4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2011, 2014 by Delphix. All rights reserved.
25 * Copyright (c) 2013, 2014, Nexenta Systems, Inc. All rights reserved.
26 * Copyright (c) 2013 Martin Matuska <mm@FreeBSD.org>. All rights reserved.
30 * SPA: Storage Pool Allocator
32 * This file contains all the routines used when modifying on-disk SPA state.
33 * This includes opening, importing, destroying, exporting a pool, and syncing a
37 #include <sys/zfs_context.h>
38 #include <sys/fm/fs/zfs.h>
39 #include <sys/spa_impl.h>
41 #include <sys/zio_checksum.h>
43 #include <sys/dmu_tx.h>
47 #include <sys/vdev_impl.h>
48 #include <sys/metaslab.h>
49 #include <sys/metaslab_impl.h>
50 #include <sys/uberblock_impl.h>
53 #include <sys/dmu_traverse.h>
54 #include <sys/dmu_objset.h>
55 #include <sys/unique.h>
56 #include <sys/dsl_pool.h>
57 #include <sys/dsl_dataset.h>
58 #include <sys/dsl_dir.h>
59 #include <sys/dsl_prop.h>
60 #include <sys/dsl_synctask.h>
61 #include <sys/fs/zfs.h>
63 #include <sys/callb.h>
64 #include <sys/spa_boot.h>
65 #include <sys/zfs_ioctl.h>
66 #include <sys/dsl_scan.h>
67 #include <sys/dmu_send.h>
68 #include <sys/dsl_destroy.h>
69 #include <sys/dsl_userhold.h>
70 #include <sys/zfeature.h>
72 #include <sys/trim_map.h>
75 #include <sys/callb.h>
76 #include <sys/cpupart.h>
81 #include "zfs_comutil.h"
83 /* Check hostid on import? */
84 static int check_hostid = 1;
86 SYSCTL_DECL(_vfs_zfs);
87 SYSCTL_INT(_vfs_zfs, OID_AUTO, check_hostid, CTLFLAG_RWTUN, &check_hostid, 0,
88 "Check hostid on import?");
91 * The interval, in seconds, at which failed configuration cache file writes
94 static int zfs_ccw_retry_interval = 300;
96 typedef enum zti_modes {
97 ZTI_MODE_FIXED, /* value is # of threads (min 1) */
98 ZTI_MODE_BATCH, /* cpu-intensive; value is ignored */
99 ZTI_MODE_NULL, /* don't create a taskq */
103 #define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) }
104 #define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 }
105 #define ZTI_NULL { ZTI_MODE_NULL, 0, 0 }
107 #define ZTI_N(n) ZTI_P(n, 1)
108 #define ZTI_ONE ZTI_N(1)
110 typedef struct zio_taskq_info {
111 zti_modes_t zti_mode;
116 static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
117 "issue", "issue_high", "intr", "intr_high"
121 * This table defines the taskq settings for each ZFS I/O type. When
122 * initializing a pool, we use this table to create an appropriately sized
123 * taskq. Some operations are low volume and therefore have a small, static
124 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE
125 * macros. Other operations process a large amount of data; the ZTI_BATCH
126 * macro causes us to create a taskq oriented for throughput. Some operations
127 * are so high frequency and short-lived that the taskq itself can become a a
128 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an
129 * additional degree of parallelism specified by the number of threads per-
130 * taskq and the number of taskqs; when dispatching an event in this case, the
131 * particular taskq is chosen at random.
133 * The different taskq priorities are to handle the different contexts (issue
134 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that
135 * need to be handled with minimum delay.
137 const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
138 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
139 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* NULL */
140 { ZTI_N(8), ZTI_NULL, ZTI_P(12, 8), ZTI_NULL }, /* READ */
141 { ZTI_BATCH, ZTI_N(5), ZTI_N(8), ZTI_N(5) }, /* WRITE */
142 { ZTI_P(12, 8), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FREE */
143 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* CLAIM */
144 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* IOCTL */
147 static void spa_sync_version(void *arg, dmu_tx_t *tx);
148 static void spa_sync_props(void *arg, dmu_tx_t *tx);
149 static boolean_t spa_has_active_shared_spare(spa_t *spa);
150 static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
151 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
153 static void spa_vdev_resilver_done(spa_t *spa);
155 uint_t zio_taskq_batch_pct = 75; /* 1 thread per cpu in pset */
157 id_t zio_taskq_psrset_bind = PS_NONE;
160 boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */
162 uint_t zio_taskq_basedc = 80; /* base duty cycle */
164 boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */
165 extern int zfs_sync_pass_deferred_free;
168 extern void spa_deadman(void *arg);
172 * This (illegal) pool name is used when temporarily importing a spa_t in order
173 * to get the vdev stats associated with the imported devices.
175 #define TRYIMPORT_NAME "$import"
178 * ==========================================================================
179 * SPA properties routines
180 * ==========================================================================
184 * Add a (source=src, propname=propval) list to an nvlist.
187 spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
188 uint64_t intval, zprop_source_t src)
190 const char *propname = zpool_prop_to_name(prop);
193 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
194 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
197 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
199 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
201 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
202 nvlist_free(propval);
206 * Get property values from the spa configuration.
209 spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
211 vdev_t *rvd = spa->spa_root_vdev;
212 dsl_pool_t *pool = spa->spa_dsl_pool;
216 uint64_t cap, version;
217 zprop_source_t src = ZPROP_SRC_NONE;
218 spa_config_dirent_t *dp;
220 ASSERT(MUTEX_HELD(&spa->spa_props_lock));
223 alloc = metaslab_class_get_alloc(spa_normal_class(spa));
224 size = metaslab_class_get_space(spa_normal_class(spa));
225 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
226 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
227 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
228 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
232 for (int c = 0; c < rvd->vdev_children; c++) {
233 vdev_t *tvd = rvd->vdev_child[c];
234 space += tvd->vdev_max_asize - tvd->vdev_asize;
236 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, space,
239 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
240 (spa_mode(spa) == FREAD), src);
242 cap = (size == 0) ? 0 : (alloc * 100 / size);
243 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
245 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
246 ddt_get_pool_dedup_ratio(spa), src);
248 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
249 rvd->vdev_state, src);
251 version = spa_version(spa);
252 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
253 src = ZPROP_SRC_DEFAULT;
255 src = ZPROP_SRC_LOCAL;
256 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
261 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
262 * when opening pools before this version freedir will be NULL.
264 if (pool->dp_free_dir != NULL) {
265 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
266 pool->dp_free_dir->dd_phys->dd_used_bytes, src);
268 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
272 if (pool->dp_leak_dir != NULL) {
273 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, NULL,
274 pool->dp_leak_dir->dd_phys->dd_used_bytes, src);
276 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED,
281 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
283 if (spa->spa_comment != NULL) {
284 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
288 if (spa->spa_root != NULL)
289 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
292 if ((dp = list_head(&spa->spa_config_list)) != NULL) {
293 if (dp->scd_path == NULL) {
294 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
295 "none", 0, ZPROP_SRC_LOCAL);
296 } else if (strcmp(dp->scd_path, spa_config_path) != 0) {
297 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
298 dp->scd_path, 0, ZPROP_SRC_LOCAL);
304 * Get zpool property values.
307 spa_prop_get(spa_t *spa, nvlist_t **nvp)
309 objset_t *mos = spa->spa_meta_objset;
314 VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
316 mutex_enter(&spa->spa_props_lock);
319 * Get properties from the spa config.
321 spa_prop_get_config(spa, nvp);
323 /* If no pool property object, no more prop to get. */
324 if (mos == NULL || spa->spa_pool_props_object == 0) {
325 mutex_exit(&spa->spa_props_lock);
330 * Get properties from the MOS pool property object.
332 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
333 (err = zap_cursor_retrieve(&zc, &za)) == 0;
334 zap_cursor_advance(&zc)) {
337 zprop_source_t src = ZPROP_SRC_DEFAULT;
340 if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
343 switch (za.za_integer_length) {
345 /* integer property */
346 if (za.za_first_integer !=
347 zpool_prop_default_numeric(prop))
348 src = ZPROP_SRC_LOCAL;
350 if (prop == ZPOOL_PROP_BOOTFS) {
352 dsl_dataset_t *ds = NULL;
354 dp = spa_get_dsl(spa);
355 dsl_pool_config_enter(dp, FTAG);
356 if (err = dsl_dataset_hold_obj(dp,
357 za.za_first_integer, FTAG, &ds)) {
358 dsl_pool_config_exit(dp, FTAG);
363 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
365 dsl_dataset_name(ds, strval);
366 dsl_dataset_rele(ds, FTAG);
367 dsl_pool_config_exit(dp, FTAG);
370 intval = za.za_first_integer;
373 spa_prop_add_list(*nvp, prop, strval, intval, src);
377 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
382 /* string property */
383 strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
384 err = zap_lookup(mos, spa->spa_pool_props_object,
385 za.za_name, 1, za.za_num_integers, strval);
387 kmem_free(strval, za.za_num_integers);
390 spa_prop_add_list(*nvp, prop, strval, 0, src);
391 kmem_free(strval, za.za_num_integers);
398 zap_cursor_fini(&zc);
399 mutex_exit(&spa->spa_props_lock);
401 if (err && err != ENOENT) {
411 * Validate the given pool properties nvlist and modify the list
412 * for the property values to be set.
415 spa_prop_validate(spa_t *spa, nvlist_t *props)
418 int error = 0, reset_bootfs = 0;
420 boolean_t has_feature = B_FALSE;
423 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
425 char *strval, *slash, *check, *fname;
426 const char *propname = nvpair_name(elem);
427 zpool_prop_t prop = zpool_name_to_prop(propname);
431 if (!zpool_prop_feature(propname)) {
432 error = SET_ERROR(EINVAL);
437 * Sanitize the input.
439 if (nvpair_type(elem) != DATA_TYPE_UINT64) {
440 error = SET_ERROR(EINVAL);
444 if (nvpair_value_uint64(elem, &intval) != 0) {
445 error = SET_ERROR(EINVAL);
450 error = SET_ERROR(EINVAL);
454 fname = strchr(propname, '@') + 1;
455 if (zfeature_lookup_name(fname, NULL) != 0) {
456 error = SET_ERROR(EINVAL);
460 has_feature = B_TRUE;
463 case ZPOOL_PROP_VERSION:
464 error = nvpair_value_uint64(elem, &intval);
466 (intval < spa_version(spa) ||
467 intval > SPA_VERSION_BEFORE_FEATURES ||
469 error = SET_ERROR(EINVAL);
472 case ZPOOL_PROP_DELEGATION:
473 case ZPOOL_PROP_AUTOREPLACE:
474 case ZPOOL_PROP_LISTSNAPS:
475 case ZPOOL_PROP_AUTOEXPAND:
476 error = nvpair_value_uint64(elem, &intval);
477 if (!error && intval > 1)
478 error = SET_ERROR(EINVAL);
481 case ZPOOL_PROP_BOOTFS:
483 * If the pool version is less than SPA_VERSION_BOOTFS,
484 * or the pool is still being created (version == 0),
485 * the bootfs property cannot be set.
487 if (spa_version(spa) < SPA_VERSION_BOOTFS) {
488 error = SET_ERROR(ENOTSUP);
493 * Make sure the vdev config is bootable
495 if (!vdev_is_bootable(spa->spa_root_vdev)) {
496 error = SET_ERROR(ENOTSUP);
502 error = nvpair_value_string(elem, &strval);
508 if (strval == NULL || strval[0] == '\0') {
509 objnum = zpool_prop_default_numeric(
514 if (error = dmu_objset_hold(strval, FTAG, &os))
517 /* Must be ZPL and not gzip compressed. */
519 if (dmu_objset_type(os) != DMU_OST_ZFS) {
520 error = SET_ERROR(ENOTSUP);
522 dsl_prop_get_int_ds(dmu_objset_ds(os),
523 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
525 !BOOTFS_COMPRESS_VALID(compress)) {
526 error = SET_ERROR(ENOTSUP);
528 objnum = dmu_objset_id(os);
530 dmu_objset_rele(os, FTAG);
534 case ZPOOL_PROP_FAILUREMODE:
535 error = nvpair_value_uint64(elem, &intval);
536 if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
537 intval > ZIO_FAILURE_MODE_PANIC))
538 error = SET_ERROR(EINVAL);
541 * This is a special case which only occurs when
542 * the pool has completely failed. This allows
543 * the user to change the in-core failmode property
544 * without syncing it out to disk (I/Os might
545 * currently be blocked). We do this by returning
546 * EIO to the caller (spa_prop_set) to trick it
547 * into thinking we encountered a property validation
550 if (!error && spa_suspended(spa)) {
551 spa->spa_failmode = intval;
552 error = SET_ERROR(EIO);
556 case ZPOOL_PROP_CACHEFILE:
557 if ((error = nvpair_value_string(elem, &strval)) != 0)
560 if (strval[0] == '\0')
563 if (strcmp(strval, "none") == 0)
566 if (strval[0] != '/') {
567 error = SET_ERROR(EINVAL);
571 slash = strrchr(strval, '/');
572 ASSERT(slash != NULL);
574 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
575 strcmp(slash, "/..") == 0)
576 error = SET_ERROR(EINVAL);
579 case ZPOOL_PROP_COMMENT:
580 if ((error = nvpair_value_string(elem, &strval)) != 0)
582 for (check = strval; *check != '\0'; check++) {
584 * The kernel doesn't have an easy isprint()
585 * check. For this kernel check, we merely
586 * check ASCII apart from DEL. Fix this if
587 * there is an easy-to-use kernel isprint().
589 if (*check >= 0x7f) {
590 error = SET_ERROR(EINVAL);
595 if (strlen(strval) > ZPROP_MAX_COMMENT)
599 case ZPOOL_PROP_DEDUPDITTO:
600 if (spa_version(spa) < SPA_VERSION_DEDUP)
601 error = SET_ERROR(ENOTSUP);
603 error = nvpair_value_uint64(elem, &intval);
605 intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
606 error = SET_ERROR(EINVAL);
614 if (!error && reset_bootfs) {
615 error = nvlist_remove(props,
616 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
619 error = nvlist_add_uint64(props,
620 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
628 spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
631 spa_config_dirent_t *dp;
633 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
637 dp = kmem_alloc(sizeof (spa_config_dirent_t),
640 if (cachefile[0] == '\0')
641 dp->scd_path = spa_strdup(spa_config_path);
642 else if (strcmp(cachefile, "none") == 0)
645 dp->scd_path = spa_strdup(cachefile);
647 list_insert_head(&spa->spa_config_list, dp);
649 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
653 spa_prop_set(spa_t *spa, nvlist_t *nvp)
656 nvpair_t *elem = NULL;
657 boolean_t need_sync = B_FALSE;
659 if ((error = spa_prop_validate(spa, nvp)) != 0)
662 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
663 zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem));
665 if (prop == ZPOOL_PROP_CACHEFILE ||
666 prop == ZPOOL_PROP_ALTROOT ||
667 prop == ZPOOL_PROP_READONLY)
670 if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) {
673 if (prop == ZPOOL_PROP_VERSION) {
674 VERIFY(nvpair_value_uint64(elem, &ver) == 0);
676 ASSERT(zpool_prop_feature(nvpair_name(elem)));
677 ver = SPA_VERSION_FEATURES;
681 /* Save time if the version is already set. */
682 if (ver == spa_version(spa))
686 * In addition to the pool directory object, we might
687 * create the pool properties object, the features for
688 * read object, the features for write object, or the
689 * feature descriptions object.
691 error = dsl_sync_task(spa->spa_name, NULL,
692 spa_sync_version, &ver,
693 6, ZFS_SPACE_CHECK_RESERVED);
704 return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props,
705 nvp, 6, ZFS_SPACE_CHECK_RESERVED));
712 * If the bootfs property value is dsobj, clear it.
715 spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
717 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
718 VERIFY(zap_remove(spa->spa_meta_objset,
719 spa->spa_pool_props_object,
720 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
727 spa_change_guid_check(void *arg, dmu_tx_t *tx)
729 uint64_t *newguid = arg;
730 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
731 vdev_t *rvd = spa->spa_root_vdev;
734 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
735 vdev_state = rvd->vdev_state;
736 spa_config_exit(spa, SCL_STATE, FTAG);
738 if (vdev_state != VDEV_STATE_HEALTHY)
739 return (SET_ERROR(ENXIO));
741 ASSERT3U(spa_guid(spa), !=, *newguid);
747 spa_change_guid_sync(void *arg, dmu_tx_t *tx)
749 uint64_t *newguid = arg;
750 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
752 vdev_t *rvd = spa->spa_root_vdev;
754 oldguid = spa_guid(spa);
756 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
757 rvd->vdev_guid = *newguid;
758 rvd->vdev_guid_sum += (*newguid - oldguid);
759 vdev_config_dirty(rvd);
760 spa_config_exit(spa, SCL_STATE, FTAG);
762 spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu",
767 * Change the GUID for the pool. This is done so that we can later
768 * re-import a pool built from a clone of our own vdevs. We will modify
769 * the root vdev's guid, our own pool guid, and then mark all of our
770 * vdevs dirty. Note that we must make sure that all our vdevs are
771 * online when we do this, or else any vdevs that weren't present
772 * would be orphaned from our pool. We are also going to issue a
773 * sysevent to update any watchers.
776 spa_change_guid(spa_t *spa)
781 mutex_enter(&spa->spa_vdev_top_lock);
782 mutex_enter(&spa_namespace_lock);
783 guid = spa_generate_guid(NULL);
785 error = dsl_sync_task(spa->spa_name, spa_change_guid_check,
786 spa_change_guid_sync, &guid, 5, ZFS_SPACE_CHECK_RESERVED);
789 spa_config_sync(spa, B_FALSE, B_TRUE);
790 spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID);
793 mutex_exit(&spa_namespace_lock);
794 mutex_exit(&spa->spa_vdev_top_lock);
800 * ==========================================================================
801 * SPA state manipulation (open/create/destroy/import/export)
802 * ==========================================================================
806 spa_error_entry_compare(const void *a, const void *b)
808 spa_error_entry_t *sa = (spa_error_entry_t *)a;
809 spa_error_entry_t *sb = (spa_error_entry_t *)b;
812 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
813 sizeof (zbookmark_phys_t));
824 * Utility function which retrieves copies of the current logs and
825 * re-initializes them in the process.
828 spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
830 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
832 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
833 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
835 avl_create(&spa->spa_errlist_scrub,
836 spa_error_entry_compare, sizeof (spa_error_entry_t),
837 offsetof(spa_error_entry_t, se_avl));
838 avl_create(&spa->spa_errlist_last,
839 spa_error_entry_compare, sizeof (spa_error_entry_t),
840 offsetof(spa_error_entry_t, se_avl));
844 spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
846 const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
847 enum zti_modes mode = ztip->zti_mode;
848 uint_t value = ztip->zti_value;
849 uint_t count = ztip->zti_count;
850 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
853 boolean_t batch = B_FALSE;
855 if (mode == ZTI_MODE_NULL) {
857 tqs->stqs_taskq = NULL;
861 ASSERT3U(count, >, 0);
863 tqs->stqs_count = count;
864 tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP);
868 ASSERT3U(value, >=, 1);
869 value = MAX(value, 1);
874 flags |= TASKQ_THREADS_CPU_PCT;
875 value = zio_taskq_batch_pct;
879 panic("unrecognized mode for %s_%s taskq (%u:%u) in "
881 zio_type_name[t], zio_taskq_types[q], mode, value);
885 for (uint_t i = 0; i < count; i++) {
889 (void) snprintf(name, sizeof (name), "%s_%s_%u",
890 zio_type_name[t], zio_taskq_types[q], i);
892 (void) snprintf(name, sizeof (name), "%s_%s",
893 zio_type_name[t], zio_taskq_types[q]);
897 if (zio_taskq_sysdc && spa->spa_proc != &p0) {
899 flags |= TASKQ_DC_BATCH;
901 tq = taskq_create_sysdc(name, value, 50, INT_MAX,
902 spa->spa_proc, zio_taskq_basedc, flags);
905 pri_t pri = maxclsyspri;
907 * The write issue taskq can be extremely CPU
908 * intensive. Run it at slightly lower priority
909 * than the other taskqs.
