4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2013 by Delphix. All rights reserved.
25 * Copyright (c) 2013, 2014, Nexenta Systems, Inc. All rights reserved.
26 * Copyright (c) 2013 Martin Matuska <mm@FreeBSD.org>. All rights reserved.
30 * SPA: Storage Pool Allocator
32 * This file contains all the routines used when modifying on-disk SPA state.
33 * This includes opening, importing, destroying, exporting a pool, and syncing a
37 #include <sys/zfs_context.h>
38 #include <sys/fm/fs/zfs.h>
39 #include <sys/spa_impl.h>
41 #include <sys/zio_checksum.h>
43 #include <sys/dmu_tx.h>
47 #include <sys/vdev_impl.h>
48 #include <sys/metaslab.h>
49 #include <sys/metaslab_impl.h>
50 #include <sys/uberblock_impl.h>
53 #include <sys/dmu_traverse.h>
54 #include <sys/dmu_objset.h>
55 #include <sys/unique.h>
56 #include <sys/dsl_pool.h>
57 #include <sys/dsl_dataset.h>
58 #include <sys/dsl_dir.h>
59 #include <sys/dsl_prop.h>
60 #include <sys/dsl_synctask.h>
61 #include <sys/fs/zfs.h>
63 #include <sys/callb.h>
64 #include <sys/spa_boot.h>
65 #include <sys/zfs_ioctl.h>
66 #include <sys/dsl_scan.h>
67 #include <sys/dmu_send.h>
68 #include <sys/dsl_destroy.h>
69 #include <sys/dsl_userhold.h>
70 #include <sys/zfeature.h>
72 #include <sys/trim_map.h>
75 #include <sys/callb.h>
76 #include <sys/cpupart.h>
81 #include "zfs_comutil.h"
83 /* Check hostid on import? */
84 static int check_hostid = 1;
86 SYSCTL_DECL(_vfs_zfs);
87 TUNABLE_INT("vfs.zfs.check_hostid", &check_hostid);
88 SYSCTL_INT(_vfs_zfs, OID_AUTO, check_hostid, CTLFLAG_RW, &check_hostid, 0,
89 "Check hostid on import?");
92 * The interval, in seconds, at which failed configuration cache file writes
95 static int zfs_ccw_retry_interval = 300;
97 typedef enum zti_modes {
98 ZTI_MODE_FIXED, /* value is # of threads (min 1) */
99 ZTI_MODE_BATCH, /* cpu-intensive; value is ignored */
100 ZTI_MODE_NULL, /* don't create a taskq */
104 #define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) }
105 #define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 }
106 #define ZTI_NULL { ZTI_MODE_NULL, 0, 0 }
108 #define ZTI_N(n) ZTI_P(n, 1)
109 #define ZTI_ONE ZTI_N(1)
111 typedef struct zio_taskq_info {
112 zti_modes_t zti_mode;
117 static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
118 "issue", "issue_high", "intr", "intr_high"
122 * This table defines the taskq settings for each ZFS I/O type. When
123 * initializing a pool, we use this table to create an appropriately sized
124 * taskq. Some operations are low volume and therefore have a small, static
125 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE
126 * macros. Other operations process a large amount of data; the ZTI_BATCH
127 * macro causes us to create a taskq oriented for throughput. Some operations
128 * are so high frequency and short-lived that the taskq itself can become a a
129 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an
130 * additional degree of parallelism specified by the number of threads per-
131 * taskq and the number of taskqs; when dispatching an event in this case, the
132 * particular taskq is chosen at random.
134 * The different taskq priorities are to handle the different contexts (issue
135 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that
136 * need to be handled with minimum delay.
138 const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
139 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
140 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* NULL */
141 { ZTI_N(8), ZTI_NULL, ZTI_BATCH, ZTI_NULL }, /* READ */
142 { ZTI_BATCH, ZTI_N(5), ZTI_N(8), ZTI_N(5) }, /* WRITE */
143 { ZTI_P(12, 8), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FREE */
144 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* CLAIM */
145 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* IOCTL */
148 static void spa_sync_version(void *arg, dmu_tx_t *tx);
149 static void spa_sync_props(void *arg, dmu_tx_t *tx);
150 static boolean_t spa_has_active_shared_spare(spa_t *spa);
151 static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
152 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
154 static void spa_vdev_resilver_done(spa_t *spa);
156 uint_t zio_taskq_batch_pct = 75; /* 1 thread per cpu in pset */
158 id_t zio_taskq_psrset_bind = PS_NONE;
161 boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */
163 uint_t zio_taskq_basedc = 80; /* base duty cycle */
165 boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */
166 extern int zfs_sync_pass_deferred_free;
169 extern void spa_deadman(void *arg);
173 * This (illegal) pool name is used when temporarily importing a spa_t in order
174 * to get the vdev stats associated with the imported devices.
176 #define TRYIMPORT_NAME "$import"
179 * ==========================================================================
180 * SPA properties routines
181 * ==========================================================================
185 * Add a (source=src, propname=propval) list to an nvlist.
188 spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
189 uint64_t intval, zprop_source_t src)
191 const char *propname = zpool_prop_to_name(prop);
194 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
195 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
198 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
200 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
202 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
203 nvlist_free(propval);
207 * Get property values from the spa configuration.
210 spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
212 vdev_t *rvd = spa->spa_root_vdev;
213 dsl_pool_t *pool = spa->spa_dsl_pool;
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);
262 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
263 * when opening pools before this version freedir will be NULL.
265 if (pool->dp_free_dir != NULL) {
266 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
267 pool->dp_free_dir->dd_phys->dd_used_bytes, src);
269 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
273 if (pool->dp_leak_dir != NULL) {
274 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, NULL,
275 pool->dp_leak_dir->dd_phys->dd_used_bytes, src);
277 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED,
282 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
284 if (spa->spa_comment != NULL) {
285 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
289 if (spa->spa_root != NULL)
290 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
293 if ((dp = list_head(&spa->spa_config_list)) != NULL) {
294 if (dp->scd_path == NULL) {
295 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
296 "none", 0, ZPROP_SRC_LOCAL);
297 } else if (strcmp(dp->scd_path, spa_config_path) != 0) {
298 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
299 dp->scd_path, 0, ZPROP_SRC_LOCAL);
305 * Get zpool property values.
308 spa_prop_get(spa_t *spa, nvlist_t **nvp)
310 objset_t *mos = spa->spa_meta_objset;
315 VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
317 mutex_enter(&spa->spa_props_lock);
320 * Get properties from the spa config.
322 spa_prop_get_config(spa, nvp);
324 /* If no pool property object, no more prop to get. */
325 if (mos == NULL || spa->spa_pool_props_object == 0) {
326 mutex_exit(&spa->spa_props_lock);
331 * Get properties from the MOS pool property object.
333 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
334 (err = zap_cursor_retrieve(&zc, &za)) == 0;
335 zap_cursor_advance(&zc)) {
338 zprop_source_t src = ZPROP_SRC_DEFAULT;
341 if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
344 switch (za.za_integer_length) {
346 /* integer property */
347 if (za.za_first_integer !=
348 zpool_prop_default_numeric(prop))
349 src = ZPROP_SRC_LOCAL;
351 if (prop == ZPOOL_PROP_BOOTFS) {
353 dsl_dataset_t *ds = NULL;
355 dp = spa_get_dsl(spa);
356 dsl_pool_config_enter(dp, FTAG);
357 if (err = dsl_dataset_hold_obj(dp,
358 za.za_first_integer, FTAG, &ds)) {
359 dsl_pool_config_exit(dp, FTAG);
364 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
366 dsl_dataset_name(ds, strval);
367 dsl_dataset_rele(ds, FTAG);
368 dsl_pool_config_exit(dp, FTAG);
371 intval = za.za_first_integer;
374 spa_prop_add_list(*nvp, prop, strval, intval, src);
378 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
383 /* string property */
384 strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
385 err = zap_lookup(mos, spa->spa_pool_props_object,
386 za.za_name, 1, za.za_num_integers, strval);
388 kmem_free(strval, za.za_num_integers);
391 spa_prop_add_list(*nvp, prop, strval, 0, src);
392 kmem_free(strval, za.za_num_integers);
399 zap_cursor_fini(&zc);
400 mutex_exit(&spa->spa_props_lock);
402 if (err && err != ENOENT) {
412 * Validate the given pool properties nvlist and modify the list
413 * for the property values to be set.
416 spa_prop_validate(spa_t *spa, nvlist_t *props)
419 int error = 0, reset_bootfs = 0;
421 boolean_t has_feature = B_FALSE;
424 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
426 char *strval, *slash, *check, *fname;
427 const char *propname = nvpair_name(elem);
428 zpool_prop_t prop = zpool_name_to_prop(propname);
432 if (!zpool_prop_feature(propname)) {
433 error = SET_ERROR(EINVAL);
438 * Sanitize the input.
440 if (nvpair_type(elem) != DATA_TYPE_UINT64) {
441 error = SET_ERROR(EINVAL);
445 if (nvpair_value_uint64(elem, &intval) != 0) {
446 error = SET_ERROR(EINVAL);
451 error = SET_ERROR(EINVAL);
455 fname = strchr(propname, '@') + 1;
456 if (zfeature_lookup_name(fname, NULL) != 0) {
457 error = SET_ERROR(EINVAL);
461 has_feature = B_TRUE;
464 case ZPOOL_PROP_VERSION:
465 error = nvpair_value_uint64(elem, &intval);
467 (intval < spa_version(spa) ||
468 intval > SPA_VERSION_BEFORE_FEATURES ||
470 error = SET_ERROR(EINVAL);
473 case ZPOOL_PROP_DELEGATION:
474 case ZPOOL_PROP_AUTOREPLACE:
475 case ZPOOL_PROP_LISTSNAPS:
476 case ZPOOL_PROP_AUTOEXPAND:
477 error = nvpair_value_uint64(elem, &intval);
478 if (!error && intval > 1)
479 error = SET_ERROR(EINVAL);
482 case ZPOOL_PROP_BOOTFS:
484 * If the pool version is less than SPA_VERSION_BOOTFS,
485 * or the pool is still being created (version == 0),
486 * the bootfs property cannot be set.
488 if (spa_version(spa) < SPA_VERSION_BOOTFS) {
489 error = SET_ERROR(ENOTSUP);
494 * Make sure the vdev config is bootable
496 if (!vdev_is_bootable(spa->spa_root_vdev)) {
497 error = SET_ERROR(ENOTSUP);
503 error = nvpair_value_string(elem, &strval);
509 if (strval == NULL || strval[0] == '\0') {
510 objnum = zpool_prop_default_numeric(
515 if (error = dmu_objset_hold(strval, FTAG, &os))
518 /* Must be ZPL and not gzip compressed. */
520 if (dmu_objset_type(os) != DMU_OST_ZFS) {
521 error = SET_ERROR(ENOTSUP);
523 dsl_prop_get_int_ds(dmu_objset_ds(os),
524 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
526 !BOOTFS_COMPRESS_VALID(compress)) {
527 error = SET_ERROR(ENOTSUP);
529 objnum = dmu_objset_id(os);
531 dmu_objset_rele(os, FTAG);
535 case ZPOOL_PROP_FAILUREMODE:
536 error = nvpair_value_uint64(elem, &intval);
537 if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
538 intval > ZIO_FAILURE_MODE_PANIC))
539 error = SET_ERROR(EINVAL);
542 * This is a special case which only occurs when
543 * the pool has completely failed. This allows
544 * the user to change the in-core failmode property
545 * without syncing it out to disk (I/Os might
546 * currently be blocked). We do this by returning
547 * EIO to the caller (spa_prop_set) to trick it
548 * into thinking we encountered a property validation
551 if (!error && spa_suspended(spa)) {
552 spa->spa_failmode = intval;
553 error = SET_ERROR(EIO);
557 case ZPOOL_PROP_CACHEFILE:
558 if ((error = nvpair_value_string(elem, &strval)) != 0)
561 if (strval[0] == '\0')
564 if (strcmp(strval, "none") == 0)
567 if (strval[0] != '/') {
568 error = SET_ERROR(EINVAL);
572 slash = strrchr(strval, '/');
573 ASSERT(slash != NULL);
575 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
576 strcmp(slash, "/..") == 0)
577 error = SET_ERROR(EINVAL);
580 case ZPOOL_PROP_COMMENT:
581 if ((error = nvpair_value_string(elem, &strval)) != 0)
583 for (check = strval; *check != '\0'; check++) {
585 * The kernel doesn't have an easy isprint()
586 * check. For this kernel check, we merely
587 * check ASCII apart from DEL. Fix this if
588 * there is an easy-to-use kernel isprint().
590 if (*check >= 0x7f) {
591 error = SET_ERROR(EINVAL);
596 if (strlen(strval) > ZPROP_MAX_COMMENT)
600 case ZPOOL_PROP_DEDUPDITTO:
601 if (spa_version(spa) < SPA_VERSION_DEDUP)
602 error = SET_ERROR(ENOTSUP);
604 error = nvpair_value_uint64(elem, &intval);
606 intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
607 error = SET_ERROR(EINVAL);
615 if (!error && reset_bootfs) {
616 error = nvlist_remove(props,
617 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
620 error = nvlist_add_uint64(props,
621 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
629 spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
632 spa_config_dirent_t *dp;
634 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
638 dp = kmem_alloc(sizeof (spa_config_dirent_t),
641 if (cachefile[0] == '\0')
642 dp->scd_path = spa_strdup(spa_config_path);
643 else if (strcmp(cachefile, "none") == 0)
646 dp->scd_path = spa_strdup(cachefile);
648 list_insert_head(&spa->spa_config_list, dp);
650 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
654 spa_prop_set(spa_t *spa, nvlist_t *nvp)
657 nvpair_t *elem = NULL;
658 boolean_t need_sync = B_FALSE;
660 if ((error = spa_prop_validate(spa, nvp)) != 0)
663 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
664 zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem));
666 if (prop == ZPOOL_PROP_CACHEFILE ||
667 prop == ZPOOL_PROP_ALTROOT ||
668 prop == ZPOOL_PROP_READONLY)
671 if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) {
674 if (prop == ZPOOL_PROP_VERSION) {
675 VERIFY(nvpair_value_uint64(elem, &ver) == 0);
677 ASSERT(zpool_prop_feature(nvpair_name(elem)));
678 ver = SPA_VERSION_FEATURES;
682 /* Save time if the version is already set. */
683 if (ver == spa_version(spa))
687 * In addition to the pool directory object, we might
688 * create the pool properties object, the features for
689 * read object, the features for write object, or the
690 * feature descriptions object.
692 error = dsl_sync_task(spa->spa_name, NULL,
693 spa_sync_version, &ver, 6);
704 return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props,
712 * If the bootfs property value is dsobj, clear it.
715 spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
717 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
718 VERIFY(zap_remove(spa->spa_meta_objset,
719 spa->spa_pool_props_object,
720 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
727 spa_change_guid_check(void *arg, dmu_tx_t *tx)
729 uint64_t *newguid = arg;
730 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
731 vdev_t *rvd = spa->spa_root_vdev;
734 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
735 vdev_state = rvd->vdev_state;
736 spa_config_exit(spa, SCL_STATE, FTAG);
738 if (vdev_state != VDEV_STATE_HEALTHY)
739 return (SET_ERROR(ENXIO));
741 ASSERT3U(spa_guid(spa), !=, *newguid);
747 spa_change_guid_sync(void *arg, dmu_tx_t *tx)
749 uint64_t *newguid = arg;
750 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
752 vdev_t *rvd = spa->spa_root_vdev;
754 oldguid = spa_guid(spa);
756 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
757 rvd->vdev_guid = *newguid;
758 rvd->vdev_guid_sum += (*newguid - oldguid);
759 vdev_config_dirty(rvd);
760 spa_config_exit(spa, SCL_STATE, FTAG);
762 spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu",
767 * Change the GUID for the pool. This is done so that we can later
768 * re-import a pool built from a clone of our own vdevs. We will modify
769 * the root vdev's guid, our own pool guid, and then mark all of our
770 * vdevs dirty. Note that we must make sure that all our vdevs are
771 * online when we do this, or else any vdevs that weren't present
772 * would be orphaned from our pool. We are also going to issue a
773 * sysevent to update any watchers.
776 spa_change_guid(spa_t *spa)
781 mutex_enter(&spa->spa_vdev_top_lock);
782 mutex_enter(&spa_namespace_lock);
783 guid = spa_generate_guid(NULL);
785 error = dsl_sync_task(spa->spa_name, spa_change_guid_check,
786 spa_change_guid_sync, &guid, 5);
789 spa_config_sync(spa, B_FALSE, B_TRUE);
790 spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID);
793 mutex_exit(&spa_namespace_lock);
794 mutex_exit(&spa->spa_vdev_top_lock);
800 * ==========================================================================
801 * SPA state manipulation (open/create/destroy/import/export)
802 * ==========================================================================
806 spa_error_entry_compare(const void *a, const void *b)
808 spa_error_entry_t *sa = (spa_error_entry_t *)a;
809 spa_error_entry_t *sb = (spa_error_entry_t *)b;
812 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
813 sizeof (zbookmark_phys_t));
824 * Utility function which retrieves copies of the current logs and
825 * re-initializes them in the process.
828 spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
830 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
832 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
833 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
835 avl_create(&spa->spa_errlist_scrub,
836 spa_error_entry_compare, sizeof (spa_error_entry_t),
837 offsetof(spa_error_entry_t, se_avl));
838 avl_create(&spa->spa_errlist_last,
839 spa_error_entry_compare, sizeof (spa_error_entry_t),
840 offsetof(spa_error_entry_t, se_avl));
844 spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
846 const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
847 enum zti_modes mode = ztip->zti_mode;
848 uint_t value = ztip->zti_value;
849 uint_t count = ztip->zti_count;
850 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
853 boolean_t batch = B_FALSE;
855 if (mode == ZTI_MODE_NULL) {
857 tqs->stqs_taskq = NULL;
861 ASSERT3U(count, >, 0);
863 tqs->stqs_count = count;
864 tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP);
868 ASSERT3U(value, >=, 1);
869 value = MAX(value, 1);
874 flags |= TASKQ_THREADS_CPU_PCT;
875 value = zio_taskq_batch_pct;
879 panic("unrecognized mode for %s_%s taskq (%u:%u) in "
881 zio_type_name[t], zio_taskq_types[q], mode, value);
885 for (uint_t i = 0; i < count; i++) {
889 (void) snprintf(name, sizeof (name), "%s_%s_%u",
890 zio_type_name[t], zio_taskq_types[q], i);
892 (void) snprintf(name, sizeof (name), "%s_%s",
893 zio_type_name[t], zio_taskq_types[q]);
897 if (zio_taskq_sysdc && spa->spa_proc != &p0) {
899 flags |= TASKQ_DC_BATCH;
901 tq = taskq_create_sysdc(name, value, 50, INT_MAX,
902 spa->spa_proc, zio_taskq_basedc, flags);
905 pri_t pri = maxclsyspri;
907 * The write issue taskq can be extremely CPU
908 * intensive. Run it at slightly lower priority
909 * than the other taskqs.
911 if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE)
914 tq = taskq_create_proc(name, value, pri, 50,
915 INT_MAX, spa->spa_proc, flags);
920 tqs->stqs_taskq[i] = tq;
925 spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
927 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
929 if (tqs->stqs_taskq == NULL) {
930 ASSERT0(tqs->stqs_count);
934 for (uint_t i = 0; i < tqs->stqs_count; i++) {
935 ASSERT3P(tqs->stqs_taskq[i], !=, NULL);
936 taskq_destroy(tqs->stqs_taskq[i]);
939 kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *));
940 tqs->stqs_taskq = NULL;
944 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
945 * Note that a type may have multiple discrete taskqs to avoid lock contention
946 * on the taskq itself. In that case we choose which taskq at random by using
947 * the low bits of gethrtime().
