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 (c) 2015, Nexenta Systems, Inc. All rights reserved.
26 * Copyright (c) 2013 Martin Matuska <mm@FreeBSD.org>. All rights reserved.
27 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
28 * Copyright 2013 Saso Kiselkov. All rights reserved.
29 * Copyright (c) 2014 Integros [integros.com]
33 * SPA: Storage Pool Allocator
35 * This file contains all the routines used when modifying on-disk SPA state.
36 * This includes opening, importing, destroying, exporting a pool, and syncing a
40 #include <sys/zfs_context.h>
41 #include <sys/fm/fs/zfs.h>
42 #include <sys/spa_impl.h>
44 #include <sys/zio_checksum.h>
46 #include <sys/dmu_tx.h>
50 #include <sys/vdev_impl.h>
51 #include <sys/metaslab.h>
52 #include <sys/metaslab_impl.h>
53 #include <sys/uberblock_impl.h>
56 #include <sys/dmu_traverse.h>
57 #include <sys/dmu_objset.h>
58 #include <sys/unique.h>
59 #include <sys/dsl_pool.h>
60 #include <sys/dsl_dataset.h>
61 #include <sys/dsl_dir.h>
62 #include <sys/dsl_prop.h>
63 #include <sys/dsl_synctask.h>
64 #include <sys/fs/zfs.h>
66 #include <sys/callb.h>
67 #include <sys/spa_boot.h>
68 #include <sys/zfs_ioctl.h>
69 #include <sys/dsl_scan.h>
70 #include <sys/dmu_send.h>
71 #include <sys/dsl_destroy.h>
72 #include <sys/dsl_userhold.h>
73 #include <sys/zfeature.h>
75 #include <sys/trim_map.h>
78 #include <sys/callb.h>
79 #include <sys/cpupart.h>
84 #include "zfs_comutil.h"
86 /* Check hostid on import? */
87 static int check_hostid = 1;
90 * The interval, in seconds, at which failed configuration cache file writes
93 static int zfs_ccw_retry_interval = 300;
95 SYSCTL_DECL(_vfs_zfs);
96 TUNABLE_INT("vfs.zfs.check_hostid", &check_hostid);
97 SYSCTL_INT(_vfs_zfs, OID_AUTO, check_hostid, CTLFLAG_RWTUN, &check_hostid, 0,
98 "Check hostid on import?");
99 TUNABLE_INT("vfs.zfs.ccw_retry_interval", &zfs_ccw_retry_interval);
100 SYSCTL_INT(_vfs_zfs, OID_AUTO, ccw_retry_interval, CTLFLAG_RW,
101 &zfs_ccw_retry_interval, 0,
102 "Configuration cache file write, retry after failure, interval (seconds)");
104 typedef enum zti_modes {
105 ZTI_MODE_FIXED, /* value is # of threads (min 1) */
106 ZTI_MODE_BATCH, /* cpu-intensive; value is ignored */
107 ZTI_MODE_NULL, /* don't create a taskq */
111 #define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) }
112 #define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 }
113 #define ZTI_NULL { ZTI_MODE_NULL, 0, 0 }
115 #define ZTI_N(n) ZTI_P(n, 1)
116 #define ZTI_ONE ZTI_N(1)
118 typedef struct zio_taskq_info {
119 zti_modes_t zti_mode;
124 static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
125 "issue", "issue_high", "intr", "intr_high"
129 * This table defines the taskq settings for each ZFS I/O type. When
130 * initializing a pool, we use this table to create an appropriately sized
131 * taskq. Some operations are low volume and therefore have a small, static
132 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE
133 * macros. Other operations process a large amount of data; the ZTI_BATCH
134 * macro causes us to create a taskq oriented for throughput. Some operations
135 * are so high frequency and short-lived that the taskq itself can become a a
136 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an
137 * additional degree of parallelism specified by the number of threads per-
138 * taskq and the number of taskqs; when dispatching an event in this case, the
139 * particular taskq is chosen at random.
141 * The different taskq priorities are to handle the different contexts (issue
142 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that
143 * need to be handled with minimum delay.
145 const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
146 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
147 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* NULL */
148 { ZTI_N(8), ZTI_NULL, ZTI_P(12, 8), ZTI_NULL }, /* READ */
149 { ZTI_BATCH, ZTI_N(5), ZTI_N(8), ZTI_N(5) }, /* WRITE */
150 { ZTI_P(12, 8), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FREE */
151 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* CLAIM */
152 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* IOCTL */
155 static void spa_sync_version(void *arg, dmu_tx_t *tx);
156 static void spa_sync_props(void *arg, dmu_tx_t *tx);
157 static boolean_t spa_has_active_shared_spare(spa_t *spa);
158 static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
159 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
161 static void spa_vdev_resilver_done(spa_t *spa);
163 uint_t zio_taskq_batch_pct = 75; /* 1 thread per cpu in pset */
165 id_t zio_taskq_psrset_bind = PS_NONE;
168 boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */
170 uint_t zio_taskq_basedc = 80; /* base duty cycle */
172 boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */
173 extern int zfs_sync_pass_deferred_free;
176 extern void spa_deadman(void *arg);
180 * This (illegal) pool name is used when temporarily importing a spa_t in order
181 * to get the vdev stats associated with the imported devices.
183 #define TRYIMPORT_NAME "$import"
186 * ==========================================================================
187 * SPA properties routines
188 * ==========================================================================
192 * Add a (source=src, propname=propval) list to an nvlist.
195 spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
196 uint64_t intval, zprop_source_t src)
198 const char *propname = zpool_prop_to_name(prop);
201 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
202 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
205 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
207 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
209 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
210 nvlist_free(propval);
214 * Get property values from the spa configuration.
217 spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
219 vdev_t *rvd = spa->spa_root_vdev;
220 dsl_pool_t *pool = spa->spa_dsl_pool;
221 uint64_t size, alloc, cap, version;
222 zprop_source_t src = ZPROP_SRC_NONE;
223 spa_config_dirent_t *dp;
224 metaslab_class_t *mc = spa_normal_class(spa);
226 ASSERT(MUTEX_HELD(&spa->spa_props_lock));
229 alloc = metaslab_class_get_alloc(spa_normal_class(spa));
230 size = metaslab_class_get_space(spa_normal_class(spa));
231 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
232 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
233 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
234 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
237 spa_prop_add_list(*nvp, ZPOOL_PROP_FRAGMENTATION, NULL,
238 metaslab_class_fragmentation(mc), src);
239 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL,
240 metaslab_class_expandable_space(mc), src);
241 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
242 (spa_mode(spa) == FREAD), src);
244 cap = (size == 0) ? 0 : (alloc * 100 / size);
245 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
247 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
248 ddt_get_pool_dedup_ratio(spa), src);
250 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
251 rvd->vdev_state, src);
253 version = spa_version(spa);
254 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
255 src = ZPROP_SRC_DEFAULT;
257 src = ZPROP_SRC_LOCAL;
258 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
263 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
264 * when opening pools before this version freedir will be NULL.
266 if (pool->dp_free_dir != NULL) {
267 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
268 dsl_dir_phys(pool->dp_free_dir)->dd_used_bytes,
271 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
275 if (pool->dp_leak_dir != NULL) {
276 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, NULL,
277 dsl_dir_phys(pool->dp_leak_dir)->dd_used_bytes,
280 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED,
285 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
287 if (spa->spa_comment != NULL) {
288 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
292 if (spa->spa_root != NULL)
293 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
296 if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) {
297 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL,
298 MIN(zfs_max_recordsize, SPA_MAXBLOCKSIZE), ZPROP_SRC_NONE);
300 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL,
301 SPA_OLD_MAXBLOCKSIZE, ZPROP_SRC_NONE);
304 if ((dp = list_head(&spa->spa_config_list)) != NULL) {
305 if (dp->scd_path == NULL) {
306 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
307 "none", 0, ZPROP_SRC_LOCAL);
308 } else if (strcmp(dp->scd_path, spa_config_path) != 0) {
309 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
310 dp->scd_path, 0, ZPROP_SRC_LOCAL);
316 * Get zpool property values.
319 spa_prop_get(spa_t *spa, nvlist_t **nvp)
321 objset_t *mos = spa->spa_meta_objset;
326 VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
328 mutex_enter(&spa->spa_props_lock);
331 * Get properties from the spa config.
333 spa_prop_get_config(spa, nvp);
335 /* If no pool property object, no more prop to get. */
336 if (mos == NULL || spa->spa_pool_props_object == 0) {
337 mutex_exit(&spa->spa_props_lock);
342 * Get properties from the MOS pool property object.
344 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
345 (err = zap_cursor_retrieve(&zc, &za)) == 0;
346 zap_cursor_advance(&zc)) {
349 zprop_source_t src = ZPROP_SRC_DEFAULT;
352 if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
355 switch (za.za_integer_length) {
357 /* integer property */
358 if (za.za_first_integer !=
359 zpool_prop_default_numeric(prop))
360 src = ZPROP_SRC_LOCAL;
362 if (prop == ZPOOL_PROP_BOOTFS) {
364 dsl_dataset_t *ds = NULL;
366 dp = spa_get_dsl(spa);
367 dsl_pool_config_enter(dp, FTAG);
368 if (err = dsl_dataset_hold_obj(dp,
369 za.za_first_integer, FTAG, &ds)) {
370 dsl_pool_config_exit(dp, FTAG);
375 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
377 dsl_dataset_name(ds, strval);
378 dsl_dataset_rele(ds, FTAG);
379 dsl_pool_config_exit(dp, FTAG);
382 intval = za.za_first_integer;
385 spa_prop_add_list(*nvp, prop, strval, intval, src);
389 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
394 /* string property */
395 strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
396 err = zap_lookup(mos, spa->spa_pool_props_object,
397 za.za_name, 1, za.za_num_integers, strval);
399 kmem_free(strval, za.za_num_integers);
402 spa_prop_add_list(*nvp, prop, strval, 0, src);
403 kmem_free(strval, za.za_num_integers);
410 zap_cursor_fini(&zc);
411 mutex_exit(&spa->spa_props_lock);
413 if (err && err != ENOENT) {
423 * Validate the given pool properties nvlist and modify the list
424 * for the property values to be set.
427 spa_prop_validate(spa_t *spa, nvlist_t *props)
430 int error = 0, reset_bootfs = 0;
432 boolean_t has_feature = B_FALSE;
435 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
437 char *strval, *slash, *check, *fname;
438 const char *propname = nvpair_name(elem);
439 zpool_prop_t prop = zpool_name_to_prop(propname);
443 if (!zpool_prop_feature(propname)) {
444 error = SET_ERROR(EINVAL);
449 * Sanitize the input.
451 if (nvpair_type(elem) != DATA_TYPE_UINT64) {
452 error = SET_ERROR(EINVAL);
456 if (nvpair_value_uint64(elem, &intval) != 0) {
457 error = SET_ERROR(EINVAL);
462 error = SET_ERROR(EINVAL);
466 fname = strchr(propname, '@') + 1;
467 if (zfeature_lookup_name(fname, NULL) != 0) {
468 error = SET_ERROR(EINVAL);
472 has_feature = B_TRUE;
475 case ZPOOL_PROP_VERSION:
476 error = nvpair_value_uint64(elem, &intval);
478 (intval < spa_version(spa) ||
479 intval > SPA_VERSION_BEFORE_FEATURES ||
481 error = SET_ERROR(EINVAL);
484 case ZPOOL_PROP_DELEGATION:
485 case ZPOOL_PROP_AUTOREPLACE:
486 case ZPOOL_PROP_LISTSNAPS:
487 case ZPOOL_PROP_AUTOEXPAND:
488 error = nvpair_value_uint64(elem, &intval);
489 if (!error && intval > 1)
490 error = SET_ERROR(EINVAL);
493 case ZPOOL_PROP_BOOTFS:
495 * If the pool version is less than SPA_VERSION_BOOTFS,
496 * or the pool is still being created (version == 0),
497 * the bootfs property cannot be set.
499 if (spa_version(spa) < SPA_VERSION_BOOTFS) {
500 error = SET_ERROR(ENOTSUP);
505 * Make sure the vdev config is bootable
507 if (!vdev_is_bootable(spa->spa_root_vdev)) {
508 error = SET_ERROR(ENOTSUP);
514 error = nvpair_value_string(elem, &strval);
520 if (strval == NULL || strval[0] == '\0') {
521 objnum = zpool_prop_default_numeric(
526 if (error = dmu_objset_hold(strval, FTAG, &os))
530 * Must be ZPL, and its property settings
531 * must be supported by GRUB (compression
532 * is not gzip, and large blocks are not used).
535 if (dmu_objset_type(os) != DMU_OST_ZFS) {
536 error = SET_ERROR(ENOTSUP);
538 dsl_prop_get_int_ds(dmu_objset_ds(os),
539 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
541 !BOOTFS_COMPRESS_VALID(propval)) {
542 error = SET_ERROR(ENOTSUP);
544 dsl_prop_get_int_ds(dmu_objset_ds(os),
545 zfs_prop_to_name(ZFS_PROP_RECORDSIZE),
547 propval > SPA_OLD_MAXBLOCKSIZE) {
548 error = SET_ERROR(ENOTSUP);
550 objnum = dmu_objset_id(os);
552 dmu_objset_rele(os, FTAG);
556 case ZPOOL_PROP_FAILUREMODE:
557 error = nvpair_value_uint64(elem, &intval);
558 if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
559 intval > ZIO_FAILURE_MODE_PANIC))
560 error = SET_ERROR(EINVAL);
563 * This is a special case which only occurs when
564 * the pool has completely failed. This allows
565 * the user to change the in-core failmode property
566 * without syncing it out to disk (I/Os might
567 * currently be blocked). We do this by returning
568 * EIO to the caller (spa_prop_set) to trick it
569 * into thinking we encountered a property validation
572 if (!error && spa_suspended(spa)) {
573 spa->spa_failmode = intval;
574 error = SET_ERROR(EIO);
578 case ZPOOL_PROP_CACHEFILE:
579 if ((error = nvpair_value_string(elem, &strval)) != 0)
582 if (strval[0] == '\0')
585 if (strcmp(strval, "none") == 0)
588 if (strval[0] != '/') {
589 error = SET_ERROR(EINVAL);
593 slash = strrchr(strval, '/');
594 ASSERT(slash != NULL);
596 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
597 strcmp(slash, "/..") == 0)
598 error = SET_ERROR(EINVAL);
601 case ZPOOL_PROP_COMMENT:
602 if ((error = nvpair_value_string(elem, &strval)) != 0)
604 for (check = strval; *check != '\0'; check++) {
606 * The kernel doesn't have an easy isprint()
607 * check. For this kernel check, we merely
608 * check ASCII apart from DEL. Fix this if
609 * there is an easy-to-use kernel isprint().