911 if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE)
914 tq = taskq_create_proc(name, value, pri, 50,
915 INT_MAX, spa->spa_proc, flags);
920 tqs->stqs_taskq[i] = tq;
925 spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
927 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
929 if (tqs->stqs_taskq == NULL) {
930 ASSERT0(tqs->stqs_count);
934 for (uint_t i = 0; i < tqs->stqs_count; i++) {
935 ASSERT3P(tqs->stqs_taskq[i], !=, NULL);
936 taskq_destroy(tqs->stqs_taskq[i]);
939 kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *));
940 tqs->stqs_taskq = NULL;
944 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
945 * Note that a type may have multiple discrete taskqs to avoid lock contention
946 * on the taskq itself. In that case we choose which taskq at random by using
947 * the low bits of gethrtime().
950 spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q,
951 task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent)
953 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
956 ASSERT3P(tqs->stqs_taskq, !=, NULL);
957 ASSERT3U(tqs->stqs_count, !=, 0);
959 if (tqs->stqs_count == 1) {
960 tq = tqs->stqs_taskq[0];
963 tq = tqs->stqs_taskq[cpu_ticks() % tqs->stqs_count];
965 tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count];
969 taskq_dispatch_ent(tq, func, arg, flags, ent);
973 spa_create_zio_taskqs(spa_t *spa)
975 for (int t = 0; t < ZIO_TYPES; t++) {
976 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
977 spa_taskqs_init(spa, t, q);
985 spa_thread(void *arg)
990 user_t *pu = PTOU(curproc);
992 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
995 ASSERT(curproc != &p0);
996 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
997 "zpool-%s", spa->spa_name);
998 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
1001 /* bind this thread to the requested psrset */
1002 if (zio_taskq_psrset_bind != PS_NONE) {
1004 mutex_enter(&cpu_lock);
1005 mutex_enter(&pidlock);
1006 mutex_enter(&curproc->p_lock);
1008 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
1009 0, NULL, NULL) == 0) {
1010 curthread->t_bind_pset = zio_taskq_psrset_bind;
1013 "Couldn't bind process for zfs pool \"%s\" to "
1014 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
1017 mutex_exit(&curproc->p_lock);
1018 mutex_exit(&pidlock);
1019 mutex_exit(&cpu_lock);
1025 if (zio_taskq_sysdc) {
1026 sysdc_thread_enter(curthread, 100, 0);
1030 spa->spa_proc = curproc;
1031 spa->spa_did = curthread->t_did;
1033 spa_create_zio_taskqs(spa);
1035 mutex_enter(&spa->spa_proc_lock);
1036 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
1038 spa->spa_proc_state = SPA_PROC_ACTIVE;
1039 cv_broadcast(&spa->spa_proc_cv);
1041 CALLB_CPR_SAFE_BEGIN(&cprinfo);
1042 while (spa->spa_proc_state == SPA_PROC_ACTIVE)
1043 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1044 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
1046 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
1047 spa->spa_proc_state = SPA_PROC_GONE;
1048 spa->spa_proc = &p0;
1049 cv_broadcast(&spa->spa_proc_cv);
1050 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */
1052 mutex_enter(&curproc->p_lock);
1055 #endif /* SPA_PROCESS */
1059 * Activate an uninitialized pool.
1062 spa_activate(spa_t *spa, int mode)
1064 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
1066 spa->spa_state = POOL_STATE_ACTIVE;
1067 spa->spa_mode = mode;
1069 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
1070 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
1072 /* Try to create a covering process */
1073 mutex_enter(&spa->spa_proc_lock);
1074 ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
1075 ASSERT(spa->spa_proc == &p0);
1079 /* Only create a process if we're going to be around a while. */
1080 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
1081 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
1083 spa->spa_proc_state = SPA_PROC_CREATED;
1084 while (spa->spa_proc_state == SPA_PROC_CREATED) {
1085 cv_wait(&spa->spa_proc_cv,
1086 &spa->spa_proc_lock);
1088 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1089 ASSERT(spa->spa_proc != &p0);
1090 ASSERT(spa->spa_did != 0);
1094 "Couldn't create process for zfs pool \"%s\"\n",
1099 #endif /* SPA_PROCESS */
1100 mutex_exit(&spa->spa_proc_lock);
1102 /* If we didn't create a process, we need to create our taskqs. */
1103 ASSERT(spa->spa_proc == &p0);
1104 if (spa->spa_proc == &p0) {
1105 spa_create_zio_taskqs(spa);
1109 * Start TRIM thread.
1111 trim_thread_create(spa);
1113 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
1114 offsetof(vdev_t, vdev_config_dirty_node));
1115 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
1116 offsetof(vdev_t, vdev_state_dirty_node));
1118 txg_list_create(&spa->spa_vdev_txg_list,
1119 offsetof(struct vdev, vdev_txg_node));
1121 avl_create(&spa->spa_errlist_scrub,
1122 spa_error_entry_compare, sizeof (spa_error_entry_t),
1123 offsetof(spa_error_entry_t, se_avl));
1124 avl_create(&spa->spa_errlist_last,
1125 spa_error_entry_compare, sizeof (spa_error_entry_t),
1126 offsetof(spa_error_entry_t, se_avl));
1130 * Opposite of spa_activate().
1133 spa_deactivate(spa_t *spa)
1135 ASSERT(spa->spa_sync_on == B_FALSE);
1136 ASSERT(spa->spa_dsl_pool == NULL);
1137 ASSERT(spa->spa_root_vdev == NULL);
1138 ASSERT(spa->spa_async_zio_root == NULL);
1139 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
1142 * Stop TRIM thread in case spa_unload() wasn't called directly
1143 * before spa_deactivate().
1145 trim_thread_destroy(spa);
1147 txg_list_destroy(&spa->spa_vdev_txg_list);
1149 list_destroy(&spa->spa_config_dirty_list);
1150 list_destroy(&spa->spa_state_dirty_list);
1152 for (int t = 0; t < ZIO_TYPES; t++) {
1153 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1154 spa_taskqs_fini(spa, t, q);
1158 metaslab_class_destroy(spa->spa_normal_class);
1159 spa->spa_normal_class = NULL;
1161 metaslab_class_destroy(spa->spa_log_class);
1162 spa->spa_log_class = NULL;
1165 * If this was part of an import or the open otherwise failed, we may
1166 * still have errors left in the queues. Empty them just in case.
1168 spa_errlog_drain(spa);
1170 avl_destroy(&spa->spa_errlist_scrub);
1171 avl_destroy(&spa->spa_errlist_last);
1173 spa->spa_state = POOL_STATE_UNINITIALIZED;
1175 mutex_enter(&spa->spa_proc_lock);
1176 if (spa->spa_proc_state != SPA_PROC_NONE) {
1177 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1178 spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1179 cv_broadcast(&spa->spa_proc_cv);
1180 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1181 ASSERT(spa->spa_proc != &p0);
1182 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1184 ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1185 spa->spa_proc_state = SPA_PROC_NONE;
1187 ASSERT(spa->spa_proc == &p0);
1188 mutex_exit(&spa->spa_proc_lock);
1192 * We want to make sure spa_thread() has actually exited the ZFS
1193 * module, so that the module can't be unloaded out from underneath
1196 if (spa->spa_did != 0) {
1197 thread_join(spa->spa_did);
1200 #endif /* SPA_PROCESS */
1204 * Verify a pool configuration, and construct the vdev tree appropriately. This
1205 * will create all the necessary vdevs in the appropriate layout, with each vdev
1206 * in the CLOSED state. This will prep the pool before open/creation/import.
1207 * All vdev validation is done by the vdev_alloc() routine.
1210 spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1211 uint_t id, int atype)
1217 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1220 if ((*vdp)->vdev_ops->vdev_op_leaf)
1223 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1226 if (error == ENOENT)
1232 return (SET_ERROR(EINVAL));
1235 for (int c = 0; c < children; c++) {
1237 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1245 ASSERT(*vdp != NULL);
1251 * Opposite of spa_load().
1254 spa_unload(spa_t *spa)
1258 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1263 trim_thread_destroy(spa);
1268 spa_async_suspend(spa);
1273 if (spa->spa_sync_on) {
1274 txg_sync_stop(spa->spa_dsl_pool);
1275 spa->spa_sync_on = B_FALSE;
1279 * Wait for any outstanding async I/O to complete.
1281 if (spa->spa_async_zio_root != NULL) {
1282 (void) zio_wait(spa->spa_async_zio_root);
1283 spa->spa_async_zio_root = NULL;
1286 bpobj_close(&spa->spa_deferred_bpobj);
1288 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1293 if (spa->spa_root_vdev)
1294 vdev_free(spa->spa_root_vdev);
1295 ASSERT(spa->spa_root_vdev == NULL);
1298 * Close the dsl pool.
1300 if (spa->spa_dsl_pool) {
1301 dsl_pool_close(spa->spa_dsl_pool);
1302 spa->spa_dsl_pool = NULL;
1303 spa->spa_meta_objset = NULL;
1310 * Drop and purge level 2 cache
1312 spa_l2cache_drop(spa);
1314 for (i = 0; i < spa->spa_spares.sav_count; i++)
1315 vdev_free(spa->spa_spares.sav_vdevs[i]);
1316 if (spa->spa_spares.sav_vdevs) {
1317 kmem_free(spa->spa_spares.sav_vdevs,
1318 spa->spa_spares.sav_count * sizeof (void *));
1319 spa->spa_spares.sav_vdevs = NULL;
1321 if (spa->spa_spares.sav_config) {
1322 nvlist_free(spa->spa_spares.sav_config);
1323 spa->spa_spares.sav_config = NULL;
1325 spa->spa_spares.sav_count = 0;
1327 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1328 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1329 vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1331 if (spa->spa_l2cache.sav_vdevs) {
1332 kmem_free(spa->spa_l2cache.sav_vdevs,
1333 spa->spa_l2cache.sav_count * sizeof (void *));
1334 spa->spa_l2cache.sav_vdevs = NULL;
1336 if (spa->spa_l2cache.sav_config) {
1337 nvlist_free(spa->spa_l2cache.sav_config);
1338 spa->spa_l2cache.sav_config = NULL;
1340 spa->spa_l2cache.sav_count = 0;
1342 spa->spa_async_suspended = 0;
1344 if (spa->spa_comment != NULL) {
1345 spa_strfree(spa->spa_comment);
1346 spa->spa_comment = NULL;
1349 spa_config_exit(spa, SCL_ALL, FTAG);
1353 * Load (or re-load) the current list of vdevs describing the active spares for
1354 * this pool. When this is called, we have some form of basic information in
1355 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1356 * then re-generate a more complete list including status information.
1359 spa_load_spares(spa_t *spa)
1366 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1369 * First, close and free any existing spare vdevs.
1371 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1372 vd = spa->spa_spares.sav_vdevs[i];
1374 /* Undo the call to spa_activate() below */
1375 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1376 B_FALSE)) != NULL && tvd->vdev_isspare)
1377 spa_spare_remove(tvd);
1382 if (spa->spa_spares.sav_vdevs)
1383 kmem_free(spa->spa_spares.sav_vdevs,
1384 spa->spa_spares.sav_count * sizeof (void *));
1386 if (spa->spa_spares.sav_config == NULL)
1389 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1390 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1392 spa->spa_spares.sav_count = (int)nspares;
1393 spa->spa_spares.sav_vdevs = NULL;
1399 * Construct the array of vdevs, opening them to get status in the
1400 * process. For each spare, there is potentially two different vdev_t
1401 * structures associated with it: one in the list of spares (used only
1402 * for basic validation purposes) and one in the active vdev
1403 * configuration (if it's spared in). During this phase we open and
1404 * validate each vdev on the spare list. If the vdev also exists in the
1405 * active configuration, then we also mark this vdev as an active spare.
1407 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1409 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1410 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1411 VDEV_ALLOC_SPARE) == 0);
1414 spa->spa_spares.sav_vdevs[i] = vd;
1416 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1417 B_FALSE)) != NULL) {
1418 if (!tvd->vdev_isspare)
1422 * We only mark the spare active if we were successfully
1423 * able to load the vdev. Otherwise, importing a pool
1424 * with a bad active spare would result in strange
1425 * behavior, because multiple pool would think the spare
1426 * is actively in use.
1428 * There is a vulnerability here to an equally bizarre
1429 * circumstance, where a dead active spare is later
1430 * brought back to life (onlined or otherwise). Given
1431 * the rarity of this scenario, and the extra complexity
1432 * it adds, we ignore the possibility.
1434 if (!vdev_is_dead(tvd))
1435 spa_spare_activate(tvd);
1439 vd->vdev_aux = &spa->spa_spares;
1441 if (vdev_open(vd) != 0)
1444 if (vdev_validate_aux(vd) == 0)
1449 * Recompute the stashed list of spares, with status information
1452 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1453 DATA_TYPE_NVLIST_ARRAY) == 0);
1455 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1457 for (i = 0; i < spa->spa_spares.sav_count; i++)
1458 spares[i] = vdev_config_generate(spa,
1459 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1460 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1461 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1462 for (i = 0; i < spa->spa_spares.sav_count; i++)
1463 nvlist_free(spares[i]);
1464 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1468 * Load (or re-load) the current list of vdevs describing the active l2cache for
1469 * this pool. When this is called, we have some form of basic information in
1470 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1471 * then re-generate a more complete list including status information.
1472 * Devices which are already active have their details maintained, and are
1476 spa_load_l2cache(spa_t *spa)
1480 int i, j, oldnvdevs;
1482 vdev_t *vd, **oldvdevs, **newvdevs;
1483 spa_aux_vdev_t *sav = &spa->spa_l2cache;
1485 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1487 if (sav->sav_config != NULL) {
1488 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1489 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1490 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1496 oldvdevs = sav->sav_vdevs;
1497 oldnvdevs = sav->sav_count;
1498 sav->sav_vdevs = NULL;
1502 * Process new nvlist of vdevs.
1504 for (i = 0; i < nl2cache; i++) {
1505 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1509 for (j = 0; j < oldnvdevs; j++) {
1511 if (vd != NULL && guid == vd->vdev_guid) {
1513 * Retain previous vdev for add/remove ops.
1521 if (newvdevs[i] == NULL) {
1525 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1526 VDEV_ALLOC_L2CACHE) == 0);
1531 * Commit this vdev as an l2cache device,
1532 * even if it fails to open.
1534 spa_l2cache_add(vd);
1539 spa_l2cache_activate(vd);
1541 if (vdev_open(vd) != 0)
1544 (void) vdev_validate_aux(vd);
1546 if (!vdev_is_dead(vd))
1547 l2arc_add_vdev(spa, vd);
1552 * Purge vdevs that were dropped
1554 for (i = 0; i < oldnvdevs; i++) {
1559 ASSERT(vd->vdev_isl2cache);
1561 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1562 pool != 0ULL && l2arc_vdev_present(vd))
1563 l2arc_remove_vdev(vd);
1564 vdev_clear_stats(vd);
1570 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1572 if (sav->sav_config == NULL)
1575 sav->sav_vdevs = newvdevs;
1576 sav->sav_count = (int)nl2cache;
1579 * Recompute the stashed list of l2cache devices, with status
1580 * information this time.
1582 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1583 DATA_TYPE_NVLIST_ARRAY) == 0);
1585 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1586 for (i = 0; i < sav->sav_count; i++)
1587 l2cache[i] = vdev_config_generate(spa,
1588 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1589 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1590 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1592 for (i = 0; i < sav->sav_count; i++)
1593 nvlist_free(l2cache[i]);
1595 kmem_free(l2cache, sav->sav_count * sizeof (void *));
1599 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1602 char *packed = NULL;
1607 error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db);
1610 nvsize = *(uint64_t *)db->db_data;
1611 dmu_buf_rele(db, FTAG);
1613 packed = kmem_alloc(nvsize, KM_SLEEP);
1614 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1617 error = nvlist_unpack(packed, nvsize, value, 0);
1618 kmem_free(packed, nvsize);
1624 * Checks to see if the given vdev could not be opened, in which case we post a
1625 * sysevent to notify the autoreplace code that the device has been removed.
1628 spa_check_removed(vdev_t *vd)
1630 for (int c = 0; c < vd->vdev_children; c++)
1631 spa_check_removed(vd->vdev_child[c]);
1633 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) &&
1635 zfs_post_autoreplace(vd->vdev_spa, vd);
1636 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1641 * Validate the current config against the MOS config
1644 spa_config_valid(spa_t *spa, nvlist_t *config)
1646 vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1649 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1651 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1652 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1654 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1657 * If we're doing a normal import, then build up any additional
1658 * diagnostic information about missing devices in this config.
1659 * We'll pass this up to the user for further processing.
1661 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1662 nvlist_t **child, *nv;
1665 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1667 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1669 for (int c = 0; c < rvd->vdev_children; c++) {
1670 vdev_t *tvd = rvd->vdev_child[c];
1671 vdev_t *mtvd = mrvd->vdev_child[c];
1673 if (tvd->vdev_ops == &vdev_missing_ops &&
1674 mtvd->vdev_ops != &vdev_missing_ops &&
1676 child[idx++] = vdev_config_generate(spa, mtvd,
1681 VERIFY(nvlist_add_nvlist_array(nv,
1682 ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1683 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1684 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1686 for (int i = 0; i < idx; i++)
1687 nvlist_free(child[i]);
1690 kmem_free(child, rvd->vdev_children * sizeof (char **));
1694 * Compare the root vdev tree with the information we have
1695 * from the MOS config (mrvd). Check each top-level vdev
1696 * with the corresponding MOS config top-level (mtvd).
1698 for (int c = 0; c < rvd->vdev_children; c++) {
1699 vdev_t *tvd = rvd->vdev_child[c];
1700 vdev_t *mtvd = mrvd->vdev_child[c];
1703 * Resolve any "missing" vdevs in the current configuration.
1704 * If we find that the MOS config has more accurate information
1705 * about the top-level vdev then use that vdev instead.
1707 if (tvd->vdev_ops == &vdev_missing_ops &&
1708 mtvd->vdev_ops != &vdev_missing_ops) {
1710 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1714 * Device specific actions.
1716 if (mtvd->vdev_islog) {
1717 spa_set_log_state(spa, SPA_LOG_CLEAR);
1720 * XXX - once we have 'readonly' pool
1721 * support we should be able to handle
1722 * missing data devices by transitioning
1723 * the pool to readonly.
1729 * Swap the missing vdev with the data we were
1730 * able to obtain from the MOS config.
1732 vdev_remove_child(rvd, tvd);
1733 vdev_remove_child(mrvd, mtvd);
1735 vdev_add_child(rvd, mtvd);
1736 vdev_add_child(mrvd, tvd);
1738 spa_config_exit(spa, SCL_ALL, FTAG);
1740 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1743 } else if (mtvd->vdev_islog) {
1745 * Load the slog device's state from the MOS config
1746 * since it's possible that the label does not
1747 * contain the most up-to-date information.
1749 vdev_load_log_state(tvd, mtvd);
1754 spa_config_exit(spa, SCL_ALL, FTAG);
1757 * Ensure we were able to validate the config.
