4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2011, 2014 by Delphix. All rights reserved.
25 * Copyright 2013 Nexenta Systems, Inc. All rights reserved.
26 * Copyright (c) 2013 Martin Matuska <mm@FreeBSD.org>. All rights reserved.
30 * SPA: Storage Pool Allocator
32 * This file contains all the routines used when modifying on-disk SPA state.
33 * This includes opening, importing, destroying, exporting a pool, and syncing a
37 #include <sys/zfs_context.h>
38 #include <sys/fm/fs/zfs.h>
39 #include <sys/spa_impl.h>
41 #include <sys/zio_checksum.h>
43 #include <sys/dmu_tx.h>
47 #include <sys/vdev_impl.h>
48 #include <sys/metaslab.h>
49 #include <sys/metaslab_impl.h>
50 #include <sys/uberblock_impl.h>
53 #include <sys/dmu_traverse.h>
54 #include <sys/dmu_objset.h>
55 #include <sys/unique.h>
56 #include <sys/dsl_pool.h>
57 #include <sys/dsl_dataset.h>
58 #include <sys/dsl_dir.h>
59 #include <sys/dsl_prop.h>
60 #include <sys/dsl_synctask.h>
61 #include <sys/fs/zfs.h>
63 #include <sys/callb.h>
64 #include <sys/spa_boot.h>
65 #include <sys/zfs_ioctl.h>
66 #include <sys/dsl_scan.h>
67 #include <sys/dmu_send.h>
68 #include <sys/dsl_destroy.h>
69 #include <sys/dsl_userhold.h>
70 #include <sys/zfeature.h>
72 #include <sys/trim_map.h>
75 #include <sys/callb.h>
76 #include <sys/cpupart.h>
81 #include "zfs_comutil.h"
83 /* Check hostid on import? */
84 static int check_hostid = 1;
86 SYSCTL_DECL(_vfs_zfs);
87 SYSCTL_INT(_vfs_zfs, OID_AUTO, check_hostid, CTLFLAG_RWTUN, &check_hostid, 0,
88 "Check hostid on import?");
91 * The interval, in seconds, at which failed configuration cache file writes
94 static int zfs_ccw_retry_interval = 300;
96 typedef enum zti_modes {
97 ZTI_MODE_FIXED, /* value is # of threads (min 1) */
98 ZTI_MODE_BATCH, /* cpu-intensive; value is ignored */
99 ZTI_MODE_NULL, /* don't create a taskq */
103 #define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) }
104 #define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 }
105 #define ZTI_NULL { ZTI_MODE_NULL, 0, 0 }
107 #define ZTI_N(n) ZTI_P(n, 1)
108 #define ZTI_ONE ZTI_N(1)
110 typedef struct zio_taskq_info {
111 zti_modes_t zti_mode;
116 static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
117 "issue", "issue_high", "intr", "intr_high"
121 * This table defines the taskq settings for each ZFS I/O type. When
122 * initializing a pool, we use this table to create an appropriately sized
123 * taskq. Some operations are low volume and therefore have a small, static
124 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE
125 * macros. Other operations process a large amount of data; the ZTI_BATCH
126 * macro causes us to create a taskq oriented for throughput. Some operations
127 * are so high frequency and short-lived that the taskq itself can become a a
128 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an
129 * additional degree of parallelism specified by the number of threads per-
130 * taskq and the number of taskqs; when dispatching an event in this case, the
131 * particular taskq is chosen at random.
133 * The different taskq priorities are to handle the different contexts (issue
134 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that
135 * need to be handled with minimum delay.
137 const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
138 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
139 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* NULL */
140 { ZTI_N(8), ZTI_NULL, ZTI_P(12, 8), ZTI_NULL }, /* READ */
141 { ZTI_BATCH, ZTI_N(5), ZTI_N(8), ZTI_N(5) }, /* WRITE */
142 { ZTI_P(12, 8), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FREE */
143 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* CLAIM */
144 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* IOCTL */
147 static void spa_sync_version(void *arg, dmu_tx_t *tx);
148 static void spa_sync_props(void *arg, dmu_tx_t *tx);
149 static boolean_t spa_has_active_shared_spare(spa_t *spa);
150 static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
151 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
153 static void spa_vdev_resilver_done(spa_t *spa);
155 uint_t zio_taskq_batch_pct = 75; /* 1 thread per cpu in pset */
157 id_t zio_taskq_psrset_bind = PS_NONE;
160 boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */
162 uint_t zio_taskq_basedc = 80; /* base duty cycle */
164 boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */
165 extern int zfs_sync_pass_deferred_free;
168 extern void spa_deadman(void *arg);
172 * This (illegal) pool name is used when temporarily importing a spa_t in order
173 * to get the vdev stats associated with the imported devices.
175 #define TRYIMPORT_NAME "$import"
178 * ==========================================================================
179 * SPA properties routines
180 * ==========================================================================
184 * Add a (source=src, propname=propval) list to an nvlist.
187 spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
188 uint64_t intval, zprop_source_t src)
190 const char *propname = zpool_prop_to_name(prop);
193 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
194 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
197 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
199 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
201 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
202 nvlist_free(propval);
206 * Get property values from the spa configuration.
209 spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
211 vdev_t *rvd = spa->spa_root_vdev;
212 dsl_pool_t *pool = spa->spa_dsl_pool;
216 uint64_t cap, version;
217 zprop_source_t src = ZPROP_SRC_NONE;
218 spa_config_dirent_t *dp;
220 ASSERT(MUTEX_HELD(&spa->spa_props_lock));
223 alloc = metaslab_class_get_alloc(spa_normal_class(spa));
224 size = metaslab_class_get_space(spa_normal_class(spa));
225 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
226 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
227 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
228 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
232 for (int c = 0; c < rvd->vdev_children; c++) {
233 vdev_t *tvd = rvd->vdev_child[c];
234 space += tvd->vdev_max_asize - tvd->vdev_asize;
236 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, space,
239 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
240 (spa_mode(spa) == FREAD), src);
242 cap = (size == 0) ? 0 : (alloc * 100 / size);
243 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
245 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
246 ddt_get_pool_dedup_ratio(spa), src);
248 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
249 rvd->vdev_state, src);
251 version = spa_version(spa);
252 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
253 src = ZPROP_SRC_DEFAULT;
255 src = ZPROP_SRC_LOCAL;
256 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
261 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
262 * when opening pools before this version freedir will be NULL.
264 if (pool->dp_free_dir != NULL) {
265 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
266 pool->dp_free_dir->dd_phys->dd_used_bytes, src);
268 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
272 if (pool->dp_leak_dir != NULL) {
273 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, NULL,
274 pool->dp_leak_dir->dd_phys->dd_used_bytes, src);
276 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED,
281 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
283 if (spa->spa_comment != NULL) {
284 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
288 if (spa->spa_root != NULL)
289 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
292 if ((dp = list_head(&spa->spa_config_list)) != NULL) {
293 if (dp->scd_path == NULL) {
294 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
295 "none", 0, ZPROP_SRC_LOCAL);
296 } else if (strcmp(dp->scd_path, spa_config_path) != 0) {
297 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
298 dp->scd_path, 0, ZPROP_SRC_LOCAL);
304 * Get zpool property values.
307 spa_prop_get(spa_t *spa, nvlist_t **nvp)
309 objset_t *mos = spa->spa_meta_objset;
314 VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
316 mutex_enter(&spa->spa_props_lock);
319 * Get properties from the spa config.
321 spa_prop_get_config(spa, nvp);
323 /* If no pool property object, no more prop to get. */
324 if (mos == NULL || spa->spa_pool_props_object == 0) {
325 mutex_exit(&spa->spa_props_lock);
330 * Get properties from the MOS pool property object.
332 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
333 (err = zap_cursor_retrieve(&zc, &za)) == 0;
334 zap_cursor_advance(&zc)) {
337 zprop_source_t src = ZPROP_SRC_DEFAULT;
340 if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
343 switch (za.za_integer_length) {
345 /* integer property */
346 if (za.za_first_integer !=
347 zpool_prop_default_numeric(prop))
348 src = ZPROP_SRC_LOCAL;
350 if (prop == ZPOOL_PROP_BOOTFS) {
352 dsl_dataset_t *ds = NULL;
354 dp = spa_get_dsl(spa);
355 dsl_pool_config_enter(dp, FTAG);
356 if (err = dsl_dataset_hold_obj(dp,
357 za.za_first_integer, FTAG, &ds)) {
358 dsl_pool_config_exit(dp, FTAG);
363 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
365 dsl_dataset_name(ds, strval);
366 dsl_dataset_rele(ds, FTAG);
367 dsl_pool_config_exit(dp, FTAG);
370 intval = za.za_first_integer;
373 spa_prop_add_list(*nvp, prop, strval, intval, src);
377 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
382 /* string property */
383 strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
384 err = zap_lookup(mos, spa->spa_pool_props_object,
385 za.za_name, 1, za.za_num_integers, strval);
387 kmem_free(strval, za.za_num_integers);
390 spa_prop_add_list(*nvp, prop, strval, 0, src);
391 kmem_free(strval, za.za_num_integers);
398 zap_cursor_fini(&zc);
399 mutex_exit(&spa->spa_props_lock);
401 if (err && err != ENOENT) {
411 * Validate the given pool properties nvlist and modify the list
412 * for the property values to be set.
415 spa_prop_validate(spa_t *spa, nvlist_t *props)
418 int error = 0, reset_bootfs = 0;
420 boolean_t has_feature = B_FALSE;
423 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
425 char *strval, *slash, *check, *fname;
426 const char *propname = nvpair_name(elem);
427 zpool_prop_t prop = zpool_name_to_prop(propname);
431 if (!zpool_prop_feature(propname)) {
432 error = SET_ERROR(EINVAL);
437 * Sanitize the input.
439 if (nvpair_type(elem) != DATA_TYPE_UINT64) {
440 error = SET_ERROR(EINVAL);
444 if (nvpair_value_uint64(elem, &intval) != 0) {
445 error = SET_ERROR(EINVAL);
450 error = SET_ERROR(EINVAL);
454 fname = strchr(propname, '@') + 1;
455 if (zfeature_lookup_name(fname, NULL) != 0) {
456 error = SET_ERROR(EINVAL);
460 has_feature = B_TRUE;
463 case ZPOOL_PROP_VERSION:
464 error = nvpair_value_uint64(elem, &intval);
466 (intval < spa_version(spa) ||
467 intval > SPA_VERSION_BEFORE_FEATURES ||
469 error = SET_ERROR(EINVAL);
472 case ZPOOL_PROP_DELEGATION:
473 case ZPOOL_PROP_AUTOREPLACE:
474 case ZPOOL_PROP_LISTSNAPS:
475 case ZPOOL_PROP_AUTOEXPAND:
476 error = nvpair_value_uint64(elem, &intval);
477 if (!error && intval > 1)
478 error = SET_ERROR(EINVAL);
481 case ZPOOL_PROP_BOOTFS:
483 * If the pool version is less than SPA_VERSION_BOOTFS,
484 * or the pool is still being created (version == 0),
485 * the bootfs property cannot be set.
487 if (spa_version(spa) < SPA_VERSION_BOOTFS) {
488 error = SET_ERROR(ENOTSUP);
493 * Make sure the vdev config is bootable
495 if (!vdev_is_bootable(spa->spa_root_vdev)) {
496 error = SET_ERROR(ENOTSUP);
502 error = nvpair_value_string(elem, &strval);
508 if (strval == NULL || strval[0] == '\0') {
509 objnum = zpool_prop_default_numeric(
514 if (error = dmu_objset_hold(strval, FTAG, &os))
517 /* Must be ZPL and not gzip compressed. */
519 if (dmu_objset_type(os) != DMU_OST_ZFS) {
520 error = SET_ERROR(ENOTSUP);
522 dsl_prop_get_int_ds(dmu_objset_ds(os),
523 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
525 !BOOTFS_COMPRESS_VALID(compress)) {
526 error = SET_ERROR(ENOTSUP);
528 objnum = dmu_objset_id(os);
530 dmu_objset_rele(os, FTAG);
534 case ZPOOL_PROP_FAILUREMODE:
535 error = nvpair_value_uint64(elem, &intval);
536 if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
537 intval > ZIO_FAILURE_MODE_PANIC))
538 error = SET_ERROR(EINVAL);
541 * This is a special case which only occurs when
542 * the pool has completely failed. This allows
543 * the user to change the in-core failmode property
544 * without syncing it out to disk (I/Os might
545 * currently be blocked). We do this by returning
546 * EIO to the caller (spa_prop_set) to trick it
547 * into thinking we encountered a property validation
550 if (!error && spa_suspended(spa)) {
551 spa->spa_failmode = intval;
552 error = SET_ERROR(EIO);
556 case ZPOOL_PROP_CACHEFILE:
557 if ((error = nvpair_value_string(elem, &strval)) != 0)
560 if (strval[0] == '\0')
563 if (strcmp(strval, "none") == 0)
566 if (strval[0] != '/') {
567 error = SET_ERROR(EINVAL);
571 slash = strrchr(strval, '/');
572 ASSERT(slash != NULL);
574 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
575 strcmp(slash, "/..") == 0)
576 error = SET_ERROR(EINVAL);
579 case ZPOOL_PROP_COMMENT:
580 if ((error = nvpair_value_string(elem, &strval)) != 0)
582 for (check = strval; *check != '\0'; check++) {
584 * The kernel doesn't have an easy isprint()
585 * check. For this kernel check, we merely
586 * check ASCII apart from DEL. Fix this if
587 * there is an easy-to-use kernel isprint().
589 if (*check >= 0x7f) {
590 error = SET_ERROR(EINVAL);
595 if (strlen(strval) > ZPROP_MAX_COMMENT)
599 case ZPOOL_PROP_DEDUPDITTO:
600 if (spa_version(spa) < SPA_VERSION_DEDUP)
601 error = SET_ERROR(ENOTSUP);
603 error = nvpair_value_uint64(elem, &intval);
605 intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
606 error = SET_ERROR(EINVAL);
614 if (!error && reset_bootfs) {
615 error = nvlist_remove(props,
616 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
619 error = nvlist_add_uint64(props,
620 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
628 spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
631 spa_config_dirent_t *dp;
633 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
637 dp = kmem_alloc(sizeof (spa_config_dirent_t),
640 if (cachefile[0] == '\0')
641 dp->scd_path = spa_strdup(spa_config_path);
642 else if (strcmp(cachefile, "none") == 0)
645 dp->scd_path = spa_strdup(cachefile);
647 list_insert_head(&spa->spa_config_list, dp);
649 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
653 spa_prop_set(spa_t *spa, nvlist_t *nvp)
656 nvpair_t *elem = NULL;
657 boolean_t need_sync = B_FALSE;
659 if ((error = spa_prop_validate(spa, nvp)) != 0)
662 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
663 zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem));
665 if (prop == ZPOOL_PROP_CACHEFILE ||
666 prop == ZPOOL_PROP_ALTROOT ||
667 prop == ZPOOL_PROP_READONLY)
670 if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) {
673 if (prop == ZPOOL_PROP_VERSION) {
674 VERIFY(nvpair_value_uint64(elem, &ver) == 0);
676 ASSERT(zpool_prop_feature(nvpair_name(elem)));
677 ver = SPA_VERSION_FEATURES;
681 /* Save time if the version is already set. */
682 if (ver == spa_version(spa))
686 * In addition to the pool directory object, we might
687 * create the pool properties object, the features for
688 * read object, the features for write object, or the
689 * feature descriptions object.
691 error = dsl_sync_task(spa->spa_name, NULL,
692 spa_sync_version, &ver, 6);
703 return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props,
711 * If the bootfs property value is dsobj, clear it.
714 spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
716 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
717 VERIFY(zap_remove(spa->spa_meta_objset,
718 spa->spa_pool_props_object,
719 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
726 spa_change_guid_check(void *arg, dmu_tx_t *tx)
728 uint64_t *newguid = arg;
729 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
730 vdev_t *rvd = spa->spa_root_vdev;
733 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
734 vdev_state = rvd->vdev_state;
735 spa_config_exit(spa, SCL_STATE, FTAG);
737 if (vdev_state != VDEV_STATE_HEALTHY)
738 return (SET_ERROR(ENXIO));
740 ASSERT3U(spa_guid(spa), !=, *newguid);
746 spa_change_guid_sync(void *arg, dmu_tx_t *tx)
748 uint64_t *newguid = arg;
749 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
751 vdev_t *rvd = spa->spa_root_vdev;
753 oldguid = spa_guid(spa);
755 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
756 rvd->vdev_guid = *newguid;
757 rvd->vdev_guid_sum += (*newguid - oldguid);
758 vdev_config_dirty(rvd);
759 spa_config_exit(spa, SCL_STATE, FTAG);
761 spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu",
766 * Change the GUID for the pool. This is done so that we can later
767 * re-import a pool built from a clone of our own vdevs. We will modify
768 * the root vdev's guid, our own pool guid, and then mark all of our
769 * vdevs dirty. Note that we must make sure that all our vdevs are
770 * online when we do this, or else any vdevs that weren't present
771 * would be orphaned from our pool. We are also going to issue a
772 * sysevent to update any watchers.
775 spa_change_guid(spa_t *spa)
780 mutex_enter(&spa->spa_vdev_top_lock);
781 mutex_enter(&spa_namespace_lock);
782 guid = spa_generate_guid(NULL);
784 error = dsl_sync_task(spa->spa_name, spa_change_guid_check,
785 spa_change_guid_sync, &guid, 5);
788 spa_config_sync(spa, B_FALSE, B_TRUE);
789 spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID);
792 mutex_exit(&spa_namespace_lock);
793 mutex_exit(&spa->spa_vdev_top_lock);
799 * ==========================================================================
800 * SPA state manipulation (open/create/destroy/import/export)
801 * ==========================================================================
805 spa_error_entry_compare(const void *a, const void *b)
807 spa_error_entry_t *sa = (spa_error_entry_t *)a;
808 spa_error_entry_t *sb = (spa_error_entry_t *)b;
811 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
812 sizeof (zbookmark_phys_t));
823 * Utility function which retrieves copies of the current logs and
824 * re-initializes them in the process.
827 spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
829 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
831 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
832 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
834 avl_create(&spa->spa_errlist_scrub,
835 spa_error_entry_compare, sizeof (spa_error_entry_t),
836 offsetof(spa_error_entry_t, se_avl));
837 avl_create(&spa->spa_errlist_last,
838 spa_error_entry_compare, sizeof (spa_error_entry_t),
839 offsetof(spa_error_entry_t, se_avl));
843 spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
845 const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
846 enum zti_modes mode = ztip->zti_mode;
847 uint_t value = ztip->zti_value;
848 uint_t count = ztip->zti_count;
849 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
852 boolean_t batch = B_FALSE;
854 if (mode == ZTI_MODE_NULL) {
856 tqs->stqs_taskq = NULL;
860 ASSERT3U(count, >, 0);
862 tqs->stqs_count = count;
863 tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP);
867 ASSERT3U(value, >=, 1);
868 value = MAX(value, 1);
873 flags |= TASKQ_THREADS_CPU_PCT;
874 value = zio_taskq_batch_pct;
878 panic("unrecognized mode for %s_%s taskq (%u:%u) in "
880 zio_type_name[t], zio_taskq_types[q], mode, value);
884 for (uint_t i = 0; i < count; i++) {
888 (void) snprintf(name, sizeof (name), "%s_%s_%u",
889 zio_type_name[t], zio_taskq_types[q], i);
891 (void) snprintf(name, sizeof (name), "%s_%s",
892 zio_type_name[t], zio_taskq_types[q]);
896 if (zio_taskq_sysdc && spa->spa_proc != &p0) {
898 flags |= TASKQ_DC_BATCH;
900 tq = taskq_create_sysdc(name, value, 50, INT_MAX,
901 spa->spa_proc, zio_taskq_basedc, flags);
904 pri_t pri = maxclsyspri;
906 * The write issue taskq can be extremely CPU
907 * intensive. Run it at slightly lower priority
908 * than the other taskqs.