611 if (*check >= 0x7f) {
612 error = SET_ERROR(EINVAL);
616 if (strlen(strval) > ZPROP_MAX_COMMENT)
620 case ZPOOL_PROP_DEDUPDITTO:
621 if (spa_version(spa) < SPA_VERSION_DEDUP)
622 error = SET_ERROR(ENOTSUP);
624 error = nvpair_value_uint64(elem, &intval);
626 intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
627 error = SET_ERROR(EINVAL);
635 if (!error && reset_bootfs) {
636 error = nvlist_remove(props,
637 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
640 error = nvlist_add_uint64(props,
641 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
649 spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
652 spa_config_dirent_t *dp;
654 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
658 dp = kmem_alloc(sizeof (spa_config_dirent_t),
661 if (cachefile[0] == '\0')
662 dp->scd_path = spa_strdup(spa_config_path);
663 else if (strcmp(cachefile, "none") == 0)
666 dp->scd_path = spa_strdup(cachefile);
668 list_insert_head(&spa->spa_config_list, dp);
670 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
674 spa_prop_set(spa_t *spa, nvlist_t *nvp)
677 nvpair_t *elem = NULL;
678 boolean_t need_sync = B_FALSE;
680 if ((error = spa_prop_validate(spa, nvp)) != 0)
683 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
684 zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem));
686 if (prop == ZPOOL_PROP_CACHEFILE ||
687 prop == ZPOOL_PROP_ALTROOT ||
688 prop == ZPOOL_PROP_READONLY)
691 if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) {
694 if (prop == ZPOOL_PROP_VERSION) {
695 VERIFY(nvpair_value_uint64(elem, &ver) == 0);
697 ASSERT(zpool_prop_feature(nvpair_name(elem)));
698 ver = SPA_VERSION_FEATURES;
702 /* Save time if the version is already set. */
703 if (ver == spa_version(spa))
707 * In addition to the pool directory object, we might
708 * create the pool properties object, the features for
709 * read object, the features for write object, or the
710 * feature descriptions object.
712 error = dsl_sync_task(spa->spa_name, NULL,
713 spa_sync_version, &ver,
714 6, ZFS_SPACE_CHECK_RESERVED);
725 return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props,
726 nvp, 6, ZFS_SPACE_CHECK_RESERVED));
733 * If the bootfs property value is dsobj, clear it.
736 spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
738 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
739 VERIFY(zap_remove(spa->spa_meta_objset,
740 spa->spa_pool_props_object,
741 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
748 spa_change_guid_check(void *arg, dmu_tx_t *tx)
750 uint64_t *newguid = arg;
751 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
752 vdev_t *rvd = spa->spa_root_vdev;
755 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
756 vdev_state = rvd->vdev_state;
757 spa_config_exit(spa, SCL_STATE, FTAG);
759 if (vdev_state != VDEV_STATE_HEALTHY)
760 return (SET_ERROR(ENXIO));
762 ASSERT3U(spa_guid(spa), !=, *newguid);
768 spa_change_guid_sync(void *arg, dmu_tx_t *tx)
770 uint64_t *newguid = arg;
771 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
773 vdev_t *rvd = spa->spa_root_vdev;
775 oldguid = spa_guid(spa);
777 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
778 rvd->vdev_guid = *newguid;
779 rvd->vdev_guid_sum += (*newguid - oldguid);
780 vdev_config_dirty(rvd);
781 spa_config_exit(spa, SCL_STATE, FTAG);
783 spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu",
788 * Change the GUID for the pool. This is done so that we can later
789 * re-import a pool built from a clone of our own vdevs. We will modify
790 * the root vdev's guid, our own pool guid, and then mark all of our
791 * vdevs dirty. Note that we must make sure that all our vdevs are
792 * online when we do this, or else any vdevs that weren't present
793 * would be orphaned from our pool. We are also going to issue a
794 * sysevent to update any watchers.
797 spa_change_guid(spa_t *spa)
802 mutex_enter(&spa->spa_vdev_top_lock);
803 mutex_enter(&spa_namespace_lock);
804 guid = spa_generate_guid(NULL);
806 error = dsl_sync_task(spa->spa_name, spa_change_guid_check,
807 spa_change_guid_sync, &guid, 5, ZFS_SPACE_CHECK_RESERVED);
810 spa_config_sync(spa, B_FALSE, B_TRUE);
811 spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID);
814 mutex_exit(&spa_namespace_lock);
815 mutex_exit(&spa->spa_vdev_top_lock);
821 * ==========================================================================
822 * SPA state manipulation (open/create/destroy/import/export)
823 * ==========================================================================
827 spa_error_entry_compare(const void *a, const void *b)
829 spa_error_entry_t *sa = (spa_error_entry_t *)a;
830 spa_error_entry_t *sb = (spa_error_entry_t *)b;
833 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
834 sizeof (zbookmark_phys_t));
845 * Utility function which retrieves copies of the current logs and
846 * re-initializes them in the process.
849 spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
851 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
853 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
854 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
856 avl_create(&spa->spa_errlist_scrub,
857 spa_error_entry_compare, sizeof (spa_error_entry_t),
858 offsetof(spa_error_entry_t, se_avl));
859 avl_create(&spa->spa_errlist_last,
860 spa_error_entry_compare, sizeof (spa_error_entry_t),
861 offsetof(spa_error_entry_t, se_avl));
865 spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
867 const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
868 enum zti_modes mode = ztip->zti_mode;
869 uint_t value = ztip->zti_value;
870 uint_t count = ztip->zti_count;
871 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
874 boolean_t batch = B_FALSE;
876 if (mode == ZTI_MODE_NULL) {
878 tqs->stqs_taskq = NULL;
882 ASSERT3U(count, >, 0);
884 tqs->stqs_count = count;
885 tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP);
889 ASSERT3U(value, >=, 1);
890 value = MAX(value, 1);
895 flags |= TASKQ_THREADS_CPU_PCT;
896 value = zio_taskq_batch_pct;
900 panic("unrecognized mode for %s_%s taskq (%u:%u) in "
902 zio_type_name[t], zio_taskq_types[q], mode, value);
906 for (uint_t i = 0; i < count; i++) {
910 (void) snprintf(name, sizeof (name), "%s_%s_%u",
911 zio_type_name[t], zio_taskq_types[q], i);
913 (void) snprintf(name, sizeof (name), "%s_%s",
914 zio_type_name[t], zio_taskq_types[q]);
918 if (zio_taskq_sysdc && spa->spa_proc != &p0) {
920 flags |= TASKQ_DC_BATCH;
922 tq = taskq_create_sysdc(name, value, 50, INT_MAX,
923 spa->spa_proc, zio_taskq_basedc, flags);
926 pri_t pri = maxclsyspri;
928 * The write issue taskq can be extremely CPU
929 * intensive. Run it at slightly lower priority
930 * than the other taskqs.
932 if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE)
935 tq = taskq_create_proc(name, value, pri, 50,
936 INT_MAX, spa->spa_proc, flags);
941 tqs->stqs_taskq[i] = tq;
946 spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
948 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
950 if (tqs->stqs_taskq == NULL) {
951 ASSERT0(tqs->stqs_count);
955 for (uint_t i = 0; i < tqs->stqs_count; i++) {
956 ASSERT3P(tqs->stqs_taskq[i], !=, NULL);
957 taskq_destroy(tqs->stqs_taskq[i]);
960 kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *));
961 tqs->stqs_taskq = NULL;
965 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
966 * Note that a type may have multiple discrete taskqs to avoid lock contention
967 * on the taskq itself. In that case we choose which taskq at random by using
968 * the low bits of gethrtime().
971 spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q,
972 task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent)
974 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
977 ASSERT3P(tqs->stqs_taskq, !=, NULL);
978 ASSERT3U(tqs->stqs_count, !=, 0);
980 if (tqs->stqs_count == 1) {
981 tq = tqs->stqs_taskq[0];
984 tq = tqs->stqs_taskq[cpu_ticks() % tqs->stqs_count];
986 tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count];
990 taskq_dispatch_ent(tq, func, arg, flags, ent);
994 spa_create_zio_taskqs(spa_t *spa)
996 for (int t = 0; t < ZIO_TYPES; t++) {
997 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
998 spa_taskqs_init(spa, t, q);
1006 spa_thread(void *arg)
1008 callb_cpr_t cprinfo;
1011 user_t *pu = PTOU(curproc);
1013 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
1016 ASSERT(curproc != &p0);
1017 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
1018 "zpool-%s", spa->spa_name);
1019 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
1022 /* bind this thread to the requested psrset */
1023 if (zio_taskq_psrset_bind != PS_NONE) {
1025 mutex_enter(&cpu_lock);
1026 mutex_enter(&pidlock);
1027 mutex_enter(&curproc->p_lock);
1029 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
1030 0, NULL, NULL) == 0) {
1031 curthread->t_bind_pset = zio_taskq_psrset_bind;
1034 "Couldn't bind process for zfs pool \"%s\" to "
1035 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
1038 mutex_exit(&curproc->p_lock);
1039 mutex_exit(&pidlock);
1040 mutex_exit(&cpu_lock);
1046 if (zio_taskq_sysdc) {
1047 sysdc_thread_enter(curthread, 100, 0);
1051 spa->spa_proc = curproc;
1052 spa->spa_did = curthread->t_did;
1054 spa_create_zio_taskqs(spa);
1056 mutex_enter(&spa->spa_proc_lock);
1057 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
1059 spa->spa_proc_state = SPA_PROC_ACTIVE;
1060 cv_broadcast(&spa->spa_proc_cv);
1062 CALLB_CPR_SAFE_BEGIN(&cprinfo);
1063 while (spa->spa_proc_state == SPA_PROC_ACTIVE)
1064 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1065 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
1067 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
1068 spa->spa_proc_state = SPA_PROC_GONE;
1069 spa->spa_proc = &p0;
1070 cv_broadcast(&spa->spa_proc_cv);
1071 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */
1073 mutex_enter(&curproc->p_lock);
1076 #endif /* SPA_PROCESS */
1080 * Activate an uninitialized pool.
1083 spa_activate(spa_t *spa, int mode)
1085 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
1087 spa->spa_state = POOL_STATE_ACTIVE;
1088 spa->spa_mode = mode;
1090 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
1091 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
1093 /* Try to create a covering process */
1094 mutex_enter(&spa->spa_proc_lock);
1095 ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
1096 ASSERT(spa->spa_proc == &p0);
1100 /* Only create a process if we're going to be around a while. */
1101 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
1102 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
1104 spa->spa_proc_state = SPA_PROC_CREATED;
1105 while (spa->spa_proc_state == SPA_PROC_CREATED) {
1106 cv_wait(&spa->spa_proc_cv,
1107 &spa->spa_proc_lock);
1109 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1110 ASSERT(spa->spa_proc != &p0);
1111 ASSERT(spa->spa_did != 0);
1115 "Couldn't create process for zfs pool \"%s\"\n",
1120 #endif /* SPA_PROCESS */
1121 mutex_exit(&spa->spa_proc_lock);
1123 /* If we didn't create a process, we need to create our taskqs. */
1124 ASSERT(spa->spa_proc == &p0);
1125 if (spa->spa_proc == &p0) {
1126 spa_create_zio_taskqs(spa);
1130 * Start TRIM thread.
1132 trim_thread_create(spa);
1134 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
1135 offsetof(vdev_t, vdev_config_dirty_node));
1136 list_create(&spa->spa_evicting_os_list, sizeof (objset_t),
1137 offsetof(objset_t, os_evicting_node));
1138 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
1139 offsetof(vdev_t, vdev_state_dirty_node));
1141 txg_list_create(&spa->spa_vdev_txg_list,
1142 offsetof(struct vdev, vdev_txg_node));
1144 avl_create(&spa->spa_errlist_scrub,
1145 spa_error_entry_compare, sizeof (spa_error_entry_t),
1146 offsetof(spa_error_entry_t, se_avl));
1147 avl_create(&spa->spa_errlist_last,
1148 spa_error_entry_compare, sizeof (spa_error_entry_t),
1149 offsetof(spa_error_entry_t, se_avl));
1153 * Opposite of spa_activate().
1156 spa_deactivate(spa_t *spa)
1158 ASSERT(spa->spa_sync_on == B_FALSE);
1159 ASSERT(spa->spa_dsl_pool == NULL);
1160 ASSERT(spa->spa_root_vdev == NULL);
1161 ASSERT(spa->spa_async_zio_root == NULL);
1162 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
1165 * Stop TRIM thread in case spa_unload() wasn't called directly
1166 * before spa_deactivate().
1168 trim_thread_destroy(spa);
1170 spa_evicting_os_wait(spa);
1172 txg_list_destroy(&spa->spa_vdev_txg_list);
1174 list_destroy(&spa->spa_config_dirty_list);
1175 list_destroy(&spa->spa_evicting_os_list);
1176 list_destroy(&spa->spa_state_dirty_list);
1178 for (int t = 0; t < ZIO_TYPES; t++) {
1179 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1180 spa_taskqs_fini(spa, t, q);
1184 metaslab_class_destroy(spa->spa_normal_class);
1185 spa->spa_normal_class = NULL;
1187 metaslab_class_destroy(spa->spa_log_class);
1188 spa->spa_log_class = NULL;
1191 * If this was part of an import or the open otherwise failed, we may
1192 * still have errors left in the queues. Empty them just in case.
1194 spa_errlog_drain(spa);
1196 avl_destroy(&spa->spa_errlist_scrub);
1197 avl_destroy(&spa->spa_errlist_last);
1199 spa->spa_state = POOL_STATE_UNINITIALIZED;
1201 mutex_enter(&spa->spa_proc_lock);
1202 if (spa->spa_proc_state != SPA_PROC_NONE) {
1203 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1204 spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1205 cv_broadcast(&spa->spa_proc_cv);
1206 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1207 ASSERT(spa->spa_proc != &p0);
1208 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1210 ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1211 spa->spa_proc_state = SPA_PROC_NONE;
1213 ASSERT(spa->spa_proc == &p0);
1214 mutex_exit(&spa->spa_proc_lock);
1218 * We want to make sure spa_thread() has actually exited the ZFS
1219 * module, so that the module can't be unloaded out from underneath
1222 if (spa->spa_did != 0) {
1223 thread_join(spa->spa_did);
1226 #endif /* SPA_PROCESS */
1230 * Verify a pool configuration, and construct the vdev tree appropriately. This
1231 * will create all the necessary vdevs in the appropriate layout, with each vdev
1232 * in the CLOSED state. This will prep the pool before open/creation/import.
1233 * All vdev validation is done by the vdev_alloc() routine.
1236 spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1237 uint_t id, int atype)
1243 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1246 if ((*vdp)->vdev_ops->vdev_op_leaf)
1249 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1252 if (error == ENOENT)
1258 return (SET_ERROR(EINVAL));
1261 for (int c = 0; c < children; c++) {
1263 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1271 ASSERT(*vdp != NULL);
1277 * Opposite of spa_load().
1280 spa_unload(spa_t *spa)
1284 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1289 trim_thread_destroy(spa);
1294 spa_async_suspend(spa);
1299 if (spa->spa_sync_on) {
1300 txg_sync_stop(spa->spa_dsl_pool);
1301 spa->spa_sync_on = B_FALSE;
1305 * Wait for any outstanding async I/O to complete.
1307 if (spa->spa_async_zio_root != NULL) {
1308 for (int i = 0; i < max_ncpus; i++)
1309 (void) zio_wait(spa->spa_async_zio_root[i]);
1310 kmem_free(spa->spa_async_zio_root, max_ncpus * sizeof (void *));
1311 spa->spa_async_zio_root = NULL;
1314 bpobj_close(&spa->spa_deferred_bpobj);
1316 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1321 if (spa->spa_root_vdev)
1322 vdev_free(spa->spa_root_vdev);
1323 ASSERT(spa->spa_root_vdev == NULL);
1326 * Close the dsl pool.