950 spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q,
951 task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent)
953 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
956 ASSERT3P(tqs->stqs_taskq, !=, NULL);
957 ASSERT3U(tqs->stqs_count, !=, 0);
959 if (tqs->stqs_count == 1) {
960 tq = tqs->stqs_taskq[0];
963 tq = tqs->stqs_taskq[cpu_ticks() % tqs->stqs_count];
965 tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count];
969 taskq_dispatch_ent(tq, func, arg, flags, ent);
973 spa_create_zio_taskqs(spa_t *spa)
975 for (int t = 0; t < ZIO_TYPES; t++) {
976 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
977 spa_taskqs_init(spa, t, q);
985 spa_thread(void *arg)
990 user_t *pu = PTOU(curproc);
992 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
995 ASSERT(curproc != &p0);
996 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
997 "zpool-%s", spa->spa_name);
998 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
1001 /* bind this thread to the requested psrset */
1002 if (zio_taskq_psrset_bind != PS_NONE) {
1004 mutex_enter(&cpu_lock);
1005 mutex_enter(&pidlock);
1006 mutex_enter(&curproc->p_lock);
1008 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
1009 0, NULL, NULL) == 0) {
1010 curthread->t_bind_pset = zio_taskq_psrset_bind;
1013 "Couldn't bind process for zfs pool \"%s\" to "
1014 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
1017 mutex_exit(&curproc->p_lock);
1018 mutex_exit(&pidlock);
1019 mutex_exit(&cpu_lock);
1025 if (zio_taskq_sysdc) {
1026 sysdc_thread_enter(curthread, 100, 0);
1030 spa->spa_proc = curproc;
1031 spa->spa_did = curthread->t_did;
1033 spa_create_zio_taskqs(spa);
1035 mutex_enter(&spa->spa_proc_lock);
1036 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
1038 spa->spa_proc_state = SPA_PROC_ACTIVE;
1039 cv_broadcast(&spa->spa_proc_cv);
1041 CALLB_CPR_SAFE_BEGIN(&cprinfo);
1042 while (spa->spa_proc_state == SPA_PROC_ACTIVE)
1043 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1044 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
1046 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
1047 spa->spa_proc_state = SPA_PROC_GONE;
1048 spa->spa_proc = &p0;
1049 cv_broadcast(&spa->spa_proc_cv);
1050 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */
1052 mutex_enter(&curproc->p_lock);
1055 #endif /* SPA_PROCESS */
1059 * Activate an uninitialized pool.
1062 spa_activate(spa_t *spa, int mode)
1064 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
1066 spa->spa_state = POOL_STATE_ACTIVE;
1067 spa->spa_mode = mode;
1069 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
1070 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
1072 /* Try to create a covering process */
1073 mutex_enter(&spa->spa_proc_lock);
1074 ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
1075 ASSERT(spa->spa_proc == &p0);
1079 /* Only create a process if we're going to be around a while. */
1080 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
1081 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
1083 spa->spa_proc_state = SPA_PROC_CREATED;
1084 while (spa->spa_proc_state == SPA_PROC_CREATED) {
1085 cv_wait(&spa->spa_proc_cv,
1086 &spa->spa_proc_lock);
1088 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1089 ASSERT(spa->spa_proc != &p0);
1090 ASSERT(spa->spa_did != 0);
1094 "Couldn't create process for zfs pool \"%s\"\n",
1099 #endif /* SPA_PROCESS */
1100 mutex_exit(&spa->spa_proc_lock);
1102 /* If we didn't create a process, we need to create our taskqs. */
1103 ASSERT(spa->spa_proc == &p0);
1104 if (spa->spa_proc == &p0) {
1105 spa_create_zio_taskqs(spa);
1109 * Start TRIM thread.
1111 trim_thread_create(spa);
1113 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
1114 offsetof(vdev_t, vdev_config_dirty_node));
1115 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
1116 offsetof(vdev_t, vdev_state_dirty_node));
1118 txg_list_create(&spa->spa_vdev_txg_list,
1119 offsetof(struct vdev, vdev_txg_node));
1121 avl_create(&spa->spa_errlist_scrub,
1122 spa_error_entry_compare, sizeof (spa_error_entry_t),
1123 offsetof(spa_error_entry_t, se_avl));
1124 avl_create(&spa->spa_errlist_last,
1125 spa_error_entry_compare, sizeof (spa_error_entry_t),
1126 offsetof(spa_error_entry_t, se_avl));
1130 * Opposite of spa_activate().
1133 spa_deactivate(spa_t *spa)
1135 ASSERT(spa->spa_sync_on == B_FALSE);
1136 ASSERT(spa->spa_dsl_pool == NULL);
1137 ASSERT(spa->spa_root_vdev == NULL);
1138 ASSERT(spa->spa_async_zio_root == NULL);
1139 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
1142 * Stop TRIM thread in case spa_unload() wasn't called directly
1143 * before spa_deactivate().
1145 trim_thread_destroy(spa);
1147 txg_list_destroy(&spa->spa_vdev_txg_list);
1149 list_destroy(&spa->spa_config_dirty_list);
1150 list_destroy(&spa->spa_state_dirty_list);
1152 for (int t = 0; t < ZIO_TYPES; t++) {
1153 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1154 spa_taskqs_fini(spa, t, q);
1158 metaslab_class_destroy(spa->spa_normal_class);
1159 spa->spa_normal_class = NULL;
1161 metaslab_class_destroy(spa->spa_log_class);
1162 spa->spa_log_class = NULL;
1165 * If this was part of an import or the open otherwise failed, we may
1166 * still have errors left in the queues. Empty them just in case.
1168 spa_errlog_drain(spa);
1170 avl_destroy(&spa->spa_errlist_scrub);
1171 avl_destroy(&spa->spa_errlist_last);
1173 spa->spa_state = POOL_STATE_UNINITIALIZED;
1175 mutex_enter(&spa->spa_proc_lock);
1176 if (spa->spa_proc_state != SPA_PROC_NONE) {
1177 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1178 spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1179 cv_broadcast(&spa->spa_proc_cv);
1180 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1181 ASSERT(spa->spa_proc != &p0);
1182 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1184 ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1185 spa->spa_proc_state = SPA_PROC_NONE;
1187 ASSERT(spa->spa_proc == &p0);
1188 mutex_exit(&spa->spa_proc_lock);
1192 * We want to make sure spa_thread() has actually exited the ZFS
1193 * module, so that the module can't be unloaded out from underneath
1196 if (spa->spa_did != 0) {
1197 thread_join(spa->spa_did);
1200 #endif /* SPA_PROCESS */
1204 * Verify a pool configuration, and construct the vdev tree appropriately. This
1205 * will create all the necessary vdevs in the appropriate layout, with each vdev
1206 * in the CLOSED state. This will prep the pool before open/creation/import.
1207 * All vdev validation is done by the vdev_alloc() routine.
1210 spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1211 uint_t id, int atype)
1217 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1220 if ((*vdp)->vdev_ops->vdev_op_leaf)
1223 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1226 if (error == ENOENT)
1232 return (SET_ERROR(EINVAL));
1235 for (int c = 0; c < children; c++) {
1237 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1245 ASSERT(*vdp != NULL);
1251 * Opposite of spa_load().
1254 spa_unload(spa_t *spa)
1258 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1263 trim_thread_destroy(spa);
1268 spa_async_suspend(spa);
1273 if (spa->spa_sync_on) {
1274 txg_sync_stop(spa->spa_dsl_pool);
1275 spa->spa_sync_on = B_FALSE;
1279 * Wait for any outstanding async I/O to complete.
1281 if (spa->spa_async_zio_root != NULL) {
1282 (void) zio_wait(spa->spa_async_zio_root);
1283 spa->spa_async_zio_root = NULL;
1286 bpobj_close(&spa->spa_deferred_bpobj);
1288 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1293 if (spa->spa_root_vdev)
1294 vdev_free(spa->spa_root_vdev);
1295 ASSERT(spa->spa_root_vdev == NULL);
1298 * Close the dsl pool.
1300 if (spa->spa_dsl_pool) {
1301 dsl_pool_close(spa->spa_dsl_pool);
1302 spa->spa_dsl_pool = NULL;
1303 spa->spa_meta_objset = NULL;
1310 * Drop and purge level 2 cache
1312 spa_l2cache_drop(spa);
1314 for (i = 0; i < spa->spa_spares.sav_count; i++)
1315 vdev_free(spa->spa_spares.sav_vdevs[i]);
1316 if (spa->spa_spares.sav_vdevs) {
1317 kmem_free(spa->spa_spares.sav_vdevs,
1318 spa->spa_spares.sav_count * sizeof (void *));
1319 spa->spa_spares.sav_vdevs = NULL;
1321 if (spa->spa_spares.sav_config) {
1322 nvlist_free(spa->spa_spares.sav_config);
1323 spa->spa_spares.sav_config = NULL;
1325 spa->spa_spares.sav_count = 0;
1327 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1328 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1329 vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1331 if (spa->spa_l2cache.sav_vdevs) {
1332 kmem_free(spa->spa_l2cache.sav_vdevs,
1333 spa->spa_l2cache.sav_count * sizeof (void *));
1334 spa->spa_l2cache.sav_vdevs = NULL;
1336 if (spa->spa_l2cache.sav_config) {
1337 nvlist_free(spa->spa_l2cache.sav_config);
1338 spa->spa_l2cache.sav_config = NULL;
1340 spa->spa_l2cache.sav_count = 0;
1342 spa->spa_async_suspended = 0;
1344 if (spa->spa_comment != NULL) {
1345 spa_strfree(spa->spa_comment);
1346 spa->spa_comment = NULL;
1349 spa_config_exit(spa, SCL_ALL, FTAG);
1353 * Load (or re-load) the current list of vdevs describing the active spares for
1354 * this pool. When this is called, we have some form of basic information in
1355 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1356 * then re-generate a more complete list including status information.
1359 spa_load_spares(spa_t *spa)
1366 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1369 * First, close and free any existing spare vdevs.
1371 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1372 vd = spa->spa_spares.sav_vdevs[i];
1374 /* Undo the call to spa_activate() below */
1375 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1376 B_FALSE)) != NULL && tvd->vdev_isspare)
1377 spa_spare_remove(tvd);
1382 if (spa->spa_spares.sav_vdevs)
1383 kmem_free(spa->spa_spares.sav_vdevs,
1384 spa->spa_spares.sav_count * sizeof (void *));
1386 if (spa->spa_spares.sav_config == NULL)
1389 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1390 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1392 spa->spa_spares.sav_count = (int)nspares;
1393 spa->spa_spares.sav_vdevs = NULL;
1399 * Construct the array of vdevs, opening them to get status in the
1400 * process. For each spare, there is potentially two different vdev_t
1401 * structures associated with it: one in the list of spares (used only
1402 * for basic validation purposes) and one in the active vdev
1403 * configuration (if it's spared in). During this phase we open and
1404 * validate each vdev on the spare list. If the vdev also exists in the
1405 * active configuration, then we also mark this vdev as an active spare.
1407 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1409 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1410 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1411 VDEV_ALLOC_SPARE) == 0);
1414 spa->spa_spares.sav_vdevs[i] = vd;
1416 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1417 B_FALSE)) != NULL) {
1418 if (!tvd->vdev_isspare)
1422 * We only mark the spare active if we were successfully
1423 * able to load the vdev. Otherwise, importing a pool
1424 * with a bad active spare would result in strange
1425 * behavior, because multiple pool would think the spare
1426 * is actively in use.
1428 * There is a vulnerability here to an equally bizarre
1429 * circumstance, where a dead active spare is later
1430 * brought back to life (onlined or otherwise). Given
1431 * the rarity of this scenario, and the extra complexity
1432 * it adds, we ignore the possibility.
1434 if (!vdev_is_dead(tvd))
1435 spa_spare_activate(tvd);
1439 vd->vdev_aux = &spa->spa_spares;
1441 if (vdev_open(vd) != 0)
1444 if (vdev_validate_aux(vd) == 0)
1449 * Recompute the stashed list of spares, with status information
1452 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1453 DATA_TYPE_NVLIST_ARRAY) == 0);
1455 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1457 for (i = 0; i < spa->spa_spares.sav_count; i++)
1458 spares[i] = vdev_config_generate(spa,
1459 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1460 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1461 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1462 for (i = 0; i < spa->spa_spares.sav_count; i++)
1463 nvlist_free(spares[i]);
1464 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1468 * Load (or re-load) the current list of vdevs describing the active l2cache for
1469 * this pool. When this is called, we have some form of basic information in
1470 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1471 * then re-generate a more complete list including status information.
1472 * Devices which are already active have their details maintained, and are
1476 spa_load_l2cache(spa_t *spa)
1480 int i, j, oldnvdevs;
1482 vdev_t *vd, **oldvdevs, **newvdevs;
1483 spa_aux_vdev_t *sav = &spa->spa_l2cache;
1485 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1487 if (sav->sav_config != NULL) {
1488 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1489 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1490 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1496 oldvdevs = sav->sav_vdevs;
1497 oldnvdevs = sav->sav_count;
1498 sav->sav_vdevs = NULL;
1502 * Process new nvlist of vdevs.
1504 for (i = 0; i < nl2cache; i++) {
1505 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1509 for (j = 0; j < oldnvdevs; j++) {
1511 if (vd != NULL && guid == vd->vdev_guid) {
1513 * Retain previous vdev for add/remove ops.
1521 if (newvdevs[i] == NULL) {
1525 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1526 VDEV_ALLOC_L2CACHE) == 0);
1531 * Commit this vdev as an l2cache device,
1532 * even if it fails to open.
1534 spa_l2cache_add(vd);
1539 spa_l2cache_activate(vd);
1541 if (vdev_open(vd) != 0)
1544 (void) vdev_validate_aux(vd);
1546 if (!vdev_is_dead(vd))
1547 l2arc_add_vdev(spa, vd);
1552 * Purge vdevs that were dropped
1554 for (i = 0; i < oldnvdevs; i++) {
1559 ASSERT(vd->vdev_isl2cache);
1561 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1562 pool != 0ULL && l2arc_vdev_present(vd))
1563 l2arc_remove_vdev(vd);
1564 vdev_clear_stats(vd);
1570 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1572 if (sav->sav_config == NULL)
1575 sav->sav_vdevs = newvdevs;
1576 sav->sav_count = (int)nl2cache;
1579 * Recompute the stashed list of l2cache devices, with status
1580 * information this time.
1582 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1583 DATA_TYPE_NVLIST_ARRAY) == 0);
1585 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1586 for (i = 0; i < sav->sav_count; i++)
1587 l2cache[i] = vdev_config_generate(spa,
1588 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1589 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1590 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1592 for (i = 0; i < sav->sav_count; i++)
1593 nvlist_free(l2cache[i]);
1595 kmem_free(l2cache, sav->sav_count * sizeof (void *));
1599 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1602 char *packed = NULL;
1607 error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db);
1610 nvsize = *(uint64_t *)db->db_data;
1611 dmu_buf_rele(db, FTAG);
1613 packed = kmem_alloc(nvsize, KM_SLEEP);
1614 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1617 error = nvlist_unpack(packed, nvsize, value, 0);
1618 kmem_free(packed, nvsize);
1624 * Checks to see if the given vdev could not be opened, in which case we post a
1625 * sysevent to notify the autoreplace code that the device has been removed.
1628 spa_check_removed(vdev_t *vd)
1630 for (int c = 0; c < vd->vdev_children; c++)
1631 spa_check_removed(vd->vdev_child[c]);
1633 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) &&
1635 zfs_post_autoreplace(vd->vdev_spa, vd);
1636 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1641 * Validate the current config against the MOS config
1644 spa_config_valid(spa_t *spa, nvlist_t *config)
1646 vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1649 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1651 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1652 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1654 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1657 * If we're doing a normal import, then build up any additional
1658 * diagnostic information about missing devices in this config.
1659 * We'll pass this up to the user for further processing.
1661 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1662 nvlist_t **child, *nv;
1665 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1667 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1669 for (int c = 0; c < rvd->vdev_children; c++) {
1670 vdev_t *tvd = rvd->vdev_child[c];
1671 vdev_t *mtvd = mrvd->vdev_child[c];
1673 if (tvd->vdev_ops == &vdev_missing_ops &&
1674 mtvd->vdev_ops != &vdev_missing_ops &&
1676 child[idx++] = vdev_config_generate(spa, mtvd,
1681 VERIFY(nvlist_add_nvlist_array(nv,
1682 ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1683 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1684 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1686 for (int i = 0; i < idx; i++)
1687 nvlist_free(child[i]);
1690 kmem_free(child, rvd->vdev_children * sizeof (char **));
1694 * Compare the root vdev tree with the information we have
1695 * from the MOS config (mrvd). Check each top-level vdev
1696 * with the corresponding MOS config top-level (mtvd).
1698 for (int c = 0; c < rvd->vdev_children; c++) {
1699 vdev_t *tvd = rvd->vdev_child[c];
1700 vdev_t *mtvd = mrvd->vdev_child[c];
1703 * Resolve any "missing" vdevs in the current configuration.
1704 * If we find that the MOS config has more accurate information
1705 * about the top-level vdev then use that vdev instead.
1707 if (tvd->vdev_ops == &vdev_missing_ops &&
1708 mtvd->vdev_ops != &vdev_missing_ops) {
1710 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1714 * Device specific actions.
1716 if (mtvd->vdev_islog) {
1717 spa_set_log_state(spa, SPA_LOG_CLEAR);
1720 * XXX - once we have 'readonly' pool
1721 * support we should be able to handle
1722 * missing data devices by transitioning
1723 * the pool to readonly.
1729 * Swap the missing vdev with the data we were
1730 * able to obtain from the MOS config.
1732 vdev_remove_child(rvd, tvd);
1733 vdev_remove_child(mrvd, mtvd);
1735 vdev_add_child(rvd, mtvd);
1736 vdev_add_child(mrvd, tvd);
1738 spa_config_exit(spa, SCL_ALL, FTAG);
1740 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1743 } else if (mtvd->vdev_islog) {
1745 * Load the slog device's state from the MOS config
1746 * since it's possible that the label does not
1747 * contain the most up-to-date information.
1749 vdev_load_log_state(tvd, mtvd);
1754 spa_config_exit(spa, SCL_ALL, FTAG);
1757 * Ensure we were able to validate the config.
