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);
1279 * Close the dsl pool.
1281 if (spa->spa_dsl_pool) {
1282 dsl_pool_close(spa->spa_dsl_pool);
1283 spa->spa_dsl_pool = NULL;
1284 spa->spa_meta_objset = NULL;
1289 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1292 * Drop and purge level 2 cache
1294 spa_l2cache_drop(spa);
1299 if (spa->spa_root_vdev)
1300 vdev_free(spa->spa_root_vdev);
1301 ASSERT(spa->spa_root_vdev == NULL);
1303 for (i = 0; i < spa->spa_spares.sav_count; i++)
1304 vdev_free(spa->spa_spares.sav_vdevs[i]);
1305 if (spa->spa_spares.sav_vdevs) {
1306 kmem_free(spa->spa_spares.sav_vdevs,
1307 spa->spa_spares.sav_count * sizeof (void *));
1308 spa->spa_spares.sav_vdevs = NULL;
1310 if (spa->spa_spares.sav_config) {
1311 nvlist_free(spa->spa_spares.sav_config);
1312 spa->spa_spares.sav_config = NULL;
1314 spa->spa_spares.sav_count = 0;
1316 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1317 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1318 vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1320 if (spa->spa_l2cache.sav_vdevs) {
1321 kmem_free(spa->spa_l2cache.sav_vdevs,
1322 spa->spa_l2cache.sav_count * sizeof (void *));
1323 spa->spa_l2cache.sav_vdevs = NULL;
1325 if (spa->spa_l2cache.sav_config) {
1326 nvlist_free(spa->spa_l2cache.sav_config);
1327 spa->spa_l2cache.sav_config = NULL;
1329 spa->spa_l2cache.sav_count = 0;
1331 spa->spa_async_suspended = 0;
1333 if (spa->spa_comment != NULL) {
1334 spa_strfree(spa->spa_comment);
1335 spa->spa_comment = NULL;
1338 spa_config_exit(spa, SCL_ALL, FTAG);
1342 * Load (or re-load) the current list of vdevs describing the active spares for
1343 * this pool. When this is called, we have some form of basic information in
1344 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1345 * then re-generate a more complete list including status information.
1348 spa_load_spares(spa_t *spa)
1355 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1358 * First, close and free any existing spare vdevs.
1360 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1361 vd = spa->spa_spares.sav_vdevs[i];
1363 /* Undo the call to spa_activate() below */
1364 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1365 B_FALSE)) != NULL && tvd->vdev_isspare)
1366 spa_spare_remove(tvd);
1371 if (spa->spa_spares.sav_vdevs)
1372 kmem_free(spa->spa_spares.sav_vdevs,
1373 spa->spa_spares.sav_count * sizeof (void *));
1375 if (spa->spa_spares.sav_config == NULL)
1378 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1379 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1381 spa->spa_spares.sav_count = (int)nspares;
1382 spa->spa_spares.sav_vdevs = NULL;
1388 * Construct the array of vdevs, opening them to get status in the
1389 * process. For each spare, there is potentially two different vdev_t
1390 * structures associated with it: one in the list of spares (used only
1391 * for basic validation purposes) and one in the active vdev
1392 * configuration (if it's spared in). During this phase we open and
1393 * validate each vdev on the spare list. If the vdev also exists in the
1394 * active configuration, then we also mark this vdev as an active spare.
1396 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1398 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1399 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1400 VDEV_ALLOC_SPARE) == 0);
1403 spa->spa_spares.sav_vdevs[i] = vd;
1405 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1406 B_FALSE)) != NULL) {
1407 if (!tvd->vdev_isspare)
1411 * We only mark the spare active if we were successfully
1412 * able to load the vdev. Otherwise, importing a pool
1413 * with a bad active spare would result in strange
1414 * behavior, because multiple pool would think the spare
1415 * is actively in use.
1417 * There is a vulnerability here to an equally bizarre
1418 * circumstance, where a dead active spare is later
1419 * brought back to life (onlined or otherwise). Given
1420 * the rarity of this scenario, and the extra complexity
1421 * it adds, we ignore the possibility.
1423 if (!vdev_is_dead(tvd))
1424 spa_spare_activate(tvd);
1428 vd->vdev_aux = &spa->spa_spares;
1430 if (vdev_open(vd) != 0)
1433 if (vdev_validate_aux(vd) == 0)
1438 * Recompute the stashed list of spares, with status information
1441 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1442 DATA_TYPE_NVLIST_ARRAY) == 0);
1444 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1446 for (i = 0; i < spa->spa_spares.sav_count; i++)
1447 spares[i] = vdev_config_generate(spa,
1448 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1449 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1450 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1451 for (i = 0; i < spa->spa_spares.sav_count; i++)
1452 nvlist_free(spares[i]);
1453 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1457 * Load (or re-load) the current list of vdevs describing the active l2cache for
1458 * this pool. When this is called, we have some form of basic information in
1459 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1460 * then re-generate a more complete list including status information.
1461 * Devices which are already active have their details maintained, and are
1465 spa_load_l2cache(spa_t *spa)
1469 int i, j, oldnvdevs;
1471 vdev_t *vd, **oldvdevs, **newvdevs;
1472 spa_aux_vdev_t *sav = &spa->spa_l2cache;
1474 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1476 if (sav->sav_config != NULL) {
1477 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1478 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1479 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1485 oldvdevs = sav->sav_vdevs;
1486 oldnvdevs = sav->sav_count;
1487 sav->sav_vdevs = NULL;
1491 * Process new nvlist of vdevs.
1493 for (i = 0; i < nl2cache; i++) {
1494 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1498 for (j = 0; j < oldnvdevs; j++) {
1500 if (vd != NULL && guid == vd->vdev_guid) {
1502 * Retain previous vdev for add/remove ops.
1510 if (newvdevs[i] == NULL) {
1514 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1515 VDEV_ALLOC_L2CACHE) == 0);
1520 * Commit this vdev as an l2cache device,
1521 * even if it fails to open.
1523 spa_l2cache_add(vd);
1528 spa_l2cache_activate(vd);
1530 if (vdev_open(vd) != 0)
1533 (void) vdev_validate_aux(vd);
1535 if (!vdev_is_dead(vd))
1536 l2arc_add_vdev(spa, vd);
1541 * Purge vdevs that were dropped
1543 for (i = 0; i < oldnvdevs; i++) {
1548 ASSERT(vd->vdev_isl2cache);
1550 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1551 pool != 0ULL && l2arc_vdev_present(vd))
1552 l2arc_remove_vdev(vd);
1553 vdev_clear_stats(vd);
1559 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1561 if (sav->sav_config == NULL)
1564 sav->sav_vdevs = newvdevs;
1565 sav->sav_count = (int)nl2cache;
1568 * Recompute the stashed list of l2cache devices, with status
1569 * information this time.
1571 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1572 DATA_TYPE_NVLIST_ARRAY) == 0);
1574 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1575 for (i = 0; i < sav->sav_count; i++)
1576 l2cache[i] = vdev_config_generate(spa,
1577 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1578 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1579 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1581 for (i = 0; i < sav->sav_count; i++)
1582 nvlist_free(l2cache[i]);
1584 kmem_free(l2cache, sav->sav_count * sizeof (void *));
1588 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1591 char *packed = NULL;
1596 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
1597 nvsize = *(uint64_t *)db->db_data;
1598 dmu_buf_rele(db, FTAG);
1600 packed = kmem_alloc(nvsize, KM_SLEEP);
1601 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1604 error = nvlist_unpack(packed, nvsize, value, 0);
1605 kmem_free(packed, nvsize);
1611 * Checks to see if the given vdev could not be opened, in which case we post a
1612 * sysevent to notify the autoreplace code that the device has been removed.
1615 spa_check_removed(vdev_t *vd)
1617 for (int c = 0; c < vd->vdev_children; c++)
1618 spa_check_removed(vd->vdev_child[c]);
1620 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) &&
1622 zfs_post_autoreplace(vd->vdev_spa, vd);
1623 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1628 * Validate the current config against the MOS config
1631 spa_config_valid(spa_t *spa, nvlist_t *config)
1633 vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1636 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1638 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1639 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1641 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1644 * If we're doing a normal import, then build up any additional
1645 * diagnostic information about missing devices in this config.
1646 * We'll pass this up to the user for further processing.
1648 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1649 nvlist_t **child, *nv;
1652 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1654 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1656 for (int c = 0; c < rvd->vdev_children; c++) {
1657 vdev_t *tvd = rvd->vdev_child[c];
1658 vdev_t *mtvd = mrvd->vdev_child[c];
1660 if (tvd->vdev_ops == &vdev_missing_ops &&
1661 mtvd->vdev_ops != &vdev_missing_ops &&
1663 child[idx++] = vdev_config_generate(spa, mtvd,
1668 VERIFY(nvlist_add_nvlist_array(nv,
1669 ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1670 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1671 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1673 for (int i = 0; i < idx; i++)
1674 nvlist_free(child[i]);
1677 kmem_free(child, rvd->vdev_children * sizeof (char **));
1681 * Compare the root vdev tree with the information we have
1682 * from the MOS config (mrvd). Check each top-level vdev
1683 * with the corresponding MOS config top-level (mtvd).
1685 for (int c = 0; c < rvd->vdev_children; c++) {
1686 vdev_t *tvd = rvd->vdev_child[c];
1687 vdev_t *mtvd = mrvd->vdev_child[c];
1690 * Resolve any "missing" vdevs in the current configuration.
1691 * If we find that the MOS config has more accurate information
1692 * about the top-level vdev then use that vdev instead.
1694 if (tvd->vdev_ops == &vdev_missing_ops &&
1695 mtvd->vdev_ops != &vdev_missing_ops) {
1697 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1701 * Device specific actions.
1703 if (mtvd->vdev_islog) {
1704 spa_set_log_state(spa, SPA_LOG_CLEAR);
1707 * XXX - once we have 'readonly' pool
1708 * support we should be able to handle
1709 * missing data devices by transitioning
1710 * the pool to readonly.
1716 * Swap the missing vdev with the data we were
1717 * able to obtain from the MOS config.
1719 vdev_remove_child(rvd, tvd);
1720 vdev_remove_child(mrvd, mtvd);
1722 vdev_add_child(rvd, mtvd);
1723 vdev_add_child(mrvd, tvd);
1725 spa_config_exit(spa, SCL_ALL, FTAG);
1727 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1730 } else if (mtvd->vdev_islog) {
1732 * Load the slog device's state from the MOS config
1733 * since it's possible that the label does not
1734 * contain the most up-to-date information.
1736 vdev_load_log_state(tvd, mtvd);
1741 spa_config_exit(spa, SCL_ALL, FTAG);
1744 * Ensure we were able to validate the config.
1746 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1750 * Check for missing log devices
1753 spa_check_logs(spa_t *spa)
1755 boolean_t rv = B_FALSE;
1757 switch (spa->spa_log_state) {
1758 case SPA_LOG_MISSING:
1759 /* need to recheck in case slog has been restored */
1760 case SPA_LOG_UNKNOWN:
1761 rv = (dmu_objset_find(spa->spa_name, zil_check_log_chain,
1762 NULL, DS_FIND_CHILDREN) != 0);
1764 spa_set_log_state(spa, SPA_LOG_MISSING);
1771 spa_passivate_log(spa_t *spa)
1773 vdev_t *rvd = spa->spa_root_vdev;
1774 boolean_t slog_found = B_FALSE;
1776 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1778 if (!spa_has_slogs(spa))
1781 for (int c = 0; c < rvd->vdev_children; c++) {
1782 vdev_t *tvd = rvd->vdev_child[c];
1783 metaslab_group_t *mg = tvd->vdev_mg;
1785 if (tvd->vdev_islog) {
1786 metaslab_group_passivate(mg);
1787 slog_found = B_TRUE;
1791 return (slog_found);
1795 spa_activate_log(spa_t *spa)
1797 vdev_t *rvd = spa->spa_root_vdev;
1799 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1801 for (int c = 0; c < rvd->vdev_children; c++) {
1802 vdev_t *tvd = rvd->vdev_child[c];
1803 metaslab_group_t *mg = tvd->vdev_mg;
1805 if (tvd->vdev_islog)
1806 metaslab_group_activate(mg);
1811 spa_offline_log(spa_t *spa)
1815 error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1816 NULL, DS_FIND_CHILDREN);
1819 * We successfully offlined the log device, sync out the
1820 * current txg so that the "stubby" block can be removed
1823 txg_wait_synced(spa->spa_dsl_pool, 0);
1829 spa_aux_check_removed(spa_aux_vdev_t *sav)
1833 for (i = 0; i < sav->sav_count; i++)
1834 spa_check_removed(sav->sav_vdevs[i]);
1838 spa_claim_notify(zio_t *zio)
1840 spa_t *spa = zio->io_spa;
1845 mutex_enter(&spa->spa_props_lock); /* any mutex will do */
1846 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1847 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1848 mutex_exit(&spa->spa_props_lock);
1851 typedef struct spa_load_error {
1852 uint64_t sle_meta_count;
1853 uint64_t sle_data_count;
1857 spa_load_verify_done(zio_t *zio)
1859 blkptr_t *bp = zio->io_bp;
1860 spa_load_error_t *sle = zio->io_private;
1861 dmu_object_type_t type = BP_GET_TYPE(bp);
1862 int error = zio->io_error;
1865 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1866 type != DMU_OT_INTENT_LOG)
1867 atomic_add_64(&sle->sle_meta_count, 1);
1869 atomic_add_64(&sle->sle_data_count, 1);
1871 zio_data_buf_free(zio->io_data, zio->io_size);
1876 spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1877 const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1881 size_t size = BP_GET_PSIZE(bp);
1882 void *data = zio_data_buf_alloc(size);
1884 zio_nowait(zio_read(rio, spa, bp, data, size,
1885 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1886 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1887 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1893 spa_load_verify(spa_t *spa)
1896 spa_load_error_t sle = { 0 };
1897 zpool_rewind_policy_t policy;
1898 boolean_t verify_ok = B_FALSE;
1901 zpool_get_rewind_policy(spa->spa_config, &policy);
1903 if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1906 rio = zio_root(spa, NULL, &sle,
1907 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1909 error = traverse_pool(spa, spa->spa_verify_min_txg,
1910 TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio);
1912 (void) zio_wait(rio);
1914 spa->spa_load_meta_errors = sle.sle_meta_count;
1915 spa->spa_load_data_errors = sle.sle_data_count;
1917 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1918 sle.sle_data_count <= policy.zrp_maxdata) {
1922 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1923 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1925 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1926 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1927 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1928 VERIFY(nvlist_add_int64(spa->spa_load_info,
1929 ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1930 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1931 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1933 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1937 if (error != ENXIO && error != EIO)
1938 error = SET_ERROR(EIO);
1942 return (verify_ok ? 0 : EIO);
1946 * Find a value in the pool props object.
1949 spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1951 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1952 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1956 * Find a value in the pool directory object.
1959 spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1961 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1962 name, sizeof (uint64_t), 1, val));
1966 spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1968 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1973 * Fix up config after a partly-completed split. This is done with the
1974 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
1975 * pool have that entry in their config, but only the splitting one contains
1976 * a list of all the guids of the vdevs that are being split off.
1978 * This function determines what to do with that list: either rejoin
1979 * all the disks to the pool, or complete the splitting process. To attempt
1980 * the rejoin, each disk that is offlined is marked online again, and
1981 * we do a reopen() call. If the vdev label for every disk that was
1982 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1983 * then we call vdev_split() on each disk, and complete the split.
1985 * Otherwise we leave the config alone, with all the vdevs in place in
1986 * the original pool.
1989 spa_try_repair(spa_t *spa, nvlist_t *config)
1996 boolean_t attempt_reopen;
1998 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
2001 /* check that the config is complete */
2002 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
2003 &glist, &gcount) != 0)
2006 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
2008 /* attempt to online all the vdevs & validate */
2009 attempt_reopen = B_TRUE;
2010 for (i = 0; i < gcount; i++) {
2011 if (glist[i] == 0) /* vdev is hole */
2014 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
2015 if (vd[i] == NULL) {
2017 * Don't bother attempting to reopen the disks;
2018 * just do the split.
