4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2013 by Delphix. All rights reserved.
25 * Copyright 2013 Nexenta Systems, Inc. All rights reserved.
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;
217 uint64_t cap, version;
218 zprop_source_t src = ZPROP_SRC_NONE;
219 spa_config_dirent_t *dp;
221 ASSERT(MUTEX_HELD(&spa->spa_props_lock));
224 alloc = metaslab_class_get_alloc(spa_normal_class(spa));
225 size = metaslab_class_get_space(spa_normal_class(spa));
226 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
227 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
228 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
229 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
233 for (int c = 0; c < rvd->vdev_children; c++) {
234 vdev_t *tvd = rvd->vdev_child[c];
235 space += tvd->vdev_max_asize - tvd->vdev_asize;
237 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, space,
240 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
241 (spa_mode(spa) == FREAD), src);
243 cap = (size == 0) ? 0 : (alloc * 100 / size);
244 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
246 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
247 ddt_get_pool_dedup_ratio(spa), src);
249 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
250 rvd->vdev_state, src);
252 version = spa_version(spa);
253 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
254 src = ZPROP_SRC_DEFAULT;
256 src = ZPROP_SRC_LOCAL;
257 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
261 dsl_dir_t *freedir = pool->dp_free_dir;
264 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
265 * when opening pools before this version freedir will be NULL.
267 if (freedir != NULL) {
268 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
269 freedir->dd_phys->dd_used_bytes, src);
271 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
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, 6);
698 return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props,
706 * If the bootfs property value is dsobj, clear it.
709 spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
711 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
712 VERIFY(zap_remove(spa->spa_meta_objset,
713 spa->spa_pool_props_object,
714 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
721 spa_change_guid_check(void *arg, dmu_tx_t *tx)
723 uint64_t *newguid = arg;
724 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
725 vdev_t *rvd = spa->spa_root_vdev;
728 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
729 vdev_state = rvd->vdev_state;
730 spa_config_exit(spa, SCL_STATE, FTAG);
732 if (vdev_state != VDEV_STATE_HEALTHY)
733 return (SET_ERROR(ENXIO));
735 ASSERT3U(spa_guid(spa), !=, *newguid);
741 spa_change_guid_sync(void *arg, dmu_tx_t *tx)
743 uint64_t *newguid = arg;
744 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
746 vdev_t *rvd = spa->spa_root_vdev;
748 oldguid = spa_guid(spa);
750 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
751 rvd->vdev_guid = *newguid;
752 rvd->vdev_guid_sum += (*newguid - oldguid);
753 vdev_config_dirty(rvd);
754 spa_config_exit(spa, SCL_STATE, FTAG);
756 spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu",
761 * Change the GUID for the pool. This is done so that we can later
762 * re-import a pool built from a clone of our own vdevs. We will modify
763 * the root vdev's guid, our own pool guid, and then mark all of our
764 * vdevs dirty. Note that we must make sure that all our vdevs are
765 * online when we do this, or else any vdevs that weren't present
766 * would be orphaned from our pool. We are also going to issue a
767 * sysevent to update any watchers.
770 spa_change_guid(spa_t *spa)
775 mutex_enter(&spa->spa_vdev_top_lock);
776 mutex_enter(&spa_namespace_lock);
777 guid = spa_generate_guid(NULL);
779 error = dsl_sync_task(spa->spa_name, spa_change_guid_check,
780 spa_change_guid_sync, &guid, 5);
783 spa_config_sync(spa, B_FALSE, B_TRUE);
784 spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID);
787 mutex_exit(&spa_namespace_lock);
788 mutex_exit(&spa->spa_vdev_top_lock);
794 * ==========================================================================
795 * SPA state manipulation (open/create/destroy/import/export)
796 * ==========================================================================
800 spa_error_entry_compare(const void *a, const void *b)
802 spa_error_entry_t *sa = (spa_error_entry_t *)a;
803 spa_error_entry_t *sb = (spa_error_entry_t *)b;
806 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
807 sizeof (zbookmark_t));
818 * Utility function which retrieves copies of the current logs and
819 * re-initializes them in the process.
822 spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
824 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
826 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
827 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
829 avl_create(&spa->spa_errlist_scrub,
830 spa_error_entry_compare, sizeof (spa_error_entry_t),
831 offsetof(spa_error_entry_t, se_avl));
832 avl_create(&spa->spa_errlist_last,
833 spa_error_entry_compare, sizeof (spa_error_entry_t),
834 offsetof(spa_error_entry_t, se_avl));
838 spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
840 const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
841 enum zti_modes mode = ztip->zti_mode;
842 uint_t value = ztip->zti_value;
843 uint_t count = ztip->zti_count;
844 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
847 boolean_t batch = B_FALSE;
849 if (mode == ZTI_MODE_NULL) {
851 tqs->stqs_taskq = NULL;
855 ASSERT3U(count, >, 0);
857 tqs->stqs_count = count;
858 tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP);
862 ASSERT3U(value, >=, 1);
863 value = MAX(value, 1);
868 flags |= TASKQ_THREADS_CPU_PCT;
869 value = zio_taskq_batch_pct;
873 panic("unrecognized mode for %s_%s taskq (%u:%u) in "
875 zio_type_name[t], zio_taskq_types[q], mode, value);
879 for (uint_t i = 0; i < count; i++) {
883 (void) snprintf(name, sizeof (name), "%s_%s_%u",
884 zio_type_name[t], zio_taskq_types[q], i);
886 (void) snprintf(name, sizeof (name), "%s_%s",
887 zio_type_name[t], zio_taskq_types[q]);
891 if (zio_taskq_sysdc && spa->spa_proc != &p0) {
893 flags |= TASKQ_DC_BATCH;
895 tq = taskq_create_sysdc(name, value, 50, INT_MAX,
896 spa->spa_proc, zio_taskq_basedc, flags);
899 pri_t pri = maxclsyspri;
901 * The write issue taskq can be extremely CPU
902 * intensive. Run it at slightly lower priority
903 * than the other taskqs.
905 if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE)
908 tq = taskq_create_proc(name, value, pri, 50,
909 INT_MAX, spa->spa_proc, flags);
914 tqs->stqs_taskq[i] = tq;
919 spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
921 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
923 if (tqs->stqs_taskq == NULL) {
924 ASSERT0(tqs->stqs_count);
928 for (uint_t i = 0; i < tqs->stqs_count; i++) {
929 ASSERT3P(tqs->stqs_taskq[i], !=, NULL);
930 taskq_destroy(tqs->stqs_taskq[i]);
933 kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *));
934 tqs->stqs_taskq = NULL;
938 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
939 * Note that a type may have multiple discrete taskqs to avoid lock contention
940 * on the taskq itself. In that case we choose which taskq at random by using
941 * the low bits of gethrtime().
944 spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q,
945 task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent)
947 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
950 ASSERT3P(tqs->stqs_taskq, !=, NULL);
951 ASSERT3U(tqs->stqs_count, !=, 0);
953 if (tqs->stqs_count == 1) {
954 tq = tqs->stqs_taskq[0];
956 tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count];
959 taskq_dispatch_ent(tq, func, arg, flags, ent);
963 spa_create_zio_taskqs(spa_t *spa)
965 for (int t = 0; t < ZIO_TYPES; t++) {
966 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
967 spa_taskqs_init(spa, t, q);
975 spa_thread(void *arg)
980 user_t *pu = PTOU(curproc);
982 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
985 ASSERT(curproc != &p0);
986 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
987 "zpool-%s", spa->spa_name);
988 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
991 /* bind this thread to the requested psrset */
992 if (zio_taskq_psrset_bind != PS_NONE) {
994 mutex_enter(&cpu_lock);
995 mutex_enter(&pidlock);
996 mutex_enter(&curproc->p_lock);
998 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
999 0, NULL, NULL) == 0) {
1000 curthread->t_bind_pset = zio_taskq_psrset_bind;
1003 "Couldn't bind process for zfs pool \"%s\" to "
1004 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
1007 mutex_exit(&curproc->p_lock);
1008 mutex_exit(&pidlock);
1009 mutex_exit(&cpu_lock);
1015 if (zio_taskq_sysdc) {
1016 sysdc_thread_enter(curthread, 100, 0);
1020 spa->spa_proc = curproc;
1021 spa->spa_did = curthread->t_did;
1023 spa_create_zio_taskqs(spa);
1025 mutex_enter(&spa->spa_proc_lock);
1026 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
1028 spa->spa_proc_state = SPA_PROC_ACTIVE;
1029 cv_broadcast(&spa->spa_proc_cv);
1031 CALLB_CPR_SAFE_BEGIN(&cprinfo);
1032 while (spa->spa_proc_state == SPA_PROC_ACTIVE)
1033 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1034 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
1036 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
1037 spa->spa_proc_state = SPA_PROC_GONE;
1038 spa->spa_proc = &p0;
1039 cv_broadcast(&spa->spa_proc_cv);
1040 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */
1042 mutex_enter(&curproc->p_lock);
1045 #endif /* SPA_PROCESS */
1049 * Activate an uninitialized pool.
1052 spa_activate(spa_t *spa, int mode)
1054 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
1056 spa->spa_state = POOL_STATE_ACTIVE;
1057 spa->spa_mode = mode;
1059 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
1060 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
1062 /* Try to create a covering process */
1063 mutex_enter(&spa->spa_proc_lock);
1064 ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
1065 ASSERT(spa->spa_proc == &p0);
1069 /* Only create a process if we're going to be around a while. */
1070 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
1071 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
1073 spa->spa_proc_state = SPA_PROC_CREATED;
1074 while (spa->spa_proc_state == SPA_PROC_CREATED) {
1075 cv_wait(&spa->spa_proc_cv,
1076 &spa->spa_proc_lock);
1078 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1079 ASSERT(spa->spa_proc != &p0);
1080 ASSERT(spa->spa_did != 0);
1084 "Couldn't create process for zfs pool \"%s\"\n",
1089 #endif /* SPA_PROCESS */
1090 mutex_exit(&spa->spa_proc_lock);
1092 /* If we didn't create a process, we need to create our taskqs. */
1093 ASSERT(spa->spa_proc == &p0);
1094 if (spa->spa_proc == &p0) {
1095 spa_create_zio_taskqs(spa);
1099 * Start TRIM thread.
1101 trim_thread_create(spa);
1103 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
1104 offsetof(vdev_t, vdev_config_dirty_node));
1105 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
1106 offsetof(vdev_t, vdev_state_dirty_node));
1108 txg_list_create(&spa->spa_vdev_txg_list,
1109 offsetof(struct vdev, vdev_txg_node));
1111 avl_create(&spa->spa_errlist_scrub,
1112 spa_error_entry_compare, sizeof (spa_error_entry_t),
1113 offsetof(spa_error_entry_t, se_avl));
1114 avl_create(&spa->spa_errlist_last,
1115 spa_error_entry_compare, sizeof (spa_error_entry_t),
1116 offsetof(spa_error_entry_t, se_avl));
1120 * Opposite of spa_activate().
1123 spa_deactivate(spa_t *spa)
1125 ASSERT(spa->spa_sync_on == B_FALSE);
1126 ASSERT(spa->spa_dsl_pool == NULL);
1127 ASSERT(spa->spa_root_vdev == NULL);
1128 ASSERT(spa->spa_async_zio_root == NULL);
1129 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
1132 * Stop TRIM thread in case spa_unload() wasn't called directly
1133 * before spa_deactivate().
1135 trim_thread_destroy(spa);
1137 txg_list_destroy(&spa->spa_vdev_txg_list);
1139 list_destroy(&spa->spa_config_dirty_list);
1140 list_destroy(&spa->spa_state_dirty_list);
1142 for (int t = 0; t < ZIO_TYPES; t++) {
1143 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1144 spa_taskqs_fini(spa, t, q);
1148 metaslab_class_destroy(spa->spa_normal_class);
1149 spa->spa_normal_class = NULL;
1151 metaslab_class_destroy(spa->spa_log_class);
1152 spa->spa_log_class = NULL;
1155 * If this was part of an import or the open otherwise failed, we may
1156 * still have errors left in the queues. Empty them just in case.
1158 spa_errlog_drain(spa);
1160 avl_destroy(&spa->spa_errlist_scrub);
1161 avl_destroy(&spa->spa_errlist_last);
1163 spa->spa_state = POOL_STATE_UNINITIALIZED;
1165 mutex_enter(&spa->spa_proc_lock);
1166 if (spa->spa_proc_state != SPA_PROC_NONE) {
1167 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1168 spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1169 cv_broadcast(&spa->spa_proc_cv);
1170 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1171 ASSERT(spa->spa_proc != &p0);
1172 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1174 ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1175 spa->spa_proc_state = SPA_PROC_NONE;
1177 ASSERT(spa->spa_proc == &p0);
1178 mutex_exit(&spa->spa_proc_lock);
1182 * We want to make sure spa_thread() has actually exited the ZFS
1183 * module, so that the module can't be unloaded out from underneath
1186 if (spa->spa_did != 0) {
1187 thread_join(spa->spa_did);
1190 #endif /* SPA_PROCESS */
1194 * Verify a pool configuration, and construct the vdev tree appropriately. This
1195 * will create all the necessary vdevs in the appropriate layout, with each vdev
1196 * in the CLOSED state. This will prep the pool before open/creation/import.
1197 * All vdev validation is done by the vdev_alloc() routine.
1200 spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1201 uint_t id, int atype)
1207 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1210 if ((*vdp)->vdev_ops->vdev_op_leaf)
1213 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1216 if (error == ENOENT)
1222 return (SET_ERROR(EINVAL));
1225 for (int c = 0; c < children; c++) {
1227 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1235 ASSERT(*vdp != NULL);
1241 * Opposite of spa_load().
1244 spa_unload(spa_t *spa)
1248 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1253 trim_thread_destroy(spa);
1258 spa_async_suspend(spa);
1263 if (spa->spa_sync_on) {
1264 txg_sync_stop(spa->spa_dsl_pool);
1265 spa->spa_sync_on = B_FALSE;
1269 * Wait for any outstanding async I/O to complete.
1271 if (spa->spa_async_zio_root != NULL) {
1272 (void) zio_wait(spa->spa_async_zio_root);
1273 spa->spa_async_zio_root = NULL;
1276 bpobj_close(&spa->spa_deferred_bpobj);
1278 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1283 if (spa->spa_root_vdev)
1284 vdev_free(spa->spa_root_vdev);
1285 ASSERT(spa->spa_root_vdev == NULL);
1288 * Close the dsl pool.
1290 if (spa->spa_dsl_pool) {
1291 dsl_pool_close(spa->spa_dsl_pool);
1292 spa->spa_dsl_pool = NULL;
1293 spa->spa_meta_objset = NULL;
1300 * Drop and purge level 2 cache
1302 spa_l2cache_drop(spa);
1304 for (i = 0; i < spa->spa_spares.sav_count; i++)
1305 vdev_free(spa->spa_spares.sav_vdevs[i]);
1306 if (spa->spa_spares.sav_vdevs) {
1307 kmem_free(spa->spa_spares.sav_vdevs,
1308 spa->spa_spares.sav_count * sizeof (void *));
1309 spa->spa_spares.sav_vdevs = NULL;
1311 if (spa->spa_spares.sav_config) {
1312 nvlist_free(spa->spa_spares.sav_config);
1313 spa->spa_spares.sav_config = NULL;
1315 spa->spa_spares.sav_count = 0;
1317 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1318 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1319 vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1321 if (spa->spa_l2cache.sav_vdevs) {
1322 kmem_free(spa->spa_l2cache.sav_vdevs,
1323 spa->spa_l2cache.sav_count * sizeof (void *));
1324 spa->spa_l2cache.sav_vdevs = NULL;
1326 if (spa->spa_l2cache.sav_config) {
1327 nvlist_free(spa->spa_l2cache.sav_config);
1328 spa->spa_l2cache.sav_config = NULL;
1330 spa->spa_l2cache.sav_count = 0;
1332 spa->spa_async_suspended = 0;
1334 if (spa->spa_comment != NULL) {
1335 spa_strfree(spa->spa_comment);
1336 spa->spa_comment = NULL;
1339 spa_config_exit(spa, SCL_ALL, FTAG);
1343 * Load (or re-load) the current list of vdevs describing the active spares for
1344 * this pool. When this is called, we have some form of basic information in
1345 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1346 * then re-generate a more complete list including status information.
1349 spa_load_spares(spa_t *spa)
1356 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1359 * First, close and free any existing spare vdevs.
1361 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1362 vd = spa->spa_spares.sav_vdevs[i];
1364 /* Undo the call to spa_activate() below */
1365 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1366 B_FALSE)) != NULL && tvd->vdev_isspare)
1367 spa_spare_remove(tvd);
1372 if (spa->spa_spares.sav_vdevs)
1373 kmem_free(spa->spa_spares.sav_vdevs,
1374 spa->spa_spares.sav_count * sizeof (void *));
1376 if (spa->spa_spares.sav_config == NULL)
1379 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1380 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1382 spa->spa_spares.sav_count = (int)nspares;
1383 spa->spa_spares.sav_vdevs = NULL;
1389 * Construct the array of vdevs, opening them to get status in the
1390 * process. For each spare, there is potentially two different vdev_t
1391 * structures associated with it: one in the list of spares (used only
1392 * for basic validation purposes) and one in the active vdev
1393 * configuration (if it's spared in). During this phase we open and
1394 * validate each vdev on the spare list. If the vdev also exists in the
1395 * active configuration, then we also mark this vdev as an active spare.
1397 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1399 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1400 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1401 VDEV_ALLOC_SPARE) == 0);
1404 spa->spa_spares.sav_vdevs[i] = vd;
1406 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1407 B_FALSE)) != NULL) {
1408 if (!tvd->vdev_isspare)
1412 * We only mark the spare active if we were successfully
1413 * able to load the vdev. Otherwise, importing a pool
1414 * with a bad active spare would result in strange
1415 * behavior, because multiple pool would think the spare
1416 * is actively in use.
