it from being exported or destroyed.
2988 spa_inject_addref(char *name)
2992 mutex_enter(&spa_namespace_lock);
2993 if ((spa = spa_lookup(name)) == NULL) {
2994 mutex_exit(&spa_namespace_lock);
2997 spa->spa_inject_ref++;
2998 mutex_exit(&spa_namespace_lock);
3004 spa_inject_delref(spa_t *spa)
3006 mutex_enter(&spa_namespace_lock);
3007 spa->spa_inject_ref--;
3008 mutex_exit(&spa_namespace_lock);
3012 * Add spares device information to the nvlist.
3015 spa_add_spares(spa_t *spa, nvlist_t *config)
3025 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3027 if (spa->spa_spares.sav_count == 0)
3030 VERIFY(nvlist_lookup_nvlist(config,
3031 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3032 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
3033 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3035 VERIFY(nvlist_add_nvlist_array(nvroot,
3036 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3037 VERIFY(nvlist_lookup_nvlist_array(nvroot,
3038 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3041 * Go through and find any spares which have since been
3042 * repurposed as an active spare. If this is the case, update
3043 * their status appropriately.
3045 for (i = 0; i < nspares; i++) {
3046 VERIFY(nvlist_lookup_uint64(spares[i],
3047 ZPOOL_CONFIG_GUID, &guid) == 0);
3048 if (spa_spare_exists(guid, &pool, NULL) &&
3050 VERIFY(nvlist_lookup_uint64_array(
3051 spares[i], ZPOOL_CONFIG_VDEV_STATS,
3052 (uint64_t **)&vs, &vsc) == 0);
3053 vs->vs_state = VDEV_STATE_CANT_OPEN;
3054 vs->vs_aux = VDEV_AUX_SPARED;
3061 * Add l2cache device information to the nvlist, including vdev stats.
3064 spa_add_l2cache(spa_t *spa, nvlist_t *config)
3067 uint_t i, j, nl2cache;
3074 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3076 if (spa->spa_l2cache.sav_count == 0)
3079 VERIFY(nvlist_lookup_nvlist(config,
3080 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3081 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
3082 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3083 if (nl2cache != 0) {
3084 VERIFY(nvlist_add_nvlist_array(nvroot,
3085 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3086 VERIFY(nvlist_lookup_nvlist_array(nvroot,
3087 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3090 * Update level 2 cache device stats.
3093 for (i = 0; i < nl2cache; i++) {
3094 VERIFY(nvlist_lookup_uint64(l2cache[i],
3095 ZPOOL_CONFIG_GUID, &guid) == 0);
3098 for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
3100 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
3101 vd = spa->spa_l2cache.sav_vdevs[j];
3107 VERIFY(nvlist_lookup_uint64_array(l2cache[i],
3108 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
3110 vdev_get_stats(vd, vs);
3116 spa_add_feature_stats(spa_t *spa, nvlist_t *config)
3122 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3123 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3125 if (spa->spa_feat_for_read_obj != 0) {
3126 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3127 spa->spa_feat_for_read_obj);
3128 zap_cursor_retrieve(&zc, &za) == 0;
3129 zap_cursor_advance(&zc)) {
3130 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3131 za.za_num_integers == 1);
3132 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3133 za.za_first_integer));
3135 zap_cursor_fini(&zc);
3138 if (spa->spa_feat_for_write_obj != 0) {
3139 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3140 spa->spa_feat_for_write_obj);
3141 zap_cursor_retrieve(&zc, &za) == 0;
3142 zap_cursor_advance(&zc)) {
3143 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3144 za.za_num_integers == 1);
3145 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3146 za.za_first_integer));
3148 zap_cursor_fini(&zc);
3151 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
3153 nvlist_free(features);
3157 spa_get_stats(const char *name, nvlist_t **config,
3158 char *altroot, size_t buflen)
3164 error = spa_open_common(name, &spa, FTAG, NULL, config);
3168 * This still leaves a window of inconsistency where the spares
3169 * or l2cache devices could change and the config would be
3170 * self-inconsistent.
3172 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3174 if (*config != NULL) {
3175 uint64_t loadtimes[2];
3177 loadtimes[0] = spa->spa_loaded_ts.tv_sec;
3178 loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
3179 VERIFY(nvlist_add_uint64_array(*config,
3180 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
3182 VERIFY(nvlist_add_uint64(*config,
3183 ZPOOL_CONFIG_ERRCOUNT,
3184 spa_get_errlog_size(spa)) == 0);
3186 if (spa_suspended(spa))
3187 VERIFY(nvlist_add_uint64(*config,
3188 ZPOOL_CONFIG_SUSPENDED,
3189 spa->spa_failmode) == 0);
3191 spa_add_spares(spa, *config);
3192 spa_add_l2cache(spa, *config);
3193 spa_add_feature_stats(spa, *config);
3198 * We want to get the alternate root even for faulted pools, so we cheat
3199 * and call spa_lookup() directly.
3203 mutex_enter(&spa_namespace_lock);
3204 spa = spa_lookup(name);
3206 spa_altroot(spa, altroot, buflen);
3210 mutex_exit(&spa_namespace_lock);
3212 spa_altroot(spa, altroot, buflen);
3217 spa_config_exit(spa, SCL_CONFIG, FTAG);
3218 spa_close(spa, FTAG);
3225 * Validate that the auxiliary device array is well formed. We must have an
3226 * array of nvlists, each which describes a valid leaf vdev. If this is an
3227 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3228 * specified, as long as they are well-formed.
3231 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
3232 spa_aux_vdev_t *sav, const char *config, uint64_t version,
3233 vdev_labeltype_t label)
3240 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3243 * It's acceptable to have no devs specified.
3245 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
3252 * Make sure the pool is formatted with a version that supports this
3255 if (spa_version(spa) < version)
3259 * Set the pending device list so we correctly handle device in-use
3262 sav->sav_pending = dev;
3263 sav->sav_npending = ndev;
3265 for (i = 0; i < ndev; i++) {
3266 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
3270 if (!vd->vdev_ops->vdev_op_leaf) {
3277 * The L2ARC currently only supports disk devices in
3278 * kernel context. For user-level testing, we allow it.
3281 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
3282 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
3290 if ((error = vdev_open(vd)) == 0 &&
3291 (error = vdev_label_init(vd, crtxg, label)) == 0) {
3292 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
3293 vd->vdev_guid) == 0);
3299 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
3306 sav->sav_pending = NULL;
3307 sav->sav_npending = 0;
3312 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
3316 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3318 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3319 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
3320 VDEV_LABEL_SPARE)) != 0) {
3324 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3325 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
3326 VDEV_LABEL_L2CACHE));
3330 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
3335 if (sav->sav_config != NULL) {
3341 * Generate new dev list by concatentating with the
3344 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
3345 &olddevs, &oldndevs) == 0);
3347 newdevs = kmem_alloc(sizeof (void *) *
3348 (ndevs + oldndevs), KM_PUSHPAGE);
3349 for (i = 0; i < oldndevs; i++)
3350 VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
3352 for (i = 0; i < ndevs; i++)
3353 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
3356 VERIFY(nvlist_remove(sav->sav_config, config,
3357 DATA_TYPE_NVLIST_ARRAY) == 0);
3359 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3360 config, newdevs, ndevs + oldndevs) == 0);
3361 for (i = 0; i < oldndevs + ndevs; i++)
3362 nvlist_free(newdevs[i]);
3363 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
3366 * Generate a new dev list.
3368 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
3370 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
3376 * Stop and drop level 2 ARC devices
3379 spa_l2cache_drop(spa_t *spa)
3383 spa_aux_vdev_t *sav = &spa->spa_l2cache;
3385 for (i = 0; i < sav->sav_count; i++) {
3388 vd = sav->sav_vdevs[i];
3391 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
3392 pool != 0ULL && l2arc_vdev_present(vd))
3393 l2arc_remove_vdev(vd);
3401 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
3402 const char *history_str, nvlist_t *zplprops)
3405 char *altroot = NULL;
3410 uint64_t txg = TXG_INITIAL;
3411 nvlist_t **spares, **l2cache;
3412 uint_t nspares, nl2cache;
3413 uint64_t version, obj;
3414 boolean_t has_features;
3419 * If this pool already exists, return failure.
3421 mutex_enter(&spa_namespace_lock);
3422 if (spa_lookup(pool) != NULL) {
3423 mutex_exit(&spa_namespace_lock);
3428 * Allocate a new spa_t structure.
3430 (void) nvlist_lookup_string(props,
3431 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3432 spa = spa_add(pool, NULL, altroot);
3433 spa_activate(spa, spa_mode_global);
3435 if (props && (error = spa_prop_validate(spa, props))) {
3436 spa_deactivate(spa);
3438 mutex_exit(&spa_namespace_lock);
3442 has_features = B_FALSE;
3443 for (elem = nvlist_next_nvpair(props, NULL);
3444 elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
3445 if (zpool_prop_feature(nvpair_name(elem)))
3446 has_features = B_TRUE;
3449 if (has_features || nvlist_lookup_uint64(props,
3450 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
3451 version = SPA_VERSION;
3453 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
3455 spa->spa_first_txg = txg;
3456 spa->spa_uberblock.ub_txg = txg - 1;
3457 spa->spa_uberblock.ub_version = version;
3458 spa->spa_ubsync = spa->spa_uberblock;
3461 * Create "The Godfather" zio to hold all async IOs
3463 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
3464 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
3467 * Create the root vdev.
3469 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3471 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3473 ASSERT(error != 0 || rvd != NULL);
3474 ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3476 if (error == 0 && !zfs_allocatable_devs(nvroot))
3480 (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3481 (error = spa_validate_aux(spa, nvroot, txg,
3482 VDEV_ALLOC_ADD)) == 0) {
3483 for (c = 0; c < rvd->vdev_children; c++) {
3484 vdev_metaslab_set_size(rvd->vdev_child[c]);
3485 vdev_expand(rvd->vdev_child[c], txg);
3489 spa_config_exit(spa, SCL_ALL, FTAG);
3493 spa_deactivate(spa);
3495 mutex_exit(&spa_namespace_lock);
3500 * Get the list of spares, if specified.
3502 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3503 &spares, &nspares) == 0) {
3504 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3506 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3507 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3508 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3509 spa_load_spares(spa);
3510 spa_config_exit(spa, SCL_ALL, FTAG);
3511 spa->spa_spares.sav_sync = B_TRUE;
3515 * Get the list of level 2 cache devices, if specified.