2020 attempt_reopen = B_FALSE;
2022 /* attempt to re-online it */
2023 vd[i]->vdev_offline = B_FALSE;
2027 if (attempt_reopen) {
2028 vdev_reopen(spa->spa_root_vdev);
2030 /* check each device to see what state it's in */
2031 for (extracted = 0, i = 0; i < gcount; i++) {
2032 if (vd[i] != NULL &&
2033 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
2040 * If every disk has been moved to the new pool, or if we never
2041 * even attempted to look at them, then we split them off for
2044 if (!attempt_reopen || gcount == extracted) {
2045 for (i = 0; i < gcount; i++)
2048 vdev_reopen(spa->spa_root_vdev);
2051 kmem_free(vd, gcount * sizeof (vdev_t *));
2055 spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
2056 boolean_t mosconfig)
2058 nvlist_t *config = spa->spa_config;
2059 char *ereport = FM_EREPORT_ZFS_POOL;
2065 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
2066 return (SET_ERROR(EINVAL));
2068 ASSERT(spa->spa_comment == NULL);
2069 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
2070 spa->spa_comment = spa_strdup(comment);
2073 * Versioning wasn't explicitly added to the label until later, so if
2074 * it's not present treat it as the initial version.
2076 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
2077 &spa->spa_ubsync.ub_version) != 0)
2078 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
2080 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
2081 &spa->spa_config_txg);
2083 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
2084 spa_guid_exists(pool_guid, 0)) {
2085 error = SET_ERROR(EEXIST);
2087 spa->spa_config_guid = pool_guid;
2089 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
2091 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
2095 nvlist_free(spa->spa_load_info);
2096 spa->spa_load_info = fnvlist_alloc();
2098 gethrestime(&spa->spa_loaded_ts);
2099 error = spa_load_impl(spa, pool_guid, config, state, type,
2100 mosconfig, &ereport);
2103 spa->spa_minref = refcount_count(&spa->spa_refcount);
2105 if (error != EEXIST) {
2106 spa->spa_loaded_ts.tv_sec = 0;
2107 spa->spa_loaded_ts.tv_nsec = 0;
2109 if (error != EBADF) {
2110 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
2113 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2120 * Load an existing storage pool, using the pool's builtin spa_config as a
2121 * source of configuration information.
2124 spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
2125 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
2129 nvlist_t *nvroot = NULL;
2132 uberblock_t *ub = &spa->spa_uberblock;
2133 uint64_t children, config_cache_txg = spa->spa_config_txg;
2134 int orig_mode = spa->spa_mode;
2137 boolean_t missing_feat_write = B_FALSE;
2140 * If this is an untrusted config, access the pool in read-only mode.
2141 * This prevents things like resilvering recently removed devices.
2144 spa->spa_mode = FREAD;
2146 ASSERT(MUTEX_HELD(&spa_namespace_lock));
2148 spa->spa_load_state = state;
2150 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
2151 return (SET_ERROR(EINVAL));
2153 parse = (type == SPA_IMPORT_EXISTING ?
2154 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2157 * Create "The Godfather" zio to hold all async IOs
2159 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2160 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2163 * Parse the configuration into a vdev tree. We explicitly set the
2164 * value that will be returned by spa_version() since parsing the
2165 * configuration requires knowing the version number.
2167 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2168 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
2169 spa_config_exit(spa, SCL_ALL, FTAG);
2174 ASSERT(spa->spa_root_vdev == rvd);
2176 if (type != SPA_IMPORT_ASSEMBLE) {
2177 ASSERT(spa_guid(spa) == pool_guid);
2181 * Try to open all vdevs, loading each label in the process.
2183 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2184 error = vdev_open(rvd);
2185 spa_config_exit(spa, SCL_ALL, FTAG);
2190 * We need to validate the vdev labels against the configuration that
2191 * we have in hand, which is dependent on the setting of mosconfig. If
2192 * mosconfig is true then we're validating the vdev labels based on
2193 * that config. Otherwise, we're validating against the cached config
2194 * (zpool.cache) that was read when we loaded the zfs module, and then
2195 * later we will recursively call spa_load() and validate against
2198 * If we're assembling a new pool that's been split off from an
2199 * existing pool, the labels haven't yet been updated so we skip
2200 * validation for now.
2202 if (type != SPA_IMPORT_ASSEMBLE) {
2203 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2204 error = vdev_validate(rvd, mosconfig);
2205 spa_config_exit(spa, SCL_ALL, FTAG);
2210 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2211 return (SET_ERROR(ENXIO));
2215 * Find the best uberblock.
2217 vdev_uberblock_load(rvd, ub, &label);
2220 * If we weren't able to find a single valid uberblock, return failure.
2222 if (ub->ub_txg == 0) {
2224 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2228 * If the pool has an unsupported version we can't open it.
2230 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2232 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2235 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2239 * If we weren't able to find what's necessary for reading the
2240 * MOS in the label, return failure.
2242 if (label == NULL || nvlist_lookup_nvlist(label,
2243 ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) {
2245 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2250 * Update our in-core representation with the definitive values
2253 nvlist_free(spa->spa_label_features);
2254 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2260 * Look through entries in the label nvlist's features_for_read. If
2261 * there is a feature listed there which we don't understand then we
2262 * cannot open a pool.
2264 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2265 nvlist_t *unsup_feat;
2267 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2270 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2272 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2273 if (!zfeature_is_supported(nvpair_name(nvp))) {
2274 VERIFY(nvlist_add_string(unsup_feat,
2275 nvpair_name(nvp), "") == 0);
2279 if (!nvlist_empty(unsup_feat)) {
2280 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2281 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2282 nvlist_free(unsup_feat);
2283 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2287 nvlist_free(unsup_feat);
2291 * If the vdev guid sum doesn't match the uberblock, we have an
2292 * incomplete configuration. We first check to see if the pool
2293 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2294 * If it is, defer the vdev_guid_sum check till later so we
2295 * can handle missing vdevs.
2297 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
2298 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
2299 rvd->vdev_guid_sum != ub->ub_guid_sum)
2300 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2302 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2303 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2304 spa_try_repair(spa, config);
2305 spa_config_exit(spa, SCL_ALL, FTAG);
2306 nvlist_free(spa->spa_config_splitting);
2307 spa->spa_config_splitting = NULL;
2311 * Initialize internal SPA structures.
2313 spa->spa_state = POOL_STATE_ACTIVE;
2314 spa->spa_ubsync = spa->spa_uberblock;
2315 spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2316 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2317 spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2318 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2319 spa->spa_claim_max_txg = spa->spa_first_txg;
2320 spa->spa_prev_software_version = ub->ub_software_version;
2322 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2324 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2325 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2327 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2328 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2330 if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2331 boolean_t missing_feat_read = B_FALSE;
2332 nvlist_t *unsup_feat, *enabled_feat;
2334 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2335 &spa->spa_feat_for_read_obj) != 0) {
2336 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2339 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2340 &spa->spa_feat_for_write_obj) != 0) {
2341 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2344 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2345 &spa->spa_feat_desc_obj) != 0) {
2346 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2349 enabled_feat = fnvlist_alloc();
2350 unsup_feat = fnvlist_alloc();
2352 if (!feature_is_supported(spa->spa_meta_objset,
2353 spa->spa_feat_for_read_obj, spa->spa_feat_desc_obj,
2354 unsup_feat, enabled_feat))
2355 missing_feat_read = B_TRUE;
2357 if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) {
2358 if (!feature_is_supported(spa->spa_meta_objset,
2359 spa->spa_feat_for_write_obj, spa->spa_feat_desc_obj,
2360 unsup_feat, enabled_feat)) {
2361 missing_feat_write = B_TRUE;
2365 fnvlist_add_nvlist(spa->spa_load_info,
2366 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
2368 if (!nvlist_empty(unsup_feat)) {
2369 fnvlist_add_nvlist(spa->spa_load_info,
2370 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
2373 fnvlist_free(enabled_feat);
2374 fnvlist_free(unsup_feat);
2376 if (!missing_feat_read) {
2377 fnvlist_add_boolean(spa->spa_load_info,
2378 ZPOOL_CONFIG_CAN_RDONLY);
2382 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2383 * twofold: to determine whether the pool is available for
2384 * import in read-write mode and (if it is not) whether the
2385 * pool is available for import in read-only mode. If the pool
2386 * is available for import in read-write mode, it is displayed
2387 * as available in userland; if it is not available for import
2388 * in read-only mode, it is displayed as unavailable in
2389 * userland. If the pool is available for import in read-only
2390 * mode but not read-write mode, it is displayed as unavailable
2391 * in userland with a special note that the pool is actually
2392 * available for open in read-only mode.
2394 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2395 * missing a feature for write, we must first determine whether
2396 * the pool can be opened read-only before returning to
2397 * userland in order to know whether to display the
2398 * abovementioned note.
2400 if (missing_feat_read || (missing_feat_write &&
2401 spa_writeable(spa))) {
2402 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2407 spa->spa_is_initializing = B_TRUE;
2408 error = dsl_pool_open(spa->spa_dsl_pool);
2409 spa->spa_is_initializing = B_FALSE;
2411 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2415 nvlist_t *policy = NULL, *nvconfig;
2417 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2418 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2420 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2421 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2423 unsigned long myhostid = 0;
2425 VERIFY(nvlist_lookup_string(nvconfig,
2426 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2429 myhostid = zone_get_hostid(NULL);
2432 * We're emulating the system's hostid in userland, so
2433 * we can't use zone_get_hostid().
2435 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2436 #endif /* _KERNEL */
2437 if (check_hostid && hostid != 0 && myhostid != 0 &&
2438 hostid != myhostid) {
2439 nvlist_free(nvconfig);
2440 cmn_err(CE_WARN, "pool '%s' could not be "
2441 "loaded as it was last accessed by "
2442 "another system (host: %s hostid: 0x%lx). "
2443 "See: http://illumos.org/msg/ZFS-8000-EY",
2444 spa_name(spa), hostname,
2445 (unsigned long)hostid);
2446 return (SET_ERROR(EBADF));
2449 if (nvlist_lookup_nvlist(spa->spa_config,
2450 ZPOOL_REWIND_POLICY, &policy) == 0)
2451 VERIFY(nvlist_add_nvlist(nvconfig,
2452 ZPOOL_REWIND_POLICY, policy) == 0);
2454 spa_config_set(spa, nvconfig);
2456 spa_deactivate(spa);
2457 spa_activate(spa, orig_mode);
2459 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2462 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2463 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2464 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2466 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2469 * Load the bit that tells us to use the new accounting function
2470 * (raid-z deflation). If we have an older pool, this will not
2473 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2474 if (error != 0 && error != ENOENT)
2475 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2477 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2478 &spa->spa_creation_version);
2479 if (error != 0 && error != ENOENT)
2480 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2483 * Load the persistent error log. If we have an older pool, this will
2486 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2487 if (error != 0 && error != ENOENT)
2488 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2490 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2491 &spa->spa_errlog_scrub);
2492 if (error != 0 && error != ENOENT)
2493 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2496 * Load the history object. If we have an older pool, this
2497 * will not be present.
2499 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2500 if (error != 0 && error != ENOENT)
2501 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2504 * If we're assembling the pool from the split-off vdevs of
2505 * an existing pool, we don't want to attach the spares & cache
2510 * Load any hot spares for this pool.
2512 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2513 if (error != 0 && error != ENOENT)
2514 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2515 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2516 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2517 if (load_nvlist(spa, spa->spa_spares.sav_object,
2518 &spa->spa_spares.sav_config) != 0)
2519 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2521 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2522 spa_load_spares(spa);
2523 spa_config_exit(spa, SCL_ALL, FTAG);
2524 } else if (error == 0) {
2525 spa->spa_spares.sav_sync = B_TRUE;
2529 * Load any level 2 ARC devices for this pool.
2531 error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2532 &spa->spa_l2cache.sav_object);
2533 if (error != 0 && error != ENOENT)
2534 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2535 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2536 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2537 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2538 &spa->spa_l2cache.sav_config) != 0)
2539 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2541 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2542 spa_load_l2cache(spa);
2543 spa_config_exit(spa, SCL_ALL, FTAG);
2544 } else if (error == 0) {
2545 spa->spa_l2cache.sav_sync = B_TRUE;
2548 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2550 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2551 if (error && error != ENOENT)
2552 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2555 uint64_t autoreplace;
2557 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2558 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2559 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2560 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2561 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2562 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2563 &spa->spa_dedup_ditto);
2565 spa->spa_autoreplace = (autoreplace != 0);
2569 * If the 'autoreplace' property is set, then post a resource notifying
2570 * the ZFS DE that it should not issue any faults for unopenable
2571 * devices. We also iterate over the vdevs, and post a sysevent for any
2572 * unopenable vdevs so that the normal autoreplace handler can take
2575 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2576 spa_check_removed(spa->spa_root_vdev);
2578 * For the import case, this is done in spa_import(), because
2579 * at this point we're using the spare definitions from
2580 * the MOS config, not necessarily from the userland config.
2582 if (state != SPA_LOAD_IMPORT) {
2583 spa_aux_check_removed(&spa->spa_spares);
2584 spa_aux_check_removed(&spa->spa_l2cache);
2589 * Load the vdev state for all toplevel vdevs.
2594 * Propagate the leaf DTLs we just loaded all the way up the tree.
2596 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2597 vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2598 spa_config_exit(spa, SCL_ALL, FTAG);
2601 * Load the DDTs (dedup tables).
2603 error = ddt_load(spa);
2605 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2607 spa_update_dspace(spa);
2610 * Validate the config, using the MOS config to fill in any
2611 * information which might be missing. If we fail to validate
2612 * the config then declare the pool unfit for use. If we're
2613 * assembling a pool from a split, the log is not transferred
2616 if (type != SPA_IMPORT_ASSEMBLE) {
2619 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2620 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2622 if (!spa_config_valid(spa, nvconfig)) {
2623 nvlist_free(nvconfig);
2624 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2627 nvlist_free(nvconfig);
2630 * Now that we've validated the config, check the state of the
2631 * root vdev. If it can't be opened, it indicates one or
2632 * more toplevel vdevs are faulted.
2634 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2635 return (SET_ERROR(ENXIO));
2637 if (spa_check_logs(spa)) {
2638 *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2639 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2643 if (missing_feat_write) {
2644 ASSERT(state == SPA_LOAD_TRYIMPORT);
2647 * At this point, we know that we can open the pool in
2648 * read-only mode but not read-write mode. We now have enough
2649 * information and can return to userland.
2651 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
2655 * We've successfully opened the pool, verify that we're ready
2656 * to start pushing transactions.
2658 if (state != SPA_LOAD_TRYIMPORT) {
2659 if (error = spa_load_verify(spa))
2660 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2664 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2665 spa->spa_load_max_txg == UINT64_MAX)) {
2667 int need_update = B_FALSE;
2669 ASSERT(state != SPA_LOAD_TRYIMPORT);
2672 * Claim log blocks that haven't been committed yet.
2673 * This must all happen in a single txg.
2674 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2675 * invoked from zil_claim_log_block()'s i/o done callback.
2676 * Price of rollback is that we abandon the log.
2678 spa->spa_claiming = B_TRUE;
2680 tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2681 spa_first_txg(spa));
2682 (void) dmu_objset_find(spa_name(spa),
2683 zil_claim, tx, DS_FIND_CHILDREN);
2686 spa->spa_claiming = B_FALSE;
2688 spa_set_log_state(spa, SPA_LOG_GOOD);
2689 spa->spa_sync_on = B_TRUE;
2690 txg_sync_start(spa->spa_dsl_pool);
2693 * Wait for all claims to sync. We sync up to the highest
2694 * claimed log block birth time so that claimed log blocks
2695 * don't appear to be from the future. spa_claim_max_txg
2696 * will have been set for us by either zil_check_log_chain()
2697 * (invoked from spa_check_logs()) or zil_claim() above.