1418 * There is a vulnerability here to an equally bizarre
1419 * circumstance, where a dead active spare is later
1420 * brought back to life (onlined or otherwise). Given
1421 * the rarity of this scenario, and the extra complexity
1422 * it adds, we ignore the possibility.
1424 if (!vdev_is_dead(tvd))
1425 spa_spare_activate(tvd);
1429 vd->vdev_aux = &spa->spa_spares;
1431 if (vdev_open(vd) != 0)
1434 if (vdev_validate_aux(vd) == 0)
1439 * Recompute the stashed list of spares, with status information
1442 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1443 DATA_TYPE_NVLIST_ARRAY) == 0);
1445 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1447 for (i = 0; i < spa->spa_spares.sav_count; i++)
1448 spares[i] = vdev_config_generate(spa,
1449 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1450 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1451 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1452 for (i = 0; i < spa->spa_spares.sav_count; i++)
1453 nvlist_free(spares[i]);
1454 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1458 * Load (or re-load) the current list of vdevs describing the active l2cache for
1459 * this pool. When this is called, we have some form of basic information in
1460 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1461 * then re-generate a more complete list including status information.
1462 * Devices which are already active have their details maintained, and are
1466 spa_load_l2cache(spa_t *spa)
1470 int i, j, oldnvdevs;
1472 vdev_t *vd, **oldvdevs, **newvdevs;
1473 spa_aux_vdev_t *sav = &spa->spa_l2cache;
1475 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1477 if (sav->sav_config != NULL) {
1478 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1479 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1480 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1486 oldvdevs = sav->sav_vdevs;
1487 oldnvdevs = sav->sav_count;
1488 sav->sav_vdevs = NULL;
1492 * Process new nvlist of vdevs.
1494 for (i = 0; i < nl2cache; i++) {
1495 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1499 for (j = 0; j < oldnvdevs; j++) {
1501 if (vd != NULL && guid == vd->vdev_guid) {
1503 * Retain previous vdev for add/remove ops.
1511 if (newvdevs[i] == NULL) {
1515 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1516 VDEV_ALLOC_L2CACHE) == 0);
1521 * Commit this vdev as an l2cache device,
1522 * even if it fails to open.
1524 spa_l2cache_add(vd);
1529 spa_l2cache_activate(vd);
1531 if (vdev_open(vd) != 0)
1534 (void) vdev_validate_aux(vd);
1536 if (!vdev_is_dead(vd))
1537 l2arc_add_vdev(spa, vd);
1542 * Purge vdevs that were dropped
1544 for (i = 0; i < oldnvdevs; i++) {
1549 ASSERT(vd->vdev_isl2cache);
1551 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1552 pool != 0ULL && l2arc_vdev_present(vd))
1553 l2arc_remove_vdev(vd);
1554 vdev_clear_stats(vd);
1560 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1562 if (sav->sav_config == NULL)
1565 sav->sav_vdevs = newvdevs;
1566 sav->sav_count = (int)nl2cache;
1569 * Recompute the stashed list of l2cache devices, with status
1570 * information this time.
1572 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1573 DATA_TYPE_NVLIST_ARRAY) == 0);
1575 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1576 for (i = 0; i < sav->sav_count; i++)
1577 l2cache[i] = vdev_config_generate(spa,
1578 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1579 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1580 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1582 for (i = 0; i < sav->sav_count; i++)
1583 nvlist_free(l2cache[i]);
1585 kmem_free(l2cache, sav->sav_count * sizeof (void *));
1589 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1592 char *packed = NULL;
1597 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
1598 nvsize = *(uint64_t *)db->db_data;
1599 dmu_buf_rele(db, FTAG);
1601 packed = kmem_alloc(nvsize, KM_SLEEP);
1602 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1605 error = nvlist_unpack(packed, nvsize, value, 0);
1606 kmem_free(packed, nvsize);
1612 * Checks to see if the given vdev could not be opened, in which case we post a
1613 * sysevent to notify the autoreplace code that the device has been removed.
1616 spa_check_removed(vdev_t *vd)
1618 for (int c = 0; c < vd->vdev_children; c++)
1619 spa_check_removed(vd->vdev_child[c]);
1621 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) &&
1623 zfs_post_autoreplace(vd->vdev_spa, vd);
1624 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1629 * Validate the current config against the MOS config
1632 spa_config_valid(spa_t *spa, nvlist_t *config)
1634 vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1637 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1639 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1640 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1642 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1645 * If we're doing a normal import, then build up any additional
1646 * diagnostic information about missing devices in this config.
1647 * We'll pass this up to the user for further processing.
1649 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1650 nvlist_t **child, *nv;
1653 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1655 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1657 for (int c = 0; c < rvd->vdev_children; c++) {
1658 vdev_t *tvd = rvd->vdev_child[c];
1659 vdev_t *mtvd = mrvd->vdev_child[c];
1661 if (tvd->vdev_ops == &vdev_missing_ops &&
1662 mtvd->vdev_ops != &vdev_missing_ops &&
1664 child[idx++] = vdev_config_generate(spa, mtvd,
1669 VERIFY(nvlist_add_nvlist_array(nv,
1670 ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1671 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1672 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1674 for (int i = 0; i < idx; i++)
1675 nvlist_free(child[i]);
1678 kmem_free(child, rvd->vdev_children * sizeof (char **));
1682 * Compare the root vdev tree with the information we have
1683 * from the MOS config (mrvd). Check each top-level vdev
1684 * with the corresponding MOS config top-level (mtvd).
1686 for (int c = 0; c < rvd->vdev_children; c++) {
1687 vdev_t *tvd = rvd->vdev_child[c];
1688 vdev_t *mtvd = mrvd->vdev_child[c];
1691 * Resolve any "missing" vdevs in the current configuration.
1692 * If we find that the MOS config has more accurate information
1693 * about the top-level vdev then use that vdev instead.
1695 if (tvd->vdev_ops == &vdev_missing_ops &&
1696 mtvd->vdev_ops != &vdev_missing_ops) {
1698 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1702 * Device specific actions.
1704 if (mtvd->vdev_islog) {
1705 spa_set_log_state(spa, SPA_LOG_CLEAR);
1708 * XXX - once we have 'readonly' pool
1709 * support we should be able to handle
1710 * missing data devices by transitioning
1711 * the pool to readonly.
1717 * Swap the missing vdev with the data we were
1718 * able to obtain from the MOS config.
1720 vdev_remove_child(rvd, tvd);
1721 vdev_remove_child(mrvd, mtvd);
1723 vdev_add_child(rvd, mtvd);
1724 vdev_add_child(mrvd, tvd);
1726 spa_config_exit(spa, SCL_ALL, FTAG);
1728 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1731 } else if (mtvd->vdev_islog) {
1733 * Load the slog device's state from the MOS config
1734 * since it's possible that the label does not
1735 * contain the most up-to-date information.
1737 vdev_load_log_state(tvd, mtvd);
1742 spa_config_exit(spa, SCL_ALL, FTAG);
1745 * Ensure we were able to validate the config.
1747 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1751 * Check for missing log devices
1754 spa_check_logs(spa_t *spa)
1756 boolean_t rv = B_FALSE;
1758 switch (spa->spa_log_state) {
1759 case SPA_LOG_MISSING:
1760 /* need to recheck in case slog has been restored */
1761 case SPA_LOG_UNKNOWN:
1762 rv = (dmu_objset_find(spa->spa_name, zil_check_log_chain,
1763 NULL, DS_FIND_CHILDREN) != 0);
1765 spa_set_log_state(spa, SPA_LOG_MISSING);
1772 spa_passivate_log(spa_t *spa)
1774 vdev_t *rvd = spa->spa_root_vdev;
1775 boolean_t slog_found = B_FALSE;
1777 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1779 if (!spa_has_slogs(spa))
1782 for (int c = 0; c < rvd->vdev_children; c++) {
1783 vdev_t *tvd = rvd->vdev_child[c];
1784 metaslab_group_t *mg = tvd->vdev_mg;
1786 if (tvd->vdev_islog) {
1787 metaslab_group_passivate(mg);
1788 slog_found = B_TRUE;
1792 return (slog_found);
1796 spa_activate_log(spa_t *spa)
1798 vdev_t *rvd = spa->spa_root_vdev;
1800 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1802 for (int c = 0; c < rvd->vdev_children; c++) {
1803 vdev_t *tvd = rvd->vdev_child[c];
1804 metaslab_group_t *mg = tvd->vdev_mg;
1806 if (tvd->vdev_islog)
1807 metaslab_group_activate(mg);
1812 spa_offline_log(spa_t *spa)
1816 error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1817 NULL, DS_FIND_CHILDREN);
1820 * We successfully offlined the log device, sync out the
1821 * current txg so that the "stubby" block can be removed
1824 txg_wait_synced(spa->spa_dsl_pool, 0);
1830 spa_aux_check_removed(spa_aux_vdev_t *sav)
1834 for (i = 0; i < sav->sav_count; i++)
1835 spa_check_removed(sav->sav_vdevs[i]);
1839 spa_claim_notify(zio_t *zio)
1841 spa_t *spa = zio->io_spa;
1846 mutex_enter(&spa->spa_props_lock); /* any mutex will do */
1847 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1848 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1849 mutex_exit(&spa->spa_props_lock);
1852 typedef struct spa_load_error {
1853 uint64_t sle_meta_count;
1854 uint64_t sle_data_count;
1858 spa_load_verify_done(zio_t *zio)
1860 blkptr_t *bp = zio->io_bp;
1861 spa_load_error_t *sle = zio->io_private;
1862 dmu_object_type_t type = BP_GET_TYPE(bp);
1863 int error = zio->io_error;
1866 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1867 type != DMU_OT_INTENT_LOG)
1868 atomic_add_64(&sle->sle_meta_count, 1);
1870 atomic_add_64(&sle->sle_data_count, 1);
1872 zio_data_buf_free(zio->io_data, zio->io_size);
1877 spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1878 const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1882 size_t size = BP_GET_PSIZE(bp);
1883 void *data = zio_data_buf_alloc(size);
1885 zio_nowait(zio_read(rio, spa, bp, data, size,
1886 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1887 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1888 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1894 spa_load_verify(spa_t *spa)
1897 spa_load_error_t sle = { 0 };
1898 zpool_rewind_policy_t policy;
1899 boolean_t verify_ok = B_FALSE;
1902 zpool_get_rewind_policy(spa->spa_config, &policy);
1904 if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1907 rio = zio_root(spa, NULL, &sle,
1908 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1910 error = traverse_pool(spa, spa->spa_verify_min_txg,
1911 TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio);
1913 (void) zio_wait(rio);
1915 spa->spa_load_meta_errors = sle.sle_meta_count;
1916 spa->spa_load_data_errors = sle.sle_data_count;
1918 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1919 sle.sle_data_count <= policy.zrp_maxdata) {
1923 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1924 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1926 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1927 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1928 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1929 VERIFY(nvlist_add_int64(spa->spa_load_info,
1930 ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1931 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1932 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1934 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1938 if (error != ENXIO && error != EIO)
1939 error = SET_ERROR(EIO);
1943 return (verify_ok ? 0 : EIO);
1947 * Find a value in the pool props object.
1950 spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1952 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1953 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1957 * Find a value in the pool directory object.
1960 spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1962 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1963 name, sizeof (uint64_t), 1, val));
1967 spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1969 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1974 * Fix up config after a partly-completed split. This is done with the
1975 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
1976 * pool have that entry in their config, but only the splitting one contains
1977 * a list of all the guids of the vdevs that are being split off.
1979 * This function determines what to do with that list: either rejoin
1980 * all the disks to the pool, or complete the splitting process. To attempt
1981 * the rejoin, each disk that is offlined is marked online again, and
1982 * we do a reopen() call. If the vdev label for every disk that was
1983 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1984 * then we call vdev_split() on each disk, and complete the split.
1986 * Otherwise we leave the config alone, with all the vdevs in place in
1987 * the original pool.
1990 spa_try_repair(spa_t *spa, nvlist_t *config)
1997 boolean_t attempt_reopen;
1999 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
2002 /* check that the config is complete */
2003 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
2004 &glist, &gcount) != 0)
2007 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
2009 /* attempt to online all the vdevs & validate */
2010 attempt_reopen = B_TRUE;
2011 for (i = 0; i < gcount; i++) {
2012 if (glist[i] == 0) /* vdev is hole */
2015 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
2016 if (vd[i] == NULL) {
2018 * Don't bother attempting to reopen the disks;
2019 * just do the split.
2021 attempt_reopen = B_FALSE;
2023 /* attempt to re-online it */
2024 vd[i]->vdev_offline = B_FALSE;
2028 if (attempt_reopen) {
2029 vdev_reopen(spa->spa_root_vdev);
2031 /* check each device to see what state it's in */
2032 for (extracted = 0, i = 0; i < gcount; i++) {
2033 if (vd[i] != NULL &&
2034 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
2041 * If every disk has been moved to the new pool, or if we never
2042 * even attempted to look at them, then we split them off for
2045 if (!attempt_reopen || gcount == extracted) {
2046 for (i = 0; i < gcount; i++)
2049 vdev_reopen(spa->spa_root_vdev);
2052 kmem_free(vd, gcount * sizeof (vdev_t *));
2056 spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
2057 boolean_t mosconfig)
2059 nvlist_t *config = spa->spa_config;
2060 char *ereport = FM_EREPORT_ZFS_POOL;
2066 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
2067 return (SET_ERROR(EINVAL));
2069 ASSERT(spa->spa_comment == NULL);
2070 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
2071 spa->spa_comment = spa_strdup(comment);
2074 * Versioning wasn't explicitly added to the label until later, so if
2075 * it's not present treat it as the initial version.
2077 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
2078 &spa->spa_ubsync.ub_version) != 0)
2079 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
2081 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
2082 &spa->spa_config_txg);
2084 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
2085 spa_guid_exists(pool_guid, 0)) {
2086 error = SET_ERROR(EEXIST);
2088 spa->spa_config_guid = pool_guid;
2090 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
2092 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
2096 nvlist_free(spa->spa_load_info);
2097 spa->spa_load_info = fnvlist_alloc();
2099 gethrestime(&spa->spa_loaded_ts);
2100 error = spa_load_impl(spa, pool_guid, config, state, type,
2101 mosconfig, &ereport);
2104 spa->spa_minref = refcount_count(&spa->spa_refcount);
2106 if (error != EEXIST) {
2107 spa->spa_loaded_ts.tv_sec = 0;
2108 spa->spa_loaded_ts.tv_nsec = 0;
2110 if (error != EBADF) {
2111 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
2114 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2121 * Load an existing storage pool, using the pool's builtin spa_config as a
2122 * source of configuration information.
2125 spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
2126 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
2130 nvlist_t *nvroot = NULL;
2133 uberblock_t *ub = &spa->spa_uberblock;
2134 uint64_t children, config_cache_txg = spa->spa_config_txg;
2135 int orig_mode = spa->spa_mode;
2138 boolean_t missing_feat_write = B_FALSE;
2141 * If this is an untrusted config, access the pool in read-only mode.
2142 * This prevents things like resilvering recently removed devices.
2145 spa->spa_mode = FREAD;
2147 ASSERT(MUTEX_HELD(&spa_namespace_lock));
2149 spa->spa_load_state = state;
2151 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
2152 return (SET_ERROR(EINVAL));
2154 parse = (type == SPA_IMPORT_EXISTING ?
2155 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2158 * Create "The Godfather" zio to hold all async IOs
2160 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2161 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2164 * Parse the configuration into a vdev tree. We explicitly set the
2165 * value that will be returned by spa_version() since parsing the
2166 * configuration requires knowing the version number.
2168 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2169 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
2170 spa_config_exit(spa, SCL_ALL, FTAG);
2175 ASSERT(spa->spa_root_vdev == rvd);
2177 if (type != SPA_IMPORT_ASSEMBLE) {
2178 ASSERT(spa_guid(spa) == pool_guid);
2182 * Try to open all vdevs, loading each label in the process.
2184 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2185 error = vdev_open(rvd);
2186 spa_config_exit(spa, SCL_ALL, FTAG);
2191 * We need to validate the vdev labels against the configuration that
2192 * we have in hand, which is dependent on the setting of mosconfig. If
2193 * mosconfig is true then we're validating the vdev labels based on
2194 * that config. Otherwise, we're validating against the cached config
2195 * (zpool.cache) that was read when we loaded the zfs module, and then
2196 * later we will recursively call spa_load() and validate against
2199 * If we're assembling a new pool that's been split off from an
2200 * existing pool, the labels haven't yet been updated so we skip
2201 * validation for now.
2203 if (type != SPA_IMPORT_ASSEMBLE) {
2204 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2205 error = vdev_validate(rvd, mosconfig);
2206 spa_config_exit(spa, SCL_ALL, FTAG);
2211 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2212 return (SET_ERROR(ENXIO));
2216 * Find the best uberblock.
2218 vdev_uberblock_load(rvd, ub, &label);
2221 * If we weren't able to find a single valid uberblock, return failure.
2223 if (ub->ub_txg == 0) {
2225 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2229 * If the pool has an unsupported version we can't open it.
2231 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2233 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2236 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2240 * If we weren't able to find what's necessary for reading the
2241 * MOS in the label, return failure.
2243 if (label == NULL || nvlist_lookup_nvlist(label,
2244 ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) {
2246 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2251 * Update our in-core representation with the definitive values
2254 nvlist_free(spa->spa_label_features);
2255 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2261 * Look through entries in the label nvlist's features_for_read. If
2262 * there is a feature listed there which we don't understand then we
2263 * cannot open a pool.