1328 if (spa->spa_dsl_pool) {
1329 dsl_pool_close(spa->spa_dsl_pool);
1330 spa->spa_dsl_pool = NULL;
1331 spa->spa_meta_objset = NULL;
1338 * Drop and purge level 2 cache
1340 spa_l2cache_drop(spa);
1342 for (i = 0; i < spa->spa_spares.sav_count; i++)
1343 vdev_free(spa->spa_spares.sav_vdevs[i]);
1344 if (spa->spa_spares.sav_vdevs) {
1345 kmem_free(spa->spa_spares.sav_vdevs,
1346 spa->spa_spares.sav_count * sizeof (void *));
1347 spa->spa_spares.sav_vdevs = NULL;
1349 if (spa->spa_spares.sav_config) {
1350 nvlist_free(spa->spa_spares.sav_config);
1351 spa->spa_spares.sav_config = NULL;
1353 spa->spa_spares.sav_count = 0;
1355 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1356 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1357 vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1359 if (spa->spa_l2cache.sav_vdevs) {
1360 kmem_free(spa->spa_l2cache.sav_vdevs,
1361 spa->spa_l2cache.sav_count * sizeof (void *));
1362 spa->spa_l2cache.sav_vdevs = NULL;
1364 if (spa->spa_l2cache.sav_config) {
1365 nvlist_free(spa->spa_l2cache.sav_config);
1366 spa->spa_l2cache.sav_config = NULL;
1368 spa->spa_l2cache.sav_count = 0;
1370 spa->spa_async_suspended = 0;
1372 if (spa->spa_comment != NULL) {
1373 spa_strfree(spa->spa_comment);
1374 spa->spa_comment = NULL;
1377 spa_config_exit(spa, SCL_ALL, FTAG);
1381 * Load (or re-load) the current list of vdevs describing the active spares for
1382 * this pool. When this is called, we have some form of basic information in
1383 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1384 * then re-generate a more complete list including status information.
1387 spa_load_spares(spa_t *spa)
1394 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1397 * First, close and free any existing spare vdevs.
1399 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1400 vd = spa->spa_spares.sav_vdevs[i];
1402 /* Undo the call to spa_activate() below */
1403 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1404 B_FALSE)) != NULL && tvd->vdev_isspare)
1405 spa_spare_remove(tvd);
1410 if (spa->spa_spares.sav_vdevs)
1411 kmem_free(spa->spa_spares.sav_vdevs,
1412 spa->spa_spares.sav_count * sizeof (void *));
1414 if (spa->spa_spares.sav_config == NULL)
1417 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1418 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1420 spa->spa_spares.sav_count = (int)nspares;
1421 spa->spa_spares.sav_vdevs = NULL;
1427 * Construct the array of vdevs, opening them to get status in the
1428 * process. For each spare, there is potentially two different vdev_t
1429 * structures associated with it: one in the list of spares (used only
1430 * for basic validation purposes) and one in the active vdev
1431 * configuration (if it's spared in). During this phase we open and
1432 * validate each vdev on the spare list. If the vdev also exists in the
1433 * active configuration, then we also mark this vdev as an active spare.
1435 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1437 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1438 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1439 VDEV_ALLOC_SPARE) == 0);
1442 spa->spa_spares.sav_vdevs[i] = vd;
1444 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1445 B_FALSE)) != NULL) {
1446 if (!tvd->vdev_isspare)
1450 * We only mark the spare active if we were successfully
1451 * able to load the vdev. Otherwise, importing a pool
1452 * with a bad active spare would result in strange
1453 * behavior, because multiple pool would think the spare
1454 * is actively in use.
1456 * There is a vulnerability here to an equally bizarre
1457 * circumstance, where a dead active spare is later
1458 * brought back to life (onlined or otherwise). Given
1459 * the rarity of this scenario, and the extra complexity
1460 * it adds, we ignore the possibility.
1462 if (!vdev_is_dead(tvd))
1463 spa_spare_activate(tvd);
1467 vd->vdev_aux = &spa->spa_spares;
1469 if (vdev_open(vd) != 0)
1472 if (vdev_validate_aux(vd) == 0)
1477 * Recompute the stashed list of spares, with status information
1480 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1481 DATA_TYPE_NVLIST_ARRAY) == 0);
1483 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1485 for (i = 0; i < spa->spa_spares.sav_count; i++)
1486 spares[i] = vdev_config_generate(spa,
1487 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1488 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1489 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1490 for (i = 0; i < spa->spa_spares.sav_count; i++)
1491 nvlist_free(spares[i]);
1492 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1496 * Load (or re-load) the current list of vdevs describing the active l2cache for
1497 * this pool. When this is called, we have some form of basic information in
1498 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1499 * then re-generate a more complete list including status information.
1500 * Devices which are already active have their details maintained, and are
1504 spa_load_l2cache(spa_t *spa)
1508 int i, j, oldnvdevs;
1510 vdev_t *vd, **oldvdevs, **newvdevs;
1511 spa_aux_vdev_t *sav = &spa->spa_l2cache;
1513 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1515 if (sav->sav_config != NULL) {
1516 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1517 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1518 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1524 oldvdevs = sav->sav_vdevs;
1525 oldnvdevs = sav->sav_count;
1526 sav->sav_vdevs = NULL;
1530 * Process new nvlist of vdevs.
1532 for (i = 0; i < nl2cache; i++) {
1533 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1537 for (j = 0; j < oldnvdevs; j++) {
1539 if (vd != NULL && guid == vd->vdev_guid) {
1541 * Retain previous vdev for add/remove ops.
1549 if (newvdevs[i] == NULL) {
1553 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1554 VDEV_ALLOC_L2CACHE) == 0);
1559 * Commit this vdev as an l2cache device,
1560 * even if it fails to open.
1562 spa_l2cache_add(vd);
1567 spa_l2cache_activate(vd);
1569 if (vdev_open(vd) != 0)
1572 (void) vdev_validate_aux(vd);
1574 if (!vdev_is_dead(vd))
1575 l2arc_add_vdev(spa, vd);
1580 * Purge vdevs that were dropped
1582 for (i = 0; i < oldnvdevs; i++) {
1587 ASSERT(vd->vdev_isl2cache);
1589 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1590 pool != 0ULL && l2arc_vdev_present(vd))
1591 l2arc_remove_vdev(vd);
1592 vdev_clear_stats(vd);
1598 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1600 if (sav->sav_config == NULL)
1603 sav->sav_vdevs = newvdevs;
1604 sav->sav_count = (int)nl2cache;
1607 * Recompute the stashed list of l2cache devices, with status
1608 * information this time.
1610 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1611 DATA_TYPE_NVLIST_ARRAY) == 0);
1613 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1614 for (i = 0; i < sav->sav_count; i++)
1615 l2cache[i] = vdev_config_generate(spa,
1616 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1617 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1618 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1620 for (i = 0; i < sav->sav_count; i++)
1621 nvlist_free(l2cache[i]);
1623 kmem_free(l2cache, sav->sav_count * sizeof (void *));
1627 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1630 char *packed = NULL;
1635 error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db);
1639 nvsize = *(uint64_t *)db->db_data;
1640 dmu_buf_rele(db, FTAG);
1642 packed = kmem_alloc(nvsize, KM_SLEEP);
1643 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1646 error = nvlist_unpack(packed, nvsize, value, 0);
1647 kmem_free(packed, nvsize);
1653 * Checks to see if the given vdev could not be opened, in which case we post a
1654 * sysevent to notify the autoreplace code that the device has been removed.
1657 spa_check_removed(vdev_t *vd)
1659 for (int c = 0; c < vd->vdev_children; c++)
1660 spa_check_removed(vd->vdev_child[c]);
1662 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) &&
1664 zfs_post_autoreplace(vd->vdev_spa, vd);
1665 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1670 * Validate the current config against the MOS config
1673 spa_config_valid(spa_t *spa, nvlist_t *config)
1675 vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1678 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1680 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1681 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1683 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1686 * If we're doing a normal import, then build up any additional
1687 * diagnostic information about missing devices in this config.
1688 * We'll pass this up to the user for further processing.
1690 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1691 nvlist_t **child, *nv;
1694 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1696 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1698 for (int c = 0; c < rvd->vdev_children; c++) {
1699 vdev_t *tvd = rvd->vdev_child[c];
1700 vdev_t *mtvd = mrvd->vdev_child[c];
1702 if (tvd->vdev_ops == &vdev_missing_ops &&
1703 mtvd->vdev_ops != &vdev_missing_ops &&
1705 child[idx++] = vdev_config_generate(spa, mtvd,
1710 VERIFY(nvlist_add_nvlist_array(nv,
1711 ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1712 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1713 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1715 for (int i = 0; i < idx; i++)
1716 nvlist_free(child[i]);
1719 kmem_free(child, rvd->vdev_children * sizeof (char **));
1723 * Compare the root vdev tree with the information we have
1724 * from the MOS config (mrvd). Check each top-level vdev
1725 * with the corresponding MOS config top-level (mtvd).
1727 for (int c = 0; c < rvd->vdev_children; c++) {
1728 vdev_t *tvd = rvd->vdev_child[c];
1729 vdev_t *mtvd = mrvd->vdev_child[c];
1732 * Resolve any "missing" vdevs in the current configuration.
1733 * If we find that the MOS config has more accurate information
1734 * about the top-level vdev then use that vdev instead.
1736 if (tvd->vdev_ops == &vdev_missing_ops &&
1737 mtvd->vdev_ops != &vdev_missing_ops) {
1739 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1743 * Device specific actions.
1745 if (mtvd->vdev_islog) {
1746 spa_set_log_state(spa, SPA_LOG_CLEAR);
1749 * XXX - once we have 'readonly' pool
1750 * support we should be able to handle
1751 * missing data devices by transitioning
1752 * the pool to readonly.
1758 * Swap the missing vdev with the data we were
1759 * able to obtain from the MOS config.
1761 vdev_remove_child(rvd, tvd);
1762 vdev_remove_child(mrvd, mtvd);
1764 vdev_add_child(rvd, mtvd);
1765 vdev_add_child(mrvd, tvd);
1767 spa_config_exit(spa, SCL_ALL, FTAG);
1769 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1772 } else if (mtvd->vdev_islog) {
1774 * Load the slog device's state from the MOS config
1775 * since it's possible that the label does not
1776 * contain the most up-to-date information.
1778 vdev_load_log_state(tvd, mtvd);
1783 spa_config_exit(spa, SCL_ALL, FTAG);
1786 * Ensure we were able to validate the config.
1788 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1792 * Check for missing log devices
1795 spa_check_logs(spa_t *spa)
1797 boolean_t rv = B_FALSE;
1798 dsl_pool_t *dp = spa_get_dsl(spa);
1800 switch (spa->spa_log_state) {
1801 case SPA_LOG_MISSING:
1802 /* need to recheck in case slog has been restored */
1803 case SPA_LOG_UNKNOWN:
1804 rv = (dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
1805 zil_check_log_chain, NULL, DS_FIND_CHILDREN) != 0);
1807 spa_set_log_state(spa, SPA_LOG_MISSING);
1814 spa_passivate_log(spa_t *spa)
1816 vdev_t *rvd = spa->spa_root_vdev;
1817 boolean_t slog_found = B_FALSE;
1819 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1821 if (!spa_has_slogs(spa))
1824 for (int c = 0; c < rvd->vdev_children; c++) {
1825 vdev_t *tvd = rvd->vdev_child[c];
1826 metaslab_group_t *mg = tvd->vdev_mg;
1828 if (tvd->vdev_islog) {
1829 metaslab_group_passivate(mg);
1830 slog_found = B_TRUE;
1834 return (slog_found);
1838 spa_activate_log(spa_t *spa)
1840 vdev_t *rvd = spa->spa_root_vdev;
1842 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1844 for (int c = 0; c < rvd->vdev_children; c++) {
1845 vdev_t *tvd = rvd->vdev_child[c];
1846 metaslab_group_t *mg = tvd->vdev_mg;
1848 if (tvd->vdev_islog)
1849 metaslab_group_activate(mg);
1854 spa_offline_log(spa_t *spa)
1858 error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1859 NULL, DS_FIND_CHILDREN);
1862 * We successfully offlined the log device, sync out the
1863 * current txg so that the "stubby" block can be removed
1866 txg_wait_synced(spa->spa_dsl_pool, 0);
1872 spa_aux_check_removed(spa_aux_vdev_t *sav)
1876 for (i = 0; i < sav->sav_count; i++)
1877 spa_check_removed(sav->sav_vdevs[i]);
1881 spa_claim_notify(zio_t *zio)
1883 spa_t *spa = zio->io_spa;
1888 mutex_enter(&spa->spa_props_lock); /* any mutex will do */
1889 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1890 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1891 mutex_exit(&spa->spa_props_lock);
1894 typedef struct spa_load_error {
1895 uint64_t sle_meta_count;
1896 uint64_t sle_data_count;
1900 spa_load_verify_done(zio_t *zio)
1902 blkptr_t *bp = zio->io_bp;
1903 spa_load_error_t *sle = zio->io_private;
1904 dmu_object_type_t type = BP_GET_TYPE(bp);
1905 int error = zio->io_error;
1906 spa_t *spa = zio->io_spa;
1909 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1910 type != DMU_OT_INTENT_LOG)
1911 atomic_inc_64(&sle->sle_meta_count);
1913 atomic_inc_64(&sle->sle_data_count);
1915 zio_data_buf_free(zio->io_data, zio->io_size);
1917 mutex_enter(&spa->spa_scrub_lock);
1918 spa->spa_scrub_inflight--;
1919 cv_broadcast(&spa->spa_scrub_io_cv);
1920 mutex_exit(&spa->spa_scrub_lock);
1924 * Maximum number of concurrent scrub i/os to create while verifying
1925 * a pool while importing it.
1927 int spa_load_verify_maxinflight = 10000;
1928 boolean_t spa_load_verify_metadata = B_TRUE;
1929 boolean_t spa_load_verify_data = B_TRUE;
1931 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_maxinflight, CTLFLAG_RWTUN,
1932 &spa_load_verify_maxinflight, 0,
1933 "Maximum number of concurrent scrub I/Os to create while verifying a "
1934 "pool while importing it");
1936 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_metadata, CTLFLAG_RWTUN,
1937 &spa_load_verify_metadata, 0,
1938 "Check metadata on import?");
1940 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_data, CTLFLAG_RWTUN,
1941 &spa_load_verify_data, 0,
1942 "Check user data on import?");
1946 spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1947 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
1949 if (bp == NULL || BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp))
1952 * Note: normally this routine will not be called if
1953 * spa_load_verify_metadata is not set. However, it may be useful
1954 * to manually set the flag after the traversal has begun.
1956 if (!spa_load_verify_metadata)
1958 if (BP_GET_BUFC_TYPE(bp) == ARC_BUFC_DATA && !spa_load_verify_data)
1962 size_t size = BP_GET_PSIZE(bp);
1963 void *data = zio_data_buf_alloc(size);
1965 mutex_enter(&spa->spa_scrub_lock);
1966 while (spa->spa_scrub_inflight >= spa_load_verify_maxinflight)
1967 cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
1968 spa->spa_scrub_inflight++;
1969 mutex_exit(&spa->spa_scrub_lock);
1971 zio_nowait(zio_read(rio, spa, bp, data, size,
1972 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1973 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1974 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1979 spa_load_verify(spa_t *spa)
1982 spa_load_error_t sle = { 0 };
1983 zpool_rewind_policy_t policy;
1984 boolean_t verify_ok = B_FALSE;
1987 zpool_get_rewind_policy(spa->spa_config, &policy);
1989 if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1992 rio = zio_root(spa, NULL, &sle,
1993 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1995 if (spa_load_verify_metadata) {
1996 error = traverse_pool(spa, spa->spa_verify_min_txg,
1997 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA,
1998 spa_load_verify_cb, rio);
2001 (void) zio_wait(rio);
2003 spa->spa_load_meta_errors = sle.sle_meta_count;
2004 spa->spa_load_data_errors = sle.sle_data_count;
2006 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
2007 sle.sle_data_count <= policy.zrp_maxdata) {
2011 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
2012 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
2014 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
2015 VERIFY(nvlist_add_uint64(spa->spa_load_info,
2016 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
2017 VERIFY(nvlist_add_int64(spa->spa_load_info,
2018 ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
2019 VERIFY(nvlist_add_uint64(spa->spa_load_info,
2020 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
2022 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
2026 if (error != ENXIO && error != EIO)
2027 error = SET_ERROR(EIO);
2031 return (verify_ok ? 0 : EIO);
2035 * Find a value in the pool props object.