2699 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2702 * If the config cache is stale, or we have uninitialized
2703 * metaslabs (see spa_vdev_add()), then update the config.
2705 * If this is a verbatim import, trust the current
2706 * in-core spa_config and update the disk labels.
2708 if (config_cache_txg != spa->spa_config_txg ||
2709 state == SPA_LOAD_IMPORT ||
2710 state == SPA_LOAD_RECOVER ||
2711 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2712 need_update = B_TRUE;
2714 for (int c = 0; c < rvd->vdev_children; c++)
2715 if (rvd->vdev_child[c]->vdev_ms_array == 0)
2716 need_update = B_TRUE;
2719 * Update the config cache asychronously in case we're the
2720 * root pool, in which case the config cache isn't writable yet.
2723 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2726 * Check all DTLs to see if anything needs resilvering.
2728 if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2729 vdev_resilver_needed(rvd, NULL, NULL))
2730 spa_async_request(spa, SPA_ASYNC_RESILVER);
2733 * Log the fact that we booted up (so that we can detect if
2734 * we rebooted in the middle of an operation).
2736 spa_history_log_version(spa, "open");
2739 * Delete any inconsistent datasets.
2741 (void) dmu_objset_find(spa_name(spa),
2742 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2745 * Clean up any stale temporary dataset userrefs.
2747 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2754 spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2756 int mode = spa->spa_mode;
2759 spa_deactivate(spa);
2761 spa->spa_load_max_txg--;
2763 spa_activate(spa, mode);
2764 spa_async_suspend(spa);
2766 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2770 * If spa_load() fails this function will try loading prior txg's. If
2771 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2772 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2773 * function will not rewind the pool and will return the same error as
2777 spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2778 uint64_t max_request, int rewind_flags)
2780 nvlist_t *loadinfo = NULL;
2781 nvlist_t *config = NULL;
2782 int load_error, rewind_error;
2783 uint64_t safe_rewind_txg;
2786 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2787 spa->spa_load_max_txg = spa->spa_load_txg;
2788 spa_set_log_state(spa, SPA_LOG_CLEAR);
2790 spa->spa_load_max_txg = max_request;
2793 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2795 if (load_error == 0)
2798 if (spa->spa_root_vdev != NULL)
2799 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2801 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2802 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2804 if (rewind_flags & ZPOOL_NEVER_REWIND) {
2805 nvlist_free(config);
2806 return (load_error);
2809 if (state == SPA_LOAD_RECOVER) {
2810 /* Price of rolling back is discarding txgs, including log */
2811 spa_set_log_state(spa, SPA_LOG_CLEAR);
2814 * If we aren't rolling back save the load info from our first
2815 * import attempt so that we can restore it after attempting
2818 loadinfo = spa->spa_load_info;
2819 spa->spa_load_info = fnvlist_alloc();
2822 spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2823 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2824 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2825 TXG_INITIAL : safe_rewind_txg;
2828 * Continue as long as we're finding errors, we're still within
2829 * the acceptable rewind range, and we're still finding uberblocks
2831 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2832 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2833 if (spa->spa_load_max_txg < safe_rewind_txg)
2834 spa->spa_extreme_rewind = B_TRUE;
2835 rewind_error = spa_load_retry(spa, state, mosconfig);
2838 spa->spa_extreme_rewind = B_FALSE;
2839 spa->spa_load_max_txg = UINT64_MAX;
2841 if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2842 spa_config_set(spa, config);
2844 if (state == SPA_LOAD_RECOVER) {
2845 ASSERT3P(loadinfo, ==, NULL);
2846 return (rewind_error);
2848 /* Store the rewind info as part of the initial load info */
2849 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
2850 spa->spa_load_info);
2852 /* Restore the initial load info */
2853 fnvlist_free(spa->spa_load_info);
2854 spa->spa_load_info = loadinfo;
2856 return (load_error);
2863 * The import case is identical to an open except that the configuration is sent
2864 * down from userland, instead of grabbed from the configuration cache. For the
2865 * case of an open, the pool configuration will exist in the
2866 * POOL_STATE_UNINITIALIZED state.
2868 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2869 * the same time open the pool, without having to keep around the spa_t in some
2873 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2877 spa_load_state_t state = SPA_LOAD_OPEN;
2879 int locked = B_FALSE;
2880 int firstopen = B_FALSE;
2885 * As disgusting as this is, we need to support recursive calls to this
2886 * function because dsl_dir_open() is called during spa_load(), and ends
2887 * up calling spa_open() again. The real fix is to figure out how to
2888 * avoid dsl_dir_open() calling this in the first place.
2890 if (mutex_owner(&spa_namespace_lock) != curthread) {
2891 mutex_enter(&spa_namespace_lock);
2895 if ((spa = spa_lookup(pool)) == NULL) {
2897 mutex_exit(&spa_namespace_lock);
2898 return (SET_ERROR(ENOENT));
2901 if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2902 zpool_rewind_policy_t policy;
2906 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2908 if (policy.zrp_request & ZPOOL_DO_REWIND)
2909 state = SPA_LOAD_RECOVER;
2911 spa_activate(spa, spa_mode_global);
2913 if (state != SPA_LOAD_RECOVER)
2914 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2916 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2917 policy.zrp_request);
2919 if (error == EBADF) {
2921 * If vdev_validate() returns failure (indicated by
2922 * EBADF), it indicates that one of the vdevs indicates
2923 * that the pool has been exported or destroyed. If
2924 * this is the case, the config cache is out of sync and
2925 * we should remove the pool from the namespace.
2928 spa_deactivate(spa);
2929 spa_config_sync(spa, B_TRUE, B_TRUE);
2932 mutex_exit(&spa_namespace_lock);
2933 return (SET_ERROR(ENOENT));
2938 * We can't open the pool, but we still have useful
2939 * information: the state of each vdev after the
2940 * attempted vdev_open(). Return this to the user.
2942 if (config != NULL && spa->spa_config) {
2943 VERIFY(nvlist_dup(spa->spa_config, config,
2945 VERIFY(nvlist_add_nvlist(*config,
2946 ZPOOL_CONFIG_LOAD_INFO,
2947 spa->spa_load_info) == 0);
2950 spa_deactivate(spa);
2951 spa->spa_last_open_failed = error;
2953 mutex_exit(&spa_namespace_lock);
2959 spa_open_ref(spa, tag);
2962 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2965 * If we've recovered the pool, pass back any information we
2966 * gathered while doing the load.
2968 if (state == SPA_LOAD_RECOVER) {
2969 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
2970 spa->spa_load_info) == 0);
2974 spa->spa_last_open_failed = 0;
2975 spa->spa_last_ubsync_txg = 0;
2976 spa->spa_load_txg = 0;
2977 mutex_exit(&spa_namespace_lock);
2981 zvol_create_minors(spa->spa_name);
2992 spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
2995 return (spa_open_common(name, spapp, tag, policy, config));
2999 spa_open(const char *name, spa_t **spapp, void *tag)
3001 return (spa_open_common(name, spapp, tag, NULL, NULL));
3005 * Lookup the given spa_t, incrementing the inject count in the process,
3006 * preventing it from being exported or destroyed.
3009 spa_inject_addref(char *name)
3013 mutex_enter(&spa_namespace_lock);
3014 if ((spa = spa_lookup(name)) == NULL) {
3015 mutex_exit(&spa_namespace_lock);
3018 spa->spa_inject_ref++;
3019 mutex_exit(&spa_namespace_lock);
3025 spa_inject_delref(spa_t *spa)
3027 mutex_enter(&spa_namespace_lock);
3028 spa->spa_inject_ref--;
3029 mutex_exit(&spa_namespace_lock);
3033 * Add spares device information to the nvlist.
3036 spa_add_spares(spa_t *spa, nvlist_t *config)
3046 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3048 if (spa->spa_spares.sav_count == 0)
3051 VERIFY(nvlist_lookup_nvlist(config,
3052 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3053 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
3054 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3056 VERIFY(nvlist_add_nvlist_array(nvroot,
3057 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3058 VERIFY(nvlist_lookup_nvlist_array(nvroot,
3059 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3062 * Go through and find any spares which have since been
3063 * repurposed as an active spare. If this is the case, update
3064 * their status appropriately.
3066 for (i = 0; i < nspares; i++) {
3067 VERIFY(nvlist_lookup_uint64(spares[i],
3068 ZPOOL_CONFIG_GUID, &guid) == 0);
3069 if (spa_spare_exists(guid, &pool, NULL) &&
3071 VERIFY(nvlist_lookup_uint64_array(
3072 spares[i], ZPOOL_CONFIG_VDEV_STATS,
3073 (uint64_t **)&vs, &vsc) == 0);
3074 vs->vs_state = VDEV_STATE_CANT_OPEN;
3075 vs->vs_aux = VDEV_AUX_SPARED;
3082 * Add l2cache device information to the nvlist, including vdev stats.
3085 spa_add_l2cache(spa_t *spa, nvlist_t *config)
3088 uint_t i, j, nl2cache;
3095 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3097 if (spa->spa_l2cache.sav_count == 0)
3100 VERIFY(nvlist_lookup_nvlist(config,
3101 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3102 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
3103 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3104 if (nl2cache != 0) {
3105 VERIFY(nvlist_add_nvlist_array(nvroot,
3106 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3107 VERIFY(nvlist_lookup_nvlist_array(nvroot,
3108 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3111 * Update level 2 cache device stats.
3114 for (i = 0; i < nl2cache; i++) {
3115 VERIFY(nvlist_lookup_uint64(l2cache[i],
3116 ZPOOL_CONFIG_GUID, &guid) == 0);
3119 for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
3121 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
3122 vd = spa->spa_l2cache.sav_vdevs[j];
3128 VERIFY(nvlist_lookup_uint64_array(l2cache[i],
3129 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
3131 vdev_get_stats(vd, vs);
3137 spa_add_feature_stats(spa_t *spa, nvlist_t *config)
3143 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3144 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3146 if (spa->spa_feat_for_read_obj != 0) {
3147 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3148 spa->spa_feat_for_read_obj);
3149 zap_cursor_retrieve(&zc, &za) == 0;
3150 zap_cursor_advance(&zc)) {
3151 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3152 za.za_num_integers == 1);
3153 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3154 za.za_first_integer));
3156 zap_cursor_fini(&zc);
3159 if (spa->spa_feat_for_write_obj != 0) {
3160 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3161 spa->spa_feat_for_write_obj);
3162 zap_cursor_retrieve(&zc, &za) == 0;
3163 zap_cursor_advance(&zc)) {
3164 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3165 za.za_num_integers == 1);
3166 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3167 za.za_first_integer));
3169 zap_cursor_fini(&zc);
3172 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
3174 nvlist_free(features);
3178 spa_get_stats(const char *name, nvlist_t **config,
3179 char *altroot, size_t buflen)
3185 error = spa_open_common(name, &spa, FTAG, NULL, config);
3189 * This still leaves a window of inconsistency where the spares
3190 * or l2cache devices could change and the config would be
3191 * self-inconsistent.
3193 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3195 if (*config != NULL) {
3196 uint64_t loadtimes[2];
3198 loadtimes[0] = spa->spa_loaded_ts.tv_sec;
3199 loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
3200 VERIFY(nvlist_add_uint64_array(*config,
3201 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
3203 VERIFY(nvlist_add_uint64(*config,
3204 ZPOOL_CONFIG_ERRCOUNT,
3205 spa_get_errlog_size(spa)) == 0);
3207 if (spa_suspended(spa))
3208 VERIFY(nvlist_add_uint64(*config,
3209 ZPOOL_CONFIG_SUSPENDED,
3210 spa->spa_failmode) == 0);
3212 spa_add_spares(spa, *config);
3213 spa_add_l2cache(spa, *config);
3214 spa_add_feature_stats(spa, *config);
3219 * We want to get the alternate root even for faulted pools, so we cheat
3220 * and call spa_lookup() directly.
3224 mutex_enter(&spa_namespace_lock);
3225 spa = spa_lookup(name);
3227 spa_altroot(spa, altroot, buflen);
3231 mutex_exit(&spa_namespace_lock);
3233 spa_altroot(spa, altroot, buflen);
3238 spa_config_exit(spa, SCL_CONFIG, FTAG);
3239 spa_close(spa, FTAG);
3246 * Validate that the auxiliary device array is well formed. We must have an
3247 * array of nvlists, each which describes a valid leaf vdev. If this is an
3248 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3249 * specified, as long as they are well-formed.
3252 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
3253 spa_aux_vdev_t *sav, const char *config, uint64_t version,
3254 vdev_labeltype_t label)
3261 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3264 * It's acceptable to have no devs specified.
3266 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
3270 return (SET_ERROR(EINVAL));
3273 * Make sure the pool is formatted with a version that supports this
3276 if (spa_version(spa) < version)
3277 return (SET_ERROR(ENOTSUP));
3280 * Set the pending device list so we correctly handle device in-use
3283 sav->sav_pending = dev;
3284 sav->sav_npending = ndev;
3286 for (i = 0; i < ndev; i++) {
3287 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
3291 if (!vd->vdev_ops->vdev_op_leaf) {
3293 error = SET_ERROR(EINVAL);
3298 * The L2ARC currently only supports disk devices in
3299 * kernel context. For user-level testing, we allow it.
3302 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
3303 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
3304 error = SET_ERROR(ENOTBLK);
3311 if ((error = vdev_open(vd)) == 0 &&
3312 (error = vdev_label_init(vd, crtxg, label)) == 0) {
3313 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
3314 vd->vdev_guid) == 0);
3320 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
3327 sav->sav_pending = NULL;
3328 sav->sav_npending = 0;
3333 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
3337 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3339 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3340 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
3341 VDEV_LABEL_SPARE)) != 0) {
3345 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3346 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
3347 VDEV_LABEL_L2CACHE));
3351 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
3356 if (sav->sav_config != NULL) {
3362 * Generate new dev list by concatentating with the
3365 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
3366 &olddevs, &oldndevs) == 0);
3368 newdevs = kmem_alloc(sizeof (void *) *
3369 (ndevs + oldndevs), KM_SLEEP);
3370 for (i = 0; i < oldndevs; i++)
3371 VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
3373 for (i = 0; i < ndevs; i++)
3374 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
3377 VERIFY(nvlist_remove(sav->sav_config, config,
3378 DATA_TYPE_NVLIST_ARRAY) == 0);
3380 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3381 config, newdevs, ndevs + oldndevs) == 0);
3382 for (i = 0; i < oldndevs + ndevs; i++)
3383 nvlist_free(newdevs[i]);
3384 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
3387 * Generate a new dev list.
3389 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
3391 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
3397 * Stop and drop level 2 ARC devices
3400 spa_l2cache_drop(spa_t *spa)
3404 spa_aux_vdev_t *sav = &spa->spa_l2cache;
3406 for (i = 0; i < sav->sav_count; i++) {
3409 vd = sav->sav_vdevs[i];
3412 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
3413 pool != 0ULL && l2arc_vdev_present(vd))
3414 l2arc_remove_vdev(vd);
3422 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
3426 char *altroot = NULL;
3431 uint64_t txg = TXG_INITIAL;
3432 nvlist_t **spares, **l2cache;
3433 uint_t nspares, nl2cache;
3434 uint64_t version, obj;
3435 boolean_t has_features;
3438 * If this pool already exists, return failure.
3440 mutex_enter(&spa_namespace_lock);
3441 if (spa_lookup(pool) != NULL) {
3442 mutex_exit(&spa_namespace_lock);
3443 return (SET_ERROR(EEXIST));
3447 * Allocate a new spa_t structure.