1759 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1763 * Check for missing log devices
1766 spa_check_logs(spa_t *spa)
1768 boolean_t rv = B_FALSE;
1770 switch (spa->spa_log_state) {
1771 case SPA_LOG_MISSING:
1772 /* need to recheck in case slog has been restored */
1773 case SPA_LOG_UNKNOWN:
1774 rv = (dmu_objset_find(spa->spa_name, zil_check_log_chain,
1775 NULL, DS_FIND_CHILDREN) != 0);
1777 spa_set_log_state(spa, SPA_LOG_MISSING);
1784 spa_passivate_log(spa_t *spa)
1786 vdev_t *rvd = spa->spa_root_vdev;
1787 boolean_t slog_found = B_FALSE;
1789 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1791 if (!spa_has_slogs(spa))
1794 for (int c = 0; c < rvd->vdev_children; c++) {
1795 vdev_t *tvd = rvd->vdev_child[c];
1796 metaslab_group_t *mg = tvd->vdev_mg;
1798 if (tvd->vdev_islog) {
1799 metaslab_group_passivate(mg);
1800 slog_found = B_TRUE;
1804 return (slog_found);
1808 spa_activate_log(spa_t *spa)
1810 vdev_t *rvd = spa->spa_root_vdev;
1812 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1814 for (int c = 0; c < rvd->vdev_children; c++) {
1815 vdev_t *tvd = rvd->vdev_child[c];
1816 metaslab_group_t *mg = tvd->vdev_mg;
1818 if (tvd->vdev_islog)
1819 metaslab_group_activate(mg);
1824 spa_offline_log(spa_t *spa)
1828 error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1829 NULL, DS_FIND_CHILDREN);
1832 * We successfully offlined the log device, sync out the
1833 * current txg so that the "stubby" block can be removed
1836 txg_wait_synced(spa->spa_dsl_pool, 0);
1842 spa_aux_check_removed(spa_aux_vdev_t *sav)
1846 for (i = 0; i < sav->sav_count; i++)
1847 spa_check_removed(sav->sav_vdevs[i]);
1851 spa_claim_notify(zio_t *zio)
1853 spa_t *spa = zio->io_spa;
1858 mutex_enter(&spa->spa_props_lock); /* any mutex will do */
1859 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1860 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1861 mutex_exit(&spa->spa_props_lock);
1864 typedef struct spa_load_error {
1865 uint64_t sle_meta_count;
1866 uint64_t sle_data_count;
1870 spa_load_verify_done(zio_t *zio)
1872 blkptr_t *bp = zio->io_bp;
1873 spa_load_error_t *sle = zio->io_private;
1874 dmu_object_type_t type = BP_GET_TYPE(bp);
1875 int error = zio->io_error;
1876 spa_t *spa = zio->io_spa;
1879 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1880 type != DMU_OT_INTENT_LOG)
1881 atomic_add_64(&sle->sle_meta_count, 1);
1883 atomic_add_64(&sle->sle_data_count, 1);
1885 zio_data_buf_free(zio->io_data, zio->io_size);
1887 mutex_enter(&spa->spa_scrub_lock);
1888 spa->spa_scrub_inflight--;
1889 cv_broadcast(&spa->spa_scrub_io_cv);
1890 mutex_exit(&spa->spa_scrub_lock);
1894 * Maximum number of concurrent scrub i/os to create while verifying
1895 * a pool while importing it.
1897 int spa_load_verify_maxinflight = 10000;
1898 boolean_t spa_load_verify_metadata = B_TRUE;
1899 boolean_t spa_load_verify_data = B_TRUE;
1901 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_maxinflight, CTLFLAG_RWTUN,
1902 &spa_load_verify_maxinflight, 0,
1903 "Maximum number of concurrent scrub I/Os to create while verifying a "
1904 "pool while importing it");
1906 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_metadata, CTLFLAG_RWTUN,
1907 &spa_load_verify_metadata, 0,
1908 "Check metadata on import?");
1910 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_data, CTLFLAG_RWTUN,
1911 &spa_load_verify_data, 0,
1912 "Check user data on import?");
1916 spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1917 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
1919 if (BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp))
1922 * Note: normally this routine will not be called if
1923 * spa_load_verify_metadata is not set. However, it may be useful
1924 * to manually set the flag after the traversal has begun.
1926 if (!spa_load_verify_metadata)
1928 if (BP_GET_BUFC_TYPE(bp) == ARC_BUFC_DATA && !spa_load_verify_data)
1932 size_t size = BP_GET_PSIZE(bp);
1933 void *data = zio_data_buf_alloc(size);
1935 mutex_enter(&spa->spa_scrub_lock);
1936 while (spa->spa_scrub_inflight >= spa_load_verify_maxinflight)
1937 cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
1938 spa->spa_scrub_inflight++;
1939 mutex_exit(&spa->spa_scrub_lock);
1941 zio_nowait(zio_read(rio, spa, bp, data, size,
1942 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1943 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1944 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1949 spa_load_verify(spa_t *spa)
1952 spa_load_error_t sle = { 0 };
1953 zpool_rewind_policy_t policy;
1954 boolean_t verify_ok = B_FALSE;
1957 zpool_get_rewind_policy(spa->spa_config, &policy);
1959 if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1962 rio = zio_root(spa, NULL, &sle,
1963 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1965 if (spa_load_verify_metadata) {
1966 error = traverse_pool(spa, spa->spa_verify_min_txg,
1967 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA,
1968 spa_load_verify_cb, rio);
1971 (void) zio_wait(rio);
1973 spa->spa_load_meta_errors = sle.sle_meta_count;
1974 spa->spa_load_data_errors = sle.sle_data_count;
1976 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1977 sle.sle_data_count <= policy.zrp_maxdata) {
1981 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1982 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1984 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1985 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1986 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1987 VERIFY(nvlist_add_int64(spa->spa_load_info,
1988 ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1989 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1990 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1992 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1996 if (error != ENXIO && error != EIO)
1997 error = SET_ERROR(EIO);
2001 return (verify_ok ? 0 : EIO);
2005 * Find a value in the pool props object.
2008 spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
2010 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
2011 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
2015 * Find a value in the pool directory object.
2018 spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
2020 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
2021 name, sizeof (uint64_t), 1, val));
2025 spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
2027 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
2032 * Fix up config after a partly-completed split. This is done with the
2033 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
2034 * pool have that entry in their config, but only the splitting one contains
2035 * a list of all the guids of the vdevs that are being split off.
2037 * This function determines what to do with that list: either rejoin
2038 * all the disks to the pool, or complete the splitting process. To attempt
2039 * the rejoin, each disk that is offlined is marked online again, and
2040 * we do a reopen() call. If the vdev label for every disk that was
2041 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
2042 * then we call vdev_split() on each disk, and complete the split.
2044 * Otherwise we leave the config alone, with all the vdevs in place in
2045 * the original pool.
2048 spa_try_repair(spa_t *spa, nvlist_t *config)
2055 boolean_t attempt_reopen;
2057 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
2060 /* check that the config is complete */
2061 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
2062 &glist, &gcount) != 0)
2065 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
2067 /* attempt to online all the vdevs & validate */
2068 attempt_reopen = B_TRUE;
2069 for (i = 0; i < gcount; i++) {
2070 if (glist[i] == 0) /* vdev is hole */
2073 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
2074 if (vd[i] == NULL) {
2076 * Don't bother attempting to reopen the disks;
2077 * just do the split.
2079 attempt_reopen = B_FALSE;
2081 /* attempt to re-online it */
2082 vd[i]->vdev_offline = B_FALSE;
2086 if (attempt_reopen) {
2087 vdev_reopen(spa->spa_root_vdev);
2089 /* check each device to see what state it's in */
2090 for (extracted = 0, i = 0; i < gcount; i++) {
2091 if (vd[i] != NULL &&
2092 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
2099 * If every disk has been moved to the new pool, or if we never
2100 * even attempted to look at them, then we split them off for
2103 if (!attempt_reopen || gcount == extracted) {
2104 for (i = 0; i < gcount; i++)
2107 vdev_reopen(spa->spa_root_vdev);
2110 kmem_free(vd, gcount * sizeof (vdev_t *));
2114 spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
2115 boolean_t mosconfig)
2117 nvlist_t *config = spa->spa_config;
2118 char *ereport = FM_EREPORT_ZFS_POOL;
2124 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
2125 return (SET_ERROR(EINVAL));
2127 ASSERT(spa->spa_comment == NULL);
2128 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
2129 spa->spa_comment = spa_strdup(comment);
2132 * Versioning wasn't explicitly added to the label until later, so if
2133 * it's not present treat it as the initial version.
2135 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
2136 &spa->spa_ubsync.ub_version) != 0)
2137 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
2139 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
2140 &spa->spa_config_txg);
2142 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
2143 spa_guid_exists(pool_guid, 0)) {
2144 error = SET_ERROR(EEXIST);
2146 spa->spa_config_guid = pool_guid;
2148 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
2150 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
2154 nvlist_free(spa->spa_load_info);
2155 spa->spa_load_info = fnvlist_alloc();
2157 gethrestime(&spa->spa_loaded_ts);
2158 error = spa_load_impl(spa, pool_guid, config, state, type,
2159 mosconfig, &ereport);
2162 spa->spa_minref = refcount_count(&spa->spa_refcount);
2164 if (error != EEXIST) {
2165 spa->spa_loaded_ts.tv_sec = 0;
2166 spa->spa_loaded_ts.tv_nsec = 0;
2168 if (error != EBADF) {
2169 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
2172 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2179 * Load an existing storage pool, using the pool's builtin spa_config as a
2180 * source of configuration information.
2183 spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
2184 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
2188 nvlist_t *nvroot = NULL;
2191 uberblock_t *ub = &spa->spa_uberblock;
2192 uint64_t children, config_cache_txg = spa->spa_config_txg;
2193 int orig_mode = spa->spa_mode;
2196 boolean_t missing_feat_write = B_FALSE;
2199 * If this is an untrusted config, access the pool in read-only mode.
2200 * This prevents things like resilvering recently removed devices.
2203 spa->spa_mode = FREAD;
2205 ASSERT(MUTEX_HELD(&spa_namespace_lock));
2207 spa->spa_load_state = state;
2209 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
2210 return (SET_ERROR(EINVAL));
2212 parse = (type == SPA_IMPORT_EXISTING ?
2213 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2216 * Create "The Godfather" zio to hold all async IOs
2218 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2219 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2222 * Parse the configuration into a vdev tree. We explicitly set the
2223 * value that will be returned by spa_version() since parsing the
2224 * configuration requires knowing the version number.
2226 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2227 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
2228 spa_config_exit(spa, SCL_ALL, FTAG);
2233 ASSERT(spa->spa_root_vdev == rvd);
2235 if (type != SPA_IMPORT_ASSEMBLE) {
2236 ASSERT(spa_guid(spa) == pool_guid);
2240 * Try to open all vdevs, loading each label in the process.
2242 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2243 error = vdev_open(rvd);
2244 spa_config_exit(spa, SCL_ALL, FTAG);
2249 * We need to validate the vdev labels against the configuration that
2250 * we have in hand, which is dependent on the setting of mosconfig. If
2251 * mosconfig is true then we're validating the vdev labels based on
2252 * that config. Otherwise, we're validating against the cached config
2253 * (zpool.cache) that was read when we loaded the zfs module, and then
2254 * later we will recursively call spa_load() and validate against
2257 * If we're assembling a new pool that's been split off from an
2258 * existing pool, the labels haven't yet been updated so we skip
2259 * validation for now.
2261 if (type != SPA_IMPORT_ASSEMBLE) {
2262 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2263 error = vdev_validate(rvd, mosconfig);
2264 spa_config_exit(spa, SCL_ALL, FTAG);
2269 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2270 return (SET_ERROR(ENXIO));
2274 * Find the best uberblock.
2276 vdev_uberblock_load(rvd, ub, &label);
2279 * If we weren't able to find a single valid uberblock, return failure.
2281 if (ub->ub_txg == 0) {
2283 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2287 * If the pool has an unsupported version we can't open it.
2289 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2291 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2294 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2298 * If we weren't able to find what's necessary for reading the
2299 * MOS in the label, return failure.
2301 if (label == NULL || nvlist_lookup_nvlist(label,
2302 ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) {
2304 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2309 * Update our in-core representation with the definitive values
2312 nvlist_free(spa->spa_label_features);
2313 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2319 * Look through entries in the label nvlist's features_for_read. If
2320 * there is a feature listed there which we don't understand then we
2321 * cannot open a pool.
2323 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2324 nvlist_t *unsup_feat;
2326 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2329 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2331 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2332 if (!zfeature_is_supported(nvpair_name(nvp))) {
2333 VERIFY(nvlist_add_string(unsup_feat,
2334 nvpair_name(nvp), "") == 0);
2338 if (!nvlist_empty(unsup_feat)) {
2339 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2340 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2341 nvlist_free(unsup_feat);
2342 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2346 nvlist_free(unsup_feat);
2350 * If the vdev guid sum doesn't match the uberblock, we have an
2351 * incomplete configuration. We first check to see if the pool
2352 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2353 * If it is, defer the vdev_guid_sum check till later so we
2354 * can handle missing vdevs.
2356 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
2357 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
2358 rvd->vdev_guid_sum != ub->ub_guid_sum)
2359 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2361 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2362 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2363 spa_try_repair(spa, config);
2364 spa_config_exit(spa, SCL_ALL, FTAG);
2365 nvlist_free(spa->spa_config_splitting);
2366 spa->spa_config_splitting = NULL;
2370 * Initialize internal SPA structures.
2372 spa->spa_state = POOL_STATE_ACTIVE;
2373 spa->spa_ubsync = spa->spa_uberblock;
2374 spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2375 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2376 spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2377 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2378 spa->spa_claim_max_txg = spa->spa_first_txg;
2379 spa->spa_prev_software_version = ub->ub_software_version;
2381 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2383 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2384 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2386 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2387 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2389 if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2390 boolean_t missing_feat_read = B_FALSE;
2391 nvlist_t *unsup_feat, *enabled_feat;
2393 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2394 &spa->spa_feat_for_read_obj) != 0) {
2395 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2398 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2399 &spa->spa_feat_for_write_obj) != 0) {
2400 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2403 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2404 &spa->spa_feat_desc_obj) != 0) {
2405 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2408 enabled_feat = fnvlist_alloc();
2409 unsup_feat = fnvlist_alloc();
2411 if (!spa_features_check(spa, B_FALSE,
2412 unsup_feat, enabled_feat))
2413 missing_feat_read = B_TRUE;
2415 if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) {
2416 if (!spa_features_check(spa, B_TRUE,
2417 unsup_feat, enabled_feat)) {
2418 missing_feat_write = B_TRUE;
2422 fnvlist_add_nvlist(spa->spa_load_info,
2423 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
2425 if (!nvlist_empty(unsup_feat)) {
2426 fnvlist_add_nvlist(spa->spa_load_info,
2427 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
2430 fnvlist_free(enabled_feat);
2431 fnvlist_free(unsup_feat);
2433 if (!missing_feat_read) {
2434 fnvlist_add_boolean(spa->spa_load_info,
2435 ZPOOL_CONFIG_CAN_RDONLY);
2439 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2440 * twofold: to determine whether the pool is available for
2441 * import in read-write mode and (if it is not) whether the
2442 * pool is available for import in read-only mode. If the pool
2443 * is available for import in read-write mode, it is displayed
2444 * as available in userland; if it is not available for import
2445 * in read-only mode, it is displayed as unavailable in
2446 * userland. If the pool is available for import in read-only
2447 * mode but not read-write mode, it is displayed as unavailable
2448 * in userland with a special note that the pool is actually
2449 * available for open in read-only mode.
2451 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2452 * missing a feature for write, we must first determine whether
2453 * the pool can be opened read-only before returning to
2454 * userland in order to know whether to display the
2455 * abovementioned note.
2457 if (missing_feat_read || (missing_feat_write &&
2458 spa_writeable(spa))) {
2459 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2464 * Load refcounts for ZFS features from disk into an in-memory
2465 * cache during SPA initialization.
2467 for (spa_feature_t i = 0; i < SPA_FEATURES; i++) {
2470 error = feature_get_refcount_from_disk(spa,
2471 &spa_feature_table[i], &refcount);
2473 spa->spa_feat_refcount_cache[i] = refcount;
2474 } else if (error == ENOTSUP) {
2475 spa->spa_feat_refcount_cache[i] =
2476 SPA_FEATURE_DISABLED;
2478 return (spa_vdev_err(rvd,
2479 VDEV_AUX_CORRUPT_DATA, EIO));
2484 if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) {
2485 if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG,
2486 &spa->spa_feat_enabled_txg_obj) != 0)
2487 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2490 spa->spa_is_initializing = B_TRUE;
2491 error = dsl_pool_open(spa->spa_dsl_pool);
2492 spa->spa_is_initializing = B_FALSE;
2494 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2498 nvlist_t *policy = NULL, *nvconfig;
2500 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2501 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2503 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2504 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2506 unsigned long myhostid = 0;
2508 VERIFY(nvlist_lookup_string(nvconfig,
2509 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2512 myhostid = zone_get_hostid(NULL);
2515 * We're emulating the system's hostid in userland, so
2516 * we can't use zone_get_hostid().
2518 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2519 #endif /* _KERNEL */
2520 if (check_hostid && hostid != 0 && myhostid != 0 &&
2521 hostid != myhostid) {
2522 nvlist_free(nvconfig);
2523 cmn_err(CE_WARN, "pool '%s' could not be "
2524 "loaded as it was last accessed by "
2525 "another system (host: %s hostid: 0x%lx). "
2526 "See: http://illumos.org/msg/ZFS-8000-EY",
2527 spa_name(spa), hostname,
2528 (unsigned long)hostid);
2529 return (SET_ERROR(EBADF));
2532 if (nvlist_lookup_nvlist(spa->spa_config,
2533 ZPOOL_REWIND_POLICY, &policy) == 0)
2534 VERIFY(nvlist_add_nvlist(nvconfig,
2535 ZPOOL_REWIND_POLICY, policy) == 0);
2537 spa_config_set(spa, nvconfig);
2539 spa_deactivate(spa);
2540 spa_activate(spa, orig_mode);
2542 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2545 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2546 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2547 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2549 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2552 * Load the bit that tells us to use the new accounting function
2553 * (raid-z deflation). If we have an older pool, this will not
2556 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2557 if (error != 0 && error != ENOENT)
2558 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2560 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2561 &spa->spa_creation_version);
2562 if (error != 0 && error != ENOENT)
2563 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2566 * Load the persistent error log. If we have an older pool, this will
2569 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2570 if (error != 0 && error != ENOENT)
2571 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2573 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2574 &spa->spa_errlog_scrub);
2575 if (error != 0 && error != ENOENT)
2576 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2579 * Load the history object. If we have an older pool, this
2580 * will not be present.
2582 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2583 if (error != 0 && error != ENOENT)
2584 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2587 * If we're assembling the pool from the split-off vdevs of
2588 * an existing pool, we don't want to attach the spares & cache
2593 * Load any hot spares for this pool.
2595 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2596 if (error != 0 && error != ENOENT)
2597 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2598 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2599 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2600 if (load_nvlist(spa, spa->spa_spares.sav_object,
2601 &spa->spa_spares.sav_config) != 0)
2602 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2604 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2605 spa_load_spares(spa);
2606 spa_config_exit(spa, SCL_ALL, FTAG);
2607 } else if (error == 0) {
2608 spa->spa_spares.sav_sync = B_TRUE;
2612 * Load any level 2 ARC devices for this pool.
2614 error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2615 &spa->spa_l2cache.sav_object);
2616 if (error != 0 && error != ENOENT)
2617 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2618 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2619 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2620 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2621 &spa->spa_l2cache.sav_config) != 0)
2622 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2624 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2625 spa_load_l2cache(spa);
2626 spa_config_exit(spa, SCL_ALL, FTAG);
2627 } else if (error == 0) {
2628 spa->spa_l2cache.sav_sync = B_TRUE;
2631 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2633 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2634 if (error && error != ENOENT)
2635 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2638 uint64_t autoreplace;
2640 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2641 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2642 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2643 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2644 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2645 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2646 &spa->spa_dedup_ditto);
2648 spa->spa_autoreplace = (autoreplace != 0);
2652 * If the 'autoreplace' property is set, then post a resource notifying
2653 * the ZFS DE that it should not issue any faults for unopenable
2654 * devices. We also iterate over the vdevs, and post a sysevent for any
2655 * unopenable vdevs so that the normal autoreplace handler can take
2658 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2659 spa_check_removed(spa->spa_root_vdev);
2661 * For the import case, this is done in spa_import(), because
2662 * at this point we're using the spare definitions from
2663 * the MOS config, not necessarily from the userland config.
2665 if (state != SPA_LOAD_IMPORT) {
2666 spa_aux_check_removed(&spa->spa_spares);
2667 spa_aux_check_removed(&spa->spa_l2cache);
2672 * Load the vdev state for all toplevel vdevs.