2038 spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
2040 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
2041 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
2045 * Find a value in the pool directory object.
2048 spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
2050 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
2051 name, sizeof (uint64_t), 1, val));
2055 spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
2057 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
2062 * Fix up config after a partly-completed split. This is done with the
2063 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
2064 * pool have that entry in their config, but only the splitting one contains
2065 * a list of all the guids of the vdevs that are being split off.
2067 * This function determines what to do with that list: either rejoin
2068 * all the disks to the pool, or complete the splitting process. To attempt
2069 * the rejoin, each disk that is offlined is marked online again, and
2070 * we do a reopen() call. If the vdev label for every disk that was
2071 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
2072 * then we call vdev_split() on each disk, and complete the split.
2074 * Otherwise we leave the config alone, with all the vdevs in place in
2075 * the original pool.
2078 spa_try_repair(spa_t *spa, nvlist_t *config)
2085 boolean_t attempt_reopen;
2087 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
2090 /* check that the config is complete */
2091 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
2092 &glist, &gcount) != 0)
2095 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
2097 /* attempt to online all the vdevs & validate */
2098 attempt_reopen = B_TRUE;
2099 for (i = 0; i < gcount; i++) {
2100 if (glist[i] == 0) /* vdev is hole */
2103 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
2104 if (vd[i] == NULL) {
2106 * Don't bother attempting to reopen the disks;
2107 * just do the split.
2109 attempt_reopen = B_FALSE;
2111 /* attempt to re-online it */
2112 vd[i]->vdev_offline = B_FALSE;
2116 if (attempt_reopen) {
2117 vdev_reopen(spa->spa_root_vdev);
2119 /* check each device to see what state it's in */
2120 for (extracted = 0, i = 0; i < gcount; i++) {
2121 if (vd[i] != NULL &&
2122 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
2129 * If every disk has been moved to the new pool, or if we never
2130 * even attempted to look at them, then we split them off for
2133 if (!attempt_reopen || gcount == extracted) {
2134 for (i = 0; i < gcount; i++)
2137 vdev_reopen(spa->spa_root_vdev);
2140 kmem_free(vd, gcount * sizeof (vdev_t *));
2144 spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
2145 boolean_t mosconfig)
2147 nvlist_t *config = spa->spa_config;
2148 char *ereport = FM_EREPORT_ZFS_POOL;
2154 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
2155 return (SET_ERROR(EINVAL));
2157 ASSERT(spa->spa_comment == NULL);
2158 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
2159 spa->spa_comment = spa_strdup(comment);
2162 * Versioning wasn't explicitly added to the label until later, so if
2163 * it's not present treat it as the initial version.
2165 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
2166 &spa->spa_ubsync.ub_version) != 0)
2167 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
2169 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
2170 &spa->spa_config_txg);
2172 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
2173 spa_guid_exists(pool_guid, 0)) {
2174 error = SET_ERROR(EEXIST);
2176 spa->spa_config_guid = pool_guid;
2178 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
2180 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
2184 nvlist_free(spa->spa_load_info);
2185 spa->spa_load_info = fnvlist_alloc();
2187 gethrestime(&spa->spa_loaded_ts);
2188 error = spa_load_impl(spa, pool_guid, config, state, type,
2189 mosconfig, &ereport);
2193 * Don't count references from objsets that are already closed
2194 * and are making their way through the eviction process.
2196 spa_evicting_os_wait(spa);
2197 spa->spa_minref = refcount_count(&spa->spa_refcount);
2199 if (error != EEXIST) {
2200 spa->spa_loaded_ts.tv_sec = 0;
2201 spa->spa_loaded_ts.tv_nsec = 0;
2203 if (error != EBADF) {
2204 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
2207 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2214 * Load an existing storage pool, using the pool's builtin spa_config as a
2215 * source of configuration information.
2218 spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
2219 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
2223 nvlist_t *nvroot = NULL;
2226 uberblock_t *ub = &spa->spa_uberblock;
2227 uint64_t children, config_cache_txg = spa->spa_config_txg;
2228 int orig_mode = spa->spa_mode;
2231 boolean_t missing_feat_write = B_FALSE;
2234 * If this is an untrusted config, access the pool in read-only mode.
2235 * This prevents things like resilvering recently removed devices.
2238 spa->spa_mode = FREAD;
2240 ASSERT(MUTEX_HELD(&spa_namespace_lock));
2242 spa->spa_load_state = state;
2244 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
2245 return (SET_ERROR(EINVAL));
2247 parse = (type == SPA_IMPORT_EXISTING ?
2248 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2251 * Create "The Godfather" zio to hold all async IOs
2253 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
2255 for (int i = 0; i < max_ncpus; i++) {
2256 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
2257 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
2258 ZIO_FLAG_GODFATHER);
2262 * Parse the configuration into a vdev tree. We explicitly set the
2263 * value that will be returned by spa_version() since parsing the
2264 * configuration requires knowing the version number.
2266 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2267 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
2268 spa_config_exit(spa, SCL_ALL, FTAG);
2273 ASSERT(spa->spa_root_vdev == rvd);
2274 ASSERT3U(spa->spa_min_ashift, >=, SPA_MINBLOCKSHIFT);
2275 ASSERT3U(spa->spa_max_ashift, <=, SPA_MAXBLOCKSHIFT);
2277 if (type != SPA_IMPORT_ASSEMBLE) {
2278 ASSERT(spa_guid(spa) == pool_guid);
2282 * Try to open all vdevs, loading each label in the process.
2284 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2285 error = vdev_open(rvd);
2286 spa_config_exit(spa, SCL_ALL, FTAG);
2291 * We need to validate the vdev labels against the configuration that
2292 * we have in hand, which is dependent on the setting of mosconfig. If
2293 * mosconfig is true then we're validating the vdev labels based on
2294 * that config. Otherwise, we're validating against the cached config
2295 * (zpool.cache) that was read when we loaded the zfs module, and then
2296 * later we will recursively call spa_load() and validate against
2299 * If we're assembling a new pool that's been split off from an
2300 * existing pool, the labels haven't yet been updated so we skip
2301 * validation for now.
2303 if (type != SPA_IMPORT_ASSEMBLE) {
2304 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2305 error = vdev_validate(rvd, mosconfig);
2306 spa_config_exit(spa, SCL_ALL, FTAG);
2311 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2312 return (SET_ERROR(ENXIO));
2316 * Find the best uberblock.
2318 vdev_uberblock_load(rvd, ub, &label);
2321 * If we weren't able to find a single valid uberblock, return failure.
2323 if (ub->ub_txg == 0) {
2325 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2329 * If the pool has an unsupported version we can't open it.
2331 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2333 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2336 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2340 * If we weren't able to find what's necessary for reading the
2341 * MOS in the label, return failure.
2343 if (label == NULL || nvlist_lookup_nvlist(label,
2344 ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) {
2346 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2351 * Update our in-core representation with the definitive values
2354 nvlist_free(spa->spa_label_features);
2355 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2361 * Look through entries in the label nvlist's features_for_read. If
2362 * there is a feature listed there which we don't understand then we
2363 * cannot open a pool.
2365 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2366 nvlist_t *unsup_feat;
2368 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2371 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2373 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2374 if (!zfeature_is_supported(nvpair_name(nvp))) {
2375 VERIFY(nvlist_add_string(unsup_feat,
2376 nvpair_name(nvp), "") == 0);
2380 if (!nvlist_empty(unsup_feat)) {
2381 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2382 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2383 nvlist_free(unsup_feat);
2384 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2388 nvlist_free(unsup_feat);
2392 * If the vdev guid sum doesn't match the uberblock, we have an
2393 * incomplete configuration. We first check to see if the pool
2394 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2395 * If it is, defer the vdev_guid_sum check till later so we
2396 * can handle missing vdevs.
2398 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
2399 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
2400 rvd->vdev_guid_sum != ub->ub_guid_sum)
2401 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2403 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2404 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2405 spa_try_repair(spa, config);
2406 spa_config_exit(spa, SCL_ALL, FTAG);
2407 nvlist_free(spa->spa_config_splitting);
2408 spa->spa_config_splitting = NULL;
2412 * Initialize internal SPA structures.
2414 spa->spa_state = POOL_STATE_ACTIVE;
2415 spa->spa_ubsync = spa->spa_uberblock;
2416 spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2417 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2418 spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2419 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2420 spa->spa_claim_max_txg = spa->spa_first_txg;
2421 spa->spa_prev_software_version = ub->ub_software_version;
2423 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2425 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2426 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2428 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2429 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2431 if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2432 boolean_t missing_feat_read = B_FALSE;
2433 nvlist_t *unsup_feat, *enabled_feat;
2435 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2436 &spa->spa_feat_for_read_obj) != 0) {
2437 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2440 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2441 &spa->spa_feat_for_write_obj) != 0) {
2442 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2445 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2446 &spa->spa_feat_desc_obj) != 0) {
2447 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2450 enabled_feat = fnvlist_alloc();
2451 unsup_feat = fnvlist_alloc();
2453 if (!spa_features_check(spa, B_FALSE,
2454 unsup_feat, enabled_feat))
2455 missing_feat_read = B_TRUE;
2457 if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) {
2458 if (!spa_features_check(spa, B_TRUE,
2459 unsup_feat, enabled_feat)) {
2460 missing_feat_write = B_TRUE;
2464 fnvlist_add_nvlist(spa->spa_load_info,
2465 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
2467 if (!nvlist_empty(unsup_feat)) {
2468 fnvlist_add_nvlist(spa->spa_load_info,
2469 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
2472 fnvlist_free(enabled_feat);
2473 fnvlist_free(unsup_feat);
2475 if (!missing_feat_read) {
2476 fnvlist_add_boolean(spa->spa_load_info,
2477 ZPOOL_CONFIG_CAN_RDONLY);
2481 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2482 * twofold: to determine whether the pool is available for
2483 * import in read-write mode and (if it is not) whether the
2484 * pool is available for import in read-only mode. If the pool
2485 * is available for import in read-write mode, it is displayed
2486 * as available in userland; if it is not available for import
2487 * in read-only mode, it is displayed as unavailable in
2488 * userland. If the pool is available for import in read-only
2489 * mode but not read-write mode, it is displayed as unavailable
2490 * in userland with a special note that the pool is actually
2491 * available for open in read-only mode.
2493 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2494 * missing a feature for write, we must first determine whether
2495 * the pool can be opened read-only before returning to
2496 * userland in order to know whether to display the
2497 * abovementioned note.
2499 if (missing_feat_read || (missing_feat_write &&
2500 spa_writeable(spa))) {
2501 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2506 * Load refcounts for ZFS features from disk into an in-memory
2507 * cache during SPA initialization.
2509 for (spa_feature_t i = 0; i < SPA_FEATURES; i++) {
2512 error = feature_get_refcount_from_disk(spa,
2513 &spa_feature_table[i], &refcount);
2515 spa->spa_feat_refcount_cache[i] = refcount;
2516 } else if (error == ENOTSUP) {
2517 spa->spa_feat_refcount_cache[i] =
2518 SPA_FEATURE_DISABLED;
2520 return (spa_vdev_err(rvd,
2521 VDEV_AUX_CORRUPT_DATA, EIO));
2526 if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) {
2527 if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG,
2528 &spa->spa_feat_enabled_txg_obj) != 0)
2529 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2532 spa->spa_is_initializing = B_TRUE;
2533 error = dsl_pool_open(spa->spa_dsl_pool);
2534 spa->spa_is_initializing = B_FALSE;
2536 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2540 nvlist_t *policy = NULL, *nvconfig;
2542 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2543 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2545 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2546 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2548 unsigned long myhostid = 0;
2550 VERIFY(nvlist_lookup_string(nvconfig,
2551 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2554 myhostid = zone_get_hostid(NULL);
2557 * We're emulating the system's hostid in userland, so
2558 * we can't use zone_get_hostid().
2560 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2561 #endif /* _KERNEL */
2562 if (check_hostid && hostid != 0 && myhostid != 0 &&
2563 hostid != myhostid) {
2564 nvlist_free(nvconfig);
2565 cmn_err(CE_WARN, "pool '%s' could not be "
2566 "loaded as it was last accessed by "
2567 "another system (host: %s hostid: 0x%lx). "
2568 "See: http://illumos.org/msg/ZFS-8000-EY",
2569 spa_name(spa), hostname,
2570 (unsigned long)hostid);
2571 return (SET_ERROR(EBADF));
2574 if (nvlist_lookup_nvlist(spa->spa_config,
2575 ZPOOL_REWIND_POLICY, &policy) == 0)
2576 VERIFY(nvlist_add_nvlist(nvconfig,
2577 ZPOOL_REWIND_POLICY, policy) == 0);
2579 spa_config_set(spa, nvconfig);
2581 spa_deactivate(spa);
2582 spa_activate(spa, orig_mode);
2584 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2587 /* Grab the secret checksum salt from the MOS. */
2588 error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
2589 DMU_POOL_CHECKSUM_SALT, 1,
2590 sizeof (spa->spa_cksum_salt.zcs_bytes),
2591 spa->spa_cksum_salt.zcs_bytes);
2592 if (error == ENOENT) {
2593 /* Generate a new salt for subsequent use */
2594 (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes,
2595 sizeof (spa->spa_cksum_salt.zcs_bytes));
2596 } else if (error != 0) {
2597 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2600 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2601 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2602 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2604 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2607 * Load the bit that tells us to use the new accounting function
2608 * (raid-z deflation). If we have an older pool, this will not
2611 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2612 if (error != 0 && error != ENOENT)
2613 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2615 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2616 &spa->spa_creation_version);
2617 if (error != 0 && error != ENOENT)
2618 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2621 * Load the persistent error log. If we have an older pool, this will
2624 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2625 if (error != 0 && error != ENOENT)
2626 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2628 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2629 &spa->spa_errlog_scrub);
2630 if (error != 0 && error != ENOENT)
2631 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2634 * Load the history object. If we have an older pool, this
2635 * will not be present.
2637 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2638 if (error != 0 && error != ENOENT)
2639 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2642 * If we're assembling the pool from the split-off vdevs of
2643 * an existing pool, we don't want to attach the spares & cache
2648 * Load any hot spares for this pool.
2650 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2651 if (error != 0 && error != ENOENT)
2652 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2653 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2654 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2655 if (load_nvlist(spa, spa->spa_spares.sav_object,
2656 &spa->spa_spares.sav_config) != 0)
2657 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2659 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2660 spa_load_spares(spa);
2661 spa_config_exit(spa, SCL_ALL, FTAG);
2662 } else if (error == 0) {
2663 spa->spa_spares.sav_sync = B_TRUE;
2667 * Load any level 2 ARC devices for this pool.
