4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2013 by Delphix. All rights reserved.
25 * Copyright (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 TUNABLE_INT("vfs.zfs.check_hostid", &check_hostid);
88 SYSCTL_INT(_vfs_zfs, OID_AUTO, check_hostid, CTLFLAG_RW, &check_hostid, 0,
89 "Check hostid on import?");
92 * The interval, in seconds, at which failed configuration cache file writes
95 static int zfs_ccw_retry_interval = 300;
97 typedef enum zti_modes {
98 ZTI_MODE_FIXED, /* value is # of threads (min 1) */
99 ZTI_MODE_BATCH, /* cpu-intensive; value is ignored */
100 ZTI_MODE_NULL, /* don't create a taskq */
104 #define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) }
105 #define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 }
106 #define ZTI_NULL { ZTI_MODE_NULL, 0, 0 }
108 #define ZTI_N(n) ZTI_P(n, 1)
109 #define ZTI_ONE ZTI_N(1)
111 typedef struct zio_taskq_info {
112 zti_modes_t zti_mode;
117 static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
118 "issue", "issue_high", "intr", "intr_high"
122 * This table defines the taskq settings for each ZFS I/O type. When
123 * initializing a pool, we use this table to create an appropriately sized
124 * taskq. Some operations are low volume and therefore have a small, static
125 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE
126 * macros. Other operations process a large amount of data; the ZTI_BATCH
127 * macro causes us to create a taskq oriented for throughput. Some operations
128 * are so high frequency and short-lived that the taskq itself can become a a
129 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an
130 * additional degree of parallelism specified by the number of threads per-
131 * taskq and the number of taskqs; when dispatching an event in this case, the
132 * particular taskq is chosen at random.
134 * The different taskq priorities are to handle the different contexts (issue
135 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that
136 * need to be handled with minimum delay.
138 const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
139 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
140 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* NULL */
141 { ZTI_N(8), ZTI_NULL, ZTI_BATCH, ZTI_NULL }, /* READ */
142 { ZTI_BATCH, ZTI_N(5), ZTI_N(8), ZTI_N(5) }, /* WRITE */
143 { ZTI_P(12, 8), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FREE */
144 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* CLAIM */
145 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* IOCTL */
148 static void spa_sync_version(void *arg, dmu_tx_t *tx);
149 static void spa_sync_props(void *arg, dmu_tx_t *tx);
150 static boolean_t spa_has_active_shared_spare(spa_t *spa);
151 static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
152 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
154 static void spa_vdev_resilver_done(spa_t *spa);
156 uint_t zio_taskq_batch_pct = 75; /* 1 thread per cpu in pset */
158 id_t zio_taskq_psrset_bind = PS_NONE;
161 boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */
163 uint_t zio_taskq_basedc = 80; /* base duty cycle */
165 boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */
166 extern int zfs_sync_pass_deferred_free;
169 extern void spa_deadman(void *arg);
173 * This (illegal) pool name is used when temporarily importing a spa_t in order
174 * to get the vdev stats associated with the imported devices.
176 #define TRYIMPORT_NAME "$import"
179 * ==========================================================================
180 * SPA properties routines
181 * ==========================================================================
185 * Add a (source=src, propname=propval) list to an nvlist.
188 spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
189 uint64_t intval, zprop_source_t src)
191 const char *propname = zpool_prop_to_name(prop);
194 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
195 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
198 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
200 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
202 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
203 nvlist_free(propval);
207 * Get property values from the spa configuration.
210 spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
212 vdev_t *rvd = spa->spa_root_vdev;
213 dsl_pool_t *pool = spa->spa_dsl_pool;
214 uint64_t size, alloc, cap, version;
215 zprop_source_t src = ZPROP_SRC_NONE;
216 spa_config_dirent_t *dp;
217 metaslab_class_t *mc = spa_normal_class(spa);
219 ASSERT(MUTEX_HELD(&spa->spa_props_lock));
222 alloc = metaslab_class_get_alloc(spa_normal_class(spa));
223 size = metaslab_class_get_space(spa_normal_class(spa));
224 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
225 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
226 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
227 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
230 spa_prop_add_list(*nvp, ZPOOL_PROP_FRAGMENTATION, NULL,
231 metaslab_class_fragmentation(mc), src);
232 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL,
233 metaslab_class_expandable_space(mc), src);
234 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
235 (spa_mode(spa) == FREAD), src);
237 cap = (size == 0) ? 0 : (alloc * 100 / size);
238 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
240 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
241 ddt_get_pool_dedup_ratio(spa), src);
243 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
244 rvd->vdev_state, src);
246 version = spa_version(spa);
247 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
248 src = ZPROP_SRC_DEFAULT;
250 src = ZPROP_SRC_LOCAL;
251 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
256 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
257 * when opening pools before this version freedir will be NULL.
259 if (pool->dp_free_dir != NULL) {
260 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
261 pool->dp_free_dir->dd_phys->dd_used_bytes, src);
263 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
267 if (pool->dp_leak_dir != NULL) {
268 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, NULL,
269 pool->dp_leak_dir->dd_phys->dd_used_bytes, src);
271 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED,
276 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
278 if (spa->spa_comment != NULL) {
279 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
283 if (spa->spa_root != NULL)
284 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
287 if ((dp = list_head(&spa->spa_config_list)) != NULL) {
288 if (dp->scd_path == NULL) {
289 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
290 "none", 0, ZPROP_SRC_LOCAL);
291 } else if (strcmp(dp->scd_path, spa_config_path) != 0) {
292 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
293 dp->scd_path, 0, ZPROP_SRC_LOCAL);
299 * Get zpool property values.
302 spa_prop_get(spa_t *spa, nvlist_t **nvp)
304 objset_t *mos = spa->spa_meta_objset;
309 VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
311 mutex_enter(&spa->spa_props_lock);
314 * Get properties from the spa config.
316 spa_prop_get_config(spa, nvp);
318 /* If no pool property object, no more prop to get. */
319 if (mos == NULL || spa->spa_pool_props_object == 0) {
320 mutex_exit(&spa->spa_props_lock);
325 * Get properties from the MOS pool property object.
327 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
328 (err = zap_cursor_retrieve(&zc, &za)) == 0;
329 zap_cursor_advance(&zc)) {
332 zprop_source_t src = ZPROP_SRC_DEFAULT;
335 if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
338 switch (za.za_integer_length) {
340 /* integer property */
341 if (za.za_first_integer !=
342 zpool_prop_default_numeric(prop))
343 src = ZPROP_SRC_LOCAL;
345 if (prop == ZPOOL_PROP_BOOTFS) {
347 dsl_dataset_t *ds = NULL;
349 dp = spa_get_dsl(spa);
350 dsl_pool_config_enter(dp, FTAG);
351 if (err = dsl_dataset_hold_obj(dp,
352 za.za_first_integer, FTAG, &ds)) {
353 dsl_pool_config_exit(dp, FTAG);
358 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
360 dsl_dataset_name(ds, strval);
361 dsl_dataset_rele(ds, FTAG);
362 dsl_pool_config_exit(dp, FTAG);
365 intval = za.za_first_integer;
368 spa_prop_add_list(*nvp, prop, strval, intval, src);
372 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
377 /* string property */
378 strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
379 err = zap_lookup(mos, spa->spa_pool_props_object,
380 za.za_name, 1, za.za_num_integers, strval);
382 kmem_free(strval, za.za_num_integers);
385 spa_prop_add_list(*nvp, prop, strval, 0, src);
386 kmem_free(strval, za.za_num_integers);
393 zap_cursor_fini(&zc);
394 mutex_exit(&spa->spa_props_lock);
396 if (err && err != ENOENT) {
406 * Validate the given pool properties nvlist and modify the list
407 * for the property values to be set.
410 spa_prop_validate(spa_t *spa, nvlist_t *props)
413 int error = 0, reset_bootfs = 0;
415 boolean_t has_feature = B_FALSE;
418 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
420 char *strval, *slash, *check, *fname;
421 const char *propname = nvpair_name(elem);
422 zpool_prop_t prop = zpool_name_to_prop(propname);
426 if (!zpool_prop_feature(propname)) {
427 error = SET_ERROR(EINVAL);
432 * Sanitize the input.
434 if (nvpair_type(elem) != DATA_TYPE_UINT64) {
435 error = SET_ERROR(EINVAL);
439 if (nvpair_value_uint64(elem, &intval) != 0) {
440 error = SET_ERROR(EINVAL);
445 error = SET_ERROR(EINVAL);
449 fname = strchr(propname, '@') + 1;
450 if (zfeature_lookup_name(fname, NULL) != 0) {
451 error = SET_ERROR(EINVAL);
455 has_feature = B_TRUE;
458 case ZPOOL_PROP_VERSION:
459 error = nvpair_value_uint64(elem, &intval);
461 (intval < spa_version(spa) ||
462 intval > SPA_VERSION_BEFORE_FEATURES ||
464 error = SET_ERROR(EINVAL);
467 case ZPOOL_PROP_DELEGATION:
468 case ZPOOL_PROP_AUTOREPLACE:
469 case ZPOOL_PROP_LISTSNAPS:
470 case ZPOOL_PROP_AUTOEXPAND:
471 error = nvpair_value_uint64(elem, &intval);
472 if (!error && intval > 1)
473 error = SET_ERROR(EINVAL);
476 case ZPOOL_PROP_BOOTFS:
478 * If the pool version is less than SPA_VERSION_BOOTFS,
479 * or the pool is still being created (version == 0),
480 * the bootfs property cannot be set.
482 if (spa_version(spa) < SPA_VERSION_BOOTFS) {
483 error = SET_ERROR(ENOTSUP);
488 * Make sure the vdev config is bootable
490 if (!vdev_is_bootable(spa->spa_root_vdev)) {
491 error = SET_ERROR(ENOTSUP);
497 error = nvpair_value_string(elem, &strval);
503 if (strval == NULL || strval[0] == '\0') {
504 objnum = zpool_prop_default_numeric(
509 if (error = dmu_objset_hold(strval, FTAG, &os))
512 /* Must be ZPL and not gzip compressed. */
514 if (dmu_objset_type(os) != DMU_OST_ZFS) {
515 error = SET_ERROR(ENOTSUP);
517 dsl_prop_get_int_ds(dmu_objset_ds(os),
518 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
520 !BOOTFS_COMPRESS_VALID(compress)) {
521 error = SET_ERROR(ENOTSUP);
523 objnum = dmu_objset_id(os);
525 dmu_objset_rele(os, FTAG);
529 case ZPOOL_PROP_FAILUREMODE:
530 error = nvpair_value_uint64(elem, &intval);
531 if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
532 intval > ZIO_FAILURE_MODE_PANIC))
533 error = SET_ERROR(EINVAL);
536 * This is a special case which only occurs when
537 * the pool has completely failed. This allows
538 * the user to change the in-core failmode property
539 * without syncing it out to disk (I/Os might
540 * currently be blocked). We do this by returning
541 * EIO to the caller (spa_prop_set) to trick it
542 * into thinking we encountered a property validation
545 if (!error && spa_suspended(spa)) {
546 spa->spa_failmode = intval;
547 error = SET_ERROR(EIO);
551 case ZPOOL_PROP_CACHEFILE:
552 if ((error = nvpair_value_string(elem, &strval)) != 0)
555 if (strval[0] == '\0')
558 if (strcmp(strval, "none") == 0)
561 if (strval[0] != '/') {
562 error = SET_ERROR(EINVAL);
566 slash = strrchr(strval, '/');
567 ASSERT(slash != NULL);
569 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
570 strcmp(slash, "/..") == 0)
571 error = SET_ERROR(EINVAL);
574 case ZPOOL_PROP_COMMENT:
575 if ((error = nvpair_value_string(elem, &strval)) != 0)
577 for (check = strval; *check != '\0'; check++) {
579 * The kernel doesn't have an easy isprint()
580 * check. For this kernel check, we merely
581 * check ASCII apart from DEL. Fix this if
582 * there is an easy-to-use kernel isprint().
584 if (*check >= 0x7f) {
585 error = SET_ERROR(EINVAL);
590 if (strlen(strval) > ZPROP_MAX_COMMENT)
594 case ZPOOL_PROP_DEDUPDITTO:
595 if (spa_version(spa) < SPA_VERSION_DEDUP)
596 error = SET_ERROR(ENOTSUP);
598 error = nvpair_value_uint64(elem, &intval);
600 intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
601 error = SET_ERROR(EINVAL);
609 if (!error && reset_bootfs) {
610 error = nvlist_remove(props,
611 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
614 error = nvlist_add_uint64(props,
615 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
623 spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
626 spa_config_dirent_t *dp;
628 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
632 dp = kmem_alloc(sizeof (spa_config_dirent_t),
635 if (cachefile[0] == '\0')
636 dp->scd_path = spa_strdup(spa_config_path);
637 else if (strcmp(cachefile, "none") == 0)
640 dp->scd_path = spa_strdup(cachefile);
642 list_insert_head(&spa->spa_config_list, dp);
644 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
648 spa_prop_set(spa_t *spa, nvlist_t *nvp)
651 nvpair_t *elem = NULL;
652 boolean_t need_sync = B_FALSE;
654 if ((error = spa_prop_validate(spa, nvp)) != 0)
657 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
658 zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem));
660 if (prop == ZPOOL_PROP_CACHEFILE ||
661 prop == ZPOOL_PROP_ALTROOT ||
662 prop == ZPOOL_PROP_READONLY)
665 if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) {
668 if (prop == ZPOOL_PROP_VERSION) {
669 VERIFY(nvpair_value_uint64(elem, &ver) == 0);
671 ASSERT(zpool_prop_feature(nvpair_name(elem)));
672 ver = SPA_VERSION_FEATURES;
676 /* Save time if the version is already set. */
677 if (ver == spa_version(spa))
681 * In addition to the pool directory object, we might
682 * create the pool properties object, the features for
683 * read object, the features for write object, or the
684 * feature descriptions object.
686 error = dsl_sync_task(spa->spa_name, NULL,
687 spa_sync_version, &ver,
688 6, ZFS_SPACE_CHECK_RESERVED);
699 return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props,
700 nvp, 6, ZFS_SPACE_CHECK_RESERVED));
707 * If the bootfs property value is dsobj, clear it.
710 spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
712 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
713 VERIFY(zap_remove(spa->spa_meta_objset,
714 spa->spa_pool_props_object,
715 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
722 spa_change_guid_check(void *arg, dmu_tx_t *tx)
724 uint64_t *newguid = arg;
725 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
726 vdev_t *rvd = spa->spa_root_vdev;
729 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
730 vdev_state = rvd->vdev_state;
731 spa_config_exit(spa, SCL_STATE, FTAG);
733 if (vdev_state != VDEV_STATE_HEALTHY)
734 return (SET_ERROR(ENXIO));
736 ASSERT3U(spa_guid(spa), !=, *newguid);
742 spa_change_guid_sync(void *arg, dmu_tx_t *tx)
744 uint64_t *newguid = arg;
745 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
747 vdev_t *rvd = spa->spa_root_vdev;
749 oldguid = spa_guid(spa);
751 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
752 rvd->vdev_guid = *newguid;
753 rvd->vdev_guid_sum += (*newguid - oldguid);
754 vdev_config_dirty(rvd);
755 spa_config_exit(spa, SCL_STATE, FTAG);
757 spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu",
762 * Change the GUID for the pool. This is done so that we can later
763 * re-import a pool built from a clone of our own vdevs. We will modify
764 * the root vdev's guid, our own pool guid, and then mark all of our
765 * vdevs dirty. Note that we must make sure that all our vdevs are
766 * online when we do this, or else any vdevs that weren't present
767 * would be orphaned from our pool. We are also going to issue a
768 * sysevent to update any watchers.
771 spa_change_guid(spa_t *spa)
776 mutex_enter(&spa->spa_vdev_top_lock);
777 mutex_enter(&spa_namespace_lock);
778 guid = spa_generate_guid(NULL);
780 error = dsl_sync_task(spa->spa_name, spa_change_guid_check,
781 spa_change_guid_sync, &guid, 5, ZFS_SPACE_CHECK_RESERVED);
784 spa_config_sync(spa, B_FALSE, B_TRUE);
785 spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID);
788 mutex_exit(&spa_namespace_lock);
789 mutex_exit(&spa->spa_vdev_top_lock);
795 * ==========================================================================
796 * SPA state manipulation (open/create/destroy/import/export)
797 * ==========================================================================
801 spa_error_entry_compare(const void *a, const void *b)
803 spa_error_entry_t *sa = (spa_error_entry_t *)a;
804 spa_error_entry_t *sb = (spa_error_entry_t *)b;
807 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
808 sizeof (zbookmark_phys_t));
819 * Utility function which retrieves copies of the current logs and
820 * re-initializes them in the process.
823 spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
825 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
827 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
828 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
830 avl_create(&spa->spa_errlist_scrub,
831 spa_error_entry_compare, sizeof (spa_error_entry_t),
832 offsetof(spa_error_entry_t, se_avl));
833 avl_create(&spa->spa_errlist_last,
834 spa_error_entry_compare, sizeof (spa_error_entry_t),
835 offsetof(spa_error_entry_t, se_avl));
839 spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
841 const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
842 enum zti_modes mode = ztip->zti_mode;
843 uint_t value = ztip->zti_value;
844 uint_t count = ztip->zti_count;
845 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
848 boolean_t batch = B_FALSE;
850 if (mode == ZTI_MODE_NULL) {
852 tqs->stqs_taskq = NULL;
856 ASSERT3U(count, >, 0);
858 tqs->stqs_count = count;
859 tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP);
863 ASSERT3U(value, >=, 1);
864 value = MAX(value, 1);
869 flags |= TASKQ_THREADS_CPU_PCT;
870 value = zio_taskq_batch_pct;
874 panic("unrecognized mode for %s_%s taskq (%u:%u) in "
876 zio_type_name[t], zio_taskq_types[q], mode, value);
880 for (uint_t i = 0; i < count; i++) {
884 (void) snprintf(name, sizeof (name), "%s_%s_%u",
885 zio_type_name[t], zio_taskq_types[q], i);
887 (void) snprintf(name, sizeof (name), "%s_%s",
888 zio_type_name[t], zio_taskq_types[q]);
892 if (zio_taskq_sysdc && spa->spa_proc != &p0) {
894 flags |= TASKQ_DC_BATCH;
896 tq = taskq_create_sysdc(name, value, 50, INT_MAX,
897 spa->spa_proc, zio_taskq_basedc, flags);
900 pri_t pri = maxclsyspri;
902 * The write issue taskq can be extremely CPU
903 * intensive. Run it at slightly lower priority
904 * than the other taskqs.
906 if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE)
909 tq = taskq_create_proc(name, value, pri, 50,
910 INT_MAX, spa->spa_proc, flags);
915 tqs->stqs_taskq[i] = tq;
920 spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
922 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
924 if (tqs->stqs_taskq == NULL) {
925 ASSERT0(tqs->stqs_count);
929 for (uint_t i = 0; i < tqs->stqs_count; i++) {
930 ASSERT3P(tqs->stqs_taskq[i], !=, NULL);
931 taskq_destroy(tqs->stqs_taskq[i]);
934 kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *));
935 tqs->stqs_taskq = NULL;
939 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
940 * Note that a type may have multiple discrete taskqs to avoid lock contention
941 * on the taskq itself. In that case we choose which taskq at random by using
942 * the low bits of gethrtime().