3449 (void) nvlist_lookup_string(props,
3450 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3451 spa = spa_add(pool, NULL, altroot);
3452 spa_activate(spa, spa_mode_global);
3454 if (props && (error = spa_prop_validate(spa, props))) {
3455 spa_deactivate(spa);
3457 mutex_exit(&spa_namespace_lock);
3461 has_features = B_FALSE;
3462 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
3463 elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
3464 if (zpool_prop_feature(nvpair_name(elem)))
3465 has_features = B_TRUE;
3468 if (has_features || nvlist_lookup_uint64(props,
3469 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
3470 version = SPA_VERSION;
3472 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
3474 spa->spa_first_txg = txg;
3475 spa->spa_uberblock.ub_txg = txg - 1;
3476 spa->spa_uberblock.ub_version = version;
3477 spa->spa_ubsync = spa->spa_uberblock;
3480 * Create "The Godfather" zio to hold all async IOs
3482 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
3483 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
3486 * Create the root vdev.
3488 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3490 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3492 ASSERT(error != 0 || rvd != NULL);
3493 ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3495 if (error == 0 && !zfs_allocatable_devs(nvroot))
3496 error = SET_ERROR(EINVAL);
3499 (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3500 (error = spa_validate_aux(spa, nvroot, txg,
3501 VDEV_ALLOC_ADD)) == 0) {
3502 for (int c = 0; c < rvd->vdev_children; c++) {
3503 vdev_metaslab_set_size(rvd->vdev_child[c]);
3504 vdev_expand(rvd->vdev_child[c], txg);
3508 spa_config_exit(spa, SCL_ALL, FTAG);
3512 spa_deactivate(spa);
3514 mutex_exit(&spa_namespace_lock);
3519 * Get the list of spares, if specified.
3521 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3522 &spares, &nspares) == 0) {
3523 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3525 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3526 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3527 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3528 spa_load_spares(spa);
3529 spa_config_exit(spa, SCL_ALL, FTAG);
3530 spa->spa_spares.sav_sync = B_TRUE;
3534 * Get the list of level 2 cache devices, if specified.
3536 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3537 &l2cache, &nl2cache) == 0) {
3538 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3539 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3540 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3541 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3542 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3543 spa_load_l2cache(spa);
3544 spa_config_exit(spa, SCL_ALL, FTAG);
3545 spa->spa_l2cache.sav_sync = B_TRUE;
3548 spa->spa_is_initializing = B_TRUE;
3549 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3550 spa->spa_meta_objset = dp->dp_meta_objset;
3551 spa->spa_is_initializing = B_FALSE;
3554 * Create DDTs (dedup tables).
3558 spa_update_dspace(spa);
3560 tx = dmu_tx_create_assigned(dp, txg);
3563 * Create the pool config object.
3565 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3566 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3567 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3569 if (zap_add(spa->spa_meta_objset,
3570 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3571 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3572 cmn_err(CE_PANIC, "failed to add pool config");
3575 if (spa_version(spa) >= SPA_VERSION_FEATURES)
3576 spa_feature_create_zap_objects(spa, tx);
3578 if (zap_add(spa->spa_meta_objset,
3579 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3580 sizeof (uint64_t), 1, &version, tx) != 0) {
3581 cmn_err(CE_PANIC, "failed to add pool version");
3584 /* Newly created pools with the right version are always deflated. */
3585 if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3586 spa->spa_deflate = TRUE;
3587 if (zap_add(spa->spa_meta_objset,
3588 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3589 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3590 cmn_err(CE_PANIC, "failed to add deflate");
3595 * Create the deferred-free bpobj. Turn off compression
3596 * because sync-to-convergence takes longer if the blocksize
3599 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3600 dmu_object_set_compress(spa->spa_meta_objset, obj,
3601 ZIO_COMPRESS_OFF, tx);
3602 if (zap_add(spa->spa_meta_objset,
3603 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3604 sizeof (uint64_t), 1, &obj, tx) != 0) {
3605 cmn_err(CE_PANIC, "failed to add bpobj");
3607 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3608 spa->spa_meta_objset, obj));
3611 * Create the pool's history object.
3613 if (version >= SPA_VERSION_ZPOOL_HISTORY)
3614 spa_history_create_obj(spa, tx);
3617 * Set pool properties.
3619 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3620 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3621 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3622 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3624 if (props != NULL) {
3625 spa_configfile_set(spa, props, B_FALSE);
3626 spa_sync_props(props, tx);
3631 spa->spa_sync_on = B_TRUE;
3632 txg_sync_start(spa->spa_dsl_pool);
3635 * We explicitly wait for the first transaction to complete so that our
3636 * bean counters are appropriately updated.
3638 txg_wait_synced(spa->spa_dsl_pool, txg);
3640 spa_config_sync(spa, B_FALSE, B_TRUE);
3642 spa_history_log_version(spa, "create");
3644 spa->spa_minref = refcount_count(&spa->spa_refcount);
3646 mutex_exit(&spa_namespace_lock);
3654 * Get the root pool information from the root disk, then import the root pool
3655 * during the system boot up time.
3657 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3660 spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3663 nvlist_t *nvtop, *nvroot;
3666 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3670 * Add this top-level vdev to the child array.
3672 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3674 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3676 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3679 * Put this pool's top-level vdevs into a root vdev.
3681 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3682 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3683 VDEV_TYPE_ROOT) == 0);
3684 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3685 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3686 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3690 * Replace the existing vdev_tree with the new root vdev in
3691 * this pool's configuration (remove the old, add the new).
3693 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3694 nvlist_free(nvroot);
3699 * Walk the vdev tree and see if we can find a device with "better"
3700 * configuration. A configuration is "better" if the label on that
3701 * device has a more recent txg.
3704 spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3706 for (int c = 0; c < vd->vdev_children; c++)
3707 spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3709 if (vd->vdev_ops->vdev_op_leaf) {
3713 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3717 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3721 * Do we have a better boot device?
3723 if (label_txg > *txg) {
3732 * Import a root pool.
3734 * For x86. devpath_list will consist of devid and/or physpath name of
3735 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3736 * The GRUB "findroot" command will return the vdev we should boot.
3738 * For Sparc, devpath_list consists the physpath name of the booting device
3739 * no matter the rootpool is a single device pool or a mirrored pool.
3741 * "/pci@1f,0/ide@d/disk@0,0:a"
3744 spa_import_rootpool(char *devpath, char *devid)
3747 vdev_t *rvd, *bvd, *avd = NULL;
3748 nvlist_t *config, *nvtop;
3754 * Read the label from the boot device and generate a configuration.
3756 config = spa_generate_rootconf(devpath, devid, &guid);
3757 #if defined(_OBP) && defined(_KERNEL)
3758 if (config == NULL) {
3759 if (strstr(devpath, "/iscsi/ssd") != NULL) {
3761 get_iscsi_bootpath_phy(devpath);
3762 config = spa_generate_rootconf(devpath, devid, &guid);
3766 if (config == NULL) {
3767 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
3769 return (SET_ERROR(EIO));
3772 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3774 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3776 mutex_enter(&spa_namespace_lock);
3777 if ((spa = spa_lookup(pname)) != NULL) {
3779 * Remove the existing root pool from the namespace so that we
3780 * can replace it with the correct config we just read in.
3785 spa = spa_add(pname, config, NULL);
3786 spa->spa_is_root = B_TRUE;
3787 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3790 * Build up a vdev tree based on the boot device's label config.
3792 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3794 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3795 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3796 VDEV_ALLOC_ROOTPOOL);
3797 spa_config_exit(spa, SCL_ALL, FTAG);
3799 mutex_exit(&spa_namespace_lock);
3800 nvlist_free(config);
3801 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3807 * Get the boot vdev.
3809 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3810 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3811 (u_longlong_t)guid);
3812 error = SET_ERROR(ENOENT);
3817 * Determine if there is a better boot device.
3820 spa_alt_rootvdev(rvd, &avd, &txg);
3822 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3823 "try booting from '%s'", avd->vdev_path);
3824 error = SET_ERROR(EINVAL);
3829 * If the boot device is part of a spare vdev then ensure that
3830 * we're booting off the active spare.
3832 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3833 !bvd->vdev_isspare) {
3834 cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3835 "try booting from '%s'",
3837 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3838 error = SET_ERROR(EINVAL);
3844 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3846 spa_config_exit(spa, SCL_ALL, FTAG);
3847 mutex_exit(&spa_namespace_lock);
3849 nvlist_free(config);
3855 extern int vdev_geom_read_pool_label(const char *name, nvlist_t ***configs,
3859 spa_generate_rootconf(const char *name)
3861 nvlist_t **configs, **tops;
3863 nvlist_t *best_cfg, *nvtop, *nvroot;
3872 if (vdev_geom_read_pool_label(name, &configs, &count) != 0)
3875 ASSERT3U(count, !=, 0);
3877 for (i = 0; i < count; i++) {
3880 VERIFY(nvlist_lookup_uint64(configs[i], ZPOOL_CONFIG_POOL_TXG,
3882 if (txg > best_txg) {
3884 best_cfg = configs[i];
3889 * Multi-vdev root pool configuration discovery is not supported yet.
3892 nvlist_lookup_uint64(best_cfg, ZPOOL_CONFIG_VDEV_CHILDREN, &nchildren);
3894 nvlist_lookup_uint64_array(best_cfg, ZPOOL_CONFIG_HOLE_ARRAY,
3897 tops = kmem_zalloc(nchildren * sizeof(void *), KM_SLEEP);
3898 for (i = 0; i < nchildren; i++) {
3901 if (configs[i] == NULL)
3903 VERIFY(nvlist_lookup_nvlist(configs[i], ZPOOL_CONFIG_VDEV_TREE,
3905 nvlist_dup(nvtop, &tops[i], KM_SLEEP);
3907 for (i = 0; holes != NULL && i < nholes; i++) {
3910 if (tops[holes[i]] != NULL)
3912 nvlist_alloc(&tops[holes[i]], NV_UNIQUE_NAME, KM_SLEEP);
3913 VERIFY(nvlist_add_string(tops[holes[i]], ZPOOL_CONFIG_TYPE,
3914 VDEV_TYPE_HOLE) == 0);
3915 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_ID,
3917 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_GUID,
3920 for (i = 0; i < nchildren; i++) {
3921 if (tops[i] != NULL)
3923 nvlist_alloc(&tops[i], NV_UNIQUE_NAME, KM_SLEEP);
3924 VERIFY(nvlist_add_string(tops[i], ZPOOL_CONFIG_TYPE,
3925 VDEV_TYPE_MISSING) == 0);
3926 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_ID,
3928 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_GUID,
3933 * Create pool config based on the best vdev config.
3935 nvlist_dup(best_cfg, &config, KM_SLEEP);
3938 * Put this pool's top-level vdevs into a root vdev.
3940 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3942 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3943 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3944 VDEV_TYPE_ROOT) == 0);
3945 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3946 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3947 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3948 tops, nchildren) == 0);
3951 * Replace the existing vdev_tree with the new root vdev in
3952 * this pool's configuration (remove the old, add the new).
3954 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3957 * Drop vdev config elements that should not be present at pool level.
3959 nvlist_remove(config, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64);
3960 nvlist_remove(config, ZPOOL_CONFIG_TOP_GUID, DATA_TYPE_UINT64);
3962 for (i = 0; i < count; i++)
3963 nvlist_free(configs[i]);
3964 kmem_free(configs, count * sizeof(void *));
3965 for (i = 0; i < nchildren; i++)
3966 nvlist_free(tops[i]);
3967 kmem_free(tops, nchildren * sizeof(void *));
3968 nvlist_free(nvroot);
3973 spa_import_rootpool(const char *name)
3976 vdev_t *rvd, *bvd, *avd = NULL;
3977 nvlist_t *config, *nvtop;
3983 * Read the label from the boot device and generate a configuration.
3985 config = spa_generate_rootconf(name);
3987 mutex_enter(&spa_namespace_lock);
3988 if (config != NULL) {
3989 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3990 &pname) == 0 && strcmp(name, pname) == 0);
3991 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg)
3994 if ((spa = spa_lookup(pname)) != NULL) {
3996 * Remove the existing root pool from the namespace so
3997 * that we can replace it with the correct config
4002 spa = spa_add(pname, config, NULL);
4005 * Set spa_ubsync.ub_version as it can be used in vdev_alloc()
4006 * via spa_version().
4008 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
4009 &spa->spa_ubsync.ub_version) != 0)
4010 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
4011 } else if ((spa = spa_lookup(name)) == NULL) {
4012 cmn_err(CE_NOTE, "Cannot find the pool label for '%s'",
4016 VERIFY(nvlist_dup(spa->spa_config, &config, KM_SLEEP) == 0);
4018 spa->spa_is_root = B_TRUE;
4019 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
4022 * Build up a vdev tree based on the boot device's label config.
4024 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4026 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4027 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
4028 VDEV_ALLOC_ROOTPOOL);
4029 spa_config_exit(spa, SCL_ALL, FTAG);
4031 mutex_exit(&spa_namespace_lock);
4032 nvlist_free(config);
4033 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
4038 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4040 spa_config_exit(spa, SCL_ALL, FTAG);
4041 mutex_exit(&spa_namespace_lock);
4043 nvlist_free(config);
4051 * Import a non-root pool into the system.
4054 spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
4057 char *altroot = NULL;
4058 spa_load_state_t state = SPA_LOAD_IMPORT;
4059 zpool_rewind_policy_t policy;
4060 uint64_t mode = spa_mode_global;
4061 uint64_t readonly = B_FALSE;
4064 nvlist_t **spares, **l2cache;
4065 uint_t nspares, nl2cache;
4068 * If a pool with this name exists, return failure.
4070 mutex_enter(&spa_namespace_lock);
4071 if (spa_lookup(pool) != NULL) {
4072 mutex_exit(&spa_namespace_lock);
4073 return (SET_ERROR(EEXIST));
4077 * Create and initialize the spa structure.
4079 (void) nvlist_lookup_string(props,
4080 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4081 (void) nvlist_lookup_uint64(props,
4082 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
4085 spa = spa_add(pool, config, altroot);
4086 spa->spa_import_flags = flags;
4089 * Verbatim import - Take a pool and insert it into the namespace
4090 * as if it had been loaded at boot.
4092 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
4094 spa_configfile_set(spa, props, B_FALSE);
4096 spa_config_sync(spa, B_FALSE, B_TRUE);
4098 mutex_exit(&spa_namespace_lock);
4099 spa_history_log_version(spa, "import");
4104 spa_activate(spa, mode);
4107 * Don't start async tasks until we know everything is healthy.
4109 spa_async_suspend(spa);
4111 zpool_get_rewind_policy(config, &policy);
4112 if (policy.zrp_request & ZPOOL_DO_REWIND)
4113 state = SPA_LOAD_RECOVER;
4116 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
4117 * because the user-supplied config is actually the one to trust when
4120 if (state != SPA_LOAD_RECOVER)
4121 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
4123 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
4124 policy.zrp_request);
4127 * Propagate anything learned while loading the pool and pass it
4128 * back to caller (i.e. rewind info, missing devices, etc).
4130 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4131 spa->spa_load_info) == 0);
4133 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4135 * Toss any existing sparelist, as it doesn't have any validity
4136 * anymore, and conflicts with spa_has_spare().
4138 if (spa->spa_spares.sav_config) {
4139 nvlist_free(spa->spa_spares.sav_config);
4140 spa->spa_spares.sav_config = NULL;
4141 spa_load_spares(spa);
4143 if (spa->spa_l2cache.sav_config) {
4144 nvlist_free(spa->spa_l2cache.sav_config);
4145 spa->spa_l2cache.sav_config = NULL;
4146 spa_load_l2cache(spa);
4149 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4152 error = spa_validate_aux(spa, nvroot, -1ULL,
4155 error = spa_validate_aux(spa, nvroot, -1ULL,
4156 VDEV_ALLOC_L2CACHE);
4157 spa_config_exit(spa, SCL_ALL, FTAG);
4160 spa_configfile_set(spa, props, B_FALSE);
4162 if (error != 0 || (props && spa_writeable(spa) &&
4163 (error = spa_prop_set(spa, props)))) {
4165 spa_deactivate(spa);
4167 mutex_exit(&spa_namespace_lock);
4171 spa_async_resume(spa);
4174 * Override any spares and level 2 cache devices as specified by
4175 * the user, as these may have correct device names/devids, etc.