2669 error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2670 &spa->spa_l2cache.sav_object);
2671 if (error != 0 && error != ENOENT)
2672 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2673 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2674 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2675 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2676 &spa->spa_l2cache.sav_config) != 0)
2677 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2679 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2680 spa_load_l2cache(spa);
2681 spa_config_exit(spa, SCL_ALL, FTAG);
2682 } else if (error == 0) {
2683 spa->spa_l2cache.sav_sync = B_TRUE;
2686 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2688 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2689 if (error && error != ENOENT)
2690 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2693 uint64_t autoreplace;
2695 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2696 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2697 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2698 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2699 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2700 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2701 &spa->spa_dedup_ditto);
2703 spa->spa_autoreplace = (autoreplace != 0);
2707 * If the 'autoreplace' property is set, then post a resource notifying
2708 * the ZFS DE that it should not issue any faults for unopenable
2709 * devices. We also iterate over the vdevs, and post a sysevent for any
2710 * unopenable vdevs so that the normal autoreplace handler can take
2713 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2714 spa_check_removed(spa->spa_root_vdev);
2716 * For the import case, this is done in spa_import(), because
2717 * at this point we're using the spare definitions from
2718 * the MOS config, not necessarily from the userland config.
2720 if (state != SPA_LOAD_IMPORT) {
2721 spa_aux_check_removed(&spa->spa_spares);
2722 spa_aux_check_removed(&spa->spa_l2cache);
2727 * Load the vdev state for all toplevel vdevs.
2732 * Propagate the leaf DTLs we just loaded all the way up the tree.
2734 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2735 vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2736 spa_config_exit(spa, SCL_ALL, FTAG);
2739 * Load the DDTs (dedup tables).
2741 error = ddt_load(spa);
2743 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2745 spa_update_dspace(spa);
2748 * Validate the config, using the MOS config to fill in any
2749 * information which might be missing. If we fail to validate
2750 * the config then declare the pool unfit for use. If we're
2751 * assembling a pool from a split, the log is not transferred
2754 if (type != SPA_IMPORT_ASSEMBLE) {
2757 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2758 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2760 if (!spa_config_valid(spa, nvconfig)) {
2761 nvlist_free(nvconfig);
2762 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2765 nvlist_free(nvconfig);
2768 * Now that we've validated the config, check the state of the
2769 * root vdev. If it can't be opened, it indicates one or
2770 * more toplevel vdevs are faulted.
2772 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2773 return (SET_ERROR(ENXIO));
2775 if (spa_writeable(spa) && spa_check_logs(spa)) {
2776 *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2777 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2781 if (missing_feat_write) {
2782 ASSERT(state == SPA_LOAD_TRYIMPORT);
2785 * At this point, we know that we can open the pool in
2786 * read-only mode but not read-write mode. We now have enough
2787 * information and can return to userland.
2789 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
2793 * We've successfully opened the pool, verify that we're ready
2794 * to start pushing transactions.
2796 if (state != SPA_LOAD_TRYIMPORT) {
2797 if (error = spa_load_verify(spa))
2798 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2802 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2803 spa->spa_load_max_txg == UINT64_MAX)) {
2805 int need_update = B_FALSE;
2806 dsl_pool_t *dp = spa_get_dsl(spa);
2808 ASSERT(state != SPA_LOAD_TRYIMPORT);
2811 * Claim log blocks that haven't been committed yet.
2812 * This must all happen in a single txg.
2813 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2814 * invoked from zil_claim_log_block()'s i/o done callback.
2815 * Price of rollback is that we abandon the log.
2817 spa->spa_claiming = B_TRUE;
2819 tx = dmu_tx_create_assigned(dp, spa_first_txg(spa));
2820 (void) dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
2821 zil_claim, tx, DS_FIND_CHILDREN);
2824 spa->spa_claiming = B_FALSE;
2826 spa_set_log_state(spa, SPA_LOG_GOOD);
2827 spa->spa_sync_on = B_TRUE;
2828 txg_sync_start(spa->spa_dsl_pool);
2831 * Wait for all claims to sync. We sync up to the highest
2832 * claimed log block birth time so that claimed log blocks
2833 * don't appear to be from the future. spa_claim_max_txg
2834 * will have been set for us by either zil_check_log_chain()
2835 * (invoked from spa_check_logs()) or zil_claim() above.
2837 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2840 * If the config cache is stale, or we have uninitialized
2841 * metaslabs (see spa_vdev_add()), then update the config.
2843 * If this is a verbatim import, trust the current
2844 * in-core spa_config and update the disk labels.
2846 if (config_cache_txg != spa->spa_config_txg ||
2847 state == SPA_LOAD_IMPORT ||
2848 state == SPA_LOAD_RECOVER ||
2849 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2850 need_update = B_TRUE;
2852 for (int c = 0; c < rvd->vdev_children; c++)
2853 if (rvd->vdev_child[c]->vdev_ms_array == 0)
2854 need_update = B_TRUE;
2857 * Update the config cache asychronously in case we're the
2858 * root pool, in which case the config cache isn't writable yet.
2861 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2864 * Check all DTLs to see if anything needs resilvering.
2866 if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2867 vdev_resilver_needed(rvd, NULL, NULL))
2868 spa_async_request(spa, SPA_ASYNC_RESILVER);
2871 * Log the fact that we booted up (so that we can detect if
2872 * we rebooted in the middle of an operation).
2874 spa_history_log_version(spa, "open");
2877 * Delete any inconsistent datasets.
2879 (void) dmu_objset_find(spa_name(spa),
2880 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2883 * Clean up any stale temporary dataset userrefs.
2885 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2892 spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2894 int mode = spa->spa_mode;
2897 spa_deactivate(spa);
2899 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg - 1;
2901 spa_activate(spa, mode);
2902 spa_async_suspend(spa);
2904 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2908 * If spa_load() fails this function will try loading prior txg's. If
2909 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2910 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2911 * function will not rewind the pool and will return the same error as
2915 spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2916 uint64_t max_request, int rewind_flags)
2918 nvlist_t *loadinfo = NULL;
2919 nvlist_t *config = NULL;
2920 int load_error, rewind_error;
2921 uint64_t safe_rewind_txg;
2924 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2925 spa->spa_load_max_txg = spa->spa_load_txg;
2926 spa_set_log_state(spa, SPA_LOG_CLEAR);
2928 spa->spa_load_max_txg = max_request;
2929 if (max_request != UINT64_MAX)
2930 spa->spa_extreme_rewind = B_TRUE;
2933 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2935 if (load_error == 0)
2938 if (spa->spa_root_vdev != NULL)
2939 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2941 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2942 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2944 if (rewind_flags & ZPOOL_NEVER_REWIND) {
2945 nvlist_free(config);
2946 return (load_error);
2949 if (state == SPA_LOAD_RECOVER) {
2950 /* Price of rolling back is discarding txgs, including log */
2951 spa_set_log_state(spa, SPA_LOG_CLEAR);
2954 * If we aren't rolling back save the load info from our first
2955 * import attempt so that we can restore it after attempting
2958 loadinfo = spa->spa_load_info;
2959 spa->spa_load_info = fnvlist_alloc();
2962 spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2963 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2964 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2965 TXG_INITIAL : safe_rewind_txg;
2968 * Continue as long as we're finding errors, we're still within
2969 * the acceptable rewind range, and we're still finding uberblocks
2971 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2972 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2973 if (spa->spa_load_max_txg < safe_rewind_txg)
2974 spa->spa_extreme_rewind = B_TRUE;
2975 rewind_error = spa_load_retry(spa, state, mosconfig);
2978 spa->spa_extreme_rewind = B_FALSE;
2979 spa->spa_load_max_txg = UINT64_MAX;
2981 if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2982 spa_config_set(spa, config);
2984 if (state == SPA_LOAD_RECOVER) {
2985 ASSERT3P(loadinfo, ==, NULL);
2986 return (rewind_error);
2988 /* Store the rewind info as part of the initial load info */
2989 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
2990 spa->spa_load_info);
2992 /* Restore the initial load info */
2993 fnvlist_free(spa->spa_load_info);
2994 spa->spa_load_info = loadinfo;
2996 return (load_error);
3003 * The import case is identical to an open except that the configuration is sent
3004 * down from userland, instead of grabbed from the configuration cache. For the
3005 * case of an open, the pool configuration will exist in the
3006 * POOL_STATE_UNINITIALIZED state.
3008 * The stats information (gen/count/ustats) is used to gather vdev statistics at
3009 * the same time open the pool, without having to keep around the spa_t in some
3013 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
3017 spa_load_state_t state = SPA_LOAD_OPEN;
3019 int locked = B_FALSE;
3020 int firstopen = B_FALSE;
3025 * As disgusting as this is, we need to support recursive calls to this
3026 * function because dsl_dir_open() is called during spa_load(), and ends
3027 * up calling spa_open() again. The real fix is to figure out how to
3028 * avoid dsl_dir_open() calling this in the first place.
3030 if (mutex_owner(&spa_namespace_lock) != curthread) {
3031 mutex_enter(&spa_namespace_lock);
3035 if ((spa = spa_lookup(pool)) == NULL) {
3037 mutex_exit(&spa_namespace_lock);
3038 return (SET_ERROR(ENOENT));
3041 if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
3042 zpool_rewind_policy_t policy;
3046 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
3048 if (policy.zrp_request & ZPOOL_DO_REWIND)
3049 state = SPA_LOAD_RECOVER;
3051 spa_activate(spa, spa_mode_global);
3053 if (state != SPA_LOAD_RECOVER)
3054 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3056 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
3057 policy.zrp_request);
3059 if (error == EBADF) {
3061 * If vdev_validate() returns failure (indicated by
3062 * EBADF), it indicates that one of the vdevs indicates
3063 * that the pool has been exported or destroyed. If
3064 * this is the case, the config cache is out of sync and
3065 * we should remove the pool from the namespace.
3068 spa_deactivate(spa);
3069 spa_config_sync(spa, B_TRUE, B_TRUE);
3072 mutex_exit(&spa_namespace_lock);
3073 return (SET_ERROR(ENOENT));
3078 * We can't open the pool, but we still have useful
3079 * information: the state of each vdev after the
3080 * attempted vdev_open(). Return this to the user.
3082 if (config != NULL && spa->spa_config) {
3083 VERIFY(nvlist_dup(spa->spa_config, config,
3085 VERIFY(nvlist_add_nvlist(*config,
3086 ZPOOL_CONFIG_LOAD_INFO,
3087 spa->spa_load_info) == 0);
3090 spa_deactivate(spa);
3091 spa->spa_last_open_failed = error;
3093 mutex_exit(&spa_namespace_lock);
3099 spa_open_ref(spa, tag);
3102 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
3105 * If we've recovered the pool, pass back any information we
3106 * gathered while doing the load.
3108 if (state == SPA_LOAD_RECOVER) {
3109 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
3110 spa->spa_load_info) == 0);
3114 spa->spa_last_open_failed = 0;
3115 spa->spa_last_ubsync_txg = 0;
3116 spa->spa_load_txg = 0;
3117 mutex_exit(&spa_namespace_lock);
3121 zvol_create_minors(spa->spa_name);
3132 spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
3135 return (spa_open_common(name, spapp, tag, policy, config));
3139 spa_open(const char *name, spa_t **spapp, void *tag)
3141 return (spa_open_common(name, spapp, tag, NULL, NULL));
3145 * Lookup the given spa_t, incrementing the inject count in the process,
3146 * preventing it from being exported or destroyed.
3149 spa_inject_addref(char *name)
3153 mutex_enter(&spa_namespace_lock);
3154 if ((spa = spa_lookup(name)) == NULL) {
3155 mutex_exit(&spa_namespace_lock);
3158 spa->spa_inject_ref++;
3159 mutex_exit(&spa_namespace_lock);
3165 spa_inject_delref(spa_t *spa)
3167 mutex_enter(&spa_namespace_lock);
3168 spa->spa_inject_ref--;
3169 mutex_exit(&spa_namespace_lock);
3173 * Add spares device information to the nvlist.
3176 spa_add_spares(spa_t *spa, nvlist_t *config)
3186 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3188 if (spa->spa_spares.sav_count == 0)
3191 VERIFY(nvlist_lookup_nvlist(config,
3192 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3193 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
3194 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3196 VERIFY(nvlist_add_nvlist_array(nvroot,
3197 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3198 VERIFY(nvlist_lookup_nvlist_array(nvroot,
3199 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3202 * Go through and find any spares which have since been
3203 * repurposed as an active spare. If this is the case, update
3204 * their status appropriately.
3206 for (i = 0; i < nspares; i++) {
3207 VERIFY(nvlist_lookup_uint64(spares[i],
3208 ZPOOL_CONFIG_GUID, &guid) == 0);
3209 if (spa_spare_exists(guid, &pool, NULL) &&
3211 VERIFY(nvlist_lookup_uint64_array(
3212 spares[i], ZPOOL_CONFIG_VDEV_STATS,
3213 (uint64_t **)&vs, &vsc) == 0);
3214 vs->vs_state = VDEV_STATE_CANT_OPEN;
3215 vs->vs_aux = VDEV_AUX_SPARED;
3222 * Add l2cache device information to the nvlist, including vdev stats.
3225 spa_add_l2cache(spa_t *spa, nvlist_t *config)
3228 uint_t i, j, nl2cache;
3235 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3237 if (spa->spa_l2cache.sav_count == 0)
3240 VERIFY(nvlist_lookup_nvlist(config,
3241 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3242 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
3243 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3244 if (nl2cache != 0) {
3245 VERIFY(nvlist_add_nvlist_array(nvroot,
3246 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3247 VERIFY(nvlist_lookup_nvlist_array(nvroot,
3248 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3251 * Update level 2 cache device stats.
3254 for (i = 0; i < nl2cache; i++) {
3255 VERIFY(nvlist_lookup_uint64(l2cache[i],
3256 ZPOOL_CONFIG_GUID, &guid) == 0);
3259 for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
3261 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
3262 vd = spa->spa_l2cache.sav_vdevs[j];
3268 VERIFY(nvlist_lookup_uint64_array(l2cache[i],
3269 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
3271 vdev_get_stats(vd, vs);
3277 spa_add_feature_stats(spa_t *spa, nvlist_t *config)
3283 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3284 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3286 /* We may be unable to read features if pool is suspended. */
3287 if (spa_suspended(spa))
3290 if (spa->spa_feat_for_read_obj != 0) {
3291 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3292 spa->spa_feat_for_read_obj);
3293 zap_cursor_retrieve(&zc, &za) == 0;
3294 zap_cursor_advance(&zc)) {
3295 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3296 za.za_num_integers == 1);
3297 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3298 za.za_first_integer));
3300 zap_cursor_fini(&zc);
3303 if (spa->spa_feat_for_write_obj != 0) {
3304 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3305 spa->spa_feat_for_write_obj);
3306 zap_cursor_retrieve(&zc, &za) == 0;
3307 zap_cursor_advance(&zc)) {
3308 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3309 za.za_num_integers == 1);
3310 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3311 za.za_first_integer));
3313 zap_cursor_fini(&zc);
3317 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
3319 nvlist_free(features);
3323 spa_get_stats(const char *name, nvlist_t **config,
3324 char *altroot, size_t buflen)
3330 error = spa_open_common(name, &spa, FTAG, NULL, config);
3334 * This still leaves a window of inconsistency where the spares
3335 * or l2cache devices could change and the config would be
3336 * self-inconsistent.
3338 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3340 if (*config != NULL) {
3341 uint64_t loadtimes[2];
3343 loadtimes[0] = spa->spa_loaded_ts.tv_sec;
3344 loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
3345 VERIFY(nvlist_add_uint64_array(*config,
3346 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
3348 VERIFY(nvlist_add_uint64(*config,
3349 ZPOOL_CONFIG_ERRCOUNT,
3350 spa_get_errlog_size(spa)) == 0);
3352 if (spa_suspended(spa))
3353 VERIFY(nvlist_add_uint64(*config,
3354 ZPOOL_CONFIG_SUSPENDED,
3355 spa->spa_failmode) == 0);
3357 spa_add_spares(spa, *config);
3358 spa_add_l2cache(spa, *config);
3359 spa_add_feature_stats(spa, *config);
3364 * We want to get the alternate root even for faulted pools, so we cheat
3365 * and call spa_lookup() directly.