3517 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3518 &l2cache, &nl2cache) == 0) {
3519 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3520 NV_UNIQUE_NAME, KM_PUSHPAGE) == 0);
3521 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3522 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3523 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3524 spa_load_l2cache(spa);
3525 spa_config_exit(spa, SCL_ALL, FTAG);
3526 spa->spa_l2cache.sav_sync = B_TRUE;
3529 spa->spa_is_initializing = B_TRUE;
3530 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3531 spa->spa_meta_objset = dp->dp_meta_objset;
3532 spa->spa_is_initializing = B_FALSE;
3535 * Create DDTs (dedup tables).
3539 spa_update_dspace(spa);
3541 tx = dmu_tx_create_assigned(dp, txg);
3544 * Create the pool config object.
3546 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3547 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3548 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3550 if (zap_add(spa->spa_meta_objset,
3551 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3552 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3553 cmn_err(CE_PANIC, "failed to add pool config");
3556 if (spa_version(spa) >= SPA_VERSION_FEATURES)
3557 spa_feature_create_zap_objects(spa, tx);
3559 if (zap_add(spa->spa_meta_objset,
3560 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3561 sizeof (uint64_t), 1, &version, tx) != 0) {
3562 cmn_err(CE_PANIC, "failed to add pool version");
3565 /* Newly created pools with the right version are always deflated. */
3566 if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3567 spa->spa_deflate = TRUE;
3568 if (zap_add(spa->spa_meta_objset,
3569 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3570 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3571 cmn_err(CE_PANIC, "failed to add deflate");
3576 * Create the deferred-free bpobj. Turn off compression
3577 * because sync-to-convergence takes longer if the blocksize
3580 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3581 dmu_object_set_compress(spa->spa_meta_objset, obj,
3582 ZIO_COMPRESS_OFF, tx);
3583 if (zap_add(spa->spa_meta_objset,
3584 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3585 sizeof (uint64_t), 1, &obj, tx) != 0) {
3586 cmn_err(CE_PANIC, "failed to add bpobj");
3588 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3589 spa->spa_meta_objset, obj));
3592 * Create the pool's history object.
3594 if (version >= SPA_VERSION_ZPOOL_HISTORY)
3595 spa_history_create_obj(spa, tx);
3598 * Set pool properties.
3600 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3601 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3602 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3603 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3605 if (props != NULL) {
3606 spa_configfile_set(spa, props, B_FALSE);
3607 spa_sync_props(spa, props, tx);
3612 spa->spa_sync_on = B_TRUE;
3613 txg_sync_start(spa->spa_dsl_pool);
3616 * We explicitly wait for the first transaction to complete so that our
3617 * bean counters are appropriately updated.
3619 txg_wait_synced(spa->spa_dsl_pool, txg);
3621 spa_config_sync(spa, B_FALSE, B_TRUE);
3623 if (version >= SPA_VERSION_ZPOOL_HISTORY && history_str != NULL)
3624 (void) spa_history_log(spa, history_str, LOG_CMD_POOL_CREATE);
3625 spa_history_log_version(spa, LOG_POOL_CREATE);
3627 spa->spa_minref = refcount_count(&spa->spa_refcount);
3629 mutex_exit(&spa_namespace_lock);
3636 * Get the root pool information from the root disk, then import the root pool
3637 * during the system boot up time.
3639 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3642 spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3645 nvlist_t *nvtop, *nvroot;
3648 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3652 * Add this top-level vdev to the child array.
3654 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3656 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3658 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3661 * Put this pool's top-level vdevs into a root vdev.
3663 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_PUSHPAGE) == 0);
3664 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3665 VDEV_TYPE_ROOT) == 0);
3666 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3667 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3668 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3672 * Replace the existing vdev_tree with the new root vdev in
3673 * this pool's configuration (remove the old, add the new).
3675 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3676 nvlist_free(nvroot);
3681 * Walk the vdev tree and see if we can find a device with "better"
3682 * configuration. A configuration is "better" if the label on that
3683 * device has a more recent txg.
3686 spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3690 for (c = 0; c < vd->vdev_children; c++)
3691 spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3693 if (vd->vdev_ops->vdev_op_leaf) {
3697 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3701 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3705 * Do we have a better boot device?
3707 if (label_txg > *txg) {
3716 * Import a root pool.
3718 * For x86. devpath_list will consist of devid and/or physpath name of
3719 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3720 * The GRUB "findroot" command will return the vdev we should boot.
3722 * For Sparc, devpath_list consists the physpath name of the booting device
3723 * no matter the rootpool is a single device pool or a mirrored pool.
3725 * "/pci@1f,0/ide@d/disk@0,0:a"
3728 spa_import_rootpool(char *devpath, char *devid)
3731 vdev_t *rvd, *bvd, *avd = NULL;
3732 nvlist_t *config, *nvtop;
3738 * Read the label from the boot device and generate a configuration.
3740 config = spa_generate_rootconf(devpath, devid, &guid);
3741 #if defined(_OBP) && defined(_KERNEL)
3742 if (config == NULL) {
3743 if (strstr(devpath, "/iscsi/ssd") != NULL) {
3745 get_iscsi_bootpath_phy(devpath);
3746 config = spa_generate_rootconf(devpath, devid, &guid);
3750 if (config == NULL) {
3751 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
3756 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3758 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3760 mutex_enter(&spa_namespace_lock);
3761 if ((spa = spa_lookup(pname)) != NULL) {
3763 * Remove the existing root pool from the namespace so that we
3764 * can replace it with the correct config we just read in.
3769 spa = spa_add(pname, config, NULL);
3770 spa->spa_is_root = B_TRUE;
3771 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3774 * Build up a vdev tree based on the boot device's label config.
3776 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3778 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3779 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3780 VDEV_ALLOC_ROOTPOOL);
3781 spa_config_exit(spa, SCL_ALL, FTAG);
3783 mutex_exit(&spa_namespace_lock);
3784 nvlist_free(config);
3785 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3791 * Get the boot vdev.
3793 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3794 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3795 (u_longlong_t)guid);
3801 * Determine if there is a better boot device.
3804 spa_alt_rootvdev(rvd, &avd, &txg);
3806 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3807 "try booting from '%s'", avd->vdev_path);
3813 * If the boot device is part of a spare vdev then ensure that
3814 * we're booting off the active spare.
3816 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3817 !bvd->vdev_isspare) {
3818 cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3819 "try booting from '%s'",
3821 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3827 spa_history_log_version(spa, LOG_POOL_IMPORT);
3829 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3831 spa_config_exit(spa, SCL_ALL, FTAG);
3832 mutex_exit(&spa_namespace_lock);
3834 nvlist_free(config);
3841 * Import a non-root pool into the system.
3844 spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
3847 char *altroot = NULL;
3848 spa_load_state_t state = SPA_LOAD_IMPORT;
3849 zpool_rewind_policy_t policy;
3850 uint64_t mode = spa_mode_global;
3851 uint64_t readonly = B_FALSE;
3854 nvlist_t **spares, **l2cache;
3855 uint_t nspares, nl2cache;
3858 * If a pool with this name exists, return failure.
3860 mutex_enter(&spa_namespace_lock);
3861 if (spa_lookup(pool) != NULL) {
3862 mutex_exit(&spa_namespace_lock);
3867 * Create and initialize the spa structure.
3869 (void) nvlist_lookup_string(props,
3870 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3871 (void) nvlist_lookup_uint64(props,
3872 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
3875 spa = spa_add(pool, config, altroot);
3876 spa->spa_import_flags = flags;
3879 * Verbatim import - Take a pool and insert it into the namespace
3880 * as if it had been loaded at boot.
3882 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
3884 spa_configfile_set(spa, props, B_FALSE);
3886 spa_config_sync(spa, B_FALSE, B_TRUE);
3888 mutex_exit(&spa_namespace_lock);
3889 spa_history_log_version(spa, LOG_POOL_IMPORT);
3894 spa_activate(spa, mode);
3897 * Don't start async tasks until we know everything is healthy.
3899 spa_async_suspend(spa);
3901 zpool_get_rewind_policy(config, &policy);
3902 if (policy.zrp_request & ZPOOL_DO_REWIND)
3903 state = SPA_LOAD_RECOVER;
3906 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
3907 * because the user-supplied config is actually the one to trust when
3910 if (state != SPA_LOAD_RECOVER)
3911 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3913 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
3914 policy.zrp_request);
3917 * Propagate anything learned while loading the pool and pass it
3918 * back to caller (i.e. rewind info, missing devices, etc).
3920 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
3921 spa->spa_load_info) == 0);
3923 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3925 * Toss any existing sparelist, as it doesn't have any validity
3926 * anymore, and conflicts with spa_has_spare().
3928 if (spa->spa_spares.sav_config) {
3929 nvlist_free(spa->spa_spares.sav_config);
3930 spa->spa_spares.sav_config = NULL;
3931 spa_load_spares(spa);
3933 if (spa->spa_l2cache.sav_config) {
3934 nvlist_free(spa->spa_l2cache.sav_config);
3935 spa->spa_l2cache.sav_config = NULL;
3936 spa_load_l2cache(spa);
3939 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3942 error = spa_validate_aux(spa, nvroot, -1ULL,
3945 error = spa_validate_aux(spa, nvroot, -1ULL,
3946 VDEV_ALLOC_L2CACHE);
3947 spa_config_exit(spa, SCL_ALL, FTAG);
3950 spa_configfile_set(spa, props, B_FALSE);
3952 if (error != 0 || (props && spa_writeable(spa) &&
3953 (error = spa_prop_set(spa, props)))) {
3955 spa_deactivate(spa);
3957 mutex_exit(&spa_namespace_lock);
3961 spa_async_resume(spa);
3964 * Override any spares and level 2 cache devices as specified by
3965 * the user, as these may have correct device names/devids, etc.
3967 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3968 &spares, &nspares) == 0) {
3969 if (spa->spa_spares.sav_config)
3970 VERIFY(nvlist_remove(spa->spa_spares.sav_config,
3971 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
3973 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
3974 NV_UNIQUE_NAME, KM_PUSHPAGE) == 0);
3975 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3976 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3977 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3978 spa_load_spares(spa);
3979 spa_config_exit(spa, SCL_ALL, FTAG);
3980 spa->spa_spares.sav_sync = B_TRUE;
3982 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3983 &l2cache, &nl2cache) == 0) {
3984 if (spa->spa_l2cache.sav_config)
3985 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
3986 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
3988 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3989 NV_UNIQUE_NAME, KM_PUSHPAGE) == 0);
3990 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3991 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3992 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3993 spa_load_l2cache(spa);
3994 spa_config_exit(spa, SCL_ALL, FTAG);
3995 spa->spa_l2cache.sav_sync = B_TRUE;
3999 * Check for any removed devices.
4001 if (spa->spa_autoreplace) {
4002 spa_aux_check_removed(&spa->spa_spares);
4003 spa_aux_check_removed(&spa->spa_l2cache);
4006 if (spa_writeable(spa)) {
4008 * Update the config cache to include the newly-imported pool.