910 if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE)
913 tq = taskq_create_proc(name, value, pri, 50,
914 INT_MAX, spa->spa_proc, flags);
919 tqs->stqs_taskq[i] = tq;
924 spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
926 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
928 if (tqs->stqs_taskq == NULL) {
929 ASSERT0(tqs->stqs_count);
933 for (uint_t i = 0; i < tqs->stqs_count; i++) {
934 ASSERT3P(tqs->stqs_taskq[i], !=, NULL);
935 taskq_destroy(tqs->stqs_taskq[i]);
938 kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *));
939 tqs->stqs_taskq = NULL;
943 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
944 * Note that a type may have multiple discrete taskqs to avoid lock contention
945 * on the taskq itself. In that case we choose which taskq at random by using
946 * the low bits of gethrtime().
949 spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q,
950 task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent)
952 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
955 ASSERT3P(tqs->stqs_taskq, !=, NULL);
956 ASSERT3U(tqs->stqs_count, !=, 0);
958 if (tqs->stqs_count == 1) {
959 tq = tqs->stqs_taskq[0];
962 tq = tqs->stqs_taskq[cpu_ticks() % tqs->stqs_count];
964 tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count];
968 taskq_dispatch_ent(tq, func, arg, flags, ent);
972 spa_create_zio_taskqs(spa_t *spa)
974 for (int t = 0; t < ZIO_TYPES; t++) {
975 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
976 spa_taskqs_init(spa, t, q);
984 spa_thread(void *arg)
989 user_t *pu = PTOU(curproc);
991 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
994 ASSERT(curproc != &p0);
995 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
996 "zpool-%s", spa->spa_name);
997 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
1000 /* bind this thread to the requested psrset */
1001 if (zio_taskq_psrset_bind != PS_NONE) {
1003 mutex_enter(&cpu_lock);
1004 mutex_enter(&pidlock);
1005 mutex_enter(&curproc->p_lock);
1007 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
1008 0, NULL, NULL) == 0) {
1009 curthread->t_bind_pset = zio_taskq_psrset_bind;
1012 "Couldn't bind process for zfs pool \"%s\" to "
1013 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
1016 mutex_exit(&curproc->p_lock);
1017 mutex_exit(&pidlock);
1018 mutex_exit(&cpu_lock);
1024 if (zio_taskq_sysdc) {
1025 sysdc_thread_enter(curthread, 100, 0);
1029 spa->spa_proc = curproc;
1030 spa->spa_did = curthread->t_did;
1032 spa_create_zio_taskqs(spa);
1034 mutex_enter(&spa->spa_proc_lock);
1035 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
1037 spa->spa_proc_state = SPA_PROC_ACTIVE;
1038 cv_broadcast(&spa->spa_proc_cv);
1040 CALLB_CPR_SAFE_BEGIN(&cprinfo);
1041 while (spa->spa_proc_state == SPA_PROC_ACTIVE)
1042 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1043 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
1045 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
1046 spa->spa_proc_state = SPA_PROC_GONE;
1047 spa->spa_proc = &p0;
1048 cv_broadcast(&spa->spa_proc_cv);
1049 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */
1051 mutex_enter(&curproc->p_lock);
1054 #endif /* SPA_PROCESS */
1058 * Activate an uninitialized pool.
1061 spa_activate(spa_t *spa, int mode)
1063 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
1065 spa->spa_state = POOL_STATE_ACTIVE;
1066 spa->spa_mode = mode;
1068 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
1069 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
1071 /* Try to create a covering process */
1072 mutex_enter(&spa->spa_proc_lock);
1073 ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
1074 ASSERT(spa->spa_proc == &p0);
1078 /* Only create a process if we're going to be around a while. */
1079 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
1080 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
1082 spa->spa_proc_state = SPA_PROC_CREATED;
1083 while (spa->spa_proc_state == SPA_PROC_CREATED) {
1084 cv_wait(&spa->spa_proc_cv,
1085 &spa->spa_proc_lock);
1087 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1088 ASSERT(spa->spa_proc != &p0);
1089 ASSERT(spa->spa_did != 0);
1093 "Couldn't create process for zfs pool \"%s\"\n",
1098 #endif /* SPA_PROCESS */
1099 mutex_exit(&spa->spa_proc_lock);
1101 /* If we didn't create a process, we need to create our taskqs. */
1102 ASSERT(spa->spa_proc == &p0);
1103 if (spa->spa_proc == &p0) {
1104 spa_create_zio_taskqs(spa);
1108 * Start TRIM thread.
1110 trim_thread_create(spa);
1112 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
1113 offsetof(vdev_t, vdev_config_dirty_node));
1114 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
1115 offsetof(vdev_t, vdev_state_dirty_node));
1117 txg_list_create(&spa->spa_vdev_txg_list,
1118 offsetof(struct vdev, vdev_txg_node));
1120 avl_create(&spa->spa_errlist_scrub,
1121 spa_error_entry_compare, sizeof (spa_error_entry_t),
1122 offsetof(spa_error_entry_t, se_avl));
1123 avl_create(&spa->spa_errlist_last,
1124 spa_error_entry_compare, sizeof (spa_error_entry_t),
1125 offsetof(spa_error_entry_t, se_avl));
1129 * Opposite of spa_activate().
1132 spa_deactivate(spa_t *spa)
1134 ASSERT(spa->spa_sync_on == B_FALSE);
1135 ASSERT(spa->spa_dsl_pool == NULL);
1136 ASSERT(spa->spa_root_vdev == NULL);
1137 ASSERT(spa->spa_async_zio_root == NULL);
1138 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
1141 * Stop TRIM thread in case spa_unload() wasn't called directly
1142 * before spa_deactivate().
1144 trim_thread_destroy(spa);
1146 txg_list_destroy(&spa->spa_vdev_txg_list);
1148 list_destroy(&spa->spa_config_dirty_list);
1149 list_destroy(&spa->spa_state_dirty_list);
1151 for (int t = 0; t < ZIO_TYPES; t++) {
1152 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1153 spa_taskqs_fini(spa, t, q);
1157 metaslab_class_destroy(spa->spa_normal_class);
1158 spa->spa_normal_class = NULL;
1160 metaslab_class_destroy(spa->spa_log_class);
1161 spa->spa_log_class = NULL;
1164 * If this was part of an import or the open otherwise failed, we may
1165 * still have errors left in the queues. Empty them just in case.
1167 spa_errlog_drain(spa);
1169 avl_destroy(&spa->spa_errlist_scrub);
1170 avl_destroy(&spa->spa_errlist_last);
1172 spa->spa_state = POOL_STATE_UNINITIALIZED;
1174 mutex_enter(&spa->spa_proc_lock);
1175 if (spa->spa_proc_state != SPA_PROC_NONE) {
1176 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1177 spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1178 cv_broadcast(&spa->spa_proc_cv);
1179 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1180 ASSERT(spa->spa_proc != &p0);
1181 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1183 ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1184 spa->spa_proc_state = SPA_PROC_NONE;
1186 ASSERT(spa->spa_proc == &p0);
1187 mutex_exit(&spa->spa_proc_lock);
1191 * We want to make sure spa_thread() has actually exited the ZFS
1192 * module, so that the module can't be unloaded out from underneath
1195 if (spa->spa_did != 0) {
1196 thread_join(spa->spa_did);
1199 #endif /* SPA_PROCESS */
1203 * Verify a pool configuration, and construct the vdev tree appropriately. This
1204 * will create all the necessary vdevs in the appropriate layout, with each vdev
1205 * in the CLOSED state. This will prep the pool before open/creation/import.
1206 * All vdev validation is done by the vdev_alloc() routine.
1209 spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1210 uint_t id, int atype)
1216 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1219 if ((*vdp)->vdev_ops->vdev_op_leaf)
1222 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1225 if (error == ENOENT)
1231 return (SET_ERROR(EINVAL));
1234 for (int c = 0; c < children; c++) {
1236 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1244 ASSERT(*vdp != NULL);
1250 * Opposite of spa_load().
1253 spa_unload(spa_t *spa)
1257 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1262 trim_thread_destroy(spa);
1267 spa_async_suspend(spa);
1272 if (spa->spa_sync_on) {
1273 txg_sync_stop(spa->spa_dsl_pool);
1274 spa->spa_sync_on = B_FALSE;
1278 * Wait for any outstanding async I/O to complete.
1280 if (spa->spa_async_zio_root != NULL) {
1281 (void) zio_wait(spa->spa_async_zio_root);
1282 spa->spa_async_zio_root = NULL;
1285 bpobj_close(&spa->spa_deferred_bpobj);
1287 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1292 if (spa->spa_root_vdev)
1293 vdev_free(spa->spa_root_vdev);
1294 ASSERT(spa->spa_root_vdev == NULL);
1297 * Close the dsl pool.
1299 if (spa->spa_dsl_pool) {
1300 dsl_pool_close(spa->spa_dsl_pool);
1301 spa->spa_dsl_pool = NULL;
1302 spa->spa_meta_objset = NULL;
1309 * Drop and purge level 2 cache
1311 spa_l2cache_drop(spa);
1313 for (i = 0; i < spa->spa_spares.sav_count; i++)
1314 vdev_free(spa->spa_spares.sav_vdevs[i]);
1315 if (spa->spa_spares.sav_vdevs) {
1316 kmem_free(spa->spa_spares.sav_vdevs,
1317 spa->spa_spares.sav_count * sizeof (void *));
1318 spa->spa_spares.sav_vdevs = NULL;
1320 if (spa->spa_spares.sav_config) {
1321 nvlist_free(spa->spa_spares.sav_config);
1322 spa->spa_spares.sav_config = NULL;
1324 spa->spa_spares.sav_count = 0;
1326 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1327 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1328 vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1330 if (spa->spa_l2cache.sav_vdevs) {
1331 kmem_free(spa->spa_l2cache.sav_vdevs,
1332 spa->spa_l2cache.sav_count * sizeof (void *));
1333 spa->spa_l2cache.sav_vdevs = NULL;
1335 if (spa->spa_l2cache.sav_config) {
1336 nvlist_free(spa->spa_l2cache.sav_config);
1337 spa->spa_l2cache.sav_config = NULL;
1339 spa->spa_l2cache.sav_count = 0;
1341 spa->spa_async_suspended = 0;
1343 if (spa->spa_comment != NULL) {
1344 spa_strfree(spa->spa_comment);
1345 spa->spa_comment = NULL;
1348 spa_config_exit(spa, SCL_ALL, FTAG);
1352 * Load (or re-load) the current list of vdevs describing the active spares for
1353 * this pool. When this is called, we have some form of basic information in
1354 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1355 * then re-generate a more complete list including status information.
1358 spa_load_spares(spa_t *spa)
1365 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1368 * First, close and free any existing spare vdevs.
1370 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1371 vd = spa->spa_spares.sav_vdevs[i];
1373 /* Undo the call to spa_activate() below */
1374 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1375 B_FALSE)) != NULL && tvd->vdev_isspare)
1376 spa_spare_remove(tvd);
1381 if (spa->spa_spares.sav_vdevs)
1382 kmem_free(spa->spa_spares.sav_vdevs,
1383 spa->spa_spares.sav_count * sizeof (void *));
1385 if (spa->spa_spares.sav_config == NULL)
1388 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1389 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1391 spa->spa_spares.sav_count = (int)nspares;
1392 spa->spa_spares.sav_vdevs = NULL;
1398 * Construct the array of vdevs, opening them to get status in the
1399 * process. For each spare, there is potentially two different vdev_t
1400 * structures associated with it: one in the list of spares (used only
1401 * for basic validation purposes) and one in the active vdev
1402 * configuration (if it's spared in). During this phase we open and
1403 * validate each vdev on the spare list. If the vdev also exists in the
1404 * active configuration, then we also mark this vdev as an active spare.
1406 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1408 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1409 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1410 VDEV_ALLOC_SPARE) == 0);
1413 spa->spa_spares.sav_vdevs[i] = vd;
1415 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1416 B_FALSE)) != NULL) {
1417 if (!tvd->vdev_isspare)
1421 * We only mark the spare active if we were successfully
1422 * able to load the vdev. Otherwise, importing a pool
1423 * with a bad active spare would result in strange
1424 * behavior, because multiple pool would think the spare
1425 * is actively in use.
1427 * There is a vulnerability here to an equally bizarre
1428 * circumstance, where a dead active spare is later
1429 * brought back to life (onlined or otherwise). Given
1430 * the rarity of this scenario, and the extra complexity
1431 * it adds, we ignore the possibility.
1433 if (!vdev_is_dead(tvd))
1434 spa_spare_activate(tvd);
1438 vd->vdev_aux = &spa->spa_spares;
1440 if (vdev_open(vd) != 0)
1443 if (vdev_validate_aux(vd) == 0)
1448 * Recompute the stashed list of spares, with status information
1451 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1452 DATA_TYPE_NVLIST_ARRAY) == 0);
1454 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1456 for (i = 0; i < spa->spa_spares.sav_count; i++)
1457 spares[i] = vdev_config_generate(spa,
1458 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1459 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1460 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1461 for (i = 0; i < spa->spa_spares.sav_count; i++)
1462 nvlist_free(spares[i]);
1463 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1467 * Load (or re-load) the current list of vdevs describing the active l2cache for
1468 * this pool. When this is called, we have some form of basic information in
1469 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1470 * then re-generate a more complete list including status information.
1471 * Devices which are already active have their details maintained, and are
1475 spa_load_l2cache(spa_t *spa)
1479 int i, j, oldnvdevs;
1481 vdev_t *vd, **oldvdevs, **newvdevs;
1482 spa_aux_vdev_t *sav = &spa->spa_l2cache;
1484 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1486 if (sav->sav_config != NULL) {
1487 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1488 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1489 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1495 oldvdevs = sav->sav_vdevs;
1496 oldnvdevs = sav->sav_count;
1497 sav->sav_vdevs = NULL;
1501 * Process new nvlist of vdevs.
1503 for (i = 0; i < nl2cache; i++) {
1504 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1508 for (j = 0; j < oldnvdevs; j++) {
1510 if (vd != NULL && guid == vd->vdev_guid) {
1512 * Retain previous vdev for add/remove ops.
1520 if (newvdevs[i] == NULL) {
1524 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1525 VDEV_ALLOC_L2CACHE) == 0);
1530 * Commit this vdev as an l2cache device,
1531 * even if it fails to open.
1533 spa_l2cache_add(vd);
1538 spa_l2cache_activate(vd);
1540 if (vdev_open(vd) != 0)
1543 (void) vdev_validate_aux(vd);
1545 if (!vdev_is_dead(vd))
1546 l2arc_add_vdev(spa, vd);
1551 * Purge vdevs that were dropped
1553 for (i = 0; i < oldnvdevs; i++) {
1558 ASSERT(vd->vdev_isl2cache);
1560 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1561 pool != 0ULL && l2arc_vdev_present(vd))
1562 l2arc_remove_vdev(vd);
1563 vdev_clear_stats(vd);
1569 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1571 if (sav->sav_config == NULL)
1574 sav->sav_vdevs = newvdevs;
1575 sav->sav_count = (int)nl2cache;
1578 * Recompute the stashed list of l2cache devices, with status
1579 * information this time.
1581 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1582 DATA_TYPE_NVLIST_ARRAY) == 0);
1584 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1585 for (i = 0; i < sav->sav_count; i++)
1586 l2cache[i] = vdev_config_generate(spa,
1587 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1588 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1589 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1591 for (i = 0; i < sav->sav_count; i++)
1592 nvlist_free(l2cache[i]);
1594 kmem_free(l2cache, sav->sav_count * sizeof (void *));
1598 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1601 char *packed = NULL;
1606 error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db);
1609 nvsize = *(uint64_t *)db->db_data;
1610 dmu_buf_rele(db, FTAG);
1612 packed = kmem_alloc(nvsize, KM_SLEEP);
1613 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1616 error = nvlist_unpack(packed, nvsize, value, 0);
1617 kmem_free(packed, nvsize);
1623 * Checks to see if the given vdev could not be opened, in which case we post a
1624 * sysevent to notify the autoreplace code that the device has been removed.
1627 spa_check_removed(vdev_t *vd)
1629 for (int c = 0; c < vd->vdev_children; c++)
1630 spa_check_removed(vd->vdev_child[c]);
1632 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) &&
1634 zfs_post_autoreplace(vd->vdev_spa, vd);
1635 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1640 * Validate the current config against the MOS config
1643 spa_config_valid(spa_t *spa, nvlist_t *config)
1645 vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1648 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1650 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1651 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1653 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1656 * If we're doing a normal import, then build up any additional
1657 * diagnostic information about missing devices in this config.
1658 * We'll pass this up to the user for further processing.
1660 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1661 nvlist_t **child, *nv;
1664 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1666 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1668 for (int c = 0; c < rvd->vdev_children; c++) {
1669 vdev_t *tvd = rvd->vdev_child[c];
1670 vdev_t *mtvd = mrvd->vdev_child[c];
1672 if (tvd->vdev_ops == &vdev_missing_ops &&
1673 mtvd->vdev_ops != &vdev_missing_ops &&
1675 child[idx++] = vdev_config_generate(spa, mtvd,
1680 VERIFY(nvlist_add_nvlist_array(nv,
1681 ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1682 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1683 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1685 for (int i = 0; i < idx; i++)
1686 nvlist_free(child[i]);
1689 kmem_free(child, rvd->vdev_children * sizeof (char **));
1693 * Compare the root vdev tree with the information we have
1694 * from the MOS config (mrvd). Check each top-level vdev
1695 * with the corresponding MOS config top-level (mtvd).
1697 for (int c = 0; c < rvd->vdev_children; c++) {
1698 vdev_t *tvd = rvd->vdev_child[c];
1699 vdev_t *mtvd = mrvd->vdev_child[c];
1702 * Resolve any "missing" vdevs in the current configuration.
1703 * If we find that the MOS config has more accurate information
1704 * about the top-level vdev then use that vdev instead.
1706 if (tvd->vdev_ops == &vdev_missing_ops &&
1707 mtvd->vdev_ops != &vdev_missing_ops) {
1709 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1713 * Device specific actions.
1715 if (mtvd->vdev_islog) {
1716 spa_set_log_state(spa, SPA_LOG_CLEAR);
1719 * XXX - once we have 'readonly' pool
1720 * support we should be able to handle
1721 * missing data devices by transitioning
1722 * the pool to readonly.
1728 * Swap the missing vdev with the data we were
1729 * able to obtain from the MOS config.
1731 vdev_remove_child(rvd, tvd);
1732 vdev_remove_child(mrvd, mtvd);
1734 vdev_add_child(rvd, mtvd);
1735 vdev_add_child(mrvd, tvd);
1737 spa_config_exit(spa, SCL_ALL, FTAG);
1739 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1742 } else if (mtvd->vdev_islog) {
1744 * Load the slog device's state from the MOS config
1745 * since it's possible that the label does not
1746 * contain the most up-to-date information.
1748 vdev_load_log_state(tvd, mtvd);
1753 spa_config_exit(spa, SCL_ALL, FTAG);
1756 * Ensure we were able to validate the config.