4177 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
4178 &spares, &nspares) == 0) {
4179 if (spa->spa_spares.sav_config)
4180 VERIFY(nvlist_remove(spa->spa_spares.sav_config,
4181 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
4183 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
4184 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4185 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
4186 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
4187 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4188 spa_load_spares(spa);
4189 spa_config_exit(spa, SCL_ALL, FTAG);
4190 spa->spa_spares.sav_sync = B_TRUE;
4192 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
4193 &l2cache, &nl2cache) == 0) {
4194 if (spa->spa_l2cache.sav_config)
4195 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
4196 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
4198 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
4199 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4200 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
4201 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
4202 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4203 spa_load_l2cache(spa);
4204 spa_config_exit(spa, SCL_ALL, FTAG);
4205 spa->spa_l2cache.sav_sync = B_TRUE;
4209 * Check for any removed devices.
4211 if (spa->spa_autoreplace) {
4212 spa_aux_check_removed(&spa->spa_spares);
4213 spa_aux_check_removed(&spa->spa_l2cache);
4216 if (spa_writeable(spa)) {
4218 * Update the config cache to include the newly-imported pool.
4220 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4224 * It's possible that the pool was expanded while it was exported.
4225 * We kick off an async task to handle this for us.
4227 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
4229 mutex_exit(&spa_namespace_lock);
4230 spa_history_log_version(spa, "import");
4234 zvol_create_minors(pool);
4241 spa_tryimport(nvlist_t *tryconfig)
4243 nvlist_t *config = NULL;
4249 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
4252 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
4256 * Create and initialize the spa structure.
4258 mutex_enter(&spa_namespace_lock);
4259 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
4260 spa_activate(spa, FREAD);
4263 * Pass off the heavy lifting to spa_load().
4264 * Pass TRUE for mosconfig because the user-supplied config
4265 * is actually the one to trust when doing an import.
4267 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
4270 * If 'tryconfig' was at least parsable, return the current config.
4272 if (spa->spa_root_vdev != NULL) {
4273 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4274 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
4276 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4278 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
4279 spa->spa_uberblock.ub_timestamp) == 0);
4280 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4281 spa->spa_load_info) == 0);
4284 * If the bootfs property exists on this pool then we
4285 * copy it out so that external consumers can tell which
4286 * pools are bootable.
4288 if ((!error || error == EEXIST) && spa->spa_bootfs) {
4289 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4292 * We have to play games with the name since the
4293 * pool was opened as TRYIMPORT_NAME.
4295 if (dsl_dsobj_to_dsname(spa_name(spa),
4296 spa->spa_bootfs, tmpname) == 0) {
4298 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4300 cp = strchr(tmpname, '/');
4302 (void) strlcpy(dsname, tmpname,
4305 (void) snprintf(dsname, MAXPATHLEN,
4306 "%s/%s", poolname, ++cp);
4308 VERIFY(nvlist_add_string(config,
4309 ZPOOL_CONFIG_BOOTFS, dsname) == 0);
4310 kmem_free(dsname, MAXPATHLEN);
4312 kmem_free(tmpname, MAXPATHLEN);
4316 * Add the list of hot spares and level 2 cache devices.
4318 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4319 spa_add_spares(spa, config);
4320 spa_add_l2cache(spa, config);
4321 spa_config_exit(spa, SCL_CONFIG, FTAG);
4325 spa_deactivate(spa);
4327 mutex_exit(&spa_namespace_lock);
4333 * Pool export/destroy
4335 * The act of destroying or exporting a pool is very simple. We make sure there
4336 * is no more pending I/O and any references to the pool are gone. Then, we
4337 * update the pool state and sync all the labels to disk, removing the
4338 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4339 * we don't sync the labels or remove the configuration cache.
4342 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
4343 boolean_t force, boolean_t hardforce)
4350 if (!(spa_mode_global & FWRITE))
4351 return (SET_ERROR(EROFS));
4353 mutex_enter(&spa_namespace_lock);
4354 if ((spa = spa_lookup(pool)) == NULL) {
4355 mutex_exit(&spa_namespace_lock);
4356 return (SET_ERROR(ENOENT));
4360 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4361 * reacquire the namespace lock, and see if we can export.
4363 spa_open_ref(spa, FTAG);
4364 mutex_exit(&spa_namespace_lock);
4365 spa_async_suspend(spa);
4366 mutex_enter(&spa_namespace_lock);
4367 spa_close(spa, FTAG);
4370 * The pool will be in core if it's openable,
4371 * in which case we can modify its state.
4373 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
4375 * Objsets may be open only because they're dirty, so we
4376 * have to force it to sync before checking spa_refcnt.
4378 txg_wait_synced(spa->spa_dsl_pool, 0);
4381 * A pool cannot be exported or destroyed if there are active
4382 * references. If we are resetting a pool, allow references by
4383 * fault injection handlers.
4385 if (!spa_refcount_zero(spa) ||
4386 (spa->spa_inject_ref != 0 &&
4387 new_state != POOL_STATE_UNINITIALIZED)) {
4388 spa_async_resume(spa);
4389 mutex_exit(&spa_namespace_lock);
4390 return (SET_ERROR(EBUSY));
4394 * A pool cannot be exported if it has an active shared spare.
4395 * This is to prevent other pools stealing the active spare
4396 * from an exported pool. At user's own will, such pool can
4397 * be forcedly exported.
4399 if (!force && new_state == POOL_STATE_EXPORTED &&
4400 spa_has_active_shared_spare(spa)) {
4401 spa_async_resume(spa);
4402 mutex_exit(&spa_namespace_lock);
4403 return (SET_ERROR(EXDEV));
4407 * We want this to be reflected on every label,
4408 * so mark them all dirty. spa_unload() will do the
4409 * final sync that pushes these changes out.
4411 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
4412 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4413 spa->spa_state = new_state;
4414 spa->spa_final_txg = spa_last_synced_txg(spa) +
4416 vdev_config_dirty(spa->spa_root_vdev);
4417 spa_config_exit(spa, SCL_ALL, FTAG);
4421 spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
4423 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4425 spa_deactivate(spa);
4428 if (oldconfig && spa->spa_config)
4429 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
4431 if (new_state != POOL_STATE_UNINITIALIZED) {
4433 spa_config_sync(spa, B_TRUE, B_TRUE);
4436 mutex_exit(&spa_namespace_lock);
4442 * Destroy a storage pool.
4445 spa_destroy(char *pool)
4447 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
4452 * Export a storage pool.
4455 spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
4456 boolean_t hardforce)
4458 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
4463 * Similar to spa_export(), this unloads the spa_t without actually removing it
4464 * from the namespace in any way.
4467 spa_reset(char *pool)
4469 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
4474 * ==========================================================================
4475 * Device manipulation
4476 * ==========================================================================
4480 * Add a device to a storage pool.
4483 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
4487 vdev_t *rvd = spa->spa_root_vdev;
4489 nvlist_t **spares, **l2cache;
4490 uint_t nspares, nl2cache;
4492 ASSERT(spa_writeable(spa));
4494 txg = spa_vdev_enter(spa);
4496 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
4497 VDEV_ALLOC_ADD)) != 0)
4498 return (spa_vdev_exit(spa, NULL, txg, error));
4500 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */
4502 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
4506 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
4510 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
4511 return (spa_vdev_exit(spa, vd, txg, EINVAL));
4513 if (vd->vdev_children != 0 &&
4514 (error = vdev_create(vd, txg, B_FALSE)) != 0)
4515 return (spa_vdev_exit(spa, vd, txg, error));
4518 * We must validate the spares and l2cache devices after checking the
4519 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
4521 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
4522 return (spa_vdev_exit(spa, vd, txg, error));
4525 * Transfer each new top-level vdev from vd to rvd.
4527 for (int c = 0; c < vd->vdev_children; c++) {
4530 * Set the vdev id to the first hole, if one exists.
4532 for (id = 0; id < rvd->vdev_children; id++) {
4533 if (rvd->vdev_child[id]->vdev_ishole) {
4534 vdev_free(rvd->vdev_child[id]);
4538 tvd = vd->vdev_child[c];
4539 vdev_remove_child(vd, tvd);
4541 vdev_add_child(rvd, tvd);
4542 vdev_config_dirty(tvd);
4546 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
4547 ZPOOL_CONFIG_SPARES);
4548 spa_load_spares(spa);
4549 spa->spa_spares.sav_sync = B_TRUE;
4552 if (nl2cache != 0) {
4553 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
4554 ZPOOL_CONFIG_L2CACHE);
4555 spa_load_l2cache(spa);
4556 spa->spa_l2cache.sav_sync = B_TRUE;
4560 * We have to be careful when adding new vdevs to an existing pool.
4561 * If other threads start allocating from these vdevs before we
4562 * sync the config cache, and we lose power, then upon reboot we may
4563 * fail to open the pool because there are DVAs that the config cache
4564 * can't translate. Therefore, we first add the vdevs without
4565 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4566 * and then let spa_config_update() initialize the new metaslabs.
4568 * spa_load() checks for added-but-not-initialized vdevs, so that
4569 * if we lose power at any point in this sequence, the remaining
4570 * steps will be completed the next time we load the pool.
4572 (void) spa_vdev_exit(spa, vd, txg, 0);
4574 mutex_enter(&spa_namespace_lock);
4575 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4576 mutex_exit(&spa_namespace_lock);
4582 * Attach a device to a mirror. The arguments are the path to any device
4583 * in the mirror, and the nvroot for the new device. If the path specifies
4584 * a device that is not mirrored, we automatically insert the mirror vdev.
4586 * If 'replacing' is specified, the new device is intended to replace the
4587 * existing device; in this case the two devices are made into their own
4588 * mirror using the 'replacing' vdev, which is functionally identical to
4589 * the mirror vdev (it actually reuses all the same ops) but has a few
4590 * extra rules: you can't attach to it after it's been created, and upon
4591 * completion of resilvering, the first disk (the one being replaced)
4592 * is automatically detached.
4595 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
4597 uint64_t txg, dtl_max_txg;
4598 vdev_t *rvd = spa->spa_root_vdev;
4599 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
4601 char *oldvdpath, *newvdpath;
4605 ASSERT(spa_writeable(spa));
4607 txg = spa_vdev_enter(spa);
4609 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
4612 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4614 if (!oldvd->vdev_ops->vdev_op_leaf)
4615 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4617 pvd = oldvd->vdev_parent;
4619 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
4620 VDEV_ALLOC_ATTACH)) != 0)
4621 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4623 if (newrootvd->vdev_children != 1)
4624 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4626 newvd = newrootvd->vdev_child[0];
4628 if (!newvd->vdev_ops->vdev_op_leaf)
4629 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4631 if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
4632 return (spa_vdev_exit(spa, newrootvd, txg, error));
4635 * Spares can't replace logs
4637 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
4638 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4642 * For attach, the only allowable parent is a mirror or the root
4645 if (pvd->vdev_ops != &vdev_mirror_ops &&
4646 pvd->vdev_ops != &vdev_root_ops)
4647 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4649 pvops = &vdev_mirror_ops;
4652 * Active hot spares can only be replaced by inactive hot
4655 if (pvd->vdev_ops == &vdev_spare_ops &&
4656 oldvd->vdev_isspare &&
4657 !spa_has_spare(spa, newvd->vdev_guid))
4658 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4661 * If the source is a hot spare, and the parent isn't already a
4662 * spare, then we want to create a new hot spare. Otherwise, we
4663 * want to create a replacing vdev. The user is not allowed to
4664 * attach to a spared vdev child unless the 'isspare' state is
4665 * the same (spare replaces spare, non-spare replaces
4668 if (pvd->vdev_ops == &vdev_replacing_ops &&
4669 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
4670 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4671 } else if (pvd->vdev_ops == &vdev_spare_ops &&
4672 newvd->vdev_isspare != oldvd->vdev_isspare) {
4673 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4676 if (newvd->vdev_isspare)
4677 pvops = &vdev_spare_ops;
4679 pvops = &vdev_replacing_ops;
4683 * Make sure the new device is big enough.
4685 if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
4686 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
4689 * The new device cannot have a higher alignment requirement
4690 * than the top-level vdev.
4692 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4693 return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4696 * If this is an in-place replacement, update oldvd's path and devid
4697 * to make it distinguishable from newvd, and unopenable from now on.
4699 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4700 spa_strfree(oldvd->vdev_path);
4701 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4703 (void) sprintf(oldvd->vdev_path, "%s/%s",
4704 newvd->vdev_path, "old");
4705 if (oldvd->vdev_devid != NULL) {
4706 spa_strfree(oldvd->vdev_devid);
4707 oldvd->vdev_devid = NULL;
4711 /* mark the device being resilvered */
4712 newvd->vdev_resilvering = B_TRUE;
4715 * If the parent is not a mirror, or if we're replacing, insert the new
4716 * mirror/replacing/spare vdev above oldvd.
4718 if (pvd->vdev_ops != pvops)
4719 pvd = vdev_add_parent(oldvd, pvops);
4721 ASSERT(pvd->vdev_top->vdev_parent == rvd);
4722 ASSERT(pvd->vdev_ops == pvops);
4723 ASSERT(oldvd->vdev_parent == pvd);
4726 * Extract the new device from its root and add it to pvd.
4728 vdev_remove_child(newrootvd, newvd);
4729 newvd->vdev_id = pvd->vdev_children;
4730 newvd->vdev_crtxg = oldvd->vdev_crtxg;
4731 vdev_add_child(pvd, newvd);
4733 tvd = newvd->vdev_top;
4734 ASSERT(pvd->vdev_top == tvd);
4735 ASSERT(tvd->vdev_parent == rvd);
4737 vdev_config_dirty(tvd);
4740 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4741 * for any dmu_sync-ed blocks. It will propagate upward when
4742 * spa_vdev_exit() calls vdev_dtl_reassess().
4744 dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4746 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4747 dtl_max_txg - TXG_INITIAL);
4749 if (newvd->vdev_isspare) {
4750 spa_spare_activate(newvd);
4751 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
4754 oldvdpath = spa_strdup(oldvd->vdev_path);
4755 newvdpath = spa_strdup(newvd->vdev_path);
4756 newvd_isspare = newvd->vdev_isspare;
4759 * Mark newvd's DTL dirty in this txg.
4761 vdev_dirty(tvd, VDD_DTL, newvd, txg);
4764 * Restart the resilver
4766 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4771 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4773 spa_history_log_internal(spa, "vdev attach", NULL,
4774 "%s vdev=%s %s vdev=%s",
4775 replacing && newvd_isspare ? "spare in" :
4776 replacing ? "replace" : "attach", newvdpath,
4777 replacing ? "for" : "to", oldvdpath);
4779 spa_strfree(oldvdpath);
4780 spa_strfree(newvdpath);
4782 if (spa->spa_bootfs)
4783 spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
4789 * Detach a device from a mirror or replacing vdev.
4791 * If 'replace_done' is specified, only detach if the parent
4792 * is a replacing vdev.
4795 spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4799 vdev_t *rvd = spa->spa_root_vdev;
4800 vdev_t *vd, *pvd, *cvd, *tvd;
4801 boolean_t unspare = B_FALSE;
4802 uint64_t unspare_guid = 0;
4805 ASSERT(spa_writeable(spa));
4807 txg = spa_vdev_enter(spa);
4809 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4812 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4814 if (!vd->vdev_ops->vdev_op_leaf)
4815 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4817 pvd = vd->vdev_parent;
4820 * If the parent/child relationship is not as expected, don't do it.
4821 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4822 * vdev that's replacing B with C. The user's intent in replacing
4823 * is to go from M(A,B) to M(A,C). If the user decides to cancel
4824 * the replace by detaching C, the expected behavior is to end up
4825 * M(A,B). But suppose that right after deciding to detach C,
4826 * the replacement of B completes. We would have M(A,C), and then
4827 * ask to detach C, which would leave us with just A -- not what
4828 * the user wanted. To prevent this, we make sure that the
4829 * parent/child relationship hasn't changed -- in this example,
4830 * that C's parent is still the replacing vdev R.