4010 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4014 * It's possible that the pool was expanded while it was exported.
4015 * We kick off an async task to handle this for us.
4017 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
4019 mutex_exit(&spa_namespace_lock);
4020 spa_history_log_version(spa, LOG_POOL_IMPORT);
4023 zvol_create_minors(pool);
4030 spa_tryimport(nvlist_t *tryconfig)
4032 nvlist_t *config = NULL;
4038 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
4041 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
4045 * Create and initialize the spa structure.
4047 mutex_enter(&spa_namespace_lock);
4048 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
4049 spa_activate(spa, FREAD);
4052 * Pass off the heavy lifting to spa_load().
4053 * Pass TRUE for mosconfig because the user-supplied config
4054 * is actually the one to trust when doing an import.
4056 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
4059 * If 'tryconfig' was at least parsable, return the current config.
4061 if (spa->spa_root_vdev != NULL) {
4062 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4063 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
4065 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4067 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
4068 spa->spa_uberblock.ub_timestamp) == 0);
4069 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4070 spa->spa_load_info) == 0);
4073 * If the bootfs property exists on this pool then we
4074 * copy it out so that external consumers can tell which
4075 * pools are bootable.
4077 if ((!error || error == EEXIST) && spa->spa_bootfs) {
4078 char *tmpname = kmem_alloc(MAXPATHLEN, KM_PUSHPAGE);
4081 * We have to play games with the name since the
4082 * pool was opened as TRYIMPORT_NAME.
4084 if (dsl_dsobj_to_dsname(spa_name(spa),
4085 spa->spa_bootfs, tmpname) == 0) {
4087 char *dsname = kmem_alloc(MAXPATHLEN, KM_PUSHPAGE);
4089 cp = strchr(tmpname, '/');
4091 (void) strlcpy(dsname, tmpname,
4094 (void) snprintf(dsname, MAXPATHLEN,
4095 "%s/%s", poolname, ++cp);
4097 VERIFY(nvlist_add_string(config,
4098 ZPOOL_CONFIG_BOOTFS, dsname) == 0);
4099 kmem_free(dsname, MAXPATHLEN);
4101 kmem_free(tmpname, MAXPATHLEN);
4105 * Add the list of hot spares and level 2 cache devices.
4107 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4108 spa_add_spares(spa, config);
4109 spa_add_l2cache(spa, config);
4110 spa_config_exit(spa, SCL_CONFIG, FTAG);
4114 spa_deactivate(spa);
4116 mutex_exit(&spa_namespace_lock);
4122 * Pool export/destroy
4124 * The act of destroying or exporting a pool is very simple. We make sure there
4125 * is no more pending I/O and any references to the pool are gone. Then, we
4126 * update the pool state and sync all the labels to disk, removing the
4127 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4128 * we don't sync the labels or remove the configuration cache.
4131 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
4132 boolean_t force, boolean_t hardforce)
4139 if (!(spa_mode_global & FWRITE))
4142 mutex_enter(&spa_namespace_lock);
4143 if ((spa = spa_lookup(pool)) == NULL) {
4144 mutex_exit(&spa_namespace_lock);
4149 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4150 * reacquire the namespace lock, and see if we can export.
4152 spa_open_ref(spa, FTAG);
4153 mutex_exit(&spa_namespace_lock);
4154 spa_async_suspend(spa);
4155 mutex_enter(&spa_namespace_lock);
4156 spa_close(spa, FTAG);
4159 * The pool will be in core if it's openable,
4160 * in which case we can modify its state.
4162 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
4164 * Objsets may be open only because they're dirty, so we
4165 * have to force it to sync before checking spa_refcnt.
4167 txg_wait_synced(spa->spa_dsl_pool, 0);
4170 * A pool cannot be exported or destroyed if there are active
4171 * references. If we are resetting a pool, allow references by
4172 * fault injection handlers.
4174 if (!spa_refcount_zero(spa) ||
4175 (spa->spa_inject_ref != 0 &&
4176 new_state != POOL_STATE_UNINITIALIZED)) {
4177 spa_async_resume(spa);
4178 mutex_exit(&spa_namespace_lock);
4183 * A pool cannot be exported if it has an active shared spare.
4184 * This is to prevent other pools stealing the active spare
4185 * from an exported pool. At user's own will, such pool can
4186 * be forcedly exported.
4188 if (!force && new_state == POOL_STATE_EXPORTED &&
4189 spa_has_active_shared_spare(spa)) {
4190 spa_async_resume(spa);
4191 mutex_exit(&spa_namespace_lock);
4196 * We want this to be reflected on every label,
4197 * so mark them all dirty. spa_unload() will do the
4198 * final sync that pushes these changes out.
4200 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
4201 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4202 spa->spa_state = new_state;
4203 spa->spa_final_txg = spa_last_synced_txg(spa) +
4205 vdev_config_dirty(spa->spa_root_vdev);
4206 spa_config_exit(spa, SCL_ALL, FTAG);
4210 spa_event_notify(spa, NULL, FM_EREPORT_ZFS_POOL_DESTROY);
4212 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4214 spa_deactivate(spa);
4217 if (oldconfig && spa->spa_config)
4218 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
4220 if (new_state != POOL_STATE_UNINITIALIZED) {
4222 spa_config_sync(spa, B_TRUE, B_TRUE);
4225 mutex_exit(&spa_namespace_lock);
4231 * Destroy a storage pool.
4234 spa_destroy(char *pool)
4236 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
4241 * Export a storage pool.
4244 spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
4245 boolean_t hardforce)
4247 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
4252 * Similar to spa_export(), this unloads the spa_t without actually removing it
4253 * from the namespace in any way.
4256 spa_reset(char *pool)
4258 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
4263 * ==========================================================================
4264 * Device manipulation
4265 * ==========================================================================
4269 * Add a device to a storage pool.
4272 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
4276 vdev_t *rvd = spa->spa_root_vdev;
4278 nvlist_t **spares, **l2cache;
4279 uint_t nspares, nl2cache;
4282 ASSERT(spa_writeable(spa));
4284 txg = spa_vdev_enter(spa);
4286 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
4287 VDEV_ALLOC_ADD)) != 0)
4288 return (spa_vdev_exit(spa, NULL, txg, error));
4290 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */
4292 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
4296 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
4300 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
4301 return (spa_vdev_exit(spa, vd, txg, EINVAL));
4303 if (vd->vdev_children != 0 &&
4304 (error = vdev_create(vd, txg, B_FALSE)) != 0)
4305 return (spa_vdev_exit(spa, vd, txg, error));
4308 * We must validate the spares and l2cache devices after checking the
4309 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
4311 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
4312 return (spa_vdev_exit(spa, vd, txg, error));
4315 * Transfer each new top-level vdev from vd to rvd.
4317 for (c = 0; c < vd->vdev_children; c++) {
4320 * Set the vdev id to the first hole, if one exists.
4322 for (id = 0; id < rvd->vdev_children; id++) {
4323 if (rvd->vdev_child[id]->vdev_ishole) {
4324 vdev_free(rvd->vdev_child[id]);
4328 tvd = vd->vdev_child[c];
4329 vdev_remove_child(vd, tvd);
4331 vdev_add_child(rvd, tvd);
4332 vdev_config_dirty(tvd);
4336 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
4337 ZPOOL_CONFIG_SPARES);
4338 spa_load_spares(spa);
4339 spa->spa_spares.sav_sync = B_TRUE;
4342 if (nl2cache != 0) {
4343 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
4344 ZPOOL_CONFIG_L2CACHE);
4345 spa_load_l2cache(spa);
4346 spa->spa_l2cache.sav_sync = B_TRUE;
4350 * We have to be careful when adding new vdevs to an existing pool.
4351 * If other threads start allocating from these vdevs before we
4352 * sync the config cache, and we lose power, then upon reboot we may
4353 * fail to open the pool because there are DVAs that the config cache
4354 * can't translate. Therefore, we first add the vdevs without
4355 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4356 * and then let spa_config_update() initialize the new metaslabs.
4358 * spa_load() checks for added-but-not-initialized vdevs, so that
4359 * if we lose power at any point in this sequence, the remaining
4360 * steps will be completed the next time we load the pool.
4362 (void) spa_vdev_exit(spa, vd, txg, 0);
4364 mutex_enter(&spa_namespace_lock);
4365 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4366 mutex_exit(&spa_namespace_lock);
4372 * Attach a device to a mirror. The arguments are the path to any device
4373 * in the mirror, and the nvroot for the new device. If the path specifies
4374 * a device that is not mirrored, we automatically insert the mirror vdev.
4376 * If 'replacing' is specified, the new device is intended to replace the
4377 * existing device; in this case the two devices are made into their own
4378 * mirror using the 'replacing' vdev, which is functionally identical to
4379 * the mirror vdev (it actually reuses all the same ops) but has a few
4380 * extra rules: you can't attach to it after it's been created, and upon
4381 * completion of resilvering, the first disk (the one being replaced)
4382 * is automatically detached.
4385 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
4387 uint64_t txg, dtl_max_txg;
4388 ASSERTV(vdev_t *rvd = spa->spa_root_vdev;)
4389 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
4391 char *oldvdpath, *newvdpath;
4395 ASSERT(spa_writeable(spa));
4397 txg = spa_vdev_enter(spa);
4399 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
4402 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4404 if (!oldvd->vdev_ops->vdev_op_leaf)
4405 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4407 pvd = oldvd->vdev_parent;
4409 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
4410 VDEV_ALLOC_ATTACH)) != 0)
4411 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4413 if (newrootvd->vdev_children != 1)
4414 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4416 newvd = newrootvd->vdev_child[0];
4418 if (!newvd->vdev_ops->vdev_op_leaf)
4419 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4421 if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
4422 return (spa_vdev_exit(spa, newrootvd, txg, error));
4425 * Spares can't replace logs
4427 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
4428 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4432 * For attach, the only allowable parent is a mirror or the root
4435 if (pvd->vdev_ops != &vdev_mirror_ops &&
4436 pvd->vdev_ops != &vdev_root_ops)
4437 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4439 pvops = &vdev_mirror_ops;
4442 * Active hot spares can only be replaced by inactive hot
4445 if (pvd->vdev_ops == &vdev_spare_ops &&
4446 oldvd->vdev_isspare &&
4447 !spa_has_spare(spa, newvd->vdev_guid))
4448 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4451 * If the source is a hot spare, and the parent isn't already a
4452 * spare, then we want to create a new hot spare. Otherwise, we
4453 * want to create a replacing vdev. The user is not allowed to
4454 * attach to a spared vdev child unless the 'isspare' state is
4455 * the same (spare replaces spare, non-spare replaces
4458 if (pvd->vdev_ops == &vdev_replacing_ops &&
4459 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
4460 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4461 } else if (pvd->vdev_ops == &vdev_spare_ops &&
4462 newvd->vdev_isspare != oldvd->vdev_isspare) {
4463 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4466 if (newvd->vdev_isspare)
4467 pvops = &vdev_spare_ops;
4469 pvops = &vdev_replacing_ops;
4473 * Make sure the new device is big enough.