1758 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1762 * Check for missing log devices
1765 spa_check_logs(spa_t *spa)
1767 boolean_t rv = B_FALSE;
1769 switch (spa->spa_log_state) {
1770 case SPA_LOG_MISSING:
1771 /* need to recheck in case slog has been restored */
1772 case SPA_LOG_UNKNOWN:
1773 rv = (dmu_objset_find(spa->spa_name, zil_check_log_chain,
1774 NULL, DS_FIND_CHILDREN) != 0);
1776 spa_set_log_state(spa, SPA_LOG_MISSING);
1783 spa_passivate_log(spa_t *spa)
1785 vdev_t *rvd = spa->spa_root_vdev;
1786 boolean_t slog_found = B_FALSE;
1788 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1790 if (!spa_has_slogs(spa))
1793 for (int c = 0; c < rvd->vdev_children; c++) {
1794 vdev_t *tvd = rvd->vdev_child[c];
1795 metaslab_group_t *mg = tvd->vdev_mg;
1797 if (tvd->vdev_islog) {
1798 metaslab_group_passivate(mg);
1799 slog_found = B_TRUE;
1803 return (slog_found);
1807 spa_activate_log(spa_t *spa)
1809 vdev_t *rvd = spa->spa_root_vdev;
1811 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1813 for (int c = 0; c < rvd->vdev_children; c++) {
1814 vdev_t *tvd = rvd->vdev_child[c];
1815 metaslab_group_t *mg = tvd->vdev_mg;
1817 if (tvd->vdev_islog)
1818 metaslab_group_activate(mg);
1823 spa_offline_log(spa_t *spa)
1827 error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1828 NULL, DS_FIND_CHILDREN);
1831 * We successfully offlined the log device, sync out the
1832 * current txg so that the "stubby" block can be removed
1835 txg_wait_synced(spa->spa_dsl_pool, 0);
1841 spa_aux_check_removed(spa_aux_vdev_t *sav)
1845 for (i = 0; i < sav->sav_count; i++)
1846 spa_check_removed(sav->sav_vdevs[i]);
1850 spa_claim_notify(zio_t *zio)
1852 spa_t *spa = zio->io_spa;
1857 mutex_enter(&spa->spa_props_lock); /* any mutex will do */
1858 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1859 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1860 mutex_exit(&spa->spa_props_lock);
1863 typedef struct spa_load_error {
1864 uint64_t sle_meta_count;
1865 uint64_t sle_data_count;
1869 spa_load_verify_done(zio_t *zio)
1871 blkptr_t *bp = zio->io_bp;
1872 spa_load_error_t *sle = zio->io_private;
1873 dmu_object_type_t type = BP_GET_TYPE(bp);
1874 int error = zio->io_error;
1877 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1878 type != DMU_OT_INTENT_LOG)
1879 atomic_add_64(&sle->sle_meta_count, 1);
1881 atomic_add_64(&sle->sle_data_count, 1);
1883 zio_data_buf_free(zio->io_data, zio->io_size);
1888 spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1889 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
1891 if (!BP_IS_HOLE(bp) && !BP_IS_EMBEDDED(bp)) {
1893 size_t size = BP_GET_PSIZE(bp);
1894 void *data = zio_data_buf_alloc(size);
1896 zio_nowait(zio_read(rio, spa, bp, data, size,
1897 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1898 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1899 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1905 spa_load_verify(spa_t *spa)
1908 spa_load_error_t sle = { 0 };
1909 zpool_rewind_policy_t policy;
1910 boolean_t verify_ok = B_FALSE;
1913 zpool_get_rewind_policy(spa->spa_config, &policy);
1915 if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1918 rio = zio_root(spa, NULL, &sle,
1919 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1921 error = traverse_pool(spa, spa->spa_verify_min_txg,
1922 TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio);
1924 (void) zio_wait(rio);
1926 spa->spa_load_meta_errors = sle.sle_meta_count;
1927 spa->spa_load_data_errors = sle.sle_data_count;
1929 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1930 sle.sle_data_count <= policy.zrp_maxdata) {
1934 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1935 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1937 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1938 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1939 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1940 VERIFY(nvlist_add_int64(spa->spa_load_info,
1941 ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1942 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1943 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1945 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1949 if (error != ENXIO && error != EIO)
1950 error = SET_ERROR(EIO);
1954 return (verify_ok ? 0 : EIO);
1958 * Find a value in the pool props object.
1961 spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1963 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1964 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1968 * Find a value in the pool directory object.
1971 spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1973 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1974 name, sizeof (uint64_t), 1, val));
1978 spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1980 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1985 * Fix up config after a partly-completed split. This is done with the
1986 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
1987 * pool have that entry in their config, but only the splitting one contains
1988 * a list of all the guids of the vdevs that are being split off.
1990 * This function determines what to do with that list: either rejoin
1991 * all the disks to the pool, or complete the splitting process. To attempt
1992 * the rejoin, each disk that is offlined is marked online again, and
1993 * we do a reopen() call. If the vdev label for every disk that was
1994 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1995 * then we call vdev_split() on each disk, and complete the split.
1997 * Otherwise we leave the config alone, with all the vdevs in place in
1998 * the original pool.
2001 spa_try_repair(spa_t *spa, nvlist_t *config)
2008 boolean_t attempt_reopen;
2010 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
2013 /* check that the config is complete */
2014 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
2015 &glist, &gcount) != 0)
2018 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
2020 /* attempt to online all the vdevs & validate */
2021 attempt_reopen = B_TRUE;
2022 for (i = 0; i < gcount; i++) {
2023 if (glist[i] == 0) /* vdev is hole */
2026 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
2027 if (vd[i] == NULL) {
2029 * Don't bother attempting to reopen the disks;
2030 * just do the split.
2032 attempt_reopen = B_FALSE;
2034 /* attempt to re-online it */
2035 vd[i]->vdev_offline = B_FALSE;
2039 if (attempt_reopen) {
2040 vdev_reopen(spa->spa_root_vdev);
2042 /* check each device to see what state it's in */
2043 for (extracted = 0, i = 0; i < gcount; i++) {
2044 if (vd[i] != NULL &&
2045 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
2052 * If every disk has been moved to the new pool, or if we never
2053 * even attempted to look at them, then we split them off for
2056 if (!attempt_reopen || gcount == extracted) {
2057 for (i = 0; i < gcount; i++)
2060 vdev_reopen(spa->spa_root_vdev);
2063 kmem_free(vd, gcount * sizeof (vdev_t *));
2067 spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
2068 boolean_t mosconfig)
2070 nvlist_t *config = spa->spa_config;
2071 char *ereport = FM_EREPORT_ZFS_POOL;
2077 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
2078 return (SET_ERROR(EINVAL));
2080 ASSERT(spa->spa_comment == NULL);
2081 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
2082 spa->spa_comment = spa_strdup(comment);
2085 * Versioning wasn't explicitly added to the label until later, so if
2086 * it's not present treat it as the initial version.
2088 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
2089 &spa->spa_ubsync.ub_version) != 0)
2090 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
2092 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
2093 &spa->spa_config_txg);
2095 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
2096 spa_guid_exists(pool_guid, 0)) {
2097 error = SET_ERROR(EEXIST);
2099 spa->spa_config_guid = pool_guid;
2101 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
2103 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
2107 nvlist_free(spa->spa_load_info);
2108 spa->spa_load_info = fnvlist_alloc();
2110 gethrestime(&spa->spa_loaded_ts);
2111 error = spa_load_impl(spa, pool_guid, config, state, type,
2112 mosconfig, &ereport);
2115 spa->spa_minref = refcount_count(&spa->spa_refcount);
2117 if (error != EEXIST) {
2118 spa->spa_loaded_ts.tv_sec = 0;
2119 spa->spa_loaded_ts.tv_nsec = 0;
2121 if (error != EBADF) {
2122 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
2125 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2132 * Load an existing storage pool, using the pool's builtin spa_config as a
2133 * source of configuration information.
2136 spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
2137 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
2141 nvlist_t *nvroot = NULL;
2144 uberblock_t *ub = &spa->spa_uberblock;
2145 uint64_t children, config_cache_txg = spa->spa_config_txg;
2146 int orig_mode = spa->spa_mode;
2149 boolean_t missing_feat_write = B_FALSE;
2152 * If this is an untrusted config, access the pool in read-only mode.
2153 * This prevents things like resilvering recently removed devices.
2156 spa->spa_mode = FREAD;
2158 ASSERT(MUTEX_HELD(&spa_namespace_lock));
2160 spa->spa_load_state = state;
2162 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
2163 return (SET_ERROR(EINVAL));
2165 parse = (type == SPA_IMPORT_EXISTING ?
2166 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2169 * Create "The Godfather" zio to hold all async IOs
2171 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2172 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2175 * Parse the configuration into a vdev tree. We explicitly set the
2176 * value that will be returned by spa_version() since parsing the
2177 * configuration requires knowing the version number.
2179 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2180 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
2181 spa_config_exit(spa, SCL_ALL, FTAG);
2186 ASSERT(spa->spa_root_vdev == rvd);
2188 if (type != SPA_IMPORT_ASSEMBLE) {
2189 ASSERT(spa_guid(spa) == pool_guid);
2193 * Try to open all vdevs, loading each label in the process.
2195 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2196 error = vdev_open(rvd);
2197 spa_config_exit(spa, SCL_ALL, FTAG);
2202 * We need to validate the vdev labels against the configuration that
2203 * we have in hand, which is dependent on the setting of mosconfig. If
2204 * mosconfig is true then we're validating the vdev labels based on
2205 * that config. Otherwise, we're validating against the cached config
2206 * (zpool.cache) that was read when we loaded the zfs module, and then
2207 * later we will recursively call spa_load() and validate against
2210 * If we're assembling a new pool that's been split off from an
2211 * existing pool, the labels haven't yet been updated so we skip
2212 * validation for now.
2214 if (type != SPA_IMPORT_ASSEMBLE) {
2215 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2216 error = vdev_validate(rvd, mosconfig);
2217 spa_config_exit(spa, SCL_ALL, FTAG);
2222 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2223 return (SET_ERROR(ENXIO));
2227 * Find the best uberblock.
2229 vdev_uberblock_load(rvd, ub, &label);
2232 * If we weren't able to find a single valid uberblock, return failure.
2234 if (ub->ub_txg == 0) {
2236 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2240 * If the pool has an unsupported version we can't open it.
2242 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2244 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2247 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2251 * If we weren't able to find what's necessary for reading the
2252 * MOS in the label, return failure.
2254 if (label == NULL || nvlist_lookup_nvlist(label,
2255 ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) {
2257 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2262 * Update our in-core representation with the definitive values
2265 nvlist_free(spa->spa_label_features);
2266 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2272 * Look through entries in the label nvlist's features_for_read. If
2273 * there is a feature listed there which we don't understand then we
2274 * cannot open a pool.
2276 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2277 nvlist_t *unsup_feat;
2279 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2282 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2284 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2285 if (!zfeature_is_supported(nvpair_name(nvp))) {
2286 VERIFY(nvlist_add_string(unsup_feat,
2287 nvpair_name(nvp), "") == 0);
2291 if (!nvlist_empty(unsup_feat)) {
2292 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2293 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2294 nvlist_free(unsup_feat);
2295 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2299 nvlist_free(unsup_feat);
2303 * If the vdev guid sum doesn't match the uberblock, we have an
2304 * incomplete configuration. We first check to see if the pool
2305 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2306 * If it is, defer the vdev_guid_sum check till later so we
2307 * can handle missing vdevs.
2309 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
2310 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
2311 rvd->vdev_guid_sum != ub->ub_guid_sum)
2312 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2314 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2315 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2316 spa_try_repair(spa, config);
2317 spa_config_exit(spa, SCL_ALL, FTAG);
2318 nvlist_free(spa->spa_config_splitting);
2319 spa->spa_config_splitting = NULL;
2323 * Initialize internal SPA structures.
2325 spa->spa_state = POOL_STATE_ACTIVE;
2326 spa->spa_ubsync = spa->spa_uberblock;
2327 spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2328 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2329 spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2330 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2331 spa->spa_claim_max_txg = spa->spa_first_txg;
2332 spa->spa_prev_software_version = ub->ub_software_version;
2334 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2336 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2337 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2339 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2340 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2342 if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2343 boolean_t missing_feat_read = B_FALSE;
2344 nvlist_t *unsup_feat, *enabled_feat;
2346 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2347 &spa->spa_feat_for_read_obj) != 0) {
2348 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2351 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2352 &spa->spa_feat_for_write_obj) != 0) {
2353 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2356 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2357 &spa->spa_feat_desc_obj) != 0) {
2358 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2361 enabled_feat = fnvlist_alloc();
2362 unsup_feat = fnvlist_alloc();
2364 if (!spa_features_check(spa, B_FALSE,
2365 unsup_feat, enabled_feat))
2366 missing_feat_read = B_TRUE;
2368 if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) {
2369 if (!spa_features_check(spa, B_TRUE,
2370 unsup_feat, enabled_feat)) {
2371 missing_feat_write = B_TRUE;
2375 fnvlist_add_nvlist(spa->spa_load_info,
2376 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
2378 if (!nvlist_empty(unsup_feat)) {
2379 fnvlist_add_nvlist(spa->spa_load_info,
2380 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
2383 fnvlist_free(enabled_feat);
2384 fnvlist_free(unsup_feat);
2386 if (!missing_feat_read) {
2387 fnvlist_add_boolean(spa->spa_load_info,
2388 ZPOOL_CONFIG_CAN_RDONLY);
2392 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2393 * twofold: to determine whether the pool is available for
2394 * import in read-write mode and (if it is not) whether the
2395 * pool is available for import in read-only mode. If the pool
2396 * is available for import in read-write mode, it is displayed
2397 * as available in userland; if it is not available for import
2398 * in read-only mode, it is displayed as unavailable in
2399 * userland. If the pool is available for import in read-only
2400 * mode but not read-write mode, it is displayed as unavailable
2401 * in userland with a special note that the pool is actually
2402 * available for open in read-only mode.
2404 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2405 * missing a feature for write, we must first determine whether
2406 * the pool can be opened read-only before returning to
2407 * userland in order to know whether to display the
2408 * abovementioned note.
2410 if (missing_feat_read || (missing_feat_write &&
2411 spa_writeable(spa))) {
2412 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2417 * Load refcounts for ZFS features from disk into an in-memory
2418 * cache during SPA initialization.
2420 for (spa_feature_t i = 0; i < SPA_FEATURES; i++) {
2423 error = feature_get_refcount_from_disk(spa,
2424 &spa_feature_table[i], &refcount);
2426 spa->spa_feat_refcount_cache[i] = refcount;
2427 } else if (error == ENOTSUP) {
2428 spa->spa_feat_refcount_cache[i] =
2429 SPA_FEATURE_DISABLED;
2431 return (spa_vdev_err(rvd,
2432 VDEV_AUX_CORRUPT_DATA, EIO));
2437 if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) {
2438 if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG,
2439 &spa->spa_feat_enabled_txg_obj) != 0)
2440 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2443 spa->spa_is_initializing = B_TRUE;
2444 error = dsl_pool_open(spa->spa_dsl_pool);
2445 spa->spa_is_initializing = B_FALSE;
2447 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2451 nvlist_t *policy = NULL, *nvconfig;
2453 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2454 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2456 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2457 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2459 unsigned long myhostid = 0;
2461 VERIFY(nvlist_lookup_string(nvconfig,
2462 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2465 myhostid = zone_get_hostid(NULL);
2468 * We're emulating the system's hostid in userland, so
2469 * we can't use zone_get_hostid().
2471 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2472 #endif /* _KERNEL */
2473 if (check_hostid && hostid != 0 && myhostid != 0 &&
2474 hostid != myhostid) {
2475 nvlist_free(nvconfig);
2476 cmn_err(CE_WARN, "pool '%s' could not be "
2477 "loaded as it was last accessed by "
2478 "another system (host: %s hostid: 0x%lx). "
2479 "See: http://illumos.org/msg/ZFS-8000-EY",
2480 spa_name(spa), hostname,
2481 (unsigned long)hostid);
2482 return (SET_ERROR(EBADF));
2485 if (nvlist_lookup_nvlist(spa->spa_config,
2486 ZPOOL_REWIND_POLICY, &policy) == 0)
2487 VERIFY(nvlist_add_nvlist(nvconfig,
2488 ZPOOL_REWIND_POLICY, policy) == 0);
2490 spa_config_set(spa, nvconfig);
2492 spa_deactivate(spa);
2493 spa_activate(spa, orig_mode);
2495 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2498 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2499 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2500 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2502 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2505 * Load the bit that tells us to use the new accounting function
2506 * (raid-z deflation). If we have an older pool, this will not
2509 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2510 if (error != 0 && error != ENOENT)
2511 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2513 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2514 &spa->spa_creation_version);
2515 if (error != 0 && error != ENOENT)
2516 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2519 * Load the persistent error log. If we have an older pool, this will
2522 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2523 if (error != 0 && error != ENOENT)
2524 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2526 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2527 &spa->spa_errlog_scrub);
2528 if (error != 0 && error != ENOENT)
2529 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2532 * Load the history object. If we have an older pool, this
2533 * will not be present.
2535 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2536 if (error != 0 && error != ENOENT)
2537 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2540 * If we're assembling the pool from the split-off vdevs of
2541 * an existing pool, we don't want to attach the spares & cache
2546 * Load any hot spares for this pool.
2548 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2549 if (error != 0 && error != ENOENT)
2550 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2551 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2552 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2553 if (load_nvlist(spa, spa->spa_spares.sav_object,
2554 &spa->spa_spares.sav_config) != 0)
2555 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2557 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2558 spa_load_spares(spa);
2559 spa_config_exit(spa, SCL_ALL, FTAG);
2560 } else if (error == 0) {
2561 spa->spa_spares.sav_sync = B_TRUE;
2565 * Load any level 2 ARC devices for this pool.
2567 error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2568 &spa->spa_l2cache.sav_object);
2569 if (error != 0 && error != ENOENT)
2570 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2571 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2572 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2573 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2574 &spa->spa_l2cache.sav_config) != 0)
2575 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2577 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2578 spa_load_l2cache(spa);
2579 spa_config_exit(spa, SCL_ALL, FTAG);
2580 } else if (error == 0) {
2581 spa->spa_l2cache.sav_sync = B_TRUE;
2584 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2586 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2587 if (error && error != ENOENT)
2588 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2591 uint64_t autoreplace;
2593 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2594 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2595 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2596 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2597 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2598 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2599 &spa->spa_dedup_ditto);
2601 spa->spa_autoreplace = (autoreplace != 0);
2605 * If the 'autoreplace' property is set, then post a resource notifying
2606 * the ZFS DE that it should not issue any faults for unopenable
2607 * devices. We also iterate over the vdevs, and post a sysevent for any
2608 * unopenable vdevs so that the normal autoreplace handler can take
2611 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2612 spa_check_removed(spa->spa_root_vdev);
2614 * For the import case, this is done in spa_import(), because
2615 * at this point we're using the spare definitions from
2616 * the MOS config, not necessarily from the userland config.
2618 if (state != SPA_LOAD_IMPORT) {
2619 spa_aux_check_removed(&spa->spa_spares);
2620 spa_aux_check_removed(&spa->spa_l2cache);
2625 * Load the vdev state for all toplevel vdevs.
2630 * Propagate the leaf DTLs we just loaded all the way up the tree.
2632 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2633 vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2634 spa_config_exit(spa, SCL_ALL, FTAG);
2637 * Load the DDTs (dedup tables).
2639 error = ddt_load(spa);
2641 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2643 spa_update_dspace(spa);
2646 * Validate the config, using the MOS config to fill in any
2647 * information which might be missing. If we fail to validate
2648 * the config then declare the pool unfit for use. If we're
2649 * assembling a pool from a split, the log is not transferred
2652 if (type != SPA_IMPORT_ASSEMBLE) {
2655 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2656 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2658 if (!spa_config_valid(spa, nvconfig)) {
2659 nvlist_free(nvconfig);
2660 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2663 nvlist_free(nvconfig);
2666 * Now that we've validated the config, check the state of the
2667 * root vdev. If it can't be opened, it indicates one or
2668 * more toplevel vdevs are faulted.