1759 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1763 * Check for missing log devices
1766 spa_check_logs(spa_t *spa)
1768 boolean_t rv = B_FALSE;
1770 switch (spa->spa_log_state) {
1771 case SPA_LOG_MISSING:
1772 /* need to recheck in case slog has been restored */
1773 case SPA_LOG_UNKNOWN:
1774 rv = (dmu_objset_find(spa->spa_name, zil_check_log_chain,
1775 NULL, DS_FIND_CHILDREN) != 0);
1777 spa_set_log_state(spa, SPA_LOG_MISSING);
1784 spa_passivate_log(spa_t *spa)
1786 vdev_t *rvd = spa->spa_root_vdev;
1787 boolean_t slog_found = B_FALSE;
1789 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1791 if (!spa_has_slogs(spa))
1794 for (int c = 0; c < rvd->vdev_children; c++) {
1795 vdev_t *tvd = rvd->vdev_child[c];
1796 metaslab_group_t *mg = tvd->vdev_mg;
1798 if (tvd->vdev_islog) {
1799 metaslab_group_passivate(mg);
1800 slog_found = B_TRUE;
1804 return (slog_found);
1808 spa_activate_log(spa_t *spa)
1810 vdev_t *rvd = spa->spa_root_vdev;
1812 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1814 for (int c = 0; c < rvd->vdev_children; c++) {
1815 vdev_t *tvd = rvd->vdev_child[c];
1816 metaslab_group_t *mg = tvd->vdev_mg;
1818 if (tvd->vdev_islog)
1819 metaslab_group_activate(mg);
1824 spa_offline_log(spa_t *spa)
1828 error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1829 NULL, DS_FIND_CHILDREN);
1832 * We successfully offlined the log device, sync out the
1833 * current txg so that the "stubby" block can be removed
1836 txg_wait_synced(spa->spa_dsl_pool, 0);
1842 spa_aux_check_removed(spa_aux_vdev_t *sav)
1846 for (i = 0; i < sav->sav_count; i++)
1847 spa_check_removed(sav->sav_vdevs[i]);
1851 spa_claim_notify(zio_t *zio)
1853 spa_t *spa = zio->io_spa;
1858 mutex_enter(&spa->spa_props_lock); /* any mutex will do */
1859 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1860 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1861 mutex_exit(&spa->spa_props_lock);
1864 typedef struct spa_load_error {
1865 uint64_t sle_meta_count;
1866 uint64_t sle_data_count;
1870 spa_load_verify_done(zio_t *zio)
1872 blkptr_t *bp = zio->io_bp;
1873 spa_load_error_t *sle = zio->io_private;
1874 dmu_object_type_t type = BP_GET_TYPE(bp);
1875 int error = zio->io_error;
1878 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1879 type != DMU_OT_INTENT_LOG)
1880 atomic_add_64(&sle->sle_meta_count, 1);
1882 atomic_add_64(&sle->sle_data_count, 1);
1884 zio_data_buf_free(zio->io_data, zio->io_size);
1889 spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1890 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
1892 if (!BP_IS_HOLE(bp) && !BP_IS_EMBEDDED(bp)) {
1894 size_t size = BP_GET_PSIZE(bp);
1895 void *data = zio_data_buf_alloc(size);
1897 zio_nowait(zio_read(rio, spa, bp, data, size,
1898 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1899 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1900 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1906 spa_load_verify(spa_t *spa)
1909 spa_load_error_t sle = { 0 };
1910 zpool_rewind_policy_t policy;
1911 boolean_t verify_ok = B_FALSE;
1914 zpool_get_rewind_policy(spa->spa_config, &policy);
1916 if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1919 rio = zio_root(spa, NULL, &sle,
1920 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1922 error = traverse_pool(spa, spa->spa_verify_min_txg,
1923 TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio);
1925 (void) zio_wait(rio);
1927 spa->spa_load_meta_errors = sle.sle_meta_count;
1928 spa->spa_load_data_errors = sle.sle_data_count;
1930 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1931 sle.sle_data_count <= policy.zrp_maxdata) {
1935 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1936 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1938 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1939 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1940 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1941 VERIFY(nvlist_add_int64(spa->spa_load_info,
1942 ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1943 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1944 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1946 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1950 if (error != ENXIO && error != EIO)
1951 error = SET_ERROR(EIO);
1955 return (verify_ok ? 0 : EIO);
1959 * Find a value in the pool props object.
1962 spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1964 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1965 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1969 * Find a value in the pool directory object.
1972 spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1974 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1975 name, sizeof (uint64_t), 1, val));
1979 spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1981 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1986 * Fix up config after a partly-completed split. This is done with the
1987 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
1988 * pool have that entry in their config, but only the splitting one contains
1989 * a list of all the guids of the vdevs that are being split off.
1991 * This function determines what to do with that list: either rejoin
1992 * all the disks to the pool, or complete the splitting process. To attempt
1993 * the rejoin, each disk that is offlined is marked online again, and
1994 * we do a reopen() call. If the vdev label for every disk that was
1995 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1996 * then we call vdev_split() on each disk, and complete the split.
1998 * Otherwise we leave the config alone, with all the vdevs in place in
1999 * the original pool.
2002 spa_try_repair(spa_t *spa, nvlist_t *config)
2009 boolean_t attempt_reopen;
2011 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
2014 /* check that the config is complete */
2015 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
2016 &glist, &gcount) != 0)
2019 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
2021 /* attempt to online all the vdevs & validate */
2022 attempt_reopen = B_TRUE;
2023 for (i = 0; i < gcount; i++) {
2024 if (glist[i] == 0) /* vdev is hole */
2027 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
2028 if (vd[i] == NULL) {
2030 * Don't bother attempting to reopen the disks;
2031 * just do the split.
2033 attempt_reopen = B_FALSE;
2035 /* attempt to re-online it */
2036 vd[i]->vdev_offline = B_FALSE;
2040 if (attempt_reopen) {
2041 vdev_reopen(spa->spa_root_vdev);
2043 /* check each device to see what state it's in */
2044 for (extracted = 0, i = 0; i < gcount; i++) {
2045 if (vd[i] != NULL &&
2046 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
2053 * If every disk has been moved to the new pool, or if we never
2054 * even attempted to look at them, then we split them off for
2057 if (!attempt_reopen || gcount == extracted) {
2058 for (i = 0; i < gcount; i++)
2061 vdev_reopen(spa->spa_root_vdev);
2064 kmem_free(vd, gcount * sizeof (vdev_t *));
2068 spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
2069 boolean_t mosconfig)
2071 nvlist_t *config = spa->spa_config;
2072 char *ereport = FM_EREPORT_ZFS_POOL;
2078 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
2079 return (SET_ERROR(EINVAL));
2081 ASSERT(spa->spa_comment == NULL);
2082 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
2083 spa->spa_comment = spa_strdup(comment);
2086 * Versioning wasn't explicitly added to the label until later, so if
2087 * it's not present treat it as the initial version.
2089 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
2090 &spa->spa_ubsync.ub_version) != 0)
2091 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
2093 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
2094 &spa->spa_config_txg);
2096 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
2097 spa_guid_exists(pool_guid, 0)) {
2098 error = SET_ERROR(EEXIST);
2100 spa->spa_config_guid = pool_guid;
2102 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
2104 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
2108 nvlist_free(spa->spa_load_info);
2109 spa->spa_load_info = fnvlist_alloc();
2111 gethrestime(&spa->spa_loaded_ts);
2112 error = spa_load_impl(spa, pool_guid, config, state, type,
2113 mosconfig, &ereport);
2116 spa->spa_minref = refcount_count(&spa->spa_refcount);
2118 if (error != EEXIST) {
2119 spa->spa_loaded_ts.tv_sec = 0;
2120 spa->spa_loaded_ts.tv_nsec = 0;
2122 if (error != EBADF) {
2123 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
2126 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2133 * Load an existing storage pool, using the pool's builtin spa_config as a
2134 * source of configuration information.
2137 spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
2138 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
2142 nvlist_t *nvroot = NULL;
2145 uberblock_t *ub = &spa->spa_uberblock;
2146 uint64_t children, config_cache_txg = spa->spa_config_txg;
2147 int orig_mode = spa->spa_mode;
2150 boolean_t missing_feat_write = B_FALSE;
2153 * If this is an untrusted config, access the pool in read-only mode.
2154 * This prevents things like resilvering recently removed devices.
2157 spa->spa_mode = FREAD;
2159 ASSERT(MUTEX_HELD(&spa_namespace_lock));
2161 spa->spa_load_state = state;
2163 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
2164 return (SET_ERROR(EINVAL));
2166 parse = (type == SPA_IMPORT_EXISTING ?
2167 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2170 * Create "The Godfather" zio to hold all async IOs
2172 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2173 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2176 * Parse the configuration into a vdev tree. We explicitly set the
2177 * value that will be returned by spa_version() since parsing the
2178 * configuration requires knowing the version number.
2180 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2181 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
2182 spa_config_exit(spa, SCL_ALL, FTAG);
2187 ASSERT(spa->spa_root_vdev == rvd);
2189 if (type != SPA_IMPORT_ASSEMBLE) {
2190 ASSERT(spa_guid(spa) == pool_guid);
2194 * Try to open all vdevs, loading each label in the process.
2196 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2197 error = vdev_open(rvd);
2198 spa_config_exit(spa, SCL_ALL, FTAG);
2203 * We need to validate the vdev labels against the configuration that
2204 * we have in hand, which is dependent on the setting of mosconfig. If
2205 * mosconfig is true then we're validating the vdev labels based on
2206 * that config. Otherwise, we're validating against the cached config
2207 * (zpool.cache) that was read when we loaded the zfs module, and then
2208 * later we will recursively call spa_load() and validate against
2211 * If we're assembling a new pool that's been split off from an
2212 * existing pool, the labels haven't yet been updated so we skip
2213 * validation for now.
2215 if (type != SPA_IMPORT_ASSEMBLE) {
2216 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2217 error = vdev_validate(rvd, mosconfig);
2218 spa_config_exit(spa, SCL_ALL, FTAG);
2223 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2224 return (SET_ERROR(ENXIO));
2228 * Find the best uberblock.
2230 vdev_uberblock_load(rvd, ub, &label);
2233 * If we weren't able to find a single valid uberblock, return failure.
2235 if (ub->ub_txg == 0) {
2237 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2241 * If the pool has an unsupported version we can't open it.
2243 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2245 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2248 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2252 * If we weren't able to find what's necessary for reading the
2253 * MOS in the label, return failure.
2255 if (label == NULL || nvlist_lookup_nvlist(label,
2256 ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) {
2258 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2263 * Update our in-core representation with the definitive values
2266 nvlist_free(spa->spa_label_features);
2267 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2273 * Look through entries in the label nvlist's features_for_read. If
2274 * there is a feature listed there which we don't understand then we
2275 * cannot open a pool.
2277 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2278 nvlist_t *unsup_feat;
2280 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2283 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2285 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2286 if (!zfeature_is_supported(nvpair_name(nvp))) {
2287 VERIFY(nvlist_add_string(unsup_feat,
2288 nvpair_name(nvp), "") == 0);
2292 if (!nvlist_empty(unsup_feat)) {
2293 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2294 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2295 nvlist_free(unsup_feat);
2296 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2300 nvlist_free(unsup_feat);
2304 * If the vdev guid sum doesn't match the uberblock, we have an
2305 * incomplete configuration. We first check to see if the pool
2306 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2307 * If it is, defer the vdev_guid_sum check till later so we
2308 * can handle missing vdevs.
2310 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
2311 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
2312 rvd->vdev_guid_sum != ub->ub_guid_sum)
2313 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2315 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2316 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2317 spa_try_repair(spa, config);
2318 spa_config_exit(spa, SCL_ALL, FTAG);
2319 nvlist_free(spa->spa_config_splitting);
2320 spa->spa_config_splitting = NULL;
2324 * Initialize internal SPA structures.
2326 spa->spa_state = POOL_STATE_ACTIVE;
2327 spa->spa_ubsync = spa->spa_uberblock;
2328 spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2329 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2330 spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2331 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2332 spa->spa_claim_max_txg = spa->spa_first_txg;
2333 spa->spa_prev_software_version = ub->ub_software_version;
2335 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2337 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2338 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2340 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2341 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2343 if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2344 boolean_t missing_feat_read = B_FALSE;
2345 nvlist_t *unsup_feat, *enabled_feat;
2347 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2348 &spa->spa_feat_for_read_obj) != 0) {
2349 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2352 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2353 &spa->spa_feat_for_write_obj) != 0) {
2354 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2357 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2358 &spa->spa_feat_desc_obj) != 0) {
2359 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2362 enabled_feat = fnvlist_alloc();
2363 unsup_feat = fnvlist_alloc();
2365 if (!spa_features_check(spa, B_FALSE,
2366 unsup_feat, enabled_feat))
2367 missing_feat_read = B_TRUE;
2369 if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) {
2370 if (!spa_features_check(spa, B_TRUE,
2371 unsup_feat, enabled_feat)) {
2372 missing_feat_write = B_TRUE;
2376 fnvlist_add_nvlist(spa->spa_load_info,
2377 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
2379 if (!nvlist_empty(unsup_feat)) {
2380 fnvlist_add_nvlist(spa->spa_load_info,
2381 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
2384 fnvlist_free(enabled_feat);
2385 fnvlist_free(unsup_feat);
2387 if (!missing_feat_read) {
2388 fnvlist_add_boolean(spa->spa_load_info,
2389 ZPOOL_CONFIG_CAN_RDONLY);
2393 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2394 * twofold: to determine whether the pool is available for
2395 * import in read-write mode and (if it is not) whether the
2396 * pool is available for import in read-only mode. If the pool
2397 * is available for import in read-write mode, it is displayed
2398 * as available in userland; if it is not available for import
2399 * in read-only mode, it is displayed as unavailable in
2400 * userland. If the pool is available for import in read-only
2401 * mode but not read-write mode, it is displayed as unavailable
2402 * in userland with a special note that the pool is actually
2403 * available for open in read-only mode.
2405 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2406 * missing a feature for write, we must first determine whether
2407 * the pool can be opened read-only before returning to
2408 * userland in order to know whether to display the
2409 * abovementioned note.
2411 if (missing_feat_read || (missing_feat_write &&
2412 spa_writeable(spa))) {
2413 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2418 * Load refcounts for ZFS features from disk into an in-memory
2419 * cache during SPA initialization.
2421 for (spa_feature_t i = 0; i < SPA_FEATURES; i++) {
2424 error = feature_get_refcount_from_disk(spa,
2425 &spa_feature_table[i], &refcount);
2427 spa->spa_feat_refcount_cache[i] = refcount;
2428 } else if (error == ENOTSUP) {
2429 spa->spa_feat_refcount_cache[i] =
2430 SPA_FEATURE_DISABLED;
2432 return (spa_vdev_err(rvd,
2433 VDEV_AUX_CORRUPT_DATA, EIO));
2438 if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) {
2439 if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG,
2440 &spa->spa_feat_enabled_txg_obj) != 0)
2441 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2444 spa->spa_is_initializing = B_TRUE;
2445 error = dsl_pool_open(spa->spa_dsl_pool);
2446 spa->spa_is_initializing = B_FALSE;
2448 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2452 nvlist_t *policy = NULL, *nvconfig;
2454 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2455 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2457 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2458 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2460 unsigned long myhostid = 0;
2462 VERIFY(nvlist_lookup_string(nvconfig,
2463 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2466 myhostid = zone_get_hostid(NULL);
2469 * We're emulating the system's hostid in userland, so
2470 * we can't use zone_get_hostid().
2472 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2473 #endif /* _KERNEL */
2474 if (check_hostid && hostid != 0 && myhostid != 0 &&
2475 hostid != myhostid) {
2476 nvlist_free(nvconfig);
2477 cmn_err(CE_WARN, "pool '%s' could not be "
2478 "loaded as it was last accessed by "
2479 "another system (host: %s hostid: 0x%lx). "
2480 "See: http://illumos.org/msg/ZFS-8000-EY",
2481 spa_name(spa), hostname,
2482 (unsigned long)hostid);
2483 return (SET_ERROR(EBADF));
2486 if (nvlist_lookup_nvlist(spa->spa_config,
2487 ZPOOL_REWIND_POLICY, &policy) == 0)
2488 VERIFY(nvlist_add_nvlist(nvconfig,
2489 ZPOOL_REWIND_POLICY, policy) == 0);
2491 spa_config_set(spa, nvconfig);
2493 spa_deactivate(spa);
2494 spa_activate(spa, orig_mode);
2496 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2499 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2500 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2501 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2503 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2506 * Load the bit that tells us to use the new accounting function
2507 * (raid-z deflation). If we have an older pool, this will not
2510 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2511 if (error != 0 && error != ENOENT)
2512 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2514 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2515 &spa->spa_creation_version);
2516 if (error != 0 && error != ENOENT)
2517 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2520 * Load the persistent error log. If we have an older pool, this will
2523 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2524 if (error != 0 && error != ENOENT)
2525 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2527 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2528 &spa->spa_errlog_scrub);
2529 if (error != 0 && error != ENOENT)
2530 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2533 * Load the history object. If we have an older pool, this
2534 * will not be present.
2536 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2537 if (error != 0 && error != ENOENT)
2538 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2541 * If we're assembling the pool from the split-off vdevs of
2542 * an existing pool, we don't want to attach the spares & cache
2547 * Load any hot spares for this pool.
2549 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2550 if (error != 0 && error != ENOENT)
2551 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2552 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2553 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2554 if (load_nvlist(spa, spa->spa_spares.sav_object,
2555 &spa->spa_spares.sav_config) != 0)
2556 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2558 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2559 spa_load_spares(spa);
2560 spa_config_exit(spa, SCL_ALL, FTAG);
2561 } else if (error == 0) {
2562 spa->spa_spares.sav_sync = B_TRUE;
2566 * Load any level 2 ARC devices for this pool.
2568 error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2569 &spa->spa_l2cache.sav_object);
2570 if (error != 0 && error != ENOENT)
2571 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2572 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2573 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2574 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2575 &spa->spa_l2cache.sav_config) != 0)
2576 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2578 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2579 spa_load_l2cache(spa);
2580 spa_config_exit(spa, SCL_ALL, FTAG);
2581 } else if (error == 0) {
2582 spa->spa_l2cache.sav_sync = B_TRUE;
2585 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2587 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2588 if (error && error != ENOENT)
2589 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2592 uint64_t autoreplace;
2594 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2595 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2596 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2597 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2598 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2599 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2600 &spa->spa_dedup_ditto);
2602 spa->spa_autoreplace = (autoreplace != 0);
2606 * If the 'autoreplace' property is set, then post a resource notifying
2607 * the ZFS DE that it should not issue any faults for unopenable
2608 * devices. We also iterate over the vdevs, and post a sysevent for any
2609 * unopenable vdevs so that the normal autoreplace handler can take
2612 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2613 spa_check_removed(spa->spa_root_vdev);
2615 * For the import case, this is done in spa_import(), because
2616 * at this point we're using the spare definitions from
2617 * the MOS config, not necessarily from the userland config.
2619 if (state != SPA_LOAD_IMPORT) {
2620 spa_aux_check_removed(&spa->spa_spares);
2621 spa_aux_check_removed(&spa->spa_l2cache);
2626 * Load the vdev state for all toplevel vdevs.
2631 * Propagate the leaf DTLs we just loaded all the way up the tree.
2633 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2634 vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2635 spa_config_exit(spa, SCL_ALL, FTAG);
2638 * Load the DDTs (dedup tables).
2640 error = ddt_load(spa);
2642 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2644 spa_update_dspace(spa);
2647 * Validate the config, using the MOS config to fill in any
2648 * information which might be missing. If we fail to validate
2649 * the config then declare the pool unfit for use. If we're
2650 * assembling a pool from a split, the log is not transferred
2653 if (type != SPA_IMPORT_ASSEMBLE) {
2656 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2657 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2659 if (!spa_config_valid(spa, nvconfig)) {
2660 nvlist_free(nvconfig);
2661 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2664 nvlist_free(nvconfig);
2667 * Now that we've validated the config, check the state of the
2668 * root vdev. If it can't be opened, it indicates one or
2669 * more toplevel vdevs are faulted.
2671 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2672 return (SET_ERROR(ENXIO));
2674 if (spa_check_logs(spa)) {
2675 *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2676 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2680 if (missing_feat_write) {
2681 ASSERT(state == SPA_LOAD_TRYIMPORT);
2684 * At this point, we know that we can open the pool in
2685 * read-only mode but not read-write mode. We now have enough
2686 * information and can return to userland.
2688 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
2692 * We've successfully opened the pool, verify that we're ready
2693 * to start pushing transactions.
2695 if (state != SPA_LOAD_TRYIMPORT) {
2696 if (error = spa_load_verify(spa))
2697 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2701 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2702 spa->spa_load_max_txg == UINT64_MAX)) {
2704 int need_update = B_FALSE;
2706 ASSERT(state != SPA_LOAD_TRYIMPORT);
2709 * Claim log blocks that haven't been committed yet.