4475 if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
4476 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
4479 * The new device cannot have a higher alignment requirement
4480 * than the top-level vdev.
4482 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4483 return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4486 * If this is an in-place replacement, update oldvd's path and devid
4487 * to make it distinguishable from newvd, and unopenable from now on.
4489 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4490 spa_strfree(oldvd->vdev_path);
4491 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4493 (void) sprintf(oldvd->vdev_path, "%s/%s",
4494 newvd->vdev_path, "old");
4495 if (oldvd->vdev_devid != NULL) {
4496 spa_strfree(oldvd->vdev_devid);
4497 oldvd->vdev_devid = NULL;
4501 /* mark the device being resilvered */
4502 newvd->vdev_resilvering = B_TRUE;
4505 * If the parent is not a mirror, or if we're replacing, insert the new
4506 * mirror/replacing/spare vdev above oldvd.
4508 if (pvd->vdev_ops != pvops)
4509 pvd = vdev_add_parent(oldvd, pvops);
4511 ASSERT(pvd->vdev_top->vdev_parent == rvd);
4512 ASSERT(pvd->vdev_ops == pvops);
4513 ASSERT(oldvd->vdev_parent == pvd);
4516 * Extract the new device from its root and add it to pvd.
4518 vdev_remove_child(newrootvd, newvd);
4519 newvd->vdev_id = pvd->vdev_children;
4520 newvd->vdev_crtxg = oldvd->vdev_crtxg;
4521 vdev_add_child(pvd, newvd);
4523 tvd = newvd->vdev_top;
4524 ASSERT(pvd->vdev_top == tvd);
4525 ASSERT(tvd->vdev_parent == rvd);
4527 vdev_config_dirty(tvd);
4530 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4531 * for any dmu_sync-ed blocks. It will propagate upward when
4532 * spa_vdev_exit() calls vdev_dtl_reassess().
4534 dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4536 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4537 dtl_max_txg - TXG_INITIAL);
4539 if (newvd->vdev_isspare) {
4540 spa_spare_activate(newvd);
4541 spa_event_notify(spa, newvd, FM_EREPORT_ZFS_DEVICE_SPARE);
4544 oldvdpath = spa_strdup(oldvd->vdev_path);
4545 newvdpath = spa_strdup(newvd->vdev_path);
4546 newvd_isspare = newvd->vdev_isspare;
4549 * Mark newvd's DTL dirty in this txg.
4551 vdev_dirty(tvd, VDD_DTL, newvd, txg);
4554 * Restart the resilver
4556 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4561 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4563 spa_history_log_internal(LOG_POOL_VDEV_ATTACH, spa, NULL,
4564 "%s vdev=%s %s vdev=%s",
4565 replacing && newvd_isspare ? "spare in" :
4566 replacing ? "replace" : "attach", newvdpath,
4567 replacing ? "for" : "to", oldvdpath);
4569 spa_strfree(oldvdpath);
4570 spa_strfree(newvdpath);
4572 if (spa->spa_bootfs)
4573 spa_event_notify(spa, newvd, FM_EREPORT_ZFS_BOOTFS_VDEV_ATTACH);
4579 * Detach a device from a mirror or replacing vdev.
4580 * If 'replace_done' is specified, only detach if the parent
4581 * is a replacing vdev.
4584 spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4588 ASSERTV(vdev_t *rvd = spa->spa_root_vdev;)
4589 vdev_t *vd, *pvd, *cvd, *tvd;
4590 boolean_t unspare = B_FALSE;
4591 uint64_t unspare_guid = 0;
4595 ASSERT(spa_writeable(spa));
4597 txg = spa_vdev_enter(spa);
4599 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4602 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4604 if (!vd->vdev_ops->vdev_op_leaf)
4605 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4607 pvd = vd->vdev_parent;
4610 * If the parent/child relationship is not as expected, don't do it.
4611 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4612 * vdev that's replacing B with C. The user's intent in replacing
4613 * is to go from M(A,B) to M(A,C). If the user decides to cancel
4614 * the replace by detaching C, the expected behavior is to end up
4615 * M(A,B). But suppose that right after deciding to detach C,
4616 * the replacement of B completes. We would have M(A,C), and then
4617 * ask to detach C, which would leave us with just A -- not what
4618 * the user wanted. To prevent this, we make sure that the
4619 * parent/child relationship hasn't changed -- in this example,
4620 * that C's parent is still the replacing vdev R.
4622 if (pvd->vdev_guid != pguid && pguid != 0)
4623 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4626 * Only 'replacing' or 'spare' vdevs can be replaced.
4628 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4629 pvd->vdev_ops != &vdev_spare_ops)
4630 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4632 ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4633 spa_version(spa) >= SPA_VERSION_SPARES);
4636 * Only mirror, replacing, and spare vdevs support detach.
4638 if (pvd->vdev_ops != &vdev_replacing_ops &&
4639 pvd->vdev_ops != &vdev_mirror_ops &&
4640 pvd->vdev_ops != &vdev_spare_ops)
4641 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4644 * If this device has the only valid copy of some data,
4645 * we cannot safely detach it.
4647 if (vdev_dtl_required(vd))
4648 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4650 ASSERT(pvd->vdev_children >= 2);
4653 * If we are detaching the second disk from a replacing vdev, then
4654 * check to see if we changed the original vdev's path to have "/old"
4655 * at the end in spa_vdev_attach(). If so, undo that change now.
4657 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4658 vd->vdev_path != NULL) {
4659 size_t len = strlen(vd->vdev_path);
4661 for (c = 0; c < pvd->vdev_children; c++) {
4662 cvd = pvd->vdev_child[c];
4664 if (cvd == vd || cvd->vdev_path == NULL)
4667 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4668 strcmp(cvd->vdev_path + len, "/old") == 0) {
4669 spa_strfree(cvd->vdev_path);
4670 cvd->vdev_path = spa_strdup(vd->vdev_path);
4677 * If we are detaching the original disk from a spare, then it implies
4678 * that the spare should become a real disk, and be removed from the
4679 * active spare list for the pool.
4681 if (pvd->vdev_ops == &vdev_spare_ops &&
4683 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4687 * Erase the disk labels so the disk can be used for other things.
4688 * This must be done after all other error cases are handled,
4689 * but before we disembowel vd (so we can still do I/O to it).
4690 * But if we can't do it, don't treat the error as fatal --
4691 * it may be that the unwritability of the disk is the reason
4692 * it's being detached!
4694 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4697 * Remove vd from its parent and compact the parent's children.
4699 vdev_remove_child(pvd, vd);
4700 vdev_compact_children(pvd);
4703 * Remember one of the remaining children so we can get tvd below.
4705 cvd = pvd->vdev_child[pvd->vdev_children - 1];
4708 * If we need to remove the remaining child from the list of hot spares,
4709 * do it now, marking the vdev as no longer a spare in the process.
4710 * We must do this before vdev_remove_parent(), because that can
4711 * change the GUID if it creates a new toplevel GUID. For a similar
4712 * reason, we must remove the spare now, in the same txg as the detach;
4713 * otherwise someone could attach a new sibling, change the GUID, and
4714 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4717 ASSERT(cvd->vdev_isspare);
4718 spa_spare_remove(cvd);
4719 unspare_guid = cvd->vdev_guid;
4720 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4721 cvd->vdev_unspare = B_TRUE;
4725 * If the parent mirror/replacing vdev only has one child,
4726 * the parent is no longer needed. Remove it from the tree.
4728 if (pvd->vdev_children == 1) {
4729 if (pvd->vdev_ops == &vdev_spare_ops)
4730 cvd->vdev_unspare = B_FALSE;
4731 vdev_remove_parent(cvd);
4732 cvd->vdev_resilvering = B_FALSE;
4737 * We don't set tvd until now because the parent we just removed
4738 * may have been the previous top-level vdev.
4740 tvd = cvd->vdev_top;
4741 ASSERT(tvd->vdev_parent == rvd);
4744 * Reevaluate the parent vdev state.
4746 vdev_propagate_state(cvd);
4749 * If the 'autoexpand' property is set on the pool then automatically
4750 * try to expand the size of the pool. For example if the device we
4751 * just detached was smaller than the others, it may be possible to
4752 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4753 * first so that we can obtain the updated sizes of the leaf vdevs.
4755 if (spa->spa_autoexpand) {
4757 vdev_expand(tvd, txg);
4760 vdev_config_dirty(tvd);
4763 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
4764 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4765 * But first make sure we're not on any *other* txg's DTL list, to
4766 * prevent vd from being accessed after it's freed.
4768 vdpath = spa_strdup(vd->vdev_path);
4769 for (t = 0; t < TXG_SIZE; t++)
4770 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4771 vd->vdev_detached = B_TRUE;
4772 vdev_dirty(tvd, VDD_DTL, vd, txg);
4774 spa_event_notify(spa, vd, FM_EREPORT_ZFS_DEVICE_REMOVE);
4776 /* hang on to the spa before we release the lock */
4777 spa_open_ref(spa, FTAG);
4779 error = spa_vdev_exit(spa, vd, txg, 0);
4781 spa_history_log_internal(LOG_POOL_VDEV_DETACH, spa, NULL,
4783 spa_strfree(vdpath);
4786 * If this was the removal of the original device in a hot spare vdev,
4787 * then we want to go through and remove the device from the hot spare
4788 * list of every other pool.
4791 spa_t *altspa = NULL;
4793 mutex_enter(&spa_namespace_lock);
4794 while ((altspa = spa_next(altspa)) != NULL) {
4795 if (altspa->spa_state != POOL_STATE_ACTIVE ||
4799 spa_open_ref(altspa, FTAG);
4800 mutex_exit(&spa_namespace_lock);
4801 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
4802 mutex_enter(&spa_namespace_lock);
4803 spa_close(altspa, FTAG);
4805 mutex_exit(&spa_namespace_lock);
4807 /* search the rest of the vdevs for spares to remove */
4808 spa_vdev_resilver_done(spa);
4811 /* all done with the spa; OK to release */
4812 mutex_enter(&spa_namespace_lock);
4813 spa_close(spa, FTAG);
4814 mutex_exit(&spa_namespace_lock);
4820 * Split a set of devices from their mirrors, and create a new pool from them.