2670 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2671 return (SET_ERROR(ENXIO));
2673 if (spa_check_logs(spa)) {
2674 *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2675 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2679 if (missing_feat_write) {
2680 ASSERT(state == SPA_LOAD_TRYIMPORT);
2683 * At this point, we know that we can open the pool in
2684 * read-only mode but not read-write mode. We now have enough
2685 * information and can return to userland.
2687 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
2691 * We've successfully opened the pool, verify that we're ready
2692 * to start pushing transactions.
2694 if (state != SPA_LOAD_TRYIMPORT) {
2695 if (error = spa_load_verify(spa))
2696 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2700 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2701 spa->spa_load_max_txg == UINT64_MAX)) {
2703 int need_update = B_FALSE;
2705 ASSERT(state != SPA_LOAD_TRYIMPORT);
2708 * Claim log blocks that haven't been committed yet.
2709 * This must all happen in a single txg.
2710 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2711 * invoked from zil_claim_log_block()'s i/o done callback.
2712 * Price of rollback is that we abandon the log.
2714 spa->spa_claiming = B_TRUE;
2716 tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2717 spa_first_txg(spa));
2718 (void) dmu_objset_find(spa_name(spa),
2719 zil_claim, tx, DS_FIND_CHILDREN);
2722 spa->spa_claiming = B_FALSE;
2724 spa_set_log_state(spa, SPA_LOG_GOOD);
2725 spa->spa_sync_on = B_TRUE;
2726 txg_sync_start(spa->spa_dsl_pool);
2729 * Wait for all claims to sync. We sync up to the highest
2730 * claimed log block birth time so that claimed log blocks
2731 * don't appear to be from the future. spa_claim_max_txg
2732 * will have been set for us by either zil_check_log_chain()
2733 * (invoked from spa_check_logs()) or zil_claim() above.
2735 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2738 * If the config cache is stale, or we have uninitialized
2739 * metaslabs (see spa_vdev_add()), then update the config.
2741 * If this is a verbatim import, trust the current
2742 * in-core spa_config and update the disk labels.
2744 if (config_cache_txg != spa->spa_config_txg ||
2745 state == SPA_LOAD_IMPORT ||
2746 state == SPA_LOAD_RECOVER ||
2747 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2748 need_update = B_TRUE;
2750 for (int c = 0; c < rvd->vdev_children; c++)
2751 if (rvd->vdev_child[c]->vdev_ms_array == 0)
2752 need_update = B_TRUE;
2755 * Update the config cache asychronously in case we're the
2756 * root pool, in which case the config cache isn't writable yet.
2759 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2762 * Check all DTLs to see if anything needs resilvering.
2764 if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2765 vdev_resilver_needed(rvd, NULL, NULL))
2766 spa_async_request(spa, SPA_ASYNC_RESILVER);
2769 * Log the fact that we booted up (so that we can detect if
2770 * we rebooted in the middle of an operation).
2772 spa_history_log_version(spa, "open");
2775 * Delete any inconsistent datasets.
2777 (void) dmu_objset_find(spa_name(spa),
2778 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2781 * Clean up any stale temporary dataset userrefs.
2783 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2790 spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2792 int mode = spa->spa_mode;
2795 spa_deactivate(spa);
2797 spa->spa_load_max_txg--;
2799 spa_activate(spa, mode);
2800 spa_async_suspend(spa);
2802 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2806 * If spa_load() fails this function will try loading prior txg's. If
2807 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2808 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2809 * function will not rewind the pool and will return the same error as
2813 spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2814 uint64_t max_request, int rewind_flags)
2816 nvlist_t *loadinfo = NULL;
2817 nvlist_t *config = NULL;
2818 int load_error, rewind_error;
2819 uint64_t safe_rewind_txg;
2822 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2823 spa->spa_load_max_txg = spa->spa_load_txg;
2824 spa_set_log_state(spa, SPA_LOG_CLEAR);
2826 spa->spa_load_max_txg = max_request;
2829 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2831 if (load_error == 0)
2834 if (spa->spa_root_vdev != NULL)
2835 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2837 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2838 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2840 if (rewind_flags & ZPOOL_NEVER_REWIND) {
2841 nvlist_free(config);
2842 return (load_error);
2845 if (state == SPA_LOAD_RECOVER) {
2846 /* Price of rolling back is discarding txgs, including log */
2847 spa_set_log_state(spa, SPA_LOG_CLEAR);
2850 * If we aren't rolling back save the load info from our first
2851 * import attempt so that we can restore it after attempting
2854 loadinfo = spa->spa_load_info;
2855 spa->spa_load_info = fnvlist_alloc();
2858 spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2859 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2860 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2861 TXG_INITIAL : safe_rewind_txg;
2864 * Continue as long as we're finding errors, we're still within
2865 * the acceptable rewind range, and we're still finding uberblocks
2867 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2868 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2869 if (spa->spa_load_max_txg < safe_rewind_txg)
2870 spa->spa_extreme_rewind = B_TRUE;
2871 rewind_error = spa_load_retry(spa, state, mosconfig);
2874 spa->spa_extreme_rewind = B_FALSE;
2875 spa->spa_load_max_txg = UINT64_MAX;
2877 if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2878 spa_config_set(spa, config);
2880 if (state == SPA_LOAD_RECOVER) {
2881 ASSERT3P(loadinfo, ==, NULL);
2882 return (rewind_error);
2884 /* Store the rewind info as part of the initial load info */
2885 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
2886 spa->spa_load_info);
2888 /* Restore the initial load info */
2889 fnvlist_free(spa->spa_load_info);
2890 spa->spa_load_info = loadinfo;
2892 return (load_error);
2899 * The import case is identical to an open except that the configuration is sent
2900 * down from userland, instead of grabbed from the configuration cache. For the
2901 * case of an open, the pool configuration will exist in the
2902 * POOL_STATE_UNINITIALIZED state.
2904 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2905 * the same time open the pool, without having to keep around the spa_t in some
2909 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2913 spa_load_state_t state = SPA_LOAD_OPEN;
2915 int locked = B_FALSE;
2916 int firstopen = B_FALSE;
2921 * As disgusting as this is, we need to support recursive calls to this
2922 * function because dsl_dir_open() is called during spa_load(), and ends
2923 * up calling spa_open() again. The real fix is to figure out how to
2924 * avoid dsl_dir_open() calling this in the first place.
2926 if (mutex_owner(&spa_namespace_lock) != curthread) {
2927 mutex_enter(&spa_namespace_lock);
2931 if ((spa = spa_lookup(pool)) == NULL) {
2933 mutex_exit(&spa_namespace_lock);
2934 return (SET_ERROR(ENOENT));
2937 if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2938 zpool_rewind_policy_t policy;
2942 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2944 if (policy.zrp_request & ZPOOL_DO_REWIND)
2945 state = SPA_LOAD_RECOVER;
2947 spa_activate(spa, spa_mode_global);
2949 if (state != SPA_LOAD_RECOVER)
2950 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2952 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2953 policy.zrp_request);
2955 if (error == EBADF) {
2957 * If vdev_validate() returns failure (indicated by
2958 * EBADF), it indicates that one of the vdevs indicates
2959 * that the pool has been exported or destroyed. If
2960 * this is the case, the config cache is out of sync and
2961 * we should remove the pool from the namespace.
2964 spa_deactivate(spa);
2965 spa_config_sync(spa, B_TRUE, B_TRUE);
2968 mutex_exit(&spa_namespace_lock);
2969 return (SET_ERROR(ENOENT));
2974 * We can't open the pool, but we still have useful
2975 * information: the state of each vdev after the
2976 * attempted vdev_open(). Return this to the user.
2978 if (config != NULL && spa->spa_config) {
2979 VERIFY(nvlist_dup(spa->spa_config, config,
2981 VERIFY(nvlist_add_nvlist(*config,
2982 ZPOOL_CONFIG_LOAD_INFO,
2983 spa->spa_load_info) == 0);
2986 spa_deactivate(spa);
2987 spa->spa_last_open_failed = error;
2989 mutex_exit(&spa_namespace_lock);
2995 spa_open_ref(spa, tag);
2998 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
3001 * If we've recovered the pool, pass back any information we
3002 * gathered while doing the load.
3004 if (state == SPA_LOAD_RECOVER) {
3005 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
3006 spa->spa_load_info) == 0);
3010 spa->spa_last_open_failed = 0;
3011 spa->spa_last_ubsync_txg = 0;
3012 spa->spa_load_txg = 0;
3013 mutex_exit(&spa_namespace_lock);
3017 zvol_create_minors(spa->spa_name);
3028 spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
3031 return (spa_open_common(name, spapp, tag, policy, config));
3035 spa_open(const char *name, spa_t **spapp, void *tag)
3037 return (spa_open_common(name, spapp, tag, NULL, NULL));
3041 * Lookup the given spa_t, incrementing the inject count in the process,
3042 * preventing it from being exported or destroyed.
3045 spa_inject_addref(char *name)
3049 mutex_enter(&spa_namespace_lock);
3050 if ((spa = spa_lookup(name)) == NULL) {
3051 mutex_exit(&spa_namespace_lock);
3054 spa->spa_inject_ref++;
3055 mutex_exit(&spa_namespace_lock);
3061 spa_inject_delref(spa_t *spa)
3063 mutex_enter(&spa_namespace_lock);
3064 spa->spa_inject_ref--;
3065 mutex_exit(&spa_namespace_lock);
3069 * Add spares device information to the nvlist.
3072 spa_add_spares(spa_t *spa, nvlist_t *config)
3082 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3084 if (spa->spa_spares.sav_count == 0)
3087 VERIFY(nvlist_lookup_nvlist(config,
3088 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3089 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
3090 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3092 VERIFY(nvlist_add_nvlist_array(nvroot,
3093 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3094 VERIFY(nvlist_lookup_nvlist_array(nvroot,
3095 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3098 * Go through and find any spares which have since been
3099 * repurposed as an active spare. If this is the case, update
3100 * their status appropriately.
3102 for (i = 0; i < nspares; i++) {
3103 VERIFY(nvlist_lookup_uint64(spares[i],
3104 ZPOOL_CONFIG_GUID, &guid) == 0);
3105 if (spa_spare_exists(guid, &pool, NULL) &&
3107 VERIFY(nvlist_lookup_uint64_array(
3108 spares[i], ZPOOL_CONFIG_VDEV_STATS,
3109 (uint64_t **)&vs, &vsc) == 0);
3110 vs->vs_state = VDEV_STATE_CANT_OPEN;
3111 vs->vs_aux = VDEV_AUX_SPARED;
3118 * Add l2cache device information to the nvlist, including vdev stats.
3121 spa_add_l2cache(spa_t *spa, nvlist_t *config)
3124 uint_t i, j, nl2cache;
3131 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3133 if (spa->spa_l2cache.sav_count == 0)
3136 VERIFY(nvlist_lookup_nvlist(config,
3137 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3138 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
3139 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3140 if (nl2cache != 0) {
3141 VERIFY(nvlist_add_nvlist_array(nvroot,
3142 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3143 VERIFY(nvlist_lookup_nvlist_array(nvroot,
3144 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3147 * Update level 2 cache device stats.
3150 for (i = 0; i < nl2cache; i++) {
3151 VERIFY(nvlist_lookup_uint64(l2cache[i],
3152 ZPOOL_CONFIG_GUID, &guid) == 0);
3155 for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
3157 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
3158 vd = spa->spa_l2cache.sav_vdevs[j];
3164 VERIFY(nvlist_lookup_uint64_array(l2cache[i],
3165 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
3167 vdev_get_stats(vd, vs);
3173 spa_add_feature_stats(spa_t *spa, nvlist_t *config)
3179 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3180 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3182 /* We may be unable to read features if pool is suspended. */
3183 if (spa_suspended(spa))
3186 if (spa->spa_feat_for_read_obj != 0) {
3187 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3188 spa->spa_feat_for_read_obj);
3189 zap_cursor_retrieve(&zc, &za) == 0;
3190 zap_cursor_advance(&zc)) {
3191 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3192 za.za_num_integers == 1);
3193 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3194 za.za_first_integer));
3196 zap_cursor_fini(&zc);
3199 if (spa->spa_feat_for_write_obj != 0) {
3200 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3201 spa->spa_feat_for_write_obj);
3202 zap_cursor_retrieve(&zc, &za) == 0;
3203 zap_cursor_advance(&zc)) {
3204 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3205 za.za_num_integers == 1);
3206 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3207 za.za_first_integer));
3209 zap_cursor_fini(&zc);
3213 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
3215 nvlist_free(features);
3219 spa_get_stats(const char *name, nvlist_t **config,
3220 char *altroot, size_t buflen)
3226 error = spa_open_common(name, &spa, FTAG, NULL, config);
3230 * This still leaves a window of inconsistency where the spares
3231 * or l2cache devices could change and the config would be
3232 * self-inconsistent.
3234 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3236 if (*config != NULL) {
3237 uint64_t loadtimes[2];
3239 loadtimes[0] = spa->spa_loaded_ts.tv_sec;
3240 loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
3241 VERIFY(nvlist_add_uint64_array(*config,
3242 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
3244 VERIFY(nvlist_add_uint64(*config,
3245 ZPOOL_CONFIG_ERRCOUNT,
3246 spa_get_errlog_size(spa)) == 0);
3248 if (spa_suspended(spa))
3249 VERIFY(nvlist_add_uint64(*config,
3250 ZPOOL_CONFIG_SUSPENDED,
3251 spa->spa_failmode) == 0);
3253 spa_add_spares(spa, *config);
3254 spa_add_l2cache(spa, *config);
3255 spa_add_feature_stats(spa, *config);
3260 * We want to get the alternate root even for faulted pools, so we cheat
3261 * and call spa_lookup() directly.
3265 mutex_enter(&spa_namespace_lock);
3266 spa = spa_lookup(name);
3268 spa_altroot(spa, altroot, buflen);
3272 mutex_exit(&spa_namespace_lock);
3274 spa_altroot(spa, altroot, buflen);
3279 spa_config_exit(spa, SCL_CONFIG, FTAG);
3280 spa_close(spa, FTAG);
3287 * Validate that the auxiliary device array is well formed. We must have an
3288 * array of nvlists, each which describes a valid leaf vdev. If this is an
3289 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3290 * specified, as long as they are well-formed.
3293 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
3294 spa_aux_vdev_t *sav, const char *config, uint64_t version,
3295 vdev_labeltype_t label)
3302 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3305 * It's acceptable to have no devs specified.
3307 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
3311 return (SET_ERROR(EINVAL));
3314 * Make sure the pool is formatted with a version that supports this
3317 if (spa_version(spa) < version)
3318 return (SET_ERROR(ENOTSUP));
3321 * Set the pending device list so we correctly handle device in-use
3324 sav->sav_pending = dev;
3325 sav->sav_npending = ndev;
3327 for (i = 0; i < ndev; i++) {
3328 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
3332 if (!vd->vdev_ops->vdev_op_leaf) {
3334 error = SET_ERROR(EINVAL);
3339 * The L2ARC currently only supports disk devices in
3340 * kernel context. For user-level testing, we allow it.
3343 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
3344 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
3345 error = SET_ERROR(ENOTBLK);
3352 if ((error = vdev_open(vd)) == 0 &&
3353 (error = vdev_label_init(vd, crtxg, label)) == 0) {
3354 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
3355 vd->vdev_guid) == 0);
3361 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
3368 sav->sav_pending = NULL;
3369 sav->sav_npending = 0;
3374 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
3378 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3380 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3381 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
3382 VDEV_LABEL_SPARE)) != 0) {
3386 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3387 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
3388 VDEV_LABEL_L2CACHE));
3392 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
3397 if (sav->sav_config != NULL) {
3403 * Generate new dev list by concatentating with the
3406 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
3407 &olddevs, &oldndevs) == 0);
3409 newdevs = kmem_alloc(sizeof (void *) *
3410 (ndevs + oldndevs), KM_SLEEP);
3411 for (i = 0; i < oldndevs; i++)
3412 VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
3414 for (i = 0; i < ndevs; i++)
3415 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
3418 VERIFY(nvlist_remove(sav->sav_config, config,
3419 DATA_TYPE_NVLIST_ARRAY) == 0);
3421 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3422 config, newdevs, ndevs + oldndevs) == 0);
3423 for (i = 0; i < oldndevs + ndevs; i++)
3424 nvlist_free(newdevs[i]);
3425 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
3428 * Generate a new dev list.
3430 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
3432 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
3438 * Stop and drop level 2 ARC devices
3441 spa_l2cache_drop(spa_t *spa)
3445 spa_aux_vdev_t *sav = &spa->spa_l2cache;
3447 for (i = 0; i < sav->sav_count; i++) {
3450 vd = sav->sav_vdevs[i];
3453 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
3454 pool != 0ULL && l2arc_vdev_present(vd))
3455 l2arc_remove_vdev(vd);
3463 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
3467 char *altroot = NULL;
3472 uint64_t txg = TXG_INITIAL;
3473 nvlist_t **spares, **l2cache;
3474 uint_t nspares, nl2cache;
3475 uint64_t version, obj;
3476 boolean_t has_features;
3479 * If this pool already exists, return failure.
3481 mutex_enter(&spa_namespace_lock);
3482 if (spa_lookup(pool) != NULL) {
3483 mutex_exit(&spa_namespace_lock);
3484 return (SET_ERROR(EEXIST));
3488 * Allocate a new spa_t structure.
3490 (void) nvlist_lookup_string(props,
3491 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3492 spa = spa_add(pool, NULL, altroot);
3493 spa_activate(spa, spa_mode_global);
3495 if (props && (error = spa_prop_validate(spa, props))) {
3496 spa_deactivate(spa);
3498 mutex_exit(&spa_namespace_lock);
3502 has_features = B_FALSE;
3503 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
3504 elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
3505 if (zpool_prop_feature(nvpair_name(elem)))
3506 has_features = B_TRUE;
3509 if (has_features || nvlist_lookup_uint64(props,
3510 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
3511 version = SPA_VERSION;
3513 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
3515 spa->spa_first_txg = txg;
3516 spa->spa_uberblock.ub_txg = txg - 1;
3517 spa->spa_uberblock.ub_version = version;
3518 spa->spa_ubsync = spa->spa_uberblock;
3521 * Create "The Godfather" zio to hold all async IOs
3523 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
3524 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
3527 * Create the root vdev.
3529 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3531 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3533 ASSERT(error != 0 || rvd != NULL);
3534 ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3536 if (error == 0 && !zfs_allocatable_devs(nvroot))
3537 error = SET_ERROR(EINVAL);
3540 (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3541 (error = spa_validate_aux(spa, nvroot, txg,
3542 VDEV_ALLOC_ADD)) == 0) {
3543 for (int c = 0; c < rvd->vdev_children; c++) {
3544 vdev_ashift_optimize(rvd->vdev_child[c]);
3545 vdev_metaslab_set_size(rvd->vdev_child[c]);
3546 vdev_expand(rvd->vdev_child[c], txg);
3550 spa_config_exit(spa, SCL_ALL, FTAG);
3554 spa_deactivate(spa);
3556 mutex_exit(&spa_namespace_lock);
3561 * Get the list of spares, if specified.
3563 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3564 &spares, &nspares) == 0) {
3565 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3567 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3568 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3569 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3570 spa_load_spares(spa);
3571 spa_config_exit(spa, SCL_ALL, FTAG);
3572 spa->spa_spares.sav_sync = B_TRUE;
3576 * Get the list of level 2 cache devices, if specified.