3369 mutex_enter(&spa_namespace_lock);
3370 spa = spa_lookup(name);
3372 spa_altroot(spa, altroot, buflen);
3376 mutex_exit(&spa_namespace_lock);
3378 spa_altroot(spa, altroot, buflen);
3383 spa_config_exit(spa, SCL_CONFIG, FTAG);
3384 spa_close(spa, FTAG);
3391 * Validate that the auxiliary device array is well formed. We must have an
3392 * array of nvlists, each which describes a valid leaf vdev. If this is an
3393 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3394 * specified, as long as they are well-formed.
3397 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
3398 spa_aux_vdev_t *sav, const char *config, uint64_t version,
3399 vdev_labeltype_t label)
3406 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3409 * It's acceptable to have no devs specified.
3411 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
3415 return (SET_ERROR(EINVAL));
3418 * Make sure the pool is formatted with a version that supports this
3421 if (spa_version(spa) < version)
3422 return (SET_ERROR(ENOTSUP));
3425 * Set the pending device list so we correctly handle device in-use
3428 sav->sav_pending = dev;
3429 sav->sav_npending = ndev;
3431 for (i = 0; i < ndev; i++) {
3432 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
3436 if (!vd->vdev_ops->vdev_op_leaf) {
3438 error = SET_ERROR(EINVAL);
3443 * The L2ARC currently only supports disk devices in
3444 * kernel context. For user-level testing, we allow it.
3447 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
3448 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
3449 error = SET_ERROR(ENOTBLK);
3456 if ((error = vdev_open(vd)) == 0 &&
3457 (error = vdev_label_init(vd, crtxg, label)) == 0) {
3458 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
3459 vd->vdev_guid) == 0);
3465 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
3472 sav->sav_pending = NULL;
3473 sav->sav_npending = 0;
3478 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
3482 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3484 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3485 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
3486 VDEV_LABEL_SPARE)) != 0) {
3490 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3491 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
3492 VDEV_LABEL_L2CACHE));
3496 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
3501 if (sav->sav_config != NULL) {
3507 * Generate new dev list by concatentating with the
3510 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
3511 &olddevs, &oldndevs) == 0);
3513 newdevs = kmem_alloc(sizeof (void *) *
3514 (ndevs + oldndevs), KM_SLEEP);
3515 for (i = 0; i < oldndevs; i++)
3516 VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
3518 for (i = 0; i < ndevs; i++)
3519 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
3522 VERIFY(nvlist_remove(sav->sav_config, config,
3523 DATA_TYPE_NVLIST_ARRAY) == 0);
3525 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3526 config, newdevs, ndevs + oldndevs) == 0);
3527 for (i = 0; i < oldndevs + ndevs; i++)
3528 nvlist_free(newdevs[i]);
3529 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
3532 * Generate a new dev list.
3534 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
3536 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
3542 * Stop and drop level 2 ARC devices
3545 spa_l2cache_drop(spa_t *spa)
3549 spa_aux_vdev_t *sav = &spa->spa_l2cache;
3551 for (i = 0; i < sav->sav_count; i++) {
3554 vd = sav->sav_vdevs[i];
3557 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
3558 pool != 0ULL && l2arc_vdev_present(vd))
3559 l2arc_remove_vdev(vd);
3567 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
3571 char *altroot = NULL;
3576 uint64_t txg = TXG_INITIAL;
3577 nvlist_t **spares, **l2cache;
3578 uint_t nspares, nl2cache;
3579 uint64_t version, obj;
3580 boolean_t has_features;
3583 * If this pool already exists, return failure.
3585 mutex_enter(&spa_namespace_lock);
3586 if (spa_lookup(pool) != NULL) {
3587 mutex_exit(&spa_namespace_lock);
3588 return (SET_ERROR(EEXIST));
3592 * Allocate a new spa_t structure.
3594 (void) nvlist_lookup_string(props,
3595 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3596 spa = spa_add(pool, NULL, altroot);
3597 spa_activate(spa, spa_mode_global);
3599 if (props && (error = spa_prop_validate(spa, props))) {
3600 spa_deactivate(spa);
3602 mutex_exit(&spa_namespace_lock);
3606 has_features = B_FALSE;
3607 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
3608 elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
3609 if (zpool_prop_feature(nvpair_name(elem)))
3610 has_features = B_TRUE;
3613 if (has_features || nvlist_lookup_uint64(props,
3614 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
3615 version = SPA_VERSION;
3617 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
3619 spa->spa_first_txg = txg;
3620 spa->spa_uberblock.ub_txg = txg - 1;
3621 spa->spa_uberblock.ub_version = version;
3622 spa->spa_ubsync = spa->spa_uberblock;
3625 * Create "The Godfather" zio to hold all async IOs
3627 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
3629 for (int i = 0; i < max_ncpus; i++) {
3630 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
3631 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
3632 ZIO_FLAG_GODFATHER);
3636 * Create the root vdev.
3638 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3640 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3642 ASSERT(error != 0 || rvd != NULL);
3643 ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3645 if (error == 0 && !zfs_allocatable_devs(nvroot))
3646 error = SET_ERROR(EINVAL);
3649 (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3650 (error = spa_validate_aux(spa, nvroot, txg,
3651 VDEV_ALLOC_ADD)) == 0) {
3652 for (int c = 0; c < rvd->vdev_children; c++) {
3653 vdev_ashift_optimize(rvd->vdev_child[c]);
3654 vdev_metaslab_set_size(rvd->vdev_child[c]);
3655 vdev_expand(rvd->vdev_child[c], txg);
3659 spa_config_exit(spa, SCL_ALL, FTAG);
3663 spa_deactivate(spa);
3665 mutex_exit(&spa_namespace_lock);
3670 * Get the list of spares, if specified.
3672 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3673 &spares, &nspares) == 0) {
3674 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3676 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3677 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3678 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3679 spa_load_spares(spa);
3680 spa_config_exit(spa, SCL_ALL, FTAG);
3681 spa->spa_spares.sav_sync = B_TRUE;
3685 * Get the list of level 2 cache devices, if specified.
3687 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3688 &l2cache, &nl2cache) == 0) {
3689 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3690 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3691 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3692 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3693 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3694 spa_load_l2cache(spa);
3695 spa_config_exit(spa, SCL_ALL, FTAG);
3696 spa->spa_l2cache.sav_sync = B_TRUE;
3699 spa->spa_is_initializing = B_TRUE;
3700 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3701 spa->spa_meta_objset = dp->dp_meta_objset;
3702 spa->spa_is_initializing = B_FALSE;
3705 * Create DDTs (dedup tables).
3709 spa_update_dspace(spa);
3711 tx = dmu_tx_create_assigned(dp, txg);
3714 * Create the pool config object.
3716 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3717 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3718 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3720 if (zap_add(spa->spa_meta_objset,
3721 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3722 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3723 cmn_err(CE_PANIC, "failed to add pool config");
3726 if (spa_version(spa) >= SPA_VERSION_FEATURES)
3727 spa_feature_create_zap_objects(spa, tx);
3729 if (zap_add(spa->spa_meta_objset,
3730 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3731 sizeof (uint64_t), 1, &version, tx) != 0) {
3732 cmn_err(CE_PANIC, "failed to add pool version");
3735 /* Newly created pools with the right version are always deflated. */
3736 if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3737 spa->spa_deflate = TRUE;
3738 if (zap_add(spa->spa_meta_objset,
3739 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3740 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3741 cmn_err(CE_PANIC, "failed to add deflate");
3746 * Create the deferred-free bpobj. Turn off compression
3747 * because sync-to-convergence takes longer if the blocksize
3750 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3751 dmu_object_set_compress(spa->spa_meta_objset, obj,
3752 ZIO_COMPRESS_OFF, tx);
3753 if (zap_add(spa->spa_meta_objset,
3754 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3755 sizeof (uint64_t), 1, &obj, tx) != 0) {
3756 cmn_err(CE_PANIC, "failed to add bpobj");
3758 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3759 spa->spa_meta_objset, obj));
3762 * Create the pool's history object.
3764 if (version >= SPA_VERSION_ZPOOL_HISTORY)
3765 spa_history_create_obj(spa, tx);
3768 * Generate some random noise for salted checksums to operate on.
3770 (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes,
3771 sizeof (spa->spa_cksum_salt.zcs_bytes));
3774 * Set pool properties.
3776 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3777 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3778 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3779 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3781 if (props != NULL) {
3782 spa_configfile_set(spa, props, B_FALSE);
3783 spa_sync_props(props, tx);
3788 spa->spa_sync_on = B_TRUE;
3789 txg_sync_start(spa->spa_dsl_pool);
3792 * We explicitly wait for the first transaction to complete so that our
3793 * bean counters are appropriately updated.
3795 txg_wait_synced(spa->spa_dsl_pool, txg);
3797 spa_config_sync(spa, B_FALSE, B_TRUE);
3798 spa_event_notify(spa, NULL, ESC_ZFS_POOL_CREATE);
3800 spa_history_log_version(spa, "create");
3803 * Don't count references from objsets that are already closed
3804 * and are making their way through the eviction process.
3806 spa_evicting_os_wait(spa);
3807 spa->spa_minref = refcount_count(&spa->spa_refcount);
3809 mutex_exit(&spa_namespace_lock);
3817 * Get the root pool information from the root disk, then import the root pool
3818 * during the system boot up time.
3820 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3823 spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3826 nvlist_t *nvtop, *nvroot;
3829 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3833 * Add this top-level vdev to the child array.
3835 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3837 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3839 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3842 * Put this pool's top-level vdevs into a root vdev.
3844 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3845 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3846 VDEV_TYPE_ROOT) == 0);
3847 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3848 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3849 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3853 * Replace the existing vdev_tree with the new root vdev in
3854 * this pool's configuration (remove the old, add the new).
3856 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3857 nvlist_free(nvroot);
3862 * Walk the vdev tree and see if we can find a device with "better"
3863 * configuration. A configuration is "better" if the label on that
3864 * device has a more recent txg.
3867 spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3869 for (int c = 0; c < vd->vdev_children; c++)
3870 spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3872 if (vd->vdev_ops->vdev_op_leaf) {
3876 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3880 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3884 * Do we have a better boot device?
3886 if (label_txg > *txg) {
3895 * Import a root pool.
3897 * For x86. devpath_list will consist of devid and/or physpath name of
3898 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3899 * The GRUB "findroot" command will return the vdev we should boot.
3901 * For Sparc, devpath_list consists the physpath name of the booting device
3902 * no matter the rootpool is a single device pool or a mirrored pool.
3904 * "/pci@1f,0/ide@d/disk@0,0:a"
3907 spa_import_rootpool(char *devpath, char *devid)
3910 vdev_t *rvd, *bvd, *avd = NULL;
3911 nvlist_t *config, *nvtop;
3917 * Read the label from the boot device and generate a configuration.
3919 config = spa_generate_rootconf(devpath, devid, &guid);
3920 #if defined(_OBP) && defined(_KERNEL)
3921 if (config == NULL) {
3922 if (strstr(devpath, "/iscsi/ssd") != NULL) {
3924 get_iscsi_bootpath_phy(devpath);
3925 config = spa_generate_rootconf(devpath, devid, &guid);
3929 if (config == NULL) {
3930 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
3932 return (SET_ERROR(EIO));
3935 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3937 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3939 mutex_enter(&spa_namespace_lock);
3940 if ((spa = spa_lookup(pname)) != NULL) {
3942 * Remove the existing root pool from the namespace so that we
3943 * can replace it with the correct config we just read in.
3948 spa = spa_add(pname, config, NULL);
3949 spa->spa_is_root = B_TRUE;
3950 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3953 * Build up a vdev tree based on the boot device's label config.
3955 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3957 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3958 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3959 VDEV_ALLOC_ROOTPOOL);
3960 spa_config_exit(spa, SCL_ALL, FTAG);
3962 mutex_exit(&spa_namespace_lock);
3963 nvlist_free(config);
3964 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3970 * Get the boot vdev.
3972 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3973 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3974 (u_longlong_t)guid);
3975 error = SET_ERROR(ENOENT);
3980 * Determine if there is a better boot device.
3983 spa_alt_rootvdev(rvd, &avd, &txg);
3985 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3986 "try booting from '%s'", avd->vdev_path);
3987 error = SET_ERROR(EINVAL);
3992 * If the boot device is part of a spare vdev then ensure that
3993 * we're booting off the active spare.
3995 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3996 !bvd->vdev_isspare) {
3997 cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3998 "try booting from '%s'",
4000 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
4001 error = SET_ERROR(EINVAL);
4007 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4009 spa_config_exit(spa, SCL_ALL, FTAG);
4010 mutex_exit(&spa_namespace_lock);
4012 nvlist_free(config);
4016 #else /* !illumos */
4018 extern int vdev_geom_read_pool_label(const char *name, nvlist_t ***configs,
4022 spa_generate_rootconf(const char *name)
4024 nvlist_t **configs, **tops;
4026 nvlist_t *best_cfg, *nvtop, *nvroot;
4035 if (vdev_geom_read_pool_label(name, &configs, &count) != 0)
4038 ASSERT3U(count, !=, 0);
4040 for (i = 0; i < count; i++) {
4043 VERIFY(nvlist_lookup_uint64(configs[i], ZPOOL_CONFIG_POOL_TXG,
4045 if (txg > best_txg) {
4047 best_cfg = configs[i];
4052 * Multi-vdev root pool configuration discovery is not supported yet.
4055 nvlist_lookup_uint64(best_cfg, ZPOOL_CONFIG_VDEV_CHILDREN, &nchildren);
4057 nvlist_lookup_uint64_array(best_cfg, ZPOOL_CONFIG_HOLE_ARRAY,
4060 tops = kmem_zalloc(nchildren * sizeof(void *), KM_SLEEP);
4061 for (i = 0; i < nchildren; i++) {
4064 if (configs[i] == NULL)
4066 VERIFY(nvlist_lookup_nvlist(configs[i], ZPOOL_CONFIG_VDEV_TREE,
4068 nvlist_dup(nvtop, &tops[i], KM_SLEEP);
4070 for (i = 0; holes != NULL && i < nholes; i++) {
4073 if (tops[holes[i]] != NULL)
4075 nvlist_alloc(&tops[holes[i]], NV_UNIQUE_NAME, KM_SLEEP);
4076 VERIFY(nvlist_add_string(tops[holes[i]], ZPOOL_CONFIG_TYPE,
4077 VDEV_TYPE_HOLE) == 0);
4078 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_ID,
4080 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_GUID,
4083 for (i = 0; i < nchildren; i++) {
4084 if (tops[i] != NULL)
4086 nvlist_alloc(&tops[i], NV_UNIQUE_NAME, KM_SLEEP);
4087 VERIFY(nvlist_add_string(tops[i], ZPOOL_CONFIG_TYPE,
4088 VDEV_TYPE_MISSING) == 0);
4089 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_ID,
4091 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_GUID,
4096 * Create pool config based on the best vdev config.
4098 nvlist_dup(best_cfg, &config, KM_SLEEP);
4101 * Put this pool's top-level vdevs into a root vdev.
4103 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4105 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4106 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
4107 VDEV_TYPE_ROOT) == 0);
4108 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
4109 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
4110 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
4111 tops, nchildren) == 0);
4114 * Replace the existing vdev_tree with the new root vdev in
4115 * this pool's configuration (remove the old, add the new).