4823 spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
4824 nvlist_t *props, boolean_t exp)
4827 uint64_t txg, *glist;
4829 uint_t c, children, lastlog;
4830 nvlist_t **child, *nvl, *tmp;
4832 char *altroot = NULL;
4833 vdev_t *rvd, **vml = NULL; /* vdev modify list */
4834 boolean_t activate_slog;
4836 ASSERT(spa_writeable(spa));
4838 txg = spa_vdev_enter(spa);
4840 /* clear the log and flush everything up to now */
4841 activate_slog = spa_passivate_log(spa);
4842 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4843 error = spa_offline_log(spa);
4844 txg = spa_vdev_config_enter(spa);
4847 spa_activate_log(spa);
4850 return (spa_vdev_exit(spa, NULL, txg, error));
4852 /* check new spa name before going any further */
4853 if (spa_lookup(newname) != NULL)
4854 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
4857 * scan through all the children to ensure they're all mirrors
4859 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
4860 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
4862 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4864 /* first, check to ensure we've got the right child count */
4865 rvd = spa->spa_root_vdev;
4867 for (c = 0; c < rvd->vdev_children; c++) {
4868 vdev_t *vd = rvd->vdev_child[c];
4870 /* don't count the holes & logs as children */
4871 if (vd->vdev_islog || vd->vdev_ishole) {
4879 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
4880 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4882 /* next, ensure no spare or cache devices are part of the split */
4883 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
4884 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
4885 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4887 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_PUSHPAGE);
4888 glist = kmem_zalloc(children * sizeof (uint64_t), KM_PUSHPAGE);
4890 /* then, loop over each vdev and validate it */
4891 for (c = 0; c < children; c++) {
4892 uint64_t is_hole = 0;
4894 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
4898 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
4899 spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
4907 /* which disk is going to be split? */
4908 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
4914 /* look it up in the spa */
4915 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
4916 if (vml[c] == NULL) {
4921 /* make sure there's nothing stopping the split */
4922 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
4923 vml[c]->vdev_islog ||
4924 vml[c]->vdev_ishole ||
4925 vml[c]->vdev_isspare ||
4926 vml[c]->vdev_isl2cache ||
4927 !vdev_writeable(vml[c]) ||
4928 vml[c]->vdev_children != 0 ||
4929 vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
4930 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
4935 if (vdev_dtl_required(vml[c])) {
4940 /* we need certain info from the top level */
4941 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
4942 vml[c]->vdev_top->vdev_ms_array) == 0);
4943 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
4944 vml[c]->vdev_top->vdev_ms_shift) == 0);
4945 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
4946 vml[c]->vdev_top->vdev_asize) == 0);
4947 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
4948 vml[c]->vdev_top->vdev_ashift) == 0);
4952 kmem_free(vml, children * sizeof (vdev_t *));
4953 kmem_free(glist, children * sizeof (uint64_t));
4954 return (spa_vdev_exit(spa, NULL, txg, error));
4957 /* stop writers from using the disks */
4958 for (c = 0; c < children; c++) {
4960 vml[c]->vdev_offline = B_TRUE;
4962 vdev_reopen(spa->spa_root_vdev);
4965 * Temporarily record the splitting vdevs in the spa config. This
4966 * will disappear once the config is regenerated.
4968 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_PUSHPAGE) == 0);
4969 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
4970 glist, children) == 0);
4971 kmem_free(glist, children * sizeof (uint64_t));
4973 mutex_enter(&spa->spa_props_lock);
4974 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
4976 mutex_exit(&spa->spa_props_lock);
4977 spa->spa_config_splitting = nvl;
4978 vdev_config_dirty(spa->spa_root_vdev);
4980 /* configure and create the new pool */
4981 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
4982 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4983 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
4984 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
4985 spa_version(spa)) == 0);
4986 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
4987 spa->spa_config_txg) == 0);
4988 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4989 spa_generate_guid(NULL)) == 0);
4990 (void) nvlist_lookup_string(props,
4991 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4993 /* add the new pool to the namespace */
4994 newspa = spa_add(newname, config, altroot);
4995 newspa->spa_config_txg = spa->spa_config_txg;
4996 spa_set_log_state(newspa, SPA_LOG_CLEAR);
4998 /* release the spa config lock, retaining the namespace lock */
4999 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5001 if (zio_injection_enabled)
5002 zio_handle_panic_injection(spa, FTAG, 1);
5004 spa_activate(newspa, spa_mode_global);
5005 spa_async_suspend(newspa);
5007 /* create the new pool from the disks of the original pool */
5008 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
5012 /* if that worked, generate a real config for the new pool */
5013 if (newspa->spa_root_vdev != NULL) {
5014 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
5015 NV_UNIQUE_NAME, KM_PUSHPAGE) == 0);
5016 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
5017 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
5018 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
5023 if (props != NULL) {
5024 spa_configfile_set(newspa, props, B_FALSE);
5025 error = spa_prop_set(newspa, props);
5030 /* flush everything */
5031 txg = spa_vdev_config_enter(newspa);
5032 vdev_config_dirty(newspa->spa_root_vdev);
5033 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
5035 if (zio_injection_enabled)
5036 zio_handle_panic_injection(spa, FTAG, 2);
5038 spa_async_resume(newspa);
5040 /* finally, update the original pool's config */
5041 txg = spa_vdev_config_enter(spa);
5042 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
5043 error = dmu_tx_assign(tx, TXG_WAIT);
5046 for (c = 0; c < children; c++) {
5047 if (vml[c] != NULL) {
5050 spa_history_log_internal(LOG_POOL_VDEV_DETACH,
5056 vdev_config_dirty(spa->spa_root_vdev);
5057 spa->spa_config_splitting = NULL;
5061 (void) spa_vdev_exit(spa, NULL, txg, 0);
5063 if (zio_injection_enabled)
5064 zio_handle_panic_injection(spa, FTAG, 3);
5066 /* split is complete; log a history record */
5067 spa_history_log_internal(LOG_POOL_SPLIT, newspa, NULL,
5068 "split new pool %s from pool %s", newname, spa_name(spa));
5070 kmem_free(vml, children * sizeof (vdev_t *));
5072 /* if we're not going to mount the filesystems in userland, export */
5074 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
5081 spa_deactivate(newspa);
5084 txg = spa_vdev_config_enter(spa);
5086 /* re-online all offlined disks */
5087 for (c = 0; c < children; c++) {
5089 vml[c]->vdev_offline = B_FALSE;
5091 vdev_reopen(spa->spa_root_vdev);
5093 nvlist_free(spa->spa_config_splitting);
5094 spa->spa_config_splitting = NULL;
5095 (void) spa_vdev_exit(spa, NULL, txg, error);
5097 kmem_free(vml, children * sizeof (vdev_t *));
5102 spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
5106 for (i = 0; i < count; i++) {
5109 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
5112 if (guid == target_guid)
5120 spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
5121 nvlist_t *dev_to_remove)
5123 nvlist_t **newdev = NULL;
5127 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_PUSHPAGE);
5129 for (i = 0, j = 0; i < count; i++) {
5130 if (dev[i] == dev_to_remove)
5132 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_PUSHPAGE) == 0);
5135 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
5136 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
5138 for (i = 0; i < count - 1; i++)
5139 nvlist_free(newdev[i]);
5142 kmem_free(newdev, (count - 1) * sizeof (void *));
5146 * Evacuate the device.
5149 spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
5154 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5155 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5156 ASSERT(vd == vd->vdev_top);
5159 * Evacuate the device. We don't hold the config lock as writer
5160 * since we need to do I/O but we do keep the
5161 * spa_namespace_lock held. Once this completes the device
5162 * should no longer have any blocks allocated on it.
5164 if (vd->vdev_islog) {
5165 if (vd->vdev_stat.vs_alloc != 0)
5166 error = spa_offline_log(spa);
5175 * The evacuation succeeded. Remove any remaining MOS metadata
5176 * associated with this vdev, and wait for these changes to sync.
5178 ASSERT0(vd->vdev_stat.vs_alloc);
5179 txg = spa_vdev_config_enter(spa);
5180 vd->vdev_removing = B_TRUE;
5181 vdev_dirty(vd, 0, NULL, txg);
5182 vdev_config_dirty(vd);
5183 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5189 * Complete the removal by cleaning up the namespace.
5192 spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
5194 vdev_t *rvd = spa->spa_root_vdev;
5195 uint64_t id = vd->vdev_id;
5196 boolean_t last_vdev = (id == (rvd->vdev_children - 1));
5198 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5199 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5200 ASSERT(vd == vd->vdev_top);
5203 * Only remove any devices which are empty.
5205 if (vd->vdev_stat.vs_alloc != 0)
5208 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5210 if (list_link_active(&vd->vdev_state_dirty_node))
5211 vdev_state_clean(vd);
5212 if (list_link_active(&vd->vdev_config_dirty_node))
5213 vdev_config_clean(vd);
5218 vdev_compact_children(rvd);
5220 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
5221 vdev_add_child(rvd, vd);
5223 vdev_config_dirty(rvd);
5226 * Reassess the health of our root vdev.
5232 * Remove a device from the pool -
5234 * Removing a device from the vdev namespace requires several steps
5235 * and can take a significant amount of time. As a result we use
5236 * the spa_vdev_config_[enter/exit] functions which allow us to
5237 * grab and release the spa_config_lock while still holding the namespace
5238 * lock. During each step the configuration is synced out.
5242 * Remove a device from the pool. Currently, this supports removing only hot
5243 * spares, slogs, and level 2 ARC devices.
5246 spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
5249 metaslab_group_t *mg;
5250 nvlist_t **spares, **l2cache, *nv;
5252 uint_t nspares, nl2cache;
5254 boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
5256 ASSERT(spa_writeable(spa));
5259 txg = spa_vdev_enter(spa);
5261 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5263 if (spa->spa_spares.sav_vdevs != NULL &&
5264 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5265 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
5266 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
5268 * Only remove the hot spare if it's not currently in use
5271 if (vd == NULL || unspare) {
5272 spa_vdev_remove_aux(spa->spa_spares.sav_config,
5273 ZPOOL_CONFIG_SPARES, spares, nspares, nv);
5274 spa_load_spares(spa);
5275 spa->spa_spares.sav_sync = B_TRUE;
5279 } else if (spa->spa_l2cache.sav_vdevs != NULL &&
5280 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5281 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
5282 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
5284 * Cache devices can always be removed.
5286 spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
5287 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
5288 spa_load_l2cache(spa);
5289 spa->spa_l2cache.sav_sync = B_TRUE;
5290 } else if (vd != NULL && vd->vdev_islog) {
5292 ASSERT(vd == vd->vdev_top);
5295 * XXX - Once we have bp-rewrite this should
5296 * become the common case.