3578 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3579 &l2cache, &nl2cache) == 0) {
3580 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3581 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3582 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3583 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3584 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3585 spa_load_l2cache(spa);
3586 spa_config_exit(spa, SCL_ALL, FTAG);
3587 spa->spa_l2cache.sav_sync = B_TRUE;
3590 spa->spa_is_initializing = B_TRUE;
3591 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3592 spa->spa_meta_objset = dp->dp_meta_objset;
3593 spa->spa_is_initializing = B_FALSE;
3596 * Create DDTs (dedup tables).
3600 spa_update_dspace(spa);
3602 tx = dmu_tx_create_assigned(dp, txg);
3605 * Create the pool config object.
3607 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3608 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3609 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3611 if (zap_add(spa->spa_meta_objset,
3612 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3613 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3614 cmn_err(CE_PANIC, "failed to add pool config");
3617 if (spa_version(spa) >= SPA_VERSION_FEATURES)
3618 spa_feature_create_zap_objects(spa, tx);
3620 if (zap_add(spa->spa_meta_objset,
3621 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3622 sizeof (uint64_t), 1, &version, tx) != 0) {
3623 cmn_err(CE_PANIC, "failed to add pool version");
3626 /* Newly created pools with the right version are always deflated. */
3627 if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3628 spa->spa_deflate = TRUE;
3629 if (zap_add(spa->spa_meta_objset,
3630 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3631 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3632 cmn_err(CE_PANIC, "failed to add deflate");
3637 * Create the deferred-free bpobj. Turn off compression
3638 * because sync-to-convergence takes longer if the blocksize
3641 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3642 dmu_object_set_compress(spa->spa_meta_objset, obj,
3643 ZIO_COMPRESS_OFF, tx);
3644 if (zap_add(spa->spa_meta_objset,
3645 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3646 sizeof (uint64_t), 1, &obj, tx) != 0) {
3647 cmn_err(CE_PANIC, "failed to add bpobj");
3649 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3650 spa->spa_meta_objset, obj));
3653 * Create the pool's history object.
3655 if (version >= SPA_VERSION_ZPOOL_HISTORY)
3656 spa_history_create_obj(spa, tx);
3659 * Set pool properties.
3661 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3662 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3663 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3664 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3666 if (props != NULL) {
3667 spa_configfile_set(spa, props, B_FALSE);
3668 spa_sync_props(props, tx);
3673 spa->spa_sync_on = B_TRUE;
3674 txg_sync_start(spa->spa_dsl_pool);
3677 * We explicitly wait for the first transaction to complete so that our
3678 * bean counters are appropriately updated.
3680 txg_wait_synced(spa->spa_dsl_pool, txg);
3682 spa_config_sync(spa, B_FALSE, B_TRUE);
3684 spa_history_log_version(spa, "create");
3686 spa->spa_minref = refcount_count(&spa->spa_refcount);
3688 mutex_exit(&spa_namespace_lock);
3696 * Get the root pool information from the root disk, then import the root pool
3697 * during the system boot up time.
3699 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3702 spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3705 nvlist_t *nvtop, *nvroot;
3708 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3712 * Add this top-level vdev to the child array.
3714 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3716 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3718 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3721 * Put this pool's top-level vdevs into a root vdev.
3723 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3724 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3725 VDEV_TYPE_ROOT) == 0);
3726 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3727 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3728 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3732 * Replace the existing vdev_tree with the new root vdev in
3733 * this pool's configuration (remove the old, add the new).
3735 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3736 nvlist_free(nvroot);
3741 * Walk the vdev tree and see if we can find a device with "better"
3742 * configuration. A configuration is "better" if the label on that
3743 * device has a more recent txg.
3746 spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3748 for (int c = 0; c < vd->vdev_children; c++)
3749 spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3751 if (vd->vdev_ops->vdev_op_leaf) {
3755 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3759 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3763 * Do we have a better boot device?
3765 if (label_txg > *txg) {
3774 * Import a root pool.
3776 * For x86. devpath_list will consist of devid and/or physpath name of
3777 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3778 * The GRUB "findroot" command will return the vdev we should boot.
3780 * For Sparc, devpath_list consists the physpath name of the booting device
3781 * no matter the rootpool is a single device pool or a mirrored pool.
3783 * "/pci@1f,0/ide@d/disk@0,0:a"
3786 spa_import_rootpool(char *devpath, char *devid)
3789 vdev_t *rvd, *bvd, *avd = NULL;
3790 nvlist_t *config, *nvtop;
3796 * Read the label from the boot device and generate a configuration.
3798 config = spa_generate_rootconf(devpath, devid, &guid);
3799 #if defined(_OBP) && defined(_KERNEL)
3800 if (config == NULL) {
3801 if (strstr(devpath, "/iscsi/ssd") != NULL) {
3803 get_iscsi_bootpath_phy(devpath);
3804 config = spa_generate_rootconf(devpath, devid, &guid);
3808 if (config == NULL) {
3809 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
3811 return (SET_ERROR(EIO));
3814 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3816 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3818 mutex_enter(&spa_namespace_lock);
3819 if ((spa = spa_lookup(pname)) != NULL) {
3821 * Remove the existing root pool from the namespace so that we
3822 * can replace it with the correct config we just read in.
3827 spa = spa_add(pname, config, NULL);
3828 spa->spa_is_root = B_TRUE;
3829 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3832 * Build up a vdev tree based on the boot device's label config.
3834 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3836 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3837 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3838 VDEV_ALLOC_ROOTPOOL);
3839 spa_config_exit(spa, SCL_ALL, FTAG);
3841 mutex_exit(&spa_namespace_lock);
3842 nvlist_free(config);
3843 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3849 * Get the boot vdev.
3851 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3852 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3853 (u_longlong_t)guid);
3854 error = SET_ERROR(ENOENT);
3859 * Determine if there is a better boot device.
3862 spa_alt_rootvdev(rvd, &avd, &txg);
3864 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3865 "try booting from '%s'", avd->vdev_path);
3866 error = SET_ERROR(EINVAL);
3871 * If the boot device is part of a spare vdev then ensure that
3872 * we're booting off the active spare.
3874 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3875 !bvd->vdev_isspare) {
3876 cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3877 "try booting from '%s'",
3879 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3880 error = SET_ERROR(EINVAL);
3886 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3888 spa_config_exit(spa, SCL_ALL, FTAG);
3889 mutex_exit(&spa_namespace_lock);
3891 nvlist_free(config);
3897 extern int vdev_geom_read_pool_label(const char *name, nvlist_t ***configs,
3901 spa_generate_rootconf(const char *name)
3903 nvlist_t **configs, **tops;
3905 nvlist_t *best_cfg, *nvtop, *nvroot;
3914 if (vdev_geom_read_pool_label(name, &configs, &count) != 0)
3917 ASSERT3U(count, !=, 0);
3919 for (i = 0; i < count; i++) {
3922 VERIFY(nvlist_lookup_uint64(configs[i], ZPOOL_CONFIG_POOL_TXG,
3924 if (txg > best_txg) {
3926 best_cfg = configs[i];
3931 nvlist_lookup_uint64(best_cfg, ZPOOL_CONFIG_VDEV_CHILDREN, &nchildren);
3933 nvlist_lookup_uint64_array(best_cfg, ZPOOL_CONFIG_HOLE_ARRAY,
3936 tops = kmem_zalloc(nchildren * sizeof(void *), KM_SLEEP);
3937 for (i = 0; i < nchildren; i++) {
3940 if (configs[i] == NULL)
3942 VERIFY(nvlist_lookup_nvlist(configs[i], ZPOOL_CONFIG_VDEV_TREE,
3944 nvlist_dup(nvtop, &tops[i], KM_SLEEP);
3946 for (i = 0; holes != NULL && i < nholes; i++) {
3949 if (tops[holes[i]] != NULL)
3951 nvlist_alloc(&tops[holes[i]], NV_UNIQUE_NAME, KM_SLEEP);
3952 VERIFY(nvlist_add_string(tops[holes[i]], ZPOOL_CONFIG_TYPE,
3953 VDEV_TYPE_HOLE) == 0);
3954 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_ID,
3956 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_GUID,
3959 for (i = 0; i < nchildren; i++) {
3960 if (tops[i] != NULL)
3962 nvlist_alloc(&tops[i], NV_UNIQUE_NAME, KM_SLEEP);
3963 VERIFY(nvlist_add_string(tops[i], ZPOOL_CONFIG_TYPE,
3964 VDEV_TYPE_MISSING) == 0);
3965 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_ID,
3967 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_GUID,
3972 * Create pool config based on the best vdev config.
3974 nvlist_dup(best_cfg, &config, KM_SLEEP);
3977 * Put this pool's top-level vdevs into a root vdev.
3979 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3981 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3982 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3983 VDEV_TYPE_ROOT) == 0);
3984 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3985 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3986 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3987 tops, nchildren) == 0);
3990 * Replace the existing vdev_tree with the new root vdev in
3991 * this pool's configuration (remove the old, add the new).
3993 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3996 * Drop vdev config elements that should not be present at pool level.
3998 nvlist_remove(config, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64);
3999 nvlist_remove(config, ZPOOL_CONFIG_TOP_GUID, DATA_TYPE_UINT64);
4001 for (i = 0; i < count; i++)
4002 nvlist_free(configs[i]);
4003 kmem_free(configs, count * sizeof(void *));
4004 for (i = 0; i < nchildren; i++)
4005 nvlist_free(tops[i]);
4006 kmem_free(tops, nchildren * sizeof(void *));
4007 nvlist_free(nvroot);
4012 spa_import_rootpool(const char *name)
4015 vdev_t *rvd, *bvd, *avd = NULL;
4016 nvlist_t *config, *nvtop;
4022 * Read the label from the boot device and generate a configuration.
4024 config = spa_generate_rootconf(name);
4026 mutex_enter(&spa_namespace_lock);
4027 if (config != NULL) {
4028 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
4029 &pname) == 0 && strcmp(name, pname) == 0);
4030 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg)
4033 if ((spa = spa_lookup(pname)) != NULL) {
4035 * Remove the existing root pool from the namespace so
4036 * that we can replace it with the correct config
4041 spa = spa_add(pname, config, NULL);
4044 * Set spa_ubsync.ub_version as it can be used in vdev_alloc()
4045 * via spa_version().
4047 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
4048 &spa->spa_ubsync.ub_version) != 0)
4049 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
4050 } else if ((spa = spa_lookup(name)) == NULL) {
4051 cmn_err(CE_NOTE, "Cannot find the pool label for '%s'",
4055 VERIFY(nvlist_dup(spa->spa_config, &config, KM_SLEEP) == 0);
4057 spa->spa_is_root = B_TRUE;
4058 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
4061 * Build up a vdev tree based on the boot device's label config.
4063 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4065 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4066 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
4067 VDEV_ALLOC_ROOTPOOL);
4068 spa_config_exit(spa, SCL_ALL, FTAG);
4070 mutex_exit(&spa_namespace_lock);
4071 nvlist_free(config);
4072 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
4077 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4079 spa_config_exit(spa, SCL_ALL, FTAG);
4080 mutex_exit(&spa_namespace_lock);
4082 nvlist_free(config);
4090 * Import a non-root pool into the system.
4093 spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
4096 char *altroot = NULL;
4097 spa_load_state_t state = SPA_LOAD_IMPORT;
4098 zpool_rewind_policy_t policy;
4099 uint64_t mode = spa_mode_global;
4100 uint64_t readonly = B_FALSE;
4103 nvlist_t **spares, **l2cache;
4104 uint_t nspares, nl2cache;
4107 * If a pool with this name exists, return failure.
4109 mutex_enter(&spa_namespace_lock);
4110 if (spa_lookup(pool) != NULL) {
4111 mutex_exit(&spa_namespace_lock);
4112 return (SET_ERROR(EEXIST));
4116 * Create and initialize the spa structure.
4118 (void) nvlist_lookup_string(props,
4119 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4120 (void) nvlist_lookup_uint64(props,
4121 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
4124 spa = spa_add(pool, config, altroot);
4125 spa->spa_import_flags = flags;
4128 * Verbatim import - Take a pool and insert it into the namespace
4129 * as if it had been loaded at boot.
4131 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
4133 spa_configfile_set(spa, props, B_FALSE);
4135 spa_config_sync(spa, B_FALSE, B_TRUE);
4137 mutex_exit(&spa_namespace_lock);
4141 spa_activate(spa, mode);
4144 * Don't start async tasks until we know everything is healthy.
4146 spa_async_suspend(spa);
4148 zpool_get_rewind_policy(config, &policy);
4149 if (policy.zrp_request & ZPOOL_DO_REWIND)
4150 state = SPA_LOAD_RECOVER;
4153 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
4154 * because the user-supplied config is actually the one to trust when
4157 if (state != SPA_LOAD_RECOVER)
4158 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
4160 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
4161 policy.zrp_request);
4164 * Propagate anything learned while loading the pool and pass it
4165 * back to caller (i.e. rewind info, missing devices, etc).
4167 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4168 spa->spa_load_info) == 0);
4170 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4172 * Toss any existing sparelist, as it doesn't have any validity
4173 * anymore, and conflicts with spa_has_spare().
4175 if (spa->spa_spares.sav_config) {
4176 nvlist_free(spa->spa_spares.sav_config);
4177 spa->spa_spares.sav_config = NULL;
4178 spa_load_spares(spa);
4180 if (spa->spa_l2cache.sav_config) {
4181 nvlist_free(spa->spa_l2cache.sav_config);
4182 spa->spa_l2cache.sav_config = NULL;
4183 spa_load_l2cache(spa);
4186 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4189 error = spa_validate_aux(spa, nvroot, -1ULL,
4192 error = spa_validate_aux(spa, nvroot, -1ULL,
4193 VDEV_ALLOC_L2CACHE);
4194 spa_config_exit(spa, SCL_ALL, FTAG);
4197 spa_configfile_set(spa, props, B_FALSE);
4199 if (error != 0 || (props && spa_writeable(spa) &&
4200 (error = spa_prop_set(spa, props)))) {
4202 spa_deactivate(spa);
4204 mutex_exit(&spa_namespace_lock);
4208 spa_async_resume(spa);
4211 * Override any spares and level 2 cache devices as specified by
4212 * the user, as these may have correct device names/devids, etc.
4214 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
4215 &spares, &nspares) == 0) {
4216 if (spa->spa_spares.sav_config)
4217 VERIFY(nvlist_remove(spa->spa_spares.sav_config,
4218 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
4220 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
4221 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4222 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
4223 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
4224 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4225 spa_load_spares(spa);
4226 spa_config_exit(spa, SCL_ALL, FTAG);
4227 spa->spa_spares.sav_sync = B_TRUE;
4229 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
4230 &l2cache, &nl2cache) == 0) {
4231 if (spa->spa_l2cache.sav_config)
4232 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
4233 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
4235 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
4236 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4237 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
4238 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
4239 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4240 spa_load_l2cache(spa);
4241 spa_config_exit(spa, SCL_ALL, FTAG);
4242 spa->spa_l2cache.sav_sync = B_TRUE;
4246 * Check for any removed devices.
4248 if (spa->spa_autoreplace) {
4249 spa_aux_check_removed(&spa->spa_spares);
4250 spa_aux_check_removed(&spa->spa_l2cache);
4253 if (spa_writeable(spa)) {
4255 * Update the config cache to include the newly-imported pool.
4257 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4261 * It's possible that the pool was expanded while it was exported.
4262 * We kick off an async task to handle this for us.
4264 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
4266 mutex_exit(&spa_namespace_lock);
4267 spa_history_log_version(spa, "import");
4271 zvol_create_minors(pool);
4278 spa_tryimport(nvlist_t *tryconfig)
4280 nvlist_t *config = NULL;
4286 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
4289 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
4293 * Create and initialize the spa structure.
4295 mutex_enter(&spa_namespace_lock);
4296 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
4297 spa_activate(spa, FREAD);
4300 * Pass off the heavy lifting to spa_load().
4301 * Pass TRUE for mosconfig because the user-supplied config
4302 * is actually the one to trust when doing an import.
4304 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
4307 * If 'tryconfig' was at least parsable, return the current config.
4309 if (spa->spa_root_vdev != NULL) {
4310 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4311 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
4313 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4315 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
4316 spa->spa_uberblock.ub_timestamp) == 0);
4317 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4318 spa->spa_load_info) == 0);
4321 * If the bootfs property exists on this pool then we
4322 * copy it out so that external consumers can tell which
4323 * pools are bootable.
4325 if ((!error || error == EEXIST) && spa->spa_bootfs) {
4326 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4329 * We have to play games with the name since the
4330 * pool was opened as TRYIMPORT_NAME.
4332 if (dsl_dsobj_to_dsname(spa_name(spa),
4333 spa->spa_bootfs, tmpname) == 0) {
4335 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4337 cp = strchr(tmpname, '/');
4339 (void) strlcpy(dsname, tmpname,
4342 (void) snprintf(dsname, MAXPATHLEN,
4343 "%s/%s", poolname, ++cp);
4345 VERIFY(nvlist_add_string(config,
4346 ZPOOL_CONFIG_BOOTFS, dsname) == 0);
4347 kmem_free(dsname, MAXPATHLEN);
4349 kmem_free(tmpname, MAXPATHLEN);
4353 * Add the list of hot spares and level 2 cache devices.
4355 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4356 spa_add_spares(spa, config);
4357 spa_add_l2cache(spa, config);
4358 spa_config_exit(spa, SCL_CONFIG, FTAG);
4362 spa_deactivate(spa);
4364 mutex_exit(&spa_namespace_lock);
4370 * Pool export/destroy
4372 * The act of destroying or exporting a pool is very simple. We make sure there
4373 * is no more pending I/O and any references to the pool are gone. Then, we
4374 * update the pool state and sync all the labels to disk, removing the
4375 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4376 * we don't sync the labels or remove the configuration cache.
4379 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
4380 boolean_t force, boolean_t hardforce)
4387 if (!(spa_mode_global & FWRITE))
4388 return (SET_ERROR(EROFS));
4390 mutex_enter(&spa_namespace_lock);
4391 if ((spa = spa_lookup(pool)) == NULL) {
4392 mutex_exit(&spa_namespace_lock);
4393 return (SET_ERROR(ENOENT));
4397 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4398 * reacquire the namespace lock, and see if we can export.
4400 spa_open_ref(spa, FTAG);
4401 mutex_exit(&spa_namespace_lock);
4402 spa_async_suspend(spa);
4403 mutex_enter(&spa_namespace_lock);
4404 spa_close(spa, FTAG);
4407 * The pool will be in core if it's openable,
4408 * in which case we can modify its state.
4410 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
4412 * Objsets may be open only because they're dirty, so we
4413 * have to force it to sync before checking spa_refcnt.
4415 txg_wait_synced(spa->spa_dsl_pool, 0);
4418 * A pool cannot be exported or destroyed if there are active
4419 * references. If we are resetting a pool, allow references by
4420 * fault injection handlers.
4422 if (!spa_refcount_zero(spa) ||
4423 (spa->spa_inject_ref != 0 &&
4424 new_state != POOL_STATE_UNINITIALIZED)) {
4425 spa_async_resume(spa);
4426 mutex_exit(&spa_namespace_lock);
4427 return (SET_ERROR(EBUSY));
4431 * A pool cannot be exported if it has an active shared spare.
4432 * This is to prevent other pools stealing the active spare
4433 * from an exported pool. At user's own will, such pool can
4434 * be forcedly exported.
4436 if (!force && new_state == POOL_STATE_EXPORTED &&
4437 spa_has_active_shared_spare(spa)) {
4438 spa_async_resume(spa);
4439 mutex_exit(&spa_namespace_lock);
4440 return (SET_ERROR(EXDEV));
4444 * We want this to be reflected on every label,
4445 * so mark them all dirty. spa_unload() will do the
4446 * final sync that pushes these changes out.