945 spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q,
946 task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent)
948 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
951 ASSERT3P(tqs->stqs_taskq, !=, NULL);
952 ASSERT3U(tqs->stqs_count, !=, 0);
954 if (tqs->stqs_count == 1) {
955 tq = tqs->stqs_taskq[0];
958 tq = tqs->stqs_taskq[cpu_ticks() % tqs->stqs_count];
960 tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count];
964 taskq_dispatch_ent(tq, func, arg, flags, ent);
968 spa_create_zio_taskqs(spa_t *spa)
970 for (int t = 0; t < ZIO_TYPES; t++) {
971 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
972 spa_taskqs_init(spa, t, q);
980 spa_thread(void *arg)
985 user_t *pu = PTOU(curproc);
987 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
990 ASSERT(curproc != &p0);
991 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
992 "zpool-%s", spa->spa_name);
993 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
996 /* bind this thread to the requested psrset */
997 if (zio_taskq_psrset_bind != PS_NONE) {
999 mutex_enter(&cpu_lock);
1000 mutex_enter(&pidlock);
1001 mutex_enter(&curproc->p_lock);
1003 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
1004 0, NULL, NULL) == 0) {
1005 curthread->t_bind_pset = zio_taskq_psrset_bind;
1008 "Couldn't bind process for zfs pool \"%s\" to "
1009 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
1012 mutex_exit(&curproc->p_lock);
1013 mutex_exit(&pidlock);
1014 mutex_exit(&cpu_lock);
1020 if (zio_taskq_sysdc) {
1021 sysdc_thread_enter(curthread, 100, 0);
1025 spa->spa_proc = curproc;
1026 spa->spa_did = curthread->t_did;
1028 spa_create_zio_taskqs(spa);
1030 mutex_enter(&spa->spa_proc_lock);
1031 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
1033 spa->spa_proc_state = SPA_PROC_ACTIVE;
1034 cv_broadcast(&spa->spa_proc_cv);
1036 CALLB_CPR_SAFE_BEGIN(&cprinfo);
1037 while (spa->spa_proc_state == SPA_PROC_ACTIVE)
1038 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1039 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
1041 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
1042 spa->spa_proc_state = SPA_PROC_GONE;
1043 spa->spa_proc = &p0;
1044 cv_broadcast(&spa->spa_proc_cv);
1045 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */
1047 mutex_enter(&curproc->p_lock);
1050 #endif /* SPA_PROCESS */
1054 * Activate an uninitialized pool.
1057 spa_activate(spa_t *spa, int mode)
1059 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
1061 spa->spa_state = POOL_STATE_ACTIVE;
1062 spa->spa_mode = mode;
1064 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
1065 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
1067 /* Try to create a covering process */
1068 mutex_enter(&spa->spa_proc_lock);
1069 ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
1070 ASSERT(spa->spa_proc == &p0);
1074 /* Only create a process if we're going to be around a while. */
1075 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
1076 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
1078 spa->spa_proc_state = SPA_PROC_CREATED;
1079 while (spa->spa_proc_state == SPA_PROC_CREATED) {
1080 cv_wait(&spa->spa_proc_cv,
1081 &spa->spa_proc_lock);
1083 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1084 ASSERT(spa->spa_proc != &p0);
1085 ASSERT(spa->spa_did != 0);
1089 "Couldn't create process for zfs pool \"%s\"\n",
1094 #endif /* SPA_PROCESS */
1095 mutex_exit(&spa->spa_proc_lock);
1097 /* If we didn't create a process, we need to create our taskqs. */
1098 ASSERT(spa->spa_proc == &p0);
1099 if (spa->spa_proc == &p0) {
1100 spa_create_zio_taskqs(spa);
1104 * Start TRIM thread.
1106 trim_thread_create(spa);
1108 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
1109 offsetof(vdev_t, vdev_config_dirty_node));
1110 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
1111 offsetof(vdev_t, vdev_state_dirty_node));
1113 txg_list_create(&spa->spa_vdev_txg_list,
1114 offsetof(struct vdev, vdev_txg_node));
1116 avl_create(&spa->spa_errlist_scrub,
1117 spa_error_entry_compare, sizeof (spa_error_entry_t),
1118 offsetof(spa_error_entry_t, se_avl));
1119 avl_create(&spa->spa_errlist_last,
1120 spa_error_entry_compare, sizeof (spa_error_entry_t),
1121 offsetof(spa_error_entry_t, se_avl));
1125 * Opposite of spa_activate().
1128 spa_deactivate(spa_t *spa)
1130 ASSERT(spa->spa_sync_on == B_FALSE);
1131 ASSERT(spa->spa_dsl_pool == NULL);
1132 ASSERT(spa->spa_root_vdev == NULL);
1133 ASSERT(spa->spa_async_zio_root == NULL);
1134 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
1137 * Stop TRIM thread in case spa_unload() wasn't called directly
1138 * before spa_deactivate().
1140 trim_thread_destroy(spa);
1142 txg_list_destroy(&spa->spa_vdev_txg_list);
1144 list_destroy(&spa->spa_config_dirty_list);
1145 list_destroy(&spa->spa_state_dirty_list);
1147 for (int t = 0; t < ZIO_TYPES; t++) {
1148 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1149 spa_taskqs_fini(spa, t, q);
1153 metaslab_class_destroy(spa->spa_normal_class);
1154 spa->spa_normal_class = NULL;
1156 metaslab_class_destroy(spa->spa_log_class);
1157 spa->spa_log_class = NULL;
1160 * If this was part of an import or the open otherwise failed, we may
1161 * still have errors left in the queues. Empty them just in case.
1163 spa_errlog_drain(spa);
1165 avl_destroy(&spa->spa_errlist_scrub);
1166 avl_destroy(&spa->spa_errlist_last);
1168 spa->spa_state = POOL_STATE_UNINITIALIZED;
1170 mutex_enter(&spa->spa_proc_lock);
1171 if (spa->spa_proc_state != SPA_PROC_NONE) {
1172 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1173 spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1174 cv_broadcast(&spa->spa_proc_cv);
1175 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1176 ASSERT(spa->spa_proc != &p0);
1177 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1179 ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1180 spa->spa_proc_state = SPA_PROC_NONE;
1182 ASSERT(spa->spa_proc == &p0);
1183 mutex_exit(&spa->spa_proc_lock);
1187 * We want to make sure spa_thread() has actually exited the ZFS
1188 * module, so that the module can't be unloaded out from underneath
1191 if (spa->spa_did != 0) {
1192 thread_join(spa->spa_did);
1195 #endif /* SPA_PROCESS */
1199 * Verify a pool configuration, and construct the vdev tree appropriately. This
1200 * will create all the necessary vdevs in the appropriate layout, with each vdev
1201 * in the CLOSED state. This will prep the pool before open/creation/import.
1202 * All vdev validation is done by the vdev_alloc() routine.
1205 spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1206 uint_t id, int atype)
1212 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1215 if ((*vdp)->vdev_ops->vdev_op_leaf)
1218 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1221 if (error == ENOENT)
1227 return (SET_ERROR(EINVAL));
1230 for (int c = 0; c < children; c++) {
1232 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1240 ASSERT(*vdp != NULL);
1246 * Opposite of spa_load().
1249 spa_unload(spa_t *spa)
1253 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1258 trim_thread_destroy(spa);
1263 spa_async_suspend(spa);
1268 if (spa->spa_sync_on) {
1269 txg_sync_stop(spa->spa_dsl_pool);
1270 spa->spa_sync_on = B_FALSE;
1274 * Wait for any outstanding async I/O to complete.
1276 if (spa->spa_async_zio_root != NULL) {
1277 (void) zio_wait(spa->spa_async_zio_root);
1278 spa->spa_async_zio_root = NULL;
1281 bpobj_close(&spa->spa_deferred_bpobj);
1283 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1288 if (spa->spa_root_vdev)
1289 vdev_free(spa->spa_root_vdev);
1290 ASSERT(spa->spa_root_vdev == NULL);
1293 * Close the dsl pool.
1295 if (spa->spa_dsl_pool) {
1296 dsl_pool_close(spa->spa_dsl_pool);
1297 spa->spa_dsl_pool = NULL;
1298 spa->spa_meta_objset = NULL;
1305 * Drop and purge level 2 cache
1307 spa_l2cache_drop(spa);
1309 for (i = 0; i < spa->spa_spares.sav_count; i++)
1310 vdev_free(spa->spa_spares.sav_vdevs[i]);
1311 if (spa->spa_spares.sav_vdevs) {
1312 kmem_free(spa->spa_spares.sav_vdevs,
1313 spa->spa_spares.sav_count * sizeof (void *));
1314 spa->spa_spares.sav_vdevs = NULL;
1316 if (spa->spa_spares.sav_config) {
1317 nvlist_free(spa->spa_spares.sav_config);
1318 spa->spa_spares.sav_config = NULL;
1320 spa->spa_spares.sav_count = 0;
1322 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1323 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1324 vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1326 if (spa->spa_l2cache.sav_vdevs) {
1327 kmem_free(spa->spa_l2cache.sav_vdevs,
1328 spa->spa_l2cache.sav_count * sizeof (void *));
1329 spa->spa_l2cache.sav_vdevs = NULL;
1331 if (spa->spa_l2cache.sav_config) {
1332 nvlist_free(spa->spa_l2cache.sav_config);
1333 spa->spa_l2cache.sav_config = NULL;
1335 spa->spa_l2cache.sav_count = 0;
1337 spa->spa_async_suspended = 0;
1339 if (spa->spa_comment != NULL) {
1340 spa_strfree(spa->spa_comment);
1341 spa->spa_comment = NULL;
1344 spa_config_exit(spa, SCL_ALL, FTAG);
1348 * Load (or re-load) the current list of vdevs describing the active spares for
1349 * this pool. When this is called, we have some form of basic information in
1350 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1351 * then re-generate a more complete list including status information.
1354 spa_load_spares(spa_t *spa)
1361 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1364 * First, close and free any existing spare vdevs.
1366 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1367 vd = spa->spa_spares.sav_vdevs[i];
1369 /* Undo the call to spa_activate() below */
1370 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1371 B_FALSE)) != NULL && tvd->vdev_isspare)
1372 spa_spare_remove(tvd);
1377 if (spa->spa_spares.sav_vdevs)
1378 kmem_free(spa->spa_spares.sav_vdevs,
1379 spa->spa_spares.sav_count * sizeof (void *));
1381 if (spa->spa_spares.sav_config == NULL)
1384 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1385 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1387 spa->spa_spares.sav_count = (int)nspares;
1388 spa->spa_spares.sav_vdevs = NULL;
1394 * Construct the array of vdevs, opening them to get status in the
1395 * process. For each spare, there is potentially two different vdev_t
1396 * structures associated with it: one in the list of spares (used only
1397 * for basic validation purposes) and one in the active vdev
1398 * configuration (if it's spared in). During this phase we open and
1399 * validate each vdev on the spare list. If the vdev also exists in the
1400 * active configuration, then we also mark this vdev as an active spare.
1402 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1404 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1405 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1406 VDEV_ALLOC_SPARE) == 0);
1409 spa->spa_spares.sav_vdevs[i] = vd;
1411 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1412 B_FALSE)) != NULL) {
1413 if (!tvd->vdev_isspare)
1417 * We only mark the spare active if we were successfully
1418 * able to load the vdev. Otherwise, importing a pool
1419 * with a bad active spare would result in strange
1420 * behavior, because multiple pool would think the spare
1421 * is actively in use.
1423 * There is a vulnerability here to an equally bizarre
1424 * circumstance, where a dead active spare is later
1425 * brought back to life (onlined or otherwise). Given
1426 * the rarity of this scenario, and the extra complexity
1427 * it adds, we ignore the possibility.
1429 if (!vdev_is_dead(tvd))
1430 spa_spare_activate(tvd);
1434 vd->vdev_aux = &spa->spa_spares;
1436 if (vdev_open(vd) != 0)
1439 if (vdev_validate_aux(vd) == 0)
1444 * Recompute the stashed list of spares, with status information
1447 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1448 DATA_TYPE_NVLIST_ARRAY) == 0);
1450 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1452 for (i = 0; i < spa->spa_spares.sav_count; i++)
1453 spares[i] = vdev_config_generate(spa,
1454 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1455 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1456 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1457 for (i = 0; i < spa->spa_spares.sav_count; i++)
1458 nvlist_free(spares[i]);
1459 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1463 * Load (or re-load) the current list of vdevs describing the active l2cache for
1464 * this pool. When this is called, we have some form of basic information in
1465 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1466 * then re-generate a more complete list including status information.
1467 * Devices which are already active have their details maintained, and are
1471 spa_load_l2cache(spa_t *spa)
1475 int i, j, oldnvdevs;
1477 vdev_t *vd, **oldvdevs, **newvdevs;
1478 spa_aux_vdev_t *sav = &spa->spa_l2cache;
1480 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1482 if (sav->sav_config != NULL) {
1483 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1484 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1485 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1491 oldvdevs = sav->sav_vdevs;
1492 oldnvdevs = sav->sav_count;
1493 sav->sav_vdevs = NULL;
1497 * Process new nvlist of vdevs.
1499 for (i = 0; i < nl2cache; i++) {
1500 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1504 for (j = 0; j < oldnvdevs; j++) {
1506 if (vd != NULL && guid == vd->vdev_guid) {
1508 * Retain previous vdev for add/remove ops.
1516 if (newvdevs[i] == NULL) {
1520 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1521 VDEV_ALLOC_L2CACHE) == 0);
1526 * Commit this vdev as an l2cache device,
1527 * even if it fails to open.
1529 spa_l2cache_add(vd);
1534 spa_l2cache_activate(vd);
1536 if (vdev_open(vd) != 0)
1539 (void) vdev_validate_aux(vd);
1541 if (!vdev_is_dead(vd))
1542 l2arc_add_vdev(spa, vd);
1547 * Purge vdevs that were dropped
1549 for (i = 0; i < oldnvdevs; i++) {
1554 ASSERT(vd->vdev_isl2cache);
1556 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1557 pool != 0ULL && l2arc_vdev_present(vd))
1558 l2arc_remove_vdev(vd);
1559 vdev_clear_stats(vd);
1565 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1567 if (sav->sav_config == NULL)
1570 sav->sav_vdevs = newvdevs;
1571 sav->sav_count = (int)nl2cache;
1574 * Recompute the stashed list of l2cache devices, with status
1575 * information this time.
1577 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1578 DATA_TYPE_NVLIST_ARRAY) == 0);
1580 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1581 for (i = 0; i < sav->sav_count; i++)
1582 l2cache[i] = vdev_config_generate(spa,
1583 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1584 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1585 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1587 for (i = 0; i < sav->sav_count; i++)
1588 nvlist_free(l2cache[i]);
1590 kmem_free(l2cache, sav->sav_count * sizeof (void *));
1594 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1597 char *packed = NULL;
1602 error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db);
1605 nvsize = *(uint64_t *)db->db_data;
1606 dmu_buf_rele(db, FTAG);
1608 packed = kmem_alloc(nvsize, KM_SLEEP);
1609 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1612 error = nvlist_unpack(packed, nvsize, value, 0);
1613 kmem_free(packed, nvsize);
1619 * Checks to see if the given vdev could not be opened, in which case we post a
1620 * sysevent to notify the autoreplace code that the device has been removed.
1623 spa_check_removed(vdev_t *vd)
1625 for (int c = 0; c < vd->vdev_children; c++)
1626 spa_check_removed(vd->vdev_child[c]);
1628 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) &&
1630 zfs_post_autoreplace(vd->vdev_spa, vd);
1631 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1636 * Validate the current config against the MOS config
1639 spa_config_valid(spa_t *spa, nvlist_t *config)
1641 vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1644 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1646 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1647 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1649 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1652 * If we're doing a normal import, then build up any additional
1653 * diagnostic information about missing devices in this config.
1654 * We'll pass this up to the user for further processing.
1656 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1657 nvlist_t **child, *nv;
1660 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1662 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1664 for (int c = 0; c < rvd->vdev_children; c++) {
1665 vdev_t *tvd = rvd->vdev_child[c];
1666 vdev_t *mtvd = mrvd->vdev_child[c];
1668 if (tvd->vdev_ops == &vdev_missing_ops &&
1669 mtvd->vdev_ops != &vdev_missing_ops &&
1671 child[idx++] = vdev_config_generate(spa, mtvd,
1676 VERIFY(nvlist_add_nvlist_array(nv,
1677 ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1678 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1679 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1681 for (int i = 0; i < idx; i++)
1682 nvlist_free(child[i]);
1685 kmem_free(child, rvd->vdev_children * sizeof (char **));
1689 * Compare the root vdev tree with the information we have
1690 * from the MOS config (mrvd). Check each top-level vdev
1691 * with the corresponding MOS config top-level (mtvd).
1693 for (int c = 0; c < rvd->vdev_children; c++) {
1694 vdev_t *tvd = rvd->vdev_child[c];
1695 vdev_t *mtvd = mrvd->vdev_child[c];
1698 * Resolve any "missing" vdevs in the current configuration.
1699 * If we find that the MOS config has more accurate information
1700 * about the top-level vdev then use that vdev instead.
1702 if (tvd->vdev_ops == &vdev_missing_ops &&
1703 mtvd->vdev_ops != &vdev_missing_ops) {
1705 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1709 * Device specific actions.
1711 if (mtvd->vdev_islog) {
1712 spa_set_log_state(spa, SPA_LOG_CLEAR);
1715 * XXX - once we have 'readonly' pool
1716 * support we should be able to handle
1717 * missing data devices by transitioning
1718 * the pool to readonly.
1724 * Swap the missing vdev with the data we were
1725 * able to obtain from the MOS config.
1727 vdev_remove_child(rvd, tvd);
1728 vdev_remove_child(mrvd, mtvd);
1730 vdev_add_child(rvd, mtvd);
1731 vdev_add_child(mrvd, tvd);
1733 spa_config_exit(spa, SCL_ALL, FTAG);
1735 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1738 } else if (mtvd->vdev_islog) {
1740 * Load the slog device's state from the MOS config
1741 * since it's possible that the label does not
1742 * contain the most up-to-date information.
1744 vdev_load_log_state(tvd, mtvd);
1749 spa_config_exit(spa, SCL_ALL, FTAG);
1752 * Ensure we were able to validate the config.
1754 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1758 * Check for missing log devices
1761 spa_check_logs(spa_t *spa)
1763 boolean_t rv = B_FALSE;
1765 switch (spa->spa_log_state) {
1766 case SPA_LOG_MISSING:
1767 /* need to recheck in case slog has been restored */
1768 case SPA_LOG_UNKNOWN:
1769 rv = (dmu_objset_find(spa->spa_name, zil_check_log_chain,
1770 NULL, DS_FIND_CHILDREN) != 0);
1772 spa_set_log_state(spa, SPA_LOG_MISSING);
1779 spa_passivate_log(spa_t *spa)
1781 vdev_t *rvd = spa->spa_root_vdev;
1782 boolean_t slog_found = B_FALSE;
1784 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1786 if (!spa_has_slogs(spa))
1789 for (int c = 0; c < rvd->vdev_children; c++) {
1790 vdev_t *tvd = rvd->vdev_child[c];
1791 metaslab_group_t *mg = tvd->vdev_mg;
1793 if (tvd->vdev_islog) {
1794 metaslab_group_passivate(mg);
1795 slog_found = B_TRUE;
1799 return (slog_found);
1803 spa_activate_log(spa_t *spa)
1805 vdev_t *rvd = spa->spa_root_vdev;
1807 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1809 for (int c = 0; c < rvd->vdev_children; c++) {
1810 vdev_t *tvd = rvd->vdev_child[c];
1811 metaslab_group_t *mg = tvd->vdev_mg;
1813 if (tvd->vdev_islog)
1814 metaslab_group_activate(mg);
1819 spa_offline_log(spa_t *spa)
1823 error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1824 NULL, DS_FIND_CHILDREN);
1827 * We successfully offlined the log device, sync out the
1828 * current txg so that the "stubby" block can be removed
1831 txg_wait_synced(spa->spa_dsl_pool, 0);
1837 spa_aux_check_removed(spa_aux_vdev_t *sav)
1841 for (i = 0; i < sav->sav_count; i++)
1842 spa_check_removed(sav->sav_vdevs[i]);
1846 spa_claim_notify(zio_t *zio)
1848 spa_t *spa = zio->io_spa;
1853 mutex_enter(&spa->spa_props_lock); /* any mutex will do */
1854 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1855 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1856 mutex_exit(&spa->spa_props_lock);
1859 typedef struct spa_load_error {
1860 uint64_t sle_meta_count;
1861 uint64_t sle_data_count;
1865 spa_load_verify_done(zio_t *zio)
1867 blkptr_t *bp = zio->io_bp;
1868 spa_load_error_t *sle = zio->io_private;
1869 dmu_object_type_t type = BP_GET_TYPE(bp);
1870 int error = zio->io_error;
1871 spa_t *spa = zio->io_spa;
1874 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1875 type != DMU_OT_INTENT_LOG)
1876 atomic_inc_64(&sle->sle_meta_count);
1878 atomic_inc_64(&sle->sle_data_count);
1880 zio_data_buf_free(zio->io_data, zio->io_size);
1882 mutex_enter(&spa->spa_scrub_lock);
1883 spa->spa_scrub_inflight--;
1884 cv_broadcast(&spa->spa_scrub_io_cv);
1885 mutex_exit(&spa->spa_scrub_lock);
1889 * Maximum number of concurrent scrub i/os to create while verifying
1890 * a pool while importing it.