4832 if (pvd->vdev_guid != pguid && pguid != 0)
4833 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4836 * Only 'replacing' or 'spare' vdevs can be replaced.
4838 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4839 pvd->vdev_ops != &vdev_spare_ops)
4840 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4842 ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4843 spa_version(spa) >= SPA_VERSION_SPARES);
4846 * Only mirror, replacing, and spare vdevs support detach.
4848 if (pvd->vdev_ops != &vdev_replacing_ops &&
4849 pvd->vdev_ops != &vdev_mirror_ops &&
4850 pvd->vdev_ops != &vdev_spare_ops)
4851 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4854 * If this device has the only valid copy of some data,
4855 * we cannot safely detach it.
4857 if (vdev_dtl_required(vd))
4858 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4860 ASSERT(pvd->vdev_children >= 2);
4863 * If we are detaching the second disk from a replacing vdev, then
4864 * check to see if we changed the original vdev's path to have "/old"
4865 * at the end in spa_vdev_attach(). If so, undo that change now.
4867 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4868 vd->vdev_path != NULL) {
4869 size_t len = strlen(vd->vdev_path);
4871 for (int c = 0; c < pvd->vdev_children; c++) {
4872 cvd = pvd->vdev_child[c];
4874 if (cvd == vd || cvd->vdev_path == NULL)
4877 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4878 strcmp(cvd->vdev_path + len, "/old") == 0) {
4879 spa_strfree(cvd->vdev_path);
4880 cvd->vdev_path = spa_strdup(vd->vdev_path);
4887 * If we are detaching the original disk from a spare, then it implies
4888 * that the spare should become a real disk, and be removed from the
4889 * active spare list for the pool.
4891 if (pvd->vdev_ops == &vdev_spare_ops &&
4893 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4897 * Erase the disk labels so the disk can be used for other things.
4898 * This must be done after all other error cases are handled,
4899 * but before we disembowel vd (so we can still do I/O to it).
4900 * But if we can't do it, don't treat the error as fatal --
4901 * it may be that the unwritability of the disk is the reason
4902 * it's being detached!
4904 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4907 * Remove vd from its parent and compact the parent's children.
4909 vdev_remove_child(pvd, vd);
4910 vdev_compact_children(pvd);
4913 * Remember one of the remaining children so we can get tvd below.
4915 cvd = pvd->vdev_child[pvd->vdev_children - 1];
4918 * If we need to remove the remaining child from the list of hot spares,
4919 * do it now, marking the vdev as no longer a spare in the process.
4920 * We must do this before vdev_remove_parent(), because that can
4921 * change the GUID if it creates a new toplevel GUID. For a similar
4922 * reason, we must remove the spare now, in the same txg as the detach;
4923 * otherwise someone could attach a new sibling, change the GUID, and
4924 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4927 ASSERT(cvd->vdev_isspare);
4928 spa_spare_remove(cvd);
4929 unspare_guid = cvd->vdev_guid;
4930 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4931 cvd->vdev_unspare = B_TRUE;
4935 * If the parent mirror/replacing vdev only has one child,
4936 * the parent is no longer needed. Remove it from the tree.
4938 if (pvd->vdev_children == 1) {
4939 if (pvd->vdev_ops == &vdev_spare_ops)
4940 cvd->vdev_unspare = B_FALSE;
4941 vdev_remove_parent(cvd);
4946 * We don't set tvd until now because the parent we just removed
4947 * may have been the previous top-level vdev.
4949 tvd = cvd->vdev_top;
4950 ASSERT(tvd->vdev_parent == rvd);
4953 * Reevaluate the parent vdev state.
4955 vdev_propagate_state(cvd);
4958 * If the 'autoexpand' property is set on the pool then automatically
4959 * try to expand the size of the pool. For example if the device we
4960 * just detached was smaller than the others, it may be possible to
4961 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4962 * first so that we can obtain the updated sizes of the leaf vdevs.
4964 if (spa->spa_autoexpand) {
4966 vdev_expand(tvd, txg);
4969 vdev_config_dirty(tvd);
4972 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
4973 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4974 * But first make sure we're not on any *other* txg's DTL list, to
4975 * prevent vd from being accessed after it's freed.
4977 vdpath = spa_strdup(vd->vdev_path);
4978 for (int t = 0; t < TXG_SIZE; t++)
4979 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4980 vd->vdev_detached = B_TRUE;
4981 vdev_dirty(tvd, VDD_DTL, vd, txg);
4983 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
4985 /* hang on to the spa before we release the lock */
4986 spa_open_ref(spa, FTAG);
4988 error = spa_vdev_exit(spa, vd, txg, 0);
4990 spa_history_log_internal(spa, "detach", NULL,
4992 spa_strfree(vdpath);
4995 * If this was the removal of the original device in a hot spare vdev,
4996 * then we want to go through and remove the device from the hot spare
4997 * list of every other pool.
5000 spa_t *altspa = NULL;
5002 mutex_enter(&spa_namespace_lock);
5003 while ((altspa = spa_next(altspa)) != NULL) {
5004 if (altspa->spa_state != POOL_STATE_ACTIVE ||
5008 spa_open_ref(altspa, FTAG);
5009 mutex_exit(&spa_namespace_lock);
5010 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
5011 mutex_enter(&spa_namespace_lock);
5012 spa_close(altspa, FTAG);
5014 mutex_exit(&spa_namespace_lock);
5016 /* search the rest of the vdevs for spares to remove */
5017 spa_vdev_resilver_done(spa);
5020 /* all done with the spa; OK to release */
5021 mutex_enter(&spa_namespace_lock);
5022 spa_close(spa, FTAG);
5023 mutex_exit(&spa_namespace_lock);
5029 * Split a set of devices from their mirrors, and create a new pool from them.
5032 spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
5033 nvlist_t *props, boolean_t exp)
5036 uint64_t txg, *glist;
5038 uint_t c, children, lastlog;
5039 nvlist_t **child, *nvl, *tmp;
5041 char *altroot = NULL;
5042 vdev_t *rvd, **vml = NULL; /* vdev modify list */
5043 boolean_t activate_slog;
5045 ASSERT(spa_writeable(spa));
5047 txg = spa_vdev_enter(spa);
5049 /* clear the log and flush everything up to now */
5050 activate_slog = spa_passivate_log(spa);
5051 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5052 error = spa_offline_log(spa);
5053 txg = spa_vdev_config_enter(spa);
5056 spa_activate_log(spa);
5059 return (spa_vdev_exit(spa, NULL, txg, error));
5061 /* check new spa name before going any further */
5062 if (spa_lookup(newname) != NULL)
5063 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
5066 * scan through all the children to ensure they're all mirrors
5068 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
5069 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
5071 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5073 /* first, check to ensure we've got the right child count */
5074 rvd = spa->spa_root_vdev;
5076 for (c = 0; c < rvd->vdev_children; c++) {
5077 vdev_t *vd = rvd->vdev_child[c];
5079 /* don't count the holes & logs as children */
5080 if (vd->vdev_islog || vd->vdev_ishole) {
5088 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
5089 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5091 /* next, ensure no spare or cache devices are part of the split */
5092 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
5093 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
5094 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5096 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
5097 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
5099 /* then, loop over each vdev and validate it */
5100 for (c = 0; c < children; c++) {
5101 uint64_t is_hole = 0;
5103 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
5107 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
5108 spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
5111 error = SET_ERROR(EINVAL);
5116 /* which disk is going to be split? */
5117 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
5119 error = SET_ERROR(EINVAL);
5123 /* look it up in the spa */
5124 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
5125 if (vml[c] == NULL) {
5126 error = SET_ERROR(ENODEV);
5130 /* make sure there's nothing stopping the split */
5131 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
5132 vml[c]->vdev_islog ||
5133 vml[c]->vdev_ishole ||
5134 vml[c]->vdev_isspare ||
5135 vml[c]->vdev_isl2cache ||
5136 !vdev_writeable(vml[c]) ||
5137 vml[c]->vdev_children != 0 ||
5138 vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
5139 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
5140 error = SET_ERROR(EINVAL);
5144 if (vdev_dtl_required(vml[c])) {
5145 error = SET_ERROR(EBUSY);
5149 /* we need certain info from the top level */
5150 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
5151 vml[c]->vdev_top->vdev_ms_array) == 0);
5152 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
5153 vml[c]->vdev_top->vdev_ms_shift) == 0);
5154 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
5155 vml[c]->vdev_top->vdev_asize) == 0);
5156 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
5157 vml[c]->vdev_top->vdev_ashift) == 0);
5161 kmem_free(vml, children * sizeof (vdev_t *));
5162 kmem_free(glist, children * sizeof (uint64_t));
5163 return (spa_vdev_exit(spa, NULL, txg, error));
5166 /* stop writers from using the disks */
5167 for (c = 0; c < children; c++) {
5169 vml[c]->vdev_offline = B_TRUE;
5171 vdev_reopen(spa->spa_root_vdev);
5174 * Temporarily record the splitting vdevs in the spa config. This
5175 * will disappear once the config is regenerated.
5177 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5178 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
5179 glist, children) == 0);
5180 kmem_free(glist, children * sizeof (uint64_t));
5182 mutex_enter(&spa->spa_props_lock);
5183 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
5185 mutex_exit(&spa->spa_props_lock);
5186 spa->spa_config_splitting = nvl;
5187 vdev_config_dirty(spa->spa_root_vdev);
5189 /* configure and create the new pool */
5190 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
5191 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
5192 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
5193 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5194 spa_version(spa)) == 0);
5195 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
5196 spa->spa_config_txg) == 0);
5197 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
5198 spa_generate_guid(NULL)) == 0);
5199 (void) nvlist_lookup_string(props,
5200 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
5202 /* add the new pool to the namespace */
5203 newspa = spa_add(newname, config, altroot);
5204 newspa->spa_config_txg = spa->spa_config_txg;
5205 spa_set_log_state(newspa, SPA_LOG_CLEAR);
5207 /* release the spa config lock, retaining the namespace lock */
5208 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5210 if (zio_injection_enabled)
5211 zio_handle_panic_injection(spa, FTAG, 1);
5213 spa_activate(newspa, spa_mode_global);
5214 spa_async_suspend(newspa);
5217 /* mark that we are creating new spa by splitting */
5218 newspa->spa_splitting_newspa = B_TRUE;
5220 /* create the new pool from the disks of the original pool */
5221 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
5223 newspa->spa_splitting_newspa = B_FALSE;
5228 /* if that worked, generate a real config for the new pool */
5229 if (newspa->spa_root_vdev != NULL) {
5230 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
5231 NV_UNIQUE_NAME, KM_SLEEP) == 0);
5232 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
5233 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
5234 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
5239 if (props != NULL) {
5240 spa_configfile_set(newspa, props, B_FALSE);
5241 error = spa_prop_set(newspa, props);
5246 /* flush everything */
5247 txg = spa_vdev_config_enter(newspa);
5248 vdev_config_dirty(newspa->spa_root_vdev);
5249 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
5251 if (zio_injection_enabled)
5252 zio_handle_panic_injection(spa, FTAG, 2);
5254 spa_async_resume(newspa);
5256 /* finally, update the original pool's config */
5257 txg = spa_vdev_config_enter(spa);
5258 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
5259 error = dmu_tx_assign(tx, TXG_WAIT);
5262 for (c = 0; c < children; c++) {
5263 if (vml[c] != NULL) {
5266 spa_history_log_internal(spa, "detach", tx,
5267 "vdev=%s", vml[c]->vdev_path);
5271 vdev_config_dirty(spa->spa_root_vdev);
5272 spa->spa_config_splitting = NULL;
5276 (void) spa_vdev_exit(spa, NULL, txg, 0);
5278 if (zio_injection_enabled)
5279 zio_handle_panic_injection(spa, FTAG, 3);
5281 /* split is complete; log a history record */
5282 spa_history_log_internal(newspa, "split", NULL,
5283 "from pool %s", spa_name(spa));
5285 kmem_free(vml, children * sizeof (vdev_t *));
5287 /* if we're not going to mount the filesystems in userland, export */
5289 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
5296 spa_deactivate(newspa);
5299 txg = spa_vdev_config_enter(spa);
5301 /* re-online all offlined disks */
5302 for (c = 0; c < children; c++) {
5304 vml[c]->vdev_offline = B_FALSE;
5306 vdev_reopen(spa->spa_root_vdev);
5308 nvlist_free(spa->spa_config_splitting);
5309 spa->spa_config_splitting = NULL;
5310 (void) spa_vdev_exit(spa, NULL, txg, error);
5312 kmem_free(vml, children * sizeof (vdev_t *));
5317 spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
5319 for (int i = 0; i < count; i++) {
5322 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
5325 if (guid == target_guid)
5333 spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
5334 nvlist_t *dev_to_remove)
5336 nvlist_t **newdev = NULL;
5339 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
5341 for (int i = 0, j = 0; i < count; i++) {
5342 if (dev[i] == dev_to_remove)
5344 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
5347 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
5348 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
5350 for (int i = 0; i < count - 1; i++)
5351 nvlist_free(newdev[i]);
5354 kmem_free(newdev, (count - 1) * sizeof (void *));
5358 * Evacuate the device.
5361 spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
5366 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5367 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5368 ASSERT(vd == vd->vdev_top);
5371 * Evacuate the device. We don't hold the config lock as writer
5372 * since we need to do I/O but we do keep the
5373 * spa_namespace_lock held. Once this completes the device
5374 * should no longer have any blocks allocated on it.
5376 if (vd->vdev_islog) {
5377 if (vd->vdev_stat.vs_alloc != 0)
5378 error = spa_offline_log(spa);
5380 error = SET_ERROR(ENOTSUP);
5387 * The evacuation succeeded. Remove any remaining MOS metadata
5388 * associated with this vdev, and wait for these changes to sync.
5390 ASSERT0(vd->vdev_stat.vs_alloc);
5391 txg = spa_vdev_config_enter(spa);
5392 vd->vdev_removing = B_TRUE;
5393 vdev_dirty(vd, 0, NULL, txg);
5394 vdev_config_dirty(vd);
5395 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5401 * Complete the removal by cleaning up the namespace.
5404 spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
5406 vdev_t *rvd = spa->spa_root_vdev;
5407 uint64_t id = vd->vdev_id;
5408 boolean_t last_vdev = (id == (rvd->vdev_children - 1));
5410 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5411 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5412 ASSERT(vd == vd->vdev_top);
5415 * Only remove any devices which are empty.
5417 if (vd->vdev_stat.vs_alloc != 0)
5420 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5422 if (list_link_active(&vd->vdev_state_dirty_node))
5423 vdev_state_clean(vd);
5424 if (list_link_active(&vd->vdev_config_dirty_node))
5425 vdev_config_clean(vd);
5430 vdev_compact_children(rvd);
5432 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
5433 vdev_add_child(rvd, vd);
5435 vdev_config_dirty(rvd);
5438 * Reassess the health of our root vdev.
5444 * Remove a device from the pool -
5446 * Removing a device from the vdev namespace requires several steps
5447 * and can take a significant amount of time. As a result we use
5448 * the spa_vdev_config_[enter/exit] functions which allow us to
5449 * grab and release the spa_config_lock while still holding the namespace
5450 * lock. During each step the configuration is synced out.
5452 * Currently, this supports removing only hot spares, slogs, and level 2 ARC
5456 spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
5459 metaslab_group_t *mg;
5460 nvlist_t **spares, **l2cache, *nv;
5462 uint_t nspares, nl2cache;
5464 boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
5466 ASSERT(spa_writeable(spa));
5469 txg = spa_vdev_enter(spa);
5471 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5473 if (spa->spa_spares.sav_vdevs != NULL &&
5474 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5475 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
5476 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
5478 * Only remove the hot spare if it's not currently in use
5481 if (vd == NULL || unspare) {
5482 spa_vdev_remove_aux(spa->spa_spares.sav_config,
5483 ZPOOL_CONFIG_SPARES, spares, nspares, nv);
5484 spa_load_spares(spa);
5485 spa->spa_spares.sav_sync = B_TRUE;
5487 error = SET_ERROR(EBUSY);
5489 } else if (spa->spa_l2cache.sav_vdevs != NULL &&
5490 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5491 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
5492 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
5494 * Cache devices can always be removed.