5302 * Stop allocating from this vdev.
5304 metaslab_group_passivate(mg);
5307 * Wait for the youngest allocations and frees to sync,
5308 * and then wait for the deferral of those frees to finish.
5310 spa_vdev_config_exit(spa, NULL,
5311 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
5314 * Attempt to evacuate the vdev.
5316 error = spa_vdev_remove_evacuate(spa, vd);
5318 txg = spa_vdev_config_enter(spa);
5321 * If we couldn't evacuate the vdev, unwind.
5324 metaslab_group_activate(mg);
5325 return (spa_vdev_exit(spa, NULL, txg, error));
5329 * Clean up the vdev namespace.
5331 spa_vdev_remove_from_namespace(spa, vd);
5333 } else if (vd != NULL) {
5335 * Normal vdevs cannot be removed (yet).
5340 * There is no vdev of any kind with the specified guid.
5346 return (spa_vdev_exit(spa, NULL, txg, error));
5352 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5353 * current spared, so we can detach it.
5356 spa_vdev_resilver_done_hunt(vdev_t *vd)
5358 vdev_t *newvd, *oldvd;
5361 for (c = 0; c < vd->vdev_children; c++) {
5362 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
5368 * Check for a completed replacement. We always consider the first
5369 * vdev in the list to be the oldest vdev, and the last one to be
5370 * the newest (see spa_vdev_attach() for how that works). In
5371 * the case where the newest vdev is faulted, we will not automatically
5372 * remove it after a resilver completes. This is OK as it will require
5373 * user intervention to determine which disk the admin wishes to keep.
5375 if (vd->vdev_ops == &vdev_replacing_ops) {
5376 ASSERT(vd->vdev_children > 1);
5378 newvd = vd->vdev_child[vd->vdev_children - 1];
5379 oldvd = vd->vdev_child[0];
5381 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
5382 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5383 !vdev_dtl_required(oldvd))
5388 * Check for a completed resilver with the 'unspare' flag set.
5390 if (vd->vdev_ops == &vdev_spare_ops) {
5391 vdev_t *first = vd->vdev_child[0];
5392 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
5394 if (last->vdev_unspare) {
5397 } else if (first->vdev_unspare) {
5404 if (oldvd != NULL &&
5405 vdev_dtl_empty(newvd, DTL_MISSING) &&
5406 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5407 !vdev_dtl_required(oldvd))
5411 * If there are more than two spares attached to a disk,
5412 * and those spares are not required, then we want to
5413 * attempt to free them up now so that they can be used
5414 * by other pools. Once we're back down to a single
5415 * disk+spare, we stop removing them.
5417 if (vd->vdev_children > 2) {
5418 newvd = vd->vdev_child[1];
5420 if (newvd->vdev_isspare && last->vdev_isspare &&
5421 vdev_dtl_empty(last, DTL_MISSING) &&
5422 vdev_dtl_empty(last, DTL_OUTAGE) &&
5423 !vdev_dtl_required(newvd))
5432 spa_vdev_resilver_done(spa_t *spa)
5434 vdev_t *vd, *pvd, *ppvd;
5435 uint64_t guid, sguid, pguid, ppguid;
5437 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5439 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
5440 pvd = vd->vdev_parent;
5441 ppvd = pvd->vdev_parent;
5442 guid = vd->vdev_guid;
5443 pguid = pvd->vdev_guid;
5444 ppguid = ppvd->vdev_guid;
5447 * If we have just finished replacing a hot spared device, then
5448 * we need to detach the parent's first child (the original hot
5451 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
5452 ppvd->vdev_children == 2) {
5453 ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
5454 sguid = ppvd->vdev_child[1]->vdev_guid;
5456 spa_config_exit(spa, SCL_ALL, FTAG);
5457 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
5459 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
5461 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5464 spa_config_exit(spa, SCL_ALL, FTAG);
5468 * Update the stored path or FRU for this vdev.
5471 spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
5475 boolean_t sync = B_FALSE;
5477 ASSERT(spa_writeable(spa));
5479 spa_vdev_state_enter(spa, SCL_ALL);
5481 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
5482 return (spa_vdev_state_exit(spa, NULL, ENOENT));
5484 if (!vd->vdev_ops->vdev_op_leaf)
5485 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
5488 if (strcmp(value, vd->vdev_path) != 0) {
5489 spa_strfree(vd->vdev_path);
5490 vd->vdev_path = spa_strdup(value);
5494 if (vd->vdev_fru == NULL) {
5495 vd->vdev_fru = spa_strdup(value);
5497 } else if (strcmp(value, vd->vdev_fru) != 0) {
5498 spa_strfree(vd->vdev_fru);
5499 vd->vdev_fru = spa_strdup(value);
5504 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
5508 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
5510 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5514 spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5516 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5520 * ==========================================================================
5522 * ==========================================================================
5526 spa_scan_stop(spa_t *spa)
5528 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5529 if (dsl_scan_resilvering(spa->spa_dsl_pool))
5531 return (dsl_scan_cancel(spa->spa_dsl_pool));
5535 spa_scan(spa_t *spa, pool_scan_func_t func)
5537 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5539 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5543 * If a resilver was requested, but there is no DTL on a
5544 * writeable leaf device, we have nothing to do.
5546 if (func == POOL_SCAN_RESILVER &&
5547 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5548 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5552 return (dsl_scan(spa->spa_dsl_pool, func));
5556 * ==========================================================================
5557 * SPA async task processing
5558 * ==========================================================================
5562 spa_async_remove(spa_t *spa, vdev_t *vd)
5566 if (vd->vdev_remove_wanted) {
5567 vd->vdev_remove_wanted = B_FALSE;
5568 vd->vdev_delayed_close = B_FALSE;
5569 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5572 * We want to clear the stats, but we don't want to do a full
5573 * vdev_clear() as that will cause us to throw away
5574 * degraded/faulted state as well as attempt to reopen the
5575 * device, all of which is a waste.
5577 vd->vdev_stat.vs_read_errors = 0;
5578 vd->vdev_stat.vs_write_errors = 0;
5579 vd->vdev_stat.vs_checksum_errors = 0;
5581 vdev_state_dirty(vd->vdev_top);
5584 for (c = 0; c < vd->vdev_children; c++)
5585 spa_async_remove(spa, vd->vdev_child[c]);
5589 spa_async_probe(spa_t *spa, vdev_t *vd)
5593 if (vd->vdev_probe_wanted) {
5594 vd->vdev_probe_wanted = B_FALSE;
5595 vdev_reopen(vd); /* vdev_open() does the actual probe */
5598 for (c = 0; c < vd->vdev_children; c++)
5599 spa_async_probe(spa, vd->vdev_child[c]);
5603 spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5607 if (!spa->spa_autoexpand)
5610 for (c = 0; c < vd->vdev_children; c++) {
5611 vdev_t *cvd = vd->vdev_child[c];
5612 spa_async_autoexpand(spa, cvd);
5615 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5618 spa_event_notify(vd->vdev_spa, vd, FM_EREPORT_ZFS_DEVICE_AUTOEXPAND);
5622 spa_async_thread(spa_t *spa)
5626 ASSERT(spa->spa_sync_on);
5628 mutex_enter(&spa->spa_async_lock);
5629 tasks = spa->spa_async_tasks;
5630 spa->spa_async_tasks = 0;
5631 mutex_exit(&spa->spa_async_lock);
5634 * See if the config needs to be updated.
5636 if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5637 uint64_t old_space, new_space;
5639 mutex_enter(&spa_namespace_lock);
5640 old_space = metaslab_class_get_space(spa_normal_class(spa));
5641 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5642 new_space = metaslab_class_get_space(spa_normal_class(spa));
5643 mutex_exit(&spa_namespace_lock);
5646 * If the pool grew as a result of the config update,
5647 * then log an internal history event.
5649 if (new_space != old_space) {
5650 spa_history_log_internal(LOG_POOL_VDEV_ONLINE,
5652 "pool '%s' size: %llu(+%llu)",
5653 spa_name(spa), new_space, new_space - old_space);
5658 * See if any devices need to be marked REMOVED.
5660 if (tasks & SPA_ASYNC_REMOVE) {
5661 spa_vdev_state_enter(spa, SCL_NONE);
5662 spa_async_remove(spa, spa->spa_root_vdev);
5663 for (i = 0; i < spa->spa_l2cache.sav_count; i++)
5664 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5665 for (i = 0; i < spa->spa_spares.sav_count; i++)
5666 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5667 (void) spa_vdev_state_exit(spa, NULL, 0);
5670 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5671 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5672 spa_async_autoexpand(spa, spa->spa_root_vdev);
5673 spa_config_exit(spa, SCL_CONFIG, FTAG);
5677 * See if any devices need to be probed.
5679 if (tasks & SPA_ASYNC_PROBE) {
5680 spa_vdev_state_enter(spa, SCL_NONE);
5681 spa_async_probe(spa, spa->spa_root_vdev);
5682 (void) spa_vdev_state_exit(spa, NULL, 0);
5686 * If any devices are done replacing, detach them.
5688 if (tasks & SPA_ASYNC_RESILVER_DONE)
5689 spa_vdev_resilver_done(spa);
5692 * Kick off a resilver.
5694 if (tasks & SPA_ASYNC_RESILVER)
5695 dsl_resilver_restart(spa->spa_dsl_pool, 0);
5698 * Let the world know that we're done.
5700 mutex_enter(&spa->spa_async_lock);
5701 spa->spa_async_thread = NULL;
5702 cv_broadcast(&spa->spa_async_cv);
5703 mutex_exit(&spa->spa_async_lock);
5708 spa_async_suspend(spa_t *spa)
5710 mutex_enter(&spa->spa_async_lock);
5711 spa->spa_async_suspended++;
5712 while (spa->spa_async_thread != NULL)
5713 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5714 mutex_exit(&spa->spa_async_lock);
5718 spa_async_resume(spa_t *spa)
5720 mutex_enter(&spa->spa_async_lock);
5721 ASSERT(spa->spa_async_suspended != 0);
5722 spa->spa_async_suspended--;
5723 mutex_exit(&spa->spa_async_lock);
5727 spa_async_dispatch(spa_t *spa)
5729 mutex_enter(&spa->spa_async_lock);
5730 if (spa->spa_async_tasks && !spa->spa_async_suspended &&
5731 spa->spa_async_thread == NULL &&
5732 rootdir != NULL && !vn_is_readonly(rootdir))
5733 spa->spa_async_thread = thread_create(NULL, 0,
5734 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
5735 mutex_exit(&spa->spa_async_lock);
5739 spa_async_request(spa_t *spa, int task)
5741 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
5742 mutex_enter(&spa->spa_async_lock);
5743 spa->spa_async_tasks |= task;
5744 mutex_exit(&spa->spa_async_lock);
5748 * ==========================================================================
5749 * SPA syncing routines
5750 * ==========================================================================
5754 bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5757 bpobj_enqueue(bpo, bp, tx);
5762 spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5766 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
5772 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
5774 char *packed = NULL;
5779 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
5782 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5783 * information. This avoids the dbuf_will_dirty() path and
5784 * saves us a pre-read to get data we don't actually care about.