1892 int spa_load_verify_maxinflight = 10000;
1893 boolean_t spa_load_verify_metadata = B_TRUE;
1894 boolean_t spa_load_verify_data = B_TRUE;
1896 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_maxinflight, CTLFLAG_RWTUN,
1897 &spa_load_verify_maxinflight, 0,
1898 "Maximum number of concurrent scrub I/Os to create while verifying a "
1899 "pool while importing it");
1901 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_metadata, CTLFLAG_RWTUN,
1902 &spa_load_verify_metadata, 0,
1903 "Check metadata on import?");
1905 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_data, CTLFLAG_RWTUN,
1906 &spa_load_verify_data, 0,
1907 "Check user data on import?");
1911 spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1912 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
1914 if (BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp))
1917 * Note: normally this routine will not be called if
1918 * spa_load_verify_metadata is not set. However, it may be useful
1919 * to manually set the flag after the traversal has begun.
1921 if (!spa_load_verify_metadata)
1923 if (BP_GET_BUFC_TYPE(bp) == ARC_BUFC_DATA && !spa_load_verify_data)
1927 size_t size = BP_GET_PSIZE(bp);
1928 void *data = zio_data_buf_alloc(size);
1930 mutex_enter(&spa->spa_scrub_lock);
1931 while (spa->spa_scrub_inflight >= spa_load_verify_maxinflight)
1932 cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
1933 spa->spa_scrub_inflight++;
1934 mutex_exit(&spa->spa_scrub_lock);
1936 zio_nowait(zio_read(rio, spa, bp, data, size,
1937 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1938 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1939 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1944 spa_load_verify(spa_t *spa)
1947 spa_load_error_t sle = { 0 };
1948 zpool_rewind_policy_t policy;
1949 boolean_t verify_ok = B_FALSE;
1952 zpool_get_rewind_policy(spa->spa_config, &policy);
1954 if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1957 rio = zio_root(spa, NULL, &sle,
1958 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1960 if (spa_load_verify_metadata) {
1961 error = traverse_pool(spa, spa->spa_verify_min_txg,
1962 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA,
1963 spa_load_verify_cb, rio);
1966 (void) zio_wait(rio);
1968 spa->spa_load_meta_errors = sle.sle_meta_count;
1969 spa->spa_load_data_errors = sle.sle_data_count;
1971 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1972 sle.sle_data_count <= policy.zrp_maxdata) {
1976 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1977 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1979 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1980 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1981 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1982 VERIFY(nvlist_add_int64(spa->spa_load_info,
1983 ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1984 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1985 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1987 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1991 if (error != ENXIO && error != EIO)
1992 error = SET_ERROR(EIO);
1996 return (verify_ok ? 0 : EIO);
2000 * Find a value in the pool props object.
2003 spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
2005 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
2006 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
2010 * Find a value in the pool directory object.
2013 spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
2015 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
2016 name, sizeof (uint64_t), 1, val));
2020 spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
2022 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
2027 * Fix up config after a partly-completed split. This is done with the
2028 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
2029 * pool have that entry in their config, but only the splitting one contains
2030 * a list of all the guids of the vdevs that are being split off.
2032 * This function determines what to do with that list: either rejoin
2033 * all the disks to the pool, or complete the splitting process. To attempt
2034 * the rejoin, each disk that is offlined is marked online again, and
2035 * we do a reopen() call. If the vdev label for every disk that was
2036 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
2037 * then we call vdev_split() on each disk, and complete the split.
2039 * Otherwise we leave the config alone, with all the vdevs in place in
2040 * the original pool.
2043 spa_try_repair(spa_t *spa, nvlist_t *config)
2050 boolean_t attempt_reopen;
2052 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
2055 /* check that the config is complete */
2056 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
2057 &glist, &gcount) != 0)
2060 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
2062 /* attempt to online all the vdevs & validate */
2063 attempt_reopen = B_TRUE;
2064 for (i = 0; i < gcount; i++) {
2065 if (glist[i] == 0) /* vdev is hole */
2068 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
2069 if (vd[i] == NULL) {
2071 * Don't bother attempting to reopen the disks;
2072 * just do the split.
2074 attempt_reopen = B_FALSE;
2076 /* attempt to re-online it */
2077 vd[i]->vdev_offline = B_FALSE;
2081 if (attempt_reopen) {
2082 vdev_reopen(spa->spa_root_vdev);
2084 /* check each device to see what state it's in */
2085 for (extracted = 0, i = 0; i < gcount; i++) {
2086 if (vd[i] != NULL &&
2087 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
2094 * If every disk has been moved to the new pool, or if we never
2095 * even attempted to look at them, then we split them off for
2098 if (!attempt_reopen || gcount == extracted) {
2099 for (i = 0; i < gcount; i++)
2102 vdev_reopen(spa->spa_root_vdev);
2105 kmem_free(vd, gcount * sizeof (vdev_t *));
2109 spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
2110 boolean_t mosconfig)
2112 nvlist_t *config = spa->spa_config;
2113 char *ereport = FM_EREPORT_ZFS_POOL;
2119 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
2120 return (SET_ERROR(EINVAL));
2122 ASSERT(spa->spa_comment == NULL);
2123 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
2124 spa->spa_comment = spa_strdup(comment);
2127 * Versioning wasn't explicitly added to the label until later, so if
2128 * it's not present treat it as the initial version.
2130 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
2131 &spa->spa_ubsync.ub_version) != 0)
2132 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
2134 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
2135 &spa->spa_config_txg);
2137 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
2138 spa_guid_exists(pool_guid, 0)) {
2139 error = SET_ERROR(EEXIST);
2141 spa->spa_config_guid = pool_guid;
2143 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
2145 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
2149 nvlist_free(spa->spa_load_info);
2150 spa->spa_load_info = fnvlist_alloc();
2152 gethrestime(&spa->spa_loaded_ts);
2153 error = spa_load_impl(spa, pool_guid, config, state, type,
2154 mosconfig, &ereport);
2157 spa->spa_minref = refcount_count(&spa->spa_refcount);
2159 if (error != EEXIST) {
2160 spa->spa_loaded_ts.tv_sec = 0;
2161 spa->spa_loaded_ts.tv_nsec = 0;
2163 if (error != EBADF) {
2164 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
2167 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2174 * Load an existing storage pool, using the pool's builtin spa_config as a
2175 * source of configuration information.
2178 spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
2179 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
2183 nvlist_t *nvroot = NULL;
2186 uberblock_t *ub = &spa->spa_uberblock;
2187 uint64_t children, config_cache_txg = spa->spa_config_txg;
2188 int orig_mode = spa->spa_mode;
2191 boolean_t missing_feat_write = B_FALSE;
2194 * If this is an untrusted config, access the pool in read-only mode.
2195 * This prevents things like resilvering recently removed devices.
2198 spa->spa_mode = FREAD;
2200 ASSERT(MUTEX_HELD(&spa_namespace_lock));
2202 spa->spa_load_state = state;
2204 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
2205 return (SET_ERROR(EINVAL));
2207 parse = (type == SPA_IMPORT_EXISTING ?
2208 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2211 * Create "The Godfather" zio to hold all async IOs
2213 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2214 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2217 * Parse the configuration into a vdev tree. We explicitly set the
2218 * value that will be returned by spa_version() since parsing the
2219 * configuration requires knowing the version number.
2221 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2222 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
2223 spa_config_exit(spa, SCL_ALL, FTAG);
2228 ASSERT(spa->spa_root_vdev == rvd);
2230 if (type != SPA_IMPORT_ASSEMBLE) {
2231 ASSERT(spa_guid(spa) == pool_guid);
2235 * Try to open all vdevs, loading each label in the process.
2237 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2238 error = vdev_open(rvd);
2239 spa_config_exit(spa, SCL_ALL, FTAG);
2244 * We need to validate the vdev labels against the configuration that
2245 * we have in hand, which is dependent on the setting of mosconfig. If
2246 * mosconfig is true then we're validating the vdev labels based on
2247 * that config. Otherwise, we're validating against the cached config
2248 * (zpool.cache) that was read when we loaded the zfs module, and then
2249 * later we will recursively call spa_load() and validate against
2252 * If we're assembling a new pool that's been split off from an
2253 * existing pool, the labels haven't yet been updated so we skip
2254 * validation for now.
2256 if (type != SPA_IMPORT_ASSEMBLE) {
2257 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2258 error = vdev_validate(rvd, mosconfig);
2259 spa_config_exit(spa, SCL_ALL, FTAG);
2264 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2265 return (SET_ERROR(ENXIO));
2269 * Find the best uberblock.
2271 vdev_uberblock_load(rvd, ub, &label);
2274 * If we weren't able to find a single valid uberblock, return failure.
2276 if (ub->ub_txg == 0) {
2278 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2282 * If the pool has an unsupported version we can't open it.
2284 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2286 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2289 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2293 * If we weren't able to find what's necessary for reading the
2294 * MOS in the label, return failure.
2296 if (label == NULL || nvlist_lookup_nvlist(label,
2297 ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) {
2299 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2304 * Update our in-core representation with the definitive values
2307 nvlist_free(spa->spa_label_features);
2308 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2314 * Look through entries in the label nvlist's features_for_read. If
2315 * there is a feature listed there which we don't understand then we
2316 * cannot open a pool.
2318 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2319 nvlist_t *unsup_feat;
2321 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2324 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2326 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2327 if (!zfeature_is_supported(nvpair_name(nvp))) {
2328 VERIFY(nvlist_add_string(unsup_feat,
2329 nvpair_name(nvp), "") == 0);
2333 if (!nvlist_empty(unsup_feat)) {
2334 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2335 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2336 nvlist_free(unsup_feat);
2337 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2341 nvlist_free(unsup_feat);
2345 * If the vdev guid sum doesn't match the uberblock, we have an
2346 * incomplete configuration. We first check to see if the pool
2347 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2348 * If it is, defer the vdev_guid_sum check till later so we
2349 * can handle missing vdevs.
2351 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
2352 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
2353 rvd->vdev_guid_sum != ub->ub_guid_sum)
2354 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2356 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2357 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2358 spa_try_repair(spa, config);
2359 spa_config_exit(spa, SCL_ALL, FTAG);
2360 nvlist_free(spa->spa_config_splitting);
2361 spa->spa_config_splitting = NULL;
2365 * Initialize internal SPA structures.
2367 spa->spa_state = POOL_STATE_ACTIVE;
2368 spa->spa_ubsync = spa->spa_uberblock;
2369 spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2370 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2371 spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2372 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2373 spa->spa_claim_max_txg = spa->spa_first_txg;
2374 spa->spa_prev_software_version = ub->ub_software_version;
2376 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2378 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2379 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2381 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2382 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2384 if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2385 boolean_t missing_feat_read = B_FALSE;
2386 nvlist_t *unsup_feat, *enabled_feat;
2388 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2389 &spa->spa_feat_for_read_obj) != 0) {
2390 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2393 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2394 &spa->spa_feat_for_write_obj) != 0) {
2395 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2398 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2399 &spa->spa_feat_desc_obj) != 0) {
2400 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2403 enabled_feat = fnvlist_alloc();
2404 unsup_feat = fnvlist_alloc();
2406 if (!spa_features_check(spa, B_FALSE,
2407 unsup_feat, enabled_feat))
2408 missing_feat_read = B_TRUE;
2410 if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) {
2411 if (!spa_features_check(spa, B_TRUE,
2412 unsup_feat, enabled_feat)) {
2413 missing_feat_write = B_TRUE;
2417 fnvlist_add_nvlist(spa->spa_load_info,
2418 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
2420 if (!nvlist_empty(unsup_feat)) {
2421 fnvlist_add_nvlist(spa->spa_load_info,
2422 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
2425 fnvlist_free(enabled_feat);
2426 fnvlist_free(unsup_feat);
2428 if (!missing_feat_read) {
2429 fnvlist_add_boolean(spa->spa_load_info,
2430 ZPOOL_CONFIG_CAN_RDONLY);
2434 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2435 * twofold: to determine whether the pool is available for
2436 * import in read-write mode and (if it is not) whether the
2437 * pool is available for import in read-only mode. If the pool
2438 * is available for import in read-write mode, it is displayed
2439 * as available in userland; if it is not available for import
2440 * in read-only mode, it is displayed as unavailable in
2441 * userland. If the pool is available for import in read-only
2442 * mode but not read-write mode, it is displayed as unavailable
2443 * in userland with a special note that the pool is actually
2444 * available for open in read-only mode.
2446 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2447 * missing a feature for write, we must first determine whether
2448 * the pool can be opened read-only before returning to
2449 * userland in order to know whether to display the
2450 * abovementioned note.
2452 if (missing_feat_read || (missing_feat_write &&
2453 spa_writeable(spa))) {
2454 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2459 * Load refcounts for ZFS features from disk into an in-memory
2460 * cache during SPA initialization.
2462 for (spa_feature_t i = 0; i < SPA_FEATURES; i++) {
2465 error = feature_get_refcount_from_disk(spa,
2466 &spa_feature_table[i], &refcount);
2468 spa->spa_feat_refcount_cache[i] = refcount;
2469 } else if (error == ENOTSUP) {
2470 spa->spa_feat_refcount_cache[i] =
2471 SPA_FEATURE_DISABLED;
2473 return (spa_vdev_err(rvd,
2474 VDEV_AUX_CORRUPT_DATA, EIO));
2479 if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) {
2480 if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG,
2481 &spa->spa_feat_enabled_txg_obj) != 0)
2482 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2485 spa->spa_is_initializing = B_TRUE;
2486 error = dsl_pool_open(spa->spa_dsl_pool);
2487 spa->spa_is_initializing = B_FALSE;
2489 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2493 nvlist_t *policy = NULL, *nvconfig;
2495 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2496 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2498 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2499 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2501 unsigned long myhostid = 0;
2503 VERIFY(nvlist_lookup_string(nvconfig,
2504 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2507 myhostid = zone_get_hostid(NULL);
2510 * We're emulating the system's hostid in userland, so
2511 * we can't use zone_get_hostid().
2513 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2514 #endif /* _KERNEL */
2515 if (check_hostid && hostid != 0 && myhostid != 0 &&
2516 hostid != myhostid) {
2517 nvlist_free(nvconfig);
2518 cmn_err(CE_WARN, "pool '%s' could not be "
2519 "loaded as it was last accessed by "
2520 "another system (host: %s hostid: 0x%lx). "
2521 "See: http://illumos.org/msg/ZFS-8000-EY",
2522 spa_name(spa), hostname,
2523 (unsigned long)hostid);
2524 return (SET_ERROR(EBADF));
2527 if (nvlist_lookup_nvlist(spa->spa_config,
2528 ZPOOL_REWIND_POLICY, &policy) == 0)
2529 VERIFY(nvlist_add_nvlist(nvconfig,
2530 ZPOOL_REWIND_POLICY, policy) == 0);
2532 spa_config_set(spa, nvconfig);
2534 spa_deactivate(spa);
2535 spa_activate(spa, orig_mode);
2537 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2540 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2541 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2542 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2544 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2547 * Load the bit that tells us to use the new accounting function
2548 * (raid-z deflation). If we have an older pool, this will not
2551 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2552 if (error != 0 && error != ENOENT)
2553 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2555 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2556 &spa->spa_creation_version);
2557 if (error != 0 && error != ENOENT)
2558 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2561 * Load the persistent error log. If we have an older pool, this will
2564 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2565 if (error != 0 && error != ENOENT)
2566 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2568 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2569 &spa->spa_errlog_scrub);
2570 if (error != 0 && error != ENOENT)
2571 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2574 * Load the history object. If we have an older pool, this
2575 * will not be present.
2577 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2578 if (error != 0 && error != ENOENT)
2579 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2582 * If we're assembling the pool from the split-off vdevs of
2583 * an existing pool, we don't want to attach the spares & cache
2588 * Load any hot spares for this pool.
2590 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2591 if (error != 0 && error != ENOENT)
2592 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2593 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2594 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2595 if (load_nvlist(spa, spa->spa_spares.sav_object,
2596 &spa->spa_spares.sav_config) != 0)
2597 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2599 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2600 spa_load_spares(spa);
2601 spa_config_exit(spa, SCL_ALL, FTAG);
2602 } else if (error == 0) {
2603 spa->spa_spares.sav_sync = B_TRUE;
2607 * Load any level 2 ARC devices for this pool.
2609 error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2610 &spa->spa_l2cache.sav_object);
2611 if (error != 0 && error != ENOENT)
2612 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2613 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2614 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2615 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2616 &spa->spa_l2cache.sav_config) != 0)
2617 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2619 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2620 spa_load_l2cache(spa);
2621 spa_config_exit(spa, SCL_ALL, FTAG);
2622 } else if (error == 0) {
2623 spa->spa_l2cache.sav_sync = B_TRUE;
2626 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2628 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2629 if (error && error != ENOENT)
2630 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2633 uint64_t autoreplace;
2635 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2636 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2637 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2638 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2639 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2640 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2641 &spa->spa_dedup_ditto);
2643 spa->spa_autoreplace = (autoreplace != 0);
2647 * If the 'autoreplace' property is set, then post a resource notifying
2648 * the ZFS DE that it should not issue any faults for unopenable
2649 * devices. We also iterate over the vdevs, and post a sysevent for any
2650 * unopenable vdevs so that the normal autoreplace handler can take
2653 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2654 spa_check_removed(spa->spa_root_vdev);
2656 * For the import case, this is done in spa_import(), because
2657 * at this point we're using the spare definitions from
2658 * the MOS config, not necessarily from the userland config.
2660 if (state != SPA_LOAD_IMPORT) {
2661 spa_aux_check_removed(&spa->spa_spares);
2662 spa_aux_check_removed(&spa->spa_l2cache);
2667 * Load the vdev state for all toplevel vdevs.
2672 * Propagate the leaf DTLs we just loaded all the way up the tree.
2674 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2675 vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2676 spa_config_exit(spa, SCL_ALL, FTAG);
2679 * Load the DDTs (dedup tables).
2681 error = ddt_load(spa);
2683 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2685 spa_update_dspace(spa);
2688 * Validate the config, using the MOS config to fill in any
2689 * information which might be missing. If we fail to validate
2690 * the config then declare the pool unfit for use. If we're
2691 * assembling a pool from a split, the log is not transferred
2694 if (type != SPA_IMPORT_ASSEMBLE) {
2697 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2698 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2700 if (!spa_config_valid(spa, nvconfig)) {
2701 nvlist_free(nvconfig);
2702 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2705 nvlist_free(nvconfig);
2708 * Now that we've validated the config, check the state of the
2709 * root vdev. If it can't be opened, it indicates one or
2710 * more toplevel vdevs are faulted.