2265 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2266 nvlist_t *unsup_feat;
2268 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2271 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2273 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2274 if (!zfeature_is_supported(nvpair_name(nvp))) {
2275 VERIFY(nvlist_add_string(unsup_feat,
2276 nvpair_name(nvp), "") == 0);
2280 if (!nvlist_empty(unsup_feat)) {
2281 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2282 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2283 nvlist_free(unsup_feat);
2284 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2288 nvlist_free(unsup_feat);
2292 * If the vdev guid sum doesn't match the uberblock, we have an
2293 * incomplete configuration. We first check to see if the pool
2294 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2295 * If it is, defer the vdev_guid_sum check till later so we
2296 * can handle missing vdevs.
2298 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
2299 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
2300 rvd->vdev_guid_sum != ub->ub_guid_sum)
2301 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2303 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2304 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2305 spa_try_repair(spa, config);
2306 spa_config_exit(spa, SCL_ALL, FTAG);
2307 nvlist_free(spa->spa_config_splitting);
2308 spa->spa_config_splitting = NULL;
2312 * Initialize internal SPA structures.
2314 spa->spa_state = POOL_STATE_ACTIVE;
2315 spa->spa_ubsync = spa->spa_uberblock;
2316 spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2317 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2318 spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2319 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2320 spa->spa_claim_max_txg = spa->spa_first_txg;
2321 spa->spa_prev_software_version = ub->ub_software_version;
2323 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2325 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2326 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2328 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2329 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2331 if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2332 boolean_t missing_feat_read = B_FALSE;
2333 nvlist_t *unsup_feat, *enabled_feat;
2335 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2336 &spa->spa_feat_for_read_obj) != 0) {
2337 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2340 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2341 &spa->spa_feat_for_write_obj) != 0) {
2342 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2345 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2346 &spa->spa_feat_desc_obj) != 0) {
2347 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2350 enabled_feat = fnvlist_alloc();
2351 unsup_feat = fnvlist_alloc();
2353 if (!feature_is_supported(spa->spa_meta_objset,
2354 spa->spa_feat_for_read_obj, spa->spa_feat_desc_obj,
2355 unsup_feat, enabled_feat))
2356 missing_feat_read = B_TRUE;
2358 if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) {
2359 if (!feature_is_supported(spa->spa_meta_objset,
2360 spa->spa_feat_for_write_obj, spa->spa_feat_desc_obj,
2361 unsup_feat, enabled_feat)) {
2362 missing_feat_write = B_TRUE;
2366 fnvlist_add_nvlist(spa->spa_load_info,
2367 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
2369 if (!nvlist_empty(unsup_feat)) {
2370 fnvlist_add_nvlist(spa->spa_load_info,
2371 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
2374 fnvlist_free(enabled_feat);
2375 fnvlist_free(unsup_feat);
2377 if (!missing_feat_read) {
2378 fnvlist_add_boolean(spa->spa_load_info,
2379 ZPOOL_CONFIG_CAN_RDONLY);
2383 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2384 * twofold: to determine whether the pool is available for
2385 * import in read-write mode and (if it is not) whether the
2386 * pool is available for import in read-only mode. If the pool
2387 * is available for import in read-write mode, it is displayed
2388 * as available in userland; if it is not available for import
2389 * in read-only mode, it is displayed as unavailable in
2390 * userland. If the pool is available for import in read-only
2391 * mode but not read-write mode, it is displayed as unavailable
2392 * in userland with a special note that the pool is actually
2393 * available for open in read-only mode.
2395 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2396 * missing a feature for write, we must first determine whether
2397 * the pool can be opened read-only before returning to
2398 * userland in order to know whether to display the
2399 * abovementioned note.
2401 if (missing_feat_read || (missing_feat_write &&
2402 spa_writeable(spa))) {
2403 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2408 spa->spa_is_initializing = B_TRUE;
2409 error = dsl_pool_open(spa->spa_dsl_pool);
2410 spa->spa_is_initializing = B_FALSE;
2412 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2416 nvlist_t *policy = NULL, *nvconfig;
2418 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2419 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2421 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2422 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2424 unsigned long myhostid = 0;
2426 VERIFY(nvlist_lookup_string(nvconfig,
2427 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2430 myhostid = zone_get_hostid(NULL);
2433 * We're emulating the system's hostid in userland, so
2434 * we can't use zone_get_hostid().
2436 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2437 #endif /* _KERNEL */
2438 if (check_hostid && hostid != 0 && myhostid != 0 &&
2439 hostid != myhostid) {
2440 nvlist_free(nvconfig);
2441 cmn_err(CE_WARN, "pool '%s' could not be "
2442 "loaded as it was last accessed by "
2443 "another system (host: %s hostid: 0x%lx). "
2444 "See: http://illumos.org/msg/ZFS-8000-EY",
2445 spa_name(spa), hostname,
2446 (unsigned long)hostid);
2447 return (SET_ERROR(EBADF));
2450 if (nvlist_lookup_nvlist(spa->spa_config,
2451 ZPOOL_REWIND_POLICY, &policy) == 0)
2452 VERIFY(nvlist_add_nvlist(nvconfig,
2453 ZPOOL_REWIND_POLICY, policy) == 0);
2455 spa_config_set(spa, nvconfig);
2457 spa_deactivate(spa);
2458 spa_activate(spa, orig_mode);
2460 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2463 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2464 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2465 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2467 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2470 * Load the bit that tells us to use the new accounting function
2471 * (raid-z deflation). If we have an older pool, this will not
2474 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2475 if (error != 0 && error != ENOENT)
2476 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2478 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2479 &spa->spa_creation_version);
2480 if (error != 0 && error != ENOENT)
2481 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2484 * Load the persistent error log. If we have an older pool, this will
2487 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2488 if (error != 0 && error != ENOENT)
2489 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2491 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2492 &spa->spa_errlog_scrub);
2493 if (error != 0 && error != ENOENT)
2494 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2497 * Load the history object. If we have an older pool, this
2498 * will not be present.
2500 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2501 if (error != 0 && error != ENOENT)
2502 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2505 * If we're assembling the pool from the split-off vdevs of
2506 * an existing pool, we don't want to attach the spares & cache
2511 * Load any hot spares for this pool.
2513 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2514 if (error != 0 && error != ENOENT)
2515 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2516 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2517 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2518 if (load_nvlist(spa, spa->spa_spares.sav_object,
2519 &spa->spa_spares.sav_config) != 0)
2520 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2522 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2523 spa_load_spares(spa);
2524 spa_config_exit(spa, SCL_ALL, FTAG);
2525 } else if (error == 0) {
2526 spa->spa_spares.sav_sync = B_TRUE;
2530 * Load any level 2 ARC devices for this pool.
2532 error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2533 &spa->spa_l2cache.sav_object);
2534 if (error != 0 && error != ENOENT)
2535 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2536 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2537 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2538 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2539 &spa->spa_l2cache.sav_config) != 0)
2540 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2542 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2543 spa_load_l2cache(spa);
2544 spa_config_exit(spa, SCL_ALL, FTAG);
2545 } else if (error == 0) {
2546 spa->spa_l2cache.sav_sync = B_TRUE;
2549 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2551 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2552 if (error && error != ENOENT)
2553 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2556 uint64_t autoreplace;
2558 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2559 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2560 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2561 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2562 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2563 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2564 &spa->spa_dedup_ditto);
2566 spa->spa_autoreplace = (autoreplace != 0);
2570 * If the 'autoreplace' property is set, then post a resource notifying
2571 * the ZFS DE that it should not issue any faults for unopenable
2572 * devices. We also iterate over the vdevs, and post a sysevent for any
2573 * unopenable vdevs so that the normal autoreplace handler can take
2576 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2577 spa_check_removed(spa->spa_root_vdev);
2579 * For the import case, this is done in spa_import(), because
2580 * at this point we're using the spare definitions from
2581 * the MOS config, not necessarily from the userland config.
2583 if (state != SPA_LOAD_IMPORT) {
2584 spa_aux_check_removed(&spa->spa_spares);
2585 spa_aux_check_removed(&spa->spa_l2cache);
2590 * Load the vdev state for all toplevel vdevs.
2595 * Propagate the leaf DTLs we just loaded all the way up the tree.
2597 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2598 vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2599 spa_config_exit(spa, SCL_ALL, FTAG);
2602 * Load the DDTs (dedup tables).
2604 error = ddt_load(spa);
2606 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2608 spa_update_dspace(spa);
2611 * Validate the config, using the MOS config to fill in any
2612 * information which might be missing. If we fail to validate
2613 * the config then declare the pool unfit for use. If we're
2614 * assembling a pool from a split, the log is not transferred
2617 if (type != SPA_IMPORT_ASSEMBLE) {
2620 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2621 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2623 if (!spa_config_valid(spa, nvconfig)) {
2624 nvlist_free(nvconfig);
2625 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2628 nvlist_free(nvconfig);
2631 * Now that we've validated the config, check the state of the
2632 * root vdev. If it can't be opened, it indicates one or
2633 * more toplevel vdevs are faulted.
2635 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2636 return (SET_ERROR(ENXIO));
2638 if (spa_check_logs(spa)) {
2639 *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2640 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2644 if (missing_feat_write) {
2645 ASSERT(state == SPA_LOAD_TRYIMPORT);
2648 * At this point, we know that we can open the pool in
2649 * read-only mode but not read-write mode. We now have enough
2650 * information and can return to userland.
2652 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
2656 * We've successfully opened the pool, verify that we're ready
2657 * to start pushing transactions.
2659 if (state != SPA_LOAD_TRYIMPORT) {
2660 if (error = spa_load_verify(spa))
2661 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2665 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2666 spa->spa_load_max_txg == UINT64_MAX)) {
2668 int need_update = B_FALSE;
2670 ASSERT(state != SPA_LOAD_TRYIMPORT);
2673 * Claim log blocks that haven't been committed yet.
2674 * This must all happen in a single txg.
2675 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2676 * invoked from zil_claim_log_block()'s i/o done callback.
2677 * Price of rollback is that we abandon the log.
2679 spa->spa_claiming = B_TRUE;
2681 tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2682 spa_first_txg(spa));
2683 (void) dmu_objset_find(spa_name(spa),
2684 zil_claim, tx, DS_FIND_CHILDREN);
2687 spa->spa_claiming = B_FALSE;
2689 spa_set_log_state(spa, SPA_LOG_GOOD);
2690 spa->spa_sync_on = B_TRUE;
2691 txg_sync_start(spa->spa_dsl_pool);
2694 * Wait for all claims to sync. We sync up to the highest
2695 * claimed log block birth time so that claimed log blocks
2696 * don't appear to be from the future. spa_claim_max_txg
2697 * will have been set for us by either zil_check_log_chain()
2698 * (invoked from spa_check_logs()) or zil_claim() above.
2700 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2703 * If the config cache is stale, or we have uninitialized
2704 * metaslabs (see spa_vdev_add()), then update the config.
2706 * If this is a verbatim import, trust the current
2707 * in-core spa_config and update the disk labels.
2709 if (config_cache_txg != spa->spa_config_txg ||
2710 state == SPA_LOAD_IMPORT ||
2711 state == SPA_LOAD_RECOVER ||
2712 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2713 need_update = B_TRUE;
2715 for (int c = 0; c < rvd->vdev_children; c++)
2716 if (rvd->vdev_child[c]->vdev_ms_array == 0)
2717 need_update = B_TRUE;
2720 * Update the config cache asychronously in case we're the
2721 * root pool, in which case the config cache isn't writable yet.
2724 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2727 * Check all DTLs to see if anything needs resilvering.
2729 if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2730 vdev_resilver_needed(rvd, NULL, NULL))
2731 spa_async_request(spa, SPA_ASYNC_RESILVER);
2734 * Log the fact that we booted up (so that we can detect if
2735 * we rebooted in the middle of an operation).
2737 spa_history_log_version(spa, "open");
2740 * Delete any inconsistent datasets.
2742 (void) dmu_objset_find(spa_name(spa),
2743 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2746 * Clean up any stale temporary dataset userrefs.
2748 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2755 spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2757 int mode = spa->spa_mode;
2760 spa_deactivate(spa);
2762 spa->spa_load_max_txg--;
2764 spa_activate(spa, mode);
2765 spa_async_suspend(spa);
2767 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2771 * If spa_load() fails this function will try loading prior txg's. If
2772 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2773 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2774 * function will not rewind the pool and will return the same error as
2778 spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2779 uint64_t max_request, int rewind_flags)
2781 nvlist_t *loadinfo = NULL;
2782 nvlist_t *config = NULL;
2783 int load_error, rewind_error;
2784 uint64_t safe_rewind_txg;
2787 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2788 spa->spa_load_max_txg = spa->spa_load_txg;
2789 spa_set_log_state(spa, SPA_LOG_CLEAR);
2791 spa->spa_load_max_txg = max_request;
2794 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2796 if (load_error == 0)
2799 if (spa->spa_root_vdev != NULL)
2800 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2802 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2803 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2805 if (rewind_flags & ZPOOL_NEVER_REWIND) {
2806 nvlist_free(config);
2807 return (load_error);
2810 if (state == SPA_LOAD_RECOVER) {
2811 /* Price of rolling back is discarding txgs, including log */
2812 spa_set_log_state(spa, SPA_LOG_CLEAR);
2815 * If we aren't rolling back save the load info from our first
2816 * import attempt so that we can restore it after attempting
2819 loadinfo = spa->spa_load_info;
2820 spa->spa_load_info = fnvlist_alloc();
2823 spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2824 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2825 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2826 TXG_INITIAL : safe_rewind_txg;
2829 * Continue as long as we're finding errors, we're still within
2830 * the acceptable rewind range, and we're still finding uberblocks
2832 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2833 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2834 if (spa->spa_load_max_txg < safe_rewind_txg)
2835 spa->spa_extreme_rewind = B_TRUE;
2836 rewind_error = spa_load_retry(spa, state, mosconfig);
2839 spa->spa_extreme_rewind = B_FALSE;
2840 spa->spa_load_max_txg = UINT64_MAX;
2842 if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2843 spa_config_set(spa, config);
2845 if (state == SPA_LOAD_RECOVER) {
2846 ASSERT3P(loadinfo, ==, NULL);
2847 return (rewind_error);
2849 /* Store the rewind info as part of the initial load info */
2850 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
2851 spa->spa_load_info);
2853 /* Restore the initial load info */
2854 fnvlist_free(spa->spa_load_info);
2855 spa->spa_load_info = loadinfo;
2857 return (load_error);
2864 * The import case is identical to an open except that the configuration is sent
2865 * down from userland, instead of grabbed from the configuration cache. For the
2866 * case of an open, the pool configuration will exist in the
2867 * POOL_STATE_UNINITIALIZED state.
2869 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2870 * the same time open the pool, without having to keep around the spa_t in some
2874 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2878 spa_load_state_t state = SPA_LOAD_OPEN;
2880 int locked = B_FALSE;
2881 int firstopen = B_FALSE;
2886 * As disgusting as this is, we need to support recursive calls to this
2887 * function because dsl_dir_open() is called during spa_load(), and ends
2888 * up calling spa_open() again. The real fix is to figure out how to
2889 * avoid dsl_dir_open() calling this in the first place.
2891 if (mutex_owner(&spa_namespace_lock) != curthread) {
2892 mutex_enter(&spa_namespace_lock);
2896 if ((spa = spa_lookup(pool)) == NULL) {
2898 mutex_exit(&spa_namespace_lock);
2899 return (SET_ERROR(ENOENT));
2902 if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2903 zpool_rewind_policy_t policy;
2907 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2909 if (policy.zrp_request & ZPOOL_DO_REWIND)
2910 state = SPA_LOAD_RECOVER;
2912 spa_activate(spa, spa_mode_global);
2914 if (state != SPA_LOAD_RECOVER)
2915 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2917 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2918 policy.zrp_request);
2920 if (error == EBADF) {
2922 * If vdev_validate() returns failure (indicated by
2923 * EBADF), it indicates that one of the vdevs indicates
2924 * that the pool has been exported or destroyed. If
2925 * this is the case, the config cache is out of sync and
2926 * we should remove the pool from the namespace.
2929 spa_deactivate(spa);
2930 spa_config_sync(spa, B_TRUE, B_TRUE);
2933 mutex_exit(&spa_namespace_lock);
2934 return (SET_ERROR(ENOENT));
2939 * We can't open the pool, but we still have useful
2940 * information: the state of each vdev after the
2941 * attempted vdev_open(). Return this to the user.
2943 if (config != NULL && spa->spa_config) {
2944 VERIFY(nvlist_dup(spa->spa_config, config,
2946 VERIFY(nvlist_add_nvlist(*config,
2947 ZPOOL_CONFIG_LOAD_INFO,
2948 spa->spa_load_info) == 0);
2951 spa_deactivate(spa);
2952 spa->spa_last_open_failed = error;
2954 mutex_exit(&spa_namespace_lock);
2960 spa_open_ref(spa, tag);
2963 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2966 * If we've recovered the pool, pass back any information we
2967 * gathered while doing the load.
2969 if (state == SPA_LOAD_RECOVER) {
2970 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
2971 spa->spa_load_info) == 0);
2975 spa->spa_last_open_failed = 0;
2976 spa->spa_last_ubsync_txg = 0;
2977 spa->spa_load_txg = 0;
2978 mutex_exit(&spa_namespace_lock);
2982 zvol_create_minors(spa->spa_name);
2993 spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
2996 return (spa_open_common(name, spapp, tag, policy, config));
3000 spa_open(const char *name, spa_t **spapp, void *tag)
3002 return (spa_open_common(name, spapp, tag, NULL, NULL));
3006 * Lookup the given spa_t, incrementing the inject count in the process,
3007 * preventing it from being exported or destroyed.
3010 spa_inject_addref(char *name)
3014 mutex_enter(&spa_namespace_lock);
3015 if ((spa = spa_lookup(name)) == NULL) {
3016 mutex_exit(&spa_namespace_lock);
3019 spa->spa_inject_ref++;
3020 mutex_exit(&spa_namespace_lock);
3026 spa_inject_delref(spa_t *spa)
3028 mutex_enter(&spa_namespace_lock);
3029 spa->spa_inject_ref--;
3030 mutex_exit(&spa_namespace_lock);
3034 * Add spares device information to the nvlist.