2677 * Propagate the leaf DTLs we just loaded all the way up the tree.
2679 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2680 vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2681 spa_config_exit(spa, SCL_ALL, FTAG);
2684 * Load the DDTs (dedup tables).
2686 error = ddt_load(spa);
2688 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2690 spa_update_dspace(spa);
2693 * Validate the config, using the MOS config to fill in any
2694 * information which might be missing. If we fail to validate
2695 * the config then declare the pool unfit for use. If we're
2696 * assembling a pool from a split, the log is not transferred
2699 if (type != SPA_IMPORT_ASSEMBLE) {
2702 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2703 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2705 if (!spa_config_valid(spa, nvconfig)) {
2706 nvlist_free(nvconfig);
2707 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2710 nvlist_free(nvconfig);
2713 * Now that we've validated the config, check the state of the
2714 * root vdev. If it can't be opened, it indicates one or
2715 * more toplevel vdevs are faulted.
2717 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2718 return (SET_ERROR(ENXIO));
2720 if (spa_check_logs(spa)) {
2721 *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2722 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2726 if (missing_feat_write) {
2727 ASSERT(state == SPA_LOAD_TRYIMPORT);
2730 * At this point, we know that we can open the pool in
2731 * read-only mode but not read-write mode. We now have enough
2732 * information and can return to userland.
2734 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
2738 * We've successfully opened the pool, verify that we're ready
2739 * to start pushing transactions.
2741 if (state != SPA_LOAD_TRYIMPORT) {
2742 if (error = spa_load_verify(spa))
2743 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2747 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2748 spa->spa_load_max_txg == UINT64_MAX)) {
2750 int need_update = B_FALSE;
2752 ASSERT(state != SPA_LOAD_TRYIMPORT);
2755 * Claim log blocks that haven't been committed yet.
2756 * This must all happen in a single txg.
2757 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2758 * invoked from zil_claim_log_block()'s i/o done callback.
2759 * Price of rollback is that we abandon the log.
2761 spa->spa_claiming = B_TRUE;
2763 tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2764 spa_first_txg(spa));
2765 (void) dmu_objset_find(spa_name(spa),
2766 zil_claim, tx, DS_FIND_CHILDREN);
2769 spa->spa_claiming = B_FALSE;
2771 spa_set_log_state(spa, SPA_LOG_GOOD);
2772 spa->spa_sync_on = B_TRUE;
2773 txg_sync_start(spa->spa_dsl_pool);
2776 * Wait for all claims to sync. We sync up to the highest
2777 * claimed log block birth time so that claimed log blocks
2778 * don't appear to be from the future. spa_claim_max_txg
2779 * will have been set for us by either zil_check_log_chain()
2780 * (invoked from spa_check_logs()) or zil_claim() above.
2782 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2785 * If the config cache is stale, or we have uninitialized
2786 * metaslabs (see spa_vdev_add()), then update the config.
2788 * If this is a verbatim import, trust the current
2789 * in-core spa_config and update the disk labels.
2791 if (config_cache_txg != spa->spa_config_txg ||
2792 state == SPA_LOAD_IMPORT ||
2793 state == SPA_LOAD_RECOVER ||
2794 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2795 need_update = B_TRUE;
2797 for (int c = 0; c < rvd->vdev_children; c++)
2798 if (rvd->vdev_child[c]->vdev_ms_array == 0)
2799 need_update = B_TRUE;
2802 * Update the config cache asychronously in case we're the
2803 * root pool, in which case the config cache isn't writable yet.
2806 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2809 * Check all DTLs to see if anything needs resilvering.
2811 if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2812 vdev_resilver_needed(rvd, NULL, NULL))
2813 spa_async_request(spa, SPA_ASYNC_RESILVER);
2816 * Log the fact that we booted up (so that we can detect if
2817 * we rebooted in the middle of an operation).
2819 spa_history_log_version(spa, "open");
2822 * Delete any inconsistent datasets.
2824 (void) dmu_objset_find(spa_name(spa),
2825 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2828 * Clean up any stale temporary dataset userrefs.
2830 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2837 spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2839 int mode = spa->spa_mode;
2842 spa_deactivate(spa);
2844 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg - 1;
2846 spa_activate(spa, mode);
2847 spa_async_suspend(spa);
2849 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2853 * If spa_load() fails this function will try loading prior txg's. If
2854 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2855 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2856 * function will not rewind the pool and will return the same error as
2860 spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2861 uint64_t max_request, int rewind_flags)
2863 nvlist_t *loadinfo = NULL;
2864 nvlist_t *config = NULL;
2865 int load_error, rewind_error;
2866 uint64_t safe_rewind_txg;
2869 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2870 spa->spa_load_max_txg = spa->spa_load_txg;
2871 spa_set_log_state(spa, SPA_LOG_CLEAR);
2873 spa->spa_load_max_txg = max_request;
2874 if (max_request != UINT64_MAX)
2875 spa->spa_extreme_rewind = B_TRUE;
2878 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2880 if (load_error == 0)
2883 if (spa->spa_root_vdev != NULL)
2884 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2886 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2887 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2889 if (rewind_flags & ZPOOL_NEVER_REWIND) {
2890 nvlist_free(config);
2891 return (load_error);
2894 if (state == SPA_LOAD_RECOVER) {
2895 /* Price of rolling back is discarding txgs, including log */
2896 spa_set_log_state(spa, SPA_LOG_CLEAR);
2899 * If we aren't rolling back save the load info from our first
2900 * import attempt so that we can restore it after attempting
2903 loadinfo = spa->spa_load_info;
2904 spa->spa_load_info = fnvlist_alloc();
2907 spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2908 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2909 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2910 TXG_INITIAL : safe_rewind_txg;
2913 * Continue as long as we're finding errors, we're still within
2914 * the acceptable rewind range, and we're still finding uberblocks
2916 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2917 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2918 if (spa->spa_load_max_txg < safe_rewind_txg)
2919 spa->spa_extreme_rewind = B_TRUE;
2920 rewind_error = spa_load_retry(spa, state, mosconfig);
2923 spa->spa_extreme_rewind = B_FALSE;
2924 spa->spa_load_max_txg = UINT64_MAX;
2926 if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2927 spa_config_set(spa, config);
2929 if (state == SPA_LOAD_RECOVER) {
2930 ASSERT3P(loadinfo, ==, NULL);
2931 return (rewind_error);
2933 /* Store the rewind info as part of the initial load info */
2934 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
2935 spa->spa_load_info);
2937 /* Restore the initial load info */
2938 fnvlist_free(spa->spa_load_info);
2939 spa->spa_load_info = loadinfo;
2941 return (load_error);
2948 * The import case is identical to an open except that the configuration is sent
2949 * down from userland, instead of grabbed from the configuration cache. For the
2950 * case of an open, the pool configuration will exist in the
2951 * POOL_STATE_UNINITIALIZED state.
2953 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2954 * the same time open the pool, without having to keep around the spa_t in some
2958 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2962 spa_load_state_t state = SPA_LOAD_OPEN;
2964 int locked = B_FALSE;
2965 int firstopen = B_FALSE;
2970 * As disgusting as this is, we need to support recursive calls to this
2971 * function because dsl_dir_open() is called during spa_load(), and ends
2972 * up calling spa_open() again. The real fix is to figure out how to
2973 * avoid dsl_dir_open() calling this in the first place.
2975 if (mutex_owner(&spa_namespace_lock) != curthread) {
2976 mutex_enter(&spa_namespace_lock);
2980 if ((spa = spa_lookup(pool)) == NULL) {
2982 mutex_exit(&spa_namespace_lock);
2983 return (SET_ERROR(ENOENT));
2986 if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2987 zpool_rewind_policy_t policy;
2991 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2993 if (policy.zrp_request & ZPOOL_DO_REWIND)
2994 state = SPA_LOAD_RECOVER;
2996 spa_activate(spa, spa_mode_global);
2998 if (state != SPA_LOAD_RECOVER)
2999 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3001 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
3002 policy.zrp_request);
3004 if (error == EBADF) {
3006 * If vdev_validate() returns failure (indicated by
3007 * EBADF), it indicates that one of the vdevs indicates
3008 * that the pool has been exported or destroyed. If
3009 * this is the case, the config cache is out of sync and
3010 * we should remove the pool from the namespace.
3013 spa_deactivate(spa);
3014 spa_config_sync(spa, B_TRUE, B_TRUE);
3017 mutex_exit(&spa_namespace_lock);
3018 return (SET_ERROR(ENOENT));
3023 * We can't open the pool, but we still have useful
3024 * information: the state of each vdev after the
3025 * attempted vdev_open(). Return this to the user.
3027 if (config != NULL && spa->spa_config) {
3028 VERIFY(nvlist_dup(spa->spa_config, config,
3030 VERIFY(nvlist_add_nvlist(*config,
3031 ZPOOL_CONFIG_LOAD_INFO,
3032 spa->spa_load_info) == 0);
3035 spa_deactivate(spa);
3036 spa->spa_last_open_failed = error;
3038 mutex_exit(&spa_namespace_lock);
3044 spa_open_ref(spa, tag);
3047 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
3050 * If we've recovered the pool, pass back any information we
3051 * gathered while doing the load.
3053 if (state == SPA_LOAD_RECOVER) {
3054 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
3055 spa->spa_load_info) == 0);
3059 spa->spa_last_open_failed = 0;
3060 spa->spa_last_ubsync_txg = 0;
3061 spa->spa_load_txg = 0;
3062 mutex_exit(&spa_namespace_lock);
3066 zvol_create_minors(spa->spa_name);
3077 spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
3080 return (spa_open_common(name, spapp, tag, policy, config));
3084 spa_open(const char *name, spa_t **spapp, void *tag)
3086 return (spa_open_common(name, spapp, tag, NULL, NULL));
3090 * Lookup the given spa_t, incrementing the inject count in the process,
3091 * preventing it from being exported or destroyed.
3094 spa_inject_addref(char *name)
3098 mutex_enter(&spa_namespace_lock);
3099 if ((spa = spa_lookup(name)) == NULL) {
3100 mutex_exit(&spa_namespace_lock);
3103 spa->spa_inject_ref++;
3104 mutex_exit(&spa_namespace_lock);
3110 spa_inject_delref(spa_t *spa)
3112 mutex_enter(&spa_namespace_lock);
3113 spa->spa_inject_ref--;
3114 mutex_exit(&spa_namespace_lock);
3118 * Add spares device information to the nvlist.
3121 spa_add_spares(spa_t *spa, nvlist_t *config)
3131 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3133 if (spa->spa_spares.sav_count == 0)
3136 VERIFY(nvlist_lookup_nvlist(config,
3137 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3138 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
3139 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3141 VERIFY(nvlist_add_nvlist_array(nvroot,
3142 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3143 VERIFY(nvlist_lookup_nvlist_array(nvroot,
3144 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3147 * Go through and find any spares which have since been
3148 * repurposed as an active spare. If this is the case, update
3149 * their status appropriately.
3151 for (i = 0; i < nspares; i++) {
3152 VERIFY(nvlist_lookup_uint64(spares[i],
3153 ZPOOL_CONFIG_GUID, &guid) == 0);
3154 if (spa_spare_exists(guid, &pool, NULL) &&
3156 VERIFY(nvlist_lookup_uint64_array(
3157 spares[i], ZPOOL_CONFIG_VDEV_STATS,
3158 (uint64_t **)&vs, &vsc) == 0);
3159 vs->vs_state = VDEV_STATE_CANT_OPEN;
3160 vs->vs_aux = VDEV_AUX_SPARED;
3167 * Add l2cache device information to the nvlist, including vdev stats.
3170 spa_add_l2cache(spa_t *spa, nvlist_t *config)
3173 uint_t i, j, nl2cache;
3180 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3182 if (spa->spa_l2cache.sav_count == 0)
3185 VERIFY(nvlist_lookup_nvlist(config,
3186 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3187 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
3188 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3189 if (nl2cache != 0) {
3190 VERIFY(nvlist_add_nvlist_array(nvroot,
3191 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3192 VERIFY(nvlist_lookup_nvlist_array(nvroot,
3193 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3196 * Update level 2 cache device stats.
3199 for (i = 0; i < nl2cache; i++) {
3200 VERIFY(nvlist_lookup_uint64(l2cache[i],
3201 ZPOOL_CONFIG_GUID, &guid) == 0);
3204 for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
3206 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
3207 vd = spa->spa_l2cache.sav_vdevs[j];
3213 VERIFY(nvlist_lookup_uint64_array(l2cache[i],
3214 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
3216 vdev_get_stats(vd, vs);
3222 spa_add_feature_stats(spa_t *spa, nvlist_t *config)
3228 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3229 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3231 /* We may be unable to read features if pool is suspended. */
3232 if (spa_suspended(spa))
3235 if (spa->spa_feat_for_read_obj != 0) {
3236 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3237 spa->spa_feat_for_read_obj);
3238 zap_cursor_retrieve(&zc, &za) == 0;
3239 zap_cursor_advance(&zc)) {
3240 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3241 za.za_num_integers == 1);
3242 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3243 za.za_first_integer));
3245 zap_cursor_fini(&zc);
3248 if (spa->spa_feat_for_write_obj != 0) {
3249 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3250 spa->spa_feat_for_write_obj);
3251 zap_cursor_retrieve(&zc, &za) == 0;
3252 zap_cursor_advance(&zc)) {
3253 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3254 za.za_num_integers == 1);
3255 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3256 za.za_first_integer));
3258 zap_cursor_fini(&zc);
3262 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
3264 nvlist_free(features);
3268 spa_get_stats(const char *name, nvlist_t **config,
3269 char *altroot, size_t buflen)
3275 error = spa_open_common(name, &spa, FTAG, NULL, config);
3279 * This still leaves a window of inconsistency where the spares
3280 * or l2cache devices could change and the config would be
3281 * self-inconsistent.
3283 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3285 if (*config != NULL) {
3286 uint64_t loadtimes[2];
3288 loadtimes[0] = spa->spa_loaded_ts.tv_sec;
3289 loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
3290 VERIFY(nvlist_add_uint64_array(*config,
3291 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
3293 VERIFY(nvlist_add_uint64(*config,
3294 ZPOOL_CONFIG_ERRCOUNT,
3295 spa_get_errlog_size(spa)) == 0);
3297 if (spa_suspended(spa))
3298 VERIFY(nvlist_add_uint64(*config,
3299 ZPOOL_CONFIG_SUSPENDED,
3300 spa->spa_failmode) == 0);
3302 spa_add_spares(spa, *config);
3303 spa_add_l2cache(spa, *config);
3304 spa_add_feature_stats(spa, *config);
3309 * We want to get the alternate root even for faulted pools, so we cheat
3310 * and call spa_lookup() directly.
3314 mutex_enter(&spa_namespace_lock);
3315 spa = spa_lookup(name);
3317 spa_altroot(spa, altroot, buflen);
3321 mutex_exit(&spa_namespace_lock);
3323 spa_altroot(spa, altroot, buflen);
3328 spa_config_exit(spa, SCL_CONFIG, FTAG);
3329 spa_close(spa, FTAG);
3336 * Validate that the auxiliary device array is well formed. We must have an
3337 * array of nvlists, each which describes a valid leaf vdev. If this is an
3338 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3339 * specified, as long as they are well-formed.
3342 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
3343 spa_aux_vdev_t *sav, const char *config, uint64_t version,
3344 vdev_labeltype_t label)
3351 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3354 * It's acceptable to have no devs specified.
3356 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
3360 return (SET_ERROR(EINVAL));
3363 * Make sure the pool is formatted with a version that supports this
3366 if (spa_version(spa) < version)
3367 return (SET_ERROR(ENOTSUP));
3370 * Set the pending device list so we correctly handle device in-use
3373 sav->sav_pending = dev;
3374 sav->sav_npending = ndev;
3376 for (i = 0; i < ndev; i++) {
3377 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
3381 if (!vd->vdev_ops->vdev_op_leaf) {
3383 error = SET_ERROR(EINVAL);
3388 * The L2ARC currently only supports disk devices in
3389 * kernel context. For user-level testing, we allow it.
3392 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
3393 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
3394 error = SET_ERROR(ENOTBLK);
3401 if ((error = vdev_open(vd)) == 0 &&
3402 (error = vdev_label_init(vd, crtxg, label)) == 0) {
3403 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
3404 vd->vdev_guid) == 0);
3410 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
3417 sav->sav_pending = NULL;
3418 sav->sav_npending = 0;
3423 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
3427 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3429 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3430 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
3431 VDEV_LABEL_SPARE)) != 0) {
3435 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3436 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
3437 VDEV_LABEL_L2CACHE));
3441 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
3446 if (sav->sav_config != NULL) {
3452 * Generate new dev list by concatentating with the
3455 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
3456 &olddevs, &oldndevs) == 0);
3458 newdevs = kmem_alloc(sizeof (void *) *
3459 (ndevs + oldndevs), KM_SLEEP);
3460 for (i = 0; i < oldndevs; i++)
3461 VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
3463 for (i = 0; i < ndevs; i++)
3464 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
3467 VERIFY(nvlist_remove(sav->sav_config, config,
3468 DATA_TYPE_NVLIST_ARRAY) == 0);
3470 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3471 config, newdevs, ndevs + oldndevs) == 0);
3472 for (i = 0; i < oldndevs + ndevs; i++)
3473 nvlist_free(newdevs[i]);
3474 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
3477 * Generate a new dev list.
3479 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
3481 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
3487 * Stop and drop level 2 ARC devices
3490 spa_l2cache_drop(spa_t *spa)
3494 spa_aux_vdev_t *sav = &spa->spa_l2cache;
3496 for (i = 0; i < sav->sav_count; i++) {
3499 vd = sav->sav_vdevs[i];
3502 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
3503 pool != 0ULL && l2arc_vdev_present(vd))
3504 l2arc_remove_vdev(vd);
3512 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
3516 char *altroot = NULL;
3521 uint64_t txg = TXG_INITIAL;
3522 nvlist_t **spares, **l2cache;
3523 uint_t nspares, nl2cache;
3524 uint64_t version, obj;
3525 boolean_t has_features;
3528 * If this pool already exists, return failure.
3530 mutex_enter(&spa_namespace_lock);
3531 if (spa_lookup(pool) != NULL) {
3532 mutex_exit(&spa_namespace_lock);
3533 return (SET_ERROR(EEXIST));
3537 * Allocate a new spa_t structure.
3539 (void) nvlist_lookup_string(props,
3540 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3541 spa = spa_add(pool, NULL, altroot);
3542 spa_activate(spa, spa_mode_global);
3544 if (props && (error = spa_prop_validate(spa, props))) {
3545 spa_deactivate(spa);
3547 mutex_exit(&spa_namespace_lock);
3551 has_features = B_FALSE;
3552 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
3553 elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
3554 if (zpool_prop_feature(nvpair_name(elem)))
3555 has_features = B_TRUE;
3558 if (has_features || nvlist_lookup_uint64(props,
3559 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
3560 version = SPA_VERSION;
3562 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
3564 spa->spa_first_txg = txg;
3565 spa->spa_uberblock.ub_txg = txg - 1;
3566 spa->spa_uberblock.ub_version = version;
3567 spa->spa_ubsync = spa->spa_uberblock;
3570 * Create "The Godfather" zio to hold all async IOs
3572 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
3573 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
3576 * Create the root vdev.
3578 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3580 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3582 ASSERT(error != 0 || rvd != NULL);
3583 ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3585 if (error == 0 && !zfs_allocatable_devs(nvroot))
3586 error = SET_ERROR(EINVAL);
3589 (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3590 (error = spa_validate_aux(spa, nvroot, txg,
3591 VDEV_ALLOC_ADD)) == 0) {
3592 for (int c = 0; c < rvd->vdev_children; c++) {
3593 vdev_ashift_optimize(rvd->vdev_child[c]);
3594 vdev_metaslab_set_size(rvd->vdev_child[c]);
3595 vdev_expand(rvd->vdev_child[c], txg);
3599 spa_config_exit(spa, SCL_ALL, FTAG);
3603 spa_deactivate(spa);
3605 mutex_exit(&spa_namespace_lock);
3610 * Get the list of spares, if specified.