2712 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2713 return (SET_ERROR(ENXIO));
2715 if (spa_check_logs(spa)) {
2716 *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2717 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2721 if (missing_feat_write) {
2722 ASSERT(state == SPA_LOAD_TRYIMPORT);
2725 * At this point, we know that we can open the pool in
2726 * read-only mode but not read-write mode. We now have enough
2727 * information and can return to userland.
2729 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
2733 * We've successfully opened the pool, verify that we're ready
2734 * to start pushing transactions.
2736 if (state != SPA_LOAD_TRYIMPORT) {
2737 if (error = spa_load_verify(spa))
2738 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2742 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2743 spa->spa_load_max_txg == UINT64_MAX)) {
2745 int need_update = B_FALSE;
2747 ASSERT(state != SPA_LOAD_TRYIMPORT);
2750 * Claim log blocks that haven't been committed yet.
2751 * This must all happen in a single txg.
2752 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2753 * invoked from zil_claim_log_block()'s i/o done callback.
2754 * Price of rollback is that we abandon the log.
2756 spa->spa_claiming = B_TRUE;
2758 tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2759 spa_first_txg(spa));
2760 (void) dmu_objset_find(spa_name(spa),
2761 zil_claim, tx, DS_FIND_CHILDREN);
2764 spa->spa_claiming = B_FALSE;
2766 spa_set_log_state(spa, SPA_LOG_GOOD);
2767 spa->spa_sync_on = B_TRUE;
2768 txg_sync_start(spa->spa_dsl_pool);
2771 * Wait for all claims to sync. We sync up to the highest
2772 * claimed log block birth time so that claimed log blocks
2773 * don't appear to be from the future. spa_claim_max_txg
2774 * will have been set for us by either zil_check_log_chain()
2775 * (invoked from spa_check_logs()) or zil_claim() above.
2777 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2780 * If the config cache is stale, or we have uninitialized
2781 * metaslabs (see spa_vdev_add()), then update the config.
2783 * If this is a verbatim import, trust the current
2784 * in-core spa_config and update the disk labels.
2786 if (config_cache_txg != spa->spa_config_txg ||
2787 state == SPA_LOAD_IMPORT ||
2788 state == SPA_LOAD_RECOVER ||
2789 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2790 need_update = B_TRUE;
2792 for (int c = 0; c < rvd->vdev_children; c++)
2793 if (rvd->vdev_child[c]->vdev_ms_array == 0)
2794 need_update = B_TRUE;
2797 * Update the config cache asychronously in case we're the
2798 * root pool, in which case the config cache isn't writable yet.
2801 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2804 * Check all DTLs to see if anything needs resilvering.
2806 if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2807 vdev_resilver_needed(rvd, NULL, NULL))
2808 spa_async_request(spa, SPA_ASYNC_RESILVER);
2811 * Log the fact that we booted up (so that we can detect if
2812 * we rebooted in the middle of an operation).
2814 spa_history_log_version(spa, "open");
2817 * Delete any inconsistent datasets.
2819 (void) dmu_objset_find(spa_name(spa),
2820 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2823 * Clean up any stale temporary dataset userrefs.
2825 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2832 spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2834 int mode = spa->spa_mode;
2837 spa_deactivate(spa);
2839 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg - 1;
2841 spa_activate(spa, mode);
2842 spa_async_suspend(spa);
2844 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2848 * If spa_load() fails this function will try loading prior txg's. If
2849 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2850 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2851 * function will not rewind the pool and will return the same error as
2855 spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2856 uint64_t max_request, int rewind_flags)
2858 nvlist_t *loadinfo = NULL;
2859 nvlist_t *config = NULL;
2860 int load_error, rewind_error;
2861 uint64_t safe_rewind_txg;
2864 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2865 spa->spa_load_max_txg = spa->spa_load_txg;
2866 spa_set_log_state(spa, SPA_LOG_CLEAR);
2868 spa->spa_load_max_txg = max_request;
2869 if (max_request != UINT64_MAX)
2870 spa->spa_extreme_rewind = B_TRUE;
2873 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2875 if (load_error == 0)
2878 if (spa->spa_root_vdev != NULL)
2879 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2881 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2882 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2884 if (rewind_flags & ZPOOL_NEVER_REWIND) {
2885 nvlist_free(config);
2886 return (load_error);
2889 if (state == SPA_LOAD_RECOVER) {
2890 /* Price of rolling back is discarding txgs, including log */
2891 spa_set_log_state(spa, SPA_LOG_CLEAR);
2894 * If we aren't rolling back save the load info from our first
2895 * import attempt so that we can restore it after attempting
2898 loadinfo = spa->spa_load_info;
2899 spa->spa_load_info = fnvlist_alloc();
2902 spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2903 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2904 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2905 TXG_INITIAL : safe_rewind_txg;
2908 * Continue as long as we're finding errors, we're still within
2909 * the acceptable rewind range, and we're still finding uberblocks
2911 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2912 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2913 if (spa->spa_load_max_txg < safe_rewind_txg)
2914 spa->spa_extreme_rewind = B_TRUE;
2915 rewind_error = spa_load_retry(spa, state, mosconfig);
2918 spa->spa_extreme_rewind = B_FALSE;
2919 spa->spa_load_max_txg = UINT64_MAX;
2921 if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2922 spa_config_set(spa, config);
2924 if (state == SPA_LOAD_RECOVER) {
2925 ASSERT3P(loadinfo, ==, NULL);
2926 return (rewind_error);
2928 /* Store the rewind info as part of the initial load info */
2929 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
2930 spa->spa_load_info);
2932 /* Restore the initial load info */
2933 fnvlist_free(spa->spa_load_info);
2934 spa->spa_load_info = loadinfo;
2936 return (load_error);
2943 * The import case is identical to an open except that the configuration is sent
2944 * down from userland, instead of grabbed from the configuration cache. For the
2945 * case of an open, the pool configuration will exist in the
2946 * POOL_STATE_UNINITIALIZED state.
2948 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2949 * the same time open the pool, without having to keep around the spa_t in some
2953 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2957 spa_load_state_t state = SPA_LOAD_OPEN;
2959 int locked = B_FALSE;
2960 int firstopen = B_FALSE;
2965 * As disgusting as this is, we need to support recursive calls to this
2966 * function because dsl_dir_open() is called during spa_load(), and ends
2967 * up calling spa_open() again. The real fix is to figure out how to
2968 * avoid dsl_dir_open() calling this in the first place.
2970 if (mutex_owner(&spa_namespace_lock) != curthread) {
2971 mutex_enter(&spa_namespace_lock);
2975 if ((spa = spa_lookup(pool)) == NULL) {
2977 mutex_exit(&spa_namespace_lock);
2978 return (SET_ERROR(ENOENT));
2981 if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2982 zpool_rewind_policy_t policy;
2986 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2988 if (policy.zrp_request & ZPOOL_DO_REWIND)
2989 state = SPA_LOAD_RECOVER;
2991 spa_activate(spa, spa_mode_global);
2993 if (state != SPA_LOAD_RECOVER)
2994 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2996 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2997 policy.zrp_request);
2999 if (error == EBADF) {
3001 * If vdev_validate() returns failure (indicated by
3002 * EBADF), it indicates that one of the vdevs indicates
3003 * that the pool has been exported or destroyed. If
3004 * this is the case, the config cache is out of sync and
3005 * we should remove the pool from the namespace.
3008 spa_deactivate(spa);
3009 spa_config_sync(spa, B_TRUE, B_TRUE);
3012 mutex_exit(&spa_namespace_lock);
3013 return (SET_ERROR(ENOENT));
3018 * We can't open the pool, but we still have useful
3019 * information: the state of each vdev after the
3020 * attempted vdev_open(). Return this to the user.
3022 if (config != NULL && spa->spa_config) {
3023 VERIFY(nvlist_dup(spa->spa_config, config,
3025 VERIFY(nvlist_add_nvlist(*config,
3026 ZPOOL_CONFIG_LOAD_INFO,
3027 spa->spa_load_info) == 0);
3030 spa_deactivate(spa);
3031 spa->spa_last_open_failed = error;
3033 mutex_exit(&spa_namespace_lock);
3039 spa_open_ref(spa, tag);
3042 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
3045 * If we've recovered the pool, pass back any information we
3046 * gathered while doing the load.
3048 if (state == SPA_LOAD_RECOVER) {
3049 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
3050 spa->spa_load_info) == 0);
3054 spa->spa_last_open_failed = 0;
3055 spa->spa_last_ubsync_txg = 0;
3056 spa->spa_load_txg = 0;
3057 mutex_exit(&spa_namespace_lock);
3061 zvol_create_minors(spa->spa_name);
3072 spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
3075 return (spa_open_common(name, spapp, tag, policy, config));
3079 spa_open(const char *name, spa_t **spapp, void *tag)
3081 return (spa_open_common(name, spapp, tag, NULL, NULL));
3085 * Lookup the given spa_t, incrementing the inject count in the process,
3086 * preventing it from being exported or destroyed.
3089 spa_inject_addref(char *name)
3093 mutex_enter(&spa_namespace_lock);
3094 if ((spa = spa_lookup(name)) == NULL) {
3095 mutex_exit(&spa_namespace_lock);
3098 spa->spa_inject_ref++;
3099 mutex_exit(&spa_namespace_lock);
3105 spa_inject_delref(spa_t *spa)
3107 mutex_enter(&spa_namespace_lock);
3108 spa->spa_inject_ref--;
3109 mutex_exit(&spa_namespace_lock);
3113 * Add spares device information to the nvlist.
3116 spa_add_spares(spa_t *spa, nvlist_t *config)
3126 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3128 if (spa->spa_spares.sav_count == 0)
3131 VERIFY(nvlist_lookup_nvlist(config,
3132 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3133 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
3134 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3136 VERIFY(nvlist_add_nvlist_array(nvroot,
3137 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3138 VERIFY(nvlist_lookup_nvlist_array(nvroot,
3139 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3142 * Go through and find any spares which have since been
3143 * repurposed as an active spare. If this is the case, update
3144 * their status appropriately.
3146 for (i = 0; i < nspares; i++) {
3147 VERIFY(nvlist_lookup_uint64(spares[i],
3148 ZPOOL_CONFIG_GUID, &guid) == 0);
3149 if (spa_spare_exists(guid, &pool, NULL) &&
3151 VERIFY(nvlist_lookup_uint64_array(
3152 spares[i], ZPOOL_CONFIG_VDEV_STATS,
3153 (uint64_t **)&vs, &vsc) == 0);
3154 vs->vs_state = VDEV_STATE_CANT_OPEN;
3155 vs->vs_aux = VDEV_AUX_SPARED;
3162 * Add l2cache device information to the nvlist, including vdev stats.
3165 spa_add_l2cache(spa_t *spa, nvlist_t *config)
3168 uint_t i, j, nl2cache;
3175 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3177 if (spa->spa_l2cache.sav_count == 0)
3180 VERIFY(nvlist_lookup_nvlist(config,
3181 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3182 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
3183 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3184 if (nl2cache != 0) {
3185 VERIFY(nvlist_add_nvlist_array(nvroot,
3186 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3187 VERIFY(nvlist_lookup_nvlist_array(nvroot,
3188 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3191 * Update level 2 cache device stats.
3194 for (i = 0; i < nl2cache; i++) {
3195 VERIFY(nvlist_lookup_uint64(l2cache[i],
3196 ZPOOL_CONFIG_GUID, &guid) == 0);
3199 for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
3201 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
3202 vd = spa->spa_l2cache.sav_vdevs[j];
3208 VERIFY(nvlist_lookup_uint64_array(l2cache[i],
3209 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
3211 vdev_get_stats(vd, vs);
3217 spa_add_feature_stats(spa_t *spa, nvlist_t *config)
3223 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3224 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3226 /* We may be unable to read features if pool is suspended. */
3227 if (spa_suspended(spa))
3230 if (spa->spa_feat_for_read_obj != 0) {
3231 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3232 spa->spa_feat_for_read_obj);
3233 zap_cursor_retrieve(&zc, &za) == 0;
3234 zap_cursor_advance(&zc)) {
3235 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3236 za.za_num_integers == 1);
3237 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3238 za.za_first_integer));
3240 zap_cursor_fini(&zc);
3243 if (spa->spa_feat_for_write_obj != 0) {
3244 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3245 spa->spa_feat_for_write_obj);
3246 zap_cursor_retrieve(&zc, &za) == 0;
3247 zap_cursor_advance(&zc)) {
3248 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3249 za.za_num_integers == 1);
3250 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3251 za.za_first_integer));
3253 zap_cursor_fini(&zc);
3257 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
3259 nvlist_free(features);
3263 spa_get_stats(const char *name, nvlist_t **config,
3264 char *altroot, size_t buflen)
3270 error = spa_open_common(name, &spa, FTAG, NULL, config);
3274 * This still leaves a window of inconsistency where the spares
3275 * or l2cache devices could change and the config would be
3276 * self-inconsistent.
3278 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3280 if (*config != NULL) {
3281 uint64_t loadtimes[2];
3283 loadtimes[0] = spa->spa_loaded_ts.tv_sec;
3284 loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
3285 VERIFY(nvlist_add_uint64_array(*config,
3286 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
3288 VERIFY(nvlist_add_uint64(*config,
3289 ZPOOL_CONFIG_ERRCOUNT,
3290 spa_get_errlog_size(spa)) == 0);
3292 if (spa_suspended(spa))
3293 VERIFY(nvlist_add_uint64(*config,
3294 ZPOOL_CONFIG_SUSPENDED,
3295 spa->spa_failmode) == 0);
3297 spa_add_spares(spa, *config);
3298 spa_add_l2cache(spa, *config);
3299 spa_add_feature_stats(spa, *config);
3304 * We want to get the alternate root even for faulted pools, so we cheat
3305 * and call spa_lookup() directly.
3309 mutex_enter(&spa_namespace_lock);
3310 spa = spa_lookup(name);
3312 spa_altroot(spa, altroot, buflen);
3316 mutex_exit(&spa_namespace_lock);
3318 spa_altroot(spa, altroot, buflen);
3323 spa_config_exit(spa, SCL_CONFIG, FTAG);
3324 spa_close(spa, FTAG);
3331 * Validate that the auxiliary device array is well formed. We must have an
3332 * array of nvlists, each which describes a valid leaf vdev. If this is an
3333 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3334 * specified, as long as they are well-formed.
3337 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
3338 spa_aux_vdev_t *sav, const char *config, uint64_t version,
3339 vdev_labeltype_t label)
3346 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3349 * It's acceptable to have no devs specified.
3351 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
3355 return (SET_ERROR(EINVAL));
3358 * Make sure the pool is formatted with a version that supports this
3361 if (spa_version(spa) < version)
3362 return (SET_ERROR(ENOTSUP));
3365 * Set the pending device list so we correctly handle device in-use
3368 sav->sav_pending = dev;
3369 sav->sav_npending = ndev;
3371 for (i = 0; i < ndev; i++) {
3372 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
3376 if (!vd->vdev_ops->vdev_op_leaf) {
3378 error = SET_ERROR(EINVAL);
3383 * The L2ARC currently only supports disk devices in
3384 * kernel context. For user-level testing, we allow it.
3387 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
3388 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
3389 error = SET_ERROR(ENOTBLK);
3396 if ((error = vdev_open(vd)) == 0 &&
3397 (error = vdev_label_init(vd, crtxg, label)) == 0) {
3398 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
3399 vd->vdev_guid) == 0);
3405 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
3412 sav->sav_pending = NULL;
3413 sav->sav_npending = 0;
3418 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
3422 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3424 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3425 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
3426 VDEV_LABEL_SPARE)) != 0) {
3430 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3431 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
3432 VDEV_LABEL_L2CACHE));
3436 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
3441 if (sav->sav_config != NULL) {
3447 * Generate new dev list by concatentating with the
3450 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
3451 &olddevs, &oldndevs) == 0);
3453 newdevs = kmem_alloc(sizeof (void *) *
3454 (ndevs + oldndevs), KM_SLEEP);
3455 for (i = 0; i < oldndevs; i++)
3456 VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
3458 for (i = 0; i < ndevs; i++)
3459 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
3462 VERIFY(nvlist_remove(sav->sav_config, config,
3463 DATA_TYPE_NVLIST_ARRAY) == 0);
3465 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3466 config, newdevs, ndevs + oldndevs) == 0);
3467 for (i = 0; i < oldndevs + ndevs; i++)
3468 nvlist_free(newdevs[i]);
3469 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
3472 * Generate a new dev list.
3474 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
3476 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
3482 * Stop and drop level 2 ARC devices
3485 spa_l2cache_drop(spa_t *spa)
3489 spa_aux_vdev_t *sav = &spa->spa_l2cache;
3491 for (i = 0; i < sav->sav_count; i++) {
3494 vd = sav->sav_vdevs[i];
3497 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
3498 pool != 0ULL && l2arc_vdev_present(vd))
3499 l2arc_remove_vdev(vd);
3507 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
3511 char *altroot = NULL;
3516 uint64_t txg = TXG_INITIAL;
3517 nvlist_t **spares, **l2cache;
3518 uint_t nspares, nl2cache;
3519 uint64_t version, obj;
3520 boolean_t has_features;
3523 * If this pool already exists, return failure.
3525 mutex_enter(&spa_namespace_lock);
3526 if (spa_lookup(pool) != NULL) {
3527 mutex_exit(&spa_namespace_lock);
3528 return (SET_ERROR(EEXIST));
3532 * Allocate a new spa_t structure.
3534 (void) nvlist_lookup_string(props,
3535 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3536 spa = spa_add(pool, NULL, altroot);
3537 spa_activate(spa, spa_mode_global);
3539 if (props && (error = spa_prop_validate(spa, props))) {
3540 spa_deactivate(spa);
3542 mutex_exit(&spa_namespace_lock);
3546 has_features = B_FALSE;
3547 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
3548 elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
3549 if (zpool_prop_feature(nvpair_name(elem)))
3550 has_features = B_TRUE;
3553 if (has_features || nvlist_lookup_uint64(props,
3554 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
3555 version = SPA_VERSION;
3557 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
3559 spa->spa_first_txg = txg;
3560 spa->spa_uberblock.ub_txg = txg - 1;
3561 spa->spa_uberblock.ub_version = version;
3562 spa->spa_ubsync = spa->spa_uberblock;
3565 * Create "The Godfather" zio to hold all async IOs
3567 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
3568 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
3571 * Create the root vdev.
3573 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3575 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3577 ASSERT(error != 0 || rvd != NULL);
3578 ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3580 if (error == 0 && !zfs_allocatable_devs(nvroot))
3581 error = SET_ERROR(EINVAL);
3584 (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3585 (error = spa_validate_aux(spa, nvroot, txg,
3586 VDEV_ALLOC_ADD)) == 0) {
3587 for (int c = 0; c < rvd->vdev_children; c++) {
3588 vdev_ashift_optimize(rvd->vdev_child[c]);
3589 vdev_metaslab_set_size(rvd->vdev_child[c]);
3590 vdev_expand(rvd->vdev_child[c], txg);
3594 spa_config_exit(spa, SCL_ALL, FTAG);
3598 spa_deactivate(spa);
3600 mutex_exit(&spa_namespace_lock);
3605 * Get the list of spares, if specified.
3607 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3608 &spares, &nspares) == 0) {
3609 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3611 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3612 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3613 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3614 spa_load_spares(spa);
3615 spa_config_exit(spa, SCL_ALL, FTAG);
3616 spa->spa_spares.sav_sync = B_TRUE;
3620 * Get the list of level 2 cache devices, if specified.
3622 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3623 &l2cache, &nl2cache) == 0) {
3624 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3625 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3626 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3627 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3628 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3629 spa_load_l2cache(spa);
3630 spa_config_exit(spa, SCL_ALL, FTAG);
3631 spa->spa_l2cache.sav_sync = B_TRUE;
3634 spa->spa_is_initializing = B_TRUE;
3635 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3636 spa->spa_meta_objset = dp->dp_meta_objset;
3637 spa->spa_is_initializing = B_FALSE;
3640 * Create DDTs (dedup tables).