3037 spa_add_spares(spa_t *spa, nvlist_t *config)
3047 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3049 if (spa->spa_spares.sav_count == 0)
3052 VERIFY(nvlist_lookup_nvlist(config,
3053 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3054 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
3055 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3057 VERIFY(nvlist_add_nvlist_array(nvroot,
3058 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3059 VERIFY(nvlist_lookup_nvlist_array(nvroot,
3060 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3063 * Go through and find any spares which have since been
3064 * repurposed as an active spare. If this is the case, update
3065 * their status appropriately.
3067 for (i = 0; i < nspares; i++) {
3068 VERIFY(nvlist_lookup_uint64(spares[i],
3069 ZPOOL_CONFIG_GUID, &guid) == 0);
3070 if (spa_spare_exists(guid, &pool, NULL) &&
3072 VERIFY(nvlist_lookup_uint64_array(
3073 spares[i], ZPOOL_CONFIG_VDEV_STATS,
3074 (uint64_t **)&vs, &vsc) == 0);
3075 vs->vs_state = VDEV_STATE_CANT_OPEN;
3076 vs->vs_aux = VDEV_AUX_SPARED;
3083 * Add l2cache device information to the nvlist, including vdev stats.
3086 spa_add_l2cache(spa_t *spa, nvlist_t *config)
3089 uint_t i, j, nl2cache;
3096 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3098 if (spa->spa_l2cache.sav_count == 0)
3101 VERIFY(nvlist_lookup_nvlist(config,
3102 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3103 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
3104 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3105 if (nl2cache != 0) {
3106 VERIFY(nvlist_add_nvlist_array(nvroot,
3107 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3108 VERIFY(nvlist_lookup_nvlist_array(nvroot,
3109 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3112 * Update level 2 cache device stats.
3115 for (i = 0; i < nl2cache; i++) {
3116 VERIFY(nvlist_lookup_uint64(l2cache[i],
3117 ZPOOL_CONFIG_GUID, &guid) == 0);
3120 for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
3122 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
3123 vd = spa->spa_l2cache.sav_vdevs[j];
3129 VERIFY(nvlist_lookup_uint64_array(l2cache[i],
3130 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
3132 vdev_get_stats(vd, vs);
3138 spa_add_feature_stats(spa_t *spa, nvlist_t *config)
3144 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3145 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3147 /* We may be unable to read features if pool is suspended. */
3148 if (spa_suspended(spa))
3151 if (spa->spa_feat_for_read_obj != 0) {
3152 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3153 spa->spa_feat_for_read_obj);
3154 zap_cursor_retrieve(&zc, &za) == 0;
3155 zap_cursor_advance(&zc)) {
3156 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3157 za.za_num_integers == 1);
3158 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3159 za.za_first_integer));
3161 zap_cursor_fini(&zc);
3164 if (spa->spa_feat_for_write_obj != 0) {
3165 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3166 spa->spa_feat_for_write_obj);
3167 zap_cursor_retrieve(&zc, &za) == 0;
3168 zap_cursor_advance(&zc)) {
3169 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3170 za.za_num_integers == 1);
3171 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3172 za.za_first_integer));
3174 zap_cursor_fini(&zc);
3178 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
3180 nvlist_free(features);
3184 spa_get_stats(const char *name, nvlist_t **config,
3185 char *altroot, size_t buflen)
3191 error = spa_open_common(name, &spa, FTAG, NULL, config);
3195 * This still leaves a window of inconsistency where the spares
3196 * or l2cache devices could change and the config would be
3197 * self-inconsistent.
3199 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3201 if (*config != NULL) {
3202 uint64_t loadtimes[2];
3204 loadtimes[0] = spa->spa_loaded_ts.tv_sec;
3205 loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
3206 VERIFY(nvlist_add_uint64_array(*config,
3207 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
3209 VERIFY(nvlist_add_uint64(*config,
3210 ZPOOL_CONFIG_ERRCOUNT,
3211 spa_get_errlog_size(spa)) == 0);
3213 if (spa_suspended(spa))
3214 VERIFY(nvlist_add_uint64(*config,
3215 ZPOOL_CONFIG_SUSPENDED,
3216 spa->spa_failmode) == 0);
3218 spa_add_spares(spa, *config);
3219 spa_add_l2cache(spa, *config);
3220 spa_add_feature_stats(spa, *config);
3225 * We want to get the alternate root even for faulted pools, so we cheat
3226 * and call spa_lookup() directly.
3230 mutex_enter(&spa_namespace_lock);
3231 spa = spa_lookup(name);
3233 spa_altroot(spa, altroot, buflen);
3237 mutex_exit(&spa_namespace_lock);
3239 spa_altroot(spa, altroot, buflen);
3244 spa_config_exit(spa, SCL_CONFIG, FTAG);
3245 spa_close(spa, FTAG);
3252 * Validate that the auxiliary device array is well formed. We must have an
3253 * array of nvlists, each which describes a valid leaf vdev. If this is an
3254 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3255 * specified, as long as they are well-formed.
3258 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
3259 spa_aux_vdev_t *sav, const char *config, uint64_t version,
3260 vdev_labeltype_t label)
3267 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3270 * It's acceptable to have no devs specified.
3272 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
3276 return (SET_ERROR(EINVAL));
3279 * Make sure the pool is formatted with a version that supports this
3282 if (spa_version(spa) < version)
3283 return (SET_ERROR(ENOTSUP));
3286 * Set the pending device list so we correctly handle device in-use
3289 sav->sav_pending = dev;
3290 sav->sav_npending = ndev;
3292 for (i = 0; i < ndev; i++) {
3293 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
3297 if (!vd->vdev_ops->vdev_op_leaf) {
3299 error = SET_ERROR(EINVAL);
3304 * The L2ARC currently only supports disk devices in
3305 * kernel context. For user-level testing, we allow it.
3308 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
3309 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
3310 error = SET_ERROR(ENOTBLK);
3317 if ((error = vdev_open(vd)) == 0 &&
3318 (error = vdev_label_init(vd, crtxg, label)) == 0) {
3319 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
3320 vd->vdev_guid) == 0);
3326 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
3333 sav->sav_pending = NULL;
3334 sav->sav_npending = 0;
3339 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
3343 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3345 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3346 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
3347 VDEV_LABEL_SPARE)) != 0) {
3351 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3352 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
3353 VDEV_LABEL_L2CACHE));
3357 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
3362 if (sav->sav_config != NULL) {
3368 * Generate new dev list by concatentating with the
3371 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
3372 &olddevs, &oldndevs) == 0);
3374 newdevs = kmem_alloc(sizeof (void *) *
3375 (ndevs + oldndevs), KM_SLEEP);
3376 for (i = 0; i < oldndevs; i++)
3377 VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
3379 for (i = 0; i < ndevs; i++)
3380 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
3383 VERIFY(nvlist_remove(sav->sav_config, config,
3384 DATA_TYPE_NVLIST_ARRAY) == 0);
3386 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3387 config, newdevs, ndevs + oldndevs) == 0);
3388 for (i = 0; i < oldndevs + ndevs; i++)
3389 nvlist_free(newdevs[i]);
3390 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
3393 * Generate a new dev list.
3395 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
3397 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
3403 * Stop and drop level 2 ARC devices
3406 spa_l2cache_drop(spa_t *spa)
3410 spa_aux_vdev_t *sav = &spa->spa_l2cache;
3412 for (i = 0; i < sav->sav_count; i++) {
3415 vd = sav->sav_vdevs[i];
3418 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
3419 pool != 0ULL && l2arc_vdev_present(vd))
3420 l2arc_remove_vdev(vd);
3428 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
3432 char *altroot = NULL;
3437 uint64_t txg = TXG_INITIAL;
3438 nvlist_t **spares, **l2cache;
3439 uint_t nspares, nl2cache;
3440 uint64_t version, obj;
3441 boolean_t has_features;
3444 * If this pool already exists, return failure.
3446 mutex_enter(&spa_namespace_lock);
3447 if (spa_lookup(pool) != NULL) {
3448 mutex_exit(&spa_namespace_lock);
3449 return (SET_ERROR(EEXIST));
3453 * Allocate a new spa_t structure.
3455 (void) nvlist_lookup_string(props,
3456 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3457 spa = spa_add(pool, NULL, altroot);
3458 spa_activate(spa, spa_mode_global);
3460 if (props && (error = spa_prop_validate(spa, props))) {
3461 spa_deactivate(spa);
3463 mutex_exit(&spa_namespace_lock);
3467 has_features = B_FALSE;
3468 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
3469 elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
3470 if (zpool_prop_feature(nvpair_name(elem)))
3471 has_features = B_TRUE;
3474 if (has_features || nvlist_lookup_uint64(props,
3475 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
3476 version = SPA_VERSION;
3478 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
3480 spa->spa_first_txg = txg;
3481 spa->spa_uberblock.ub_txg = txg - 1;
3482 spa->spa_uberblock.ub_version = version;
3483 spa->spa_ubsync = spa->spa_uberblock;
3486 * Create "The Godfather" zio to hold all async IOs
3488 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
3489 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
3492 * Create the root vdev.
3494 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3496 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3498 ASSERT(error != 0 || rvd != NULL);
3499 ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3501 if (error == 0 && !zfs_allocatable_devs(nvroot))
3502 error = SET_ERROR(EINVAL);
3505 (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3506 (error = spa_validate_aux(spa, nvroot, txg,
3507 VDEV_ALLOC_ADD)) == 0) {
3508 for (int c = 0; c < rvd->vdev_children; c++) {
3509 vdev_ashift_optimize(rvd->vdev_child[c]);
3510 vdev_metaslab_set_size(rvd->vdev_child[c]);
3511 vdev_expand(rvd->vdev_child[c], txg);
3515 spa_config_exit(spa, SCL_ALL, FTAG);
3519 spa_deactivate(spa);
3521 mutex_exit(&spa_namespace_lock);
3526 * Get the list of spares, if specified.
3528 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3529 &spares, &nspares) == 0) {
3530 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3532 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3533 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3534 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3535 spa_load_spares(spa);
3536 spa_config_exit(spa, SCL_ALL, FTAG);
3537 spa->spa_spares.sav_sync = B_TRUE;
3541 * Get the list of level 2 cache devices, if specified.
3543 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3544 &l2cache, &nl2cache) == 0) {
3545 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3546 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3547 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3548 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3549 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3550 spa_load_l2cache(spa);
3551 spa_config_exit(spa, SCL_ALL, FTAG);
3552 spa->spa_l2cache.sav_sync = B_TRUE;
3555 spa->spa_is_initializing = B_TRUE;
3556 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3557 spa->spa_meta_objset = dp->dp_meta_objset;
3558 spa->spa_is_initializing = B_FALSE;
3561 * Create DDTs (dedup tables).
3565 spa_update_dspace(spa);
3567 tx = dmu_tx_create_assigned(dp, txg);
3570 * Create the pool config object.
3572 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3573 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3574 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3576 if (zap_add(spa->spa_meta_objset,
3577 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3578 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3579 cmn_err(CE_PANIC, "failed to add pool config");
3582 if (spa_version(spa) >= SPA_VERSION_FEATURES)
3583 spa_feature_create_zap_objects(spa, tx);
3585 if (zap_add(spa->spa_meta_objset,
3586 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3587 sizeof (uint64_t), 1, &version, tx) != 0) {
3588 cmn_err(CE_PANIC, "failed to add pool version");
3591 /* Newly created pools with the right version are always deflated. */
3592 if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3593 spa->spa_deflate = TRUE;
3594 if (zap_add(spa->spa_meta_objset,
3595 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3596 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3597 cmn_err(CE_PANIC, "failed to add deflate");
3602 * Create the deferred-free bpobj. Turn off compression
3603 * because sync-to-convergence takes longer if the blocksize
3606 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3607 dmu_object_set_compress(spa->spa_meta_objset, obj,
3608 ZIO_COMPRESS_OFF, tx);
3609 if (zap_add(spa->spa_meta_objset,
3610 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3611 sizeof (uint64_t), 1, &obj, tx) != 0) {
3612 cmn_err(CE_PANIC, "failed to add bpobj");
3614 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3615 spa->spa_meta_objset, obj));
3618 * Create the pool's history object.
3620 if (version >= SPA_VERSION_ZPOOL_HISTORY)
3621 spa_history_create_obj(spa, tx);
3624 * Set pool properties.
3626 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3627 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3628 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3629 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3631 if (props != NULL) {
3632 spa_configfile_set(spa, props, B_FALSE);
3633 spa_sync_props(props, tx);
3638 spa->spa_sync_on = B_TRUE;
3639 txg_sync_start(spa->spa_dsl_pool);
3642 * We explicitly wait for the first transaction to complete so that our
3643 * bean counters are appropriately updated.
3645 txg_wait_synced(spa->spa_dsl_pool, txg);
3647 spa_config_sync(spa, B_FALSE, B_TRUE);
3649 spa_history_log_version(spa, "create");
3651 spa->spa_minref = refcount_count(&spa->spa_refcount);
3653 mutex_exit(&spa_namespace_lock);
3661 * Get the root pool information from the root disk, then import the root pool
3662 * during the system boot up time.
3664 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3667 spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3670 nvlist_t *nvtop, *nvroot;
3673 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3677 * Add this top-level vdev to the child array.
3679 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3681 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3683 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3686 * Put this pool's top-level vdevs into a root vdev.
3688 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3689 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3690 VDEV_TYPE_ROOT) == 0);
3691 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3692 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3693 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3697 * Replace the existing vdev_tree with the new root vdev in
3698 * this pool's configuration (remove the old, add the new).
3700 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3701 nvlist_free(nvroot);
3706 * Walk the vdev tree and see if we can find a device with "better"
3707 * configuration. A configuration is "better" if the label on that
3708 * device has a more recent txg.
3711 spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3713 for (int c = 0; c < vd->vdev_children; c++)
3714 spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3716 if (vd->vdev_ops->vdev_op_leaf) {
3720 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3724 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3728 * Do we have a better boot device?
3730 if (label_txg > *txg) {
3739 * Import a root pool.
3741 * For x86. devpath_list will consist of devid and/or physpath name of
3742 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3743 * The GRUB "findroot" command will return the vdev we should boot.
3745 * For Sparc, devpath_list consists the physpath name of the booting device
3746 * no matter the rootpool is a single device pool or a mirrored pool.
3748 * "/pci@1f,0/ide@d/disk@0,0:a"
3751 spa_import_rootpool(char *devpath, char *devid)
3754 vdev_t *rvd, *bvd, *avd = NULL;
3755 nvlist_t *config, *nvtop;
3761 * Read the label from the boot device and generate a configuration.
3763 config = spa_generate_rootconf(devpath, devid, &guid);
3764 #if defined(_OBP) && defined(_KERNEL)
3765 if (config == NULL) {
3766 if (strstr(devpath, "/iscsi/ssd") != NULL) {
3768 get_iscsi_bootpath_phy(devpath);
3769 config = spa_generate_rootconf(devpath, devid, &guid);
3773 if (config == NULL) {
3774 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
3776 return (SET_ERROR(EIO));
3779 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3781 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3783 mutex_enter(&spa_namespace_lock);
3784 if ((spa = spa_lookup(pname)) != NULL) {
3786 * Remove the existing root pool from the namespace so that we
3787 * can replace it with the correct config we just read in.
3792 spa = spa_add(pname, config, NULL);
3793 spa->spa_is_root = B_TRUE;
3794 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3797 * Build up a vdev tree based on the boot device's label config.
3799 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3801 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3802 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3803 VDEV_ALLOC_ROOTPOOL);
3804 spa_config_exit(spa, SCL_ALL, FTAG);
3806 mutex_exit(&spa_namespace_lock);
3807 nvlist_free(config);
3808 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3814 * Get the boot vdev.
3816 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3817 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3818 (u_longlong_t)guid);
3819 error = SET_ERROR(ENOENT);
3824 * Determine if there is a better boot device.
3827 spa_alt_rootvdev(rvd, &avd, &txg);
3829 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3830 "try booting from '%s'", avd->vdev_path);
3831 error = SET_ERROR(EINVAL);
3836 * If the boot device is part of a spare vdev then ensure that
3837 * we're booting off the active spare.
3839 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3840 !bvd->vdev_isspare) {
3841 cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3842 "try booting from '%s'",
3844 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3845 error = SET_ERROR(EINVAL);
3851 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3853 spa_config_exit(spa, SCL_ALL, FTAG);
3854 mutex_exit(&spa_namespace_lock);
3856 nvlist_free(config);
3862 extern int vdev_geom_read_pool_label(const char *name, nvlist_t ***configs,
3866 spa_generate_rootconf(const char *name)
3868 nvlist_t **configs, **tops;
3870 nvlist_t *best_cfg, *nvtop, *nvroot;
3879 if (vdev_geom_read_pool_label(name, &configs, &count) != 0)
3882 ASSERT3U(count, !=, 0);
3884 for (i = 0; i < count; i++) {
3887 VERIFY(nvlist_lookup_uint64(configs[i], ZPOOL_CONFIG_POOL_TXG,
3889 if (txg > best_txg) {
3891 best_cfg = configs[i];
3896 * Multi-vdev root pool configuration discovery is not supported yet.