5786 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
5787 packed = vmem_alloc(bufsize, KM_PUSHPAGE);
5789 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
5791 bzero(packed + nvsize, bufsize - nvsize);
5793 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
5795 vmem_free(packed, bufsize);
5797 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
5798 dmu_buf_will_dirty(db, tx);
5799 *(uint64_t *)db->db_data = nvsize;
5800 dmu_buf_rele(db, FTAG);
5804 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
5805 const char *config, const char *entry)
5815 * Update the MOS nvlist describing the list of available devices.
5816 * spa_validate_aux() will have already made sure this nvlist is
5817 * valid and the vdevs are labeled appropriately.
5819 if (sav->sav_object == 0) {
5820 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
5821 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
5822 sizeof (uint64_t), tx);
5823 VERIFY(zap_update(spa->spa_meta_objset,
5824 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
5825 &sav->sav_object, tx) == 0);
5828 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_PUSHPAGE) == 0);
5829 if (sav->sav_count == 0) {
5830 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
5832 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_PUSHPAGE);
5833 for (i = 0; i < sav->sav_count; i++)
5834 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
5835 B_FALSE, VDEV_CONFIG_L2CACHE);
5836 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
5837 sav->sav_count) == 0);
5838 for (i = 0; i < sav->sav_count; i++)
5839 nvlist_free(list[i]);
5840 kmem_free(list, sav->sav_count * sizeof (void *));
5843 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
5844 nvlist_free(nvroot);
5846 sav->sav_sync = B_FALSE;
5850 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
5854 if (list_is_empty(&spa->spa_config_dirty_list))
5857 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5859 config = spa_config_generate(spa, spa->spa_root_vdev,
5860 dmu_tx_get_txg(tx), B_FALSE);
5863 * If we're upgrading the spa version then make sure that
5864 * the config object gets updated with the correct version.
5866 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
5867 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5868 spa->spa_uberblock.ub_version);
5870 spa_config_exit(spa, SCL_STATE, FTAG);
5872 if (spa->spa_config_syncing)
5873 nvlist_free(spa->spa_config_syncing);
5874 spa->spa_config_syncing = config;
5876 spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
5880 spa_sync_version(void *arg1, void *arg2, dmu_tx_t *tx)
5883 uint64_t version = *(uint64_t *)arg2;
5886 * Setting the version is special cased when first creating the pool.
5888 ASSERT(tx->tx_txg != TXG_INITIAL);
5890 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
5891 ASSERT(version >= spa_version(spa));
5893 spa->spa_uberblock.ub_version = version;
5894 vdev_config_dirty(spa->spa_root_vdev);
5898 * Set zpool properties.
5901 spa_sync_props(void *arg1, void *arg2, dmu_tx_t *tx)
5904 objset_t *mos = spa->spa_meta_objset;
5905 nvlist_t *nvp = arg2;
5906 nvpair_t *elem = NULL;
5908 mutex_enter(&spa->spa_props_lock);
5910 while ((elem = nvlist_next_nvpair(nvp, elem))) {
5912 char *strval, *fname;
5914 const char *propname;
5915 zprop_type_t proptype;
5916 zfeature_info_t *feature;
5918 prop = zpool_name_to_prop(nvpair_name(elem));
5919 switch ((int)prop) {
5922 * We checked this earlier in spa_prop_validate().
5924 ASSERT(zpool_prop_feature(nvpair_name(elem)));
5926 fname = strchr(nvpair_name(elem), '@') + 1;
5927 VERIFY3U(0, ==, zfeature_lookup_name(fname, &feature));
5929 spa_feature_enable(spa, feature, tx);
5932 case ZPOOL_PROP_VERSION:
5933 VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5935 * The version is synced seperatly before other
5936 * properties and should be correct by now.
5938 ASSERT3U(spa_version(spa), >=, intval);
5941 case ZPOOL_PROP_ALTROOT:
5943 * 'altroot' is a non-persistent property. It should
5944 * have been set temporarily at creation or import time.
5946 ASSERT(spa->spa_root != NULL);
5949 case ZPOOL_PROP_READONLY:
5950 case ZPOOL_PROP_CACHEFILE:
5952 * 'readonly' and 'cachefile' are also non-persisitent
5956 case ZPOOL_PROP_COMMENT:
5957 VERIFY(nvpair_value_string(elem, &strval) == 0);
5958 if (spa->spa_comment != NULL)
5959 spa_strfree(spa->spa_comment);
5960 spa->spa_comment = spa_strdup(strval);
5962 * We need to dirty the configuration on all the vdevs
5963 * so that their labels get updated. It's unnecessary
5964 * to do this for pool creation since the vdev's
5965 * configuratoin has already been dirtied.
5967 if (tx->tx_txg != TXG_INITIAL)
5968 vdev_config_dirty(spa->spa_root_vdev);
5972 * Set pool property values in the poolprops mos object.
5974 if (spa->spa_pool_props_object == 0) {
5975 spa->spa_pool_props_object =
5976 zap_create_link(mos, DMU_OT_POOL_PROPS,
5977 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
5981 /* normalize the property name */
5982 propname = zpool_prop_to_name(prop);
5983 proptype = zpool_prop_get_type(prop);
5985 if (nvpair_type(elem) == DATA_TYPE_STRING) {
5986 ASSERT(proptype == PROP_TYPE_STRING);
5987 VERIFY(nvpair_value_string(elem, &strval) == 0);
5988 VERIFY(zap_update(mos,
5989 spa->spa_pool_props_object, propname,
5990 1, strlen(strval) + 1, strval, tx) == 0);
5992 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
5993 VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5995 if (proptype == PROP_TYPE_INDEX) {
5997 VERIFY(zpool_prop_index_to_string(
5998 prop, intval, &unused) == 0);
6000 VERIFY(zap_update(mos,
6001 spa->spa_pool_props_object, propname,
6002 8, 1, &intval, tx) == 0);
6004 ASSERT(0); /* not allowed */
6008 case ZPOOL_PROP_DELEGATION:
6009 spa->spa_delegation = intval;
6011 case ZPOOL_PROP_BOOTFS:
6012 spa->spa_bootfs = intval;
6014 case ZPOOL_PROP_FAILUREMODE:
6015 spa->spa_failmode = intval;
6017 case ZPOOL_PROP_AUTOEXPAND:
6018 spa->spa_autoexpand = intval;
6019 if (tx->tx_txg != TXG_INITIAL)
6020 spa_async_request(spa,
6021 SPA_ASYNC_AUTOEXPAND);
6023 case ZPOOL_PROP_DEDUPDITTO:
6024 spa->spa_dedup_ditto = intval;
6031 /* log internal history if this is not a zpool create */
6032 if (spa_version(spa) >= SPA_VERSION_ZPOOL_HISTORY &&
6033 tx->tx_txg != TXG_INITIAL) {
6034 spa_history_log_internal(LOG_POOL_PROPSET,
6035 spa, tx, "%s %lld %s",
6036 nvpair_name(elem), intval, spa_name(spa));
6040 mutex_exit(&spa->spa_props_lock);
6044 * Perform one-time upgrade on-disk changes. spa_version() does not
6045 * reflect the new version this txg, so there must be no changes this
6046 * txg to anything that the upgrade code depends on after it executes.
6047 * Therefore this must be called after dsl_pool_sync() does the sync
6051 spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
6053 dsl_pool_t *dp = spa->spa_dsl_pool;
6055 ASSERT(spa->spa_sync_pass == 1);
6057 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
6058 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
6059 dsl_pool_create_origin(dp, tx);
6061 /* Keeping the origin open increases spa_minref */
6062 spa->spa_minref += 3;
6065 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
6066 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
6067 dsl_pool_upgrade_clones(dp, tx);
6070 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
6071 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
6072 dsl_pool_upgrade_dir_clones(dp, tx);
6074 /* Keeping the freedir open increases spa_minref */
6075 spa->spa_minref += 3;
6078 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES &&
6079 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
6080 spa_feature_create_zap_objects(spa, tx);
6085 * Sync the specified transaction group. New blocks may be dirtied as
6086 * part of the process, so we iterate until it converges.
6089 spa_sync(spa_t *spa, uint64_t txg)
6091 dsl_pool_t *dp = spa->spa_dsl_pool;
6092 objset_t *mos = spa->spa_meta_objset;
6093 bpobj_t *defer_bpo = &spa->spa_deferred_bpobj;
6094 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
6095 vdev_t *rvd = spa->spa_root_vdev;
6101 VERIFY(spa_writeable(spa));
6104 * Lock out configuration changes.
6106 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
6108 spa->spa_syncing_txg = txg;
6109 spa->spa_sync_pass = 0;
6112 * If there are any pending vdev state changes, convert them
6113 * into config changes that go out with this transaction group.
6115 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6116 while (list_head(&spa->spa_state_dirty_list) != NULL) {
6118 * We need the write lock here because, for aux vdevs,
6119 * calling vdev_config_dirty() modifies sav_config.
6120 * This is ugly and will become unnecessary when we
6121 * eliminate the aux vdev wart by integrating all vdevs
6122 * into the root vdev tree.
6124 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6125 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
6126 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
6127 vdev_state_clean(vd);
6128 vdev_config_dirty(vd);
6130 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6131 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
6133 spa_config_exit(spa, SCL_STATE, FTAG);
6135 tx = dmu_tx_create_assigned(dp, txg);
6137 spa->spa_sync_starttime = gethrtime();
6138 taskq_cancel_id(system_taskq, spa->spa_deadman_tqid);
6139 spa->spa_deadman_tqid = taskq_dispatch_delay(system_taskq,
6140 spa_deadman, spa, TQ_SLEEP, ddi_get_lbolt() +
6141 NSEC_TO_TICK(spa->spa_deadman_synctime));
6144 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
6145 * set spa_deflate if we have no raid-z vdevs.