3644 spa_update_dspace(spa);
3646 tx = dmu_tx_create_assigned(dp, txg);
3649 * Create the pool config object.
3651 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3652 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3653 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3655 if (zap_add(spa->spa_meta_objset,
3656 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3657 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3658 cmn_err(CE_PANIC, "failed to add pool config");
3661 if (spa_version(spa) >= SPA_VERSION_FEATURES)
3662 spa_feature_create_zap_objects(spa, tx);
3664 if (zap_add(spa->spa_meta_objset,
3665 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3666 sizeof (uint64_t), 1, &version, tx) != 0) {
3667 cmn_err(CE_PANIC, "failed to add pool version");
3670 /* Newly created pools with the right version are always deflated. */
3671 if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3672 spa->spa_deflate = TRUE;
3673 if (zap_add(spa->spa_meta_objset,
3674 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3675 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3676 cmn_err(CE_PANIC, "failed to add deflate");
3681 * Create the deferred-free bpobj. Turn off compression
3682 * because sync-to-convergence takes longer if the blocksize
3685 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3686 dmu_object_set_compress(spa->spa_meta_objset, obj,
3687 ZIO_COMPRESS_OFF, tx);
3688 if (zap_add(spa->spa_meta_objset,
3689 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3690 sizeof (uint64_t), 1, &obj, tx) != 0) {
3691 cmn_err(CE_PANIC, "failed to add bpobj");
3693 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3694 spa->spa_meta_objset, obj));
3697 * Create the pool's history object.
3699 if (version >= SPA_VERSION_ZPOOL_HISTORY)
3700 spa_history_create_obj(spa, tx);
3703 * Set pool properties.
3705 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3706 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3707 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3708 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3710 if (props != NULL) {
3711 spa_configfile_set(spa, props, B_FALSE);
3712 spa_sync_props(props, tx);
3717 spa->spa_sync_on = B_TRUE;
3718 txg_sync_start(spa->spa_dsl_pool);
3721 * We explicitly wait for the first transaction to complete so that our
3722 * bean counters are appropriately updated.
3724 txg_wait_synced(spa->spa_dsl_pool, txg);
3726 spa_config_sync(spa, B_FALSE, B_TRUE);
3728 spa_history_log_version(spa, "create");
3730 spa->spa_minref = refcount_count(&spa->spa_refcount);
3732 mutex_exit(&spa_namespace_lock);
3740 * Get the root pool information from the root disk, then import the root pool
3741 * during the system boot up time.
3743 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3746 spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3749 nvlist_t *nvtop, *nvroot;
3752 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3756 * Add this top-level vdev to the child array.
3758 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3760 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3762 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3765 * Put this pool's top-level vdevs into a root vdev.
3767 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3768 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3769 VDEV_TYPE_ROOT) == 0);
3770 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3771 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3772 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3776 * Replace the existing vdev_tree with the new root vdev in
3777 * this pool's configuration (remove the old, add the new).
3779 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3780 nvlist_free(nvroot);
3785 * Walk the vdev tree and see if we can find a device with "better"
3786 * configuration. A configuration is "better" if the label on that
3787 * device has a more recent txg.
3790 spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3792 for (int c = 0; c < vd->vdev_children; c++)
3793 spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3795 if (vd->vdev_ops->vdev_op_leaf) {
3799 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3803 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3807 * Do we have a better boot device?
3809 if (label_txg > *txg) {
3818 * Import a root pool.
3820 * For x86. devpath_list will consist of devid and/or physpath name of
3821 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3822 * The GRUB "findroot" command will return the vdev we should boot.
3824 * For Sparc, devpath_list consists the physpath name of the booting device
3825 * no matter the rootpool is a single device pool or a mirrored pool.
3827 * "/pci@1f,0/ide@d/disk@0,0:a"
3830 spa_import_rootpool(char *devpath, char *devid)
3833 vdev_t *rvd, *bvd, *avd = NULL;
3834 nvlist_t *config, *nvtop;
3840 * Read the label from the boot device and generate a configuration.
3842 config = spa_generate_rootconf(devpath, devid, &guid);
3843 #if defined(_OBP) && defined(_KERNEL)
3844 if (config == NULL) {
3845 if (strstr(devpath, "/iscsi/ssd") != NULL) {
3847 get_iscsi_bootpath_phy(devpath);
3848 config = spa_generate_rootconf(devpath, devid, &guid);
3852 if (config == NULL) {
3853 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
3855 return (SET_ERROR(EIO));
3858 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3860 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3862 mutex_enter(&spa_namespace_lock);
3863 if ((spa = spa_lookup(pname)) != NULL) {
3865 * Remove the existing root pool from the namespace so that we
3866 * can replace it with the correct config we just read in.
3871 spa = spa_add(pname, config, NULL);
3872 spa->spa_is_root = B_TRUE;
3873 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3876 * Build up a vdev tree based on the boot device's label config.
3878 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3880 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3881 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3882 VDEV_ALLOC_ROOTPOOL);
3883 spa_config_exit(spa, SCL_ALL, FTAG);
3885 mutex_exit(&spa_namespace_lock);
3886 nvlist_free(config);
3887 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3893 * Get the boot vdev.
3895 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3896 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3897 (u_longlong_t)guid);
3898 error = SET_ERROR(ENOENT);
3903 * Determine if there is a better boot device.
3906 spa_alt_rootvdev(rvd, &avd, &txg);
3908 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3909 "try booting from '%s'", avd->vdev_path);
3910 error = SET_ERROR(EINVAL);
3915 * If the boot device is part of a spare vdev then ensure that
3916 * we're booting off the active spare.
3918 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3919 !bvd->vdev_isspare) {
3920 cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3921 "try booting from '%s'",
3923 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3924 error = SET_ERROR(EINVAL);
3930 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3932 spa_config_exit(spa, SCL_ALL, FTAG);
3933 mutex_exit(&spa_namespace_lock);
3935 nvlist_free(config);
3941 extern int vdev_geom_read_pool_label(const char *name, nvlist_t ***configs,
3945 spa_generate_rootconf(const char *name)
3947 nvlist_t **configs, **tops;
3949 nvlist_t *best_cfg, *nvtop, *nvroot;
3958 if (vdev_geom_read_pool_label(name, &configs, &count) != 0)
3961 ASSERT3U(count, !=, 0);
3963 for (i = 0; i < count; i++) {
3966 VERIFY(nvlist_lookup_uint64(configs[i], ZPOOL_CONFIG_POOL_TXG,
3968 if (txg > best_txg) {
3970 best_cfg = configs[i];
3975 * Multi-vdev root pool configuration discovery is not supported yet.
3978 nvlist_lookup_uint64(best_cfg, ZPOOL_CONFIG_VDEV_CHILDREN, &nchildren);
3980 nvlist_lookup_uint64_array(best_cfg, ZPOOL_CONFIG_HOLE_ARRAY,
3983 tops = kmem_zalloc(nchildren * sizeof(void *), KM_SLEEP);
3984 for (i = 0; i < nchildren; i++) {
3987 if (configs[i] == NULL)
3989 VERIFY(nvlist_lookup_nvlist(configs[i], ZPOOL_CONFIG_VDEV_TREE,
3991 nvlist_dup(nvtop, &tops[i], KM_SLEEP);
3993 for (i = 0; holes != NULL && i < nholes; i++) {
3996 if (tops[holes[i]] != NULL)
3998 nvlist_alloc(&tops[holes[i]], NV_UNIQUE_NAME, KM_SLEEP);
3999 VERIFY(nvlist_add_string(tops[holes[i]], ZPOOL_CONFIG_TYPE,
4000 VDEV_TYPE_HOLE) == 0);
4001 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_ID,
4003 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_GUID,
4006 for (i = 0; i < nchildren; i++) {
4007 if (tops[i] != NULL)
4009 nvlist_alloc(&tops[i], NV_UNIQUE_NAME, KM_SLEEP);
4010 VERIFY(nvlist_add_string(tops[i], ZPOOL_CONFIG_TYPE,
4011 VDEV_TYPE_MISSING) == 0);
4012 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_ID,
4014 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_GUID,
4019 * Create pool config based on the best vdev config.
4021 nvlist_dup(best_cfg, &config, KM_SLEEP);
4024 * Put this pool's top-level vdevs into a root vdev.
4026 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4028 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4029 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
4030 VDEV_TYPE_ROOT) == 0);
4031 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
4032 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
4033 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
4034 tops, nchildren) == 0);
4037 * Replace the existing vdev_tree with the new root vdev in
4038 * this pool's configuration (remove the old, add the new).
4040 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
4043 * Drop vdev config elements that should not be present at pool level.
4045 nvlist_remove(config, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64);
4046 nvlist_remove(config, ZPOOL_CONFIG_TOP_GUID, DATA_TYPE_UINT64);
4048 for (i = 0; i < count; i++)
4049 nvlist_free(configs[i]);
4050 kmem_free(configs, count * sizeof(void *));
4051 for (i = 0; i < nchildren; i++)
4052 nvlist_free(tops[i]);
4053 kmem_free(tops, nchildren * sizeof(void *));
4054 nvlist_free(nvroot);
4059 spa_import_rootpool(const char *name)
4062 vdev_t *rvd, *bvd, *avd = NULL;
4063 nvlist_t *config, *nvtop;
4069 * Read the label from the boot device and generate a configuration.
4071 config = spa_generate_rootconf(name);
4073 mutex_enter(&spa_namespace_lock);
4074 if (config != NULL) {
4075 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
4076 &pname) == 0 && strcmp(name, pname) == 0);
4077 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg)
4080 if ((spa = spa_lookup(pname)) != NULL) {
4082 * Remove the existing root pool from the namespace so
4083 * that we can replace it with the correct config
4088 spa = spa_add(pname, config, NULL);
4091 * Set spa_ubsync.ub_version as it can be used in vdev_alloc()
4092 * via spa_version().
4094 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
4095 &spa->spa_ubsync.ub_version) != 0)
4096 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
4097 } else if ((spa = spa_lookup(name)) == NULL) {
4098 cmn_err(CE_NOTE, "Cannot find the pool label for '%s'",
4102 VERIFY(nvlist_dup(spa->spa_config, &config, KM_SLEEP) == 0);
4104 spa->spa_is_root = B_TRUE;
4105 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
4108 * Build up a vdev tree based on the boot device's label config.
4110 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4112 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4113 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
4114 VDEV_ALLOC_ROOTPOOL);
4115 spa_config_exit(spa, SCL_ALL, FTAG);
4117 mutex_exit(&spa_namespace_lock);
4118 nvlist_free(config);
4119 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
4124 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4126 spa_config_exit(spa, SCL_ALL, FTAG);
4127 mutex_exit(&spa_namespace_lock);
4129 nvlist_free(config);
4137 * Import a non-root pool into the system.
4140 spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
4143 char *altroot = NULL;
4144 spa_load_state_t state = SPA_LOAD_IMPORT;
4145 zpool_rewind_policy_t policy;
4146 uint64_t mode = spa_mode_global;
4147 uint64_t readonly = B_FALSE;
4150 nvlist_t **spares, **l2cache;
4151 uint_t nspares, nl2cache;
4154 * If a pool with this name exists, return failure.
4156 mutex_enter(&spa_namespace_lock);
4157 if (spa_lookup(pool) != NULL) {
4158 mutex_exit(&spa_namespace_lock);
4159 return (SET_ERROR(EEXIST));
4163 * Create and initialize the spa structure.
4165 (void) nvlist_lookup_string(props,
4166 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4167 (void) nvlist_lookup_uint64(props,
4168 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
4171 spa = spa_add(pool, config, altroot);
4172 spa->spa_import_flags = flags;
4175 * Verbatim import - Take a pool and insert it into the namespace
4176 * as if it had been loaded at boot.
4178 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
4180 spa_configfile_set(spa, props, B_FALSE);
4182 spa_config_sync(spa, B_FALSE, B_TRUE);
4184 mutex_exit(&spa_namespace_lock);
4188 spa_activate(spa, mode);
4191 * Don't start async tasks until we know everything is healthy.
4193 spa_async_suspend(spa);
4195 zpool_get_rewind_policy(config, &policy);
4196 if (policy.zrp_request & ZPOOL_DO_REWIND)
4197 state = SPA_LOAD_RECOVER;
4200 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
4201 * because the user-supplied config is actually the one to trust when
4204 if (state != SPA_LOAD_RECOVER)
4205 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
4207 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
4208 policy.zrp_request);
4211 * Propagate anything learned while loading the pool and pass it
4212 * back to caller (i.e. rewind info, missing devices, etc).
4214 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4215 spa->spa_load_info) == 0);
4217 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4219 * Toss any existing sparelist, as it doesn't have any validity
4220 * anymore, and conflicts with spa_has_spare().
4222 if (spa->spa_spares.sav_config) {
4223 nvlist_free(spa->spa_spares.sav_config);
4224 spa->spa_spares.sav_config = NULL;
4225 spa_load_spares(spa);
4227 if (spa->spa_l2cache.sav_config) {
4228 nvlist_free(spa->spa_l2cache.sav_config);
4229 spa->spa_l2cache.sav_config = NULL;
4230 spa_load_l2cache(spa);
4233 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4236 error = spa_validate_aux(spa, nvroot, -1ULL,
4239 error = spa_validate_aux(spa, nvroot, -1ULL,
4240 VDEV_ALLOC_L2CACHE);
4241 spa_config_exit(spa, SCL_ALL, FTAG);
4244 spa_configfile_set(spa, props, B_FALSE);
4246 if (error != 0 || (props && spa_writeable(spa) &&
4247 (error = spa_prop_set(spa, props)))) {
4249 spa_deactivate(spa);
4251 mutex_exit(&spa_namespace_lock);
4255 spa_async_resume(spa);
4258 * Override any spares and level 2 cache devices as specified by
4259 * the user, as these may have correct device names/devids, etc.
4261 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
4262 &spares, &nspares) == 0) {
4263 if (spa->spa_spares.sav_config)
4264 VERIFY(nvlist_remove(spa->spa_spares.sav_config,
4265 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
4267 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
4268 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4269 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
4270 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
4271 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4272 spa_load_spares(spa);
4273 spa_config_exit(spa, SCL_ALL, FTAG);
4274 spa->spa_spares.sav_sync = B_TRUE;
4276 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
4277 &l2cache, &nl2cache) == 0) {
4278 if (spa->spa_l2cache.sav_config)
4279 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
4280 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
4282 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
4283 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4284 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
4285 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
4286 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4287 spa_load_l2cache(spa);
4288 spa_config_exit(spa, SCL_ALL, FTAG);
4289 spa->spa_l2cache.sav_sync = B_TRUE;
4293 * Check for any removed devices.
4295 if (spa->spa_autoreplace) {
4296 spa_aux_check_removed(&spa->spa_spares);
4297 spa_aux_check_removed(&spa->spa_l2cache);
4300 if (spa_writeable(spa)) {
4302 * Update the config cache to include the newly-imported pool.
4304 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4308 * It's possible that the pool was expanded while it was exported.
4309 * We kick off an async task to handle this for us.
4311 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
4313 mutex_exit(&spa_namespace_lock);
4314 spa_history_log_version(spa, "import");
4318 zvol_create_minors(pool);
4325 spa_tryimport(nvlist_t *tryconfig)
4327 nvlist_t *config = NULL;
4333 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
4336 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
4340 * Create and initialize the spa structure.
4342 mutex_enter(&spa_namespace_lock);
4343 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
4344 spa_activate(spa, FREAD);
4347 * Pass off the heavy lifting to spa_load().
4348 * Pass TRUE for mosconfig because the user-supplied config
4349 * is actually the one to trust when doing an import.
4351 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
4354 * If 'tryconfig' was at least parsable, return the current config.
4356 if (spa->spa_root_vdev != NULL) {
4357 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4358 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
4360 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4362 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
4363 spa->spa_uberblock.ub_timestamp) == 0);
4364 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4365 spa->spa_load_info) == 0);
4368 * If the bootfs property exists on this pool then we
4369 * copy it out so that external consumers can tell which
4370 * pools are bootable.
4372 if ((!error || error == EEXIST) && spa->spa_bootfs) {
4373 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4376 * We have to play games with the name since the
4377 * pool was opened as TRYIMPORT_NAME.
4379 if (dsl_dsobj_to_dsname(spa_name(spa),
4380 spa->spa_bootfs, tmpname) == 0) {
4382 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4384 cp = strchr(tmpname, '/');
4386 (void) strlcpy(dsname, tmpname,
4389 (void) snprintf(dsname, MAXPATHLEN,
4390 "%s/%s", poolname, ++cp);
4392 VERIFY(nvlist_add_string(config,
4393 ZPOOL_CONFIG_BOOTFS, dsname) == 0);
4394 kmem_free(dsname, MAXPATHLEN);
4396 kmem_free(tmpname, MAXPATHLEN);
4400 * Add the list of hot spares and level 2 cache devices.
4402 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4403 spa_add_spares(spa, config);
4404 spa_add_l2cache(spa, config);
4405 spa_config_exit(spa, SCL_CONFIG, FTAG);
4409 spa_deactivate(spa);
4411 mutex_exit(&spa_namespace_lock);
4417 * Pool export/destroy
4419 * The act of destroying or exporting a pool is very simple. We make sure there
4420 * is no more pending I/O and any references to the pool are gone. Then, we
4421 * update the pool state and sync all the labels to disk, removing the
4422 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4423 * we don't sync the labels or remove the configuration cache.
4426 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
4427 boolean_t force, boolean_t hardforce)
4434 if (!(spa_mode_global & FWRITE))
4435 return (SET_ERROR(EROFS));
4437 mutex_enter(&spa_namespace_lock);
4438 if ((spa = spa_lookup(pool)) == NULL) {
4439 mutex_exit(&spa_namespace_lock);
4440 return (SET_ERROR(ENOENT));
4444 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4445 * reacquire the namespace lock, and see if we can export.
4447 spa_open_ref(spa, FTAG);
4448 mutex_exit(&spa_namespace_lock);
4449 spa_async_suspend(spa);
4450 mutex_enter(&spa_namespace_lock);
4451 spa_close(spa, FTAG);
4454 * The pool will be in core if it's openable,
4455 * in which case we can modify its state.
4457 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
4459 * Objsets may be open only because they're dirty, so we
4460 * have to force it to sync before checking spa_refcnt.
4462 txg_wait_synced(spa->spa_dsl_pool, 0);
4465 * A pool cannot be exported or destroyed if there are active
4466 * references. If we are resetting a pool, allow references by
4467 * fault injection handlers.
4469 if (!spa_refcount_zero(spa) ||
4470 (spa->spa_inject_ref != 0 &&
4471 new_state != POOL_STATE_UNINITIALIZED)) {
4472 spa_async_resume(spa);
4473 mutex_exit(&spa_namespace_lock);
4474 return (SET_ERROR(EBUSY));
4478 * A pool cannot be exported if it has an active shared spare.
4479 * This is to prevent other pools stealing the active spare
4480 * from an exported pool. At user's own will, such pool can
4481 * be forcedly exported.
4483 if (!force && new_state == POOL_STATE_EXPORTED &&
4484 spa_has_active_shared_spare(spa)) {
4485 spa_async_resume(spa);
4486 mutex_exit(&spa_namespace_lock);
4487 return (SET_ERROR(EXDEV));
4491 * We want this to be reflected on every label,
4492 * so mark them all dirty. spa_unload() will do the
4493 * final sync that pushes these changes out.
4495 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
4496 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4497 spa->spa_state = new_state;
4498 spa->spa_final_txg = spa_last_synced_txg(spa) +
4500 vdev_config_dirty(spa->spa_root_vdev);
4501 spa_config_exit(spa, SCL_ALL, FTAG);
4505 spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
4507 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4509 spa_deactivate(spa);
4512 if (oldconfig && spa->spa_config)
4513 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
4515 if (new_state != POOL_STATE_UNINITIALIZED) {
4517 spa_config_sync(spa, B_TRUE, B_TRUE);
4520 mutex_exit(&spa_namespace_lock);
4526 * Destroy a storage pool.