3899 nvlist_lookup_uint64(best_cfg, ZPOOL_CONFIG_VDEV_CHILDREN, &nchildren);
3901 nvlist_lookup_uint64_array(best_cfg, ZPOOL_CONFIG_HOLE_ARRAY,
3904 tops = kmem_zalloc(nchildren * sizeof(void *), KM_SLEEP);
3905 for (i = 0; i < nchildren; i++) {
3908 if (configs[i] == NULL)
3910 VERIFY(nvlist_lookup_nvlist(configs[i], ZPOOL_CONFIG_VDEV_TREE,
3912 nvlist_dup(nvtop, &tops[i], KM_SLEEP);
3914 for (i = 0; holes != NULL && i < nholes; i++) {
3917 if (tops[holes[i]] != NULL)
3919 nvlist_alloc(&tops[holes[i]], NV_UNIQUE_NAME, KM_SLEEP);
3920 VERIFY(nvlist_add_string(tops[holes[i]], ZPOOL_CONFIG_TYPE,
3921 VDEV_TYPE_HOLE) == 0);
3922 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_ID,
3924 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_GUID,
3927 for (i = 0; i < nchildren; i++) {
3928 if (tops[i] != NULL)
3930 nvlist_alloc(&tops[i], NV_UNIQUE_NAME, KM_SLEEP);
3931 VERIFY(nvlist_add_string(tops[i], ZPOOL_CONFIG_TYPE,
3932 VDEV_TYPE_MISSING) == 0);
3933 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_ID,
3935 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_GUID,
3940 * Create pool config based on the best vdev config.
3942 nvlist_dup(best_cfg, &config, KM_SLEEP);
3945 * Put this pool's top-level vdevs into a root vdev.
3947 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3949 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3950 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3951 VDEV_TYPE_ROOT) == 0);
3952 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3953 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3954 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3955 tops, nchildren) == 0);
3958 * Replace the existing vdev_tree with the new root vdev in
3959 * this pool's configuration (remove the old, add the new).
3961 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3964 * Drop vdev config elements that should not be present at pool level.
3966 nvlist_remove(config, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64);
3967 nvlist_remove(config, ZPOOL_CONFIG_TOP_GUID, DATA_TYPE_UINT64);
3969 for (i = 0; i < count; i++)
3970 nvlist_free(configs[i]);
3971 kmem_free(configs, count * sizeof(void *));
3972 for (i = 0; i < nchildren; i++)
3973 nvlist_free(tops[i]);
3974 kmem_free(tops, nchildren * sizeof(void *));
3975 nvlist_free(nvroot);
3980 spa_import_rootpool(const char *name)
3983 vdev_t *rvd, *bvd, *avd = NULL;
3984 nvlist_t *config, *nvtop;
3990 * Read the label from the boot device and generate a configuration.
3992 config = spa_generate_rootconf(name);
3994 mutex_enter(&spa_namespace_lock);
3995 if (config != NULL) {
3996 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3997 &pname) == 0 && strcmp(name, pname) == 0);
3998 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg)
4001 if ((spa = spa_lookup(pname)) != NULL) {
4003 * Remove the existing root pool from the namespace so
4004 * that we can replace it with the correct config
4009 spa = spa_add(pname, config, NULL);
4012 * Set spa_ubsync.ub_version as it can be used in vdev_alloc()
4013 * via spa_version().
4015 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
4016 &spa->spa_ubsync.ub_version) != 0)
4017 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
4018 } else if ((spa = spa_lookup(name)) == NULL) {
4019 cmn_err(CE_NOTE, "Cannot find the pool label for '%s'",
4023 VERIFY(nvlist_dup(spa->spa_config, &config, KM_SLEEP) == 0);
4025 spa->spa_is_root = B_TRUE;
4026 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
4029 * Build up a vdev tree based on the boot device's label config.
4031 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4033 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4034 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
4035 VDEV_ALLOC_ROOTPOOL);
4036 spa_config_exit(spa, SCL_ALL, FTAG);
4038 mutex_exit(&spa_namespace_lock);
4039 nvlist_free(config);
4040 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
4045 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4047 spa_config_exit(spa, SCL_ALL, FTAG);
4048 mutex_exit(&spa_namespace_lock);
4050 nvlist_free(config);
4058 * Import a non-root pool into the system.
4061 spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
4064 char *altroot = NULL;
4065 spa_load_state_t state = SPA_LOAD_IMPORT;
4066 zpool_rewind_policy_t policy;
4067 uint64_t mode = spa_mode_global;
4068 uint64_t readonly = B_FALSE;
4071 nvlist_t **spares, **l2cache;
4072 uint_t nspares, nl2cache;
4075 * If a pool with this name exists, return failure.
4077 mutex_enter(&spa_namespace_lock);
4078 if (spa_lookup(pool) != NULL) {
4079 mutex_exit(&spa_namespace_lock);
4080 return (SET_ERROR(EEXIST));
4084 * Create and initialize the spa structure.
4086 (void) nvlist_lookup_string(props,
4087 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4088 (void) nvlist_lookup_uint64(props,
4089 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
4092 spa = spa_add(pool, config, altroot);
4093 spa->spa_import_flags = flags;
4096 * Verbatim import - Take a pool and insert it into the namespace
4097 * as if it had been loaded at boot.
4099 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
4101 spa_configfile_set(spa, props, B_FALSE);
4103 spa_config_sync(spa, B_FALSE, B_TRUE);
4105 mutex_exit(&spa_namespace_lock);
4106 spa_history_log_version(spa, "import");
4111 spa_activate(spa, mode);
4114 * Don't start async tasks until we know everything is healthy.
4116 spa_async_suspend(spa);
4118 zpool_get_rewind_policy(config, &policy);
4119 if (policy.zrp_request & ZPOOL_DO_REWIND)
4120 state = SPA_LOAD_RECOVER;
4123 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
4124 * because the user-supplied config is actually the one to trust when
4127 if (state != SPA_LOAD_RECOVER)
4128 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
4130 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
4131 policy.zrp_request);
4134 * Propagate anything learned while loading the pool and pass it
4135 * back to caller (i.e. rewind info, missing devices, etc).
4137 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4138 spa->spa_load_info) == 0);
4140 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4142 * Toss any existing sparelist, as it doesn't have any validity
4143 * anymore, and conflicts with spa_has_spare().
4145 if (spa->spa_spares.sav_config) {
4146 nvlist_free(spa->spa_spares.sav_config);
4147 spa->spa_spares.sav_config = NULL;
4148 spa_load_spares(spa);
4150 if (spa->spa_l2cache.sav_config) {
4151 nvlist_free(spa->spa_l2cache.sav_config);
4152 spa->spa_l2cache.sav_config = NULL;
4153 spa_load_l2cache(spa);
4156 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4159 error = spa_validate_aux(spa, nvroot, -1ULL,
4162 error = spa_validate_aux(spa, nvroot, -1ULL,
4163 VDEV_ALLOC_L2CACHE);
4164 spa_config_exit(spa, SCL_ALL, FTAG);
4167 spa_configfile_set(spa, props, B_FALSE);
4169 if (error != 0 || (props && spa_writeable(spa) &&
4170 (error = spa_prop_set(spa, props)))) {
4172 spa_deactivate(spa);
4174 mutex_exit(&spa_namespace_lock);
4178 spa_async_resume(spa);
4181 * Override any spares and level 2 cache devices as specified by
4182 * the user, as these may have correct device names/devids, etc.
4184 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
4185 &spares, &nspares) == 0) {
4186 if (spa->spa_spares.sav_config)
4187 VERIFY(nvlist_remove(spa->spa_spares.sav_config,
4188 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
4190 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
4191 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4192 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
4193 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
4194 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4195 spa_load_spares(spa);
4196 spa_config_exit(spa, SCL_ALL, FTAG);
4197 spa->spa_spares.sav_sync = B_TRUE;
4199 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
4200 &l2cache, &nl2cache) == 0) {
4201 if (spa->spa_l2cache.sav_config)
4202 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
4203 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
4205 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
4206 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4207 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
4208 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
4209 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4210 spa_load_l2cache(spa);
4211 spa_config_exit(spa, SCL_ALL, FTAG);
4212 spa->spa_l2cache.sav_sync = B_TRUE;
4216 * Check for any removed devices.
4218 if (spa->spa_autoreplace) {
4219 spa_aux_check_removed(&spa->spa_spares);
4220 spa_aux_check_removed(&spa->spa_l2cache);
4223 if (spa_writeable(spa)) {
4225 * Update the config cache to include the newly-imported pool.
4227 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4231 * It's possible that the pool was expanded while it was exported.
4232 * We kick off an async task to handle this for us.
4234 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
4236 mutex_exit(&spa_namespace_lock);
4237 spa_history_log_version(spa, "import");
4241 zvol_create_minors(pool);
4248 spa_tryimport(nvlist_t *tryconfig)
4250 nvlist_t *config = NULL;
4256 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
4259 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
4263 * Create and initialize the spa structure.
4265 mutex_enter(&spa_namespace_lock);
4266 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
4267 spa_activate(spa, FREAD);
4270 * Pass off the heavy lifting to spa_load().
4271 * Pass TRUE for mosconfig because the user-supplied config
4272 * is actually the one to trust when doing an import.
4274 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
4277 * If 'tryconfig' was at least parsable, return the current config.
4279 if (spa->spa_root_vdev != NULL) {
4280 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4281 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
4283 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4285 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
4286 spa->spa_uberblock.ub_timestamp) == 0);
4287 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4288 spa->spa_load_info) == 0);
4291 * If the bootfs property exists on this pool then we
4292 * copy it out so that external consumers can tell which
4293 * pools are bootable.
4295 if ((!error || error == EEXIST) && spa->spa_bootfs) {
4296 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4299 * We have to play games with the name since the
4300 * pool was opened as TRYIMPORT_NAME.
4302 if (dsl_dsobj_to_dsname(spa_name(spa),
4303 spa->spa_bootfs, tmpname) == 0) {
4305 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4307 cp = strchr(tmpname, '/');
4309 (void) strlcpy(dsname, tmpname,
4312 (void) snprintf(dsname, MAXPATHLEN,
4313 "%s/%s", poolname, ++cp);
4315 VERIFY(nvlist_add_string(config,
4316 ZPOOL_CONFIG_BOOTFS, dsname) == 0);
4317 kmem_free(dsname, MAXPATHLEN);
4319 kmem_free(tmpname, MAXPATHLEN);
4323 * Add the list of hot spares and level 2 cache devices.
4325 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4326 spa_add_spares(spa, config);
4327 spa_add_l2cache(spa, config);
4328 spa_config_exit(spa, SCL_CONFIG, FTAG);
4332 spa_deactivate(spa);
4334 mutex_exit(&spa_namespace_lock);
4340 * Pool export/destroy
4342 * The act of destroying or exporting a pool is very simple. We make sure there
4343 * is no more pending I/O and any references to the pool are gone. Then, we
4344 * update the pool state and sync all the labels to disk, removing the
4345 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4346 * we don't sync the labels or remove the configuration cache.
4349 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
4350 boolean_t force, boolean_t hardforce)
4357 if (!(spa_mode_global & FWRITE))
4358 return (SET_ERROR(EROFS));
4360 mutex_enter(&spa_namespace_lock);
4361 if ((spa = spa_lookup(pool)) == NULL) {
4362 mutex_exit(&spa_namespace_lock);
4363 return (SET_ERROR(ENOENT));
4367 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4368 * reacquire the namespace lock, and see if we can export.
4370 spa_open_ref(spa, FTAG);
4371 mutex_exit(&spa_namespace_lock);
4372 spa_async_suspend(spa);
4373 mutex_enter(&spa_namespace_lock);
4374 spa_close(spa, FTAG);
4377 * The pool will be in core if it's openable,
4378 * in which case we can modify its state.
4380 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
4382 * Objsets may be open only because they're dirty, so we
4383 * have to force it to sync before checking spa_refcnt.
4385 txg_wait_synced(spa->spa_dsl_pool, 0);
4388 * A pool cannot be exported or destroyed if there are active
4389 * references. If we are resetting a pool, allow references by
4390 * fault injection handlers.
4392 if (!spa_refcount_zero(spa) ||
4393 (spa->spa_inject_ref != 0 &&
4394 new_state != POOL_STATE_UNINITIALIZED)) {
4395 spa_async_resume(spa);
4396 mutex_exit(&spa_namespace_lock);
4397 return (SET_ERROR(EBUSY));
4401 * A pool cannot be exported if it has an active shared spare.
4402 * This is to prevent other pools stealing the active spare
4403 * from an exported pool. At user's own will, such pool can
4404 * be forcedly exported.
4406 if (!force && new_state == POOL_STATE_EXPORTED &&
4407 spa_has_active_shared_spare(spa)) {
4408 spa_async_resume(spa);
4409 mutex_exit(&spa_namespace_lock);
4410 return (SET_ERROR(EXDEV));
4414 * We want this to be reflected on every label,
4415 * so mark them all dirty. spa_unload() will do the
4416 * final sync that pushes these changes out.
4418 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
4419 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4420 spa->spa_state = new_state;
4421 spa->spa_final_txg = spa_last_synced_txg(spa) +
4423 vdev_config_dirty(spa->spa_root_vdev);
4424 spa_config_exit(spa, SCL_ALL, FTAG);
4428 spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
4430 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4432 spa_deactivate(spa);
4435 if (oldconfig && spa->spa_config)
4436 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
4438 if (new_state != POOL_STATE_UNINITIALIZED) {
4440 spa_config_sync(spa, B_TRUE, B_TRUE);
4443 mutex_exit(&spa_namespace_lock);
4449 * Destroy a storage pool.
4452 spa_destroy(char *pool)
4454 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
4459 * Export a storage pool.
4462 spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
4463 boolean_t hardforce)
4465 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
4470 * Similar to spa_export(), this unloads the spa_t without actually removing it
4471 * from the namespace in any way.
4474 spa_reset(char *pool)
4476 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
4481 * ==========================================================================
4482 * Device manipulation
4483 * ==========================================================================
4487 * Add a device to a storage pool.
4490 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
4494 vdev_t *rvd = spa->spa_root_vdev;
4496 nvlist_t **spares, **l2cache;
4497 uint_t nspares, nl2cache;
4499 ASSERT(spa_writeable(spa));
4501 txg = spa_vdev_enter(spa);
4503 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
4504 VDEV_ALLOC_ADD)) != 0)
4505 return (spa_vdev_exit(spa, NULL, txg, error));
4507 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */
4509 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
4513 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
4517 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
4518 return (spa_vdev_exit(spa, vd, txg, EINVAL));
4520 if (vd->vdev_children != 0 &&
4521 (error = vdev_create(vd, txg, B_FALSE)) != 0)
4522 return (spa_vdev_exit(spa, vd, txg, error));
4525 * We must validate the spares and l2cache devices after checking the
4526 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
4528 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
4529 return (spa_vdev_exit(spa, vd, txg, error));
4532 * Transfer each new top-level vdev from vd to rvd.
4534 for (int c = 0; c < vd->vdev_children; c++) {
4537 * Set the vdev id to the first hole, if one exists.
4539 for (id = 0; id < rvd->vdev_children; id++) {
4540 if (rvd->vdev_child[id]->vdev_ishole) {
4541 vdev_free(rvd->vdev_child[id]);
4545 tvd = vd->vdev_child[c];
4546 vdev_remove_child(vd, tvd);
4548 vdev_add_child(rvd, tvd);
4549 vdev_config_dirty(tvd);
4553 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
4554 ZPOOL_CONFIG_SPARES);
4555 spa_load_spares(spa);
4556 spa->spa_spares.sav_sync = B_TRUE;
4559 if (nl2cache != 0) {
4560 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
4561 ZPOOL_CONFIG_L2CACHE);
4562 spa_load_l2cache(spa);
4563 spa->spa_l2cache.sav_sync = B_TRUE;
4567 * We have to be careful when adding new vdevs to an existing pool.
4568 * If other threads start allocating from these vdevs before we
4569 * sync the config cache, and we lose power, then upon reboot we may
4570 * fail to open the pool because there are DVAs that the config cache
4571 * can't translate. Therefore, we first add the vdevs without
4572 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4573 * and then let spa_config_update() initialize the new metaslabs.
4575 * spa_load() checks for added-but-not-initialized vdevs, so that
4576 * if we lose power at any point in this sequence, the remaining
4577 * steps will be completed the next time we load the pool.
4579 (void) spa_vdev_exit(spa, vd, txg, 0);
4581 mutex_enter(&spa_namespace_lock);
4582 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4583 mutex_exit(&spa_namespace_lock);
4589 * Attach a device to a mirror. The arguments are the path to any device
4590 * in the mirror, and the nvroot for the new device. If the path specifies
4591 * a device that is not mirrored, we automatically insert the mirror vdev.
4593 * If 'replacing' is specified, the new device is intended to replace the
4594 * existing device; in this case the two devices are made into their own
4595 * mirror using the 'replacing' vdev, which is functionally identical to
4596 * the mirror vdev (it actually reuses all the same ops) but has a few
4597 * extra rules: you can't attach to it after it's been created, and upon
4598 * completion of resilvering, the first disk (the one being replaced)
4599 * is automatically detached.
4602 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
4604 uint64_t txg, dtl_max_txg;
4605 vdev_t *rvd = spa->spa_root_vdev;
4606 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
4608 char *oldvdpath, *newvdpath;
4612 ASSERT(spa_writeable(spa));
4614 txg = spa_vdev_enter(spa);
4616 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
4619 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4621 if (!oldvd->vdev_ops->vdev_op_leaf)
4622 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4624 pvd = oldvd->vdev_parent;
4626 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
4627 VDEV_ALLOC_ATTACH)) != 0)
4628 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4630 if (newrootvd->vdev_children != 1)
4631 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4633 newvd = newrootvd->vdev_child[0];
4635 if (!newvd->vdev_ops->vdev_op_leaf)
4636 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4638 if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
4639 return (spa_vdev_exit(spa, newrootvd, txg, error));
4642 * Spares can't replace logs
4644 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
4645 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4649 * For attach, the only allowable parent is a mirror or the root
4652 if (pvd->vdev_ops != &vdev_mirror_ops &&
4653 pvd->vdev_ops != &vdev_root_ops)
4654 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4656 pvops = &vdev_mirror_ops;
4659 * Active hot spares can only be replaced by inactive hot
4662 if (pvd->vdev_ops == &vdev_spare_ops &&
4663 oldvd->vdev_isspare &&
4664 !spa_has_spare(spa, newvd->vdev_guid))
4665 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4668 * If the source is a hot spare, and the parent isn't already a
4669 * spare, then we want to create a new hot spare. Otherwise, we
4670 * want to create a replacing vdev. The user is not allowed to
4671 * attach to a spared vdev child unless the 'isspare' state is
4672 * the same (spare replaces spare, non-spare replaces
4675 if (pvd->vdev_ops == &vdev_replacing_ops &&
4676 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
4677 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4678 } else if (pvd->vdev_ops == &vdev_spare_ops &&
4679 newvd->vdev_isspare != oldvd->vdev_isspare) {
4680 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4683 if (newvd->vdev_isspare)
4684 pvops = &vdev_spare_ops;
4686 pvops = &vdev_replacing_ops;
4690 * Make sure the new device is big enough.