5496 spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
5497 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
5498 spa_load_l2cache(spa);
5499 spa->spa_l2cache.sav_sync = B_TRUE;
5500 } else if (vd != NULL && vd->vdev_islog) {
5502 ASSERT(vd == vd->vdev_top);
5505 * XXX - Once we have bp-rewrite this should
5506 * become the common case.
5512 * Stop allocating from this vdev.
5514 metaslab_group_passivate(mg);
5517 * Wait for the youngest allocations and frees to sync,
5518 * and then wait for the deferral of those frees to finish.
5520 spa_vdev_config_exit(spa, NULL,
5521 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
5524 * Attempt to evacuate the vdev.
5526 error = spa_vdev_remove_evacuate(spa, vd);
5528 txg = spa_vdev_config_enter(spa);
5531 * If we couldn't evacuate the vdev, unwind.
5534 metaslab_group_activate(mg);
5535 return (spa_vdev_exit(spa, NULL, txg, error));
5539 * Clean up the vdev namespace.
5541 spa_vdev_remove_from_namespace(spa, vd);
5543 } else if (vd != NULL) {
5545 * Normal vdevs cannot be removed (yet).
5547 error = SET_ERROR(ENOTSUP);
5550 * There is no vdev of any kind with the specified guid.
5552 error = SET_ERROR(ENOENT);
5556 return (spa_vdev_exit(spa, NULL, txg, error));
5562 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5563 * currently spared, so we can detach it.
5566 spa_vdev_resilver_done_hunt(vdev_t *vd)
5568 vdev_t *newvd, *oldvd;
5570 for (int c = 0; c < vd->vdev_children; c++) {
5571 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
5576 if (vd->vdev_resilvering && vdev_dtl_empty(vd, DTL_MISSING) &&
5577 vdev_dtl_empty(vd, DTL_OUTAGE)) {
5578 ASSERT(vd->vdev_ops->vdev_op_leaf);
5579 vd->vdev_resilvering = B_FALSE;
5580 vdev_config_dirty(vd->vdev_top);
5584 * Check for a completed replacement. We always consider the first
5585 * vdev in the list to be the oldest vdev, and the last one to be
5586 * the newest (see spa_vdev_attach() for how that works). In
5587 * the case where the newest vdev is faulted, we will not automatically
5588 * remove it after a resilver completes. This is OK as it will require
5589 * user intervention to determine which disk the admin wishes to keep.
5591 if (vd->vdev_ops == &vdev_replacing_ops) {
5592 ASSERT(vd->vdev_children > 1);
5594 newvd = vd->vdev_child[vd->vdev_children - 1];
5595 oldvd = vd->vdev_child[0];
5597 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
5598 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5599 !vdev_dtl_required(oldvd))
5604 * Check for a completed resilver with the 'unspare' flag set.
5606 if (vd->vdev_ops == &vdev_spare_ops) {
5607 vdev_t *first = vd->vdev_child[0];
5608 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
5610 if (last->vdev_unspare) {
5613 } else if (first->vdev_unspare) {
5620 if (oldvd != NULL &&
5621 vdev_dtl_empty(newvd, DTL_MISSING) &&
5622 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5623 !vdev_dtl_required(oldvd))
5627 * If there are more than two spares attached to a disk,
5628 * and those spares are not required, then we want to
5629 * attempt to free them up now so that they can be used
5630 * by other pools. Once we're back down to a single
5631 * disk+spare, we stop removing them.
5633 if (vd->vdev_children > 2) {
5634 newvd = vd->vdev_child[1];
5636 if (newvd->vdev_isspare && last->vdev_isspare &&
5637 vdev_dtl_empty(last, DTL_MISSING) &&
5638 vdev_dtl_empty(last, DTL_OUTAGE) &&
5639 !vdev_dtl_required(newvd))
5648 spa_vdev_resilver_done(spa_t *spa)
5650 vdev_t *vd, *pvd, *ppvd;
5651 uint64_t guid, sguid, pguid, ppguid;
5653 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5655 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
5656 pvd = vd->vdev_parent;
5657 ppvd = pvd->vdev_parent;
5658 guid = vd->vdev_guid;
5659 pguid = pvd->vdev_guid;
5660 ppguid = ppvd->vdev_guid;
5663 * If we have just finished replacing a hot spared device, then
5664 * we need to detach the parent's first child (the original hot
5667 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
5668 ppvd->vdev_children == 2) {
5669 ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
5670 sguid = ppvd->vdev_child[1]->vdev_guid;
5672 spa_config_exit(spa, SCL_ALL, FTAG);
5673 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
5675 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
5677 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5680 spa_config_exit(spa, SCL_ALL, FTAG);
5684 * Update the stored path or FRU for this vdev.
5687 spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
5691 boolean_t sync = B_FALSE;
5693 ASSERT(spa_writeable(spa));
5695 spa_vdev_state_enter(spa, SCL_ALL);
5697 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
5698 return (spa_vdev_state_exit(spa, NULL, ENOENT));
5700 if (!vd->vdev_ops->vdev_op_leaf)
5701 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
5704 if (strcmp(value, vd->vdev_path) != 0) {
5705 spa_strfree(vd->vdev_path);
5706 vd->vdev_path = spa_strdup(value);
5710 if (vd->vdev_fru == NULL) {
5711 vd->vdev_fru = spa_strdup(value);
5713 } else if (strcmp(value, vd->vdev_fru) != 0) {
5714 spa_strfree(vd->vdev_fru);
5715 vd->vdev_fru = spa_strdup(value);
5720 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
5724 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
5726 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5730 spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5732 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5736 * ==========================================================================
5738 * ==========================================================================
5742 spa_scan_stop(spa_t *spa)
5744 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5745 if (dsl_scan_resilvering(spa->spa_dsl_pool))
5746 return (SET_ERROR(EBUSY));
5747 return (dsl_scan_cancel(spa->spa_dsl_pool));
5751 spa_scan(spa_t *spa, pool_scan_func_t func)
5753 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5755 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5756 return (SET_ERROR(ENOTSUP));
5759 * If a resilver was requested, but there is no DTL on a
5760 * writeable leaf device, we have nothing to do.
5762 if (func == POOL_SCAN_RESILVER &&
5763 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5764 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5768 return (dsl_scan(spa->spa_dsl_pool, func));
5772 * ==========================================================================
5773 * SPA async task processing
5774 * ==========================================================================
5778 spa_async_remove(spa_t *spa, vdev_t *vd)
5780 if (vd->vdev_remove_wanted) {
5781 vd->vdev_remove_wanted = B_FALSE;
5782 vd->vdev_delayed_close = B_FALSE;
5783 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5786 * We want to clear the stats, but we don't want to do a full
5787 * vdev_clear() as that will cause us to throw away
5788 * degraded/faulted state as well as attempt to reopen the
5789 * device, all of which is a waste.
5791 vd->vdev_stat.vs_read_errors = 0;
5792 vd->vdev_stat.vs_write_errors = 0;
5793 vd->vdev_stat.vs_checksum_errors = 0;
5795 vdev_state_dirty(vd->vdev_top);
5798 for (int c = 0; c < vd->vdev_children; c++)
5799 spa_async_remove(spa, vd->vdev_child[c]);
5803 spa_async_probe(spa_t *spa, vdev_t *vd)
5805 if (vd->vdev_probe_wanted) {
5806 vd->vdev_probe_wanted = B_FALSE;
5807 vdev_reopen(vd); /* vdev_open() does the actual probe */
5810 for (int c = 0; c < vd->vdev_children; c++)
5811 spa_async_probe(spa, vd->vdev_child[c]);
5815 spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5821 if (!spa->spa_autoexpand)
5824 for (int c = 0; c < vd->vdev_children; c++) {
5825 vdev_t *cvd = vd->vdev_child[c];
5826 spa_async_autoexpand(spa, cvd);
5829 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5832 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
5833 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
5835 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5836 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
5838 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
5839 ESC_ZFS_VDEV_AUTOEXPAND, attr, &eid, DDI_SLEEP);
5842 kmem_free(physpath, MAXPATHLEN);
5846 spa_async_thread(void *arg)
5851 ASSERT(spa->spa_sync_on);
5853 mutex_enter(&spa->spa_async_lock);
5854 tasks = spa->spa_async_tasks;
5855 spa->spa_async_tasks = 0;
5856 mutex_exit(&spa->spa_async_lock);
5859 * See if the config needs to be updated.
5861 if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5862 uint64_t old_space, new_space;
5864 mutex_enter(&spa_namespace_lock);
5865 old_space = metaslab_class_get_space(spa_normal_class(spa));
5866 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5867 new_space = metaslab_class_get_space(spa_normal_class(spa));
5868 mutex_exit(&spa_namespace_lock);
5871 * If the pool grew as a result of the config update,
5872 * then log an internal history event.
5874 if (new_space != old_space) {
5875 spa_history_log_internal(spa, "vdev online", NULL,
5876 "pool '%s' size: %llu(+%llu)",
5877 spa_name(spa), new_space, new_space - old_space);
5882 * See if any devices need to be marked REMOVED.
5884 if (tasks & SPA_ASYNC_REMOVE) {
5885 spa_vdev_state_enter(spa, SCL_NONE);
5886 spa_async_remove(spa, spa->spa_root_vdev);
5887 for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
5888 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5889 for (int i = 0; i < spa->spa_spares.sav_count; i++)
5890 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5891 (void) spa_vdev_state_exit(spa, NULL, 0);
5894 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5895 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5896 spa_async_autoexpand(spa, spa->spa_root_vdev);
5897 spa_config_exit(spa, SCL_CONFIG, FTAG);
5901 * See if any devices need to be probed.
5903 if (tasks & SPA_ASYNC_PROBE) {
5904 spa_vdev_state_enter(spa, SCL_NONE);
5905 spa_async_probe(spa, spa->spa_root_vdev);
5906 (void) spa_vdev_state_exit(spa, NULL, 0);
5910 * If any devices are done replacing, detach them.
5912 if (tasks & SPA_ASYNC_RESILVER_DONE)
5913 spa_vdev_resilver_done(spa);
5916 * Kick off a resilver.
5918 if (tasks & SPA_ASYNC_RESILVER)
5919 dsl_resilver_restart(spa->spa_dsl_pool, 0);
5922 * Let the world know that we're done.
5924 mutex_enter(&spa->spa_async_lock);
5925 spa->spa_async_thread = NULL;
5926 cv_broadcast(&spa->spa_async_cv);
5927 mutex_exit(&spa->spa_async_lock);
5932 spa_async_suspend(spa_t *spa)
5934 mutex_enter(&spa->spa_async_lock);
5935 spa->spa_async_suspended++;
5936 while (spa->spa_async_thread != NULL)
5937 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5938 mutex_exit(&spa->spa_async_lock);
5942 spa_async_resume(spa_t *spa)
5944 mutex_enter(&spa->spa_async_lock);
5945 ASSERT(spa->spa_async_suspended != 0);
5946 spa->spa_async_suspended--;
5947 mutex_exit(&spa->spa_async_lock);
5951 spa_async_tasks_pending(spa_t *spa)
5953 uint_t non_config_tasks;
5955 boolean_t config_task_suspended;
5957 non_config_tasks = spa->spa_async_tasks & ~SPA_ASYNC_CONFIG_UPDATE;
5958 config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE;
5959 if (spa->spa_ccw_fail_time == 0) {
5960 config_task_suspended = B_FALSE;
5962 config_task_suspended =
5963 (gethrtime() - spa->spa_ccw_fail_time) <
5964 (zfs_ccw_retry_interval * NANOSEC);
5967 return (non_config_tasks || (config_task && !config_task_suspended));
5971 spa_async_dispatch(spa_t *spa)
5973 mutex_enter(&spa->spa_async_lock);
5974 if (spa_async_tasks_pending(spa) &&
5975 !spa->spa_async_suspended &&
5976 spa->spa_async_thread == NULL &&
5978 spa->spa_async_thread = thread_create(NULL, 0,
5979 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
5980 mutex_exit(&spa->spa_async_lock);
5984 spa_async_request(spa_t *spa, int task)
5986 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
5987 mutex_enter(&spa->spa_async_lock);
5988 spa->spa_async_tasks |= task;
5989 mutex_exit(&spa->spa_async_lock);
5993 * ==========================================================================
5994 * SPA syncing routines
5995 * ==========================================================================
5999 bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
6002 bpobj_enqueue(bpo, bp, tx);
6007 spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
6011 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
6012 BP_GET_PSIZE(bp), zio->io_flags));
6017 * Note: this simple function is not inlined to make it easier to dtrace the
6018 * amount of time spent syncing frees.
6021 spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx)
6023 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6024 bplist_iterate(bpl, spa_free_sync_cb, zio, tx);
6025 VERIFY(zio_wait(zio) == 0);
6029 * Note: this simple function is not inlined to make it easier to dtrace the
6030 * amount of time spent syncing deferred frees.
6033 spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx)
6035 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6036 VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj,
6037 spa_free_sync_cb, zio, tx), ==, 0);
6038 VERIFY0(zio_wait(zio));
6043 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
6045 char *packed = NULL;
6050 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
6053 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
6054 * information. This avoids the dbuf_will_dirty() path and
6055 * saves us a pre-read to get data we don't actually care about.
6057 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
6058 packed = kmem_alloc(bufsize, KM_SLEEP);
6060 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
6062 bzero(packed + nvsize, bufsize - nvsize);
6064 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
6066 kmem_free(packed, bufsize);
6068 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
6069 dmu_buf_will_dirty(db, tx);
6070 *(uint64_t *)db->db_data = nvsize;
6071 dmu_buf_rele(db, FTAG);
6075 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
6076 const char *config, const char *entry)
6086 * Update the MOS nvlist describing the list of available devices.
6087 * spa_validate_aux() will have already made sure this nvlist is
6088 * valid and the vdevs are labeled appropriately.
6090 if (sav->sav_object == 0) {
6091 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
6092 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
6093 sizeof (uint64_t), tx);
6094 VERIFY(zap_update(spa->spa_meta_objset,
6095 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
6096 &sav->sav_object, tx) == 0);
6099 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
6100 if (sav->sav_count == 0) {
6101 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
6103 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
6104 for (i = 0; i < sav->sav_count; i++)
6105 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
6106 B_FALSE, VDEV_CONFIG_L2CACHE);
6107 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
6108 sav->sav_count) == 0);
6109 for (i = 0; i < sav->sav_count; i++)
6110 nvlist_free(list[i]);
6111 kmem_free(list, sav->sav_count * sizeof (void *));
6114 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
6115 nvlist_free(nvroot);
6117 sav->sav_sync = B_FALSE;
6121 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
6125 if (list_is_empty(&spa->spa_config_dirty_list))
6128 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6130 config = spa_config_generate(spa, spa->spa_root_vdev,
6131 dmu_tx_get_txg(tx), B_FALSE);
6134 * If we're upgrading the spa version then make sure that
6135 * the config object gets updated with the correct version.
6137 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
6138 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
6139 spa->spa_uberblock.ub_version);
6141 spa_config_exit(spa, SCL_STATE, FTAG);
6143 if (spa->spa_config_syncing)
6144 nvlist_free(spa->spa_config_syncing);
6145 spa->spa_config_syncing = config;
6147 spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
6151 spa_sync_version(void *arg, dmu_tx_t *tx)
6153 uint64_t *versionp = arg;
6154 uint64_t version = *versionp;
6155 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
6158 * Setting the version is special cased when first creating the pool.
6160 ASSERT(tx->tx_txg != TXG_INITIAL);
6162 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
6163 ASSERT(version >= spa_version(spa));
6165 spa->spa_uberblock.ub_version = version;
6166 vdev_config_dirty(spa->spa_root_vdev);
6167 spa_history_log_internal(spa, "set", tx, "version=%lld", version);
6171 * Set zpool properties.