4529 spa_destroy(char *pool)
4531 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
4536 * Export a storage pool.
4539 spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
4540 boolean_t hardforce)
4542 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
4547 * Similar to spa_export(), this unloads the spa_t without actually removing it
4548 * from the namespace in any way.
4551 spa_reset(char *pool)
4553 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
4558 * ==========================================================================
4559 * Device manipulation
4560 * ==========================================================================
4564 * Add a device to a storage pool.
4567 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
4571 vdev_t *rvd = spa->spa_root_vdev;
4573 nvlist_t **spares, **l2cache;
4574 uint_t nspares, nl2cache;
4576 ASSERT(spa_writeable(spa));
4578 txg = spa_vdev_enter(spa);
4580 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
4581 VDEV_ALLOC_ADD)) != 0)
4582 return (spa_vdev_exit(spa, NULL, txg, error));
4584 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */
4586 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
4590 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
4594 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
4595 return (spa_vdev_exit(spa, vd, txg, EINVAL));
4597 if (vd->vdev_children != 0 &&
4598 (error = vdev_create(vd, txg, B_FALSE)) != 0)
4599 return (spa_vdev_exit(spa, vd, txg, error));
4602 * We must validate the spares and l2cache devices after checking the
4603 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
4605 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
4606 return (spa_vdev_exit(spa, vd, txg, error));
4609 * Transfer each new top-level vdev from vd to rvd.
4611 for (int c = 0; c < vd->vdev_children; c++) {
4614 * Set the vdev id to the first hole, if one exists.
4616 for (id = 0; id < rvd->vdev_children; id++) {
4617 if (rvd->vdev_child[id]->vdev_ishole) {
4618 vdev_free(rvd->vdev_child[id]);
4622 tvd = vd->vdev_child[c];
4623 vdev_remove_child(vd, tvd);
4625 vdev_add_child(rvd, tvd);
4626 vdev_config_dirty(tvd);
4630 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
4631 ZPOOL_CONFIG_SPARES);
4632 spa_load_spares(spa);
4633 spa->spa_spares.sav_sync = B_TRUE;
4636 if (nl2cache != 0) {
4637 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
4638 ZPOOL_CONFIG_L2CACHE);
4639 spa_load_l2cache(spa);
4640 spa->spa_l2cache.sav_sync = B_TRUE;
4644 * We have to be careful when adding new vdevs to an existing pool.
4645 * If other threads start allocating from these vdevs before we
4646 * sync the config cache, and we lose power, then upon reboot we may
4647 * fail to open the pool because there are DVAs that the config cache
4648 * can't translate. Therefore, we first add the vdevs without
4649 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4650 * and then let spa_config_update() initialize the new metaslabs.
4652 * spa_load() checks for added-but-not-initialized vdevs, so that
4653 * if we lose power at any point in this sequence, the remaining
4654 * steps will be completed the next time we load the pool.
4656 (void) spa_vdev_exit(spa, vd, txg, 0);
4658 mutex_enter(&spa_namespace_lock);
4659 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4660 mutex_exit(&spa_namespace_lock);
4666 * Attach a device to a mirror. The arguments are the path to any device
4667 * in the mirror, and the nvroot for the new device. If the path specifies
4668 * a device that is not mirrored, we automatically insert the mirror vdev.
4670 * If 'replacing' is specified, the new device is intended to replace the
4671 * existing device; in this case the two devices are made into their own
4672 * mirror using the 'replacing' vdev, which is functionally identical to
4673 * the mirror vdev (it actually reuses all the same ops) but has a few
4674 * extra rules: you can't attach to it after it's been created, and upon
4675 * completion of resilvering, the first disk (the one being replaced)
4676 * is automatically detached.
4679 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
4681 uint64_t txg, dtl_max_txg;
4682 vdev_t *rvd = spa->spa_root_vdev;
4683 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
4685 char *oldvdpath, *newvdpath;
4689 ASSERT(spa_writeable(spa));
4691 txg = spa_vdev_enter(spa);
4693 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
4696 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4698 if (!oldvd->vdev_ops->vdev_op_leaf)
4699 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4701 pvd = oldvd->vdev_parent;
4703 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
4704 VDEV_ALLOC_ATTACH)) != 0)
4705 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4707 if (newrootvd->vdev_children != 1)
4708 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4710 newvd = newrootvd->vdev_child[0];
4712 if (!newvd->vdev_ops->vdev_op_leaf)
4713 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4715 if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
4716 return (spa_vdev_exit(spa, newrootvd, txg, error));
4719 * Spares can't replace logs
4721 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
4722 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4726 * For attach, the only allowable parent is a mirror or the root
4729 if (pvd->vdev_ops != &vdev_mirror_ops &&
4730 pvd->vdev_ops != &vdev_root_ops)
4731 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4733 pvops = &vdev_mirror_ops;
4736 * Active hot spares can only be replaced by inactive hot
4739 if (pvd->vdev_ops == &vdev_spare_ops &&
4740 oldvd->vdev_isspare &&
4741 !spa_has_spare(spa, newvd->vdev_guid))
4742 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4745 * If the source is a hot spare, and the parent isn't already a
4746 * spare, then we want to create a new hot spare. Otherwise, we
4747 * want to create a replacing vdev. The user is not allowed to
4748 * attach to a spared vdev child unless the 'isspare' state is
4749 * the same (spare replaces spare, non-spare replaces
4752 if (pvd->vdev_ops == &vdev_replacing_ops &&
4753 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
4754 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4755 } else if (pvd->vdev_ops == &vdev_spare_ops &&
4756 newvd->vdev_isspare != oldvd->vdev_isspare) {
4757 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4760 if (newvd->vdev_isspare)
4761 pvops = &vdev_spare_ops;
4763 pvops = &vdev_replacing_ops;
4767 * Make sure the new device is big enough.
4769 if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
4770 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
4773 * The new device cannot have a higher alignment requirement
4774 * than the top-level vdev.
4776 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4777 return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4780 * If this is an in-place replacement, update oldvd's path and devid
4781 * to make it distinguishable from newvd, and unopenable from now on.
4783 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4784 spa_strfree(oldvd->vdev_path);
4785 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4787 (void) sprintf(oldvd->vdev_path, "%s/%s",
4788 newvd->vdev_path, "old");
4789 if (oldvd->vdev_devid != NULL) {
4790 spa_strfree(oldvd->vdev_devid);
4791 oldvd->vdev_devid = NULL;
4795 /* mark the device being resilvered */
4796 newvd->vdev_resilver_txg = txg;
4799 * If the parent is not a mirror, or if we're replacing, insert the new
4800 * mirror/replacing/spare vdev above oldvd.
4802 if (pvd->vdev_ops != pvops)
4803 pvd = vdev_add_parent(oldvd, pvops);
4805 ASSERT(pvd->vdev_top->vdev_parent == rvd);
4806 ASSERT(pvd->vdev_ops == pvops);
4807 ASSERT(oldvd->vdev_parent == pvd);
4810 * Extract the new device from its root and add it to pvd.
4812 vdev_remove_child(newrootvd, newvd);
4813 newvd->vdev_id = pvd->vdev_children;
4814 newvd->vdev_crtxg = oldvd->vdev_crtxg;
4815 vdev_add_child(pvd, newvd);
4817 tvd = newvd->vdev_top;
4818 ASSERT(pvd->vdev_top == tvd);
4819 ASSERT(tvd->vdev_parent == rvd);
4821 vdev_config_dirty(tvd);
4824 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4825 * for any dmu_sync-ed blocks. It will propagate upward when
4826 * spa_vdev_exit() calls vdev_dtl_reassess().
4828 dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4830 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4831 dtl_max_txg - TXG_INITIAL);
4833 if (newvd->vdev_isspare) {
4834 spa_spare_activate(newvd);
4835 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
4838 oldvdpath = spa_strdup(oldvd->vdev_path);
4839 newvdpath = spa_strdup(newvd->vdev_path);
4840 newvd_isspare = newvd->vdev_isspare;
4843 * Mark newvd's DTL dirty in this txg.
4845 vdev_dirty(tvd, VDD_DTL, newvd, txg);
4848 * Schedule the resilver to restart in the future. We do this to
4849 * ensure that dmu_sync-ed blocks have been stitched into the
4850 * respective datasets.
4852 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4857 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4859 spa_history_log_internal(spa, "vdev attach", NULL,
4860 "%s vdev=%s %s vdev=%s",
4861 replacing && newvd_isspare ? "spare in" :
4862 replacing ? "replace" : "attach", newvdpath,
4863 replacing ? "for" : "to", oldvdpath);
4865 spa_strfree(oldvdpath);
4866 spa_strfree(newvdpath);
4868 if (spa->spa_bootfs)
4869 spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
4875 * Detach a device from a mirror or replacing vdev.
4877 * If 'replace_done' is specified, only detach if the parent
4878 * is a replacing vdev.
4881 spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4885 vdev_t *rvd = spa->spa_root_vdev;
4886 vdev_t *vd, *pvd, *cvd, *tvd;
4887 boolean_t unspare = B_FALSE;
4888 uint64_t unspare_guid = 0;
4891 ASSERT(spa_writeable(spa));
4893 txg = spa_vdev_enter(spa);
4895 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4898 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4900 if (!vd->vdev_ops->vdev_op_leaf)
4901 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4903 pvd = vd->vdev_parent;
4906 * If the parent/child relationship is not as expected, don't do it.
4907 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4908 * vdev that's replacing B with C. The user's intent in replacing
4909 * is to go from M(A,B) to M(A,C). If the user decides to cancel
4910 * the replace by detaching C, the expected behavior is to end up
4911 * M(A,B). But suppose that right after deciding to detach C,
4912 * the replacement of B completes. We would have M(A,C), and then
4913 * ask to detach C, which would leave us with just A -- not what
4914 * the user wanted. To prevent this, we make sure that the
4915 * parent/child relationship hasn't changed -- in this example,
4916 * that C's parent is still the replacing vdev R.
4918 if (pvd->vdev_guid != pguid && pguid != 0)
4919 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4922 * Only 'replacing' or 'spare' vdevs can be replaced.
4924 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4925 pvd->vdev_ops != &vdev_spare_ops)
4926 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4928 ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4929 spa_version(spa) >= SPA_VERSION_SPARES);
4932 * Only mirror, replacing, and spare vdevs support detach.
4934 if (pvd->vdev_ops != &vdev_replacing_ops &&
4935 pvd->vdev_ops != &vdev_mirror_ops &&
4936 pvd->vdev_ops != &vdev_spare_ops)
4937 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4940 * If this device has the only valid copy of some data,
4941 * we cannot safely detach it.
4943 if (vdev_dtl_required(vd))
4944 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4946 ASSERT(pvd->vdev_children >= 2);
4949 * If we are detaching the second disk from a replacing vdev, then
4950 * check to see if we changed the original vdev's path to have "/old"
4951 * at the end in spa_vdev_attach(). If so, undo that change now.
4953 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4954 vd->vdev_path != NULL) {
4955 size_t len = strlen(vd->vdev_path);
4957 for (int c = 0; c < pvd->vdev_children; c++) {
4958 cvd = pvd->vdev_child[c];
4960 if (cvd == vd || cvd->vdev_path == NULL)
4963 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4964 strcmp(cvd->vdev_path + len, "/old") == 0) {
4965 spa_strfree(cvd->vdev_path);
4966 cvd->vdev_path = spa_strdup(vd->vdev_path);
4973 * If we are detaching the original disk from a spare, then it implies
4974 * that the spare should become a real disk, and be removed from the
4975 * active spare list for the pool.
4977 if (pvd->vdev_ops == &vdev_spare_ops &&
4979 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4983 * Erase the disk labels so the disk can be used for other things.
4984 * This must be done after all other error cases are handled,
4985 * but before we disembowel vd (so we can still do I/O to it).
4986 * But if we can't do it, don't treat the error as fatal --
4987 * it may be that the unwritability of the disk is the reason
4988 * it's being detached!
4990 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4993 * Remove vd from its parent and compact the parent's children.
4995 vdev_remove_child(pvd, vd);
4996 vdev_compact_children(pvd);
4999 * Remember one of the remaining children so we can get tvd below.
5001 cvd = pvd->vdev_child[pvd->vdev_children - 1];
5004 * If we need to remove the remaining child from the list of hot spares,
5005 * do it now, marking the vdev as no longer a spare in the process.
5006 * We must do this before vdev_remove_parent(), because that can
5007 * change the GUID if it creates a new toplevel GUID. For a similar
5008 * reason, we must remove the spare now, in the same txg as the detach;
5009 * otherwise someone could attach a new sibling, change the GUID, and
5010 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
5013 ASSERT(cvd->vdev_isspare);
5014 spa_spare_remove(cvd);
5015 unspare_guid = cvd->vdev_guid;
5016 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
5017 cvd->vdev_unspare = B_TRUE;
5021 * If the parent mirror/replacing vdev only has one child,
5022 * the parent is no longer needed. Remove it from the tree.
5024 if (pvd->vdev_children == 1) {
5025 if (pvd->vdev_ops == &vdev_spare_ops)
5026 cvd->vdev_unspare = B_FALSE;
5027 vdev_remove_parent(cvd);
5032 * We don't set tvd until now because the parent we just removed
5033 * may have been the previous top-level vdev.
5035 tvd = cvd->vdev_top;
5036 ASSERT(tvd->vdev_parent == rvd);
5039 * Reevaluate the parent vdev state.
5041 vdev_propagate_state(cvd);
5044 * If the 'autoexpand' property is set on the pool then automatically
5045 * try to expand the size of the pool. For example if the device we
5046 * just detached was smaller than the others, it may be possible to
5047 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
5048 * first so that we can obtain the updated sizes of the leaf vdevs.
5050 if (spa->spa_autoexpand) {
5052 vdev_expand(tvd, txg);
5055 vdev_config_dirty(tvd);
5058 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
5059 * vd->vdev_detached is set and free vd's DTL object in syncing context.
5060 * But first make sure we're not on any *other* txg's DTL list, to
5061 * prevent vd from being accessed after it's freed.
5063 vdpath = spa_strdup(vd->vdev_path);
5064 for (int t = 0; t < TXG_SIZE; t++)
5065 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
5066 vd->vdev_detached = B_TRUE;
5067 vdev_dirty(tvd, VDD_DTL, vd, txg);
5069 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
5071 /* hang on to the spa before we release the lock */
5072 spa_open_ref(spa, FTAG);
5074 error = spa_vdev_exit(spa, vd, txg, 0);
5076 spa_history_log_internal(spa, "detach", NULL,
5078 spa_strfree(vdpath);
5081 * If this was the removal of the original device in a hot spare vdev,
5082 * then we want to go through and remove the device from the hot spare
5083 * list of every other pool.
5086 spa_t *altspa = NULL;
5088 mutex_enter(&spa_namespace_lock);
5089 while ((altspa = spa_next(altspa)) != NULL) {
5090 if (altspa->spa_state != POOL_STATE_ACTIVE ||
5094 spa_open_ref(altspa, FTAG);
5095 mutex_exit(&spa_namespace_lock);
5096 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
5097 mutex_enter(&spa_namespace_lock);
5098 spa_close(altspa, FTAG);
5100 mutex_exit(&spa_namespace_lock);
5102 /* search the rest of the vdevs for spares to remove */
5103 spa_vdev_resilver_done(spa);
5106 /* all done with the spa; OK to release */
5107 mutex_enter(&spa_namespace_lock);
5108 spa_close(spa, FTAG);
5109 mutex_exit(&spa_namespace_lock);
5115 * Split a set of devices from their mirrors, and create a new pool from them.
5118 spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
5119 nvlist_t *props, boolean_t exp)
5122 uint64_t txg, *glist;
5124 uint_t c, children, lastlog;
5125 nvlist_t **child, *nvl, *tmp;
5127 char *altroot = NULL;
5128 vdev_t *rvd, **vml = NULL; /* vdev modify list */
5129 boolean_t activate_slog;
5131 ASSERT(spa_writeable(spa));
5133 txg = spa_vdev_enter(spa);
5135 /* clear the log and flush everything up to now */
5136 activate_slog = spa_passivate_log(spa);
5137 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5138 error = spa_offline_log(spa);
5139 txg = spa_vdev_config_enter(spa);
5142 spa_activate_log(spa);
5145 return (spa_vdev_exit(spa, NULL, txg, error));
5147 /* check new spa name before going any further */
5148 if (spa_lookup(newname) != NULL)
5149 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
5152 * scan through all the children to ensure they're all mirrors
5154 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
5155 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
5157 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5159 /* first, check to ensure we've got the right child count */
5160 rvd = spa->spa_root_vdev;
5162 for (c = 0; c < rvd->vdev_children; c++) {
5163 vdev_t *vd = rvd->vdev_child[c];
5165 /* don't count the holes & logs as children */
5166 if (vd->vdev_islog || vd->vdev_ishole) {
5174 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
5175 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5177 /* next, ensure no spare or cache devices are part of the split */
5178 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
5179 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
5180 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5182 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
5183 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
5185 /* then, loop over each vdev and validate it */
5186 for (c = 0; c < children; c++) {
5187 uint64_t is_hole = 0;
5189 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
5193 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
5194 spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
5197 error = SET_ERROR(EINVAL);
5202 /* which disk is going to be split? */
5203 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
5205 error = SET_ERROR(EINVAL);
5209 /* look it up in the spa */
5210 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
5211 if (vml[c] == NULL) {
5212 error = SET_ERROR(ENODEV);
5216 /* make sure there's nothing stopping the split */
5217 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
5218 vml[c]->vdev_islog ||
5219 vml[c]->vdev_ishole ||
5220 vml[c]->vdev_isspare ||
5221 vml[c]->vdev_isl2cache ||
5222 !vdev_writeable(vml[c]) ||
5223 vml[c]->vdev_children != 0 ||
5224 vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
5225 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
5226 error = SET_ERROR(EINVAL);
5230 if (vdev_dtl_required(vml[c])) {
5231 error = SET_ERROR(EBUSY);
5235 /* we need certain info from the top level */
5236 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
5237 vml[c]->vdev_top->vdev_ms_array) == 0);
5238 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
5239 vml[c]->vdev_top->vdev_ms_shift) == 0);
5240 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
5241 vml[c]->vdev_top->vdev_asize) == 0);
5242 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
5243 vml[c]->vdev_top->vdev_ashift) == 0);
5247 kmem_free(vml, children * sizeof (vdev_t *));
5248 kmem_free(glist, children * sizeof (uint64_t));
5249 return (spa_vdev_exit(spa, NULL, txg, error));
5252 /* stop writers from using the disks */
5253 for (c = 0; c < children; c++) {
5255 vml[c]->vdev_offline = B_TRUE;
5257 vdev_reopen(spa->spa_root_vdev);
5260 * Temporarily record the splitting vdevs in the spa config. This
5261 * will disappear once the config is regenerated.