4692 if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
4693 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
4696 * The new device cannot have a higher alignment requirement
4697 * than the top-level vdev.
4699 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4700 return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4703 * If this is an in-place replacement, update oldvd's path and devid
4704 * to make it distinguishable from newvd, and unopenable from now on.
4706 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4707 spa_strfree(oldvd->vdev_path);
4708 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4710 (void) sprintf(oldvd->vdev_path, "%s/%s",
4711 newvd->vdev_path, "old");
4712 if (oldvd->vdev_devid != NULL) {
4713 spa_strfree(oldvd->vdev_devid);
4714 oldvd->vdev_devid = NULL;
4718 /* mark the device being resilvered */
4719 newvd->vdev_resilver_txg = txg;
4722 * If the parent is not a mirror, or if we're replacing, insert the new
4723 * mirror/replacing/spare vdev above oldvd.
4725 if (pvd->vdev_ops != pvops)
4726 pvd = vdev_add_parent(oldvd, pvops);
4728 ASSERT(pvd->vdev_top->vdev_parent == rvd);
4729 ASSERT(pvd->vdev_ops == pvops);
4730 ASSERT(oldvd->vdev_parent == pvd);
4733 * Extract the new device from its root and add it to pvd.
4735 vdev_remove_child(newrootvd, newvd);
4736 newvd->vdev_id = pvd->vdev_children;
4737 newvd->vdev_crtxg = oldvd->vdev_crtxg;
4738 vdev_add_child(pvd, newvd);
4740 tvd = newvd->vdev_top;
4741 ASSERT(pvd->vdev_top == tvd);
4742 ASSERT(tvd->vdev_parent == rvd);
4744 vdev_config_dirty(tvd);
4747 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4748 * for any dmu_sync-ed blocks. It will propagate upward when
4749 * spa_vdev_exit() calls vdev_dtl_reassess().
4751 dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4753 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4754 dtl_max_txg - TXG_INITIAL);
4756 if (newvd->vdev_isspare) {
4757 spa_spare_activate(newvd);
4758 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
4761 oldvdpath = spa_strdup(oldvd->vdev_path);
4762 newvdpath = spa_strdup(newvd->vdev_path);
4763 newvd_isspare = newvd->vdev_isspare;
4766 * Mark newvd's DTL dirty in this txg.
4768 vdev_dirty(tvd, VDD_DTL, newvd, txg);
4771 * Schedule the resilver to restart in the future. We do this to
4772 * ensure that dmu_sync-ed blocks have been stitched into the
4773 * respective datasets.
4775 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4780 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4782 spa_history_log_internal(spa, "vdev attach", NULL,
4783 "%s vdev=%s %s vdev=%s",
4784 replacing && newvd_isspare ? "spare in" :
4785 replacing ? "replace" : "attach", newvdpath,
4786 replacing ? "for" : "to", oldvdpath);
4788 spa_strfree(oldvdpath);
4789 spa_strfree(newvdpath);
4791 if (spa->spa_bootfs)
4792 spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
4798 * Detach a device from a mirror or replacing vdev.
4800 * If 'replace_done' is specified, only detach if the parent
4801 * is a replacing vdev.
4804 spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4808 vdev_t *rvd = spa->spa_root_vdev;
4809 vdev_t *vd, *pvd, *cvd, *tvd;
4810 boolean_t unspare = B_FALSE;
4811 uint64_t unspare_guid = 0;
4814 ASSERT(spa_writeable(spa));
4816 txg = spa_vdev_enter(spa);
4818 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4821 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4823 if (!vd->vdev_ops->vdev_op_leaf)
4824 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4826 pvd = vd->vdev_parent;
4829 * If the parent/child relationship is not as expected, don't do it.
4830 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4831 * vdev that's replacing B with C. The user's intent in replacing
4832 * is to go from M(A,B) to M(A,C). If the user decides to cancel
4833 * the replace by detaching C, the expected behavior is to end up
4834 * M(A,B). But suppose that right after deciding to detach C,
4835 * the replacement of B completes. We would have M(A,C), and then
4836 * ask to detach C, which would leave us with just A -- not what
4837 * the user wanted. To prevent this, we make sure that the
4838 * parent/child relationship hasn't changed -- in this example,
4839 * that C's parent is still the replacing vdev R.
4841 if (pvd->vdev_guid != pguid && pguid != 0)
4842 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4845 * Only 'replacing' or 'spare' vdevs can be replaced.
4847 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4848 pvd->vdev_ops != &vdev_spare_ops)
4849 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4851 ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4852 spa_version(spa) >= SPA_VERSION_SPARES);
4855 * Only mirror, replacing, and spare vdevs support detach.
4857 if (pvd->vdev_ops != &vdev_replacing_ops &&
4858 pvd->vdev_ops != &vdev_mirror_ops &&
4859 pvd->vdev_ops != &vdev_spare_ops)
4860 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4863 * If this device has the only valid copy of some data,
4864 * we cannot safely detach it.
4866 if (vdev_dtl_required(vd))
4867 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4869 ASSERT(pvd->vdev_children >= 2);
4872 * If we are detaching the second disk from a replacing vdev, then
4873 * check to see if we changed the original vdev's path to have "/old"
4874 * at the end in spa_vdev_attach(). If so, undo that change now.
4876 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4877 vd->vdev_path != NULL) {
4878 size_t len = strlen(vd->vdev_path);
4880 for (int c = 0; c < pvd->vdev_children; c++) {
4881 cvd = pvd->vdev_child[c];
4883 if (cvd == vd || cvd->vdev_path == NULL)
4886 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4887 strcmp(cvd->vdev_path + len, "/old") == 0) {
4888 spa_strfree(cvd->vdev_path);
4889 cvd->vdev_path = spa_strdup(vd->vdev_path);
4896 * If we are detaching the original disk from a spare, then it implies
4897 * that the spare should become a real disk, and be removed from the
4898 * active spare list for the pool.
4900 if (pvd->vdev_ops == &vdev_spare_ops &&
4902 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4906 * Erase the disk labels so the disk can be used for other things.
4907 * This must be done after all other error cases are handled,
4908 * but before we disembowel vd (so we can still do I/O to it).
4909 * But if we can't do it, don't treat the error as fatal --
4910 * it may be that the unwritability of the disk is the reason
4911 * it's being detached!
4913 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4916 * Remove vd from its parent and compact the parent's children.
4918 vdev_remove_child(pvd, vd);
4919 vdev_compact_children(pvd);
4922 * Remember one of the remaining children so we can get tvd below.
4924 cvd = pvd->vdev_child[pvd->vdev_children - 1];
4927 * If we need to remove the remaining child from the list of hot spares,
4928 * do it now, marking the vdev as no longer a spare in the process.
4929 * We must do this before vdev_remove_parent(), because that can
4930 * change the GUID if it creates a new toplevel GUID. For a similar
4931 * reason, we must remove the spare now, in the same txg as the detach;
4932 * otherwise someone could attach a new sibling, change the GUID, and
4933 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4936 ASSERT(cvd->vdev_isspare);
4937 spa_spare_remove(cvd);
4938 unspare_guid = cvd->vdev_guid;
4939 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4940 cvd->vdev_unspare = B_TRUE;
4944 * If the parent mirror/replacing vdev only has one child,
4945 * the parent is no longer needed. Remove it from the tree.
4947 if (pvd->vdev_children == 1) {
4948 if (pvd->vdev_ops == &vdev_spare_ops)
4949 cvd->vdev_unspare = B_FALSE;
4950 vdev_remove_parent(cvd);
4955 * We don't set tvd until now because the parent we just removed
4956 * may have been the previous top-level vdev.
4958 tvd = cvd->vdev_top;
4959 ASSERT(tvd->vdev_parent == rvd);
4962 * Reevaluate the parent vdev state.
4964 vdev_propagate_state(cvd);
4967 * If the 'autoexpand' property is set on the pool then automatically
4968 * try to expand the size of the pool. For example if the device we
4969 * just detached was smaller than the others, it may be possible to
4970 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4971 * first so that we can obtain the updated sizes of the leaf vdevs.
4973 if (spa->spa_autoexpand) {
4975 vdev_expand(tvd, txg);
4978 vdev_config_dirty(tvd);
4981 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
4982 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4983 * But first make sure we're not on any *other* txg's DTL list, to
4984 * prevent vd from being accessed after it's freed.
4986 vdpath = spa_strdup(vd->vdev_path);
4987 for (int t = 0; t < TXG_SIZE; t++)
4988 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4989 vd->vdev_detached = B_TRUE;
4990 vdev_dirty(tvd, VDD_DTL, vd, txg);
4992 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
4994 /* hang on to the spa before we release the lock */
4995 spa_open_ref(spa, FTAG);
4997 error = spa_vdev_exit(spa, vd, txg, 0);
4999 spa_history_log_internal(spa, "detach", NULL,
5001 spa_strfree(vdpath);
5004 * If this was the removal of the original device in a hot spare vdev,
5005 * then we want to go through and remove the device from the hot spare
5006 * list of every other pool.
5009 spa_t *altspa = NULL;
5011 mutex_enter(&spa_namespace_lock);
5012 while ((altspa = spa_next(altspa)) != NULL) {
5013 if (altspa->spa_state != POOL_STATE_ACTIVE ||
5017 spa_open_ref(altspa, FTAG);
5018 mutex_exit(&spa_namespace_lock);
5019 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
5020 mutex_enter(&spa_namespace_lock);
5021 spa_close(altspa, FTAG);
5023 mutex_exit(&spa_namespace_lock);
5025 /* search the rest of the vdevs for spares to remove */
5026 spa_vdev_resilver_done(spa);
5029 /* all done with the spa; OK to release */
5030 mutex_enter(&spa_namespace_lock);
5031 spa_close(spa, FTAG);
5032 mutex_exit(&spa_namespace_lock);
5038 * Split a set of devices from their mirrors, and create a new pool from them.
5041 spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
5042 nvlist_t *props, boolean_t exp)
5045 uint64_t txg, *glist;
5047 uint_t c, children, lastlog;
5048 nvlist_t **child, *nvl, *tmp;
5050 char *altroot = NULL;
5051 vdev_t *rvd, **vml = NULL; /* vdev modify list */
5052 boolean_t activate_slog;
5054 ASSERT(spa_writeable(spa));
5056 txg = spa_vdev_enter(spa);
5058 /* clear the log and flush everything up to now */
5059 activate_slog = spa_passivate_log(spa);
5060 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5061 error = spa_offline_log(spa);
5062 txg = spa_vdev_config_enter(spa);
5065 spa_activate_log(spa);
5068 return (spa_vdev_exit(spa, NULL, txg, error));
5070 /* check new spa name before going any further */
5071 if (spa_lookup(newname) != NULL)
5072 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
5075 * scan through all the children to ensure they're all mirrors
5077 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
5078 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
5080 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5082 /* first, check to ensure we've got the right child count */
5083 rvd = spa->spa_root_vdev;
5085 for (c = 0; c < rvd->vdev_children; c++) {
5086 vdev_t *vd = rvd->vdev_child[c];
5088 /* don't count the holes & logs as children */
5089 if (vd->vdev_islog || vd->vdev_ishole) {
5097 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
5098 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5100 /* next, ensure no spare or cache devices are part of the split */
5101 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
5102 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
5103 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5105 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
5106 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
5108 /* then, loop over each vdev and validate it */
5109 for (c = 0; c < children; c++) {
5110 uint64_t is_hole = 0;
5112 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
5116 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
5117 spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
5120 error = SET_ERROR(EINVAL);
5125 /* which disk is going to be split? */
5126 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
5128 error = SET_ERROR(EINVAL);
5132 /* look it up in the spa */
5133 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
5134 if (vml[c] == NULL) {
5135 error = SET_ERROR(ENODEV);
5139 /* make sure there's nothing stopping the split */
5140 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
5141 vml[c]->vdev_islog ||
5142 vml[c]->vdev_ishole ||
5143 vml[c]->vdev_isspare ||
5144 vml[c]->vdev_isl2cache ||
5145 !vdev_writeable(vml[c]) ||
5146 vml[c]->vdev_children != 0 ||
5147 vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
5148 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
5149 error = SET_ERROR(EINVAL);
5153 if (vdev_dtl_required(vml[c])) {
5154 error = SET_ERROR(EBUSY);
5158 /* we need certain info from the top level */
5159 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
5160 vml[c]->vdev_top->vdev_ms_array) == 0);
5161 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
5162 vml[c]->vdev_top->vdev_ms_shift) == 0);
5163 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
5164 vml[c]->vdev_top->vdev_asize) == 0);
5165 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
5166 vml[c]->vdev_top->vdev_ashift) == 0);
5170 kmem_free(vml, children * sizeof (vdev_t *));
5171 kmem_free(glist, children * sizeof (uint64_t));
5172 return (spa_vdev_exit(spa, NULL, txg, error));
5175 /* stop writers from using the disks */
5176 for (c = 0; c < children; c++) {
5178 vml[c]->vdev_offline = B_TRUE;
5180 vdev_reopen(spa->spa_root_vdev);
5183 * Temporarily record the splitting vdevs in the spa config. This
5184 * will disappear once the config is regenerated.
5186 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5187 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
5188 glist, children) == 0);
5189 kmem_free(glist, children * sizeof (uint64_t));
5191 mutex_enter(&spa->spa_props_lock);
5192 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
5194 mutex_exit(&spa->spa_props_lock);
5195 spa->spa_config_splitting = nvl;
5196 vdev_config_dirty(spa->spa_root_vdev);
5198 /* configure and create the new pool */
5199 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
5200 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
5201 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
5202 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5203 spa_version(spa)) == 0);
5204 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
5205 spa->spa_config_txg) == 0);
5206 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
5207 spa_generate_guid(NULL)) == 0);
5208 (void) nvlist_lookup_string(props,
5209 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
5211 /* add the new pool to the namespace */
5212 newspa = spa_add(newname, config, altroot);
5213 newspa->spa_config_txg = spa->spa_config_txg;
5214 spa_set_log_state(newspa, SPA_LOG_CLEAR);
5216 /* release the spa config lock, retaining the namespace lock */
5217 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5219 if (zio_injection_enabled)
5220 zio_handle_panic_injection(spa, FTAG, 1);
5222 spa_activate(newspa, spa_mode_global);
5223 spa_async_suspend(newspa);
5226 /* mark that we are creating new spa by splitting */
5227 newspa->spa_splitting_newspa = B_TRUE;
5229 /* create the new pool from the disks of the original pool */
5230 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
5232 newspa->spa_splitting_newspa = B_FALSE;
5237 /* if that worked, generate a real config for the new pool */
5238 if (newspa->spa_root_vdev != NULL) {
5239 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
5240 NV_UNIQUE_NAME, KM_SLEEP) == 0);
5241 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
5242 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
5243 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
5248 if (props != NULL) {
5249 spa_configfile_set(newspa, props, B_FALSE);
5250 error = spa_prop_set(newspa, props);
5255 /* flush everything */
5256 txg = spa_vdev_config_enter(newspa);
5257 vdev_config_dirty(newspa->spa_root_vdev);
5258 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
5260 if (zio_injection_enabled)
5261 zio_handle_panic_injection(spa, FTAG, 2);
5263 spa_async_resume(newspa);
5265 /* finally, update the original pool's config */
5266 txg = spa_vdev_config_enter(spa);
5267 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
5268 error = dmu_tx_assign(tx, TXG_WAIT);
5271 for (c = 0; c < children; c++) {
5272 if (vml[c] != NULL) {
5275 spa_history_log_internal(spa, "detach", tx,
5276 "vdev=%s", vml[c]->vdev_path);
5280 vdev_config_dirty(spa->spa_root_vdev);
5281 spa->spa_config_splitting = NULL;
5285 (void) spa_vdev_exit(spa, NULL, txg, 0);
5287 if (zio_injection_enabled)
5288 zio_handle_panic_injection(spa, FTAG, 3);
5290 /* split is complete; log a history record */
5291 spa_history_log_internal(newspa, "split", NULL,
5292 "from pool %s", spa_name(spa));
5294 kmem_free(vml, children * sizeof (vdev_t *));
5296 /* if we're not going to mount the filesystems in userland, export */
5298 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
5305 spa_deactivate(newspa);
5308 txg = spa_vdev_config_enter(spa);
5310 /* re-online all offlined disks */
5311 for (c = 0; c < children; c++) {
5313 vml[c]->vdev_offline = B_FALSE;
5315 vdev_reopen(spa->spa_root_vdev);
5317 nvlist_free(spa->spa_config_splitting);
5318 spa->spa_config_splitting = NULL;
5319 (void) spa_vdev_exit(spa, NULL, txg, error);
5321 kmem_free(vml, children * sizeof (vdev_t *));
5326 spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
5328 for (int i = 0; i < count; i++) {
5331 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
5334 if (guid == target_guid)
5342 spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
5343 nvlist_t *dev_to_remove)
5345 nvlist_t **newdev = NULL;
5348 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
5350 for (int i = 0, j = 0; i < count; i++) {
5351 if (dev[i] == dev_to_remove)
5353 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
5356 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
5357 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
5359 for (int i = 0; i < count - 1; i++)
5360 nvlist_free(newdev[i]);
5363 kmem_free(newdev, (count - 1) * sizeof (void *));
5367 * Evacuate the device.