3612 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3613 &spares, &nspares) == 0) {
3614 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3616 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3617 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3618 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3619 spa_load_spares(spa);
3620 spa_config_exit(spa, SCL_ALL, FTAG);
3621 spa->spa_spares.sav_sync = B_TRUE;
3625 * Get the list of level 2 cache devices, if specified.
3627 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3628 &l2cache, &nl2cache) == 0) {
3629 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3630 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3631 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3632 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3633 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3634 spa_load_l2cache(spa);
3635 spa_config_exit(spa, SCL_ALL, FTAG);
3636 spa->spa_l2cache.sav_sync = B_TRUE;
3639 spa->spa_is_initializing = B_TRUE;
3640 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3641 spa->spa_meta_objset = dp->dp_meta_objset;
3642 spa->spa_is_initializing = B_FALSE;
3645 * Create DDTs (dedup tables).
3649 spa_update_dspace(spa);
3651 tx = dmu_tx_create_assigned(dp, txg);
3654 * Create the pool config object.
3656 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3657 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3658 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3660 if (zap_add(spa->spa_meta_objset,
3661 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3662 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3663 cmn_err(CE_PANIC, "failed to add pool config");
3666 if (spa_version(spa) >= SPA_VERSION_FEATURES)
3667 spa_feature_create_zap_objects(spa, tx);
3669 if (zap_add(spa->spa_meta_objset,
3670 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3671 sizeof (uint64_t), 1, &version, tx) != 0) {
3672 cmn_err(CE_PANIC, "failed to add pool version");
3675 /* Newly created pools with the right version are always deflated. */
3676 if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3677 spa->spa_deflate = TRUE;
3678 if (zap_add(spa->spa_meta_objset,
3679 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3680 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3681 cmn_err(CE_PANIC, "failed to add deflate");
3686 * Create the deferred-free bpobj. Turn off compression
3687 * because sync-to-convergence takes longer if the blocksize
3690 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3691 dmu_object_set_compress(spa->spa_meta_objset, obj,
3692 ZIO_COMPRESS_OFF, tx);
3693 if (zap_add(spa->spa_meta_objset,
3694 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3695 sizeof (uint64_t), 1, &obj, tx) != 0) {
3696 cmn_err(CE_PANIC, "failed to add bpobj");
3698 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3699 spa->spa_meta_objset, obj));
3702 * Create the pool's history object.
3704 if (version >= SPA_VERSION_ZPOOL_HISTORY)
3705 spa_history_create_obj(spa, tx);
3708 * Set pool properties.
3710 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3711 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3712 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3713 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3715 if (props != NULL) {
3716 spa_configfile_set(spa, props, B_FALSE);
3717 spa_sync_props(props, tx);
3722 spa->spa_sync_on = B_TRUE;
3723 txg_sync_start(spa->spa_dsl_pool);
3726 * We explicitly wait for the first transaction to complete so that our
3727 * bean counters are appropriately updated.
3729 txg_wait_synced(spa->spa_dsl_pool, txg);
3731 spa_config_sync(spa, B_FALSE, B_TRUE);
3733 spa_history_log_version(spa, "create");
3735 spa->spa_minref = refcount_count(&spa->spa_refcount);
3737 mutex_exit(&spa_namespace_lock);
3745 * Get the root pool information from the root disk, then import the root pool
3746 * during the system boot up time.
3748 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3751 spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3754 nvlist_t *nvtop, *nvroot;
3757 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3761 * Add this top-level vdev to the child array.
3763 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3765 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3767 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3770 * Put this pool's top-level vdevs into a root vdev.
3772 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3773 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3774 VDEV_TYPE_ROOT) == 0);
3775 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3776 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3777 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3781 * Replace the existing vdev_tree with the new root vdev in
3782 * this pool's configuration (remove the old, add the new).
3784 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3785 nvlist_free(nvroot);
3790 * Walk the vdev tree and see if we can find a device with "better"
3791 * configuration. A configuration is "better" if the label on that
3792 * device has a more recent txg.
3795 spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3797 for (int c = 0; c < vd->vdev_children; c++)
3798 spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3800 if (vd->vdev_ops->vdev_op_leaf) {
3804 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3808 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3812 * Do we have a better boot device?
3814 if (label_txg > *txg) {
3823 * Import a root pool.
3825 * For x86. devpath_list will consist of devid and/or physpath name of
3826 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3827 * The GRUB "findroot" command will return the vdev we should boot.
3829 * For Sparc, devpath_list consists the physpath name of the booting device
3830 * no matter the rootpool is a single device pool or a mirrored pool.
3832 * "/pci@1f,0/ide@d/disk@0,0:a"
3835 spa_import_rootpool(char *devpath, char *devid)
3838 vdev_t *rvd, *bvd, *avd = NULL;
3839 nvlist_t *config, *nvtop;
3845 * Read the label from the boot device and generate a configuration.
3847 config = spa_generate_rootconf(devpath, devid, &guid);
3848 #if defined(_OBP) && defined(_KERNEL)
3849 if (config == NULL) {
3850 if (strstr(devpath, "/iscsi/ssd") != NULL) {
3852 get_iscsi_bootpath_phy(devpath);
3853 config = spa_generate_rootconf(devpath, devid, &guid);
3857 if (config == NULL) {
3858 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
3860 return (SET_ERROR(EIO));
3863 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3865 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3867 mutex_enter(&spa_namespace_lock);
3868 if ((spa = spa_lookup(pname)) != NULL) {
3870 * Remove the existing root pool from the namespace so that we
3871 * can replace it with the correct config we just read in.
3876 spa = spa_add(pname, config, NULL);
3877 spa->spa_is_root = B_TRUE;
3878 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3881 * Build up a vdev tree based on the boot device's label config.
3883 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3885 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3886 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3887 VDEV_ALLOC_ROOTPOOL);
3888 spa_config_exit(spa, SCL_ALL, FTAG);
3890 mutex_exit(&spa_namespace_lock);
3891 nvlist_free(config);
3892 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3898 * Get the boot vdev.
3900 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3901 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3902 (u_longlong_t)guid);
3903 error = SET_ERROR(ENOENT);
3908 * Determine if there is a better boot device.
3911 spa_alt_rootvdev(rvd, &avd, &txg);
3913 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3914 "try booting from '%s'", avd->vdev_path);
3915 error = SET_ERROR(EINVAL);
3920 * If the boot device is part of a spare vdev then ensure that
3921 * we're booting off the active spare.
3923 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3924 !bvd->vdev_isspare) {
3925 cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3926 "try booting from '%s'",
3928 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3929 error = SET_ERROR(EINVAL);
3935 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3937 spa_config_exit(spa, SCL_ALL, FTAG);
3938 mutex_exit(&spa_namespace_lock);
3940 nvlist_free(config);
3946 extern int vdev_geom_read_pool_label(const char *name, nvlist_t ***configs,
3950 spa_generate_rootconf(const char *name)
3952 nvlist_t **configs, **tops;
3954 nvlist_t *best_cfg, *nvtop, *nvroot;
3963 if (vdev_geom_read_pool_label(name, &configs, &count) != 0)
3966 ASSERT3U(count, !=, 0);
3968 for (i = 0; i < count; i++) {
3971 VERIFY(nvlist_lookup_uint64(configs[i], ZPOOL_CONFIG_POOL_TXG,
3973 if (txg > best_txg) {
3975 best_cfg = configs[i];
3980 nvlist_lookup_uint64(best_cfg, ZPOOL_CONFIG_VDEV_CHILDREN, &nchildren);
3982 nvlist_lookup_uint64_array(best_cfg, ZPOOL_CONFIG_HOLE_ARRAY,
3985 tops = kmem_zalloc(nchildren * sizeof(void *), KM_SLEEP);
3986 for (i = 0; i < nchildren; i++) {
3989 if (configs[i] == NULL)
3991 VERIFY(nvlist_lookup_nvlist(configs[i], ZPOOL_CONFIG_VDEV_TREE,
3993 nvlist_dup(nvtop, &tops[i], KM_SLEEP);
3995 for (i = 0; holes != NULL && i < nholes; i++) {
3998 if (tops[holes[i]] != NULL)
4000 nvlist_alloc(&tops[holes[i]], NV_UNIQUE_NAME, KM_SLEEP);
4001 VERIFY(nvlist_add_string(tops[holes[i]], ZPOOL_CONFIG_TYPE,
4002 VDEV_TYPE_HOLE) == 0);
4003 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_ID,
4005 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_GUID,
4008 for (i = 0; i < nchildren; i++) {
4009 if (tops[i] != NULL)
4011 nvlist_alloc(&tops[i], NV_UNIQUE_NAME, KM_SLEEP);
4012 VERIFY(nvlist_add_string(tops[i], ZPOOL_CONFIG_TYPE,
4013 VDEV_TYPE_MISSING) == 0);
4014 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_ID,
4016 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_GUID,
4021 * Create pool config based on the best vdev config.
4023 nvlist_dup(best_cfg, &config, KM_SLEEP);
4026 * Put this pool's top-level vdevs into a root vdev.
4028 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4030 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4031 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
4032 VDEV_TYPE_ROOT) == 0);
4033 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
4034 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
4035 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
4036 tops, nchildren) == 0);
4039 * Replace the existing vdev_tree with the new root vdev in
4040 * this pool's configuration (remove the old, add the new).
4042 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
4045 * Drop vdev config elements that should not be present at pool level.
4047 nvlist_remove(config, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64);
4048 nvlist_remove(config, ZPOOL_CONFIG_TOP_GUID, DATA_TYPE_UINT64);
4050 for (i = 0; i < count; i++)
4051 nvlist_free(configs[i]);
4052 kmem_free(configs, count * sizeof(void *));
4053 for (i = 0; i < nchildren; i++)
4054 nvlist_free(tops[i]);
4055 kmem_free(tops, nchildren * sizeof(void *));
4056 nvlist_free(nvroot);
4061 spa_import_rootpool(const char *name)
4064 vdev_t *rvd, *bvd, *avd = NULL;
4065 nvlist_t *config, *nvtop;
4071 * Read the label from the boot device and generate a configuration.
4073 config = spa_generate_rootconf(name);
4075 mutex_enter(&spa_namespace_lock);
4076 if (config != NULL) {
4077 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
4078 &pname) == 0 && strcmp(name, pname) == 0);
4079 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg)
4082 if ((spa = spa_lookup(pname)) != NULL) {
4084 * Remove the existing root pool from the namespace so
4085 * that we can replace it with the correct config
4090 spa = spa_add(pname, config, NULL);
4093 * Set spa_ubsync.ub_version as it can be used in vdev_alloc()
4094 * via spa_version().
4096 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
4097 &spa->spa_ubsync.ub_version) != 0)
4098 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
4099 } else if ((spa = spa_lookup(name)) == NULL) {
4100 cmn_err(CE_NOTE, "Cannot find the pool label for '%s'",
4104 VERIFY(nvlist_dup(spa->spa_config, &config, KM_SLEEP) == 0);
4106 spa->spa_is_root = B_TRUE;
4107 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
4110 * Build up a vdev tree based on the boot device's label config.
4112 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4114 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4115 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
4116 VDEV_ALLOC_ROOTPOOL);
4117 spa_config_exit(spa, SCL_ALL, FTAG);
4119 mutex_exit(&spa_namespace_lock);
4120 nvlist_free(config);
4121 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
4126 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4128 spa_config_exit(spa, SCL_ALL, FTAG);
4129 mutex_exit(&spa_namespace_lock);
4131 nvlist_free(config);
4139 * Import a non-root pool into the system.
4142 spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
4145 char *altroot = NULL;
4146 spa_load_state_t state = SPA_LOAD_IMPORT;
4147 zpool_rewind_policy_t policy;
4148 uint64_t mode = spa_mode_global;
4149 uint64_t readonly = B_FALSE;
4152 nvlist_t **spares, **l2cache;
4153 uint_t nspares, nl2cache;
4156 * If a pool with this name exists, return failure.
4158 mutex_enter(&spa_namespace_lock);
4159 if (spa_lookup(pool) != NULL) {
4160 mutex_exit(&spa_namespace_lock);
4161 return (SET_ERROR(EEXIST));
4165 * Create and initialize the spa structure.
4167 (void) nvlist_lookup_string(props,
4168 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4169 (void) nvlist_lookup_uint64(props,
4170 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
4173 spa = spa_add(pool, config, altroot);
4174 spa->spa_import_flags = flags;
4177 * Verbatim import - Take a pool and insert it into the namespace
4178 * as if it had been loaded at boot.
4180 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
4182 spa_configfile_set(spa, props, B_FALSE);
4184 spa_config_sync(spa, B_FALSE, B_TRUE);
4186 mutex_exit(&spa_namespace_lock);
4190 spa_activate(spa, mode);
4193 * Don't start async tasks until we know everything is healthy.
4195 spa_async_suspend(spa);
4197 zpool_get_rewind_policy(config, &policy);
4198 if (policy.zrp_request & ZPOOL_DO_REWIND)
4199 state = SPA_LOAD_RECOVER;
4202 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
4203 * because the user-supplied config is actually the one to trust when
4206 if (state != SPA_LOAD_RECOVER)
4207 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
4209 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
4210 policy.zrp_request);
4213 * Propagate anything learned while loading the pool and pass it
4214 * back to caller (i.e. rewind info, missing devices, etc).
4216 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4217 spa->spa_load_info) == 0);
4219 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4221 * Toss any existing sparelist, as it doesn't have any validity
4222 * anymore, and conflicts with spa_has_spare().
4224 if (spa->spa_spares.sav_config) {
4225 nvlist_free(spa->spa_spares.sav_config);
4226 spa->spa_spares.sav_config = NULL;
4227 spa_load_spares(spa);
4229 if (spa->spa_l2cache.sav_config) {
4230 nvlist_free(spa->spa_l2cache.sav_config);
4231 spa->spa_l2cache.sav_config = NULL;
4232 spa_load_l2cache(spa);
4235 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4238 error = spa_validate_aux(spa, nvroot, -1ULL,
4241 error = spa_validate_aux(spa, nvroot, -1ULL,
4242 VDEV_ALLOC_L2CACHE);
4243 spa_config_exit(spa, SCL_ALL, FTAG);
4246 spa_configfile_set(spa, props, B_FALSE);
4248 if (error != 0 || (props && spa_writeable(spa) &&
4249 (error = spa_prop_set(spa, props)))) {
4251 spa_deactivate(spa);
4253 mutex_exit(&spa_namespace_lock);
4257 spa_async_resume(spa);
4260 * Override any spares and level 2 cache devices as specified by
4261 * the user, as these may have correct device names/devids, etc.
4263 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
4264 &spares, &nspares) == 0) {
4265 if (spa->spa_spares.sav_config)
4266 VERIFY(nvlist_remove(spa->spa_spares.sav_config,
4267 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
4269 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
4270 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4271 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
4272 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
4273 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4274 spa_load_spares(spa);
4275 spa_config_exit(spa, SCL_ALL, FTAG);
4276 spa->spa_spares.sav_sync = B_TRUE;
4278 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
4279 &l2cache, &nl2cache) == 0) {
4280 if (spa->spa_l2cache.sav_config)
4281 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
4282 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
4284 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
4285 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4286 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
4287 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
4288 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4289 spa_load_l2cache(spa);
4290 spa_config_exit(spa, SCL_ALL, FTAG);
4291 spa->spa_l2cache.sav_sync = B_TRUE;
4295 * Check for any removed devices.
4297 if (spa->spa_autoreplace) {
4298 spa_aux_check_removed(&spa->spa_spares);
4299 spa_aux_check_removed(&spa->spa_l2cache);
4302 if (spa_writeable(spa)) {
4304 * Update the config cache to include the newly-imported pool.
4306 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4310 * It's possible that the pool was expanded while it was exported.
4311 * We kick off an async task to handle this for us.
4313 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
4315 mutex_exit(&spa_namespace_lock);
4316 spa_history_log_version(spa, "import");
4320 zvol_create_minors(pool);
4327 spa_tryimport(nvlist_t *tryconfig)
4329 nvlist_t *config = NULL;
4335 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
4338 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
4342 * Create and initialize the spa structure.
4344 mutex_enter(&spa_namespace_lock);
4345 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
4346 spa_activate(spa, FREAD);
4349 * Pass off the heavy lifting to spa_load().
4350 * Pass TRUE for mosconfig because the user-supplied config
4351 * is actually the one to trust when doing an import.
4353 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
4356 * If 'tryconfig' was at least parsable, return the current config.
4358 if (spa->spa_root_vdev != NULL) {
4359 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4360 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
4362 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4364 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
4365 spa->spa_uberblock.ub_timestamp) == 0);
4366 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4367 spa->spa_load_info) == 0);
4370 * If the bootfs property exists on this pool then we
4371 * copy it out so that external consumers can tell which
4372 * pools are bootable.
4374 if ((!error || error == EEXIST) && spa->spa_bootfs) {
4375 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4378 * We have to play games with the name since the
4379 * pool was opened as TRYIMPORT_NAME.
4381 if (dsl_dsobj_to_dsname(spa_name(spa),
4382 spa->spa_bootfs, tmpname) == 0) {
4384 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4386 cp = strchr(tmpname, '/');
4388 (void) strlcpy(dsname, tmpname,
4391 (void) snprintf(dsname, MAXPATHLEN,
4392 "%s/%s", poolname, ++cp);
4394 VERIFY(nvlist_add_string(config,
4395 ZPOOL_CONFIG_BOOTFS, dsname) == 0);
4396 kmem_free(dsname, MAXPATHLEN);
4398 kmem_free(tmpname, MAXPATHLEN);
4402 * Add the list of hot spares and level 2 cache devices.
4404 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4405 spa_add_spares(spa, config);
4406 spa_add_l2cache(spa, config);
4407 spa_config_exit(spa, SCL_CONFIG, FTAG);
4411 spa_deactivate(spa);
4413 mutex_exit(&spa_namespace_lock);
4419 * Pool export/destroy
4421 * The act of destroying or exporting a pool is very simple. We make sure there
4422 * is no more pending I/O and any references to the pool are gone. Then, we
4423 * update the pool state and sync all the labels to disk, removing the
4424 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4425 * we don't sync the labels or remove the configuration cache.
4428 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
4429 boolean_t force, boolean_t hardforce)
4436 if (!(spa_mode_global & FWRITE))
4437 return (SET_ERROR(EROFS));
4439 mutex_enter(&spa_namespace_lock);
4440 if ((spa = spa_lookup(pool)) == NULL) {
4441 mutex_exit(&spa_namespace_lock);
4442 return (SET_ERROR(ENOENT));
4446 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4447 * reacquire the namespace lock, and see if we can export.
4449 spa_open_ref(spa, FTAG);
4450 mutex_exit(&spa_namespace_lock);
4451 spa_async_suspend(spa);
4452 mutex_enter(&spa_namespace_lock);
4453 spa_close(spa, FTAG);
4456 * The pool will be in core if it's openable,
4457 * in which case we can modify its state.
4459 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
4461 * Objsets may be open only because they're dirty, so we
4462 * have to force it to sync before checking spa_refcnt.
4464 txg_wait_synced(spa->spa_dsl_pool, 0);
4467 * A pool cannot be exported or destroyed if there are active
4468 * references. If we are resetting a pool, allow references by
4469 * fault injection handlers.
4471 if (!spa_refcount_zero(spa) ||
4472 (spa->spa_inject_ref != 0 &&
4473 new_state != POOL_STATE_UNINITIALIZED)) {
4474 spa_async_resume(spa);
4475 mutex_exit(&spa_namespace_lock);
4476 return (SET_ERROR(EBUSY));
4480 * A pool cannot be exported if it has an active shared spare.
4481 * This is to prevent other pools stealing the active spare
4482 * from an exported pool. At user's own will, such pool can
4483 * be forcedly exported.