4448 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
4449 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4450 spa->spa_state = new_state;
4451 spa->spa_final_txg = spa_last_synced_txg(spa) +
4453 vdev_config_dirty(spa->spa_root_vdev);
4454 spa_config_exit(spa, SCL_ALL, FTAG);
4458 spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
4460 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4462 spa_deactivate(spa);
4465 if (oldconfig && spa->spa_config)
4466 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
4468 if (new_state != POOL_STATE_UNINITIALIZED) {
4470 spa_config_sync(spa, B_TRUE, B_TRUE);
4473 mutex_exit(&spa_namespace_lock);
4479 * Destroy a storage pool.
4482 spa_destroy(char *pool)
4484 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
4489 * Export a storage pool.
4492 spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
4493 boolean_t hardforce)
4495 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
4500 * Similar to spa_export(), this unloads the spa_t without actually removing it
4501 * from the namespace in any way.
4504 spa_reset(char *pool)
4506 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
4511 * ==========================================================================
4512 * Device manipulation
4513 * ==========================================================================
4517 * Add a device to a storage pool.
4520 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
4524 vdev_t *rvd = spa->spa_root_vdev;
4526 nvlist_t **spares, **l2cache;
4527 uint_t nspares, nl2cache;
4529 ASSERT(spa_writeable(spa));
4531 txg = spa_vdev_enter(spa);
4533 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
4534 VDEV_ALLOC_ADD)) != 0)
4535 return (spa_vdev_exit(spa, NULL, txg, error));
4537 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */
4539 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
4543 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
4547 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
4548 return (spa_vdev_exit(spa, vd, txg, EINVAL));
4550 if (vd->vdev_children != 0 &&
4551 (error = vdev_create(vd, txg, B_FALSE)) != 0)
4552 return (spa_vdev_exit(spa, vd, txg, error));
4555 * We must validate the spares and l2cache devices after checking the
4556 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
4558 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
4559 return (spa_vdev_exit(spa, vd, txg, error));
4562 * Transfer each new top-level vdev from vd to rvd.
4564 for (int c = 0; c < vd->vdev_children; c++) {
4567 * Set the vdev id to the first hole, if one exists.
4569 for (id = 0; id < rvd->vdev_children; id++) {
4570 if (rvd->vdev_child[id]->vdev_ishole) {
4571 vdev_free(rvd->vdev_child[id]);
4575 tvd = vd->vdev_child[c];
4576 vdev_remove_child(vd, tvd);
4578 vdev_add_child(rvd, tvd);
4579 vdev_config_dirty(tvd);
4583 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
4584 ZPOOL_CONFIG_SPARES);
4585 spa_load_spares(spa);
4586 spa->spa_spares.sav_sync = B_TRUE;
4589 if (nl2cache != 0) {
4590 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
4591 ZPOOL_CONFIG_L2CACHE);
4592 spa_load_l2cache(spa);
4593 spa->spa_l2cache.sav_sync = B_TRUE;
4597 * We have to be careful when adding new vdevs to an existing pool.
4598 * If other threads start allocating from these vdevs before we
4599 * sync the config cache, and we lose power, then upon reboot we may
4600 * fail to open the pool because there are DVAs that the config cache
4601 * can't translate. Therefore, we first add the vdevs without
4602 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4603 * and then let spa_config_update() initialize the new metaslabs.
4605 * spa_load() checks for added-but-not-initialized vdevs, so that
4606 * if we lose power at any point in this sequence, the remaining
4607 * steps will be completed the next time we load the pool.
4609 (void) spa_vdev_exit(spa, vd, txg, 0);
4611 mutex_enter(&spa_namespace_lock);
4612 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4613 mutex_exit(&spa_namespace_lock);
4619 * Attach a device to a mirror. The arguments are the path to any device
4620 * in the mirror, and the nvroot for the new device. If the path specifies
4621 * a device that is not mirrored, we automatically insert the mirror vdev.
4623 * If 'replacing' is specified, the new device is intended to replace the
4624 * existing device; in this case the two devices are made into their own
4625 * mirror using the 'replacing' vdev, which is functionally identical to
4626 * the mirror vdev (it actually reuses all the same ops) but has a few
4627 * extra rules: you can't attach to it after it's been created, and upon
4628 * completion of resilvering, the first disk (the one being replaced)
4629 * is automatically detached.
4632 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
4634 uint64_t txg, dtl_max_txg;
4635 vdev_t *rvd = spa->spa_root_vdev;
4636 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
4638 char *oldvdpath, *newvdpath;
4642 ASSERT(spa_writeable(spa));
4644 txg = spa_vdev_enter(spa);
4646 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
4649 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4651 if (!oldvd->vdev_ops->vdev_op_leaf)
4652 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4654 pvd = oldvd->vdev_parent;
4656 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
4657 VDEV_ALLOC_ATTACH)) != 0)
4658 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4660 if (newrootvd->vdev_children != 1)
4661 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4663 newvd = newrootvd->vdev_child[0];
4665 if (!newvd->vdev_ops->vdev_op_leaf)
4666 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4668 if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
4669 return (spa_vdev_exit(spa, newrootvd, txg, error));
4672 * Spares can't replace logs
4674 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
4675 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4679 * For attach, the only allowable parent is a mirror or the root
4682 if (pvd->vdev_ops != &vdev_mirror_ops &&
4683 pvd->vdev_ops != &vdev_root_ops)
4684 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4686 pvops = &vdev_mirror_ops;
4689 * Active hot spares can only be replaced by inactive hot
4692 if (pvd->vdev_ops == &vdev_spare_ops &&
4693 oldvd->vdev_isspare &&
4694 !spa_has_spare(spa, newvd->vdev_guid))
4695 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4698 * If the source is a hot spare, and the parent isn't already a
4699 * spare, then we want to create a new hot spare. Otherwise, we
4700 * want to create a replacing vdev. The user is not allowed to
4701 * attach to a spared vdev child unless the 'isspare' state is
4702 * the same (spare replaces spare, non-spare replaces
4705 if (pvd->vdev_ops == &vdev_replacing_ops &&
4706 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
4707 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4708 } else if (pvd->vdev_ops == &vdev_spare_ops &&
4709 newvd->vdev_isspare != oldvd->vdev_isspare) {
4710 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4713 if (newvd->vdev_isspare)
4714 pvops = &vdev_spare_ops;
4716 pvops = &vdev_replacing_ops;
4720 * Make sure the new device is big enough.
4722 if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
4723 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
4726 * The new device cannot have a higher alignment requirement
4727 * than the top-level vdev.
4729 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4730 return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4733 * If this is an in-place replacement, update oldvd's path and devid
4734 * to make it distinguishable from newvd, and unopenable from now on.
4736 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4737 spa_strfree(oldvd->vdev_path);
4738 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4740 (void) sprintf(oldvd->vdev_path, "%s/%s",
4741 newvd->vdev_path, "old");
4742 if (oldvd->vdev_devid != NULL) {
4743 spa_strfree(oldvd->vdev_devid);
4744 oldvd->vdev_devid = NULL;
4748 /* mark the device being resilvered */
4749 newvd->vdev_resilver_txg = txg;
4752 * If the parent is not a mirror, or if we're replacing, insert the new
4753 * mirror/replacing/spare vdev above oldvd.
4755 if (pvd->vdev_ops != pvops)
4756 pvd = vdev_add_parent(oldvd, pvops);
4758 ASSERT(pvd->vdev_top->vdev_parent == rvd);
4759 ASSERT(pvd->vdev_ops == pvops);
4760 ASSERT(oldvd->vdev_parent == pvd);
4763 * Extract the new device from its root and add it to pvd.
4765 vdev_remove_child(newrootvd, newvd);
4766 newvd->vdev_id = pvd->vdev_children;
4767 newvd->vdev_crtxg = oldvd->vdev_crtxg;
4768 vdev_add_child(pvd, newvd);
4770 tvd = newvd->vdev_top;
4771 ASSERT(pvd->vdev_top == tvd);
4772 ASSERT(tvd->vdev_parent == rvd);
4774 vdev_config_dirty(tvd);
4777 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4778 * for any dmu_sync-ed blocks. It will propagate upward when
4779 * spa_vdev_exit() calls vdev_dtl_reassess().
4781 dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4783 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4784 dtl_max_txg - TXG_INITIAL);
4786 if (newvd->vdev_isspare) {
4787 spa_spare_activate(newvd);
4788 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
4791 oldvdpath = spa_strdup(oldvd->vdev_path);
4792 newvdpath = spa_strdup(newvd->vdev_path);
4793 newvd_isspare = newvd->vdev_isspare;
4796 * Mark newvd's DTL dirty in this txg.
4798 vdev_dirty(tvd, VDD_DTL, newvd, txg);
4801 * Schedule the resilver to restart in the future. We do this to
4802 * ensure that dmu_sync-ed blocks have been stitched into the
4803 * respective datasets.
4805 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4810 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4812 spa_history_log_internal(spa, "vdev attach", NULL,
4813 "%s vdev=%s %s vdev=%s",
4814 replacing && newvd_isspare ? "spare in" :
4815 replacing ? "replace" : "attach", newvdpath,
4816 replacing ? "for" : "to", oldvdpath);
4818 spa_strfree(oldvdpath);
4819 spa_strfree(newvdpath);
4821 if (spa->spa_bootfs)
4822 spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
4828 * Detach a device from a mirror or replacing vdev.
4830 * If 'replace_done' is specified, only detach if the parent
4831 * is a replacing vdev.
4834 spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4838 vdev_t *rvd = spa->spa_root_vdev;
4839 vdev_t *vd, *pvd, *cvd, *tvd;
4840 boolean_t unspare = B_FALSE;
4841 uint64_t unspare_guid = 0;
4844 ASSERT(spa_writeable(spa));
4846 txg = spa_vdev_enter(spa);
4848 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4851 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4853 if (!vd->vdev_ops->vdev_op_leaf)
4854 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4856 pvd = vd->vdev_parent;
4859 * If the parent/child relationship is not as expected, don't do it.
4860 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4861 * vdev that's replacing B with C. The user's intent in replacing
4862 * is to go from M(A,B) to M(A,C). If the user decides to cancel
4863 * the replace by detaching C, the expected behavior is to end up
4864 * M(A,B). But suppose that right after deciding to detach C,
4865 * the replacement of B completes. We would have M(A,C), and then
4866 * ask to detach C, which would leave us with just A -- not what
4867 * the user wanted. To prevent this, we make sure that the
4868 * parent/child relationship hasn't changed -- in this example,
4869 * that C's parent is still the replacing vdev R.
4871 if (pvd->vdev_guid != pguid && pguid != 0)
4872 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4875 * Only 'replacing' or 'spare' vdevs can be replaced.
4877 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4878 pvd->vdev_ops != &vdev_spare_ops)
4879 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4881 ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4882 spa_version(spa) >= SPA_VERSION_SPARES);
4885 * Only mirror, replacing, and spare vdevs support detach.
4887 if (pvd->vdev_ops != &vdev_replacing_ops &&
4888 pvd->vdev_ops != &vdev_mirror_ops &&
4889 pvd->vdev_ops != &vdev_spare_ops)
4890 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4893 * If this device has the only valid copy of some data,
4894 * we cannot safely detach it.
4896 if (vdev_dtl_required(vd))
4897 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4899 ASSERT(pvd->vdev_children >= 2);
4902 * If we are detaching the second disk from a replacing vdev, then
4903 * check to see if we changed the original vdev's path to have "/old"
4904 * at the end in spa_vdev_attach(). If so, undo that change now.
4906 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4907 vd->vdev_path != NULL) {
4908 size_t len = strlen(vd->vdev_path);
4910 for (int c = 0; c < pvd->vdev_children; c++) {
4911 cvd = pvd->vdev_child[c];
4913 if (cvd == vd || cvd->vdev_path == NULL)
4916 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4917 strcmp(cvd->vdev_path + len, "/old") == 0) {
4918 spa_strfree(cvd->vdev_path);
4919 cvd->vdev_path = spa_strdup(vd->vdev_path);
4926 * If we are detaching the original disk from a spare, then it implies
4927 * that the spare should become a real disk, and be removed from the
4928 * active spare list for the pool.
4930 if (pvd->vdev_ops == &vdev_spare_ops &&
4932 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4936 * Erase the disk labels so the disk can be used for other things.
4937 * This must be done after all other error cases are handled,
4938 * but before we disembowel vd (so we can still do I/O to it).
4939 * But if we can't do it, don't treat the error as fatal --
4940 * it may be that the unwritability of the disk is the reason
4941 * it's being detached!
4943 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4946 * Remove vd from its parent and compact the parent's children.
4948 vdev_remove_child(pvd, vd);
4949 vdev_compact_children(pvd);
4952 * Remember one of the remaining children so we can get tvd below.
4954 cvd = pvd->vdev_child[pvd->vdev_children - 1];
4957 * If we need to remove the remaining child from the list of hot spares,
4958 * do it now, marking the vdev as no longer a spare in the process.
4959 * We must do this before vdev_remove_parent(), because that can
4960 * change the GUID if it creates a new toplevel GUID. For a similar
4961 * reason, we must remove the spare now, in the same txg as the detach;
4962 * otherwise someone could attach a new sibling, change the GUID, and
4963 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4966 ASSERT(cvd->vdev_isspare);
4967 spa_spare_remove(cvd);
4968 unspare_guid = cvd->vdev_guid;
4969 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4970 cvd->vdev_unspare = B_TRUE;
4974 * If the parent mirror/replacing vdev only has one child,
4975 * the parent is no longer needed. Remove it from the tree.
4977 if (pvd->vdev_children == 1) {
4978 if (pvd->vdev_ops == &vdev_spare_ops)
4979 cvd->vdev_unspare = B_FALSE;
4980 vdev_remove_parent(cvd);
4985 * We don't set tvd until now because the parent we just removed
4986 * may have been the previous top-level vdev.
4988 tvd = cvd->vdev_top;
4989 ASSERT(tvd->vdev_parent == rvd);
4992 * Reevaluate the parent vdev state.
4994 vdev_propagate_state(cvd);
4997 * If the 'autoexpand' property is set on the pool then automatically
4998 * try to expand the size of the pool. For example if the device we
4999 * just detached was smaller than the others, it may be possible to
5000 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
5001 * first so that we can obtain the updated sizes of the leaf vdevs.
5003 if (spa->spa_autoexpand) {
5005 vdev_expand(tvd, txg);
5008 vdev_config_dirty(tvd);
5011 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
5012 * vd->vdev_detached is set and free vd's DTL object in syncing context.
5013 * But first make sure we're not on any *other* txg's DTL list, to
5014 * prevent vd from being accessed after it's freed.
5016 vdpath = spa_strdup(vd->vdev_path);
5017 for (int t = 0; t < TXG_SIZE; t++)
5018 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
5019 vd->vdev_detached = B_TRUE;
5020 vdev_dirty(tvd, VDD_DTL, vd, txg);
5022 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
5024 /* hang on to the spa before we release the lock */
5025 spa_open_ref(spa, FTAG);
5027 error = spa_vdev_exit(spa, vd, txg, 0);
5029 spa_history_log_internal(spa, "detach", NULL,
5031 spa_strfree(vdpath);
5034 * If this was the removal of the original device in a hot spare vdev,
5035 * then we want to go through and remove the device from the hot spare
5036 * list of every other pool.
5039 spa_t *altspa = NULL;
5041 mutex_enter(&spa_namespace_lock);
5042 while ((altspa = spa_next(altspa)) != NULL) {
5043 if (altspa->spa_state != POOL_STATE_ACTIVE ||
5047 spa_open_ref(altspa, FTAG);
5048 mutex_exit(&spa_namespace_lock);
5049 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
5050 mutex_enter(&spa_namespace_lock);
5051 spa_close(altspa, FTAG);
5053 mutex_exit(&spa_namespace_lock);
5055 /* search the rest of the vdevs for spares to remove */
5056 spa_vdev_resilver_done(spa);
5059 /* all done with the spa; OK to release */
5060 mutex_enter(&spa_namespace_lock);
5061 spa_close(spa, FTAG);
5062 mutex_exit(&spa_namespace_lock);
5068 * Split a set of devices from their mirrors, and create a new pool from them.
5071 spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
5072 nvlist_t *props, boolean_t exp)
5075 uint64_t txg, *glist;
5077 uint_t c, children, lastlog;
5078 nvlist_t **child, *nvl, *tmp;
5080 char *altroot = NULL;
5081 vdev_t *rvd, **vml = NULL; /* vdev modify list */
5082 boolean_t activate_slog;
5084 ASSERT(spa_writeable(spa));
5086 txg = spa_vdev_enter(spa);
5088 /* clear the log and flush everything up to now */
5089 activate_slog = spa_passivate_log(spa);
5090 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5091 error = spa_offline_log(spa);
5092 txg = spa_vdev_config_enter(spa);
5095 spa_activate_log(spa);
5098 return (spa_vdev_exit(spa, NULL, txg, error));
5100 /* check new spa name before going any further */
5101 if (spa_lookup(newname) != NULL)
5102 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
5105 * scan through all the children to ensure they're all mirrors
5107 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
5108 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
5110 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5112 /* first, check to ensure we've got the right child count */
5113 rvd = spa->spa_root_vdev;
5115 for (c = 0; c < rvd->vdev_children; c++) {
5116 vdev_t *vd = rvd->vdev_child[c];
5118 /* don't count the holes & logs as children */
5119 if (vd->vdev_islog || vd->vdev_ishole) {
5127 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
5128 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5130 /* next, ensure no spare or cache devices are part of the split */
5131 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
5132 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
5133 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5135 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
5136 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
5138 /* then, loop over each vdev and validate it */
5139 for (c = 0; c < children; c++) {
5140 uint64_t is_hole = 0;
5142 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
5146 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
5147 spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
5150 error = SET_ERROR(EINVAL);
5155 /* which disk is going to be split? */
5156 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
5158 error = SET_ERROR(EINVAL);
5162 /* look it up in the spa */
5163 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
5164 if (vml[c] == NULL) {
5165 error = SET_ERROR(ENODEV);
5169 /* make sure there's nothing stopping the split */
5170 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
5171 vml[c]->vdev_islog ||
5172 vml[c]->vdev_ishole ||
5173 vml[c]->vdev_isspare ||
5174 vml[c]->vdev_isl2cache ||
5175 !vdev_writeable(vml[c]) ||
5176 vml[c]->vdev_children != 0 ||
5177 vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
5178 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
5179 error = SET_ERROR(EINVAL);
5183 if (vdev_dtl_required(vml[c])) {
5184 error = SET_ERROR(EBUSY);
5188 /* we need certain info from the top level */
5189 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
5190 vml[c]->vdev_top->vdev_ms_array) == 0);
5191 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
5192 vml[c]->vdev_top->vdev_ms_shift) == 0);
5193 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
5194 vml[c]->vdev_top->vdev_asize) == 0);
5195 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
5196 vml[c]->vdev_top->vdev_ashift) == 0);
5200 kmem_free(vml, children * sizeof (vdev_t *));
5201 kmem_free(glist, children * sizeof (uint64_t));
5202 return (spa_vdev_exit(spa, NULL, txg, error));
5205 /* stop writers from using the disks */
5206 for (c = 0; c < children; c++) {
5208 vml[c]->vdev_offline = B_TRUE;
5210 vdev_reopen(spa->spa_root_vdev);
5213 * Temporarily record the splitting vdevs in the spa config. This
5214 * will disappear once the config is regenerated.