5370 spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
5375 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5376 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5377 ASSERT(vd == vd->vdev_top);
5380 * Evacuate the device. We don't hold the config lock as writer
5381 * since we need to do I/O but we do keep the
5382 * spa_namespace_lock held. Once this completes the device
5383 * should no longer have any blocks allocated on it.
5385 if (vd->vdev_islog) {
5386 if (vd->vdev_stat.vs_alloc != 0)
5387 error = spa_offline_log(spa);
5389 error = SET_ERROR(ENOTSUP);
5396 * The evacuation succeeded. Remove any remaining MOS metadata
5397 * associated with this vdev, and wait for these changes to sync.
5399 ASSERT0(vd->vdev_stat.vs_alloc);
5400 txg = spa_vdev_config_enter(spa);
5401 vd->vdev_removing = B_TRUE;
5402 vdev_dirty_leaves(vd, VDD_DTL, txg);
5403 vdev_config_dirty(vd);
5404 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5410 * Complete the removal by cleaning up the namespace.
5413 spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
5415 vdev_t *rvd = spa->spa_root_vdev;
5416 uint64_t id = vd->vdev_id;
5417 boolean_t last_vdev = (id == (rvd->vdev_children - 1));
5419 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5420 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5421 ASSERT(vd == vd->vdev_top);
5424 * Only remove any devices which are empty.
5426 if (vd->vdev_stat.vs_alloc != 0)
5429 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5431 if (list_link_active(&vd->vdev_state_dirty_node))
5432 vdev_state_clean(vd);
5433 if (list_link_active(&vd->vdev_config_dirty_node))
5434 vdev_config_clean(vd);
5439 vdev_compact_children(rvd);
5441 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
5442 vdev_add_child(rvd, vd);
5444 vdev_config_dirty(rvd);
5447 * Reassess the health of our root vdev.
5453 * Remove a device from the pool -
5455 * Removing a device from the vdev namespace requires several steps
5456 * and can take a significant amount of time. As a result we use
5457 * the spa_vdev_config_[enter/exit] functions which allow us to
5458 * grab and release the spa_config_lock while still holding the namespace
5459 * lock. During each step the configuration is synced out.
5461 * Currently, this supports removing only hot spares, slogs, and level 2 ARC
5465 spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
5468 metaslab_group_t *mg;
5469 nvlist_t **spares, **l2cache, *nv;
5471 uint_t nspares, nl2cache;
5473 boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
5475 ASSERT(spa_writeable(spa));
5478 txg = spa_vdev_enter(spa);
5480 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5482 if (spa->spa_spares.sav_vdevs != NULL &&
5483 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5484 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
5485 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
5487 * Only remove the hot spare if it's not currently in use
5490 if (vd == NULL || unspare) {
5491 spa_vdev_remove_aux(spa->spa_spares.sav_config,
5492 ZPOOL_CONFIG_SPARES, spares, nspares, nv);
5493 spa_load_spares(spa);
5494 spa->spa_spares.sav_sync = B_TRUE;
5496 error = SET_ERROR(EBUSY);
5498 } else if (spa->spa_l2cache.sav_vdevs != NULL &&
5499 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5500 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
5501 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
5503 * Cache devices can always be removed.
5505 spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
5506 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
5507 spa_load_l2cache(spa);
5508 spa->spa_l2cache.sav_sync = B_TRUE;
5509 } else if (vd != NULL && vd->vdev_islog) {
5511 ASSERT(vd == vd->vdev_top);
5514 * XXX - Once we have bp-rewrite this should
5515 * become the common case.
5521 * Stop allocating from this vdev.
5523 metaslab_group_passivate(mg);
5526 * Wait for the youngest allocations and frees to sync,
5527 * and then wait for the deferral of those frees to finish.
5529 spa_vdev_config_exit(spa, NULL,
5530 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
5533 * Attempt to evacuate the vdev.
5535 error = spa_vdev_remove_evacuate(spa, vd);
5537 txg = spa_vdev_config_enter(spa);
5540 * If we couldn't evacuate the vdev, unwind.
5543 metaslab_group_activate(mg);
5544 return (spa_vdev_exit(spa, NULL, txg, error));
5548 * Clean up the vdev namespace.
5550 spa_vdev_remove_from_namespace(spa, vd);
5552 } else if (vd != NULL) {
5554 * Normal vdevs cannot be removed (yet).
5556 error = SET_ERROR(ENOTSUP);
5559 * There is no vdev of any kind with the specified guid.
5561 error = SET_ERROR(ENOENT);
5565 return (spa_vdev_exit(spa, NULL, txg, error));
5571 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5572 * currently spared, so we can detach it.
5575 spa_vdev_resilver_done_hunt(vdev_t *vd)
5577 vdev_t *newvd, *oldvd;
5579 for (int c = 0; c < vd->vdev_children; c++) {
5580 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
5586 * Check for a completed replacement. We always consider the first
5587 * vdev in the list to be the oldest vdev, and the last one to be
5588 * the newest (see spa_vdev_attach() for how that works). In
5589 * the case where the newest vdev is faulted, we will not automatically
5590 * remove it after a resilver completes. This is OK as it will require
5591 * user intervention to determine which disk the admin wishes to keep.
5593 if (vd->vdev_ops == &vdev_replacing_ops) {
5594 ASSERT(vd->vdev_children > 1);
5596 newvd = vd->vdev_child[vd->vdev_children - 1];
5597 oldvd = vd->vdev_child[0];
5599 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
5600 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5601 !vdev_dtl_required(oldvd))
5606 * Check for a completed resilver with the 'unspare' flag set.
5608 if (vd->vdev_ops == &vdev_spare_ops) {
5609 vdev_t *first = vd->vdev_child[0];
5610 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
5612 if (last->vdev_unspare) {
5615 } else if (first->vdev_unspare) {
5622 if (oldvd != NULL &&
5623 vdev_dtl_empty(newvd, DTL_MISSING) &&
5624 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5625 !vdev_dtl_required(oldvd))
5629 * If there are more than two spares attached to a disk,
5630 * and those spares are not required, then we want to
5631 * attempt to free them up now so that they can be used
5632 * by other pools. Once we're back down to a single
5633 * disk+spare, we stop removing them.
5635 if (vd->vdev_children > 2) {
5636 newvd = vd->vdev_child[1];
5638 if (newvd->vdev_isspare && last->vdev_isspare &&
5639 vdev_dtl_empty(last, DTL_MISSING) &&
5640 vdev_dtl_empty(last, DTL_OUTAGE) &&
5641 !vdev_dtl_required(newvd))
5650 spa_vdev_resilver_done(spa_t *spa)
5652 vdev_t *vd, *pvd, *ppvd;
5653 uint64_t guid, sguid, pguid, ppguid;
5655 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5657 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
5658 pvd = vd->vdev_parent;
5659 ppvd = pvd->vdev_parent;
5660 guid = vd->vdev_guid;
5661 pguid = pvd->vdev_guid;
5662 ppguid = ppvd->vdev_guid;
5665 * If we have just finished replacing a hot spared device, then
5666 * we need to detach the parent's first child (the original hot
5669 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
5670 ppvd->vdev_children == 2) {
5671 ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
5672 sguid = ppvd->vdev_child[1]->vdev_guid;
5674 ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd));
5676 spa_config_exit(spa, SCL_ALL, FTAG);
5677 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
5679 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
5681 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5684 spa_config_exit(spa, SCL_ALL, FTAG);
5688 * Update the stored path or FRU for this vdev.
5691 spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
5695 boolean_t sync = B_FALSE;
5697 ASSERT(spa_writeable(spa));
5699 spa_vdev_state_enter(spa, SCL_ALL);
5701 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
5702 return (spa_vdev_state_exit(spa, NULL, ENOENT));
5704 if (!vd->vdev_ops->vdev_op_leaf)
5705 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
5708 if (strcmp(value, vd->vdev_path) != 0) {
5709 spa_strfree(vd->vdev_path);
5710 vd->vdev_path = spa_strdup(value);
5714 if (vd->vdev_fru == NULL) {
5715 vd->vdev_fru = spa_strdup(value);
5717 } else if (strcmp(value, vd->vdev_fru) != 0) {
5718 spa_strfree(vd->vdev_fru);
5719 vd->vdev_fru = spa_strdup(value);
5724 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
5728 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
5730 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5734 spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5736 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5740 * ==========================================================================
5742 * ==========================================================================
5746 spa_scan_stop(spa_t *spa)
5748 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5749 if (dsl_scan_resilvering(spa->spa_dsl_pool))
5750 return (SET_ERROR(EBUSY));
5751 return (dsl_scan_cancel(spa->spa_dsl_pool));
5755 spa_scan(spa_t *spa, pool_scan_func_t func)
5757 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5759 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5760 return (SET_ERROR(ENOTSUP));
5763 * If a resilver was requested, but there is no DTL on a
5764 * writeable leaf device, we have nothing to do.
5766 if (func == POOL_SCAN_RESILVER &&
5767 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5768 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5772 return (dsl_scan(spa->spa_dsl_pool, func));
5776 * ==========================================================================
5777 * SPA async task processing
5778 * ==========================================================================
5782 spa_async_remove(spa_t *spa, vdev_t *vd)
5784 if (vd->vdev_remove_wanted) {
5785 vd->vdev_remove_wanted = B_FALSE;
5786 vd->vdev_delayed_close = B_FALSE;
5787 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5790 * We want to clear the stats, but we don't want to do a full
5791 * vdev_clear() as that will cause us to throw away
5792 * degraded/faulted state as well as attempt to reopen the
5793 * device, all of which is a waste.
5795 vd->vdev_stat.vs_read_errors = 0;
5796 vd->vdev_stat.vs_write_errors = 0;
5797 vd->vdev_stat.vs_checksum_errors = 0;
5799 vdev_state_dirty(vd->vdev_top);
5802 for (int c = 0; c < vd->vdev_children; c++)
5803 spa_async_remove(spa, vd->vdev_child[c]);
5807 spa_async_probe(spa_t *spa, vdev_t *vd)
5809 if (vd->vdev_probe_wanted) {
5810 vd->vdev_probe_wanted = B_FALSE;
5811 vdev_reopen(vd); /* vdev_open() does the actual probe */
5814 for (int c = 0; c < vd->vdev_children; c++)
5815 spa_async_probe(spa, vd->vdev_child[c]);
5819 spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5825 if (!spa->spa_autoexpand)
5828 for (int c = 0; c < vd->vdev_children; c++) {
5829 vdev_t *cvd = vd->vdev_child[c];
5830 spa_async_autoexpand(spa, cvd);
5833 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5836 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
5837 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
5839 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5840 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
5842 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
5843 ESC_ZFS_VDEV_AUTOEXPAND, attr, &eid, DDI_SLEEP);
5846 kmem_free(physpath, MAXPATHLEN);
5850 spa_async_thread(void *arg)
5855 ASSERT(spa->spa_sync_on);
5857 mutex_enter(&spa->spa_async_lock);
5858 tasks = spa->spa_async_tasks;
5859 spa->spa_async_tasks &= SPA_ASYNC_REMOVE;
5860 mutex_exit(&spa->spa_async_lock);
5863 * See if the config needs to be updated.
5865 if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5866 uint64_t old_space, new_space;
5868 mutex_enter(&spa_namespace_lock);
5869 old_space = metaslab_class_get_space(spa_normal_class(spa));
5870 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5871 new_space = metaslab_class_get_space(spa_normal_class(spa));
5872 mutex_exit(&spa_namespace_lock);
5875 * If the pool grew as a result of the config update,
5876 * then log an internal history event.
5878 if (new_space != old_space) {
5879 spa_history_log_internal(spa, "vdev online", NULL,
5880 "pool '%s' size: %llu(+%llu)",
5881 spa_name(spa), new_space, new_space - old_space);
5885 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5886 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5887 spa_async_autoexpand(spa, spa->spa_root_vdev);
5888 spa_config_exit(spa, SCL_CONFIG, FTAG);
5892 * See if any devices need to be probed.
5894 if (tasks & SPA_ASYNC_PROBE) {
5895 spa_vdev_state_enter(spa, SCL_NONE);
5896 spa_async_probe(spa, spa->spa_root_vdev);
5897 (void) spa_vdev_state_exit(spa, NULL, 0);
5901 * If any devices are done replacing, detach them.
5903 if (tasks & SPA_ASYNC_RESILVER_DONE)
5904 spa_vdev_resilver_done(spa);
5907 * Kick off a resilver.
5909 if (tasks & SPA_ASYNC_RESILVER)
5910 dsl_resilver_restart(spa->spa_dsl_pool, 0);
5913 * Let the world know that we're done.
5915 mutex_enter(&spa->spa_async_lock);
5916 spa->spa_async_thread = NULL;
5917 cv_broadcast(&spa->spa_async_cv);
5918 mutex_exit(&spa->spa_async_lock);
5923 spa_async_thread_vd(void *arg)
5928 ASSERT(spa->spa_sync_on);
5930 mutex_enter(&spa->spa_async_lock);
5931 tasks = spa->spa_async_tasks;
5933 spa->spa_async_tasks &= ~SPA_ASYNC_REMOVE;
5934 mutex_exit(&spa->spa_async_lock);
5937 * See if any devices need to be marked REMOVED.
5939 if (tasks & SPA_ASYNC_REMOVE) {
5940 spa_vdev_state_enter(spa, SCL_NONE);
5941 spa_async_remove(spa, spa->spa_root_vdev);
5942 for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
5943 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5944 for (int i = 0; i < spa->spa_spares.sav_count; i++)
5945 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5946 (void) spa_vdev_state_exit(spa, NULL, 0);
5950 * Let the world know that we're done.
5952 mutex_enter(&spa->spa_async_lock);
5953 tasks = spa->spa_async_tasks;
5954 if ((tasks & SPA_ASYNC_REMOVE) != 0)
5956 spa->spa_async_thread_vd = NULL;
5957 cv_broadcast(&spa->spa_async_cv);
5958 mutex_exit(&spa->spa_async_lock);
5963 spa_async_suspend(spa_t *spa)
5965 mutex_enter(&spa->spa_async_lock);
5966 spa->spa_async_suspended++;
5967 while (spa->spa_async_thread != NULL &&
5968 spa->spa_async_thread_vd != NULL)
5969 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5970 mutex_exit(&spa->spa_async_lock);
5974 spa_async_resume(spa_t *spa)
5976 mutex_enter(&spa->spa_async_lock);
5977 ASSERT(spa->spa_async_suspended != 0);
5978 spa->spa_async_suspended--;
5979 mutex_exit(&spa->spa_async_lock);
5983 spa_async_tasks_pending(spa_t *spa)
5985 uint_t non_config_tasks;
5987 boolean_t config_task_suspended;
5989 non_config_tasks = spa->spa_async_tasks & ~(SPA_ASYNC_CONFIG_UPDATE |
5991 config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE;
5992 if (spa->spa_ccw_fail_time == 0) {
5993 config_task_suspended = B_FALSE;
5995 config_task_suspended =
5996 (gethrtime() - spa->spa_ccw_fail_time) <
5997 (zfs_ccw_retry_interval * NANOSEC);
6000 return (non_config_tasks || (config_task && !config_task_suspended));
6004 spa_async_dispatch(spa_t *spa)
6006 mutex_enter(&spa->spa_async_lock);
6007 if (spa_async_tasks_pending(spa) &&
6008 !spa->spa_async_suspended &&
6009 spa->spa_async_thread == NULL &&
6011 spa->spa_async_thread = thread_create(NULL, 0,
6012 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
6013 mutex_exit(&spa->spa_async_lock);
6017 spa_async_dispatch_vd(spa_t *spa)
6019 mutex_enter(&spa->spa_async_lock);
6020 if ((spa->spa_async_tasks & SPA_ASYNC_REMOVE) != 0 &&
6021 !spa->spa_async_suspended &&
6022 spa->spa_async_thread_vd == NULL &&
6024 spa->spa_async_thread_vd = thread_create(NULL, 0,
6025 spa_async_thread_vd, spa, 0, &p0, TS_RUN, maxclsyspri);
6026 mutex_exit(&spa->spa_async_lock);
6030 spa_async_request(spa_t *spa, int task)
6032 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
6033 mutex_enter(&spa->spa_async_lock);
6034 spa->spa_async_tasks |= task;
6035 mutex_exit(&spa->spa_async_lock);
6036 spa_async_dispatch_vd(spa);
6040 * ==========================================================================
6041 * SPA syncing routines
6042 * ==========================================================================
6046 bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
6049 bpobj_enqueue(bpo, bp, tx);
6054 spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
6058 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
6059 BP_GET_PSIZE(bp), zio->io_flags));
6064 * Note: this simple function is not inlined to make it easier to dtrace the
6065 * amount of time spent syncing frees.
6068 spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx)
6070 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6071 bplist_iterate(bpl, spa_free_sync_cb, zio, tx);
6072 VERIFY(zio_wait(zio) == 0);
6076 * Note: this simple function is not inlined to make it easier to dtrace the
6077 * amount of time spent syncing deferred frees.
6080 spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx)
6082 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6083 VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj,
6084 spa_free_sync_cb, zio, tx), ==, 0);
6085 VERIFY0(zio_wait(zio));
6090 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
6092 char *packed = NULL;
6097 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
6100 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
6101 * information. This avoids the dbuf_will_dirty() path and
6102 * saves us a pre-read to get data we don't actually care about.
6104 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
6105 packed = kmem_alloc(bufsize, KM_SLEEP);
6107 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
6109 bzero(packed + nvsize, bufsize - nvsize);
6111 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
6113 kmem_free(packed, bufsize);
6115 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
6116 dmu_buf_will_dirty(db, tx);
6117 *(uint64_t *)db->db_data = nvsize;
6118 dmu_buf_rele(db, FTAG);
6122 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
6123 const char *config, const char *entry)
6133 * Update the MOS nvlist describing the list of available devices.
6134 * spa_validate_aux() will have already made sure this nvlist is
6135 * valid and the vdevs are labeled appropriately.