5263 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5264 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
5265 glist, children) == 0);
5266 kmem_free(glist, children * sizeof (uint64_t));
5268 mutex_enter(&spa->spa_props_lock);
5269 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
5271 mutex_exit(&spa->spa_props_lock);
5272 spa->spa_config_splitting = nvl;
5273 vdev_config_dirty(spa->spa_root_vdev);
5275 /* configure and create the new pool */
5276 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
5277 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
5278 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
5279 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5280 spa_version(spa)) == 0);
5281 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
5282 spa->spa_config_txg) == 0);
5283 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
5284 spa_generate_guid(NULL)) == 0);
5285 (void) nvlist_lookup_string(props,
5286 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
5288 /* add the new pool to the namespace */
5289 newspa = spa_add(newname, config, altroot);
5290 newspa->spa_config_txg = spa->spa_config_txg;
5291 spa_set_log_state(newspa, SPA_LOG_CLEAR);
5293 /* release the spa config lock, retaining the namespace lock */
5294 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5296 if (zio_injection_enabled)
5297 zio_handle_panic_injection(spa, FTAG, 1);
5299 spa_activate(newspa, spa_mode_global);
5300 spa_async_suspend(newspa);
5303 /* mark that we are creating new spa by splitting */
5304 newspa->spa_splitting_newspa = B_TRUE;
5306 /* create the new pool from the disks of the original pool */
5307 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
5309 newspa->spa_splitting_newspa = B_FALSE;
5314 /* if that worked, generate a real config for the new pool */
5315 if (newspa->spa_root_vdev != NULL) {
5316 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
5317 NV_UNIQUE_NAME, KM_SLEEP) == 0);
5318 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
5319 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
5320 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
5325 if (props != NULL) {
5326 spa_configfile_set(newspa, props, B_FALSE);
5327 error = spa_prop_set(newspa, props);
5332 /* flush everything */
5333 txg = spa_vdev_config_enter(newspa);
5334 vdev_config_dirty(newspa->spa_root_vdev);
5335 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
5337 if (zio_injection_enabled)
5338 zio_handle_panic_injection(spa, FTAG, 2);
5340 spa_async_resume(newspa);
5342 /* finally, update the original pool's config */
5343 txg = spa_vdev_config_enter(spa);
5344 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
5345 error = dmu_tx_assign(tx, TXG_WAIT);
5348 for (c = 0; c < children; c++) {
5349 if (vml[c] != NULL) {
5352 spa_history_log_internal(spa, "detach", tx,
5353 "vdev=%s", vml[c]->vdev_path);
5357 vdev_config_dirty(spa->spa_root_vdev);
5358 spa->spa_config_splitting = NULL;
5362 (void) spa_vdev_exit(spa, NULL, txg, 0);
5364 if (zio_injection_enabled)
5365 zio_handle_panic_injection(spa, FTAG, 3);
5367 /* split is complete; log a history record */
5368 spa_history_log_internal(newspa, "split", NULL,
5369 "from pool %s", spa_name(spa));
5371 kmem_free(vml, children * sizeof (vdev_t *));
5373 /* if we're not going to mount the filesystems in userland, export */
5375 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
5382 spa_deactivate(newspa);
5385 txg = spa_vdev_config_enter(spa);
5387 /* re-online all offlined disks */
5388 for (c = 0; c < children; c++) {
5390 vml[c]->vdev_offline = B_FALSE;
5392 vdev_reopen(spa->spa_root_vdev);
5394 nvlist_free(spa->spa_config_splitting);
5395 spa->spa_config_splitting = NULL;
5396 (void) spa_vdev_exit(spa, NULL, txg, error);
5398 kmem_free(vml, children * sizeof (vdev_t *));
5403 spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
5405 for (int i = 0; i < count; i++) {
5408 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
5411 if (guid == target_guid)
5419 spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
5420 nvlist_t *dev_to_remove)
5422 nvlist_t **newdev = NULL;
5425 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
5427 for (int i = 0, j = 0; i < count; i++) {
5428 if (dev[i] == dev_to_remove)
5430 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
5433 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
5434 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
5436 for (int i = 0; i < count - 1; i++)
5437 nvlist_free(newdev[i]);
5440 kmem_free(newdev, (count - 1) * sizeof (void *));
5444 * Evacuate the device.
5447 spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
5452 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5453 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5454 ASSERT(vd == vd->vdev_top);
5457 * Evacuate the device. We don't hold the config lock as writer
5458 * since we need to do I/O but we do keep the
5459 * spa_namespace_lock held. Once this completes the device
5460 * should no longer have any blocks allocated on it.
5462 if (vd->vdev_islog) {
5463 if (vd->vdev_stat.vs_alloc != 0)
5464 error = spa_offline_log(spa);
5466 error = SET_ERROR(ENOTSUP);
5473 * The evacuation succeeded. Remove any remaining MOS metadata
5474 * associated with this vdev, and wait for these changes to sync.
5476 ASSERT0(vd->vdev_stat.vs_alloc);
5477 txg = spa_vdev_config_enter(spa);
5478 vd->vdev_removing = B_TRUE;
5479 vdev_dirty_leaves(vd, VDD_DTL, txg);
5480 vdev_config_dirty(vd);
5481 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5487 * Complete the removal by cleaning up the namespace.
5490 spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
5492 vdev_t *rvd = spa->spa_root_vdev;
5493 uint64_t id = vd->vdev_id;
5494 boolean_t last_vdev = (id == (rvd->vdev_children - 1));
5496 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5497 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5498 ASSERT(vd == vd->vdev_top);
5501 * Only remove any devices which are empty.
5503 if (vd->vdev_stat.vs_alloc != 0)
5506 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5508 if (list_link_active(&vd->vdev_state_dirty_node))
5509 vdev_state_clean(vd);
5510 if (list_link_active(&vd->vdev_config_dirty_node))
5511 vdev_config_clean(vd);
5516 vdev_compact_children(rvd);
5518 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
5519 vdev_add_child(rvd, vd);
5521 vdev_config_dirty(rvd);
5524 * Reassess the health of our root vdev.
5530 * Remove a device from the pool -
5532 * Removing a device from the vdev namespace requires several steps
5533 * and can take a significant amount of time. As a result we use
5534 * the spa_vdev_config_[enter/exit] functions which allow us to
5535 * grab and release the spa_config_lock while still holding the namespace
5536 * lock. During each step the configuration is synced out.
5538 * Currently, this supports removing only hot spares, slogs, and level 2 ARC
5542 spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
5545 metaslab_group_t *mg;
5546 nvlist_t **spares, **l2cache, *nv;
5548 uint_t nspares, nl2cache;
5550 boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
5552 ASSERT(spa_writeable(spa));
5555 txg = spa_vdev_enter(spa);
5557 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5559 if (spa->spa_spares.sav_vdevs != NULL &&
5560 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5561 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
5562 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
5564 * Only remove the hot spare if it's not currently in use
5567 if (vd == NULL || unspare) {
5568 spa_vdev_remove_aux(spa->spa_spares.sav_config,
5569 ZPOOL_CONFIG_SPARES, spares, nspares, nv);
5570 spa_load_spares(spa);
5571 spa->spa_spares.sav_sync = B_TRUE;
5573 error = SET_ERROR(EBUSY);
5575 } else if (spa->spa_l2cache.sav_vdevs != NULL &&
5576 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5577 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
5578 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
5580 * Cache devices can always be removed.
5582 spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
5583 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
5584 spa_load_l2cache(spa);
5585 spa->spa_l2cache.sav_sync = B_TRUE;
5586 } else if (vd != NULL && vd->vdev_islog) {
5588 ASSERT(vd == vd->vdev_top);
5593 * Stop allocating from this vdev.
5595 metaslab_group_passivate(mg);
5598 * Wait for the youngest allocations and frees to sync,
5599 * and then wait for the deferral of those frees to finish.
5601 spa_vdev_config_exit(spa, NULL,
5602 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
5605 * Attempt to evacuate the vdev.
5607 error = spa_vdev_remove_evacuate(spa, vd);
5609 txg = spa_vdev_config_enter(spa);
5612 * If we couldn't evacuate the vdev, unwind.
5615 metaslab_group_activate(mg);
5616 return (spa_vdev_exit(spa, NULL, txg, error));
5620 * Clean up the vdev namespace.
5622 spa_vdev_remove_from_namespace(spa, vd);
5624 } else if (vd != NULL) {
5626 * Normal vdevs cannot be removed (yet).
5628 error = SET_ERROR(ENOTSUP);
5631 * There is no vdev of any kind with the specified guid.
5633 error = SET_ERROR(ENOENT);
5637 return (spa_vdev_exit(spa, NULL, txg, error));
5643 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5644 * currently spared, so we can detach it.
5647 spa_vdev_resilver_done_hunt(vdev_t *vd)
5649 vdev_t *newvd, *oldvd;
5651 for (int c = 0; c < vd->vdev_children; c++) {
5652 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
5658 * Check for a completed replacement. We always consider the first
5659 * vdev in the list to be the oldest vdev, and the last one to be
5660 * the newest (see spa_vdev_attach() for how that works). In
5661 * the case where the newest vdev is faulted, we will not automatically
5662 * remove it after a resilver completes. This is OK as it will require
5663 * user intervention to determine which disk the admin wishes to keep.
5665 if (vd->vdev_ops == &vdev_replacing_ops) {
5666 ASSERT(vd->vdev_children > 1);
5668 newvd = vd->vdev_child[vd->vdev_children - 1];
5669 oldvd = vd->vdev_child[0];
5671 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
5672 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5673 !vdev_dtl_required(oldvd))
5678 * Check for a completed resilver with the 'unspare' flag set.
5680 if (vd->vdev_ops == &vdev_spare_ops) {
5681 vdev_t *first = vd->vdev_child[0];
5682 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
5684 if (last->vdev_unspare) {
5687 } else if (first->vdev_unspare) {
5694 if (oldvd != NULL &&
5695 vdev_dtl_empty(newvd, DTL_MISSING) &&
5696 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5697 !vdev_dtl_required(oldvd))
5701 * If there are more than two spares attached to a disk,
5702 * and those spares are not required, then we want to
5703 * attempt to free them up now so that they can be used
5704 * by other pools. Once we're back down to a single
5705 * disk+spare, we stop removing them.
5707 if (vd->vdev_children > 2) {
5708 newvd = vd->vdev_child[1];
5710 if (newvd->vdev_isspare && last->vdev_isspare &&
5711 vdev_dtl_empty(last, DTL_MISSING) &&
5712 vdev_dtl_empty(last, DTL_OUTAGE) &&
5713 !vdev_dtl_required(newvd))
5722 spa_vdev_resilver_done(spa_t *spa)
5724 vdev_t *vd, *pvd, *ppvd;
5725 uint64_t guid, sguid, pguid, ppguid;
5727 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5729 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
5730 pvd = vd->vdev_parent;
5731 ppvd = pvd->vdev_parent;
5732 guid = vd->vdev_guid;
5733 pguid = pvd->vdev_guid;
5734 ppguid = ppvd->vdev_guid;
5737 * If we have just finished replacing a hot spared device, then
5738 * we need to detach the parent's first child (the original hot
5741 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
5742 ppvd->vdev_children == 2) {
5743 ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
5744 sguid = ppvd->vdev_child[1]->vdev_guid;
5746 ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd));
5748 spa_config_exit(spa, SCL_ALL, FTAG);
5749 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
5751 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
5753 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5756 spa_config_exit(spa, SCL_ALL, FTAG);
5760 * Update the stored path or FRU for this vdev.
5763 spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
5767 boolean_t sync = B_FALSE;
5769 ASSERT(spa_writeable(spa));
5771 spa_vdev_state_enter(spa, SCL_ALL);
5773 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
5774 return (spa_vdev_state_exit(spa, NULL, ENOENT));
5776 if (!vd->vdev_ops->vdev_op_leaf)
5777 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
5780 if (strcmp(value, vd->vdev_path) != 0) {
5781 spa_strfree(vd->vdev_path);
5782 vd->vdev_path = spa_strdup(value);
5786 if (vd->vdev_fru == NULL) {
5787 vd->vdev_fru = spa_strdup(value);
5789 } else if (strcmp(value, vd->vdev_fru) != 0) {
5790 spa_strfree(vd->vdev_fru);
5791 vd->vdev_fru = spa_strdup(value);
5796 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
5800 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
5802 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5806 spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5808 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5812 * ==========================================================================
5814 * ==========================================================================
5818 spa_scan_stop(spa_t *spa)
5820 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5821 if (dsl_scan_resilvering(spa->spa_dsl_pool))
5822 return (SET_ERROR(EBUSY));
5823 return (dsl_scan_cancel(spa->spa_dsl_pool));
5827 spa_scan(spa_t *spa, pool_scan_func_t func)
5829 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5831 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5832 return (SET_ERROR(ENOTSUP));
5835 * If a resilver was requested, but there is no DTL on a
5836 * writeable leaf device, we have nothing to do.
5838 if (func == POOL_SCAN_RESILVER &&
5839 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5840 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5844 return (dsl_scan(spa->spa_dsl_pool, func));
5848 * ==========================================================================
5849 * SPA async task processing
5850 * ==========================================================================
5854 spa_async_remove(spa_t *spa, vdev_t *vd)
5856 if (vd->vdev_remove_wanted) {
5857 vd->vdev_remove_wanted = B_FALSE;
5858 vd->vdev_delayed_close = B_FALSE;
5859 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5862 * We want to clear the stats, but we don't want to do a full
5863 * vdev_clear() as that will cause us to throw away
5864 * degraded/faulted state as well as attempt to reopen the
5865 * device, all of which is a waste.
5867 vd->vdev_stat.vs_read_errors = 0;
5868 vd->vdev_stat.vs_write_errors = 0;
5869 vd->vdev_stat.vs_checksum_errors = 0;
5871 vdev_state_dirty(vd->vdev_top);
5874 for (int c = 0; c < vd->vdev_children; c++)
5875 spa_async_remove(spa, vd->vdev_child[c]);
5879 spa_async_probe(spa_t *spa, vdev_t *vd)
5881 if (vd->vdev_probe_wanted) {
5882 vd->vdev_probe_wanted = B_FALSE;
5883 vdev_reopen(vd); /* vdev_open() does the actual probe */
5886 for (int c = 0; c < vd->vdev_children; c++)
5887 spa_async_probe(spa, vd->vdev_child[c]);
5891 spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5897 if (!spa->spa_autoexpand)
5900 for (int c = 0; c < vd->vdev_children; c++) {
5901 vdev_t *cvd = vd->vdev_child[c];
5902 spa_async_autoexpand(spa, cvd);
5905 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5908 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
5909 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
5911 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5912 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
5914 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
5915 ESC_ZFS_VDEV_AUTOEXPAND, attr, &eid, DDI_SLEEP);
5918 kmem_free(physpath, MAXPATHLEN);
5922 spa_async_thread(void *arg)
5927 ASSERT(spa->spa_sync_on);
5929 mutex_enter(&spa->spa_async_lock);
5930 tasks = spa->spa_async_tasks;
5931 spa->spa_async_tasks &= SPA_ASYNC_REMOVE;
5932 mutex_exit(&spa->spa_async_lock);
5935 * See if the config needs to be updated.
5937 if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5938 uint64_t old_space, new_space;
5940 mutex_enter(&spa_namespace_lock);
5941 old_space = metaslab_class_get_space(spa_normal_class(spa));
5942 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5943 new_space = metaslab_class_get_space(spa_normal_class(spa));
5944 mutex_exit(&spa_namespace_lock);
5947 * If the pool grew as a result of the config update,
5948 * then log an internal history event.
5950 if (new_space != old_space) {
5951 spa_history_log_internal(spa, "vdev online", NULL,
5952 "pool '%s' size: %llu(+%llu)",
5953 spa_name(spa), new_space, new_space - old_space);
5957 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5958 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5959 spa_async_autoexpand(spa, spa->spa_root_vdev);
5960 spa_config_exit(spa, SCL_CONFIG, FTAG);
5964 * See if any devices need to be probed.
5966 if (tasks & SPA_ASYNC_PROBE) {
5967 spa_vdev_state_enter(spa, SCL_NONE);
5968 spa_async_probe(spa, spa->spa_root_vdev);
5969 (void) spa_vdev_state_exit(spa, NULL, 0);
5973 * If any devices are done replacing, detach them.
5975 if (tasks & SPA_ASYNC_RESILVER_DONE)
5976 spa_vdev_resilver_done(spa);
5979 * Kick off a resilver.
5981 if (tasks & SPA_ASYNC_RESILVER)
5982 dsl_resilver_restart(spa->spa_dsl_pool, 0);
5985 * Let the world know that we're done.
5987 mutex_enter(&spa->spa_async_lock);
5988 spa->spa_async_thread = NULL;
5989 cv_broadcast(&spa->spa_async_cv);
5990 mutex_exit(&spa->spa_async_lock);
5995 spa_async_thread_vd(void *arg)
6000 ASSERT(spa->spa_sync_on);
6002 mutex_enter(&spa->spa_async_lock);
6003 tasks = spa->spa_async_tasks;
6005 spa->spa_async_tasks &= ~SPA_ASYNC_REMOVE;
6006 mutex_exit(&spa->spa_async_lock);
6009 * See if any devices need to be marked REMOVED.
6011 if (tasks & SPA_ASYNC_REMOVE) {
6012 spa_vdev_state_enter(spa, SCL_NONE);
6013 spa_async_remove(spa, spa->spa_root_vdev);
6014 for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
6015 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
6016 for (int i = 0; i < spa->spa_spares.sav_count; i++)
6017 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
6018 (void) spa_vdev_state_exit(spa, NULL, 0);
6022 * Let the world know that we're done.
6024 mutex_enter(&spa->spa_async_lock);
6025 tasks = spa->spa_async_tasks;
6026 if ((tasks & SPA_ASYNC_REMOVE) != 0)
6028 spa->spa_async_thread_vd = NULL;
6029 cv_broadcast(&spa->spa_async_cv);
6030 mutex_exit(&spa->spa_async_lock);
6035 spa_async_suspend(spa_t *spa)
6037 mutex_enter(&spa->spa_async_lock);
6038 spa->spa_async_suspended++;
6039 while (spa->spa_async_thread != NULL &&
6040 spa->spa_async_thread_vd != NULL)
6041 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
6042 mutex_exit(&spa->spa_async_lock);
6046 spa_async_resume(spa_t *spa)
6048 mutex_enter(&spa->spa_async_lock);
6049 ASSERT(spa->spa_async_suspended != 0);
6050 spa->spa_async_suspended--;
6051 mutex_exit(&spa->spa_async_lock);
6055 spa_async_tasks_pending(spa_t *spa)
6057 uint_t non_config_tasks;
6059 boolean_t config_task_suspended;
6061 non_config_tasks = spa->spa_async_tasks & ~(SPA_ASYNC_CONFIG_UPDATE |
6063 config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE;
6064 if (spa->spa_ccw_fail_time == 0) {
6065 config_task_suspended = B_FALSE;
6067 config_task_suspended =
6068 (gethrtime() - spa->spa_ccw_fail_time) <
6069 (zfs_ccw_retry_interval * NANOSEC);
6072 return (non_config_tasks || (config_task && !config_task_suspended));
6076 spa_async_dispatch(spa_t *spa)
6078 mutex_enter(&spa->spa_async_lock);
6079 if (spa_async_tasks_pending(spa) &&
6080 !spa->spa_async_suspended &&
6081 spa->spa_async_thread == NULL &&
6083 spa->spa_async_thread = thread_create(NULL, 0,
6084 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
6085 mutex_exit(&spa->spa_async_lock);
6089 spa_async_dispatch_vd(spa_t *spa)
6091 mutex_enter(&spa->spa_async_lock);
6092 if ((spa->spa_async_tasks & SPA_ASYNC_REMOVE) != 0 &&
6093 !spa->spa_async_suspended &&
6094 spa->spa_async_thread_vd == NULL &&
6096 spa->spa_async_thread_vd = thread_create(NULL, 0,
6097 spa_async_thread_vd, spa, 0, &p0, TS_RUN, maxclsyspri);
6098 mutex_exit(&spa->spa_async_lock);
6102 spa_async_request(spa_t *spa, int task)
6104 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
6105 mutex_enter(&spa->spa_async_lock);
6106 spa->spa_async_tasks |= task;
6107 mutex_exit(&spa->spa_async_lock);
6108 spa_async_dispatch_vd(spa);
6112 * ==========================================================================
6113 * SPA syncing routines
6114 * ==========================================================================
6118 bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
6121 bpobj_enqueue(bpo, bp, tx);
6126 spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
6130 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
6131 BP_GET_PSIZE(bp), zio->io_flags));
6136 * Note: this simple function is not inlined to make it easier to dtrace the
6137 * amount of time spent syncing frees.