5216 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5217 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
5218 glist, children) == 0);
5219 kmem_free(glist, children * sizeof (uint64_t));
5221 mutex_enter(&spa->spa_props_lock);
5222 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
5224 mutex_exit(&spa->spa_props_lock);
5225 spa->spa_config_splitting = nvl;
5226 vdev_config_dirty(spa->spa_root_vdev);
5228 /* configure and create the new pool */
5229 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
5230 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
5231 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
5232 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5233 spa_version(spa)) == 0);
5234 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
5235 spa->spa_config_txg) == 0);
5236 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
5237 spa_generate_guid(NULL)) == 0);
5238 (void) nvlist_lookup_string(props,
5239 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
5241 /* add the new pool to the namespace */
5242 newspa = spa_add(newname, config, altroot);
5243 newspa->spa_config_txg = spa->spa_config_txg;
5244 spa_set_log_state(newspa, SPA_LOG_CLEAR);
5246 /* release the spa config lock, retaining the namespace lock */
5247 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5249 if (zio_injection_enabled)
5250 zio_handle_panic_injection(spa, FTAG, 1);
5252 spa_activate(newspa, spa_mode_global);
5253 spa_async_suspend(newspa);
5256 /* mark that we are creating new spa by splitting */
5257 newspa->spa_splitting_newspa = B_TRUE;
5259 /* create the new pool from the disks of the original pool */
5260 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
5262 newspa->spa_splitting_newspa = B_FALSE;
5267 /* if that worked, generate a real config for the new pool */
5268 if (newspa->spa_root_vdev != NULL) {
5269 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
5270 NV_UNIQUE_NAME, KM_SLEEP) == 0);
5271 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
5272 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
5273 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
5278 if (props != NULL) {
5279 spa_configfile_set(newspa, props, B_FALSE);
5280 error = spa_prop_set(newspa, props);
5285 /* flush everything */
5286 txg = spa_vdev_config_enter(newspa);
5287 vdev_config_dirty(newspa->spa_root_vdev);
5288 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
5290 if (zio_injection_enabled)
5291 zio_handle_panic_injection(spa, FTAG, 2);
5293 spa_async_resume(newspa);
5295 /* finally, update the original pool's config */
5296 txg = spa_vdev_config_enter(spa);
5297 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
5298 error = dmu_tx_assign(tx, TXG_WAIT);
5301 for (c = 0; c < children; c++) {
5302 if (vml[c] != NULL) {
5305 spa_history_log_internal(spa, "detach", tx,
5306 "vdev=%s", vml[c]->vdev_path);
5310 vdev_config_dirty(spa->spa_root_vdev);
5311 spa->spa_config_splitting = NULL;
5315 (void) spa_vdev_exit(spa, NULL, txg, 0);
5317 if (zio_injection_enabled)
5318 zio_handle_panic_injection(spa, FTAG, 3);
5320 /* split is complete; log a history record */
5321 spa_history_log_internal(newspa, "split", NULL,
5322 "from pool %s", spa_name(spa));
5324 kmem_free(vml, children * sizeof (vdev_t *));
5326 /* if we're not going to mount the filesystems in userland, export */
5328 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
5335 spa_deactivate(newspa);
5338 txg = spa_vdev_config_enter(spa);
5340 /* re-online all offlined disks */
5341 for (c = 0; c < children; c++) {
5343 vml[c]->vdev_offline = B_FALSE;
5345 vdev_reopen(spa->spa_root_vdev);
5347 nvlist_free(spa->spa_config_splitting);
5348 spa->spa_config_splitting = NULL;
5349 (void) spa_vdev_exit(spa, NULL, txg, error);
5351 kmem_free(vml, children * sizeof (vdev_t *));
5356 spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
5358 for (int i = 0; i < count; i++) {
5361 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
5364 if (guid == target_guid)
5372 spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
5373 nvlist_t *dev_to_remove)
5375 nvlist_t **newdev = NULL;
5378 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
5380 for (int i = 0, j = 0; i < count; i++) {
5381 if (dev[i] == dev_to_remove)
5383 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
5386 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
5387 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
5389 for (int i = 0; i < count - 1; i++)
5390 nvlist_free(newdev[i]);
5393 kmem_free(newdev, (count - 1) * sizeof (void *));
5397 * Evacuate the device.
5400 spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
5405 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5406 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5407 ASSERT(vd == vd->vdev_top);
5410 * Evacuate the device. We don't hold the config lock as writer
5411 * since we need to do I/O but we do keep the
5412 * spa_namespace_lock held. Once this completes the device
5413 * should no longer have any blocks allocated on it.
5415 if (vd->vdev_islog) {
5416 if (vd->vdev_stat.vs_alloc != 0)
5417 error = spa_offline_log(spa);
5419 error = SET_ERROR(ENOTSUP);
5426 * The evacuation succeeded. Remove any remaining MOS metadata
5427 * associated with this vdev, and wait for these changes to sync.
5429 ASSERT0(vd->vdev_stat.vs_alloc);
5430 txg = spa_vdev_config_enter(spa);
5431 vd->vdev_removing = B_TRUE;
5432 vdev_dirty_leaves(vd, VDD_DTL, txg);
5433 vdev_config_dirty(vd);
5434 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5440 * Complete the removal by cleaning up the namespace.
5443 spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
5445 vdev_t *rvd = spa->spa_root_vdev;
5446 uint64_t id = vd->vdev_id;
5447 boolean_t last_vdev = (id == (rvd->vdev_children - 1));
5449 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5450 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5451 ASSERT(vd == vd->vdev_top);
5454 * Only remove any devices which are empty.
5456 if (vd->vdev_stat.vs_alloc != 0)
5459 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5461 if (list_link_active(&vd->vdev_state_dirty_node))
5462 vdev_state_clean(vd);
5463 if (list_link_active(&vd->vdev_config_dirty_node))
5464 vdev_config_clean(vd);
5469 vdev_compact_children(rvd);
5471 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
5472 vdev_add_child(rvd, vd);
5474 vdev_config_dirty(rvd);
5477 * Reassess the health of our root vdev.
5483 * Remove a device from the pool -
5485 * Removing a device from the vdev namespace requires several steps
5486 * and can take a significant amount of time. As a result we use
5487 * the spa_vdev_config_[enter/exit] functions which allow us to
5488 * grab and release the spa_config_lock while still holding the namespace
5489 * lock. During each step the configuration is synced out.
5491 * Currently, this supports removing only hot spares, slogs, and level 2 ARC
5495 spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
5498 metaslab_group_t *mg;
5499 nvlist_t **spares, **l2cache, *nv;
5501 uint_t nspares, nl2cache;
5503 boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
5505 ASSERT(spa_writeable(spa));
5508 txg = spa_vdev_enter(spa);
5510 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5512 if (spa->spa_spares.sav_vdevs != NULL &&
5513 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5514 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
5515 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
5517 * Only remove the hot spare if it's not currently in use
5520 if (vd == NULL || unspare) {
5521 spa_vdev_remove_aux(spa->spa_spares.sav_config,
5522 ZPOOL_CONFIG_SPARES, spares, nspares, nv);
5523 spa_load_spares(spa);
5524 spa->spa_spares.sav_sync = B_TRUE;
5526 error = SET_ERROR(EBUSY);
5528 } else if (spa->spa_l2cache.sav_vdevs != NULL &&
5529 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5530 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
5531 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
5533 * Cache devices can always be removed.
5535 spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
5536 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
5537 spa_load_l2cache(spa);
5538 spa->spa_l2cache.sav_sync = B_TRUE;
5539 } else if (vd != NULL && vd->vdev_islog) {
5541 ASSERT(vd == vd->vdev_top);
5546 * Stop allocating from this vdev.
5548 metaslab_group_passivate(mg);
5551 * Wait for the youngest allocations and frees to sync,
5552 * and then wait for the deferral of those frees to finish.
5554 spa_vdev_config_exit(spa, NULL,
5555 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
5558 * Attempt to evacuate the vdev.
5560 error = spa_vdev_remove_evacuate(spa, vd);
5562 txg = spa_vdev_config_enter(spa);
5565 * If we couldn't evacuate the vdev, unwind.
5568 metaslab_group_activate(mg);
5569 return (spa_vdev_exit(spa, NULL, txg, error));
5573 * Clean up the vdev namespace.
5575 spa_vdev_remove_from_namespace(spa, vd);
5577 } else if (vd != NULL) {
5579 * Normal vdevs cannot be removed (yet).
5581 error = SET_ERROR(ENOTSUP);
5584 * There is no vdev of any kind with the specified guid.
5586 error = SET_ERROR(ENOENT);
5590 return (spa_vdev_exit(spa, NULL, txg, error));
5596 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5597 * currently spared, so we can detach it.
5600 spa_vdev_resilver_done_hunt(vdev_t *vd)
5602 vdev_t *newvd, *oldvd;
5604 for (int c = 0; c < vd->vdev_children; c++) {
5605 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
5611 * Check for a completed replacement. We always consider the first
5612 * vdev in the list to be the oldest vdev, and the last one to be
5613 * the newest (see spa_vdev_attach() for how that works). In
5614 * the case where the newest vdev is faulted, we will not automatically
5615 * remove it after a resilver completes. This is OK as it will require
5616 * user intervention to determine which disk the admin wishes to keep.
5618 if (vd->vdev_ops == &vdev_replacing_ops) {
5619 ASSERT(vd->vdev_children > 1);
5621 newvd = vd->vdev_child[vd->vdev_children - 1];
5622 oldvd = vd->vdev_child[0];
5624 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
5625 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5626 !vdev_dtl_required(oldvd))
5631 * Check for a completed resilver with the 'unspare' flag set.
5633 if (vd->vdev_ops == &vdev_spare_ops) {
5634 vdev_t *first = vd->vdev_child[0];
5635 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
5637 if (last->vdev_unspare) {
5640 } else if (first->vdev_unspare) {
5647 if (oldvd != NULL &&
5648 vdev_dtl_empty(newvd, DTL_MISSING) &&
5649 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5650 !vdev_dtl_required(oldvd))
5654 * If there are more than two spares attached to a disk,
5655 * and those spares are not required, then we want to
5656 * attempt to free them up now so that they can be used
5657 * by other pools. Once we're back down to a single
5658 * disk+spare, we stop removing them.
5660 if (vd->vdev_children > 2) {
5661 newvd = vd->vdev_child[1];
5663 if (newvd->vdev_isspare && last->vdev_isspare &&
5664 vdev_dtl_empty(last, DTL_MISSING) &&
5665 vdev_dtl_empty(last, DTL_OUTAGE) &&
5666 !vdev_dtl_required(newvd))
5675 spa_vdev_resilver_done(spa_t *spa)
5677 vdev_t *vd, *pvd, *ppvd;
5678 uint64_t guid, sguid, pguid, ppguid;
5680 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5682 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
5683 pvd = vd->vdev_parent;
5684 ppvd = pvd->vdev_parent;
5685 guid = vd->vdev_guid;
5686 pguid = pvd->vdev_guid;
5687 ppguid = ppvd->vdev_guid;
5690 * If we have just finished replacing a hot spared device, then
5691 * we need to detach the parent's first child (the original hot
5694 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
5695 ppvd->vdev_children == 2) {
5696 ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
5697 sguid = ppvd->vdev_child[1]->vdev_guid;
5699 ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd));
5701 spa_config_exit(spa, SCL_ALL, FTAG);
5702 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
5704 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
5706 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5709 spa_config_exit(spa, SCL_ALL, FTAG);
5713 * Update the stored path or FRU for this vdev.
5716 spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
5720 boolean_t sync = B_FALSE;
5722 ASSERT(spa_writeable(spa));
5724 spa_vdev_state_enter(spa, SCL_ALL);
5726 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
5727 return (spa_vdev_state_exit(spa, NULL, ENOENT));
5729 if (!vd->vdev_ops->vdev_op_leaf)
5730 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
5733 if (strcmp(value, vd->vdev_path) != 0) {
5734 spa_strfree(vd->vdev_path);
5735 vd->vdev_path = spa_strdup(value);
5739 if (vd->vdev_fru == NULL) {
5740 vd->vdev_fru = spa_strdup(value);
5742 } else if (strcmp(value, vd->vdev_fru) != 0) {
5743 spa_strfree(vd->vdev_fru);
5744 vd->vdev_fru = spa_strdup(value);
5749 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
5753 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
5755 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5759 spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5761 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5765 * ==========================================================================
5767 * ==========================================================================
5771 spa_scan_stop(spa_t *spa)
5773 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5774 if (dsl_scan_resilvering(spa->spa_dsl_pool))
5775 return (SET_ERROR(EBUSY));
5776 return (dsl_scan_cancel(spa->spa_dsl_pool));
5780 spa_scan(spa_t *spa, pool_scan_func_t func)
5782 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5784 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5785 return (SET_ERROR(ENOTSUP));
5788 * If a resilver was requested, but there is no DTL on a
5789 * writeable leaf device, we have nothing to do.
5791 if (func == POOL_SCAN_RESILVER &&
5792 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5793 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5797 return (dsl_scan(spa->spa_dsl_pool, func));
5801 * ==========================================================================
5802 * SPA async task processing
5803 * ==========================================================================
5807 spa_async_remove(spa_t *spa, vdev_t *vd)
5809 if (vd->vdev_remove_wanted) {
5810 vd->vdev_remove_wanted = B_FALSE;
5811 vd->vdev_delayed_close = B_FALSE;
5812 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5815 * We want to clear the stats, but we don't want to do a full
5816 * vdev_clear() as that will cause us to throw away
5817 * degraded/faulted state as well as attempt to reopen the
5818 * device, all of which is a waste.
5820 vd->vdev_stat.vs_read_errors = 0;
5821 vd->vdev_stat.vs_write_errors = 0;
5822 vd->vdev_stat.vs_checksum_errors = 0;
5824 vdev_state_dirty(vd->vdev_top);
5827 for (int c = 0; c < vd->vdev_children; c++)
5828 spa_async_remove(spa, vd->vdev_child[c]);
5832 spa_async_probe(spa_t *spa, vdev_t *vd)
5834 if (vd->vdev_probe_wanted) {
5835 vd->vdev_probe_wanted = B_FALSE;
5836 vdev_reopen(vd); /* vdev_open() does the actual probe */
5839 for (int c = 0; c < vd->vdev_children; c++)
5840 spa_async_probe(spa, vd->vdev_child[c]);
5844 spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5850 if (!spa->spa_autoexpand)
5853 for (int c = 0; c < vd->vdev_children; c++) {
5854 vdev_t *cvd = vd->vdev_child[c];
5855 spa_async_autoexpand(spa, cvd);
5858 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5861 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
5862 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
5864 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5865 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
5867 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
5868 ESC_ZFS_VDEV_AUTOEXPAND, attr, &eid, DDI_SLEEP);
5871 kmem_free(physpath, MAXPATHLEN);
5875 spa_async_thread(void *arg)
5880 ASSERT(spa->spa_sync_on);
5882 mutex_enter(&spa->spa_async_lock);
5883 tasks = spa->spa_async_tasks;
5884 spa->spa_async_tasks &= SPA_ASYNC_REMOVE;
5885 mutex_exit(&spa->spa_async_lock);
5888 * See if the config needs to be updated.
5890 if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5891 uint64_t old_space, new_space;
5893 mutex_enter(&spa_namespace_lock);
5894 old_space = metaslab_class_get_space(spa_normal_class(spa));
5895 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5896 new_space = metaslab_class_get_space(spa_normal_class(spa));
5897 mutex_exit(&spa_namespace_lock);
5900 * If the pool grew as a result of the config update,
5901 * then log an internal history event.
5903 if (new_space != old_space) {
5904 spa_history_log_internal(spa, "vdev online", NULL,
5905 "pool '%s' size: %llu(+%llu)",
5906 spa_name(spa), new_space, new_space - old_space);
5910 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5911 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5912 spa_async_autoexpand(spa, spa->spa_root_vdev);
5913 spa_config_exit(spa, SCL_CONFIG, FTAG);
5917 * See if any devices need to be probed.
5919 if (tasks & SPA_ASYNC_PROBE) {
5920 spa_vdev_state_enter(spa, SCL_NONE);
5921 spa_async_probe(spa, spa->spa_root_vdev);
5922 (void) spa_vdev_state_exit(spa, NULL, 0);
5926 * If any devices are done replacing, detach them.
5928 if (tasks & SPA_ASYNC_RESILVER_DONE)
5929 spa_vdev_resilver_done(spa);
5932 * Kick off a resilver.
5934 if (tasks & SPA_ASYNC_RESILVER)
5935 dsl_resilver_restart(spa->spa_dsl_pool, 0);
5938 * Let the world know that we're done.
5940 mutex_enter(&spa->spa_async_lock);
5941 spa->spa_async_thread = NULL;
5942 cv_broadcast(&spa->spa_async_cv);
5943 mutex_exit(&spa->spa_async_lock);
5948 spa_async_thread_vd(void *arg)
5953 ASSERT(spa->spa_sync_on);
5955 mutex_enter(&spa->spa_async_lock);
5956 tasks = spa->spa_async_tasks;
5958 spa->spa_async_tasks &= ~SPA_ASYNC_REMOVE;
5959 mutex_exit(&spa->spa_async_lock);
5962 * See if any devices need to be marked REMOVED.
5964 if (tasks & SPA_ASYNC_REMOVE) {
5965 spa_vdev_state_enter(spa, SCL_NONE);
5966 spa_async_remove(spa, spa->spa_root_vdev);
5967 for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
5968 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5969 for (int i = 0; i < spa->spa_spares.sav_count; i++)
5970 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5971 (void) spa_vdev_state_exit(spa, NULL, 0);
5975 * Let the world know that we're done.
5977 mutex_enter(&spa->spa_async_lock);
5978 tasks = spa->spa_async_tasks;
5979 if ((tasks & SPA_ASYNC_REMOVE) != 0)
5981 spa->spa_async_thread_vd = NULL;
5982 cv_broadcast(&spa->spa_async_cv);
5983 mutex_exit(&spa->spa_async_lock);
5988 spa_async_suspend(spa_t *spa)
5990 mutex_enter(&spa->spa_async_lock);
5991 spa->spa_async_suspended++;
5992 while (spa->spa_async_thread != NULL &&
5993 spa->spa_async_thread_vd != NULL)
5994 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5995 mutex_exit(&spa->spa_async_lock);
5999 spa_async_resume(spa_t *spa)
6001 mutex_enter(&spa->spa_async_lock);
6002 ASSERT(spa->spa_async_suspended != 0);
6003 spa->spa_async_suspended--;
6004 mutex_exit(&spa->spa_async_lock);
6008 spa_async_tasks_pending(spa_t *spa)
6010 uint_t non_config_tasks;
6012 boolean_t config_task_suspended;
6014 non_config_tasks = spa->spa_async_tasks & ~(SPA_ASYNC_CONFIG_UPDATE |
6016 config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE;
6017 if (spa->spa_ccw_fail_time == 0) {
6018 config_task_suspended = B_FALSE;
6020 config_task_suspended =
6021 (gethrtime() - spa->spa_ccw_fail_time) <
6022 (zfs_ccw_retry_interval * NANOSEC);
6025 return (non_config_tasks || (config_task && !config_task_suspended));
6029 spa_async_dispatch(spa_t *spa)
6031 mutex_enter(&spa->spa_async_lock);
6032 if (spa_async_tasks_pending(spa) &&
6033 !spa->spa_async_suspended &&
6034 spa->spa_async_thread == NULL &&
6036 spa->spa_async_thread = thread_create(NULL, 0,
6037 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
6038 mutex_exit(&spa->spa_async_lock);
6042 spa_async_dispatch_vd(spa_t *spa)
6044 mutex_enter(&spa->spa_async_lock);
6045 if ((spa->spa_async_tasks & SPA_ASYNC_REMOVE) != 0 &&
6046 !spa->spa_async_suspended &&
6047 spa->spa_async_thread_vd == NULL &&
6049 spa->spa_async_thread_vd = thread_create(NULL, 0,
6050 spa_async_thread_vd, spa, 0, &p0, TS_RUN, maxclsyspri);
6051 mutex_exit(&spa->spa_async_lock);
6055 spa_async_request(spa_t *spa, int task)
6057 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
6058 mutex_enter(&spa->spa_async_lock);
6059 spa->spa_async_tasks |= task;
6060 mutex_exit(&spa->spa_async_lock);
6061 spa_async_dispatch_vd(spa);
6065 * ==========================================================================
6066 * SPA syncing routines
6067 * ==========================================================================
6071 bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
6074 bpobj_enqueue(bpo, bp, tx);
6079 spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
6083 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
6084 BP_GET_PSIZE(bp), zio->io_flags));
6089 * Note: this simple function is not inlined to make it easier to dtrace the
6090 * amount of time spent syncing frees.