2710 * This must all happen in a single txg.
2711 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2712 * invoked from zil_claim_log_block()'s i/o done callback.
2713 * Price of rollback is that we abandon the log.
2715 spa->spa_claiming = B_TRUE;
2717 tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2718 spa_first_txg(spa));
2719 (void) dmu_objset_find(spa_name(spa),
2720 zil_claim, tx, DS_FIND_CHILDREN);
2723 spa->spa_claiming = B_FALSE;
2725 spa_set_log_state(spa, SPA_LOG_GOOD);
2726 spa->spa_sync_on = B_TRUE;
2727 txg_sync_start(spa->spa_dsl_pool);
2730 * Wait for all claims to sync. We sync up to the highest
2731 * claimed log block birth time so that claimed log blocks
2732 * don't appear to be from the future. spa_claim_max_txg
2733 * will have been set for us by either zil_check_log_chain()
2734 * (invoked from spa_check_logs()) or zil_claim() above.
2736 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2739 * If the config cache is stale, or we have uninitialized
2740 * metaslabs (see spa_vdev_add()), then update the config.
2742 * If this is a verbatim import, trust the current
2743 * in-core spa_config and update the disk labels.
2745 if (config_cache_txg != spa->spa_config_txg ||
2746 state == SPA_LOAD_IMPORT ||
2747 state == SPA_LOAD_RECOVER ||
2748 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2749 need_update = B_TRUE;
2751 for (int c = 0; c < rvd->vdev_children; c++)
2752 if (rvd->vdev_child[c]->vdev_ms_array == 0)
2753 need_update = B_TRUE;
2756 * Update the config cache asychronously in case we're the
2757 * root pool, in which case the config cache isn't writable yet.
2760 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2763 * Check all DTLs to see if anything needs resilvering.
2765 if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2766 vdev_resilver_needed(rvd, NULL, NULL))
2767 spa_async_request(spa, SPA_ASYNC_RESILVER);
2770 * Log the fact that we booted up (so that we can detect if
2771 * we rebooted in the middle of an operation).
2773 spa_history_log_version(spa, "open");
2776 * Delete any inconsistent datasets.
2778 (void) dmu_objset_find(spa_name(spa),
2779 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2782 * Clean up any stale temporary dataset userrefs.
2784 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2791 spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2793 int mode = spa->spa_mode;
2796 spa_deactivate(spa);
2798 spa->spa_load_max_txg--;
2800 spa_activate(spa, mode);
2801 spa_async_suspend(spa);
2803 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2807 * If spa_load() fails this function will try loading prior txg's. If
2808 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2809 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2810 * function will not rewind the pool and will return the same error as
2814 spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2815 uint64_t max_request, int rewind_flags)
2817 nvlist_t *loadinfo = NULL;
2818 nvlist_t *config = NULL;
2819 int load_error, rewind_error;
2820 uint64_t safe_rewind_txg;
2823 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2824 spa->spa_load_max_txg = spa->spa_load_txg;
2825 spa_set_log_state(spa, SPA_LOG_CLEAR);
2827 spa->spa_load_max_txg = max_request;
2830 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2832 if (load_error == 0)
2835 if (spa->spa_root_vdev != NULL)
2836 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2838 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2839 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2841 if (rewind_flags & ZPOOL_NEVER_REWIND) {
2842 nvlist_free(config);
2843 return (load_error);
2846 if (state == SPA_LOAD_RECOVER) {
2847 /* Price of rolling back is discarding txgs, including log */
2848 spa_set_log_state(spa, SPA_LOG_CLEAR);
2851 * If we aren't rolling back save the load info from our first
2852 * import attempt so that we can restore it after attempting
2855 loadinfo = spa->spa_load_info;
2856 spa->spa_load_info = fnvlist_alloc();
2859 spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2860 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2861 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2862 TXG_INITIAL : safe_rewind_txg;
2865 * Continue as long as we're finding errors, we're still within
2866 * the acceptable rewind range, and we're still finding uberblocks
2868 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2869 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2870 if (spa->spa_load_max_txg < safe_rewind_txg)
2871 spa->spa_extreme_rewind = B_TRUE;
2872 rewind_error = spa_load_retry(spa, state, mosconfig);
2875 spa->spa_extreme_rewind = B_FALSE;
2876 spa->spa_load_max_txg = UINT64_MAX;
2878 if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2879 spa_config_set(spa, config);
2881 if (state == SPA_LOAD_RECOVER) {
2882 ASSERT3P(loadinfo, ==, NULL);
2883 return (rewind_error);
2885 /* Store the rewind info as part of the initial load info */
2886 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
2887 spa->spa_load_info);
2889 /* Restore the initial load info */
2890 fnvlist_free(spa->spa_load_info);
2891 spa->spa_load_info = loadinfo;
2893 return (load_error);
2900 * The import case is identical to an open except that the configuration is sent
2901 * down from userland, instead of grabbed from the configuration cache. For the
2902 * case of an open, the pool configuration will exist in the
2903 * POOL_STATE_UNINITIALIZED state.
2905 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2906 * the same time open the pool, without having to keep around the spa_t in some
2910 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2914 spa_load_state_t state = SPA_LOAD_OPEN;
2916 int locked = B_FALSE;
2917 int firstopen = B_FALSE;
2922 * As disgusting as this is, we need to support recursive calls to this
2923 * function because dsl_dir_open() is called during spa_load(), and ends
2924 * up calling spa_open() again. The real fix is to figure out how to
2925 * avoid dsl_dir_open() calling this in the first place.
2927 if (mutex_owner(&spa_namespace_lock) != curthread) {
2928 mutex_enter(&spa_namespace_lock);
2932 if ((spa = spa_lookup(pool)) == NULL) {
2934 mutex_exit(&spa_namespace_lock);
2935 return (SET_ERROR(ENOENT));
2938 if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2939 zpool_rewind_policy_t policy;
2943 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2945 if (policy.zrp_request & ZPOOL_DO_REWIND)
2946 state = SPA_LOAD_RECOVER;
2948 spa_activate(spa, spa_mode_global);
2950 if (state != SPA_LOAD_RECOVER)
2951 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2953 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2954 policy.zrp_request);
2956 if (error == EBADF) {
2958 * If vdev_validate() returns failure (indicated by
2959 * EBADF), it indicates that one of the vdevs indicates
2960 * that the pool has been exported or destroyed. If
2961 * this is the case, the config cache is out of sync and
2962 * we should remove the pool from the namespace.
2965 spa_deactivate(spa);
2966 spa_config_sync(spa, B_TRUE, B_TRUE);
2969 mutex_exit(&spa_namespace_lock);
2970 return (SET_ERROR(ENOENT));
2975 * We can't open the pool, but we still have useful
2976 * information: the state of each vdev after the
2977 * attempted vdev_open(). Return this to the user.
2979 if (config != NULL && spa->spa_config) {
2980 VERIFY(nvlist_dup(spa->spa_config, config,
2982 VERIFY(nvlist_add_nvlist(*config,
2983 ZPOOL_CONFIG_LOAD_INFO,
2984 spa->spa_load_info) == 0);
2987 spa_deactivate(spa);
2988 spa->spa_last_open_failed = error;
2990 mutex_exit(&spa_namespace_lock);
2996 spa_open_ref(spa, tag);
2999 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
3002 * If we've recovered the pool, pass back any information we
3003 * gathered while doing the load.
3005 if (state == SPA_LOAD_RECOVER) {
3006 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
3007 spa->spa_load_info) == 0);
3011 spa->spa_last_open_failed = 0;
3012 spa->spa_last_ubsync_txg = 0;
3013 spa->spa_load_txg = 0;
3014 mutex_exit(&spa_namespace_lock);
3018 zvol_create_minors(spa->spa_name);
3029 spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
3032 return (spa_open_common(name, spapp, tag, policy, config));
3036 spa_open(const char *name, spa_t **spapp, void *tag)
3038 return (spa_open_common(name, spapp, tag, NULL, NULL));
3042 * Lookup the given spa_t, incrementing the inject count in the process,
3043 * preventing it from being exported or destroyed.
3046 spa_inject_addref(char *name)
3050 mutex_enter(&spa_namespace_lock);
3051 if ((spa = spa_lookup(name)) == NULL) {
3052 mutex_exit(&spa_namespace_lock);
3055 spa->spa_inject_ref++;
3056 mutex_exit(&spa_namespace_lock);
3062 spa_inject_delref(spa_t *spa)
3064 mutex_enter(&spa_namespace_lock);
3065 spa->spa_inject_ref--;
3066 mutex_exit(&spa_namespace_lock);
3070 * Add spares device information to the nvlist.
3073 spa_add_spares(spa_t *spa, nvlist_t *config)
3083 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3085 if (spa->spa_spares.sav_count == 0)
3088 VERIFY(nvlist_lookup_nvlist(config,
3089 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3090 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
3091 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3093 VERIFY(nvlist_add_nvlist_array(nvroot,
3094 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3095 VERIFY(nvlist_lookup_nvlist_array(nvroot,
3096 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3099 * Go through and find any spares which have since been
3100 * repurposed as an active spare. If this is the case, update
3101 * their status appropriately.
3103 for (i = 0; i < nspares; i++) {
3104 VERIFY(nvlist_lookup_uint64(spares[i],
3105 ZPOOL_CONFIG_GUID, &guid) == 0);
3106 if (spa_spare_exists(guid, &pool, NULL) &&
3108 VERIFY(nvlist_lookup_uint64_array(
3109 spares[i], ZPOOL_CONFIG_VDEV_STATS,
3110 (uint64_t **)&vs, &vsc) == 0);
3111 vs->vs_state = VDEV_STATE_CANT_OPEN;
3112 vs->vs_aux = VDEV_AUX_SPARED;
3119 * Add l2cache device information to the nvlist, including vdev stats.
3122 spa_add_l2cache(spa_t *spa, nvlist_t *config)
3125 uint_t i, j, nl2cache;
3132 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3134 if (spa->spa_l2cache.sav_count == 0)
3137 VERIFY(nvlist_lookup_nvlist(config,
3138 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3139 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
3140 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3141 if (nl2cache != 0) {
3142 VERIFY(nvlist_add_nvlist_array(nvroot,
3143 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3144 VERIFY(nvlist_lookup_nvlist_array(nvroot,
3145 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3148 * Update level 2 cache device stats.
3151 for (i = 0; i < nl2cache; i++) {
3152 VERIFY(nvlist_lookup_uint64(l2cache[i],
3153 ZPOOL_CONFIG_GUID, &guid) == 0);
3156 for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
3158 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
3159 vd = spa->spa_l2cache.sav_vdevs[j];
3165 VERIFY(nvlist_lookup_uint64_array(l2cache[i],
3166 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
3168 vdev_get_stats(vd, vs);
3174 spa_add_feature_stats(spa_t *spa, nvlist_t *config)
3180 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3181 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3183 /* We may be unable to read features if pool is suspended. */
3184 if (spa_suspended(spa))
3187 if (spa->spa_feat_for_read_obj != 0) {
3188 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3189 spa->spa_feat_for_read_obj);
3190 zap_cursor_retrieve(&zc, &za) == 0;
3191 zap_cursor_advance(&zc)) {
3192 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3193 za.za_num_integers == 1);
3194 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3195 za.za_first_integer));
3197 zap_cursor_fini(&zc);
3200 if (spa->spa_feat_for_write_obj != 0) {
3201 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3202 spa->spa_feat_for_write_obj);
3203 zap_cursor_retrieve(&zc, &za) == 0;
3204 zap_cursor_advance(&zc)) {
3205 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3206 za.za_num_integers == 1);
3207 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3208 za.za_first_integer));
3210 zap_cursor_fini(&zc);
3214 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
3216 nvlist_free(features);
3220 spa_get_stats(const char *name, nvlist_t **config,
3221 char *altroot, size_t buflen)
3227 error = spa_open_common(name, &spa, FTAG, NULL, config);
3231 * This still leaves a window of inconsistency where the spares
3232 * or l2cache devices could change and the config would be
3233 * self-inconsistent.
3235 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3237 if (*config != NULL) {
3238 uint64_t loadtimes[2];
3240 loadtimes[0] = spa->spa_loaded_ts.tv_sec;
3241 loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
3242 VERIFY(nvlist_add_uint64_array(*config,
3243 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
3245 VERIFY(nvlist_add_uint64(*config,
3246 ZPOOL_CONFIG_ERRCOUNT,
3247 spa_get_errlog_size(spa)) == 0);
3249 if (spa_suspended(spa))
3250 VERIFY(nvlist_add_uint64(*config,
3251 ZPOOL_CONFIG_SUSPENDED,
3252 spa->spa_failmode) == 0);
3254 spa_add_spares(spa, *config);
3255 spa_add_l2cache(spa, *config);
3256 spa_add_feature_stats(spa, *config);
3261 * We want to get the alternate root even for faulted pools, so we cheat
3262 * and call spa_lookup() directly.
3266 mutex_enter(&spa_namespace_lock);
3267 spa = spa_lookup(name);
3269 spa_altroot(spa, altroot, buflen);
3273 mutex_exit(&spa_namespace_lock);
3275 spa_altroot(spa, altroot, buflen);
3280 spa_config_exit(spa, SCL_CONFIG, FTAG);
3281 spa_close(spa, FTAG);
3288 * Validate that the auxiliary device array is well formed. We must have an
3289 * array of nvlists, each which describes a valid leaf vdev. If this is an
3290 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3291 * specified, as long as they are well-formed.
3294 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
3295 spa_aux_vdev_t *sav, const char *config, uint64_t version,
3296 vdev_labeltype_t label)
3303 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3306 * It's acceptable to have no devs specified.
3308 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
3312 return (SET_ERROR(EINVAL));
3315 * Make sure the pool is formatted with a version that supports this
3318 if (spa_version(spa) < version)
3319 return (SET_ERROR(ENOTSUP));
3322 * Set the pending device list so we correctly handle device in-use
3325 sav->sav_pending = dev;
3326 sav->sav_npending = ndev;
3328 for (i = 0; i < ndev; i++) {
3329 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
3333 if (!vd->vdev_ops->vdev_op_leaf) {
3335 error = SET_ERROR(EINVAL);
3340 * The L2ARC currently only supports disk devices in
3341 * kernel context. For user-level testing, we allow it.
3344 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
3345 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
3346 error = SET_ERROR(ENOTBLK);
3353 if ((error = vdev_open(vd)) == 0 &&
3354 (error = vdev_label_init(vd, crtxg, label)) == 0) {
3355 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
3356 vd->vdev_guid) == 0);
3362 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
3369 sav->sav_pending = NULL;
3370 sav->sav_npending = 0;
3375 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
3379 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3381 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3382 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
3383 VDEV_LABEL_SPARE)) != 0) {
3387 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3388 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
3389 VDEV_LABEL_L2CACHE));
3393 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
3398 if (sav->sav_config != NULL) {
3404 * Generate new dev list by concatentating with the
3407 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
3408 &olddevs, &oldndevs) == 0);
3410 newdevs = kmem_alloc(sizeof (void *) *
3411 (ndevs + oldndevs), KM_SLEEP);
3412 for (i = 0; i < oldndevs; i++)
3413 VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
3415 for (i = 0; i < ndevs; i++)
3416 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
3419 VERIFY(nvlist_remove(sav->sav_config, config,
3420 DATA_TYPE_NVLIST_ARRAY) == 0);
3422 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3423 config, newdevs, ndevs + oldndevs) == 0);
3424 for (i = 0; i < oldndevs + ndevs; i++)
3425 nvlist_free(newdevs[i]);
3426 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
3429 * Generate a new dev list.
3431 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
3433 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
3439 * Stop and drop level 2 ARC devices
3442 spa_l2cache_drop(spa_t *spa)
3446 spa_aux_vdev_t *sav = &spa->spa_l2cache;
3448 for (i = 0; i < sav->sav_count; i++) {
3451 vd = sav->sav_vdevs[i];
3454 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
3455 pool != 0ULL && l2arc_vdev_present(vd))
3456 l2arc_remove_vdev(vd);
3464 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
3468 char *altroot = NULL;
3473 uint64_t txg = TXG_INITIAL;
3474 nvlist_t **spares, **l2cache;
3475 uint_t nspares, nl2cache;
3476 uint64_t version, obj;
3477 boolean_t has_features;
3480 * If this pool already exists, return failure.
3482 mutex_enter(&spa_namespace_lock);
3483 if (spa_lookup(pool) != NULL) {
3484 mutex_exit(&spa_namespace_lock);
3485 return (SET_ERROR(EEXIST));
3489 * Allocate a new spa_t structure.
3491 (void) nvlist_lookup_string(props,
3492 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3493 spa = spa_add(pool, NULL, altroot);
3494 spa_activate(spa, spa_mode_global);
3496 if (props && (error = spa_prop_validate(spa, props))) {
3497 spa_deactivate(spa);
3499 mutex_exit(&spa_namespace_lock);
3503 has_features = B_FALSE;
3504 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
3505 elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
3506 if (zpool_prop_feature(nvpair_name(elem)))
3507 has_features = B_TRUE;
3510 if (has_features || nvlist_lookup_uint64(props,
3511 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
3512 version = SPA_VERSION;
3514 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
3516 spa->spa_first_txg = txg;
3517 spa->spa_uberblock.ub_txg = txg - 1;
3518 spa->spa_uberblock.ub_version = version;
3519 spa->spa_ubsync = spa->spa_uberblock;
3522 * Create "The Godfather" zio to hold all async IOs
3524 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
3525 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
3528 * Create the root vdev.
3530 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3532 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3534 ASSERT(error != 0 || rvd != NULL);
3535 ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3537 if (error == 0 && !zfs_allocatable_devs(nvroot))
3538 error = SET_ERROR(EINVAL);
3541 (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3542 (error = spa_validate_aux(spa, nvroot, txg,
3543 VDEV_ALLOC_ADD)) == 0) {
3544 for (int c = 0; c < rvd->vdev_children; c++) {
3545 vdev_ashift_optimize(rvd->vdev_child[c]);
3546 vdev_metaslab_set_size(rvd->vdev_child[c]);
3547 vdev_expand(rvd->vdev_child[c], txg);
3551 spa_config_exit(spa, SCL_ALL, FTAG);
3555 spa_deactivate(spa);
3557 mutex_exit(&spa_namespace_lock);
3562 * Get the list of spares, if specified.
3564 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3565 &spares, &nspares) == 0) {
3566 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3568 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3569 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3570 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3571 spa_load_spares(spa);
3572 spa_config_exit(spa, SCL_ALL, FTAG);
3573 spa->spa_spares.sav_sync = B_TRUE;
3577 * Get the list of level 2 cache devices, if specified.
3579 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3580 &l2cache, &nl2cache) == 0) {
3581 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3582 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3583 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3584 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3585 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3586 spa_load_l2cache(spa);
3587 spa_config_exit(spa, SCL_ALL, FTAG);
3588 spa->spa_l2cache.sav_sync = B_TRUE;
3591 spa->spa_is_initializing = B_TRUE;
3592 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3593 spa->spa_meta_objset = dp->dp_meta_objset;
3594 spa->spa_is_initializing = B_FALSE;
3597 * Create DDTs (dedup tables).
3601 spa_update_dspace(spa);
3603 tx = dmu_tx_create_assigned(dp, txg);
3606 * Create the pool config object.