4117 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
4120 * Drop vdev config elements that should not be present at pool level.
4122 nvlist_remove(config, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64);
4123 nvlist_remove(config, ZPOOL_CONFIG_TOP_GUID, DATA_TYPE_UINT64);
4125 for (i = 0; i < count; i++)
4126 nvlist_free(configs[i]);
4127 kmem_free(configs, count * sizeof(void *));
4128 for (i = 0; i < nchildren; i++)
4129 nvlist_free(tops[i]);
4130 kmem_free(tops, nchildren * sizeof(void *));
4131 nvlist_free(nvroot);
4136 spa_import_rootpool(const char *name)
4139 vdev_t *rvd, *bvd, *avd = NULL;
4140 nvlist_t *config, *nvtop;
4146 * Read the label from the boot device and generate a configuration.
4148 config = spa_generate_rootconf(name);
4150 mutex_enter(&spa_namespace_lock);
4151 if (config != NULL) {
4152 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
4153 &pname) == 0 && strcmp(name, pname) == 0);
4154 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg)
4157 if ((spa = spa_lookup(pname)) != NULL) {
4159 * Remove the existing root pool from the namespace so
4160 * that we can replace it with the correct config
4165 spa = spa_add(pname, config, NULL);
4168 * Set spa_ubsync.ub_version as it can be used in vdev_alloc()
4169 * via spa_version().
4171 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
4172 &spa->spa_ubsync.ub_version) != 0)
4173 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
4174 } else if ((spa = spa_lookup(name)) == NULL) {
4175 mutex_exit(&spa_namespace_lock);
4176 nvlist_free(config);
4177 cmn_err(CE_NOTE, "Cannot find the pool label for '%s'",
4181 VERIFY(nvlist_dup(spa->spa_config, &config, KM_SLEEP) == 0);
4183 spa->spa_is_root = B_TRUE;
4184 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
4187 * Build up a vdev tree based on the boot device's label config.
4189 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4191 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4192 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
4193 VDEV_ALLOC_ROOTPOOL);
4194 spa_config_exit(spa, SCL_ALL, FTAG);
4196 mutex_exit(&spa_namespace_lock);
4197 nvlist_free(config);
4198 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
4203 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4205 spa_config_exit(spa, SCL_ALL, FTAG);
4206 mutex_exit(&spa_namespace_lock);
4208 nvlist_free(config);
4212 #endif /* illumos */
4213 #endif /* _KERNEL */
4216 * Import a non-root pool into the system.
4219 spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
4222 char *altroot = NULL;
4223 spa_load_state_t state = SPA_LOAD_IMPORT;
4224 zpool_rewind_policy_t policy;
4225 uint64_t mode = spa_mode_global;
4226 uint64_t readonly = B_FALSE;
4229 nvlist_t **spares, **l2cache;
4230 uint_t nspares, nl2cache;
4233 * If a pool with this name exists, return failure.
4235 mutex_enter(&spa_namespace_lock);
4236 if (spa_lookup(pool) != NULL) {
4237 mutex_exit(&spa_namespace_lock);
4238 return (SET_ERROR(EEXIST));
4242 * Create and initialize the spa structure.
4244 (void) nvlist_lookup_string(props,
4245 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4246 (void) nvlist_lookup_uint64(props,
4247 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
4250 spa = spa_add(pool, config, altroot);
4251 spa->spa_import_flags = flags;
4254 * Verbatim import - Take a pool and insert it into the namespace
4255 * as if it had been loaded at boot.
4257 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
4259 spa_configfile_set(spa, props, B_FALSE);
4261 spa_config_sync(spa, B_FALSE, B_TRUE);
4262 spa_event_notify(spa, NULL, ESC_ZFS_POOL_IMPORT);
4264 mutex_exit(&spa_namespace_lock);
4268 spa_activate(spa, mode);
4271 * Don't start async tasks until we know everything is healthy.
4273 spa_async_suspend(spa);
4275 zpool_get_rewind_policy(config, &policy);
4276 if (policy.zrp_request & ZPOOL_DO_REWIND)
4277 state = SPA_LOAD_RECOVER;
4280 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
4281 * because the user-supplied config is actually the one to trust when
4284 if (state != SPA_LOAD_RECOVER)
4285 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
4287 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
4288 policy.zrp_request);
4291 * Propagate anything learned while loading the pool and pass it
4292 * back to caller (i.e. rewind info, missing devices, etc).
4294 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4295 spa->spa_load_info) == 0);
4297 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4299 * Toss any existing sparelist, as it doesn't have any validity
4300 * anymore, and conflicts with spa_has_spare().
4302 if (spa->spa_spares.sav_config) {
4303 nvlist_free(spa->spa_spares.sav_config);
4304 spa->spa_spares.sav_config = NULL;
4305 spa_load_spares(spa);
4307 if (spa->spa_l2cache.sav_config) {
4308 nvlist_free(spa->spa_l2cache.sav_config);
4309 spa->spa_l2cache.sav_config = NULL;
4310 spa_load_l2cache(spa);
4313 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4316 error = spa_validate_aux(spa, nvroot, -1ULL,
4319 error = spa_validate_aux(spa, nvroot, -1ULL,
4320 VDEV_ALLOC_L2CACHE);
4321 spa_config_exit(spa, SCL_ALL, FTAG);
4324 spa_configfile_set(spa, props, B_FALSE);
4326 if (error != 0 || (props && spa_writeable(spa) &&
4327 (error = spa_prop_set(spa, props)))) {
4329 spa_deactivate(spa);
4331 mutex_exit(&spa_namespace_lock);
4335 spa_async_resume(spa);
4338 * Override any spares and level 2 cache devices as specified by
4339 * the user, as these may have correct device names/devids, etc.
4341 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
4342 &spares, &nspares) == 0) {
4343 if (spa->spa_spares.sav_config)
4344 VERIFY(nvlist_remove(spa->spa_spares.sav_config,
4345 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
4347 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
4348 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4349 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
4350 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
4351 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4352 spa_load_spares(spa);
4353 spa_config_exit(spa, SCL_ALL, FTAG);
4354 spa->spa_spares.sav_sync = B_TRUE;
4356 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
4357 &l2cache, &nl2cache) == 0) {
4358 if (spa->spa_l2cache.sav_config)
4359 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
4360 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
4362 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
4363 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4364 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
4365 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
4366 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4367 spa_load_l2cache(spa);
4368 spa_config_exit(spa, SCL_ALL, FTAG);
4369 spa->spa_l2cache.sav_sync = B_TRUE;
4373 * Check for any removed devices.
4375 if (spa->spa_autoreplace) {
4376 spa_aux_check_removed(&spa->spa_spares);
4377 spa_aux_check_removed(&spa->spa_l2cache);
4380 if (spa_writeable(spa)) {
4382 * Update the config cache to include the newly-imported pool.
4384 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4388 * It's possible that the pool was expanded while it was exported.
4389 * We kick off an async task to handle this for us.
4391 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
4393 spa_history_log_version(spa, "import");
4395 spa_event_notify(spa, NULL, ESC_ZFS_POOL_IMPORT);
4397 mutex_exit(&spa_namespace_lock);
4401 zvol_create_minors(pool);
4408 spa_tryimport(nvlist_t *tryconfig)
4410 nvlist_t *config = NULL;
4416 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
4419 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
4423 * Create and initialize the spa structure.
4425 mutex_enter(&spa_namespace_lock);
4426 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
4427 spa_activate(spa, FREAD);
4430 * Pass off the heavy lifting to spa_load().
4431 * Pass TRUE for mosconfig because the user-supplied config
4432 * is actually the one to trust when doing an import.
4434 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
4437 * If 'tryconfig' was at least parsable, return the current config.
4439 if (spa->spa_root_vdev != NULL) {
4440 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4441 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
4443 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4445 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
4446 spa->spa_uberblock.ub_timestamp) == 0);
4447 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4448 spa->spa_load_info) == 0);
4451 * If the bootfs property exists on this pool then we
4452 * copy it out so that external consumers can tell which
4453 * pools are bootable.
4455 if ((!error || error == EEXIST) && spa->spa_bootfs) {
4456 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4459 * We have to play games with the name since the
4460 * pool was opened as TRYIMPORT_NAME.
4462 if (dsl_dsobj_to_dsname(spa_name(spa),
4463 spa->spa_bootfs, tmpname) == 0) {
4465 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4467 cp = strchr(tmpname, '/');
4469 (void) strlcpy(dsname, tmpname,
4472 (void) snprintf(dsname, MAXPATHLEN,
4473 "%s/%s", poolname, ++cp);
4475 VERIFY(nvlist_add_string(config,
4476 ZPOOL_CONFIG_BOOTFS, dsname) == 0);
4477 kmem_free(dsname, MAXPATHLEN);
4479 kmem_free(tmpname, MAXPATHLEN);
4483 * Add the list of hot spares and level 2 cache devices.
4485 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4486 spa_add_spares(spa, config);
4487 spa_add_l2cache(spa, config);
4488 spa_config_exit(spa, SCL_CONFIG, FTAG);
4492 spa_deactivate(spa);
4494 mutex_exit(&spa_namespace_lock);
4500 * Pool export/destroy
4502 * The act of destroying or exporting a pool is very simple. We make sure there
4503 * is no more pending I/O and any references to the pool are gone. Then, we
4504 * update the pool state and sync all the labels to disk, removing the
4505 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4506 * we don't sync the labels or remove the configuration cache.
4509 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
4510 boolean_t force, boolean_t hardforce)
4517 if (!(spa_mode_global & FWRITE))
4518 return (SET_ERROR(EROFS));
4520 mutex_enter(&spa_namespace_lock);
4521 if ((spa = spa_lookup(pool)) == NULL) {
4522 mutex_exit(&spa_namespace_lock);
4523 return (SET_ERROR(ENOENT));
4527 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4528 * reacquire the namespace lock, and see if we can export.
4530 spa_open_ref(spa, FTAG);
4531 mutex_exit(&spa_namespace_lock);
4532 spa_async_suspend(spa);
4533 mutex_enter(&spa_namespace_lock);
4534 spa_close(spa, FTAG);
4537 * The pool will be in core if it's openable,
4538 * in which case we can modify its state.
4540 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
4542 * Objsets may be open only because they're dirty, so we
4543 * have to force it to sync before checking spa_refcnt.
4545 txg_wait_synced(spa->spa_dsl_pool, 0);
4546 spa_evicting_os_wait(spa);
4549 * A pool cannot be exported or destroyed if there are active
4550 * references. If we are resetting a pool, allow references by
4551 * fault injection handlers.
4553 if (!spa_refcount_zero(spa) ||
4554 (spa->spa_inject_ref != 0 &&
4555 new_state != POOL_STATE_UNINITIALIZED)) {
4556 spa_async_resume(spa);
4557 mutex_exit(&spa_namespace_lock);
4558 return (SET_ERROR(EBUSY));
4562 * A pool cannot be exported if it has an active shared spare.
4563 * This is to prevent other pools stealing the active spare
4564 * from an exported pool. At user's own will, such pool can
4565 * be forcedly exported.
4567 if (!force && new_state == POOL_STATE_EXPORTED &&
4568 spa_has_active_shared_spare(spa)) {
4569 spa_async_resume(spa);
4570 mutex_exit(&spa_namespace_lock);
4571 return (SET_ERROR(EXDEV));
4575 * We want this to be reflected on every label,
4576 * so mark them all dirty. spa_unload() will do the
4577 * final sync that pushes these changes out.
4579 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
4580 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4581 spa->spa_state = new_state;
4582 spa->spa_final_txg = spa_last_synced_txg(spa) +
4584 vdev_config_dirty(spa->spa_root_vdev);
4585 spa_config_exit(spa, SCL_ALL, FTAG);
4589 spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
4591 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4593 spa_deactivate(spa);
4596 if (oldconfig && spa->spa_config)
4597 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
4599 if (new_state != POOL_STATE_UNINITIALIZED) {
4601 spa_config_sync(spa, B_TRUE, B_TRUE);
4604 mutex_exit(&spa_namespace_lock);
4610 * Destroy a storage pool.
4613 spa_destroy(char *pool)
4615 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
4620 * Export a storage pool.
4623 spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
4624 boolean_t hardforce)
4626 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
4631 * Similar to spa_export(), this unloads the spa_t without actually removing it
4632 * from the namespace in any way.
4635 spa_reset(char *pool)
4637 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
4642 * ==========================================================================
4643 * Device manipulation
4644 * ==========================================================================
4648 * Add a device to a storage pool.
4651 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
4655 vdev_t *rvd = spa->spa_root_vdev;
4657 nvlist_t **spares, **l2cache;
4658 uint_t nspares, nl2cache;
4660 ASSERT(spa_writeable(spa));
4662 txg = spa_vdev_enter(spa);
4664 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
4665 VDEV_ALLOC_ADD)) != 0)
4666 return (spa_vdev_exit(spa, NULL, txg, error));
4668 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */
4670 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
4674 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
4678 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
4679 return (spa_vdev_exit(spa, vd, txg, EINVAL));
4681 if (vd->vdev_children != 0 &&
4682 (error = vdev_create(vd, txg, B_FALSE)) != 0)
4683 return (spa_vdev_exit(spa, vd, txg, error));
4686 * We must validate the spares and l2cache devices after checking the
4687 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
4689 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
4690 return (spa_vdev_exit(spa, vd, txg, error));
4693 * Transfer each new top-level vdev from vd to rvd.
4695 for (int c = 0; c < vd->vdev_children; c++) {
4698 * Set the vdev id to the first hole, if one exists.
4700 for (id = 0; id < rvd->vdev_children; id++) {
4701 if (rvd->vdev_child[id]->vdev_ishole) {
4702 vdev_free(rvd->vdev_child[id]);
4706 tvd = vd->vdev_child[c];
4707 vdev_remove_child(vd, tvd);
4709 vdev_add_child(rvd, tvd);
4710 vdev_config_dirty(tvd);
4714 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
4715 ZPOOL_CONFIG_SPARES);
4716 spa_load_spares(spa);
4717 spa->spa_spares.sav_sync = B_TRUE;
4720 if (nl2cache != 0) {
4721 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
4722 ZPOOL_CONFIG_L2CACHE);
4723 spa_load_l2cache(spa);
4724 spa->spa_l2cache.sav_sync = B_TRUE;
4728 * We have to be careful when adding new vdevs to an existing pool.
4729 * If other threads start allocating from these vdevs before we
4730 * sync the config cache, and we lose power, then upon reboot we may
4731 * fail to open the pool because there are DVAs that the config cache
4732 * can't translate. Therefore, we first add the vdevs without
4733 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4734 * and then let spa_config_update() initialize the new metaslabs.
4736 * spa_load() checks for added-but-not-initialized vdevs, so that
4737 * if we lose power at any point in this sequence, the remaining
4738 * steps will be completed the next time we load the pool.
4740 (void) spa_vdev_exit(spa, vd, txg, 0);
4742 mutex_enter(&spa_namespace_lock);
4743 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4744 spa_event_notify(spa, NULL, ESC_ZFS_VDEV_ADD);
4745 mutex_exit(&spa_namespace_lock);
4751 * Attach a device to a mirror. The arguments are the path to any device
4752 * in the mirror, and the nvroot for the new device. If the path specifies
4753 * a device that is not mirrored, we automatically insert the mirror vdev.