6093 spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx)
6095 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6096 bplist_iterate(bpl, spa_free_sync_cb, zio, tx);
6097 VERIFY(zio_wait(zio) == 0);
6101 * Note: this simple function is not inlined to make it easier to dtrace the
6102 * amount of time spent syncing deferred frees.
6105 spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx)
6107 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6108 VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj,
6109 spa_free_sync_cb, zio, tx), ==, 0);
6110 VERIFY0(zio_wait(zio));
6115 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
6117 char *packed = NULL;
6122 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
6125 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
6126 * information. This avoids the dmu_buf_will_dirty() path and
6127 * saves us a pre-read to get data we don't actually care about.
6129 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
6130 packed = kmem_alloc(bufsize, KM_SLEEP);
6132 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
6134 bzero(packed + nvsize, bufsize - nvsize);
6136 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
6138 kmem_free(packed, bufsize);
6140 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
6141 dmu_buf_will_dirty(db, tx);
6142 *(uint64_t *)db->db_data = nvsize;
6143 dmu_buf_rele(db, FTAG);
6147 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
6148 const char *config, const char *entry)
6158 * Update the MOS nvlist describing the list of available devices.
6159 * spa_validate_aux() will have already made sure this nvlist is
6160 * valid and the vdevs are labeled appropriately.
6162 if (sav->sav_object == 0) {
6163 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
6164 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
6165 sizeof (uint64_t), tx);
6166 VERIFY(zap_update(spa->spa_meta_objset,
6167 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
6168 &sav->sav_object, tx) == 0);
6171 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
6172 if (sav->sav_count == 0) {
6173 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
6175 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
6176 for (i = 0; i < sav->sav_count; i++)
6177 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
6178 B_FALSE, VDEV_CONFIG_L2CACHE);
6179 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
6180 sav->sav_count) == 0);
6181 for (i = 0; i < sav->sav_count; i++)
6182 nvlist_free(list[i]);
6183 kmem_free(list, sav->sav_count * sizeof (void *));
6186 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
6187 nvlist_free(nvroot);
6189 sav->sav_sync = B_FALSE;
6193 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
6197 if (list_is_empty(&spa->spa_config_dirty_list))
6200 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6202 config = spa_config_generate(spa, spa->spa_root_vdev,
6203 dmu_tx_get_txg(tx), B_FALSE);
6206 * If we're upgrading the spa version then make sure that
6207 * the config object gets updated with the correct version.
6209 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
6210 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
6211 spa->spa_uberblock.ub_version);
6213 spa_config_exit(spa, SCL_STATE, FTAG);
6215 if (spa->spa_config_syncing)
6216 nvlist_free(spa->spa_config_syncing);
6217 spa->spa_config_syncing = config;
6219 spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
6223 spa_sync_version(void *arg, dmu_tx_t *tx)
6225 uint64_t *versionp = arg;
6226 uint64_t version = *versionp;
6227 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
6230 * Setting the version is special cased when first creating the pool.
6232 ASSERT(tx->tx_txg != TXG_INITIAL);
6234 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
6235 ASSERT(version >= spa_version(spa));
6237 spa->spa_uberblock.ub_version = version;
6238 vdev_config_dirty(spa->spa_root_vdev);
6239 spa_history_log_internal(spa, "set", tx, "version=%lld", version);
6243 * Set zpool properties.
6246 spa_sync_props(void *arg, dmu_tx_t *tx)
6248 nvlist_t *nvp = arg;
6249 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
6250 objset_t *mos = spa->spa_meta_objset;
6251 nvpair_t *elem = NULL;
6253 mutex_enter(&spa->spa_props_lock);
6255 while ((elem = nvlist_next_nvpair(nvp, elem))) {
6257 char *strval, *fname;
6259 const char *propname;
6260 zprop_type_t proptype;
6263 switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
6266 * We checked this earlier in spa_prop_validate().
6268 ASSERT(zpool_prop_feature(nvpair_name(elem)));
6270 fname = strchr(nvpair_name(elem), '@') + 1;
6271 VERIFY0(zfeature_lookup_name(fname, &fid));
6273 spa_feature_enable(spa, fid, tx);
6274 spa_history_log_internal(spa, "set", tx,
6275 "%s=enabled", nvpair_name(elem));
6278 case ZPOOL_PROP_VERSION:
6279 intval = fnvpair_value_uint64(elem);
6281 * The version is synced seperatly before other
6282 * properties and should be correct by now.
6284 ASSERT3U(spa_version(spa), >=, intval);
6287 case ZPOOL_PROP_ALTROOT:
6289 * 'altroot' is a non-persistent property. It should
6290 * have been set temporarily at creation or import time.
6292 ASSERT(spa->spa_root != NULL);
6295 case ZPOOL_PROP_READONLY:
6296 case ZPOOL_PROP_CACHEFILE:
6298 * 'readonly' and 'cachefile' are also non-persisitent
6302 case ZPOOL_PROP_COMMENT:
6303 strval = fnvpair_value_string(elem);
6304 if (spa->spa_comment != NULL)
6305 spa_strfree(spa->spa_comment);
6306 spa->spa_comment = spa_strdup(strval);
6308 * We need to dirty the configuration on all the vdevs
6309 * so that their labels get updated. It's unnecessary
6310 * to do this for pool creation since the vdev's
6311 * configuratoin has already been dirtied.
6313 if (tx->tx_txg != TXG_INITIAL)
6314 vdev_config_dirty(spa->spa_root_vdev);
6315 spa_history_log_internal(spa, "set", tx,
6316 "%s=%s", nvpair_name(elem), strval);
6320 * Set pool property values in the poolprops mos object.
6322 if (spa->spa_pool_props_object == 0) {
6323 spa->spa_pool_props_object =
6324 zap_create_link(mos, DMU_OT_POOL_PROPS,
6325 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
6329 /* normalize the property name */
6330 propname = zpool_prop_to_name(prop);
6331 proptype = zpool_prop_get_type(prop);
6333 if (nvpair_type(elem) == DATA_TYPE_STRING) {
6334 ASSERT(proptype == PROP_TYPE_STRING);
6335 strval = fnvpair_value_string(elem);
6336 VERIFY0(zap_update(mos,
6337 spa->spa_pool_props_object, propname,
6338 1, strlen(strval) + 1, strval, tx));
6339 spa_history_log_internal(spa, "set", tx,
6340 "%s=%s", nvpair_name(elem), strval);
6341 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
6342 intval = fnvpair_value_uint64(elem);
6344 if (proptype == PROP_TYPE_INDEX) {
6346 VERIFY0(zpool_prop_index_to_string(
6347 prop, intval, &unused));
6349 VERIFY0(zap_update(mos,
6350 spa->spa_pool_props_object, propname,
6351 8, 1, &intval, tx));
6352 spa_history_log_internal(spa, "set", tx,
6353 "%s=%lld", nvpair_name(elem), intval);
6355 ASSERT(0); /* not allowed */
6359 case ZPOOL_PROP_DELEGATION:
6360 spa->spa_delegation = intval;
6362 case ZPOOL_PROP_BOOTFS:
6363 spa->spa_bootfs = intval;
6365 case ZPOOL_PROP_FAILUREMODE:
6366 spa->spa_failmode = intval;
6368 case ZPOOL_PROP_AUTOEXPAND:
6369 spa->spa_autoexpand = intval;
6370 if (tx->tx_txg != TXG_INITIAL)
6371 spa_async_request(spa,
6372 SPA_ASYNC_AUTOEXPAND);
6374 case ZPOOL_PROP_DEDUPDITTO:
6375 spa->spa_dedup_ditto = intval;
6384 mutex_exit(&spa->spa_props_lock);
6388 * Perform one-time upgrade on-disk changes. spa_version() does not
6389 * reflect the new version this txg, so there must be no changes this
6390 * txg to anything that the upgrade code depends on after it executes.
6391 * Therefore this must be called after dsl_pool_sync() does the sync
6395 spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
6397 dsl_pool_t *dp = spa->spa_dsl_pool;
6399 ASSERT(spa->spa_sync_pass == 1);
6401 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
6403 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
6404 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
6405 dsl_pool_create_origin(dp, tx);
6407 /* Keeping the origin open increases spa_minref */
6408 spa->spa_minref += 3;
6411 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
6412 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
6413 dsl_pool_upgrade_clones(dp, tx);
6416 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
6417 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
6418 dsl_pool_upgrade_dir_clones(dp, tx);
6420 /* Keeping the freedir open increases spa_minref */
6421 spa->spa_minref += 3;
6424 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES &&
6425 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
6426 spa_feature_create_zap_objects(spa, tx);
6428 rrw_exit(&dp->dp_config_rwlock, FTAG);
6432 * Sync the specified transaction group. New blocks may be dirtied as
6433 * part of the process, so we iterate until it converges.
6436 spa_sync(spa_t *spa, uint64_t txg)
6438 dsl_pool_t *dp = spa->spa_dsl_pool;
6439 objset_t *mos = spa->spa_meta_objset;
6440 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
6441 vdev_t *rvd = spa->spa_root_vdev;
6446 VERIFY(spa_writeable(spa));
6449 * Lock out configuration changes.
6451 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
6453 spa->spa_syncing_txg = txg;
6454 spa->spa_sync_pass = 0;
6457 * If there are any pending vdev state changes, convert them
6458 * into config changes that go out with this transaction group.
6460 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6461 while (list_head(&spa->spa_state_dirty_list) != NULL) {
6463 * We need the write lock here because, for aux vdevs,
6464 * calling vdev_config_dirty() modifies sav_config.
6465 * This is ugly and will become unnecessary when we
6466 * eliminate the aux vdev wart by integrating all vdevs
6467 * into the root vdev tree.
6469 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6470 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
6471 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
6472 vdev_state_clean(vd);
6473 vdev_config_dirty(vd);
6475 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6476 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
6478 spa_config_exit(spa, SCL_STATE, FTAG);
6480 tx = dmu_tx_create_assigned(dp, txg);
6482 spa->spa_sync_starttime = gethrtime();
6484 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid,
6485 spa->spa_sync_starttime + spa->spa_deadman_synctime));
6488 callout_reset(&spa->spa_deadman_cycid,
6489 hz * spa->spa_deadman_synctime / NANOSEC, spa_deadman, spa);
6494 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
6495 * set spa_deflate if we have no raid-z vdevs.
6497 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
6498 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
6501 for (i = 0; i < rvd->vdev_children; i++) {
6502 vd = rvd->vdev_child[i];
6503 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
6506 if (i == rvd->vdev_children) {
6507 spa->spa_deflate = TRUE;
6508 VERIFY(0 == zap_add(spa->spa_meta_objset,
6509 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
6510 sizeof (uint64_t), 1, &spa->spa_deflate, tx));
6515 * If anything has changed in this txg, or if someone is waiting
6516 * for this txg to sync (eg, spa_vdev_remove()), push the
6517 * deferred frees from the previous txg. If not, leave them
6518 * alone so that we don't generate work on an otherwise idle
6521 if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
6522 !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
6523 !txg_list_empty(&dp->dp_sync_tasks, txg) ||
6524 ((dsl_scan_active(dp->dp_scan) ||
6525 txg_sync_waiting(dp)) && !spa_shutting_down(spa))) {
6526 spa_sync_deferred_frees(spa, tx);
6530 * Iterate to convergence.
6533 int pass = ++spa->spa_sync_pass;
6535 spa_sync_config_object(spa, tx);
6536 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
6537 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
6538 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
6539 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
6540 spa_errlog_sync(spa, txg);
6541 dsl_pool_sync(dp, txg);
6543 if (pass < zfs_sync_pass_deferred_free) {
6544 spa_sync_frees(spa, free_bpl, tx);
6546 bplist_iterate(free_bpl, bpobj_enqueue_cb,
6547 &spa->spa_deferred_bpobj, tx);
6551 dsl_scan_sync(dp, tx);
6553 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
6557 spa_sync_upgrades(spa, tx);
6559 } while (dmu_objset_is_dirty(mos, txg));
6562 * Rewrite the vdev configuration (which includes the uberblock)
6563 * to commit the transaction group.
6565 * If there are no dirty vdevs, we sync the uberblock to a few
6566 * random top-level vdevs that are known to be visible in the
6567 * config cache (see spa_vdev_add() for a complete description).
6568 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
6572 * We hold SCL_STATE to prevent vdev open/close/etc.
6573 * while we're attempting to write the vdev labels.
6575 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6577 if (list_is_empty(&spa->spa_config_dirty_list)) {
6578 vdev_t *svd[SPA_DVAS_PER_BP];
6580 int children = rvd->vdev_children;
6581 int c0 = spa_get_random(children);
6583 for (int c = 0; c < children; c++) {
6584 vd = rvd->vdev_child[(c0 + c) % children];
6585 if (vd->vdev_ms_array == 0 || vd->vdev_islog)
6587 svd[svdcount++] = vd;
6588 if (svdcount == SPA_DVAS_PER_BP)
6591 error = vdev_config_sync(svd, svdcount, txg, B_FALSE);
6593 error = vdev_config_sync(svd, svdcount, txg,
6596 error = vdev_config_sync(rvd->vdev_child,
6597 rvd->vdev_children, txg, B_FALSE);
6599 error = vdev_config_sync(rvd->vdev_child,
6600 rvd->vdev_children, txg, B_TRUE);
6604 spa->spa_last_synced_guid = rvd->vdev_guid;
6606 spa_config_exit(spa, SCL_STATE, FTAG);
6610 zio_suspend(spa, NULL);
6611 zio_resume_wait(spa);
6616 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY));
6619 callout_drain(&spa->spa_deadman_cycid);
6624 * Clear the dirty config list.
6626 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
6627 vdev_config_clean(vd);
6630 * Now that the new config has synced transactionally,
6631 * let it become visible to the config cache.
6633 if (spa->spa_config_syncing != NULL) {
6634 spa_config_set(spa, spa->spa_config_syncing);
6635 spa->spa_config_txg = txg;
6636 spa->spa_config_syncing = NULL;
6639 spa->spa_ubsync = spa->spa_uberblock;
6641 dsl_pool_sync_done(dp, txg);
6644 * Update usable space statistics.
6646 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
6647 vdev_sync_done(vd, txg);
6649 spa_update_dspace(spa);
6652 * It had better be the case that we didn't dirty anything
6653 * since vdev_config_sync().
6655 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
6656 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
6657 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
6659 spa->spa_sync_pass = 0;
6661 spa_config_exit(spa, SCL_CONFIG, FTAG);
6663 spa_handle_ignored_writes(spa);
6666 * If any async tasks have been requested, kick them off.
6668 spa_async_dispatch(spa);
6669 spa_async_dispatch_vd(spa);
6673 * Sync all pools. We don't want to hold the namespace lock across these
6674 * operations, so we take a reference on the spa_t and drop the lock during the
6678 spa_sync_allpools(void)
6681 mutex_enter(&spa_namespace_lock);
6682 while ((spa = spa_next(spa)) != NULL) {
6683 if (spa_state(spa) != POOL_STATE_ACTIVE ||
6684 !spa_writeable(spa) || spa_suspended(spa))
6686 spa_open_ref(spa, FTAG);
6687 mutex_exit(&spa_namespace_lock);
6688 txg_wait_synced(spa_get_dsl(spa), 0);
6689 mutex_enter(&spa_namespace_lock);
6690 spa_close(spa, FTAG);
6692 mutex_exit(&spa_namespace_lock);
6696 * ==========================================================================
6697 * Miscellaneous routines
6698 * ==========================================================================
6702 * Remove all pools in the system.
6710 * Remove all cached state. All pools should be closed now,
6711 * so every spa in the AVL tree should be unreferenced.
6713 mutex_enter(&spa_namespace_lock);
6714 while ((spa = spa_next(NULL)) != NULL) {
6716 * Stop async tasks. The async thread may need to detach
6717 * a device that's been replaced, which requires grabbing
6718 * spa_namespace_lock, so we must drop it here.
6720 spa_open_ref(spa, FTAG);
6721 mutex_exit(&spa_namespace_lock);
6722 spa_async_suspend(spa);
6723 mutex_enter(&spa_namespace_lock);
6724 spa_close(spa, FTAG);
6726 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
6728 spa_deactivate(spa);
6732 mutex_exit(&spa_namespace_lock);
6736 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
6741 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
6745 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
6746 vd = spa->spa_l2cache.sav_vdevs[i];
6747 if (vd->vdev_guid == guid)
6751 for (i = 0; i < spa->spa_spares.sav_count; i++) {
6752 vd = spa->spa_spares.sav_vdevs[i];
6753 if (vd->vdev_guid == guid)
6762 spa_upgrade(spa_t *spa, uint64_t version)
6764 ASSERT(spa_writeable(spa));
6766 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6769 * This should only be called for a non-faulted pool, and since a
6770 * future version would result in an unopenable pool, this shouldn't be
6773 ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version));
6774 ASSERT3U(version, >=, spa->spa_uberblock.ub_version);
6776 spa->spa_uberblock.ub_version = version;
6777 vdev_config_dirty(spa->spa_root_vdev);
6779 spa_config_exit(spa, SCL_ALL, FTAG);
6781 txg_wait_synced(spa_get_dsl(spa), 0);
6785 spa_has_spare(spa_t *spa, uint64_t guid)
6789 spa_aux_vdev_t *sav = &spa->spa_spares;
6791 for (i = 0; i < sav->sav_count; i++)
6792 if (sav->sav_vdevs[i]->vdev_guid == guid)
6795 for (i = 0; i < sav->sav_npending; i++) {
6796 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
6797 &spareguid) == 0 && spareguid == guid)
6805 * Check if a pool has an active shared spare device.
6806 * Note: reference count of an active spare is 2, as a spare and as a replace
6809 spa_has_active_shared_spare(spa_t *spa)
6813 spa_aux_vdev_t *sav = &spa->spa_spares;
6815 for (i = 0; i < sav->sav_count; i++) {
6816 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
6817 &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
6826 * Post a sysevent corresponding to the given event. The 'name' must be one of
6827 * the event definitions in sys/sysevent/eventdefs.h. The payload will be
6828 * filled in from the spa and (optionally) the vdev. This doesn't do anything
6829 * in the userland libzpool, as we don't want consumers to misinterpret ztest
6830 * or zdb as real changes.
6833 spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
6837 sysevent_attr_list_t *attr = NULL;
6838 sysevent_value_t value;
6841 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
6844 value.value_type = SE_DATA_TYPE_STRING;
6845 value.value.sv_string = spa_name(spa);
6846 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
6849 value.value_type = SE_DATA_TYPE_UINT64;
6850 value.value.sv_uint64 = spa_guid(spa);
6851 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
6855 value.value_type = SE_DATA_TYPE_UINT64;
6856 value.value.sv_uint64 = vd->vdev_guid;
6857 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
6861 if (vd->vdev_path) {
6862 value.value_type = SE_DATA_TYPE_STRING;
6863 value.value.sv_string = vd->vdev_path;
6864 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
6865 &value, SE_SLEEP) != 0)
6870 if (sysevent_attach_attributes(ev, attr) != 0)
6874 (void) log_sysevent(ev, SE_SLEEP, &eid);
6878 sysevent_free_attr(attr);