6137 if (sav->sav_object == 0) {
6138 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
6139 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
6140 sizeof (uint64_t), tx);
6141 VERIFY(zap_update(spa->spa_meta_objset,
6142 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
6143 &sav->sav_object, tx) == 0);
6146 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
6147 if (sav->sav_count == 0) {
6148 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
6150 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
6151 for (i = 0; i < sav->sav_count; i++)
6152 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
6153 B_FALSE, VDEV_CONFIG_L2CACHE);
6154 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
6155 sav->sav_count) == 0);
6156 for (i = 0; i < sav->sav_count; i++)
6157 nvlist_free(list[i]);
6158 kmem_free(list, sav->sav_count * sizeof (void *));
6161 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
6162 nvlist_free(nvroot);
6164 sav->sav_sync = B_FALSE;
6168 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
6172 if (list_is_empty(&spa->spa_config_dirty_list))
6175 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6177 config = spa_config_generate(spa, spa->spa_root_vdev,
6178 dmu_tx_get_txg(tx), B_FALSE);
6181 * If we're upgrading the spa version then make sure that
6182 * the config object gets updated with the correct version.
6184 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
6185 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
6186 spa->spa_uberblock.ub_version);
6188 spa_config_exit(spa, SCL_STATE, FTAG);
6190 if (spa->spa_config_syncing)
6191 nvlist_free(spa->spa_config_syncing);
6192 spa->spa_config_syncing = config;
6194 spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
6198 spa_sync_version(void *arg, dmu_tx_t *tx)
6200 uint64_t *versionp = arg;
6201 uint64_t version = *versionp;
6202 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
6205 * Setting the version is special cased when first creating the pool.
6207 ASSERT(tx->tx_txg != TXG_INITIAL);
6209 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
6210 ASSERT(version >= spa_version(spa));
6212 spa->spa_uberblock.ub_version = version;
6213 vdev_config_dirty(spa->spa_root_vdev);
6214 spa_history_log_internal(spa, "set", tx, "version=%lld", version);
6218 * Set zpool properties.
6221 spa_sync_props(void *arg, dmu_tx_t *tx)
6223 nvlist_t *nvp = arg;
6224 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
6225 objset_t *mos = spa->spa_meta_objset;
6226 nvpair_t *elem = NULL;
6228 mutex_enter(&spa->spa_props_lock);
6230 while ((elem = nvlist_next_nvpair(nvp, elem))) {
6232 char *strval, *fname;
6234 const char *propname;
6235 zprop_type_t proptype;
6236 zfeature_info_t *feature;
6238 switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
6241 * We checked this earlier in spa_prop_validate().
6243 ASSERT(zpool_prop_feature(nvpair_name(elem)));
6245 fname = strchr(nvpair_name(elem), '@') + 1;
6246 VERIFY0(zfeature_lookup_name(fname, &feature));
6248 spa_feature_enable(spa, feature, tx);
6249 spa_history_log_internal(spa, "set", tx,
6250 "%s=enabled", nvpair_name(elem));
6253 case ZPOOL_PROP_VERSION:
6254 intval = fnvpair_value_uint64(elem);
6256 * The version is synced seperatly before other
6257 * properties and should be correct by now.
6259 ASSERT3U(spa_version(spa), >=, intval);
6262 case ZPOOL_PROP_ALTROOT:
6264 * 'altroot' is a non-persistent property. It should
6265 * have been set temporarily at creation or import time.
6267 ASSERT(spa->spa_root != NULL);
6270 case ZPOOL_PROP_READONLY:
6271 case ZPOOL_PROP_CACHEFILE:
6273 * 'readonly' and 'cachefile' are also non-persisitent
6277 case ZPOOL_PROP_COMMENT:
6278 strval = fnvpair_value_string(elem);
6279 if (spa->spa_comment != NULL)
6280 spa_strfree(spa->spa_comment);
6281 spa->spa_comment = spa_strdup(strval);
6283 * We need to dirty the configuration on all the vdevs
6284 * so that their labels get updated. It's unnecessary
6285 * to do this for pool creation since the vdev's
6286 * configuratoin has already been dirtied.
6288 if (tx->tx_txg != TXG_INITIAL)
6289 vdev_config_dirty(spa->spa_root_vdev);
6290 spa_history_log_internal(spa, "set", tx,
6291 "%s=%s", nvpair_name(elem), strval);
6295 * Set pool property values in the poolprops mos object.
6297 if (spa->spa_pool_props_object == 0) {
6298 spa->spa_pool_props_object =
6299 zap_create_link(mos, DMU_OT_POOL_PROPS,
6300 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
6304 /* normalize the property name */
6305 propname = zpool_prop_to_name(prop);
6306 proptype = zpool_prop_get_type(prop);
6308 if (nvpair_type(elem) == DATA_TYPE_STRING) {
6309 ASSERT(proptype == PROP_TYPE_STRING);
6310 strval = fnvpair_value_string(elem);
6311 VERIFY0(zap_update(mos,
6312 spa->spa_pool_props_object, propname,
6313 1, strlen(strval) + 1, strval, tx));
6314 spa_history_log_internal(spa, "set", tx,
6315 "%s=%s", nvpair_name(elem), strval);
6316 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
6317 intval = fnvpair_value_uint64(elem);
6319 if (proptype == PROP_TYPE_INDEX) {
6321 VERIFY0(zpool_prop_index_to_string(
6322 prop, intval, &unused));
6324 VERIFY0(zap_update(mos,
6325 spa->spa_pool_props_object, propname,
6326 8, 1, &intval, tx));
6327 spa_history_log_internal(spa, "set", tx,
6328 "%s=%lld", nvpair_name(elem), intval);
6330 ASSERT(0); /* not allowed */
6334 case ZPOOL_PROP_DELEGATION:
6335 spa->spa_delegation = intval;
6337 case ZPOOL_PROP_BOOTFS:
6338 spa->spa_bootfs = intval;
6340 case ZPOOL_PROP_FAILUREMODE:
6341 spa->spa_failmode = intval;
6343 case ZPOOL_PROP_AUTOEXPAND:
6344 spa->spa_autoexpand = intval;
6345 if (tx->tx_txg != TXG_INITIAL)
6346 spa_async_request(spa,
6347 SPA_ASYNC_AUTOEXPAND);
6349 case ZPOOL_PROP_DEDUPDITTO:
6350 spa->spa_dedup_ditto = intval;
6359 mutex_exit(&spa->spa_props_lock);
6363 * Perform one-time upgrade on-disk changes. spa_version() does not
6364 * reflect the new version this txg, so there must be no changes this
6365 * txg to anything that the upgrade code depends on after it executes.
6366 * Therefore this must be called after dsl_pool_sync() does the sync
6370 spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
6372 dsl_pool_t *dp = spa->spa_dsl_pool;
6374 ASSERT(spa->spa_sync_pass == 1);
6376 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
6378 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
6379 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
6380 dsl_pool_create_origin(dp, tx);
6382 /* Keeping the origin open increases spa_minref */
6383 spa->spa_minref += 3;
6386 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
6387 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
6388 dsl_pool_upgrade_clones(dp, tx);
6391 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
6392 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
6393 dsl_pool_upgrade_dir_clones(dp, tx);
6395 /* Keeping the freedir open increases spa_minref */
6396 spa->spa_minref += 3;
6399 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES &&
6400 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
6401 spa_feature_create_zap_objects(spa, tx);
6403 rrw_exit(&dp->dp_config_rwlock, FTAG);
6407 * Sync the specified transaction group. New blocks may be dirtied as
6408 * part of the process, so we iterate until it converges.
6411 spa_sync(spa_t *spa, uint64_t txg)
6413 dsl_pool_t *dp = spa->spa_dsl_pool;
6414 objset_t *mos = spa->spa_meta_objset;
6415 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
6416 vdev_t *rvd = spa->spa_root_vdev;
6421 VERIFY(spa_writeable(spa));
6424 * Lock out configuration changes.
6426 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
6428 spa->spa_syncing_txg = txg;
6429 spa->spa_sync_pass = 0;
6432 * If there are any pending vdev state changes, convert them
6433 * into config changes that go out with this transaction group.
6435 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6436 while (list_head(&spa->spa_state_dirty_list) != NULL) {
6438 * We need the write lock here because, for aux vdevs,
6439 * calling vdev_config_dirty() modifies sav_config.
6440 * This is ugly and will become unnecessary when we
6441 * eliminate the aux vdev wart by integrating all vdevs
6442 * into the root vdev tree.
6444 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6445 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
6446 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
6447 vdev_state_clean(vd);
6448 vdev_config_dirty(vd);
6450 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6451 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
6453 spa_config_exit(spa, SCL_STATE, FTAG);
6455 tx = dmu_tx_create_assigned(dp, txg);
6457 spa->spa_sync_starttime = gethrtime();
6459 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid,
6460 spa->spa_sync_starttime + spa->spa_deadman_synctime));
6463 callout_reset(&spa->spa_deadman_cycid,
6464 hz * spa->spa_deadman_synctime / NANOSEC, spa_deadman, spa);
6469 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
6470 * set spa_deflate if we have no raid-z vdevs.
6472 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
6473 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
6476 for (i = 0; i < rvd->vdev_children; i++) {
6477 vd = rvd->vdev_child[i];
6478 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
6481 if (i == rvd->vdev_children) {
6482 spa->spa_deflate = TRUE;
6483 VERIFY(0 == zap_add(spa->spa_meta_objset,
6484 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
6485 sizeof (uint64_t), 1, &spa->spa_deflate, tx));
6490 * If anything has changed in this txg, or if someone is waiting
6491 * for this txg to sync (eg, spa_vdev_remove()), push the
6492 * deferred frees from the previous txg. If not, leave them
6493 * alone so that we don't generate work on an otherwise idle
6496 if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
6497 !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
6498 !txg_list_empty(&dp->dp_sync_tasks, txg) ||
6499 ((dsl_scan_active(dp->dp_scan) ||
6500 txg_sync_waiting(dp)) && !spa_shutting_down(spa))) {
6501 spa_sync_deferred_frees(spa, tx);
6505 * Iterate to convergence.
6508 int pass = ++spa->spa_sync_pass;
6510 spa_sync_config_object(spa, tx);
6511 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
6512 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
6513 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
6514 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
6515 spa_errlog_sync(spa, txg);
6516 dsl_pool_sync(dp, txg);
6518 if (pass < zfs_sync_pass_deferred_free) {
6519 spa_sync_frees(spa, free_bpl, tx);
6521 bplist_iterate(free_bpl, bpobj_enqueue_cb,
6522 &spa->spa_deferred_bpobj, tx);
6526 dsl_scan_sync(dp, tx);
6528 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
6532 spa_sync_upgrades(spa, tx);
6534 } while (dmu_objset_is_dirty(mos, txg));
6537 * Rewrite the vdev configuration (which includes the uberblock)
6538 * to commit the transaction group.
6540 * If there are no dirty vdevs, we sync the uberblock to a few
6541 * random top-level vdevs that are known to be visible in the
6542 * config cache (see spa_vdev_add() for a complete description).
6543 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
6547 * We hold SCL_STATE to prevent vdev open/close/etc.
6548 * while we're attempting to write the vdev labels.
6550 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6552 if (list_is_empty(&spa->spa_config_dirty_list)) {
6553 vdev_t *svd[SPA_DVAS_PER_BP];
6555 int children = rvd->vdev_children;
6556 int c0 = spa_get_random(children);
6558 for (int c = 0; c < children; c++) {
6559 vd = rvd->vdev_child[(c0 + c) % children];
6560 if (vd->vdev_ms_array == 0 || vd->vdev_islog)
6562 svd[svdcount++] = vd;
6563 if (svdcount == SPA_DVAS_PER_BP)
6566 error = vdev_config_sync(svd, svdcount, txg, B_FALSE);
6568 error = vdev_config_sync(svd, svdcount, txg,
6571 error = vdev_config_sync(rvd->vdev_child,
6572 rvd->vdev_children, txg, B_FALSE);
6574 error = vdev_config_sync(rvd->vdev_child,
6575 rvd->vdev_children, txg, B_TRUE);
6579 spa->spa_last_synced_guid = rvd->vdev_guid;
6581 spa_config_exit(spa, SCL_STATE, FTAG);
6585 zio_suspend(spa, NULL);
6586 zio_resume_wait(spa);
6591 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY));
6594 callout_drain(&spa->spa_deadman_cycid);
6599 * Clear the dirty config list.
6601 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
6602 vdev_config_clean(vd);
6605 * Now that the new config has synced transactionally,
6606 * let it become visible to the config cache.
6608 if (spa->spa_config_syncing != NULL) {
6609 spa_config_set(spa, spa->spa_config_syncing);
6610 spa->spa_config_txg = txg;
6611 spa->spa_config_syncing = NULL;
6614 spa->spa_ubsync = spa->spa_uberblock;
6616 dsl_pool_sync_done(dp, txg);
6619 * Update usable space statistics.
6621 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
6622 vdev_sync_done(vd, txg);
6624 spa_update_dspace(spa);
6627 * It had better be the case that we didn't dirty anything
6628 * since vdev_config_sync().
6630 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
6631 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
6632 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
6634 spa->spa_sync_pass = 0;
6636 spa_config_exit(spa, SCL_CONFIG, FTAG);
6638 spa_handle_ignored_writes(spa);
6641 * If any async tasks have been requested, kick them off.
6643 spa_async_dispatch(spa);
6644 spa_async_dispatch_vd(spa);
6648 * Sync all pools. We don't want to hold the namespace lock across these
6649 * operations, so we take a reference on the spa_t and drop the lock during the
6653 spa_sync_allpools(void)
6656 mutex_enter(&spa_namespace_lock);
6657 while ((spa = spa_next(spa)) != NULL) {
6658 if (spa_state(spa) != POOL_STATE_ACTIVE ||
6659 !spa_writeable(spa) || spa_suspended(spa))
6661 spa_open_ref(spa, FTAG);
6662 mutex_exit(&spa_namespace_lock);
6663 txg_wait_synced(spa_get_dsl(spa), 0);
6664 mutex_enter(&spa_namespace_lock);
6665 spa_close(spa, FTAG);
6667 mutex_exit(&spa_namespace_lock);
6671 * ==========================================================================
6672 * Miscellaneous routines
6673 * ==========================================================================
6677 * Remove all pools in the system.
6685 * Remove all cached state. All pools should be closed now,
6686 * so every spa in the AVL tree should be unreferenced.
6688 mutex_enter(&spa_namespace_lock);
6689 while ((spa = spa_next(NULL)) != NULL) {
6691 * Stop async tasks. The async thread may need to detach
6692 * a device that's been replaced, which requires grabbing
6693 * spa_namespace_lock, so we must drop it here.
6695 spa_open_ref(spa, FTAG);
6696 mutex_exit(&spa_namespace_lock);
6697 spa_async_suspend(spa);
6698 mutex_enter(&spa_namespace_lock);
6699 spa_close(spa, FTAG);
6701 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
6703 spa_deactivate(spa);
6707 mutex_exit(&spa_namespace_lock);
6711 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
6716 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
6720 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
6721 vd = spa->spa_l2cache.sav_vdevs[i];
6722 if (vd->vdev_guid == guid)
6726 for (i = 0; i < spa->spa_spares.sav_count; i++) {
6727 vd = spa->spa_spares.sav_vdevs[i];
6728 if (vd->vdev_guid == guid)
6737 spa_upgrade(spa_t *spa, uint64_t version)
6739 ASSERT(spa_writeable(spa));
6741 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6744 * This should only be called for a non-faulted pool, and since a
6745 * future version would result in an unopenable pool, this shouldn't be
6748 ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version));
6749 ASSERT(version >= spa->spa_uberblock.ub_version);
6751 spa->spa_uberblock.ub_version = version;
6752 vdev_config_dirty(spa->spa_root_vdev);
6754 spa_config_exit(spa, SCL_ALL, FTAG);
6756 txg_wait_synced(spa_get_dsl(spa), 0);
6760 spa_has_spare(spa_t *spa, uint64_t guid)
6764 spa_aux_vdev_t *sav = &spa->spa_spares;
6766 for (i = 0; i < sav->sav_count; i++)
6767 if (sav->sav_vdevs[i]->vdev_guid == guid)
6770 for (i = 0; i < sav->sav_npending; i++) {
6771 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
6772 &spareguid) == 0 && spareguid == guid)
6780 * Check if a pool has an active shared spare device.
6781 * Note: reference count of an active spare is 2, as a spare and as a replace
6784 spa_has_active_shared_spare(spa_t *spa)
6788 spa_aux_vdev_t *sav = &spa->spa_spares;
6790 for (i = 0; i < sav->sav_count; i++) {
6791 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
6792 &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
6801 * Post a sysevent corresponding to the given event. The 'name' must be one of
6802 * the event definitions in sys/sysevent/eventdefs.h. The payload will be
6803 * filled in from the spa and (optionally) the vdev. This doesn't do anything
6804 * in the userland libzpool, as we don't want consumers to misinterpret ztest
6805 * or zdb as real changes.
6808 spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
6812 sysevent_attr_list_t *attr = NULL;
6813 sysevent_value_t value;
6816 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
6819 value.value_type = SE_DATA_TYPE_STRING;
6820 value.value.sv_string = spa_name(spa);
6821 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
6824 value.value_type = SE_DATA_TYPE_UINT64;
6825 value.value.sv_uint64 = spa_guid(spa);
6826 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
6830 value.value_type = SE_DATA_TYPE_UINT64;
6831 value.value.sv_uint64 = vd->vdev_guid;
6832 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
6836 if (vd->vdev_path) {
6837 value.value_type = SE_DATA_TYPE_STRING;
6838 value.value.sv_string = vd->vdev_path;
6839 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
6840 &value, SE_SLEEP) != 0)
6845 if (sysevent_attach_attributes(ev, attr) != 0)
6849 (void) log_sysevent(ev, SE_SLEEP, &eid);
6853 sysevent_free_attr(attr);