4485 if (!force && new_state == POOL_STATE_EXPORTED &&
4486 spa_has_active_shared_spare(spa)) {
4487 spa_async_resume(spa);
4488 mutex_exit(&spa_namespace_lock);
4489 return (SET_ERROR(EXDEV));
4493 * We want this to be reflected on every label,
4494 * so mark them all dirty. spa_unload() will do the
4495 * final sync that pushes these changes out.
4497 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
4498 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4499 spa->spa_state = new_state;
4500 spa->spa_final_txg = spa_last_synced_txg(spa) +
4502 vdev_config_dirty(spa->spa_root_vdev);
4503 spa_config_exit(spa, SCL_ALL, FTAG);
4507 spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
4509 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4511 spa_deactivate(spa);
4514 if (oldconfig && spa->spa_config)
4515 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
4517 if (new_state != POOL_STATE_UNINITIALIZED) {
4519 spa_config_sync(spa, B_TRUE, B_TRUE);
4522 mutex_exit(&spa_namespace_lock);
4528 * Destroy a storage pool.
4531 spa_destroy(char *pool)
4533 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
4538 * Export a storage pool.
4541 spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
4542 boolean_t hardforce)
4544 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
4549 * Similar to spa_export(), this unloads the spa_t without actually removing it
4550 * from the namespace in any way.
4553 spa_reset(char *pool)
4555 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
4560 * ==========================================================================
4561 * Device manipulation
4562 * ==========================================================================
4566 * Add a device to a storage pool.
4569 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
4573 vdev_t *rvd = spa->spa_root_vdev;
4575 nvlist_t **spares, **l2cache;
4576 uint_t nspares, nl2cache;
4578 ASSERT(spa_writeable(spa));
4580 txg = spa_vdev_enter(spa);
4582 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
4583 VDEV_ALLOC_ADD)) != 0)
4584 return (spa_vdev_exit(spa, NULL, txg, error));
4586 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */
4588 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
4592 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
4596 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
4597 return (spa_vdev_exit(spa, vd, txg, EINVAL));
4599 if (vd->vdev_children != 0 &&
4600 (error = vdev_create(vd, txg, B_FALSE)) != 0)
4601 return (spa_vdev_exit(spa, vd, txg, error));
4604 * We must validate the spares and l2cache devices after checking the
4605 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
4607 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
4608 return (spa_vdev_exit(spa, vd, txg, error));
4611 * Transfer each new top-level vdev from vd to rvd.
4613 for (int c = 0; c < vd->vdev_children; c++) {
4616 * Set the vdev id to the first hole, if one exists.
4618 for (id = 0; id < rvd->vdev_children; id++) {
4619 if (rvd->vdev_child[id]->vdev_ishole) {
4620 vdev_free(rvd->vdev_child[id]);
4624 tvd = vd->vdev_child[c];
4625 vdev_remove_child(vd, tvd);
4627 vdev_add_child(rvd, tvd);
4628 vdev_config_dirty(tvd);
4632 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
4633 ZPOOL_CONFIG_SPARES);
4634 spa_load_spares(spa);
4635 spa->spa_spares.sav_sync = B_TRUE;
4638 if (nl2cache != 0) {
4639 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
4640 ZPOOL_CONFIG_L2CACHE);
4641 spa_load_l2cache(spa);
4642 spa->spa_l2cache.sav_sync = B_TRUE;
4646 * We have to be careful when adding new vdevs to an existing pool.
4647 * If other threads start allocating from these vdevs before we
4648 * sync the config cache, and we lose power, then upon reboot we may
4649 * fail to open the pool because there are DVAs that the config cache
4650 * can't translate. Therefore, we first add the vdevs without
4651 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4652 * and then let spa_config_update() initialize the new metaslabs.
4654 * spa_load() checks for added-but-not-initialized vdevs, so that
4655 * if we lose power at any point in this sequence, the remaining
4656 * steps will be completed the next time we load the pool.
4658 (void) spa_vdev_exit(spa, vd, txg, 0);
4660 mutex_enter(&spa_namespace_lock);
4661 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4662 mutex_exit(&spa_namespace_lock);
4668 * Attach a device to a mirror. The arguments are the path to any device
4669 * in the mirror, and the nvroot for the new device. If the path specifies
4670 * a device that is not mirrored, we automatically insert the mirror vdev.
4672 * If 'replacing' is specified, the new device is intended to replace the
4673 * existing device; in this case the two devices are made into their own
4674 * mirror using the 'replacing' vdev, which is functionally identical to
4675 * the mirror vdev (it actually reuses all the same ops) but has a few
4676 * extra rules: you can't attach to it after it's been created, and upon
4677 * completion of resilvering, the first disk (the one being replaced)
4678 * is automatically detached.
4681 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
4683 uint64_t txg, dtl_max_txg;
4684 vdev_t *rvd = spa->spa_root_vdev;
4685 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
4687 char *oldvdpath, *newvdpath;
4691 ASSERT(spa_writeable(spa));
4693 txg = spa_vdev_enter(spa);
4695 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
4698 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4700 if (!oldvd->vdev_ops->vdev_op_leaf)
4701 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4703 pvd = oldvd->vdev_parent;
4705 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
4706 VDEV_ALLOC_ATTACH)) != 0)
4707 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4709 if (newrootvd->vdev_children != 1)
4710 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4712 newvd = newrootvd->vdev_child[0];
4714 if (!newvd->vdev_ops->vdev_op_leaf)
4715 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4717 if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
4718 return (spa_vdev_exit(spa, newrootvd, txg, error));
4721 * Spares can't replace logs
4723 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
4724 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4728 * For attach, the only allowable parent is a mirror or the root
4731 if (pvd->vdev_ops != &vdev_mirror_ops &&
4732 pvd->vdev_ops != &vdev_root_ops)
4733 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4735 pvops = &vdev_mirror_ops;
4738 * Active hot spares can only be replaced by inactive hot
4741 if (pvd->vdev_ops == &vdev_spare_ops &&
4742 oldvd->vdev_isspare &&
4743 !spa_has_spare(spa, newvd->vdev_guid))
4744 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4747 * If the source is a hot spare, and the parent isn't already a
4748 * spare, then we want to create a new hot spare. Otherwise, we
4749 * want to create a replacing vdev. The user is not allowed to
4750 * attach to a spared vdev child unless the 'isspare' state is
4751 * the same (spare replaces spare, non-spare replaces
4754 if (pvd->vdev_ops == &vdev_replacing_ops &&
4755 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
4756 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4757 } else if (pvd->vdev_ops == &vdev_spare_ops &&
4758 newvd->vdev_isspare != oldvd->vdev_isspare) {
4759 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4762 if (newvd->vdev_isspare)
4763 pvops = &vdev_spare_ops;
4765 pvops = &vdev_replacing_ops;
4769 * Make sure the new device is big enough.
4771 if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
4772 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
4775 * The new device cannot have a higher alignment requirement
4776 * than the top-level vdev.
4778 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4779 return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4782 * If this is an in-place replacement, update oldvd's path and devid
4783 * to make it distinguishable from newvd, and unopenable from now on.
4785 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4786 spa_strfree(oldvd->vdev_path);
4787 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4789 (void) sprintf(oldvd->vdev_path, "%s/%s",
4790 newvd->vdev_path, "old");
4791 if (oldvd->vdev_devid != NULL) {
4792 spa_strfree(oldvd->vdev_devid);
4793 oldvd->vdev_devid = NULL;
4797 /* mark the device being resilvered */
4798 newvd->vdev_resilver_txg = txg;
4801 * If the parent is not a mirror, or if we're replacing, insert the new
4802 * mirror/replacing/spare vdev above oldvd.
4804 if (pvd->vdev_ops != pvops)
4805 pvd = vdev_add_parent(oldvd, pvops);
4807 ASSERT(pvd->vdev_top->vdev_parent == rvd);
4808 ASSERT(pvd->vdev_ops == pvops);
4809 ASSERT(oldvd->vdev_parent == pvd);
4812 * Extract the new device from its root and add it to pvd.
4814 vdev_remove_child(newrootvd, newvd);
4815 newvd->vdev_id = pvd->vdev_children;
4816 newvd->vdev_crtxg = oldvd->vdev_crtxg;
4817 vdev_add_child(pvd, newvd);
4819 tvd = newvd->vdev_top;
4820 ASSERT(pvd->vdev_top == tvd);
4821 ASSERT(tvd->vdev_parent == rvd);
4823 vdev_config_dirty(tvd);
4826 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4827 * for any dmu_sync-ed blocks. It will propagate upward when
4828 * spa_vdev_exit() calls vdev_dtl_reassess().
4830 dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4832 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4833 dtl_max_txg - TXG_INITIAL);
4835 if (newvd->vdev_isspare) {
4836 spa_spare_activate(newvd);
4837 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
4840 oldvdpath = spa_strdup(oldvd->vdev_path);
4841 newvdpath = spa_strdup(newvd->vdev_path);
4842 newvd_isspare = newvd->vdev_isspare;
4845 * Mark newvd's DTL dirty in this txg.
4847 vdev_dirty(tvd, VDD_DTL, newvd, txg);
4850 * Schedule the resilver to restart in the future. We do this to
4851 * ensure that dmu_sync-ed blocks have been stitched into the
4852 * respective datasets.
4854 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4859 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4861 spa_history_log_internal(spa, "vdev attach", NULL,
4862 "%s vdev=%s %s vdev=%s",
4863 replacing && newvd_isspare ? "spare in" :
4864 replacing ? "replace" : "attach", newvdpath,
4865 replacing ? "for" : "to", oldvdpath);
4867 spa_strfree(oldvdpath);
4868 spa_strfree(newvdpath);
4870 if (spa->spa_bootfs)
4871 spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
4877 * Detach a device from a mirror or replacing vdev.
4879 * If 'replace_done' is specified, only detach if the parent
4880 * is a replacing vdev.
4883 spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4887 vdev_t *rvd = spa->spa_root_vdev;
4888 vdev_t *vd, *pvd, *cvd, *tvd;
4889 boolean_t unspare = B_FALSE;
4890 uint64_t unspare_guid = 0;
4893 ASSERT(spa_writeable(spa));
4895 txg = spa_vdev_enter(spa);
4897 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4900 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4902 if (!vd->vdev_ops->vdev_op_leaf)
4903 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4905 pvd = vd->vdev_parent;
4908 * If the parent/child relationship is not as expected, don't do it.
4909 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4910 * vdev that's replacing B with C. The user's intent in replacing
4911 * is to go from M(A,B) to M(A,C). If the user decides to cancel
4912 * the replace by detaching C, the expected behavior is to end up
4913 * M(A,B). But suppose that right after deciding to detach C,
4914 * the replacement of B completes. We would have M(A,C), and then
4915 * ask to detach C, which would leave us with just A -- not what
4916 * the user wanted. To prevent this, we make sure that the
4917 * parent/child relationship hasn't changed -- in this example,
4918 * that C's parent is still the replacing vdev R.
4920 if (pvd->vdev_guid != pguid && pguid != 0)
4921 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4924 * Only 'replacing' or 'spare' vdevs can be replaced.
4926 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4927 pvd->vdev_ops != &vdev_spare_ops)
4928 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4930 ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4931 spa_version(spa) >= SPA_VERSION_SPARES);
4934 * Only mirror, replacing, and spare vdevs support detach.
4936 if (pvd->vdev_ops != &vdev_replacing_ops &&
4937 pvd->vdev_ops != &vdev_mirror_ops &&
4938 pvd->vdev_ops != &vdev_spare_ops)
4939 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4942 * If this device has the only valid copy of some data,
4943 * we cannot safely detach it.
4945 if (vdev_dtl_required(vd))
4946 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4948 ASSERT(pvd->vdev_children >= 2);
4951 * If we are detaching the second disk from a replacing vdev, then
4952 * check to see if we changed the original vdev's path to have "/old"
4953 * at the end in spa_vdev_attach(). If so, undo that change now.
4955 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4956 vd->vdev_path != NULL) {
4957 size_t len = strlen(vd->vdev_path);
4959 for (int c = 0; c < pvd->vdev_children; c++) {
4960 cvd = pvd->vdev_child[c];
4962 if (cvd == vd || cvd->vdev_path == NULL)
4965 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4966 strcmp(cvd->vdev_path + len, "/old") == 0) {
4967 spa_strfree(cvd->vdev_path);
4968 cvd->vdev_path = spa_strdup(vd->vdev_path);
4975 * If we are detaching the original disk from a spare, then it implies
4976 * that the spare should become a real disk, and be removed from the
4977 * active spare list for the pool.
4979 if (pvd->vdev_ops == &vdev_spare_ops &&
4981 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4985 * Erase the disk labels so the disk can be used for other things.
4986 * This must be done after all other error cases are handled,
4987 * but before we disembowel vd (so we can still do I/O to it).
4988 * But if we can't do it, don't treat the error as fatal --
4989 * it may be that the unwritability of the disk is the reason
4990 * it's being detached!
4992 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4995 * Remove vd from its parent and compact the parent's children.
4997 vdev_remove_child(pvd, vd);
4998 vdev_compact_children(pvd);
5001 * Remember one of the remaining children so we can get tvd below.
5003 cvd = pvd->vdev_child[pvd->vdev_children - 1];
5006 * If we need to remove the remaining child from the list of hot spares,
5007 * do it now, marking the vdev as no longer a spare in the process.
5008 * We must do this before vdev_remove_parent(), because that can
5009 * change the GUID if it creates a new toplevel GUID. For a similar
5010 * reason, we must remove the spare now, in the same txg as the detach;
5011 * otherwise someone could attach a new sibling, change the GUID, and
5012 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
5015 ASSERT(cvd->vdev_isspare);
5016 spa_spare_remove(cvd);
5017 unspare_guid = cvd->vdev_guid;
5018 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
5019 cvd->vdev_unspare = B_TRUE;
5023 * If the parent mirror/replacing vdev only has one child,
5024 * the parent is no longer needed. Remove it from the tree.
5026 if (pvd->vdev_children == 1) {
5027 if (pvd->vdev_ops == &vdev_spare_ops)
5028 cvd->vdev_unspare = B_FALSE;
5029 vdev_remove_parent(cvd);
5034 * We don't set tvd until now because the parent we just removed
5035 * may have been the previous top-level vdev.
5037 tvd = cvd->vdev_top;
5038 ASSERT(tvd->vdev_parent == rvd);
5041 * Reevaluate the parent vdev state.
5043 vdev_propagate_state(cvd);
5046 * If the 'autoexpand' property is set on the pool then automatically
5047 * try to expand the size of the pool. For example if the device we
5048 * just detached was smaller than the others, it may be possible to
5049 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
5050 * first so that we can obtain the updated sizes of the leaf vdevs.
5052 if (spa->spa_autoexpand) {
5054 vdev_expand(tvd, txg);
5057 vdev_config_dirty(tvd);
5060 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
5061 * vd->vdev_detached is set and free vd's DTL object in syncing context.
5062 * But first make sure we're not on any *other* txg's DTL list, to
5063 * prevent vd from being accessed after it's freed.
5065 vdpath = spa_strdup(vd->vdev_path);
5066 for (int t = 0; t < TXG_SIZE; t++)
5067 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
5068 vd->vdev_detached = B_TRUE;
5069 vdev_dirty(tvd, VDD_DTL, vd, txg);
5071 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
5073 /* hang on to the spa before we release the lock */
5074 spa_open_ref(spa, FTAG);
5076 error = spa_vdev_exit(spa, vd, txg, 0);
5078 spa_history_log_internal(spa, "detach", NULL,
5080 spa_strfree(vdpath);
5083 * If this was the removal of the original device in a hot spare vdev,
5084 * then we want to go through and remove the device from the hot spare
5085 * list of every other pool.
5088 spa_t *altspa = NULL;
5090 mutex_enter(&spa_namespace_lock);
5091 while ((altspa = spa_next(altspa)) != NULL) {
5092 if (altspa->spa_state != POOL_STATE_ACTIVE ||
5096 spa_open_ref(altspa, FTAG);
5097 mutex_exit(&spa_namespace_lock);
5098 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
5099 mutex_enter(&spa_namespace_lock);
5100 spa_close(altspa, FTAG);
5102 mutex_exit(&spa_namespace_lock);
5104 /* search the rest of the vdevs for spares to remove */
5105 spa_vdev_resilver_done(spa);
5108 /* all done with the spa; OK to release */
5109 mutex_enter(&spa_namespace_lock);
5110 spa_close(spa, FTAG);
5111 mutex_exit(&spa_namespace_lock);
5117 * Split a set of devices from their mirrors, and create a new pool from them.
5120 spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
5121 nvlist_t *props, boolean_t exp)
5124 uint64_t txg, *glist;
5126 uint_t c, children, lastlog;
5127 nvlist_t **child, *nvl, *tmp;
5129 char *altroot = NULL;
5130 vdev_t *rvd, **vml = NULL; /* vdev modify list */
5131 boolean_t activate_slog;
5133 ASSERT(spa_writeable(spa));
5135 txg = spa_vdev_enter(spa);
5137 /* clear the log and flush everything up to now */
5138 activate_slog = spa_passivate_log(spa);
5139 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5140 error = spa_offline_log(spa);
5141 txg = spa_vdev_config_enter(spa);
5144 spa_activate_log(spa);
5147 return (spa_vdev_exit(spa, NULL, txg, error));
5149 /* check new spa name before going any further */
5150 if (spa_lookup(newname) != NULL)
5151 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
5154 * scan through all the children to ensure they're all mirrors
5156 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
5157 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
5159 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5161 /* first, check to ensure we've got the right child count */
5162 rvd = spa->spa_root_vdev;
5164 for (c = 0; c < rvd->vdev_children; c++) {
5165 vdev_t *vd = rvd->vdev_child[c];
5167 /* don't count the holes & logs as children */
5168 if (vd->vdev_islog || vd->vdev_ishole) {
5176 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
5177 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5179 /* next, ensure no spare or cache devices are part of the split */
5180 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
5181 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
5182 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5184 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
5185 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
5187 /* then, loop over each vdev and validate it */
5188 for (c = 0; c < children; c++) {
5189 uint64_t is_hole = 0;
5191 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
5195 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
5196 spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
5199 error = SET_ERROR(EINVAL);
5204 /* which disk is going to be split? */
5205 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
5207 error = SET_ERROR(EINVAL);
5211 /* look it up in the spa */
5212 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
5213 if (vml[c] == NULL) {
5214 error = SET_ERROR(ENODEV);
5218 /* make sure there's nothing stopping the split */
5219 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
5220 vml[c]->vdev_islog ||
5221 vml[c]->vdev_ishole ||
5222 vml[c]->vdev_isspare ||
5223 vml[c]->vdev_isl2cache ||
5224 !vdev_writeable(vml[c]) ||
5225 vml[c]->vdev_children != 0 ||
5226 vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
5227 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
5228 error = SET_ERROR(EINVAL);
5232 if (vdev_dtl_required(vml[c])) {
5233 error = SET_ERROR(EBUSY);
5237 /* we need certain info from the top level */
5238 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
5239 vml[c]->vdev_top->vdev_ms_array) == 0);
5240 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
5241 vml[c]->vdev_top->vdev_ms_shift) == 0);
5242 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
5243 vml[c]->vdev_top->vdev_asize) == 0);
5244 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
5245 vml[c]->vdev_top->vdev_ashift) == 0);
5249 kmem_free(vml, children * sizeof (vdev_t *));
5250 kmem_free(glist, children * sizeof (uint64_t));
5251 return (spa_vdev_exit(spa, NULL, txg, error));
5254 /* stop writers from using the disks */
5255 for (c = 0; c < children; c++) {
5257 vml[c]->vdev_offline = B_TRUE;
5259 vdev_reopen(spa->spa_root_vdev);
5262 * Temporarily record the splitting vdevs in the spa config. This
5263 * will disappear once the config is regenerated.