6174 spa_sync_props(void *arg, dmu_tx_t *tx)
6176 nvlist_t *nvp = arg;
6177 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
6178 objset_t *mos = spa->spa_meta_objset;
6179 nvpair_t *elem = NULL;
6181 mutex_enter(&spa->spa_props_lock);
6183 while ((elem = nvlist_next_nvpair(nvp, elem))) {
6185 char *strval, *fname;
6187 const char *propname;
6188 zprop_type_t proptype;
6189 zfeature_info_t *feature;
6191 switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
6194 * We checked this earlier in spa_prop_validate().
6196 ASSERT(zpool_prop_feature(nvpair_name(elem)));
6198 fname = strchr(nvpair_name(elem), '@') + 1;
6199 VERIFY3U(0, ==, zfeature_lookup_name(fname, &feature));
6201 spa_feature_enable(spa, feature, tx);
6202 spa_history_log_internal(spa, "set", tx,
6203 "%s=enabled", nvpair_name(elem));
6206 case ZPOOL_PROP_VERSION:
6207 VERIFY(nvpair_value_uint64(elem, &intval) == 0);
6209 * The version is synced seperatly before other
6210 * properties and should be correct by now.
6212 ASSERT3U(spa_version(spa), >=, intval);
6215 case ZPOOL_PROP_ALTROOT:
6217 * 'altroot' is a non-persistent property. It should
6218 * have been set temporarily at creation or import time.
6220 ASSERT(spa->spa_root != NULL);
6223 case ZPOOL_PROP_READONLY:
6224 case ZPOOL_PROP_CACHEFILE:
6226 * 'readonly' and 'cachefile' are also non-persisitent
6230 case ZPOOL_PROP_COMMENT:
6231 VERIFY(nvpair_value_string(elem, &strval) == 0);
6232 if (spa->spa_comment != NULL)
6233 spa_strfree(spa->spa_comment);
6234 spa->spa_comment = spa_strdup(strval);
6236 * We need to dirty the configuration on all the vdevs
6237 * so that their labels get updated. It's unnecessary
6238 * to do this for pool creation since the vdev's
6239 * configuratoin has already been dirtied.
6241 if (tx->tx_txg != TXG_INITIAL)
6242 vdev_config_dirty(spa->spa_root_vdev);
6243 spa_history_log_internal(spa, "set", tx,
6244 "%s=%s", nvpair_name(elem), strval);
6248 * Set pool property values in the poolprops mos object.
6250 if (spa->spa_pool_props_object == 0) {
6251 spa->spa_pool_props_object =
6252 zap_create_link(mos, DMU_OT_POOL_PROPS,
6253 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
6257 /* normalize the property name */
6258 propname = zpool_prop_to_name(prop);
6259 proptype = zpool_prop_get_type(prop);
6261 if (nvpair_type(elem) == DATA_TYPE_STRING) {
6262 ASSERT(proptype == PROP_TYPE_STRING);
6263 VERIFY(nvpair_value_string(elem, &strval) == 0);
6264 VERIFY(zap_update(mos,
6265 spa->spa_pool_props_object, propname,
6266 1, strlen(strval) + 1, strval, tx) == 0);
6267 spa_history_log_internal(spa, "set", tx,
6268 "%s=%s", nvpair_name(elem), strval);
6269 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
6270 VERIFY(nvpair_value_uint64(elem, &intval) == 0);
6272 if (proptype == PROP_TYPE_INDEX) {
6274 VERIFY(zpool_prop_index_to_string(
6275 prop, intval, &unused) == 0);
6277 VERIFY(zap_update(mos,
6278 spa->spa_pool_props_object, propname,
6279 8, 1, &intval, tx) == 0);
6280 spa_history_log_internal(spa, "set", tx,
6281 "%s=%lld", nvpair_name(elem), intval);
6283 ASSERT(0); /* not allowed */
6287 case ZPOOL_PROP_DELEGATION:
6288 spa->spa_delegation = intval;
6290 case ZPOOL_PROP_BOOTFS:
6291 spa->spa_bootfs = intval;
6293 case ZPOOL_PROP_FAILUREMODE:
6294 spa->spa_failmode = intval;
6296 case ZPOOL_PROP_AUTOEXPAND:
6297 spa->spa_autoexpand = intval;
6298 if (tx->tx_txg != TXG_INITIAL)
6299 spa_async_request(spa,
6300 SPA_ASYNC_AUTOEXPAND);
6302 case ZPOOL_PROP_DEDUPDITTO:
6303 spa->spa_dedup_ditto = intval;
6312 mutex_exit(&spa->spa_props_lock);
6316 * Perform one-time upgrade on-disk changes. spa_version() does not
6317 * reflect the new version this txg, so there must be no changes this
6318 * txg to anything that the upgrade code depends on after it executes.
6319 * Therefore this must be called after dsl_pool_sync() does the sync
6323 spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
6325 dsl_pool_t *dp = spa->spa_dsl_pool;
6327 ASSERT(spa->spa_sync_pass == 1);
6329 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
6331 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
6332 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
6333 dsl_pool_create_origin(dp, tx);
6335 /* Keeping the origin open increases spa_minref */
6336 spa->spa_minref += 3;
6339 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
6340 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
6341 dsl_pool_upgrade_clones(dp, tx);
6344 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
6345 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
6346 dsl_pool_upgrade_dir_clones(dp, tx);
6348 /* Keeping the freedir open increases spa_minref */
6349 spa->spa_minref += 3;
6352 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES &&
6353 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
6354 spa_feature_create_zap_objects(spa, tx);
6356 rrw_exit(&dp->dp_config_rwlock, FTAG);
6360 * Sync the specified transaction group. New blocks may be dirtied as
6361 * part of the process, so we iterate until it converges.
6364 spa_sync(spa_t *spa, uint64_t txg)
6366 dsl_pool_t *dp = spa->spa_dsl_pool;
6367 objset_t *mos = spa->spa_meta_objset;
6368 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
6369 vdev_t *rvd = spa->spa_root_vdev;
6374 VERIFY(spa_writeable(spa));
6377 * Lock out configuration changes.
6379 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
6381 spa->spa_syncing_txg = txg;
6382 spa->spa_sync_pass = 0;
6385 * If there are any pending vdev state changes, convert them
6386 * into config changes that go out with this transaction group.
6388 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6389 while (list_head(&spa->spa_state_dirty_list) != NULL) {
6391 * We need the write lock here because, for aux vdevs,
6392 * calling vdev_config_dirty() modifies sav_config.
6393 * This is ugly and will become unnecessary when we
6394 * eliminate the aux vdev wart by integrating all vdevs
6395 * into the root vdev tree.
6397 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6398 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
6399 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
6400 vdev_state_clean(vd);
6401 vdev_config_dirty(vd);
6403 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6404 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
6406 spa_config_exit(spa, SCL_STATE, FTAG);
6408 tx = dmu_tx_create_assigned(dp, txg);
6410 spa->spa_sync_starttime = gethrtime();
6412 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid,
6413 spa->spa_sync_starttime + spa->spa_deadman_synctime));
6416 callout_reset(&spa->spa_deadman_cycid,
6417 hz * spa->spa_deadman_synctime / NANOSEC, spa_deadman, spa);
6422 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
6423 * set spa_deflate if we have no raid-z vdevs.
6425 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
6426 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
6429 for (i = 0; i < rvd->vdev_children; i++) {
6430 vd = rvd->vdev_child[i];
6431 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
6434 if (i == rvd->vdev_children) {
6435 spa->spa_deflate = TRUE;
6436 VERIFY(0 == zap_add(spa->spa_meta_objset,
6437 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
6438 sizeof (uint64_t), 1, &spa->spa_deflate, tx));
6443 * If anything has changed in this txg, or if someone is waiting
6444 * for this txg to sync (eg, spa_vdev_remove()), push the
6445 * deferred frees from the previous txg. If not, leave them
6446 * alone so that we don't generate work on an otherwise idle
6449 if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
6450 !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
6451 !txg_list_empty(&dp->dp_sync_tasks, txg) ||
6452 ((dsl_scan_active(dp->dp_scan) ||
6453 txg_sync_waiting(dp)) && !spa_shutting_down(spa))) {
6454 spa_sync_deferred_frees(spa, tx);
6458 * Iterate to convergence.
6461 int pass = ++spa->spa_sync_pass;
6463 spa_sync_config_object(spa, tx);
6464 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
6465 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
6466 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
6467 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
6468 spa_errlog_sync(spa, txg);
6469 dsl_pool_sync(dp, txg);
6471 if (pass < zfs_sync_pass_deferred_free) {
6472 spa_sync_frees(spa, free_bpl, tx);
6474 bplist_iterate(free_bpl, bpobj_enqueue_cb,
6475 &spa->spa_deferred_bpobj, tx);
6479 dsl_scan_sync(dp, tx);
6481 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
6485 spa_sync_upgrades(spa, tx);
6487 } while (dmu_objset_is_dirty(mos, txg));
6490 * Rewrite the vdev configuration (which includes the uberblock)
6491 * to commit the transaction group.
6493 * If there are no dirty vdevs, we sync the uberblock to a few
6494 * random top-level vdevs that are known to be visible in the
6495 * config cache (see spa_vdev_add() for a complete description).
6496 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
6500 * We hold SCL_STATE to prevent vdev open/close/etc.
6501 * while we're attempting to write the vdev labels.
6503 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6505 if (list_is_empty(&spa->spa_config_dirty_list)) {
6506 vdev_t *svd[SPA_DVAS_PER_BP];
6508 int children = rvd->vdev_children;
6509 int c0 = spa_get_random(children);
6511 for (int c = 0; c < children; c++) {
6512 vd = rvd->vdev_child[(c0 + c) % children];
6513 if (vd->vdev_ms_array == 0 || vd->vdev_islog)
6515 svd[svdcount++] = vd;
6516 if (svdcount == SPA_DVAS_PER_BP)
6519 error = vdev_config_sync(svd, svdcount, txg, B_FALSE);
6521 error = vdev_config_sync(svd, svdcount, txg,
6524 error = vdev_config_sync(rvd->vdev_child,
6525 rvd->vdev_children, txg, B_FALSE);
6527 error = vdev_config_sync(rvd->vdev_child,
6528 rvd->vdev_children, txg, B_TRUE);
6532 spa->spa_last_synced_guid = rvd->vdev_guid;
6534 spa_config_exit(spa, SCL_STATE, FTAG);
6538 zio_suspend(spa, NULL);
6539 zio_resume_wait(spa);
6544 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY));
6547 callout_drain(&spa->spa_deadman_cycid);
6552 * Clear the dirty config list.
6554 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
6555 vdev_config_clean(vd);
6558 * Now that the new config has synced transactionally,
6559 * let it become visible to the config cache.
6561 if (spa->spa_config_syncing != NULL) {
6562 spa_config_set(spa, spa->spa_config_syncing);
6563 spa->spa_config_txg = txg;
6564 spa->spa_config_syncing = NULL;
6567 spa->spa_ubsync = spa->spa_uberblock;
6569 dsl_pool_sync_done(dp, txg);
6572 * Update usable space statistics.
6574 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
6575 vdev_sync_done(vd, txg);
6577 spa_update_dspace(spa);
6580 * It had better be the case that we didn't dirty anything
6581 * since vdev_config_sync().
6583 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
6584 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
6585 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
6587 spa->spa_sync_pass = 0;
6589 spa_config_exit(spa, SCL_CONFIG, FTAG);
6591 spa_handle_ignored_writes(spa);
6594 * If any async tasks have been requested, kick them off.
6596 spa_async_dispatch(spa);
6600 * Sync all pools. We don't want to hold the namespace lock across these
6601 * operations, so we take a reference on the spa_t and drop the lock during the
6605 spa_sync_allpools(void)
6608 mutex_enter(&spa_namespace_lock);
6609 while ((spa = spa_next(spa)) != NULL) {
6610 if (spa_state(spa) != POOL_STATE_ACTIVE ||
6611 !spa_writeable(spa) || spa_suspended(spa))
6613 spa_open_ref(spa, FTAG);
6614 mutex_exit(&spa_namespace_lock);
6615 txg_wait_synced(spa_get_dsl(spa), 0);
6616 mutex_enter(&spa_namespace_lock);
6617 spa_close(spa, FTAG);
6619 mutex_exit(&spa_namespace_lock);
6623 * ==========================================================================
6624 * Miscellaneous routines
6625 * ==========================================================================
6629 * Remove all pools in the system.
6637 * Remove all cached state. All pools should be closed now,
6638 * so every spa in the AVL tree should be unreferenced.
6640 mutex_enter(&spa_namespace_lock);
6641 while ((spa = spa_next(NULL)) != NULL) {
6643 * Stop async tasks. The async thread may need to detach
6644 * a device that's been replaced, which requires grabbing
6645 * spa_namespace_lock, so we must drop it here.
6647 spa_open_ref(spa, FTAG);
6648 mutex_exit(&spa_namespace_lock);
6649 spa_async_suspend(spa);
6650 mutex_enter(&spa_namespace_lock);
6651 spa_close(spa, FTAG);
6653 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
6655 spa_deactivate(spa);
6659 mutex_exit(&spa_namespace_lock);
6663 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
6668 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
6672 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
6673 vd = spa->spa_l2cache.sav_vdevs[i];
6674 if (vd->vdev_guid == guid)
6678 for (i = 0; i < spa->spa_spares.sav_count; i++) {
6679 vd = spa->spa_spares.sav_vdevs[i];
6680 if (vd->vdev_guid == guid)
6689 spa_upgrade(spa_t *spa, uint64_t version)
6691 ASSERT(spa_writeable(spa));
6693 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6696 * This should only be called for a non-faulted pool, and since a
6697 * future version would result in an unopenable pool, this shouldn't be
6700 ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version));
6701 ASSERT(version >= spa->spa_uberblock.ub_version);
6703 spa->spa_uberblock.ub_version = version;
6704 vdev_config_dirty(spa->spa_root_vdev);
6706 spa_config_exit(spa, SCL_ALL, FTAG);
6708 txg_wait_synced(spa_get_dsl(spa), 0);
6712 spa_has_spare(spa_t *spa, uint64_t guid)
6716 spa_aux_vdev_t *sav = &spa->spa_spares;
6718 for (i = 0; i < sav->sav_count; i++)
6719 if (sav->sav_vdevs[i]->vdev_guid == guid)
6722 for (i = 0; i < sav->sav_npending; i++) {
6723 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
6724 &spareguid) == 0 && spareguid == guid)
6732 * Check if a pool has an active shared spare device.
6733 * Note: reference count of an active spare is 2, as a spare and as a replace
6736 spa_has_active_shared_spare(spa_t *spa)
6740 spa_aux_vdev_t *sav = &spa->spa_spares;
6742 for (i = 0; i < sav->sav_count; i++) {
6743 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
6744 &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
6753 * Post a sysevent corresponding to the given event. The 'name' must be one of
6754 * the event definitions in sys/sysevent/eventdefs.h. The payload will be
6755 * filled in from the spa and (optionally) the vdev. This doesn't do anything
6756 * in the userland libzpool, as we don't want consumers to misinterpret ztest
6757 * or zdb as real changes.
6760 spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
6764 sysevent_attr_list_t *attr = NULL;
6765 sysevent_value_t value;
6768 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
6771 value.value_type = SE_DATA_TYPE_STRING;
6772 value.value.sv_string = spa_name(spa);
6773 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
6776 value.value_type = SE_DATA_TYPE_UINT64;
6777 value.value.sv_uint64 = spa_guid(spa);
6778 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
6782 value.value_type = SE_DATA_TYPE_UINT64;
6783 value.value.sv_uint64 = vd->vdev_guid;
6784 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
6788 if (vd->vdev_path) {
6789 value.value_type = SE_DATA_TYPE_STRING;
6790 value.value.sv_string = vd->vdev_path;
6791 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
6792 &value, SE_SLEEP) != 0)
6797 if (sysevent_attach_attributes(ev, attr) != 0)
6801 (void) log_sysevent(ev, SE_SLEEP, &eid);
6805 sysevent_free_attr(attr);