3608 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3609 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3610 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3612 if (zap_add(spa->spa_meta_objset,
3613 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3614 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3615 cmn_err(CE_PANIC, "failed to add pool config");
3618 if (spa_version(spa) >= SPA_VERSION_FEATURES)
3619 spa_feature_create_zap_objects(spa, tx);
3621 if (zap_add(spa->spa_meta_objset,
3622 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3623 sizeof (uint64_t), 1, &version, tx) != 0) {
3624 cmn_err(CE_PANIC, "failed to add pool version");
3627 /* Newly created pools with the right version are always deflated. */
3628 if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3629 spa->spa_deflate = TRUE;
3630 if (zap_add(spa->spa_meta_objset,
3631 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3632 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3633 cmn_err(CE_PANIC, "failed to add deflate");
3638 * Create the deferred-free bpobj. Turn off compression
3639 * because sync-to-convergence takes longer if the blocksize
3642 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3643 dmu_object_set_compress(spa->spa_meta_objset, obj,
3644 ZIO_COMPRESS_OFF, tx);
3645 if (zap_add(spa->spa_meta_objset,
3646 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3647 sizeof (uint64_t), 1, &obj, tx) != 0) {
3648 cmn_err(CE_PANIC, "failed to add bpobj");
3650 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3651 spa->spa_meta_objset, obj));
3654 * Create the pool's history object.
3656 if (version >= SPA_VERSION_ZPOOL_HISTORY)
3657 spa_history_create_obj(spa, tx);
3660 * Set pool properties.
3662 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3663 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3664 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3665 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3667 if (props != NULL) {
3668 spa_configfile_set(spa, props, B_FALSE);
3669 spa_sync_props(props, tx);
3674 spa->spa_sync_on = B_TRUE;
3675 txg_sync_start(spa->spa_dsl_pool);
3678 * We explicitly wait for the first transaction to complete so that our
3679 * bean counters are appropriately updated.
3681 txg_wait_synced(spa->spa_dsl_pool, txg);
3683 spa_config_sync(spa, B_FALSE, B_TRUE);
3685 spa_history_log_version(spa, "create");
3687 spa->spa_minref = refcount_count(&spa->spa_refcount);
3689 mutex_exit(&spa_namespace_lock);
3697 * Get the root pool information from the root disk, then import the root pool
3698 * during the system boot up time.
3700 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3703 spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3706 nvlist_t *nvtop, *nvroot;
3709 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3713 * Add this top-level vdev to the child array.
3715 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3717 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3719 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3722 * Put this pool's top-level vdevs into a root vdev.
3724 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3725 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3726 VDEV_TYPE_ROOT) == 0);
3727 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3728 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3729 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3733 * Replace the existing vdev_tree with the new root vdev in
3734 * this pool's configuration (remove the old, add the new).
3736 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3737 nvlist_free(nvroot);
3742 * Walk the vdev tree and see if we can find a device with "better"
3743 * configuration. A configuration is "better" if the label on that
3744 * device has a more recent txg.
3747 spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3749 for (int c = 0; c < vd->vdev_children; c++)
3750 spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3752 if (vd->vdev_ops->vdev_op_leaf) {
3756 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3760 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3764 * Do we have a better boot device?
3766 if (label_txg > *txg) {
3775 * Import a root pool.
3777 * For x86. devpath_list will consist of devid and/or physpath name of
3778 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3779 * The GRUB "findroot" command will return the vdev we should boot.
3781 * For Sparc, devpath_list consists the physpath name of the booting device
3782 * no matter the rootpool is a single device pool or a mirrored pool.
3784 * "/pci@1f,0/ide@d/disk@0,0:a"
3787 spa_import_rootpool(char *devpath, char *devid)
3790 vdev_t *rvd, *bvd, *avd = NULL;
3791 nvlist_t *config, *nvtop;
3797 * Read the label from the boot device and generate a configuration.
3799 config = spa_generate_rootconf(devpath, devid, &guid);
3800 #if defined(_OBP) && defined(_KERNEL)
3801 if (config == NULL) {
3802 if (strstr(devpath, "/iscsi/ssd") != NULL) {
3804 get_iscsi_bootpath_phy(devpath);
3805 config = spa_generate_rootconf(devpath, devid, &guid);
3809 if (config == NULL) {
3810 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
3812 return (SET_ERROR(EIO));
3815 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3817 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3819 mutex_enter(&spa_namespace_lock);
3820 if ((spa = spa_lookup(pname)) != NULL) {
3822 * Remove the existing root pool from the namespace so that we
3823 * can replace it with the correct config we just read in.
3828 spa = spa_add(pname, config, NULL);
3829 spa->spa_is_root = B_TRUE;
3830 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3833 * Build up a vdev tree based on the boot device's label config.
3835 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3837 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3838 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3839 VDEV_ALLOC_ROOTPOOL);
3840 spa_config_exit(spa, SCL_ALL, FTAG);
3842 mutex_exit(&spa_namespace_lock);
3843 nvlist_free(config);
3844 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3850 * Get the boot vdev.
3852 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3853 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3854 (u_longlong_t)guid);
3855 error = SET_ERROR(ENOENT);
3860 * Determine if there is a better boot device.
3863 spa_alt_rootvdev(rvd, &avd, &txg);
3865 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3866 "try booting from '%s'", avd->vdev_path);
3867 error = SET_ERROR(EINVAL);
3872 * If the boot device is part of a spare vdev then ensure that
3873 * we're booting off the active spare.
3875 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3876 !bvd->vdev_isspare) {
3877 cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3878 "try booting from '%s'",
3880 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3881 error = SET_ERROR(EINVAL);
3887 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3889 spa_config_exit(spa, SCL_ALL, FTAG);
3890 mutex_exit(&spa_namespace_lock);
3892 nvlist_free(config);
3898 extern int vdev_geom_read_pool_label(const char *name, nvlist_t ***configs,
3902 spa_generate_rootconf(const char *name)
3904 nvlist_t **configs, **tops;
3906 nvlist_t *best_cfg, *nvtop, *nvroot;
3915 if (vdev_geom_read_pool_label(name, &configs, &count) != 0)
3918 ASSERT3U(count, !=, 0);
3920 for (i = 0; i < count; i++) {
3923 VERIFY(nvlist_lookup_uint64(configs[i], ZPOOL_CONFIG_POOL_TXG,
3925 if (txg > best_txg) {
3927 best_cfg = configs[i];
3932 * Multi-vdev root pool configuration discovery is not supported yet.
3935 nvlist_lookup_uint64(best_cfg, ZPOOL_CONFIG_VDEV_CHILDREN, &nchildren);
3937 nvlist_lookup_uint64_array(best_cfg, ZPOOL_CONFIG_HOLE_ARRAY,
3940 tops = kmem_zalloc(nchildren * sizeof(void *), KM_SLEEP);
3941 for (i = 0; i < nchildren; i++) {
3944 if (configs[i] == NULL)
3946 VERIFY(nvlist_lookup_nvlist(configs[i], ZPOOL_CONFIG_VDEV_TREE,
3948 nvlist_dup(nvtop, &tops[i], KM_SLEEP);
3950 for (i = 0; holes != NULL && i < nholes; i++) {
3953 if (tops[holes[i]] != NULL)
3955 nvlist_alloc(&tops[holes[i]], NV_UNIQUE_NAME, KM_SLEEP);
3956 VERIFY(nvlist_add_string(tops[holes[i]], ZPOOL_CONFIG_TYPE,
3957 VDEV_TYPE_HOLE) == 0);
3958 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_ID,
3960 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_GUID,
3963 for (i = 0; i < nchildren; i++) {
3964 if (tops[i] != NULL)
3966 nvlist_alloc(&tops[i], NV_UNIQUE_NAME, KM_SLEEP);
3967 VERIFY(nvlist_add_string(tops[i], ZPOOL_CONFIG_TYPE,
3968 VDEV_TYPE_MISSING) == 0);
3969 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_ID,
3971 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_GUID,
3976 * Create pool config based on the best vdev config.
3978 nvlist_dup(best_cfg, &config, KM_SLEEP);
3981 * Put this pool's top-level vdevs into a root vdev.
3983 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3985 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3986 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3987 VDEV_TYPE_ROOT) == 0);
3988 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3989 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3990 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3991 tops, nchildren) == 0);
3994 * Replace the existing vdev_tree with the new root vdev in
3995 * this pool's configuration (remove the old, add the new).
3997 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
4000 * Drop vdev config elements that should not be present at pool level.
4002 nvlist_remove(config, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64);
4003 nvlist_remove(config, ZPOOL_CONFIG_TOP_GUID, DATA_TYPE_UINT64);
4005 for (i = 0; i < count; i++)
4006 nvlist_free(configs[i]);
4007 kmem_free(configs, count * sizeof(void *));
4008 for (i = 0; i < nchildren; i++)
4009 nvlist_free(tops[i]);
4010 kmem_free(tops, nchildren * sizeof(void *));
4011 nvlist_free(nvroot);
4016 spa_import_rootpool(const char *name)
4019 vdev_t *rvd, *bvd, *avd = NULL;
4020 nvlist_t *config, *nvtop;
4026 * Read the label from the boot device and generate a configuration.
4028 config = spa_generate_rootconf(name);
4030 mutex_enter(&spa_namespace_lock);
4031 if (config != NULL) {
4032 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
4033 &pname) == 0 && strcmp(name, pname) == 0);
4034 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg)
4037 if ((spa = spa_lookup(pname)) != NULL) {
4039 * Remove the existing root pool from the namespace so
4040 * that we can replace it with the correct config
4045 spa = spa_add(pname, config, NULL);
4048 * Set spa_ubsync.ub_version as it can be used in vdev_alloc()
4049 * via spa_version().
4051 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
4052 &spa->spa_ubsync.ub_version) != 0)
4053 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
4054 } else if ((spa = spa_lookup(name)) == NULL) {
4055 cmn_err(CE_NOTE, "Cannot find the pool label for '%s'",
4059 VERIFY(nvlist_dup(spa->spa_config, &config, KM_SLEEP) == 0);
4061 spa->spa_is_root = B_TRUE;
4062 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
4065 * Build up a vdev tree based on the boot device's label config.
4067 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4069 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4070 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
4071 VDEV_ALLOC_ROOTPOOL);
4072 spa_config_exit(spa, SCL_ALL, FTAG);
4074 mutex_exit(&spa_namespace_lock);
4075 nvlist_free(config);
4076 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
4081 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4083 spa_config_exit(spa, SCL_ALL, FTAG);
4084 mutex_exit(&spa_namespace_lock);
4086 nvlist_free(config);
4094 * Import a non-root pool into the system.
4097 spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
4100 char *altroot = NULL;
4101 spa_load_state_t state = SPA_LOAD_IMPORT;
4102 zpool_rewind_policy_t policy;
4103 uint64_t mode = spa_mode_global;
4104 uint64_t readonly = B_FALSE;
4107 nvlist_t **spares, **l2cache;
4108 uint_t nspares, nl2cache;
4111 * If a pool with this name exists, return failure.
4113 mutex_enter(&spa_namespace_lock);
4114 if (spa_lookup(pool) != NULL) {
4115 mutex_exit(&spa_namespace_lock);
4116 return (SET_ERROR(EEXIST));
4120 * Create and initialize the spa structure.
4122 (void) nvlist_lookup_string(props,
4123 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4124 (void) nvlist_lookup_uint64(props,
4125 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
4128 spa = spa_add(pool, config, altroot);
4129 spa->spa_import_flags = flags;
4132 * Verbatim import - Take a pool and insert it into the namespace
4133 * as if it had been loaded at boot.
4135 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
4137 spa_configfile_set(spa, props, B_FALSE);
4139 spa_config_sync(spa, B_FALSE, B_TRUE);
4141 mutex_exit(&spa_namespace_lock);
4145 spa_activate(spa, mode);
4148 * Don't start async tasks until we know everything is healthy.
4150 spa_async_suspend(spa);
4152 zpool_get_rewind_policy(config, &policy);
4153 if (policy.zrp_request & ZPOOL_DO_REWIND)
4154 state = SPA_LOAD_RECOVER;
4157 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
4158 * because the user-supplied config is actually the one to trust when
4161 if (state != SPA_LOAD_RECOVER)
4162 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
4164 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
4165 policy.zrp_request);
4168 * Propagate anything learned while loading the pool and pass it
4169 * back to caller (i.e. rewind info, missing devices, etc).
4171 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4172 spa->spa_load_info) == 0);
4174 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4176 * Toss any existing sparelist, as it doesn't have any validity
4177 * anymore, and conflicts with spa_has_spare().
4179 if (spa->spa_spares.sav_config) {
4180 nvlist_free(spa->spa_spares.sav_config);
4181 spa->spa_spares.sav_config = NULL;
4182 spa_load_spares(spa);
4184 if (spa->spa_l2cache.sav_config) {
4185 nvlist_free(spa->spa_l2cache.sav_config);
4186 spa->spa_l2cache.sav_config = NULL;
4187 spa_load_l2cache(spa);
4190 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4193 error = spa_validate_aux(spa, nvroot, -1ULL,
4196 error = spa_validate_aux(spa, nvroot, -1ULL,
4197 VDEV_ALLOC_L2CACHE);
4198 spa_config_exit(spa, SCL_ALL, FTAG);
4201 spa_configfile_set(spa, props, B_FALSE);
4203 if (error != 0 || (props && spa_writeable(spa) &&
4204 (error = spa_prop_set(spa, props)))) {
4206 spa_deactivate(spa);
4208 mutex_exit(&spa_namespace_lock);
4212 spa_async_resume(spa);
4215 * Override any spares and level 2 cache devices as specified by
4216 * the user, as these may have correct device names/devids, etc.
4218 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
4219 &spares, &nspares) == 0) {
4220 if (spa->spa_spares.sav_config)
4221 VERIFY(nvlist_remove(spa->spa_spares.sav_config,
4222 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
4224 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
4225 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4226 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
4227 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
4228 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4229 spa_load_spares(spa);
4230 spa_config_exit(spa, SCL_ALL, FTAG);
4231 spa->spa_spares.sav_sync = B_TRUE;
4233 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
4234 &l2cache, &nl2cache) == 0) {
4235 if (spa->spa_l2cache.sav_config)
4236 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
4237 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
4239 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
4240 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4241 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
4242 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
4243 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4244 spa_load_l2cache(spa);
4245 spa_config_exit(spa, SCL_ALL, FTAG);
4246 spa->spa_l2cache.sav_sync = B_TRUE;
4250 * Check for any removed devices.
4252 if (spa->spa_autoreplace) {
4253 spa_aux_check_removed(&spa->spa_spares);
4254 spa_aux_check_removed(&spa->spa_l2cache);
4257 if (spa_writeable(spa)) {
4259 * Update the config cache to include the newly-imported pool.
4261 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4265 * It's possible that the pool was expanded while it was exported.
4266 * We kick off an async task to handle this for us.
4268 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
4270 mutex_exit(&spa_namespace_lock);
4271 spa_history_log_version(spa, "import");
4275 zvol_create_minors(pool);
4282 spa_tryimport(nvlist_t *tryconfig)
4284 nvlist_t *config = NULL;
4290 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
4293 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
4297 * Create and initialize the spa structure.
4299 mutex_enter(&spa_namespace_lock);
4300 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
4301 spa_activate(spa, FREAD);
4304 * Pass off the heavy lifting to spa_load().
4305 * Pass TRUE for mosconfig because the user-supplied config
4306 * is actually the one to trust when doing an import.
4308 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
4311 * If 'tryconfig' was at least parsable, return the current config.
4313 if (spa->spa_root_vdev != NULL) {
4314 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4315 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
4317 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4319 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
4320 spa->spa_uberblock.ub_timestamp) == 0);
4321 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4322 spa->spa_load_info) == 0);
4325 * If the bootfs property exists on this pool then we
4326 * copy it out so that external consumers can tell which
4327 * pools are bootable.
4329 if ((!error || error == EEXIST) && spa->spa_bootfs) {
4330 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4333 * We have to play games with the name since the
4334 * pool was opened as TRYIMPORT_NAME.
4336 if (dsl_dsobj_to_dsname(spa_name(spa),
4337 spa->spa_bootfs, tmpname) == 0) {
4339 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4341 cp = strchr(tmpname, '/');
4343 (void) strlcpy(dsname, tmpname,
4346 (void) snprintf(dsname, MAXPATHLEN,
4347 "%s/%s", poolname, ++cp);
4349 VERIFY(nvlist_add_string(config,
4350 ZPOOL_CONFIG_BOOTFS, dsname) == 0);
4351 kmem_free(dsname, MAXPATHLEN);
4353 kmem_free(tmpname, MAXPATHLEN);
4357 * Add the list of hot spares and level 2 cache devices.
4359 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4360 spa_add_spares(spa, config);
4361 spa_add_l2cache(spa, config);
4362 spa_config_exit(spa, SCL_CONFIG, FTAG);
4366 spa_deactivate(spa);
4368 mutex_exit(&spa_namespace_lock);
4374 * Pool export/destroy
4376 * The act of destroying or exporting a pool is very simple. We make sure there
4377 * is no more pending I/O and any references to the pool are gone. Then, we
4378 * update the pool state and sync all the labels to disk, removing the
4379 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4380 * we don't sync the labels or remove the configuration cache.
4383 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
4384 boolean_t force, boolean_t hardforce)
4391 if (!(spa_mode_global & FWRITE))
4392 return (SET_ERROR(EROFS));
4394 mutex_enter(&spa_namespace_lock);
4395 if ((spa = spa_lookup(pool)) == NULL) {
4396 mutex_exit(&spa_namespace_lock);
4397 return (SET_ERROR(ENOENT));
4401 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4402 * reacquire the namespace lock, and see if we can export.
4404 spa_open_ref(spa, FTAG);
4405 mutex_exit(&spa_namespace_lock);
4406 spa_async_suspend(spa);
4407 mutex_enter(&spa_namespace_lock);
4408 spa_close(spa, FTAG);
4411 * The pool will be in core if it's openable,
4412 * in which case we can modify its state.
4414 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
4416 * Objsets may be open only because they're dirty, so we
4417 * have to force it to sync before checking spa_refcnt.
4419 txg_wait_synced(spa->spa_dsl_pool, 0);
4422 * A pool cannot be exported or destroyed if there are active
4423 * references. If we are resetting a pool, allow references by
4424 * fault injection handlers.
4426 if (!spa_refcount_zero(spa) ||
4427 (spa->spa_inject_ref != 0 &&
4428 new_state != POOL_STATE_UNINITIALIZED)) {
4429 spa_async_resume(spa);
4430 mutex_exit(&spa_namespace_lock);
4431 return (SET_ERROR(EBUSY));
4435 * A pool cannot be exported if it has an active shared spare.
4436 * This is to prevent other pools stealing the active spare
4437 * from an exported pool. At user's own will, such pool can
4438 * be forcedly exported.
4440 if (!force && new_state == POOL_STATE_EXPORTED &&
4441 spa_has_active_shared_spare(spa)) {
4442 spa_async_resume(spa);
4443 mutex_exit(&spa_namespace_lock);
4444 return (SET_ERROR(EXDEV));
4448 * We want this to be reflected on every label,
4449 * so mark them all dirty. spa_unload() will do the
4450 * final sync that pushes these changes out.
4452 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
4453 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4454 spa->spa_state = new_state;
4455 spa->spa_final_txg = spa_last_synced_txg(spa) +
4457 vdev_config_dirty(spa->spa_root_vdev);
4458 spa_config_exit(spa, SCL_ALL, FTAG);
4462 spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
4464 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4466 spa_deactivate(spa);
4469 if (oldconfig && spa->spa_config)
4470 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
4472 if (new_state != POOL_STATE_UNINITIALIZED) {
4474 spa_config_sync(spa, B_TRUE, B_TRUE);
4477 mutex_exit(&spa_namespace_lock);
4483 * Destroy a storage pool.
4486 spa_destroy(char *pool)
4488 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
4493 * Export a storage pool.
4496 spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
4497 boolean_t hardforce)
4499 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
4504 * Similar to spa_export(), this unloads the spa_t without actually removing it
4505 * from the namespace in any way.