5265 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5266 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
5267 glist, children) == 0);
5268 kmem_free(glist, children * sizeof (uint64_t));
5270 mutex_enter(&spa->spa_props_lock);
5271 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
5273 mutex_exit(&spa->spa_props_lock);
5274 spa->spa_config_splitting = nvl;
5275 vdev_config_dirty(spa->spa_root_vdev);
5277 /* configure and create the new pool */
5278 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
5279 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
5280 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
5281 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5282 spa_version(spa)) == 0);
5283 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
5284 spa->spa_config_txg) == 0);
5285 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
5286 spa_generate_guid(NULL)) == 0);
5287 (void) nvlist_lookup_string(props,
5288 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
5290 /* add the new pool to the namespace */
5291 newspa = spa_add(newname, config, altroot);
5292 newspa->spa_config_txg = spa->spa_config_txg;
5293 spa_set_log_state(newspa, SPA_LOG_CLEAR);
5295 /* release the spa config lock, retaining the namespace lock */
5296 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5298 if (zio_injection_enabled)
5299 zio_handle_panic_injection(spa, FTAG, 1);
5301 spa_activate(newspa, spa_mode_global);
5302 spa_async_suspend(newspa);
5305 /* mark that we are creating new spa by splitting */
5306 newspa->spa_splitting_newspa = B_TRUE;
5308 /* create the new pool from the disks of the original pool */
5309 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
5311 newspa->spa_splitting_newspa = B_FALSE;
5316 /* if that worked, generate a real config for the new pool */
5317 if (newspa->spa_root_vdev != NULL) {
5318 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
5319 NV_UNIQUE_NAME, KM_SLEEP) == 0);
5320 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
5321 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
5322 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
5327 if (props != NULL) {
5328 spa_configfile_set(newspa, props, B_FALSE);
5329 error = spa_prop_set(newspa, props);
5334 /* flush everything */
5335 txg = spa_vdev_config_enter(newspa);
5336 vdev_config_dirty(newspa->spa_root_vdev);
5337 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
5339 if (zio_injection_enabled)
5340 zio_handle_panic_injection(spa, FTAG, 2);
5342 spa_async_resume(newspa);
5344 /* finally, update the original pool's config */
5345 txg = spa_vdev_config_enter(spa);
5346 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
5347 error = dmu_tx_assign(tx, TXG_WAIT);
5350 for (c = 0; c < children; c++) {
5351 if (vml[c] != NULL) {
5354 spa_history_log_internal(spa, "detach", tx,
5355 "vdev=%s", vml[c]->vdev_path);
5359 vdev_config_dirty(spa->spa_root_vdev);
5360 spa->spa_config_splitting = NULL;
5364 (void) spa_vdev_exit(spa, NULL, txg, 0);
5366 if (zio_injection_enabled)
5367 zio_handle_panic_injection(spa, FTAG, 3);
5369 /* split is complete; log a history record */
5370 spa_history_log_internal(newspa, "split", NULL,
5371 "from pool %s", spa_name(spa));
5373 kmem_free(vml, children * sizeof (vdev_t *));
5375 /* if we're not going to mount the filesystems in userland, export */
5377 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
5384 spa_deactivate(newspa);
5387 txg = spa_vdev_config_enter(spa);
5389 /* re-online all offlined disks */
5390 for (c = 0; c < children; c++) {
5392 vml[c]->vdev_offline = B_FALSE;
5394 vdev_reopen(spa->spa_root_vdev);
5396 nvlist_free(spa->spa_config_splitting);
5397 spa->spa_config_splitting = NULL;
5398 (void) spa_vdev_exit(spa, NULL, txg, error);
5400 kmem_free(vml, children * sizeof (vdev_t *));
5405 spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
5407 for (int i = 0; i < count; i++) {
5410 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
5413 if (guid == target_guid)
5421 spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
5422 nvlist_t *dev_to_remove)
5424 nvlist_t **newdev = NULL;
5427 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
5429 for (int i = 0, j = 0; i < count; i++) {
5430 if (dev[i] == dev_to_remove)
5432 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
5435 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
5436 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
5438 for (int i = 0; i < count - 1; i++)
5439 nvlist_free(newdev[i]);
5442 kmem_free(newdev, (count - 1) * sizeof (void *));
5446 * Evacuate the device.
5449 spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
5454 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5455 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5456 ASSERT(vd == vd->vdev_top);
5459 * Evacuate the device. We don't hold the config lock as writer
5460 * since we need to do I/O but we do keep the
5461 * spa_namespace_lock held. Once this completes the device
5462 * should no longer have any blocks allocated on it.
5464 if (vd->vdev_islog) {
5465 if (vd->vdev_stat.vs_alloc != 0)
5466 error = spa_offline_log(spa);
5468 error = SET_ERROR(ENOTSUP);
5475 * The evacuation succeeded. Remove any remaining MOS metadata
5476 * associated with this vdev, and wait for these changes to sync.
5478 ASSERT0(vd->vdev_stat.vs_alloc);
5479 txg = spa_vdev_config_enter(spa);
5480 vd->vdev_removing = B_TRUE;
5481 vdev_dirty_leaves(vd, VDD_DTL, txg);
5482 vdev_config_dirty(vd);
5483 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5489 * Complete the removal by cleaning up the namespace.
5492 spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
5494 vdev_t *rvd = spa->spa_root_vdev;
5495 uint64_t id = vd->vdev_id;
5496 boolean_t last_vdev = (id == (rvd->vdev_children - 1));
5498 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5499 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5500 ASSERT(vd == vd->vdev_top);
5503 * Only remove any devices which are empty.
5505 if (vd->vdev_stat.vs_alloc != 0)
5508 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5510 if (list_link_active(&vd->vdev_state_dirty_node))
5511 vdev_state_clean(vd);
5512 if (list_link_active(&vd->vdev_config_dirty_node))
5513 vdev_config_clean(vd);
5518 vdev_compact_children(rvd);
5520 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
5521 vdev_add_child(rvd, vd);
5523 vdev_config_dirty(rvd);
5526 * Reassess the health of our root vdev.
5532 * Remove a device from the pool -
5534 * Removing a device from the vdev namespace requires several steps
5535 * and can take a significant amount of time. As a result we use
5536 * the spa_vdev_config_[enter/exit] functions which allow us to
5537 * grab and release the spa_config_lock while still holding the namespace
5538 * lock. During each step the configuration is synced out.
5540 * Currently, this supports removing only hot spares, slogs, and level 2 ARC
5544 spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
5547 metaslab_group_t *mg;
5548 nvlist_t **spares, **l2cache, *nv;
5550 uint_t nspares, nl2cache;
5552 boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
5554 ASSERT(spa_writeable(spa));
5557 txg = spa_vdev_enter(spa);
5559 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5561 if (spa->spa_spares.sav_vdevs != NULL &&
5562 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5563 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
5564 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
5566 * Only remove the hot spare if it's not currently in use
5569 if (vd == NULL || unspare) {
5570 spa_vdev_remove_aux(spa->spa_spares.sav_config,
5571 ZPOOL_CONFIG_SPARES, spares, nspares, nv);
5572 spa_load_spares(spa);
5573 spa->spa_spares.sav_sync = B_TRUE;
5575 error = SET_ERROR(EBUSY);
5577 } else if (spa->spa_l2cache.sav_vdevs != NULL &&
5578 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5579 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
5580 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
5582 * Cache devices can always be removed.
5584 spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
5585 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
5586 spa_load_l2cache(spa);
5587 spa->spa_l2cache.sav_sync = B_TRUE;
5588 } else if (vd != NULL && vd->vdev_islog) {
5590 ASSERT(vd == vd->vdev_top);
5595 * Stop allocating from this vdev.
5597 metaslab_group_passivate(mg);
5600 * Wait for the youngest allocations and frees to sync,
5601 * and then wait for the deferral of those frees to finish.
5603 spa_vdev_config_exit(spa, NULL,
5604 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
5607 * Attempt to evacuate the vdev.
5609 error = spa_vdev_remove_evacuate(spa, vd);
5611 txg = spa_vdev_config_enter(spa);
5614 * If we couldn't evacuate the vdev, unwind.
5617 metaslab_group_activate(mg);
5618 return (spa_vdev_exit(spa, NULL, txg, error));
5622 * Clean up the vdev namespace.
5624 spa_vdev_remove_from_namespace(spa, vd);
5626 } else if (vd != NULL) {
5628 * Normal vdevs cannot be removed (yet).
5630 error = SET_ERROR(ENOTSUP);
5633 * There is no vdev of any kind with the specified guid.
5635 error = SET_ERROR(ENOENT);
5639 return (spa_vdev_exit(spa, NULL, txg, error));
5645 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5646 * currently spared, so we can detach it.
5649 spa_vdev_resilver_done_hunt(vdev_t *vd)
5651 vdev_t *newvd, *oldvd;
5653 for (int c = 0; c < vd->vdev_children; c++) {
5654 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
5660 * Check for a completed replacement. We always consider the first
5661 * vdev in the list to be the oldest vdev, and the last one to be
5662 * the newest (see spa_vdev_attach() for how that works). In
5663 * the case where the newest vdev is faulted, we will not automatically
5664 * remove it after a resilver completes. This is OK as it will require
5665 * user intervention to determine which disk the admin wishes to keep.
5667 if (vd->vdev_ops == &vdev_replacing_ops) {
5668 ASSERT(vd->vdev_children > 1);
5670 newvd = vd->vdev_child[vd->vdev_children - 1];
5671 oldvd = vd->vdev_child[0];
5673 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
5674 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5675 !vdev_dtl_required(oldvd))
5680 * Check for a completed resilver with the 'unspare' flag set.
5682 if (vd->vdev_ops == &vdev_spare_ops) {
5683 vdev_t *first = vd->vdev_child[0];
5684 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
5686 if (last->vdev_unspare) {
5689 } else if (first->vdev_unspare) {
5696 if (oldvd != NULL &&
5697 vdev_dtl_empty(newvd, DTL_MISSING) &&
5698 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5699 !vdev_dtl_required(oldvd))
5703 * If there are more than two spares attached to a disk,
5704 * and those spares are not required, then we want to
5705 * attempt to free them up now so that they can be used
5706 * by other pools. Once we're back down to a single
5707 * disk+spare, we stop removing them.
5709 if (vd->vdev_children > 2) {
5710 newvd = vd->vdev_child[1];
5712 if (newvd->vdev_isspare && last->vdev_isspare &&
5713 vdev_dtl_empty(last, DTL_MISSING) &&
5714 vdev_dtl_empty(last, DTL_OUTAGE) &&
5715 !vdev_dtl_required(newvd))
5724 spa_vdev_resilver_done(spa_t *spa)
5726 vdev_t *vd, *pvd, *ppvd;
5727 uint64_t guid, sguid, pguid, ppguid;
5729 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5731 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
5732 pvd = vd->vdev_parent;
5733 ppvd = pvd->vdev_parent;
5734 guid = vd->vdev_guid;
5735 pguid = pvd->vdev_guid;
5736 ppguid = ppvd->vdev_guid;
5739 * If we have just finished replacing a hot spared device, then
5740 * we need to detach the parent's first child (the original hot
5743 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
5744 ppvd->vdev_children == 2) {
5745 ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
5746 sguid = ppvd->vdev_child[1]->vdev_guid;
5748 ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd));
5750 spa_config_exit(spa, SCL_ALL, FTAG);
5751 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
5753 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
5755 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5758 spa_config_exit(spa, SCL_ALL, FTAG);
5762 * Update the stored path or FRU for this vdev.
5765 spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
5769 boolean_t sync = B_FALSE;
5771 ASSERT(spa_writeable(spa));
5773 spa_vdev_state_enter(spa, SCL_ALL);
5775 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
5776 return (spa_vdev_state_exit(spa, NULL, ENOENT));
5778 if (!vd->vdev_ops->vdev_op_leaf)
5779 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
5782 if (strcmp(value, vd->vdev_path) != 0) {
5783 spa_strfree(vd->vdev_path);
5784 vd->vdev_path = spa_strdup(value);
5788 if (vd->vdev_fru == NULL) {
5789 vd->vdev_fru = spa_strdup(value);
5791 } else if (strcmp(value, vd->vdev_fru) != 0) {
5792 spa_strfree(vd->vdev_fru);
5793 vd->vdev_fru = spa_strdup(value);
5798 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
5802 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
5804 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5808 spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5810 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5814 * ==========================================================================
5816 * ==========================================================================
5820 spa_scan_stop(spa_t *spa)
5822 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5823 if (dsl_scan_resilvering(spa->spa_dsl_pool))
5824 return (SET_ERROR(EBUSY));
5825 return (dsl_scan_cancel(spa->spa_dsl_pool));
5829 spa_scan(spa_t *spa, pool_scan_func_t func)
5831 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5833 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5834 return (SET_ERROR(ENOTSUP));
5837 * If a resilver was requested, but there is no DTL on a
5838 * writeable leaf device, we have nothing to do.
5840 if (func == POOL_SCAN_RESILVER &&
5841 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5842 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5846 return (dsl_scan(spa->spa_dsl_pool, func));
5850 * ==========================================================================
5851 * SPA async task processing
5852 * ==========================================================================
5856 spa_async_remove(spa_t *spa, vdev_t *vd)
5858 if (vd->vdev_remove_wanted) {
5859 vd->vdev_remove_wanted = B_FALSE;
5860 vd->vdev_delayed_close = B_FALSE;
5861 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5864 * We want to clear the stats, but we don't want to do a full
5865 * vdev_clear() as that will cause us to throw away
5866 * degraded/faulted state as well as attempt to reopen the
5867 * device, all of which is a waste.
5869 vd->vdev_stat.vs_read_errors = 0;
5870 vd->vdev_stat.vs_write_errors = 0;
5871 vd->vdev_stat.vs_checksum_errors = 0;
5873 vdev_state_dirty(vd->vdev_top);
5876 for (int c = 0; c < vd->vdev_children; c++)
5877 spa_async_remove(spa, vd->vdev_child[c]);
5881 spa_async_probe(spa_t *spa, vdev_t *vd)
5883 if (vd->vdev_probe_wanted) {
5884 vd->vdev_probe_wanted = B_FALSE;
5885 vdev_reopen(vd); /* vdev_open() does the actual probe */
5888 for (int c = 0; c < vd->vdev_children; c++)
5889 spa_async_probe(spa, vd->vdev_child[c]);
5893 spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5899 if (!spa->spa_autoexpand)
5902 for (int c = 0; c < vd->vdev_children; c++) {
5903 vdev_t *cvd = vd->vdev_child[c];
5904 spa_async_autoexpand(spa, cvd);
5907 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5910 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
5911 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
5913 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5914 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
5916 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
5917 ESC_ZFS_VDEV_AUTOEXPAND, attr, &eid, DDI_SLEEP);
5920 kmem_free(physpath, MAXPATHLEN);
5924 spa_async_thread(void *arg)
5929 ASSERT(spa->spa_sync_on);
5931 mutex_enter(&spa->spa_async_lock);
5932 tasks = spa->spa_async_tasks;
5933 spa->spa_async_tasks &= SPA_ASYNC_REMOVE;
5934 mutex_exit(&spa->spa_async_lock);
5937 * See if the config needs to be updated.
5939 if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5940 uint64_t old_space, new_space;
5942 mutex_enter(&spa_namespace_lock);
5943 old_space = metaslab_class_get_space(spa_normal_class(spa));
5944 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5945 new_space = metaslab_class_get_space(spa_normal_class(spa));
5946 mutex_exit(&spa_namespace_lock);
5949 * If the pool grew as a result of the config update,
5950 * then log an internal history event.
5952 if (new_space != old_space) {
5953 spa_history_log_internal(spa, "vdev online", NULL,
5954 "pool '%s' size: %llu(+%llu)",
5955 spa_name(spa), new_space, new_space - old_space);
5959 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5960 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5961 spa_async_autoexpand(spa, spa->spa_root_vdev);
5962 spa_config_exit(spa, SCL_CONFIG, FTAG);
5966 * See if any devices need to be probed.
5968 if (tasks & SPA_ASYNC_PROBE) {
5969 spa_vdev_state_enter(spa, SCL_NONE);
5970 spa_async_probe(spa, spa->spa_root_vdev);
5971 (void) spa_vdev_state_exit(spa, NULL, 0);
5975 * If any devices are done replacing, detach them.
5977 if (tasks & SPA_ASYNC_RESILVER_DONE)
5978 spa_vdev_resilver_done(spa);
5981 * Kick off a resilver.
5983 if (tasks & SPA_ASYNC_RESILVER)
5984 dsl_resilver_restart(spa->spa_dsl_pool, 0);
5987 * Let the world know that we're done.
5989 mutex_enter(&spa->spa_async_lock);
5990 spa->spa_async_thread = NULL;
5991 cv_broadcast(&spa->spa_async_cv);
5992 mutex_exit(&spa->spa_async_lock);
5997 spa_async_thread_vd(void *arg)
6002 ASSERT(spa->spa_sync_on);
6004 mutex_enter(&spa->spa_async_lock);
6005 tasks = spa->spa_async_tasks;
6007 spa->spa_async_tasks &= ~SPA_ASYNC_REMOVE;
6008 mutex_exit(&spa->spa_async_lock);
6011 * See if any devices need to be marked REMOVED.
6013 if (tasks & SPA_ASYNC_REMOVE) {
6014 spa_vdev_state_enter(spa, SCL_NONE);
6015 spa_async_remove(spa, spa->spa_root_vdev);
6016 for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
6017 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
6018 for (int i = 0; i < spa->spa_spares.sav_count; i++)
6019 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
6020 (void) spa_vdev_state_exit(spa, NULL, 0);
6024 * Let the world know that we're done.
6026 mutex_enter(&spa->spa_async_lock);
6027 tasks = spa->spa_async_tasks;
6028 if ((tasks & SPA_ASYNC_REMOVE) != 0)
6030 spa->spa_async_thread_vd = NULL;
6031 cv_broadcast(&spa->spa_async_cv);
6032 mutex_exit(&spa->spa_async_lock);
6037 spa_async_suspend(spa_t *spa)
6039 mutex_enter(&spa->spa_async_lock);
6040 spa->spa_async_suspended++;
6041 while (spa->spa_async_thread != NULL &&
6042 spa->spa_async_thread_vd != NULL)
6043 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
6044 mutex_exit(&spa->spa_async_lock);
6048 spa_async_resume(spa_t *spa)
6050 mutex_enter(&spa->spa_async_lock);
6051 ASSERT(spa->spa_async_suspended != 0);
6052 spa->spa_async_suspended--;
6053 mutex_exit(&spa->spa_async_lock);
6057 spa_async_tasks_pending(spa_t *spa)
6059 uint_t non_config_tasks;
6061 boolean_t config_task_suspended;
6063 non_config_tasks = spa->spa_async_tasks & ~(SPA_ASYNC_CONFIG_UPDATE |
6065 config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE;
6066 if (spa->spa_ccw_fail_time == 0) {
6067 config_task_suspended = B_FALSE;
6069 config_task_suspended =
6070 (gethrtime() - spa->spa_ccw_fail_time) <
6071 (zfs_ccw_retry_interval * NANOSEC);
6074 return (non_config_tasks || (config_task && !config_task_suspended));
6078 spa_async_dispatch(spa_t *spa)
6080 mutex_enter(&spa->spa_async_lock);
6081 if (spa_async_tasks_pending(spa) &&
6082 !spa->spa_async_suspended &&
6083 spa->spa_async_thread == NULL &&
6085 spa->spa_async_thread = thread_create(NULL, 0,
6086 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
6087 mutex_exit(&spa->spa_async_lock);
6091 spa_async_dispatch_vd(spa_t *spa)
6093 mutex_enter(&spa->spa_async_lock);
6094 if ((spa->spa_async_tasks & SPA_ASYNC_REMOVE) != 0 &&
6095 !spa->spa_async_suspended &&
6096 spa->spa_async_thread_vd == NULL &&
6098 spa->spa_async_thread_vd = thread_create(NULL, 0,
6099 spa_async_thread_vd, spa, 0, &p0, TS_RUN, maxclsyspri);
6100 mutex_exit(&spa->spa_async_lock);
6104 spa_async_request(spa_t *spa, int task)
6106 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
6107 mutex_enter(&spa->spa_async_lock);
6108 spa->spa_async_tasks |= task;
6109 mutex_exit(&spa->spa_async_lock);
6110 spa_async_dispatch_vd(spa);
6114 * ==========================================================================
6115 * SPA syncing routines
6116 * ==========================================================================
6120 bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
6123 bpobj_enqueue(bpo, bp, tx);
6128 spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
6132 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
6133 BP_GET_PSIZE(bp), zio->io_flags));
6138 * Note: this simple function is not inlined to make it easier to dtrace the
6139 * amount of time spent syncing frees.