6140 spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx)
6142 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6143 bplist_iterate(bpl, spa_free_sync_cb, zio, tx);
6144 VERIFY(zio_wait(zio) == 0);
6148 * Note: this simple function is not inlined to make it easier to dtrace the
6149 * amount of time spent syncing deferred frees.
6152 spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx)
6154 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6155 VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj,
6156 spa_free_sync_cb, zio, tx), ==, 0);
6157 VERIFY0(zio_wait(zio));
6162 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
6164 char *packed = NULL;
6169 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
6172 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
6173 * information. This avoids the dmu_buf_will_dirty() path and
6174 * saves us a pre-read to get data we don't actually care about.
6176 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
6177 packed = kmem_alloc(bufsize, KM_SLEEP);
6179 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
6181 bzero(packed + nvsize, bufsize - nvsize);
6183 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
6185 kmem_free(packed, bufsize);
6187 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
6188 dmu_buf_will_dirty(db, tx);
6189 *(uint64_t *)db->db_data = nvsize;
6190 dmu_buf_rele(db, FTAG);
6194 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
6195 const char *config, const char *entry)
6205 * Update the MOS nvlist describing the list of available devices.
6206 * spa_validate_aux() will have already made sure this nvlist is
6207 * valid and the vdevs are labeled appropriately.
6209 if (sav->sav_object == 0) {
6210 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
6211 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
6212 sizeof (uint64_t), tx);
6213 VERIFY(zap_update(spa->spa_meta_objset,
6214 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
6215 &sav->sav_object, tx) == 0);
6218 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
6219 if (sav->sav_count == 0) {
6220 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
6222 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
6223 for (i = 0; i < sav->sav_count; i++)
6224 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
6225 B_FALSE, VDEV_CONFIG_L2CACHE);
6226 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
6227 sav->sav_count) == 0);
6228 for (i = 0; i < sav->sav_count; i++)
6229 nvlist_free(list[i]);
6230 kmem_free(list, sav->sav_count * sizeof (void *));
6233 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
6234 nvlist_free(nvroot);
6236 sav->sav_sync = B_FALSE;
6240 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
6244 if (list_is_empty(&spa->spa_config_dirty_list))
6247 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6249 config = spa_config_generate(spa, spa->spa_root_vdev,
6250 dmu_tx_get_txg(tx), B_FALSE);
6253 * If we're upgrading the spa version then make sure that
6254 * the config object gets updated with the correct version.
6256 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
6257 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
6258 spa->spa_uberblock.ub_version);
6260 spa_config_exit(spa, SCL_STATE, FTAG);
6262 if (spa->spa_config_syncing)
6263 nvlist_free(spa->spa_config_syncing);
6264 spa->spa_config_syncing = config;
6266 spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
6270 spa_sync_version(void *arg, dmu_tx_t *tx)
6272 uint64_t *versionp = arg;
6273 uint64_t version = *versionp;
6274 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
6277 * Setting the version is special cased when first creating the pool.
6279 ASSERT(tx->tx_txg != TXG_INITIAL);
6281 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
6282 ASSERT(version >= spa_version(spa));
6284 spa->spa_uberblock.ub_version = version;
6285 vdev_config_dirty(spa->spa_root_vdev);
6286 spa_history_log_internal(spa, "set", tx, "version=%lld", version);
6290 * Set zpool properties.
6293 spa_sync_props(void *arg, dmu_tx_t *tx)
6295 nvlist_t *nvp = arg;
6296 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
6297 objset_t *mos = spa->spa_meta_objset;
6298 nvpair_t *elem = NULL;
6300 mutex_enter(&spa->spa_props_lock);
6302 while ((elem = nvlist_next_nvpair(nvp, elem))) {
6304 char *strval, *fname;
6306 const char *propname;
6307 zprop_type_t proptype;
6310 switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
6313 * We checked this earlier in spa_prop_validate().
6315 ASSERT(zpool_prop_feature(nvpair_name(elem)));
6317 fname = strchr(nvpair_name(elem), '@') + 1;
6318 VERIFY0(zfeature_lookup_name(fname, &fid));
6320 spa_feature_enable(spa, fid, tx);
6321 spa_history_log_internal(spa, "set", tx,
6322 "%s=enabled", nvpair_name(elem));
6325 case ZPOOL_PROP_VERSION:
6326 intval = fnvpair_value_uint64(elem);
6328 * The version is synced seperatly before other
6329 * properties and should be correct by now.
6331 ASSERT3U(spa_version(spa), >=, intval);
6334 case ZPOOL_PROP_ALTROOT:
6336 * 'altroot' is a non-persistent property. It should
6337 * have been set temporarily at creation or import time.
6339 ASSERT(spa->spa_root != NULL);
6342 case ZPOOL_PROP_READONLY:
6343 case ZPOOL_PROP_CACHEFILE:
6345 * 'readonly' and 'cachefile' are also non-persisitent
6349 case ZPOOL_PROP_COMMENT:
6350 strval = fnvpair_value_string(elem);
6351 if (spa->spa_comment != NULL)
6352 spa_strfree(spa->spa_comment);
6353 spa->spa_comment = spa_strdup(strval);
6355 * We need to dirty the configuration on all the vdevs
6356 * so that their labels get updated. It's unnecessary
6357 * to do this for pool creation since the vdev's
6358 * configuratoin has already been dirtied.
6360 if (tx->tx_txg != TXG_INITIAL)
6361 vdev_config_dirty(spa->spa_root_vdev);
6362 spa_history_log_internal(spa, "set", tx,
6363 "%s=%s", nvpair_name(elem), strval);
6367 * Set pool property values in the poolprops mos object.
6369 if (spa->spa_pool_props_object == 0) {
6370 spa->spa_pool_props_object =
6371 zap_create_link(mos, DMU_OT_POOL_PROPS,
6372 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
6376 /* normalize the property name */
6377 propname = zpool_prop_to_name(prop);
6378 proptype = zpool_prop_get_type(prop);
6380 if (nvpair_type(elem) == DATA_TYPE_STRING) {
6381 ASSERT(proptype == PROP_TYPE_STRING);
6382 strval = fnvpair_value_string(elem);
6383 VERIFY0(zap_update(mos,
6384 spa->spa_pool_props_object, propname,
6385 1, strlen(strval) + 1, strval, tx));
6386 spa_history_log_internal(spa, "set", tx,
6387 "%s=%s", nvpair_name(elem), strval);
6388 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
6389 intval = fnvpair_value_uint64(elem);
6391 if (proptype == PROP_TYPE_INDEX) {
6393 VERIFY0(zpool_prop_index_to_string(
6394 prop, intval, &unused));
6396 VERIFY0(zap_update(mos,
6397 spa->spa_pool_props_object, propname,
6398 8, 1, &intval, tx));
6399 spa_history_log_internal(spa, "set", tx,
6400 "%s=%lld", nvpair_name(elem), intval);
6402 ASSERT(0); /* not allowed */
6406 case ZPOOL_PROP_DELEGATION:
6407 spa->spa_delegation = intval;
6409 case ZPOOL_PROP_BOOTFS:
6410 spa->spa_bootfs = intval;
6412 case ZPOOL_PROP_FAILUREMODE:
6413 spa->spa_failmode = intval;
6415 case ZPOOL_PROP_AUTOEXPAND:
6416 spa->spa_autoexpand = intval;
6417 if (tx->tx_txg != TXG_INITIAL)
6418 spa_async_request(spa,
6419 SPA_ASYNC_AUTOEXPAND);
6421 case ZPOOL_PROP_DEDUPDITTO:
6422 spa->spa_dedup_ditto = intval;
6431 mutex_exit(&spa->spa_props_lock);
6435 * Perform one-time upgrade on-disk changes. spa_version() does not
6436 * reflect the new version this txg, so there must be no changes this
6437 * txg to anything that the upgrade code depends on after it executes.
6438 * Therefore this must be called after dsl_pool_sync() does the sync
6442 spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
6444 dsl_pool_t *dp = spa->spa_dsl_pool;
6446 ASSERT(spa->spa_sync_pass == 1);
6448 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
6450 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
6451 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
6452 dsl_pool_create_origin(dp, tx);
6454 /* Keeping the origin open increases spa_minref */
6455 spa->spa_minref += 3;
6458 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
6459 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
6460 dsl_pool_upgrade_clones(dp, tx);
6463 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
6464 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
6465 dsl_pool_upgrade_dir_clones(dp, tx);
6467 /* Keeping the freedir open increases spa_minref */
6468 spa->spa_minref += 3;
6471 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES &&
6472 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
6473 spa_feature_create_zap_objects(spa, tx);
6477 * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable
6478 * when possibility to use lz4 compression for metadata was added
6479 * Old pools that have this feature enabled must be upgraded to have
6480 * this feature active
6482 if (spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
6483 boolean_t lz4_en = spa_feature_is_enabled(spa,
6484 SPA_FEATURE_LZ4_COMPRESS);
6485 boolean_t lz4_ac = spa_feature_is_active(spa,
6486 SPA_FEATURE_LZ4_COMPRESS);
6488 if (lz4_en && !lz4_ac)
6489 spa_feature_incr(spa, SPA_FEATURE_LZ4_COMPRESS, tx);
6491 rrw_exit(&dp->dp_config_rwlock, FTAG);
6495 * Sync the specified transaction group. New blocks may be dirtied as
6496 * part of the process, so we iterate until it converges.
6499 spa_sync(spa_t *spa, uint64_t txg)
6501 dsl_pool_t *dp = spa->spa_dsl_pool;
6502 objset_t *mos = spa->spa_meta_objset;
6503 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
6504 vdev_t *rvd = spa->spa_root_vdev;
6509 VERIFY(spa_writeable(spa));
6512 * Lock out configuration changes.
6514 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
6516 spa->spa_syncing_txg = txg;
6517 spa->spa_sync_pass = 0;
6520 * If there are any pending vdev state changes, convert them
6521 * into config changes that go out with this transaction group.
6523 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6524 while (list_head(&spa->spa_state_dirty_list) != NULL) {
6526 * We need the write lock here because, for aux vdevs,
6527 * calling vdev_config_dirty() modifies sav_config.
6528 * This is ugly and will become unnecessary when we
6529 * eliminate the aux vdev wart by integrating all vdevs
6530 * into the root vdev tree.
6532 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6533 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
6534 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
6535 vdev_state_clean(vd);
6536 vdev_config_dirty(vd);
6538 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6539 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
6541 spa_config_exit(spa, SCL_STATE, FTAG);
6543 tx = dmu_tx_create_assigned(dp, txg);
6545 spa->spa_sync_starttime = gethrtime();
6547 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid,
6548 spa->spa_sync_starttime + spa->spa_deadman_synctime));
6551 callout_reset(&spa->spa_deadman_cycid,
6552 hz * spa->spa_deadman_synctime / NANOSEC, spa_deadman, spa);
6557 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
6558 * set spa_deflate if we have no raid-z vdevs.
6560 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
6561 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
6564 for (i = 0; i < rvd->vdev_children; i++) {
6565 vd = rvd->vdev_child[i];
6566 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
6569 if (i == rvd->vdev_children) {
6570 spa->spa_deflate = TRUE;
6571 VERIFY(0 == zap_add(spa->spa_meta_objset,
6572 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
6573 sizeof (uint64_t), 1, &spa->spa_deflate, tx));
6578 * If anything has changed in this txg, or if someone is waiting
6579 * for this txg to sync (eg, spa_vdev_remove()), push the
6580 * deferred frees from the previous txg. If not, leave them
6581 * alone so that we don't generate work on an otherwise idle
6584 if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
6585 !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
6586 !txg_list_empty(&dp->dp_sync_tasks, txg) ||
6587 ((dsl_scan_active(dp->dp_scan) ||
6588 txg_sync_waiting(dp)) && !spa_shutting_down(spa))) {
6589 spa_sync_deferred_frees(spa, tx);
6593 * Iterate to convergence.
6596 int pass = ++spa->spa_sync_pass;
6598 spa_sync_config_object(spa, tx);
6599 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
6600 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
6601 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
6602 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
6603 spa_errlog_sync(spa, txg);
6604 dsl_pool_sync(dp, txg);
6606 if (pass < zfs_sync_pass_deferred_free) {
6607 spa_sync_frees(spa, free_bpl, tx);
6609 bplist_iterate(free_bpl, bpobj_enqueue_cb,
6610 &spa->spa_deferred_bpobj, tx);
6614 dsl_scan_sync(dp, tx);
6616 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
6620 spa_sync_upgrades(spa, tx);
6622 } while (dmu_objset_is_dirty(mos, txg));
6625 * Rewrite the vdev configuration (which includes the uberblock)
6626 * to commit the transaction group.
6628 * If there are no dirty vdevs, we sync the uberblock to a few
6629 * random top-level vdevs that are known to be visible in the
6630 * config cache (see spa_vdev_add() for a complete description).
6631 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
6635 * We hold SCL_STATE to prevent vdev open/close/etc.
6636 * while we're attempting to write the vdev labels.
6638 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6640 if (list_is_empty(&spa->spa_config_dirty_list)) {
6641 vdev_t *svd[SPA_DVAS_PER_BP];
6643 int children = rvd->vdev_children;
6644 int c0 = spa_get_random(children);
6646 for (int c = 0; c < children; c++) {
6647 vd = rvd->vdev_child[(c0 + c) % children];
6648 if (vd->vdev_ms_array == 0 || vd->vdev_islog)
6650 svd[svdcount++] = vd;
6651 if (svdcount == SPA_DVAS_PER_BP)
6654 error = vdev_config_sync(svd, svdcount, txg, B_FALSE);
6656 error = vdev_config_sync(svd, svdcount, txg,
6659 error = vdev_config_sync(rvd->vdev_child,
6660 rvd->vdev_children, txg, B_FALSE);
6662 error = vdev_config_sync(rvd->vdev_child,
6663 rvd->vdev_children, txg, B_TRUE);
6667 spa->spa_last_synced_guid = rvd->vdev_guid;
6669 spa_config_exit(spa, SCL_STATE, FTAG);
6673 zio_suspend(spa, NULL);
6674 zio_resume_wait(spa);
6679 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY));
6682 callout_drain(&spa->spa_deadman_cycid);
6687 * Clear the dirty config list.
6689 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
6690 vdev_config_clean(vd);
6693 * Now that the new config has synced transactionally,
6694 * let it become visible to the config cache.
6696 if (spa->spa_config_syncing != NULL) {
6697 spa_config_set(spa, spa->spa_config_syncing);
6698 spa->spa_config_txg = txg;
6699 spa->spa_config_syncing = NULL;
6702 spa->spa_ubsync = spa->spa_uberblock;
6704 dsl_pool_sync_done(dp, txg);
6707 * Update usable space statistics.
6709 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
6710 vdev_sync_done(vd, txg);
6712 spa_update_dspace(spa);
6715 * It had better be the case that we didn't dirty anything
6716 * since vdev_config_sync().
6718 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
6719 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
6720 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
6722 spa->spa_sync_pass = 0;
6724 spa_config_exit(spa, SCL_CONFIG, FTAG);
6726 spa_handle_ignored_writes(spa);
6729 * If any async tasks have been requested, kick them off.
6731 spa_async_dispatch(spa);
6732 spa_async_dispatch_vd(spa);
6736 * Sync all pools. We don't want to hold the namespace lock across these
6737 * operations, so we take a reference on the spa_t and drop the lock during the
6741 spa_sync_allpools(void)
6744 mutex_enter(&spa_namespace_lock);
6745 while ((spa = spa_next(spa)) != NULL) {
6746 if (spa_state(spa) != POOL_STATE_ACTIVE ||
6747 !spa_writeable(spa) || spa_suspended(spa))
6749 spa_open_ref(spa, FTAG);
6750 mutex_exit(&spa_namespace_lock);
6751 txg_wait_synced(spa_get_dsl(spa), 0);
6752 mutex_enter(&spa_namespace_lock);
6753 spa_close(spa, FTAG);
6755 mutex_exit(&spa_namespace_lock);
6759 * ==========================================================================
6760 * Miscellaneous routines
6761 * ==========================================================================
6765 * Remove all pools in the system.
6773 * Remove all cached state. All pools should be closed now,
6774 * so every spa in the AVL tree should be unreferenced.
6776 mutex_enter(&spa_namespace_lock);
6777 while ((spa = spa_next(NULL)) != NULL) {
6779 * Stop async tasks. The async thread may need to detach
6780 * a device that's been replaced, which requires grabbing
6781 * spa_namespace_lock, so we must drop it here.
6783 spa_open_ref(spa, FTAG);
6784 mutex_exit(&spa_namespace_lock);
6785 spa_async_suspend(spa);
6786 mutex_enter(&spa_namespace_lock);
6787 spa_close(spa, FTAG);
6789 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
6791 spa_deactivate(spa);
6795 mutex_exit(&spa_namespace_lock);
6799 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
6804 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
6808 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
6809 vd = spa->spa_l2cache.sav_vdevs[i];
6810 if (vd->vdev_guid == guid)
6814 for (i = 0; i < spa->spa_spares.sav_count; i++) {
6815 vd = spa->spa_spares.sav_vdevs[i];
6816 if (vd->vdev_guid == guid)
6825 spa_upgrade(spa_t *spa, uint64_t version)
6827 ASSERT(spa_writeable(spa));
6829 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6832 * This should only be called for a non-faulted pool, and since a
6833 * future version would result in an unopenable pool, this shouldn't be
6836 ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version));
6837 ASSERT3U(version, >=, spa->spa_uberblock.ub_version);
6839 spa->spa_uberblock.ub_version = version;
6840 vdev_config_dirty(spa->spa_root_vdev);
6842 spa_config_exit(spa, SCL_ALL, FTAG);
6844 txg_wait_synced(spa_get_dsl(spa), 0);
6848 spa_has_spare(spa_t *spa, uint64_t guid)
6852 spa_aux_vdev_t *sav = &spa->spa_spares;
6854 for (i = 0; i < sav->sav_count; i++)
6855 if (sav->sav_vdevs[i]->vdev_guid == guid)
6858 for (i = 0; i < sav->sav_npending; i++) {
6859 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
6860 &spareguid) == 0 && spareguid == guid)
6868 * Check if a pool has an active shared spare device.
6869 * Note: reference count of an active spare is 2, as a spare and as a replace
6872 spa_has_active_shared_spare(spa_t *spa)
6876 spa_aux_vdev_t *sav = &spa->spa_spares;
6878 for (i = 0; i < sav->sav_count; i++) {
6879 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
6880 &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
6889 * Post a sysevent corresponding to the given event. The 'name' must be one of
6890 * the event definitions in sys/sysevent/eventdefs.h. The payload will be
6891 * filled in from the spa and (optionally) the vdev. This doesn't do anything
6892 * in the userland libzpool, as we don't want consumers to misinterpret ztest
6893 * or zdb as real changes.
6896 spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
6900 sysevent_attr_list_t *attr = NULL;
6901 sysevent_value_t value;
6904 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
6907 value.value_type = SE_DATA_TYPE_STRING;
6908 value.value.sv_string = spa_name(spa);
6909 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
6912 value.value_type = SE_DATA_TYPE_UINT64;
6913 value.value.sv_uint64 = spa_guid(spa);
6914 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
6918 value.value_type = SE_DATA_TYPE_UINT64;
6919 value.value.sv_uint64 = vd->vdev_guid;
6920 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
6924 if (vd->vdev_path) {
6925 value.value_type = SE_DATA_TYPE_STRING;
6926 value.value.sv_string = vd->vdev_path;
6927 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
6928 &value, SE_SLEEP) != 0)
6933 if (sysevent_attach_attributes(ev, attr) != 0)
6937 (void) log_sysevent(ev, SE_SLEEP, &eid);
6941 sysevent_free_attr(attr);