6147 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
6148 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
6151 for (i = 0; i < rvd->vdev_children; i++) {
6152 vd = rvd->vdev_child[i];
6153 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
6156 if (i == rvd->vdev_children) {
6157 spa->spa_deflate = TRUE;
6158 VERIFY(0 == zap_add(spa->spa_meta_objset,
6159 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
6160 sizeof (uint64_t), 1, &spa->spa_deflate, tx));
6165 * If anything has changed in this txg, or if someone is waiting
6166 * for this txg to sync (eg, spa_vdev_remove()), push the
6167 * deferred frees from the previous txg. If not, leave them
6168 * alone so that we don't generate work on an otherwise idle
6171 if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
6172 !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
6173 !txg_list_empty(&dp->dp_sync_tasks, txg) ||
6174 ((dsl_scan_active(dp->dp_scan) ||
6175 txg_sync_waiting(dp)) && !spa_shutting_down(spa))) {
6176 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6177 VERIFY3U(bpobj_iterate(defer_bpo,
6178 spa_free_sync_cb, zio, tx), ==, 0);
6179 VERIFY0(zio_wait(zio));
6183 * Iterate to convergence.
6186 int pass = ++spa->spa_sync_pass;
6188 spa_sync_config_object(spa, tx);
6189 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
6190 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
6191 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
6192 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
6193 spa_errlog_sync(spa, txg);
6194 dsl_pool_sync(dp, txg);
6196 if (pass < zfs_sync_pass_deferred_free) {
6197 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6198 bplist_iterate(free_bpl, spa_free_sync_cb,
6200 VERIFY(zio_wait(zio) == 0);
6202 bplist_iterate(free_bpl, bpobj_enqueue_cb,
6207 dsl_scan_sync(dp, tx);
6209 while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)))
6213 spa_sync_upgrades(spa, tx);
6215 } while (dmu_objset_is_dirty(mos, txg));
6218 * Rewrite the vdev configuration (which includes the uberblock)
6219 * to commit the transaction group.
6221 * If there are no dirty vdevs, we sync the uberblock to a few
6222 * random top-level vdevs that are known to be visible in the
6223 * config cache (see spa_vdev_add() for a complete description).
6224 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
6228 * We hold SCL_STATE to prevent vdev open/close/etc.
6229 * while we're attempting to write the vdev labels.
6231 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6233 if (list_is_empty(&spa->spa_config_dirty_list)) {
6234 vdev_t *svd[SPA_DVAS_PER_BP];
6236 int children = rvd->vdev_children;
6237 int c0 = spa_get_random(children);
6239 for (c = 0; c < children; c++) {
6240 vd = rvd->vdev_child[(c0 + c) % children];
6241 if (vd->vdev_ms_array == 0 || vd->vdev_islog)
6243 svd[svdcount++] = vd;
6244 if (svdcount == SPA_DVAS_PER_BP)
6247 error = vdev_config_sync(svd, svdcount, txg, B_FALSE);
6249 error = vdev_config_sync(svd, svdcount, txg,
6252 error = vdev_config_sync(rvd->vdev_child,
6253 rvd->vdev_children, txg, B_FALSE);
6255 error = vdev_config_sync(rvd->vdev_child,
6256 rvd->vdev_children, txg, B_TRUE);
6260 spa->spa_last_synced_guid = rvd->vdev_guid;
6262 spa_config_exit(spa, SCL_STATE, FTAG);
6266 zio_suspend(spa, NULL);
6267 zio_resume_wait(spa);
6271 taskq_cancel_id(system_taskq, spa->spa_deadman_tqid);
6272 spa->spa_deadman_tqid = 0;
6275 * Clear the dirty config list.
6277 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
6278 vdev_config_clean(vd);
6281 * Now that the new config has synced transactionally,
6282 * let it become visible to the config cache.
6284 if (spa->spa_config_syncing != NULL) {
6285 spa_config_set(spa, spa->spa_config_syncing);
6286 spa->spa_config_txg = txg;
6287 spa->spa_config_syncing = NULL;
6290 spa->spa_ubsync = spa->spa_uberblock;
6292 dsl_pool_sync_done(dp, txg);
6295 * Update usable space statistics.
6297 while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg))))
6298 vdev_sync_done(vd, txg);
6300 spa_update_dspace(spa);
6303 * It had better be the case that we didn't dirty anything
6304 * since vdev_config_sync().
6306 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
6307 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
6308 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
6310 spa->spa_sync_pass = 0;
6312 spa_config_exit(spa, SCL_CONFIG, FTAG);
6314 spa_handle_ignored_writes(spa);
6317 * If any async tasks have been requested, kick them off.
6319 spa_async_dispatch(spa);
6323 * Sync all pools. We don't want to hold the namespace lock across these
6324 * operations, so we take a reference on the spa_t and drop the lock during the
6328 spa_sync_allpools(void)
6331 mutex_enter(&spa_namespace_lock);
6332 while ((spa = spa_next(spa)) != NULL) {
6333 if (spa_state(spa) != POOL_STATE_ACTIVE ||
6334 !spa_writeable(spa) || spa_suspended(spa))
6336 spa_open_ref(spa, FTAG);
6337 mutex_exit(&spa_namespace_lock);
6338 txg_wait_synced(spa_get_dsl(spa), 0);
6339 mutex_enter(&spa_namespace_lock);
6340 spa_close(spa, FTAG);
6342 mutex_exit(&spa_namespace_lock);
6346 * ==========================================================================
6347 * Miscellaneous routines
6348 * ==========================================================================
6352 * Remove all pools in the system.
6360 * Remove all cached state. All pools should be closed now,
6361 * so every spa in the AVL tree should be unreferenced.
6363 mutex_enter(&spa_namespace_lock);
6364 while ((spa = spa_next(NULL)) != NULL) {
6366 * Stop async tasks. The async thread may need to detach
6367 * a device that's been replaced, which requires grabbing
6368 * spa_namespace_lock, so we must drop it here.
6370 spa_open_ref(spa, FTAG);
6371 mutex_exit(&spa_namespace_lock);
6372 spa_async_suspend(spa);
6373 mutex_enter(&spa_namespace_lock);
6374 spa_close(spa, FTAG);
6376 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
6378 spa_deactivate(spa);
6382 mutex_exit(&spa_namespace_lock);
6386 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
6391 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
6395 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
6396 vd = spa->spa_l2cache.sav_vdevs[i];
6397 if (vd->vdev_guid == guid)
6401 for (i = 0; i < spa->spa_spares.sav_count; i++) {
6402 vd = spa->spa_spares.sav_vdevs[i];
6403 if (vd->vdev_guid == guid)
6412 spa_upgrade(spa_t *spa, uint64_t version)
6414 ASSERT(spa_writeable(spa));
6416 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6419 * This should only be called for a non-faulted pool, and since a
6420 * future version would result in an unopenable pool, this shouldn't be
6423 ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version));
6424 ASSERT(version >= spa->spa_uberblock.ub_version);
6426 spa->spa_uberblock.ub_version = version;
6427 vdev_config_dirty(spa->spa_root_vdev);
6429 spa_config_exit(spa, SCL_ALL, FTAG);
6431 txg_wait_synced(spa_get_dsl(spa), 0);
6435 spa_has_spare(spa_t *spa, uint64_t guid)
6439 spa_aux_vdev_t *sav = &spa->spa_spares;
6441 for (i = 0; i < sav->sav_count; i++)
6442 if (sav->sav_vdevs[i]->vdev_guid == guid)
6445 for (i = 0; i < sav->sav_npending; i++) {
6446 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
6447 &spareguid) == 0 && spareguid == guid)
6455 * Check if a pool has an active shared spare device.
6456 * Note: reference count of an active spare is 2, as a spare and as a replace
6459 spa_has_active_shared_spare(spa_t *spa)
6463 spa_aux_vdev_t *sav = &spa->spa_spares;
6465 for (i = 0; i < sav->sav_count; i++) {
6466 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
6467 &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
6476 * Post a FM_EREPORT_ZFS_* event from sys/fm/fs/zfs.h. The payload will be
6477 * filled in from the spa and (optionally) the vdev. This doesn't do anything
6478 * in the userland libzpool, as we don't want consumers to misinterpret ztest
6479 * or zdb as real changes.
6482 spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
6485 zfs_ereport_post(name, spa, vd, NULL, 0, 0);
6489 #if defined(_KERNEL) && defined(HAVE_SPL)
6490 /* state manipulation functions */
6491 EXPORT_SYMBOL(spa_open);
6492 EXPORT_SYMBOL(spa_open_rewind);
6493 EXPORT_SYMBOL(spa_get_stats);
6494 EXPORT_SYMBOL(spa_create);
6495 EXPORT_SYMBOL(spa_import_rootpool);
6496 EXPORT_SYMBOL(spa_import);
6497 EXPORT_SYMBOL(spa_tryimport);
6498 EXPORT_SYMBOL(spa_destroy);
6499 EXPORT_SYMBOL(spa_export);
6500 EXPORT_SYMBOL(spa_reset);
6501 EXPORT_SYMBOL(spa_async_request);
6502 EXPORT_SYMBOL(spa_async_suspend);
6503 EXPORT_SYMBOL(spa_async_resume);
6504 EXPORT_SYMBOL(spa_inject_addref);
6505 EXPORT_SYMBOL(spa_inject_delref);
6506 EXPORT_SYMBOL(spa_scan_stat_init);
6507 EXPORT_SYMBOL(spa_scan_get_stats);
6509 /* device maniion */
6510 EXPORT_SYMBOL(spa_vdev_add);
6511 EXPORT_SYMBOL(spa_vdev_attach);
6512 EXPORT_SYMBOL(spa_vdev_detach);
6513 EXPORT_SYMBOL(spa_vdev_remove);
6514 EXPORT_SYMBOL(spa_vdev_setpath);
6515 EXPORT_SYMBOL(spa_vdev_setfru);
6516 EXPORT_SYMBOL(spa_vdev_split_mirror);
6518 /* spare statech is global across all pools) */
6519 EXPORT_SYMBOL(spa_spare_add);
6520 EXPORT_SYMBOL(spa_spare_remove);
6521 EXPORT_SYMBOL(spa_spare_exists);
6522 EXPORT_SYMBOL(spa_spare_activate);
6524 /* L2ARC statech is global across all pools) */
6525 EXPORT_SYMBOL(spa_l2cache_add);
6526 EXPORT_SYMBOL(spa_l2cache_remove);
6527 EXPORT_SYMBOL(spa_l2cache_exists);
6528 EXPORT_SYMBOL(spa_l2cache_activate);
6529 EXPORT_SYMBOL(spa_l2cache_drop);
6532 EXPORT_SYMBOL(spa_scan);
6533 EXPORT_SYMBOL(spa_scan_stop);
6536 EXPORT_SYMBOL(spa_sync); /* only for DMU use */
6537 EXPORT_SYMBOL(spa_sync_allpools);
6540 EXPORT_SYMBOL(spa_prop_set);
6541 EXPORT_SYMBOL(spa_prop_get);
6542 EXPORT_SYMBOL(spa_prop_clear_bootfs);
6544 /* asynchronous event notification */
6545 EXPORT_SYMBOL(spa_event_notify);