6142 spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx)
6144 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6145 bplist_iterate(bpl, spa_free_sync_cb, zio, tx);
6146 VERIFY(zio_wait(zio) == 0);
6150 * Note: this simple function is not inlined to make it easier to dtrace the
6151 * amount of time spent syncing deferred frees.
6154 spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx)
6156 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6157 VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj,
6158 spa_free_sync_cb, zio, tx), ==, 0);
6159 VERIFY0(zio_wait(zio));
6164 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
6166 char *packed = NULL;
6171 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
6174 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
6175 * information. This avoids the dmu_buf_will_dirty() path and
6176 * saves us a pre-read to get data we don't actually care about.
6178 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
6179 packed = kmem_alloc(bufsize, KM_SLEEP);
6181 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
6183 bzero(packed + nvsize, bufsize - nvsize);
6185 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
6187 kmem_free(packed, bufsize);
6189 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
6190 dmu_buf_will_dirty(db, tx);
6191 *(uint64_t *)db->db_data = nvsize;
6192 dmu_buf_rele(db, FTAG);
6196 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
6197 const char *config, const char *entry)
6207 * Update the MOS nvlist describing the list of available devices.
6208 * spa_validate_aux() will have already made sure this nvlist is
6209 * valid and the vdevs are labeled appropriately.
6211 if (sav->sav_object == 0) {
6212 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
6213 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
6214 sizeof (uint64_t), tx);
6215 VERIFY(zap_update(spa->spa_meta_objset,
6216 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
6217 &sav->sav_object, tx) == 0);
6220 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
6221 if (sav->sav_count == 0) {
6222 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
6224 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
6225 for (i = 0; i < sav->sav_count; i++)
6226 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
6227 B_FALSE, VDEV_CONFIG_L2CACHE);
6228 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
6229 sav->sav_count) == 0);
6230 for (i = 0; i < sav->sav_count; i++)
6231 nvlist_free(list[i]);
6232 kmem_free(list, sav->sav_count * sizeof (void *));
6235 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
6236 nvlist_free(nvroot);
6238 sav->sav_sync = B_FALSE;
6242 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
6246 if (list_is_empty(&spa->spa_config_dirty_list))
6249 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6251 config = spa_config_generate(spa, spa->spa_root_vdev,
6252 dmu_tx_get_txg(tx), B_FALSE);
6255 * If we're upgrading the spa version then make sure that
6256 * the config object gets updated with the correct version.
6258 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
6259 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
6260 spa->spa_uberblock.ub_version);
6262 spa_config_exit(spa, SCL_STATE, FTAG);
6264 if (spa->spa_config_syncing)
6265 nvlist_free(spa->spa_config_syncing);
6266 spa->spa_config_syncing = config;
6268 spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
6272 spa_sync_version(void *arg, dmu_tx_t *tx)
6274 uint64_t *versionp = arg;
6275 uint64_t version = *versionp;
6276 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
6279 * Setting the version is special cased when first creating the pool.
6281 ASSERT(tx->tx_txg != TXG_INITIAL);
6283 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
6284 ASSERT(version >= spa_version(spa));
6286 spa->spa_uberblock.ub_version = version;
6287 vdev_config_dirty(spa->spa_root_vdev);
6288 spa_history_log_internal(spa, "set", tx, "version=%lld", version);
6292 * Set zpool properties.
6295 spa_sync_props(void *arg, dmu_tx_t *tx)
6297 nvlist_t *nvp = arg;
6298 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
6299 objset_t *mos = spa->spa_meta_objset;
6300 nvpair_t *elem = NULL;
6302 mutex_enter(&spa->spa_props_lock);
6304 while ((elem = nvlist_next_nvpair(nvp, elem))) {
6306 char *strval, *fname;
6308 const char *propname;
6309 zprop_type_t proptype;
6312 switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
6315 * We checked this earlier in spa_prop_validate().
6317 ASSERT(zpool_prop_feature(nvpair_name(elem)));
6319 fname = strchr(nvpair_name(elem), '@') + 1;
6320 VERIFY0(zfeature_lookup_name(fname, &fid));
6322 spa_feature_enable(spa, fid, tx);
6323 spa_history_log_internal(spa, "set", tx,
6324 "%s=enabled", nvpair_name(elem));
6327 case ZPOOL_PROP_VERSION:
6328 intval = fnvpair_value_uint64(elem);
6330 * The version is synced seperatly before other
6331 * properties and should be correct by now.
6333 ASSERT3U(spa_version(spa), >=, intval);
6336 case ZPOOL_PROP_ALTROOT:
6338 * 'altroot' is a non-persistent property. It should
6339 * have been set temporarily at creation or import time.
6341 ASSERT(spa->spa_root != NULL);
6344 case ZPOOL_PROP_READONLY:
6345 case ZPOOL_PROP_CACHEFILE:
6347 * 'readonly' and 'cachefile' are also non-persisitent
6351 case ZPOOL_PROP_COMMENT:
6352 strval = fnvpair_value_string(elem);
6353 if (spa->spa_comment != NULL)
6354 spa_strfree(spa->spa_comment);
6355 spa->spa_comment = spa_strdup(strval);
6357 * We need to dirty the configuration on all the vdevs
6358 * so that their labels get updated. It's unnecessary
6359 * to do this for pool creation since the vdev's
6360 * configuratoin has already been dirtied.
6362 if (tx->tx_txg != TXG_INITIAL)
6363 vdev_config_dirty(spa->spa_root_vdev);
6364 spa_history_log_internal(spa, "set", tx,
6365 "%s=%s", nvpair_name(elem), strval);
6369 * Set pool property values in the poolprops mos object.
6371 if (spa->spa_pool_props_object == 0) {
6372 spa->spa_pool_props_object =
6373 zap_create_link(mos, DMU_OT_POOL_PROPS,
6374 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
6378 /* normalize the property name */
6379 propname = zpool_prop_to_name(prop);
6380 proptype = zpool_prop_get_type(prop);
6382 if (nvpair_type(elem) == DATA_TYPE_STRING) {
6383 ASSERT(proptype == PROP_TYPE_STRING);
6384 strval = fnvpair_value_string(elem);
6385 VERIFY0(zap_update(mos,
6386 spa->spa_pool_props_object, propname,
6387 1, strlen(strval) + 1, strval, tx));
6388 spa_history_log_internal(spa, "set", tx,
6389 "%s=%s", nvpair_name(elem), strval);
6390 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
6391 intval = fnvpair_value_uint64(elem);
6393 if (proptype == PROP_TYPE_INDEX) {
6395 VERIFY0(zpool_prop_index_to_string(
6396 prop, intval, &unused));
6398 VERIFY0(zap_update(mos,
6399 spa->spa_pool_props_object, propname,
6400 8, 1, &intval, tx));
6401 spa_history_log_internal(spa, "set", tx,
6402 "%s=%lld", nvpair_name(elem), intval);
6404 ASSERT(0); /* not allowed */
6408 case ZPOOL_PROP_DELEGATION:
6409 spa->spa_delegation = intval;
6411 case ZPOOL_PROP_BOOTFS:
6412 spa->spa_bootfs = intval;
6414 case ZPOOL_PROP_FAILUREMODE:
6415 spa->spa_failmode = intval;
6417 case ZPOOL_PROP_AUTOEXPAND:
6418 spa->spa_autoexpand = intval;
6419 if (tx->tx_txg != TXG_INITIAL)
6420 spa_async_request(spa,
6421 SPA_ASYNC_AUTOEXPAND);
6423 case ZPOOL_PROP_DEDUPDITTO:
6424 spa->spa_dedup_ditto = intval;
6433 mutex_exit(&spa->spa_props_lock);
6437 * Perform one-time upgrade on-disk changes. spa_version() does not
6438 * reflect the new version this txg, so there must be no changes this
6439 * txg to anything that the upgrade code depends on after it executes.
6440 * Therefore this must be called after dsl_pool_sync() does the sync
6444 spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
6446 dsl_pool_t *dp = spa->spa_dsl_pool;
6448 ASSERT(spa->spa_sync_pass == 1);
6450 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
6452 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
6453 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
6454 dsl_pool_create_origin(dp, tx);
6456 /* Keeping the origin open increases spa_minref */
6457 spa->spa_minref += 3;
6460 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
6461 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
6462 dsl_pool_upgrade_clones(dp, tx);
6465 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
6466 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
6467 dsl_pool_upgrade_dir_clones(dp, tx);
6469 /* Keeping the freedir open increases spa_minref */
6470 spa->spa_minref += 3;
6473 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES &&
6474 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
6475 spa_feature_create_zap_objects(spa, tx);
6479 * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable
6480 * when possibility to use lz4 compression for metadata was added
6481 * Old pools that have this feature enabled must be upgraded to have
6482 * this feature active
6484 if (spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
6485 boolean_t lz4_en = spa_feature_is_enabled(spa,
6486 SPA_FEATURE_LZ4_COMPRESS);
6487 boolean_t lz4_ac = spa_feature_is_active(spa,
6488 SPA_FEATURE_LZ4_COMPRESS);
6490 if (lz4_en && !lz4_ac)
6491 spa_feature_incr(spa, SPA_FEATURE_LZ4_COMPRESS, tx);
6493 rrw_exit(&dp->dp_config_rwlock, FTAG);
6497 * Sync the specified transaction group. New blocks may be dirtied as
6498 * part of the process, so we iterate until it converges.
6501 spa_sync(spa_t *spa, uint64_t txg)
6503 dsl_pool_t *dp = spa->spa_dsl_pool;
6504 objset_t *mos = spa->spa_meta_objset;
6505 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
6506 vdev_t *rvd = spa->spa_root_vdev;
6511 VERIFY(spa_writeable(spa));
6514 * Lock out configuration changes.
6516 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
6518 spa->spa_syncing_txg = txg;
6519 spa->spa_sync_pass = 0;
6522 * If there are any pending vdev state changes, convert them
6523 * into config changes that go out with this transaction group.
6525 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6526 while (list_head(&spa->spa_state_dirty_list) != NULL) {
6528 * We need the write lock here because, for aux vdevs,
6529 * calling vdev_config_dirty() modifies sav_config.
6530 * This is ugly and will become unnecessary when we
6531 * eliminate the aux vdev wart by integrating all vdevs
6532 * into the root vdev tree.
6534 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6535 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
6536 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
6537 vdev_state_clean(vd);
6538 vdev_config_dirty(vd);
6540 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6541 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
6543 spa_config_exit(spa, SCL_STATE, FTAG);
6545 tx = dmu_tx_create_assigned(dp, txg);
6547 spa->spa_sync_starttime = gethrtime();
6549 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid,
6550 spa->spa_sync_starttime + spa->spa_deadman_synctime));
6553 callout_reset(&spa->spa_deadman_cycid,
6554 hz * spa->spa_deadman_synctime / NANOSEC, spa_deadman, spa);
6559 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
6560 * set spa_deflate if we have no raid-z vdevs.
6562 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
6563 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
6566 for (i = 0; i < rvd->vdev_children; i++) {
6567 vd = rvd->vdev_child[i];
6568 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
6571 if (i == rvd->vdev_children) {
6572 spa->spa_deflate = TRUE;
6573 VERIFY(0 == zap_add(spa->spa_meta_objset,
6574 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
6575 sizeof (uint64_t), 1, &spa->spa_deflate, tx));
6580 * If anything has changed in this txg, or if someone is waiting
6581 * for this txg to sync (eg, spa_vdev_remove()), push the
6582 * deferred frees from the previous txg. If not, leave them
6583 * alone so that we don't generate work on an otherwise idle
6586 if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
6587 !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
6588 !txg_list_empty(&dp->dp_sync_tasks, txg) ||
6589 ((dsl_scan_active(dp->dp_scan) ||
6590 txg_sync_waiting(dp)) && !spa_shutting_down(spa))) {
6591 spa_sync_deferred_frees(spa, tx);
6595 * Iterate to convergence.
6598 int pass = ++spa->spa_sync_pass;
6600 spa_sync_config_object(spa, tx);
6601 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
6602 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
6603 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
6604 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
6605 spa_errlog_sync(spa, txg);
6606 dsl_pool_sync(dp, txg);
6608 if (pass < zfs_sync_pass_deferred_free) {
6609 spa_sync_frees(spa, free_bpl, tx);
6611 bplist_iterate(free_bpl, bpobj_enqueue_cb,
6612 &spa->spa_deferred_bpobj, tx);
6616 dsl_scan_sync(dp, tx);
6618 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
6622 spa_sync_upgrades(spa, tx);
6624 } while (dmu_objset_is_dirty(mos, txg));
6627 * Rewrite the vdev configuration (which includes the uberblock)
6628 * to commit the transaction group.
6630 * If there are no dirty vdevs, we sync the uberblock to a few
6631 * random top-level vdevs that are known to be visible in the
6632 * config cache (see spa_vdev_add() for a complete description).
6633 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
6637 * We hold SCL_STATE to prevent vdev open/close/etc.
6638 * while we're attempting to write the vdev labels.
6640 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6642 if (list_is_empty(&spa->spa_config_dirty_list)) {
6643 vdev_t *svd[SPA_DVAS_PER_BP];
6645 int children = rvd->vdev_children;
6646 int c0 = spa_get_random(children);
6648 for (int c = 0; c < children; c++) {
6649 vd = rvd->vdev_child[(c0 + c) % children];
6650 if (vd->vdev_ms_array == 0 || vd->vdev_islog)
6652 svd[svdcount++] = vd;
6653 if (svdcount == SPA_DVAS_PER_BP)
6656 error = vdev_config_sync(svd, svdcount, txg, B_FALSE);
6658 error = vdev_config_sync(svd, svdcount, txg,
6661 error = vdev_config_sync(rvd->vdev_child,
6662 rvd->vdev_children, txg, B_FALSE);
6664 error = vdev_config_sync(rvd->vdev_child,
6665 rvd->vdev_children, txg, B_TRUE);
6669 spa->spa_last_synced_guid = rvd->vdev_guid;
6671 spa_config_exit(spa, SCL_STATE, FTAG);
6675 zio_suspend(spa, NULL);
6676 zio_resume_wait(spa);
6681 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY));
6684 callout_drain(&spa->spa_deadman_cycid);
6689 * Clear the dirty config list.
6691 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
6692 vdev_config_clean(vd);
6695 * Now that the new config has synced transactionally,
6696 * let it become visible to the config cache.
6698 if (spa->spa_config_syncing != NULL) {
6699 spa_config_set(spa, spa->spa_config_syncing);
6700 spa->spa_config_txg = txg;
6701 spa->spa_config_syncing = NULL;
6704 spa->spa_ubsync = spa->spa_uberblock;
6706 dsl_pool_sync_done(dp, txg);
6709 * Update usable space statistics.
6711 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
6712 vdev_sync_done(vd, txg);
6714 spa_update_dspace(spa);
6717 * It had better be the case that we didn't dirty anything
6718 * since vdev_config_sync().
6720 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
6721 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
6722 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
6724 spa->spa_sync_pass = 0;
6726 spa_config_exit(spa, SCL_CONFIG, FTAG);
6728 spa_handle_ignored_writes(spa);
6731 * If any async tasks have been requested, kick them off.
6733 spa_async_dispatch(spa);
6734 spa_async_dispatch_vd(spa);
6738 * Sync all pools. We don't want to hold the namespace lock across these
6739 * operations, so we take a reference on the spa_t and drop the lock during the
6743 spa_sync_allpools(void)
6746 mutex_enter(&spa_namespace_lock);
6747 while ((spa = spa_next(spa)) != NULL) {
6748 if (spa_state(spa) != POOL_STATE_ACTIVE ||
6749 !spa_writeable(spa) || spa_suspended(spa))
6751 spa_open_ref(spa, FTAG);
6752 mutex_exit(&spa_namespace_lock);
6753 txg_wait_synced(spa_get_dsl(spa), 0);
6754 mutex_enter(&spa_namespace_lock);
6755 spa_close(spa, FTAG);
6757 mutex_exit(&spa_namespace_lock);
6761 * ==========================================================================
6762 * Miscellaneous routines
6763 * ==========================================================================
6767 * Remove all pools in the system.
6775 * Remove all cached state. All pools should be closed now,
6776 * so every spa in the AVL tree should be unreferenced.
6778 mutex_enter(&spa_namespace_lock);
6779 while ((spa = spa_next(NULL)) != NULL) {
6781 * Stop async tasks. The async thread may need to detach
6782 * a device that's been replaced, which requires grabbing
6783 * spa_namespace_lock, so we must drop it here.
6785 spa_open_ref(spa, FTAG);
6786 mutex_exit(&spa_namespace_lock);
6787 spa_async_suspend(spa);
6788 mutex_enter(&spa_namespace_lock);
6789 spa_close(spa, FTAG);
6791 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
6793 spa_deactivate(spa);
6797 mutex_exit(&spa_namespace_lock);
6801 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
6806 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
6810 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
6811 vd = spa->spa_l2cache.sav_vdevs[i];
6812 if (vd->vdev_guid == guid)
6816 for (i = 0; i < spa->spa_spares.sav_count; i++) {
6817 vd = spa->spa_spares.sav_vdevs[i];
6818 if (vd->vdev_guid == guid)
6827 spa_upgrade(spa_t *spa, uint64_t version)
6829 ASSERT(spa_writeable(spa));
6831 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6834 * This should only be called for a non-faulted pool, and since a
6835 * future version would result in an unopenable pool, this shouldn't be
6838 ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version));
6839 ASSERT3U(version, >=, spa->spa_uberblock.ub_version);
6841 spa->spa_uberblock.ub_version = version;
6842 vdev_config_dirty(spa->spa_root_vdev);
6844 spa_config_exit(spa, SCL_ALL, FTAG);
6846 txg_wait_synced(spa_get_dsl(spa), 0);
6850 spa_has_spare(spa_t *spa, uint64_t guid)
6854 spa_aux_vdev_t *sav = &spa->spa_spares;
6856 for (i = 0; i < sav->sav_count; i++)
6857 if (sav->sav_vdevs[i]->vdev_guid == guid)
6860 for (i = 0; i < sav->sav_npending; i++) {
6861 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
6862 &spareguid) == 0 && spareguid == guid)
6870 * Check if a pool has an active shared spare device.
6871 * Note: reference count of an active spare is 2, as a spare and as a replace
6874 spa_has_active_shared_spare(spa_t *spa)
6878 spa_aux_vdev_t *sav = &spa->spa_spares;
6880 for (i = 0; i < sav->sav_count; i++) {
6881 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
6882 &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
6891 * Post a sysevent corresponding to the given event. The 'name' must be one of
6892 * the event definitions in sys/sysevent/eventdefs.h. The payload will be
6893 * filled in from the spa and (optionally) the vdev. This doesn't do anything
6894 * in the userland libzpool, as we don't want consumers to misinterpret ztest
6895 * or zdb as real changes.
6898 spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
6902 sysevent_attr_list_t *attr = NULL;
6903 sysevent_value_t value;
6906 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
6909 value.value_type = SE_DATA_TYPE_STRING;
6910 value.value.sv_string = spa_name(spa);
6911 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
6914 value.value_type = SE_DATA_TYPE_UINT64;
6915 value.value.sv_uint64 = spa_guid(spa);
6916 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
6920 value.value_type = SE_DATA_TYPE_UINT64;
6921 value.value.sv_uint64 = vd->vdev_guid;
6922 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
6926 if (vd->vdev_path) {
6927 value.value_type = SE_DATA_TYPE_STRING;
6928 value.value.sv_string = vd->vdev_path;
6929 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
6930 &value, SE_SLEEP) != 0)
6935 if (sysevent_attach_attributes(ev, attr) != 0)
6939 (void) log_sysevent(ev, SE_SLEEP, &eid);
6943 sysevent_free_attr(attr);