4508 spa_reset(char *pool)
4510 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
4515 * ==========================================================================
4516 * Device manipulation
4517 * ==========================================================================
4521 * Add a device to a storage pool.
4524 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
4528 vdev_t *rvd = spa->spa_root_vdev;
4530 nvlist_t **spares, **l2cache;
4531 uint_t nspares, nl2cache;
4533 ASSERT(spa_writeable(spa));
4535 txg = spa_vdev_enter(spa);
4537 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
4538 VDEV_ALLOC_ADD)) != 0)
4539 return (spa_vdev_exit(spa, NULL, txg, error));
4541 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */
4543 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
4547 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
4551 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
4552 return (spa_vdev_exit(spa, vd, txg, EINVAL));
4554 if (vd->vdev_children != 0 &&
4555 (error = vdev_create(vd, txg, B_FALSE)) != 0)
4556 return (spa_vdev_exit(spa, vd, txg, error));
4559 * We must validate the spares and l2cache devices after checking the
4560 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
4562 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
4563 return (spa_vdev_exit(spa, vd, txg, error));
4566 * Transfer each new top-level vdev from vd to rvd.
4568 for (int c = 0; c < vd->vdev_children; c++) {
4571 * Set the vdev id to the first hole, if one exists.
4573 for (id = 0; id < rvd->vdev_children; id++) {
4574 if (rvd->vdev_child[id]->vdev_ishole) {
4575 vdev_free(rvd->vdev_child[id]);
4579 tvd = vd->vdev_child[c];
4580 vdev_remove_child(vd, tvd);
4582 vdev_add_child(rvd, tvd);
4583 vdev_config_dirty(tvd);
4587 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
4588 ZPOOL_CONFIG_SPARES);
4589 spa_load_spares(spa);
4590 spa->spa_spares.sav_sync = B_TRUE;
4593 if (nl2cache != 0) {
4594 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
4595 ZPOOL_CONFIG_L2CACHE);
4596 spa_load_l2cache(spa);
4597 spa->spa_l2cache.sav_sync = B_TRUE;
4601 * We have to be careful when adding new vdevs to an existing pool.
4602 * If other threads start allocating from these vdevs before we
4603 * sync the config cache, and we lose power, then upon reboot we may
4604 * fail to open the pool because there are DVAs that the config cache
4605 * can't translate. Therefore, we first add the vdevs without
4606 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4607 * and then let spa_config_update() initialize the new metaslabs.
4609 * spa_load() checks for added-but-not-initialized vdevs, so that
4610 * if we lose power at any point in this sequence, the remaining
4611 * steps will be completed the next time we load the pool.
4613 (void) spa_vdev_exit(spa, vd, txg, 0);
4615 mutex_enter(&spa_namespace_lock);
4616 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4617 mutex_exit(&spa_namespace_lock);
4623 * Attach a device to a mirror. The arguments are the path to any device
4624 * in the mirror, and the nvroot for the new device. If the path specifies
4625 * a device that is not mirrored, we automatically insert the mirror vdev.
4627 * If 'replacing' is specified, the new device is intended to replace the
4628 * existing device; in this case the two devices are made into their own
4629 * mirror using the 'replacing' vdev, which is functionally identical to
4630 * the mirror vdev (it actually reuses all the same ops) but has a few
4631 * extra rules: you can't attach to it after it's been created, and upon
4632 * completion of resilvering, the first disk (the one being replaced)
4633 * is automatically detached.
4636 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
4638 uint64_t txg, dtl_max_txg;
4639 vdev_t *rvd = spa->spa_root_vdev;
4640 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
4642 char *oldvdpath, *newvdpath;
4646 ASSERT(spa_writeable(spa));
4648 txg = spa_vdev_enter(spa);
4650 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
4653 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4655 if (!oldvd->vdev_ops->vdev_op_leaf)
4656 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4658 pvd = oldvd->vdev_parent;
4660 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
4661 VDEV_ALLOC_ATTACH)) != 0)
4662 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4664 if (newrootvd->vdev_children != 1)
4665 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4667 newvd = newrootvd->vdev_child[0];
4669 if (!newvd->vdev_ops->vdev_op_leaf)
4670 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4672 if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
4673 return (spa_vdev_exit(spa, newrootvd, txg, error));
4676 * Spares can't replace logs
4678 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
4679 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4683 * For attach, the only allowable parent is a mirror or the root
4686 if (pvd->vdev_ops != &vdev_mirror_ops &&
4687 pvd->vdev_ops != &vdev_root_ops)
4688 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4690 pvops = &vdev_mirror_ops;
4693 * Active hot spares can only be replaced by inactive hot
4696 if (pvd->vdev_ops == &vdev_spare_ops &&
4697 oldvd->vdev_isspare &&
4698 !spa_has_spare(spa, newvd->vdev_guid))
4699 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4702 * If the source is a hot spare, and the parent isn't already a
4703 * spare, then we want to create a new hot spare. Otherwise, we
4704 * want to create a replacing vdev. The user is not allowed to
4705 * attach to a spared vdev child unless the 'isspare' state is
4706 * the same (spare replaces spare, non-spare replaces
4709 if (pvd->vdev_ops == &vdev_replacing_ops &&
4710 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
4711 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4712 } else if (pvd->vdev_ops == &vdev_spare_ops &&
4713 newvd->vdev_isspare != oldvd->vdev_isspare) {
4714 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4717 if (newvd->vdev_isspare)
4718 pvops = &vdev_spare_ops;
4720 pvops = &vdev_replacing_ops;
4724 * Make sure the new device is big enough.
4726 if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
4727 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
4730 * The new device cannot have a higher alignment requirement
4731 * than the top-level vdev.
4733 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4734 return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4737 * If this is an in-place replacement, update oldvd's path and devid
4738 * to make it distinguishable from newvd, and unopenable from now on.
4740 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4741 spa_strfree(oldvd->vdev_path);
4742 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4744 (void) sprintf(oldvd->vdev_path, "%s/%s",
4745 newvd->vdev_path, "old");
4746 if (oldvd->vdev_devid != NULL) {
4747 spa_strfree(oldvd->vdev_devid);
4748 oldvd->vdev_devid = NULL;
4752 /* mark the device being resilvered */
4753 newvd->vdev_resilver_txg = txg;
4756 * If the parent is not a mirror, or if we're replacing, insert the new
4757 * mirror/replacing/spare vdev above oldvd.
4759 if (pvd->vdev_ops != pvops)
4760 pvd = vdev_add_parent(oldvd, pvops);
4762 ASSERT(pvd->vdev_top->vdev_parent == rvd);
4763 ASSERT(pvd->vdev_ops == pvops);
4764 ASSERT(oldvd->vdev_parent == pvd);
4767 * Extract the new device from its root and add it to pvd.
4769 vdev_remove_child(newrootvd, newvd);
4770 newvd->vdev_id = pvd->vdev_children;
4771 newvd->vdev_crtxg = oldvd->vdev_crtxg;
4772 vdev_add_child(pvd, newvd);
4774 tvd = newvd->vdev_top;
4775 ASSERT(pvd->vdev_top == tvd);
4776 ASSERT(tvd->vdev_parent == rvd);
4778 vdev_config_dirty(tvd);
4781 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4782 * for any dmu_sync-ed blocks. It will propagate upward when
4783 * spa_vdev_exit() calls vdev_dtl_reassess().
4785 dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4787 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4788 dtl_max_txg - TXG_INITIAL);
4790 if (newvd->vdev_isspare) {
4791 spa_spare_activate(newvd);
4792 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
4795 oldvdpath = spa_strdup(oldvd->vdev_path);
4796 newvdpath = spa_strdup(newvd->vdev_path);
4797 newvd_isspare = newvd->vdev_isspare;
4800 * Mark newvd's DTL dirty in this txg.
4802 vdev_dirty(tvd, VDD_DTL, newvd, txg);
4805 * Schedule the resilver to restart in the future. We do this to
4806 * ensure that dmu_sync-ed blocks have been stitched into the
4807 * respective datasets.
4809 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4814 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4816 spa_history_log_internal(spa, "vdev attach", NULL,
4817 "%s vdev=%s %s vdev=%s",
4818 replacing && newvd_isspare ? "spare in" :
4819 replacing ? "replace" : "attach", newvdpath,
4820 replacing ? "for" : "to", oldvdpath);
4822 spa_strfree(oldvdpath);
4823 spa_strfree(newvdpath);
4825 if (spa->spa_bootfs)
4826 spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
4832 * Detach a device from a mirror or replacing vdev.
4834 * If 'replace_done' is specified, only detach if the parent
4835 * is a replacing vdev.
4838 spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4842 vdev_t *rvd = spa->spa_root_vdev;
4843 vdev_t *vd, *pvd, *cvd, *tvd;
4844 boolean_t unspare = B_FALSE;
4845 uint64_t unspare_guid = 0;
4848 ASSERT(spa_writeable(spa));
4850 txg = spa_vdev_enter(spa);
4852 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4855 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4857 if (!vd->vdev_ops->vdev_op_leaf)
4858 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4860 pvd = vd->vdev_parent;
4863 * If the parent/child relationship is not as expected, don't do it.
4864 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4865 * vdev that's replacing B with C. The user's intent in replacing
4866 * is to go from M(A,B) to M(A,C). If the user decides to cancel
4867 * the replace by detaching C, the expected behavior is to end up
4868 * M(A,B). But suppose that right after deciding to detach C,
4869 * the replacement of B completes. We would have M(A,C), and then
4870 * ask to detach C, which would leave us with just A -- not what
4871 * the user wanted. To prevent this, we make sure that the
4872 * parent/child relationship hasn't changed -- in this example,
4873 * that C's parent is still the replacing vdev R.
4875 if (pvd->vdev_guid != pguid && pguid != 0)
4876 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4879 * Only 'replacing' or 'spare' vdevs can be replaced.
4881 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4882 pvd->vdev_ops != &vdev_spare_ops)
4883 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4885 ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4886 spa_version(spa) >= SPA_VERSION_SPARES);
4889 * Only mirror, replacing, and spare vdevs support detach.
4891 if (pvd->vdev_ops != &vdev_replacing_ops &&
4892 pvd->vdev_ops != &vdev_mirror_ops &&
4893 pvd->vdev_ops != &vdev_spare_ops)
4894 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4897 * If this device has the only valid copy of some data,
4898 * we cannot safely detach it.
4900 if (vdev_dtl_required(vd))
4901 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4903 ASSERT(pvd->vdev_children >= 2);
4906 * If we are detaching the second disk from a replacing vdev, then
4907 * check to see if we changed the original vdev's path to have "/old"
4908 * at the end in spa_vdev_attach(). If so, undo that change now.
4910 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4911 vd->vdev_path != NULL) {
4912 size_t len = strlen(vd->vdev_path);
4914 for (int c = 0; c < pvd->vdev_children; c++) {
4915 cvd = pvd->vdev_child[c];
4917 if (cvd == vd || cvd->vdev_path == NULL)
4920 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4921 strcmp(cvd->vdev_path + len, "/old") == 0) {
4922 spa_strfree(cvd->vdev_path);
4923 cvd->vdev_path = spa_strdup(vd->vdev_path);
4930 * If we are detaching the original disk from a spare, then it implies
4931 * that the spare should become a real disk, and be removed from the
4932 * active spare list for the pool.
4934 if (pvd->vdev_ops == &vdev_spare_ops &&
4936 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4940 * Erase the disk labels so the disk can be used for other things.
4941 * This must be done after all other error cases are handled,
4942 * but before we disembowel vd (so we can still do I/O to it).
4943 * But if we can't do it, don't treat the error as fatal --
4944 * it may be that the unwritability of the disk is the reason
4945 * it's being detached!
4947 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4950 * Remove vd from its parent and compact the parent's children.
4952 vdev_remove_child(pvd, vd);
4953 vdev_compact_children(pvd);
4956 * Remember one of the remaining children so we can get tvd below.
4958 cvd = pvd->vdev_child[pvd->vdev_children - 1];
4961 * If we need to remove the remaining child from the list of hot spares,
4962 * do it now, marking the vdev as no longer a spare in the process.
4963 * We must do this before vdev_remove_parent(), because that can
4964 * change the GUID if it creates a new toplevel GUID. For a similar
4965 * reason, we must remove the spare now, in the same txg as the detach;
4966 * otherwise someone could attach a new sibling, change the GUID, and
4967 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4970 ASSERT(cvd->vdev_isspare);
4971 spa_spare_remove(cvd);
4972 unspare_guid = cvd->vdev_guid;
4973 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4974 cvd->vdev_unspare = B_TRUE;
4978 * If the parent mirror/replacing vdev only has one child,
4979 * the parent is no longer needed. Remove it from the tree.
4981 if (pvd->vdev_children == 1) {
4982 if (pvd->vdev_ops == &vdev_spare_ops)
4983 cvd->vdev_unspare = B_FALSE;
4984 vdev_remove_parent(cvd);
4989 * We don't set tvd until now because the parent we just removed
4990 * may have been the previous top-level vdev.
4992 tvd = cvd->vdev_top;
4993 ASSERT(tvd->vdev_parent == rvd);
4996 * Reevaluate the parent vdev state.
4998 vdev_propagate_state(cvd);
5001 * If the 'autoexpand' property is set on the pool then automatically
5002 * try to expand the size of the pool. For example if the device we
5003 * just detached was smaller than the others, it may be possible to
5004 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
5005 * first so that we can obtain the updated sizes of the leaf vdevs.
5007 if (spa->spa_autoexpand) {
5009 vdev_expand(tvd, txg);
5012 vdev_config_dirty(tvd);
5015 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
5016 * vd->vdev_detached is set and free vd's DTL object in syncing context.
5017 * But first make sure we're not on any *other* txg's DTL list, to
5018 * prevent vd from being accessed after it's freed.
5020 vdpath = spa_strdup(vd->vdev_path);
5021 for (int t = 0; t < TXG_SIZE; t++)
5022 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
5023 vd->vdev_detached = B_TRUE;
5024 vdev_dirty(tvd, VDD_DTL, vd, txg);
5026 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
5028 /* hang on to the spa before we release the lock */
5029 spa_open_ref(spa, FTAG);
5031 error = spa_vdev_exit(spa, vd, txg, 0);
5033 spa_history_log_internal(spa, "detach", NULL,
5035 spa_strfree(vdpath);
5038 * If this was the removal of the original device in a hot spare vdev,
5039 * then we want to go through and remove the device from the hot spare
5040 * list of every other pool.
5043 spa_t *altspa = NULL;
5045 mutex_enter(&spa_namespace_lock);
5046 while ((altspa = spa_next(altspa)) != NULL) {
5047 if (altspa->spa_state != POOL_STATE_ACTIVE ||
5051 spa_open_ref(altspa, FTAG);
5052 mutex_exit(&spa_namespace_lock);
5053 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
5054 mutex_enter(&spa_namespace_lock);
5055 spa_close(altspa, FTAG);
5057 mutex_exit(&spa_namespace_lock);
5059 /* search the rest of the vdevs for spares to remove */
5060 spa_vdev_resilver_done(spa);
5063 /* all done with the spa; OK to release */
5064 mutex_enter(&spa_namespace_lock);
5065 spa_close(spa, FTAG);
5066 mutex_exit(&spa_namespace_lock);
5072 * Split a set of devices from their mirrors, and create a new pool from them.
5075 spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
5076 nvlist_t *props, boolean_t exp)
5079 uint64_t txg, *glist;
5081 uint_t c, children, lastlog;
5082 nvlist_t **child, *nvl, *tmp;
5084 char *altroot = NULL;
5085 vdev_t *rvd, **vml = NULL; /* vdev modify list */
5086 boolean_t activate_slog;
5088 ASSERT(spa_writeable(spa));
5090 txg = spa_vdev_enter(spa);
5092 /* clear the log and flush everything up to now */
5093 activate_slog = spa_passivate_log(spa);
5094 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5095 error = spa_offline_log(spa);
5096 txg = spa_vdev_config_enter(spa);
5099 spa_activate_log(spa);
5102 return (spa_vdev_exit(spa, NULL, txg, error));
5104 /* check new spa name before going any further */
5105 if (spa_lookup(newname) != NULL)
5106 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
5109 * scan through all the children to ensure they're all mirrors
5111 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
5112 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
5114 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5116 /* first, check to ensure we've got the right child count */
5117 rvd = spa->spa_root_vdev;
5119 for (c = 0; c < rvd->vdev_children; c++) {
5120 vdev_t *vd = rvd->vdev_child[c];
5122 /* don't count the holes & logs as children */
5123 if (vd->vdev_islog || vd->vdev_ishole) {
5131 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
5132 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5134 /* next, ensure no spare or cache devices are part of the split */
5135 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
5136 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
5137 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5139 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
5140 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
5142 /* then, loop over each vdev and validate it */
5143 for (c = 0; c < children; c++) {
5144 uint64_t is_hole = 0;
5146 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
5150 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
5151 spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
5154 error = SET_ERROR(EINVAL);
5159 /* which disk is going to be split? */
5160 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
5162 error = SET_ERROR(EINVAL);
5166 /* look it up in the spa */
5167 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
5168 if (vml[c] == NULL) {
5169 error = SET_ERROR(ENODEV);
5173 /* make sure there's nothing stopping the split */
5174 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
5175 vml[c]->vdev_islog ||
5176 vml[c]->vdev_ishole ||
5177 vml[c]->vdev_isspare ||
5178 vml[c]->vdev_isl2cache ||
5179 !vdev_writeable(vml[c]) ||
5180 vml[c]->vdev_children != 0 ||
5181 vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
5182 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
5183 error = SET_ERROR(EINVAL);
5187 if (vdev_dtl_required(vml[c])) {
5188 error = SET_ERROR(EBUSY);
5192 /* we need certain info from the top level */
5193 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
5194 vml[c]->vdev_top->vdev_ms_array) == 0);
5195 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
5196 vml[c]->vdev_top->vdev_ms_shift) == 0);
5197 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
5198 vml[c]->vdev_top->vdev_asize) == 0);
5199 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
5200 vml[c]->vdev_top->vdev_ashift) == 0);
5204 kmem_free(vml, children * sizeof (vdev_t *));
5205 kmem_free(glist, children * sizeof (uint64_t));
5206 return (spa_vdev_exit(spa, NULL, txg, error));
5209 /* stop writers from using the disks */
5210 for (c = 0; c < children; c++) {
5212 vml[c]->vdev_offline = B_TRUE;
5214 vdev_reopen(spa->spa_root_vdev);
5217 * Temporarily record the splitting vdevs in the spa config. This
5218 * will disappear once the config is regenerated.
5220 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5221 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
5222 glist, children) == 0);
5223 kmem_free(glist, children * sizeof (uint64_t));
5225 mutex_enter(&spa->spa_props_lock);
5226 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
5228 mutex_exit(&spa->spa_props_lock);
5229 spa->spa_config_splitting = nvl;
5230 vdev_config_dirty(spa->spa_root_vdev);
5232 /* configure and create the new pool */
5233 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
5234 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
5235 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
5236 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5237 spa_version(spa)) == 0);
5238 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
5239 spa->spa_config_txg) == 0);
5240 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
5241 spa_generate_guid(NULL)) == 0);
5242 (void) nvlist_lookup_string(props,
5243 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
5245 /* add the new pool to the namespace */
5246 newspa = spa_add(newname, config, altroot);
5247 newspa->spa_config_txg = spa->spa_config_txg;
5248 spa_set_log_state(newspa, SPA_LOG_CLEAR);
5250 /* release the spa config lock, retaining the namespace lock */
5251 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5253 if (zio_injection_enabled)
5254 zio_handle_panic_injection(spa, FTAG, 1);
5256 spa_activate(newspa, spa_mode_global);
5257 spa_async_suspend(newspa);
5260 /* mark that we are creating new spa by splitting */
5261 newspa->spa_splitting_newspa = B_TRUE;
5263 /* create the new pool from the disks of the original pool */
5264 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
5266 newspa->spa_splitting_newspa = B_FALSE;
5271 /* if that worked, generate a real config for the new pool */
5272 if (newspa->spa_root_vdev != NULL) {
5273 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
5274 NV_UNIQUE_NAME, KM_SLEEP) == 0);
5275 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
5276 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
5277 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
5282 if (props != NULL) {
5283 spa_configfile_set(newspa, props, B_FALSE);
5284 error = spa_prop_set(newspa, props);
5289 /* flush everything */
5290 txg = spa_vdev_config_enter(newspa);
5291 vdev_config_dirty(newspa->spa_root_vdev);
5292 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
5294 if (zio_injection_enabled)
5295 zio_handle_panic_injection(spa, FTAG, 2);
5297 spa_async_resume(newspa);
5299 /* finally, update the original pool's config */
5300 txg = spa_vdev_config_enter(spa);
5301 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
5302 error = dmu_tx_assign(tx, TXG_WAIT);
5305 for (c = 0; c < children; c++) {
5306 if (vml[c] != NULL) {
5309 spa_history_log_internal(spa, "detach", tx,
5310 "vdev=%s", vml[c]->vdev_path);
5314 vdev_config_dirty(spa->spa_root_vdev);
5315 spa->spa_config_splitting = NULL;
5319 (void) spa_vdev_exit(spa, NULL, txg, 0);
5321 if (zio_injection_enabled)
5322 zio_handle_panic_injection(spa, FTAG, 3);
5324 /* split is complete; log a history record */
5325 spa_history_log_internal(newspa, "split", NULL,
5326 "from pool %s", spa_name(spa));
5328 kmem_free(vml, children * sizeof (vdev_t *));
5330 /* if we're not going to mount the filesystems in userland, export */
5332 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
5339 spa_deactivate(newspa);
5342 txg = spa_vdev_config_enter(spa);
5344 /* re-online all offlined disks */
5345 for (c = 0; c < children; c++) {
5347 vml[c]->vdev_offline = B_FALSE;
5349 vdev_reopen(spa->spa_root_vdev);
5351 nvlist_free(spa->spa_config_splitting);
5352 spa->spa_config_splitting = NULL;
5353 (void) spa_vdev_exit(spa, NULL, txg, error);
5355 kmem_free(vml, children * sizeof (vdev_t *));
5360 spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
5362 for (int i = 0; i < count; i++) {
5365 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
5368 if (guid == target_guid)
5376 spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
5377 nvlist_t *dev_to_remove)
5379 nvlist_t **newdev = NULL;
5382 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
5384 for (int i = 0, j = 0; i < count; i++) {
5385 if (dev[i] == dev_to_remove)
5387 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
5390 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
5391 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
5393 for (int i = 0; i < count - 1; i++)
5394 nvlist_free(newdev[i]);
5397 kmem_free(newdev, (count - 1) * sizeof (void *));
5401 * Evacuate the device.
5404 spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
5409 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5410 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5411 ASSERT(vd == vd->vdev_top);
5414 * Evacuate the device. We don't hold the config lock as writer
5415 * since we need to do I/O but we do keep the
5416 * spa_namespace_lock held. Once this completes the device
5417 * should no longer have any blocks allocated on it.
5419 if (vd->vdev_islog) {
5420 if (vd->vdev_stat.vs_alloc != 0)
5421 error = spa_offline_log(spa);
5423 error = SET_ERROR(ENOTSUP);
5430 * The evacuation succeeded. Remove any remaining MOS metadata
5431 * associated with this vdev, and wait for these changes to sync.
5433 ASSERT0(vd->vdev_stat.vs_alloc);
5434 txg = spa_vdev_config_enter(spa);
5435 vd->vdev_removing = B_TRUE;
5436 vdev_dirty_leaves(vd, VDD_DTL, txg);
5437 vdev_config_dirty(vd);
5438 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5444 * Complete the removal by cleaning up the namespace.
5447 spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
5449 vdev_t *rvd = spa->spa_root_vdev;
5450 uint64_t id = vd->vdev_id;
5451 boolean_t last_vdev = (id == (rvd->vdev_children - 1));
5453 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5454 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5455 ASSERT(vd == vd->vdev_top);
5458 * Only remove any devices which are empty.
5460 if (vd->vdev_stat.vs_alloc != 0)
5463 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5465 if (list_link_active(&vd->vdev_state_dirty_node))
5466 vdev_state_clean(vd);
5467 if (list_link_active(&vd->vdev_config_dirty_node))
5468 vdev_config_clean(vd);
5473 vdev_compact_children(rvd);
5475 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
5476 vdev_add_child(rvd, vd);
5478 vdev_config_dirty(rvd);
5481 * Reassess the health of our root vdev.
5487 * Remove a device from the pool -
5489 * Removing a device from the vdev namespace requires several steps
5490 * and can take a significant amount of time. As a result we use
5491 * the spa_vdev_config_[enter/exit] functions which allow us to
5492 * grab and release the spa_config_lock while still holding the namespace
5493 * lock. During each step the configuration is synced out.
5495 * Currently, this supports removing only hot spares, slogs, and level 2 ARC
5499 spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
5502 metaslab_group_t *mg;
5503 nvlist_t **spares, **l2cache, *nv;
5505 uint_t nspares, nl2cache;
5507 boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
5509 ASSERT(spa_writeable(spa));
5512 txg = spa_vdev_enter(spa);
5514 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5516 if (spa->spa_spares.sav_vdevs != NULL &&
5517 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5518 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
5519 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
5521 * Only remove the hot spare if it's not currently in use
5524 if (vd == NULL || unspare) {
5525 spa_vdev_remove_aux(spa->spa_spares.sav_config,
5526 ZPOOL_CONFIG_SPARES, spares, nspares, nv);
5527 spa_load_spares(spa);
5528 spa->spa_spares.sav_sync = B_TRUE;
5530 error = SET_ERROR(EBUSY);
5532 } else if (spa->spa_l2cache.sav_vdevs != NULL &&
5533 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5534 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
5535 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
5537 * Cache devices can always be removed.
5539 spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
5540 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
5541 spa_load_l2cache(spa);
5542 spa->spa_l2cache.sav_sync = B_TRUE;
5543 } else if (vd != NULL && vd->vdev_islog) {
5545 ASSERT(vd == vd->vdev_top);
5550 * Stop allocating from this vdev.
5552 metaslab_group_passivate(mg);
5555 * Wait for the youngest allocations and frees to sync,
5556 * and then wait for the deferral of those frees to finish.
5558 spa_vdev_config_exit(spa, NULL,
5559 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
5562 * Attempt to evacuate the vdev.
5564 error = spa_vdev_remove_evacuate(spa, vd);
5566 txg = spa_vdev_config_enter(spa);
5569 * If we couldn't evacuate the vdev, unwind.
5572 metaslab_group_activate(mg);
5573 return (spa_vdev_exit(spa, NULL, txg, error));
5577 * Clean up the vdev namespace.
5579 spa_vdev_remove_from_namespace(spa, vd);
5581 } else if (vd != NULL) {
5583 * Normal vdevs cannot be removed (yet).
5585 error = SET_ERROR(ENOTSUP);
5588 * There is no vdev of any kind with the specified guid.
5590 error = SET_ERROR(ENOENT);
5594 return (spa_vdev_exit(spa, NULL, txg, error));
5600 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5601 * currently spared, so we can detach it.
5604 spa_vdev_resilver_done_hunt(vdev_t *vd)
5606 vdev_t *newvd, *oldvd;
5608 for (int c = 0; c < vd->vdev_children; c++) {
5609 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
5615 * Check for a completed replacement. We always consider the first
5616 * vdev in the list to be the oldest vdev, and the last one to be
5617 * the newest (see spa_vdev_attach() for how that works). In
5618 * the case where the newest vdev is faulted, we will not automatically
5619 * remove it after a resilver completes. This is OK as it will require
5620 * user intervention to determine which disk the admin wishes to keep.
5622 if (vd->vdev_ops == &vdev_replacing_ops) {
5623 ASSERT(vd->vdev_children > 1);
5625 newvd = vd->vdev_child[vd->vdev_children - 1];
5626 oldvd = vd->vdev_child[0];
5628 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
5629 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5630 !vdev_dtl_required(oldvd))
5635 * Check for a completed resilver with the 'unspare' flag set.
5637 if (vd->vdev_ops == &vdev_spare_ops) {
5638 vdev_t *first = vd->vdev_child[0];
5639 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
5641 if (last->vdev_unspare) {
5644 } else if (first->vdev_unspare) {
5651 if (oldvd != NULL &&
5652 vdev_dtl_empty(newvd, DTL_MISSING) &&
5653 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5654 !vdev_dtl_required(oldvd))
5658 * If there are more than two spares attached to a disk,
5659 * and those spares are not required, then we want to
5660 * attempt to free them up now so that they can be used
5661 * by other pools. Once we're back down to a single
5662 * disk+spare, we stop removing them.
5664 if (vd->vdev_children > 2) {
5665 newvd = vd->vdev_child[1];
5667 if (newvd->vdev_isspare && last->vdev_isspare &&
5668 vdev_dtl_empty(last, DTL_MISSING) &&
5669 vdev_dtl_empty(last, DTL_OUTAGE) &&
5670 !vdev_dtl_required(newvd))
5679 spa_vdev_resilver_done(spa_t *spa)
5681 vdev_t *vd, *pvd, *ppvd;
5682 uint64_t guid, sguid, pguid, ppguid;
5684 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5686 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
5687 pvd = vd->vdev_parent;
5688 ppvd = pvd->vdev_parent;
5689 guid = vd->vdev_guid;
5690 pguid = pvd->vdev_guid;
5691 ppguid = ppvd->vdev_guid;
5694 * If we have just finished replacing a hot spared device, then
5695 * we need to detach the parent's first child (the original hot
5698 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
5699 ppvd->vdev_children == 2) {
5700 ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
5701 sguid = ppvd->vdev_child[1]->vdev_guid;
5703 ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd));
5705 spa_config_exit(spa, SCL_ALL, FTAG);
5706 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
5708 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
5710 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5713 spa_config_exit(spa, SCL_ALL, FTAG);
5717 * Update the stored path or FRU for this vdev.
5720 spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
5724 boolean_t sync = B_FALSE;
5726 ASSERT(spa_writeable(spa));
5728 spa_vdev_state_enter(spa, SCL_ALL);
5730 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
5731 return (spa_vdev_state_exit(spa, NULL, ENOENT));
5733 if (!vd->vdev_ops->vdev_op_leaf)
5734 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
5737 if (strcmp(value, vd->vdev_path) != 0) {
5738 spa_strfree(vd->vdev_path);
5739 vd->vdev_path = spa_strdup(value);
5743 if (vd->vdev_fru == NULL) {
5744 vd->vdev_fru = spa_strdup(value);
5746 } else if (strcmp(value, vd->vdev_fru) != 0) {
5747 spa_strfree(vd->vdev_fru);
5748 vd->vdev_fru = spa_strdup(value);
5753 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
5757 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
5759 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5763 spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5765 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5769 * ==========================================================================
5771 * ==========================================================================
5775 spa_scan_stop(spa_t *spa)
5777 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5778 if (dsl_scan_resilvering(spa->spa_dsl_pool))
5779 return (SET_ERROR(EBUSY));
5780 return (dsl_scan_cancel(spa->spa_dsl_pool));
5784 spa_scan(spa_t *spa, pool_scan_func_t func)
5786 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5788 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5789 return (SET_ERROR(ENOTSUP));
5792 * If a resilver was requested, but there is no DTL on a
5793 * writeable leaf device, we have nothing to do.
5795 if (func == POOL_SCAN_RESILVER &&
5796 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5797 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5801 return (dsl_scan(spa->spa_dsl_pool, func));
5805 * ==========================================================================
5806 * SPA async task processing
5807 * ==========================================================================
5811 spa_async_remove(spa_t *spa, vdev_t *vd)
5813 if (vd->vdev_remove_wanted) {
5814 vd->vdev_remove_wanted = B_FALSE;
5815 vd->vdev_delayed_close = B_FALSE;
5816 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5819 * We want to clear the stats, but we don't want to do a full
5820 * vdev_clear() as that will cause us to throw away
5821 * degraded/faulted state as well as attempt to reopen the
5822 * device, all of which is a waste.
5824 vd->vdev_stat.vs_read_errors = 0;
5825 vd->vdev_stat.vs_write_errors = 0;
5826 vd->vdev_stat.vs_checksum_errors = 0;
5828 vdev_state_dirty(vd->vdev_top);
5831 for (int c = 0; c < vd->vdev_children; c++)
5832 spa_async_remove(spa, vd->vdev_child[c]);
5836 spa_async_probe(spa_t *spa, vdev_t *vd)
5838 if (vd->vdev_probe_wanted) {
5839 vd->vdev_probe_wanted = B_FALSE;
5840 vdev_reopen(vd); /* vdev_open() does the actual probe */
5843 for (int c = 0; c < vd->vdev_children; c++)
5844 spa_async_probe(spa, vd->vdev_child[c]);
5848 spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5854 if (!spa->spa_autoexpand)
5857 for (int c = 0; c < vd->vdev_children; c++) {
5858 vdev_t *cvd = vd->vdev_child[c];
5859 spa_async_autoexpand(spa, cvd);
5862 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5865 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
5866 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
5868 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5869 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
5871 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
5872 ESC_ZFS_VDEV_AUTOEXPAND, attr, &eid, DDI_SLEEP);
5875 kmem_free(physpath, MAXPATHLEN);
5879 spa_async_thread(void *arg)
5884 ASSERT(spa->spa_sync_on);
5886 mutex_enter(&spa->spa_async_lock);
5887 tasks = spa->spa_async_tasks;
5888 spa->spa_async_tasks &= SPA_ASYNC_REMOVE;
5889 mutex_exit(&spa->spa_async_lock);
5892 * See if the config needs to be updated.
5894 if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5895 uint64_t old_space, new_space;
5897 mutex_enter(&spa_namespace_lock);
5898 old_space = metaslab_class_get_space(spa_normal_class(spa));
5899 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5900 new_space = metaslab_class_get_space(spa_normal_class(spa));
5901 mutex_exit(&spa_namespace_lock);
5904 * If the pool grew as a result of the config update,
5905 * then log an internal history event.
5907 if (new_space != old_space) {
5908 spa_history_log_internal(spa, "vdev online", NULL,
5909 "pool '%s' size: %llu(+%llu)",
5910 spa_name(spa), new_space, new_space - old_space);
5914 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5915 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5916 spa_async_autoexpand(spa, spa->spa_root_vdev);
5917 spa_config_exit(spa, SCL_CONFIG, FTAG);
5921 * See if any devices need to be probed.
5923 if (tasks & SPA_ASYNC_PROBE) {
5924 spa_vdev_state_enter(spa, SCL_NONE);
5925 spa_async_probe(spa, spa->spa_root_vdev);
5926 (void) spa_vdev_state_exit(spa, NULL, 0);
5930 * If any devices are done replacing, detach them.
5932 if (tasks & SPA_ASYNC_RESILVER_DONE)
5933 spa_vdev_resilver_done(spa);
5936 * Kick off a resilver.
5938 if (tasks & SPA_ASYNC_RESILVER)
5939 dsl_resilver_restart(spa->spa_dsl_pool, 0);
5942 * Let the world know that we're done.
5944 mutex_enter(&spa->spa_async_lock);
5945 spa->spa_async_thread = NULL;
5946 cv_broadcast(&spa->spa_async_cv);
5947 mutex_exit(&spa->spa_async_lock);
5952 spa_async_thread_vd(void *arg)
5957 ASSERT(spa->spa_sync_on);
5959 mutex_enter(&spa->spa_async_lock);
5960 tasks = spa->spa_async_tasks;
5962 spa->spa_async_tasks &= ~SPA_ASYNC_REMOVE;
5963 mutex_exit(&spa->spa_async_lock);
5966 * See if any devices need to be marked REMOVED.
5968 if (tasks & SPA_ASYNC_REMOVE) {
5969 spa_vdev_state_enter(spa, SCL_NONE);
5970 spa_async_remove(spa, spa->spa_root_vdev);
5971 for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
5972 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5973 for (int i = 0; i < spa->spa_spares.sav_count; i++)
5974 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5975 (void) spa_vdev_state_exit(spa, NULL, 0);
5979 * Let the world know that we're done.
5981 mutex_enter(&spa->spa_async_lock);
5982 tasks = spa->spa_async_tasks;
5983 if ((tasks & SPA_ASYNC_REMOVE) != 0)
5985 spa->spa_async_thread_vd = NULL;
5986 cv_broadcast(&spa->spa_async_cv);
5987 mutex_exit(&spa->spa_async_lock);
5992 spa_async_suspend(spa_t *spa)
5994 mutex_enter(&spa->spa_async_lock);
5995 spa->spa_async_suspended++;
5996 while (spa->spa_async_thread != NULL &&
5997 spa->spa_async_thread_vd != NULL)
5998 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5999 mutex_exit(&spa->spa_async_lock);
6003 spa_async_resume(spa_t *spa)
6005 mutex_enter(&spa->spa_async_lock);
6006 ASSERT(spa->spa_async_suspended != 0);
6007 spa->spa_async_suspended--;
6008 mutex_exit(&spa->spa_async_lock);
6012 spa_async_tasks_pending(spa_t *spa)
6014 uint_t non_config_tasks;
6016 boolean_t config_task_suspended;
6018 non_config_tasks = spa->spa_async_tasks & ~(SPA_ASYNC_CONFIG_UPDATE |
6020 config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE;
6021 if (spa->spa_ccw_fail_time == 0) {
6022 config_task_suspended = B_FALSE;
6024 config_task_suspended =
6025 (gethrtime() - spa->spa_ccw_fail_time) <
6026 (zfs_ccw_retry_interval * NANOSEC);
6029 return (non_config_tasks || (config_task && !config_task_suspended));
6033 spa_async_dispatch(spa_t *spa)
6035 mutex_enter(&spa->spa_async_lock);
6036 if (spa_async_tasks_pending(spa) &&
6037 !spa->spa_async_suspended &&
6038 spa->spa_async_thread == NULL &&
6040 spa->spa_async_thread = thread_create(NULL, 0,
6041 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
6042 mutex_exit(&spa->spa_async_lock);
6046 spa_async_dispatch_vd(spa_t *spa)
6048 mutex_enter(&spa->spa_async_lock);
6049 if ((spa->spa_async_tasks & SPA_ASYNC_REMOVE) != 0 &&
6050 !spa->spa_async_suspended &&
6051 spa->spa_async_thread_vd == NULL &&
6053 spa->spa_async_thread_vd = thread_create(NULL, 0,
6054 spa_async_thread_vd, spa, 0, &p0, TS_RUN, maxclsyspri);
6055 mutex_exit(&spa->spa_async_lock);
6059 spa_async_request(spa_t *spa, int task)
6061 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
6062 mutex_enter(&spa->spa_async_lock);
6063 spa->spa_async_tasks |= task;
6064 mutex_exit(&spa->spa_async_lock);
6065 spa_async_dispatch_vd(spa);
6069 * ==========================================================================
6070 * SPA syncing routines
6071 * ==========================================================================
6075 bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
6078 bpobj_enqueue(bpo, bp, tx);
6083 spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
6087 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
6088 BP_GET_PSIZE(bp), zio->io_flags));
6093 * Note: this simple function is not inlined to make it easier to dtrace the
6094 * amount of time spent syncing frees.
6097 spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx)
6099 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6100 bplist_iterate(bpl, spa_free_sync_cb, zio, tx);
6101 VERIFY(zio_wait(zio) == 0);
6105 * Note: this simple function is not inlined to make it easier to dtrace the
6106 * amount of time spent syncing deferred frees.
6109 spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx)
6111 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6112 VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj,
6113 spa_free_sync_cb, zio, tx), ==, 0);
6114 VERIFY0(zio_wait(zio));
6119 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
6121 char *packed = NULL;
6126 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
6129 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
6130 * information. This avoids the dmu_buf_will_dirty() path and
6131 * saves us a pre-read to get data we don't actually care about.
6133 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
6134 packed = kmem_alloc(bufsize, KM_SLEEP);
6136 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
6138 bzero(packed + nvsize, bufsize - nvsize);
6140 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
6142 kmem_free(packed, bufsize);
6144 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
6145 dmu_buf_will_dirty(db, tx);
6146 *(uint64_t *)db->db_data = nvsize;
6147 dmu_buf_rele(db, FTAG);
6151 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
6152 const char *config, const char *entry)
6162 * Update the MOS nvlist describing the list of available devices.
6163 * spa_validate_aux() will have already made sure this nvlist is
6164 * valid and the vdevs are labeled appropriately.
6166 if (sav->sav_object == 0) {
6167 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
6168 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
6169 sizeof (uint64_t), tx);
6170 VERIFY(zap_update(spa->spa_meta_objset,
6171 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
6172 &sav->sav_object, tx) == 0);
6175 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
6176 if (sav->sav_count == 0) {
6177 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
6179 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
6180 for (i = 0; i < sav->sav_count; i++)
6181 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
6182 B_FALSE, VDEV_CONFIG_L2CACHE);
6183 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
6184 sav->sav_count) == 0);
6185 for (i = 0; i < sav->sav_count; i++)
6186 nvlist_free(list[i]);
6187 kmem_free(list, sav->sav_count * sizeof (void *));
6190 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
6191 nvlist_free(nvroot);
6193 sav->sav_sync = B_FALSE;
6197 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
6201 if (list_is_empty(&spa->spa_config_dirty_list))
6204 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6206 config = spa_config_generate(spa, spa->spa_root_vdev,
6207 dmu_tx_get_txg(tx), B_FALSE);
6210 * If we're upgrading the spa version then make sure that
6211 * the config object gets updated with the correct version.
6213 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
6214 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
6215 spa->spa_uberblock.ub_version);
6217 spa_config_exit(spa, SCL_STATE, FTAG);
6219 if (spa->spa_config_syncing)
6220 nvlist_free(spa->spa_config_syncing);
6221 spa->spa_config_syncing = config;
6223 spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
6227 spa_sync_version(void *arg, dmu_tx_t *tx)
6229 uint64_t *versionp = arg;
6230 uint64_t version = *versionp;
6231 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
6234 * Setting the version is special cased when first creating the pool.
6236 ASSERT(tx->tx_txg != TXG_INITIAL);
6238 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
6239 ASSERT(version >= spa_version(spa));
6241 spa->spa_uberblock.ub_version = version;
6242 vdev_config_dirty(spa->spa_root_vdev);
6243 spa_history_log_internal(spa, "set", tx, "version=%lld", version);
6247 * Set zpool properties.
6250 spa_sync_props(void *arg, dmu_tx_t *tx)
6252 nvlist_t *nvp = arg;
6253 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
6254 objset_t *mos = spa->spa_meta_objset;
6255 nvpair_t *elem = NULL;
6257 mutex_enter(&spa->spa_props_lock);
6259 while ((elem = nvlist_next_nvpair(nvp, elem))) {
6261 char *strval, *fname;
6263 const char *propname;
6264 zprop_type_t proptype;
6267 switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
6270 * We checked this earlier in spa_prop_validate().
6272 ASSERT(zpool_prop_feature(nvpair_name(elem)));
6274 fname = strchr(nvpair_name(elem), '@') + 1;
6275 VERIFY0(zfeature_lookup_name(fname, &fid));
6277 spa_feature_enable(spa, fid, tx);
6278 spa_history_log_internal(spa, "set", tx,
6279 "%s=enabled", nvpair_name(elem));
6282 case ZPOOL_PROP_VERSION:
6283 intval = fnvpair_value_uint64(elem);
6285 * The version is synced seperatly before other
6286 * properties and should be correct by now.
6288 ASSERT3U(spa_version(spa), >=, intval);
6291 case ZPOOL_PROP_ALTROOT:
6293 * 'altroot' is a non-persistent property. It should
6294 * have been set temporarily at creation or import time.
6296 ASSERT(spa->spa_root != NULL);
6299 case ZPOOL_PROP_READONLY:
6300 case ZPOOL_PROP_CACHEFILE:
6302 * 'readonly' and 'cachefile' are also non-persisitent
6306 case ZPOOL_PROP_COMMENT:
6307 strval = fnvpair_value_string(elem);
6308 if (spa->spa_comment != NULL)
6309 spa_strfree(spa->spa_comment);
6310 spa->spa_comment = spa_strdup(strval);
6312 * We need to dirty the configuration on all the vdevs
6313 * so that their labels get updated. It's unnecessary
6314 * to do this for pool creation since the vdev's
6315 * configuratoin has already been dirtied.
6317 if (tx->tx_txg != TXG_INITIAL)
6318 vdev_config_dirty(spa->spa_root_vdev);
6319 spa_history_log_internal(spa, "set", tx,
6320 "%s=%s", nvpair_name(elem), strval);
6324 * Set pool property values in the poolprops mos object.
6326 if (spa->spa_pool_props_object == 0) {
6327 spa->spa_pool_props_object =
6328 zap_create_link(mos, DMU_OT_POOL_PROPS,
6329 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
6333 /* normalize the property name */
6334 propname = zpool_prop_to_name(prop);
6335 proptype = zpool_prop_get_type(prop);
6337 if (nvpair_type(elem) == DATA_TYPE_STRING) {
6338 ASSERT(proptype == PROP_TYPE_STRING);
6339 strval = fnvpair_value_string(elem);
6340 VERIFY0(zap_update(mos,
6341 spa->spa_pool_props_object, propname,
6342 1, strlen(strval) + 1, strval, tx));
6343 spa_history_log_internal(spa, "set", tx,
6344 "%s=%s", nvpair_name(elem), strval);
6345 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
6346 intval = fnvpair_value_uint64(elem);
6348 if (proptype == PROP_TYPE_INDEX) {
6350 VERIFY0(zpool_prop_index_to_string(
6351 prop, intval, &unused));
6353 VERIFY0(zap_update(mos,
6354 spa->spa_pool_props_object, propname,
6355 8, 1, &intval, tx));
6356 spa_history_log_internal(spa, "set", tx,
6357 "%s=%lld", nvpair_name(elem), intval);
6359 ASSERT(0); /* not allowed */
6363 case ZPOOL_PROP_DELEGATION:
6364 spa->spa_delegation = intval;
6366 case ZPOOL_PROP_BOOTFS:
6367 spa->spa_bootfs = intval;
6369 case ZPOOL_PROP_FAILUREMODE:
6370 spa->spa_failmode = intval;
6372 case ZPOOL_PROP_AUTOEXPAND:
6373 spa->spa_autoexpand = intval;
6374 if (tx->tx_txg != TXG_INITIAL)
6375 spa_async_request(spa,
6376 SPA_ASYNC_AUTOEXPAND);
6378 case ZPOOL_PROP_DEDUPDITTO:
6379 spa->spa_dedup_ditto = intval;
6388 mutex_exit(&spa->spa_props_lock);
6392 * Perform one-time upgrade on-disk changes. spa_version() does not
6393 * reflect the new version this txg, so there must be no changes this
6394 * txg to anything that the upgrade code depends on after it executes.
6395 * Therefore this must be called after dsl_pool_sync() does the sync
6399 spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
6401 dsl_pool_t *dp = spa->spa_dsl_pool;
6403 ASSERT(spa->spa_sync_pass == 1);
6405 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
6407 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
6408 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
6409 dsl_pool_create_origin(dp, tx);
6411 /* Keeping the origin open increases spa_minref */
6412 spa->spa_minref += 3;
6415 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
6416 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
6417 dsl_pool_upgrade_clones(dp, tx);
6420 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
6421 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
6422 dsl_pool_upgrade_dir_clones(dp, tx);
6424 /* Keeping the freedir open increases spa_minref */
6425 spa->spa_minref += 3;
6428 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES &&
6429 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
6430 spa_feature_create_zap_objects(spa, tx);
6434 * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable
6435 * when possibility to use lz4 compression for metadata was added
6436 * Old pools that have this feature enabled must be upgraded to have
6437 * this feature active
6439 if (spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
6440 boolean_t lz4_en = spa_feature_is_enabled(spa,
6441 SPA_FEATURE_LZ4_COMPRESS);
6442 boolean_t lz4_ac = spa_feature_is_active(spa,
6443 SPA_FEATURE_LZ4_COMPRESS);
6445 if (lz4_en && !lz4_ac)
6446 spa_feature_incr(spa, SPA_FEATURE_LZ4_COMPRESS, tx);
6448 rrw_exit(&dp->dp_config_rwlock, FTAG);
6452 * Sync the specified transaction group. New blocks may be dirtied as
6453 * part of the process, so we iterate until it converges.
6456 spa_sync(spa_t *spa, uint64_t txg)
6458 dsl_pool_t *dp = spa->spa_dsl_pool;
6459 objset_t *mos = spa->spa_meta_objset;
6460 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
6461 vdev_t *rvd = spa->spa_root_vdev;
6466 VERIFY(spa_writeable(spa));
6469 * Lock out configuration changes.
6471 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
6473 spa->spa_syncing_txg = txg;
6474 spa->spa_sync_pass = 0;
6477 * If there are any pending vdev state changes, convert them
6478 * into config changes that go out with this transaction group.
6480 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6481 while (list_head(&spa->spa_state_dirty_list) != NULL) {
6483 * We need the write lock here because, for aux vdevs,
6484 * calling vdev_config_dirty() modifies sav_config.
6485 * This is ugly and will become unnecessary when we
6486 * eliminate the aux vdev wart by integrating all vdevs
6487 * into the root vdev tree.
6489 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6490 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
6491 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
6492 vdev_state_clean(vd);
6493 vdev_config_dirty(vd);
6495 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6496 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
6498 spa_config_exit(spa, SCL_STATE, FTAG);
6500 tx = dmu_tx_create_assigned(dp, txg);
6502 spa->spa_sync_starttime = gethrtime();
6504 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid,
6505 spa->spa_sync_starttime + spa->spa_deadman_synctime));
6508 callout_reset(&spa->spa_deadman_cycid,
6509 hz * spa->spa_deadman_synctime / NANOSEC, spa_deadman, spa);
6514 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
6515 * set spa_deflate if we have no raid-z vdevs.
6517 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
6518 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
6521 for (i = 0; i < rvd->vdev_children; i++) {
6522 vd = rvd->vdev_child[i];
6523 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
6526 if (i == rvd->vdev_children) {
6527 spa->spa_deflate = TRUE;
6528 VERIFY(0 == zap_add(spa->spa_meta_objset,
6529 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
6530 sizeof (uint64_t), 1, &spa->spa_deflate, tx));
6535 * If anything has changed in this txg, or if someone is waiting
6536 * for this txg to sync (eg, spa_vdev_remove()), push the
6537 * deferred frees from the previous txg. If not, leave them
6538 * alone so that we don't generate work on an otherwise idle
6541 if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
6542 !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
6543 !txg_list_empty(&dp->dp_sync_tasks, txg) ||
6544 ((dsl_scan_active(dp->dp_scan) ||
6545 txg_sync_waiting(dp)) && !spa_shutting_down(spa))) {
6546 spa_sync_deferred_frees(spa, tx);
6550 * Iterate to convergence.
6553 int pass = ++spa->spa_sync_pass;
6555 spa_sync_config_object(spa, tx);
6556 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
6557 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
6558 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
6559 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
6560 spa_errlog_sync(spa, txg);
6561 dsl_pool_sync(dp, txg);
6563 if (pass < zfs_sync_pass_deferred_free) {
6564 spa_sync_frees(spa, free_bpl, tx);
6566 bplist_iterate(free_bpl, bpobj_enqueue_cb,
6567 &spa->spa_deferred_bpobj, tx);
6571 dsl_scan_sync(dp, tx);
6573 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
6577 spa_sync_upgrades(spa, tx);
6579 } while (dmu_objset_is_dirty(mos, txg));
6582 * Rewrite the vdev configuration (which includes the uberblock)
6583 * to commit the transaction group.
6585 * If there are no dirty vdevs, we sync the uberblock to a few
6586 * random top-level vdevs that are known to be visible in the
6587 * config cache (see spa_vdev_add() for a complete description).
6588 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
6592 * We hold SCL_STATE to prevent vdev open/close/etc.
6593 * while we're attempting to write the vdev labels.
6595 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6597 if (list_is_empty(&spa->spa_config_dirty_list)) {
6598 vdev_t *svd[SPA_DVAS_PER_BP];
6600 int children = rvd->vdev_children;
6601 int c0 = spa_get_random(children);
6603 for (int c = 0; c < children; c++) {
6604 vd = rvd->vdev_child[(c0 + c) % children];
6605 if (vd->vdev_ms_array == 0 || vd->vdev_islog)
6607 svd[svdcount++] = vd;
6608 if (svdcount == SPA_DVAS_PER_BP)
6611 error = vdev_config_sync(svd, svdcount, txg, B_FALSE);
6613 error = vdev_config_sync(svd, svdcount, txg,
6616 error = vdev_config_sync(rvd->vdev_child,
6617 rvd->vdev_children, txg, B_FALSE);
6619 error = vdev_config_sync(rvd->vdev_child,
6620 rvd->vdev_children, txg, B_TRUE);
6624 spa->spa_last_synced_guid = rvd->vdev_guid;
6626 spa_config_exit(spa, SCL_STATE, FTAG);
6630 zio_suspend(spa, NULL);
6631 zio_resume_wait(spa);
6636 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY));
6639 callout_drain(&spa->spa_deadman_cycid);
6644 * Clear the dirty config list.
6646 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
6647 vdev_config_clean(vd);
6650 * Now that the new config has synced transactionally,
6651 * let it become visible to the config cache.
6653 if (spa->spa_config_syncing != NULL) {
6654 spa_config_set(spa, spa->spa_config_syncing);
6655 spa->spa_config_txg = txg;
6656 spa->spa_config_syncing = NULL;
6659 spa->spa_ubsync = spa->spa_uberblock;
6661 dsl_pool_sync_done(dp, txg);
6664 * Update usable space statistics.
6666 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
6667 vdev_sync_done(vd, txg);
6669 spa_update_dspace(spa);
6672 * It had better be the case that we didn't dirty anything
6673 * since vdev_config_sync().
6675 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
6676 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
6677 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
6679 spa->spa_sync_pass = 0;
6681 spa_config_exit(spa, SCL_CONFIG, FTAG);
6683 spa_handle_ignored_writes(spa);
6686 * If any async tasks have been requested, kick them off.
6688 spa_async_dispatch(spa);
6689 spa_async_dispatch_vd(spa);
6693 * Sync all pools. We don't want to hold the namespace lock across these
6694 * operations, so we take a reference on the spa_t and drop the lock during the
6698 spa_sync_allpools(void)
6701 mutex_enter(&spa_namespace_lock);
6702 while ((spa = spa_next(spa)) != NULL) {
6703 if (spa_state(spa) != POOL_STATE_ACTIVE ||
6704 !spa_writeable(spa) || spa_suspended(spa))
6706 spa_open_ref(spa, FTAG);
6707 mutex_exit(&spa_namespace_lock);
6708 txg_wait_synced(spa_get_dsl(spa), 0);
6709 mutex_enter(&spa_namespace_lock);
6710 spa_close(spa, FTAG);
6712 mutex_exit(&spa_namespace_lock);
6716 * ==========================================================================
6717 * Miscellaneous routines
6718 * ==========================================================================
6722 * Remove all pools in the system.
6730 * Remove all cached state. All pools should be closed now,
6731 * so every spa in the AVL tree should be unreferenced.
6733 mutex_enter(&spa_namespace_lock);
6734 while ((spa = spa_next(NULL)) != NULL) {
6736 * Stop async tasks. The async thread may need to detach
6737 * a device that's been replaced, which requires grabbing
6738 * spa_namespace_lock, so we must drop it here.
6740 spa_open_ref(spa, FTAG);
6741 mutex_exit(&spa_namespace_lock);
6742 spa_async_suspend(spa);
6743 mutex_enter(&spa_namespace_lock);
6744 spa_close(spa, FTAG);
6746 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
6748 spa_deactivate(spa);
6752 mutex_exit(&spa_namespace_lock);
6756 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
6761 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
6765 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
6766 vd = spa->spa_l2cache.sav_vdevs[i];
6767 if (vd->vdev_guid == guid)
6771 for (i = 0; i < spa->spa_spares.sav_count; i++) {
6772 vd = spa->spa_spares.sav_vdevs[i];
6773 if (vd->vdev_guid == guid)
6782 spa_upgrade(spa_t *spa, uint64_t version)
6784 ASSERT(spa_writeable(spa));
6786 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6789 * This should only be called for a non-faulted pool, and since a
6790 * future version would result in an unopenable pool, this shouldn't be
6793 ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version));
6794 ASSERT3U(version, >=, spa->spa_uberblock.ub_version);
6796 spa->spa_uberblock.ub_version = version;
6797 vdev_config_dirty(spa->spa_root_vdev);
6799 spa_config_exit(spa, SCL_ALL, FTAG);
6801 txg_wait_synced(spa_get_dsl(spa), 0);
6805 spa_has_spare(spa_t *spa, uint64_t guid)
6809 spa_aux_vdev_t *sav = &spa->spa_spares;
6811 for (i = 0; i < sav->sav_count; i++)
6812 if (sav->sav_vdevs[i]->vdev_guid == guid)
6815 for (i = 0; i < sav->sav_npending; i++) {
6816 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
6817 &spareguid) == 0 && spareguid == guid)
6825 * Check if a pool has an active shared spare device.
6826 * Note: reference count of an active spare is 2, as a spare and as a replace
6829 spa_has_active_shared_spare(spa_t *spa)
6833 spa_aux_vdev_t *sav = &spa->spa_spares;
6835 for (i = 0; i < sav->sav_count; i++) {
6836 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
6837 &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
6846 * Post a sysevent corresponding to the given event. The 'name' must be one of
6847 * the event definitions in sys/sysevent/eventdefs.h. The payload will be
6848 * filled in from the spa and (optionally) the vdev. This doesn't do anything
6849 * in the userland libzpool, as we don't want consumers to misinterpret ztest
6850 * or zdb as real changes.
6853 spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
6857 sysevent_attr_list_t *attr = NULL;
6858 sysevent_value_t value;
6861 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
6864 value.value_type = SE_DATA_TYPE_STRING;
6865 value.value.sv_string = spa_name(spa);
6866 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
6869 value.value_type = SE_DATA_TYPE_UINT64;
6870 value.value.sv_uint64 = spa_guid(spa);
6871 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
6875 value.value_type = SE_DATA_TYPE_UINT64;
6876 value.value.sv_uint64 = vd->vdev_guid;
6877 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
6881 if (vd->vdev_path) {
6882 value.value_type = SE_DATA_TYPE_STRING;
6883 value.value.sv_string = vd->vdev_path;
6884 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
6885 &value, SE_SLEEP) != 0)
6890 if (sysevent_attach_attributes(ev, attr) != 0)
6894 (void) log_sysevent(ev, SE_SLEEP, &eid);
6898 sysevent_free_attr(attr);