4755 * If 'replacing' is specified, the new device is intended to replace the
4756 * existing device; in this case the two devices are made into their own
4757 * mirror using the 'replacing' vdev, which is functionally identical to
4758 * the mirror vdev (it actually reuses all the same ops) but has a few
4759 * extra rules: you can't attach to it after it's been created, and upon
4760 * completion of resilvering, the first disk (the one being replaced)
4761 * is automatically detached.
4764 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
4766 uint64_t txg, dtl_max_txg;
4767 vdev_t *rvd = spa->spa_root_vdev;
4768 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
4770 char *oldvdpath, *newvdpath;
4774 ASSERT(spa_writeable(spa));
4776 txg = spa_vdev_enter(spa);
4778 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
4781 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4783 if (!oldvd->vdev_ops->vdev_op_leaf)
4784 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4786 pvd = oldvd->vdev_parent;
4788 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
4789 VDEV_ALLOC_ATTACH)) != 0)
4790 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4792 if (newrootvd->vdev_children != 1)
4793 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4795 newvd = newrootvd->vdev_child[0];
4797 if (!newvd->vdev_ops->vdev_op_leaf)
4798 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4800 if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
4801 return (spa_vdev_exit(spa, newrootvd, txg, error));
4804 * Spares can't replace logs
4806 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
4807 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4811 * For attach, the only allowable parent is a mirror or the root
4814 if (pvd->vdev_ops != &vdev_mirror_ops &&
4815 pvd->vdev_ops != &vdev_root_ops)
4816 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4818 pvops = &vdev_mirror_ops;
4821 * Active hot spares can only be replaced by inactive hot
4824 if (pvd->vdev_ops == &vdev_spare_ops &&
4825 oldvd->vdev_isspare &&
4826 !spa_has_spare(spa, newvd->vdev_guid))
4827 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4830 * If the source is a hot spare, and the parent isn't already a
4831 * spare, then we want to create a new hot spare. Otherwise, we
4832 * want to create a replacing vdev. The user is not allowed to
4833 * attach to a spared vdev child unless the 'isspare' state is
4834 * the same (spare replaces spare, non-spare replaces
4837 if (pvd->vdev_ops == &vdev_replacing_ops &&
4838 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
4839 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4840 } else if (pvd->vdev_ops == &vdev_spare_ops &&
4841 newvd->vdev_isspare != oldvd->vdev_isspare) {
4842 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4845 if (newvd->vdev_isspare)
4846 pvops = &vdev_spare_ops;
4848 pvops = &vdev_replacing_ops;
4852 * Make sure the new device is big enough.
4854 if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
4855 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
4858 * The new device cannot have a higher alignment requirement
4859 * than the top-level vdev.
4861 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4862 return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4865 * If this is an in-place replacement, update oldvd's path and devid
4866 * to make it distinguishable from newvd, and unopenable from now on.
4868 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4869 spa_strfree(oldvd->vdev_path);
4870 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4872 (void) sprintf(oldvd->vdev_path, "%s/%s",
4873 newvd->vdev_path, "old");
4874 if (oldvd->vdev_devid != NULL) {
4875 spa_strfree(oldvd->vdev_devid);
4876 oldvd->vdev_devid = NULL;
4880 /* mark the device being resilvered */
4881 newvd->vdev_resilver_txg = txg;
4884 * If the parent is not a mirror, or if we're replacing, insert the new
4885 * mirror/replacing/spare vdev above oldvd.
4887 if (pvd->vdev_ops != pvops)
4888 pvd = vdev_add_parent(oldvd, pvops);
4890 ASSERT(pvd->vdev_top->vdev_parent == rvd);
4891 ASSERT(pvd->vdev_ops == pvops);
4892 ASSERT(oldvd->vdev_parent == pvd);
4895 * Extract the new device from its root and add it to pvd.
4897 vdev_remove_child(newrootvd, newvd);
4898 newvd->vdev_id = pvd->vdev_children;
4899 newvd->vdev_crtxg = oldvd->vdev_crtxg;
4900 vdev_add_child(pvd, newvd);
4902 tvd = newvd->vdev_top;
4903 ASSERT(pvd->vdev_top == tvd);
4904 ASSERT(tvd->vdev_parent == rvd);
4906 vdev_config_dirty(tvd);
4909 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4910 * for any dmu_sync-ed blocks. It will propagate upward when
4911 * spa_vdev_exit() calls vdev_dtl_reassess().
4913 dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4915 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4916 dtl_max_txg - TXG_INITIAL);
4918 if (newvd->vdev_isspare) {
4919 spa_spare_activate(newvd);
4920 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
4923 oldvdpath = spa_strdup(oldvd->vdev_path);
4924 newvdpath = spa_strdup(newvd->vdev_path);
4925 newvd_isspare = newvd->vdev_isspare;
4928 * Mark newvd's DTL dirty in this txg.
4930 vdev_dirty(tvd, VDD_DTL, newvd, txg);
4933 * Schedule the resilver to restart in the future. We do this to
4934 * ensure that dmu_sync-ed blocks have been stitched into the
4935 * respective datasets.
4937 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4939 if (spa->spa_bootfs)
4940 spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
4942 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_ATTACH);
4947 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4949 spa_history_log_internal(spa, "vdev attach", NULL,
4950 "%s vdev=%s %s vdev=%s",
4951 replacing && newvd_isspare ? "spare in" :
4952 replacing ? "replace" : "attach", newvdpath,
4953 replacing ? "for" : "to", oldvdpath);
4955 spa_strfree(oldvdpath);
4956 spa_strfree(newvdpath);
4962 * Detach a device from a mirror or replacing vdev.
4964 * If 'replace_done' is specified, only detach if the parent
4965 * is a replacing vdev.
4968 spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4972 vdev_t *rvd = spa->spa_root_vdev;
4973 vdev_t *vd, *pvd, *cvd, *tvd;
4974 boolean_t unspare = B_FALSE;
4975 uint64_t unspare_guid = 0;
4978 ASSERT(spa_writeable(spa));
4980 txg = spa_vdev_enter(spa);
4982 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4985 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4987 if (!vd->vdev_ops->vdev_op_leaf)
4988 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4990 pvd = vd->vdev_parent;
4993 * If the parent/child relationship is not as expected, don't do it.
4994 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4995 * vdev that's replacing B with C. The user's intent in replacing
4996 * is to go from M(A,B) to M(A,C). If the user decides to cancel
4997 * the replace by detaching C, the expected behavior is to end up
4998 * M(A,B). But suppose that right after deciding to detach C,
4999 * the replacement of B completes. We would have M(A,C), and then
5000 * ask to detach C, which would leave us with just A -- not what
5001 * the user wanted. To prevent this, we make sure that the
5002 * parent/child relationship hasn't changed -- in this example,
5003 * that C's parent is still the replacing vdev R.
5005 if (pvd->vdev_guid != pguid && pguid != 0)
5006 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
5009 * Only 'replacing' or 'spare' vdevs can be replaced.
5011 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
5012 pvd->vdev_ops != &vdev_spare_ops)
5013 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
5015 ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
5016 spa_version(spa) >= SPA_VERSION_SPARES);
5019 * Only mirror, replacing, and spare vdevs support detach.
5021 if (pvd->vdev_ops != &vdev_replacing_ops &&
5022 pvd->vdev_ops != &vdev_mirror_ops &&
5023 pvd->vdev_ops != &vdev_spare_ops)
5024 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
5027 * If this device has the only valid copy of some data,
5028 * we cannot safely detach it.
5030 if (vdev_dtl_required(vd))
5031 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
5033 ASSERT(pvd->vdev_children >= 2);
5036 * If we are detaching the second disk from a replacing vdev, then
5037 * check to see if we changed the original vdev's path to have "/old"
5038 * at the end in spa_vdev_attach(). If so, undo that change now.
5040 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
5041 vd->vdev_path != NULL) {
5042 size_t len = strlen(vd->vdev_path);
5044 for (int c = 0; c < pvd->vdev_children; c++) {
5045 cvd = pvd->vdev_child[c];
5047 if (cvd == vd || cvd->vdev_path == NULL)
5050 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
5051 strcmp(cvd->vdev_path + len, "/old") == 0) {
5052 spa_strfree(cvd->vdev_path);
5053 cvd->vdev_path = spa_strdup(vd->vdev_path);
5060 * If we are detaching the original disk from a spare, then it implies
5061 * that the spare should become a real disk, and be removed from the
5062 * active spare list for the pool.
5064 if (pvd->vdev_ops == &vdev_spare_ops &&
5066 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
5070 * Erase the disk labels so the disk can be used for other things.
5071 * This must be done after all other error cases are handled,
5072 * but before we disembowel vd (so we can still do I/O to it).
5073 * But if we can't do it, don't treat the error as fatal --
5074 * it may be that the unwritability of the disk is the reason
5075 * it's being detached!
5077 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5080 * Remove vd from its parent and compact the parent's children.
5082 vdev_remove_child(pvd, vd);
5083 vdev_compact_children(pvd);
5086 * Remember one of the remaining children so we can get tvd below.
5088 cvd = pvd->vdev_child[pvd->vdev_children - 1];
5091 * If we need to remove the remaining child from the list of hot spares,
5092 * do it now, marking the vdev as no longer a spare in the process.
5093 * We must do this before vdev_remove_parent(), because that can
5094 * change the GUID if it creates a new toplevel GUID. For a similar
5095 * reason, we must remove the spare now, in the same txg as the detach;
5096 * otherwise someone could attach a new sibling, change the GUID, and
5097 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
5100 ASSERT(cvd->vdev_isspare);
5101 spa_spare_remove(cvd);
5102 unspare_guid = cvd->vdev_guid;
5103 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
5104 cvd->vdev_unspare = B_TRUE;
5108 * If the parent mirror/replacing vdev only has one child,
5109 * the parent is no longer needed. Remove it from the tree.
5111 if (pvd->vdev_children == 1) {
5112 if (pvd->vdev_ops == &vdev_spare_ops)
5113 cvd->vdev_unspare = B_FALSE;
5114 vdev_remove_parent(cvd);
5119 * We don't set tvd until now because the parent we just removed
5120 * may have been the previous top-level vdev.
5122 tvd = cvd->vdev_top;
5123 ASSERT(tvd->vdev_parent == rvd);
5126 * Reevaluate the parent vdev state.
5128 vdev_propagate_state(cvd);
5131 * If the 'autoexpand' property is set on the pool then automatically
5132 * try to expand the size of the pool. For example if the device we
5133 * just detached was smaller than the others, it may be possible to
5134 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
5135 * first so that we can obtain the updated sizes of the leaf vdevs.
5137 if (spa->spa_autoexpand) {
5139 vdev_expand(tvd, txg);
5142 vdev_config_dirty(tvd);
5145 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
5146 * vd->vdev_detached is set and free vd's DTL object in syncing context.
5147 * But first make sure we're not on any *other* txg's DTL list, to
5148 * prevent vd from being accessed after it's freed.
5150 vdpath = spa_strdup(vd->vdev_path);
5151 for (int t = 0; t < TXG_SIZE; t++)
5152 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
5153 vd->vdev_detached = B_TRUE;
5154 vdev_dirty(tvd, VDD_DTL, vd, txg);
5156 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
5158 /* hang on to the spa before we release the lock */
5159 spa_open_ref(spa, FTAG);
5161 error = spa_vdev_exit(spa, vd, txg, 0);
5163 spa_history_log_internal(spa, "detach", NULL,
5165 spa_strfree(vdpath);
5168 * If this was the removal of the original device in a hot spare vdev,
5169 * then we want to go through and remove the device from the hot spare
5170 * list of every other pool.
5173 spa_t *altspa = NULL;
5175 mutex_enter(&spa_namespace_lock);
5176 while ((altspa = spa_next(altspa)) != NULL) {
5177 if (altspa->spa_state != POOL_STATE_ACTIVE ||
5181 spa_open_ref(altspa, FTAG);
5182 mutex_exit(&spa_namespace_lock);
5183 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
5184 mutex_enter(&spa_namespace_lock);
5185 spa_close(altspa, FTAG);
5187 mutex_exit(&spa_namespace_lock);
5189 /* search the rest of the vdevs for spares to remove */
5190 spa_vdev_resilver_done(spa);
5193 /* all done with the spa; OK to release */
5194 mutex_enter(&spa_namespace_lock);
5195 spa_close(spa, FTAG);
5196 mutex_exit(&spa_namespace_lock);
5202 * Split a set of devices from their mirrors, and create a new pool from them.
5205 spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
5206 nvlist_t *props, boolean_t exp)
5209 uint64_t txg, *glist;
5211 uint_t c, children, lastlog;
5212 nvlist_t **child, *nvl, *tmp;
5214 char *altroot = NULL;
5215 vdev_t *rvd, **vml = NULL; /* vdev modify list */
5216 boolean_t activate_slog;
5218 ASSERT(spa_writeable(spa));
5220 txg = spa_vdev_enter(spa);
5222 /* clear the log and flush everything up to now */
5223 activate_slog = spa_passivate_log(spa);
5224 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5225 error = spa_offline_log(spa);
5226 txg = spa_vdev_config_enter(spa);
5229 spa_activate_log(spa);
5232 return (spa_vdev_exit(spa, NULL, txg, error));
5234 /* check new spa name before going any further */
5235 if (spa_lookup(newname) != NULL)
5236 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
5239 * scan through all the children to ensure they're all mirrors
5241 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
5242 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
5244 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5246 /* first, check to ensure we've got the right child count */
5247 rvd = spa->spa_root_vdev;
5249 for (c = 0; c < rvd->vdev_children; c++) {
5250 vdev_t *vd = rvd->vdev_child[c];
5252 /* don't count the holes & logs as children */
5253 if (vd->vdev_islog || vd->vdev_ishole) {
5261 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
5262 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5264 /* next, ensure no spare or cache devices are part of the split */
5265 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
5266 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
5267 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5269 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
5270 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
5272 /* then, loop over each vdev and validate it */
5273 for (c = 0; c < children; c++) {
5274 uint64_t is_hole = 0;
5276 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
5280 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
5281 spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
5284 error = SET_ERROR(EINVAL);
5289 /* which disk is going to be split? */
5290 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
5292 error = SET_ERROR(EINVAL);
5296 /* look it up in the spa */
5297 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
5298 if (vml[c] == NULL) {
5299 error = SET_ERROR(ENODEV);
5303 /* make sure there's nothing stopping the split */
5304 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
5305 vml[c]->vdev_islog ||
5306 vml[c]->vdev_ishole ||
5307 vml[c]->vdev_isspare ||
5308 vml[c]->vdev_isl2cache ||
5309 !vdev_writeable(vml[c]) ||
5310 vml[c]->vdev_children != 0 ||
5311 vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
5312 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
5313 error = SET_ERROR(EINVAL);
5317 if (vdev_dtl_required(vml[c])) {
5318 error = SET_ERROR(EBUSY);
5322 /* we need certain info from the top level */
5323 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
5324 vml[c]->vdev_top->vdev_ms_array) == 0);
5325 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
5326 vml[c]->vdev_top->vdev_ms_shift) == 0);
5327 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
5328 vml[c]->vdev_top->vdev_asize) == 0);
5329 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
5330 vml[c]->vdev_top->vdev_ashift) == 0);
5334 kmem_free(vml, children * sizeof (vdev_t *));
5335 kmem_free(glist, children * sizeof (uint64_t));
5336 return (spa_vdev_exit(spa, NULL, txg, error));
5339 /* stop writers from using the disks */
5340 for (c = 0; c < children; c++) {
5342 vml[c]->vdev_offline = B_TRUE;
5344 vdev_reopen(spa->spa_root_vdev);
5347 * Temporarily record the splitting vdevs in the spa config. This
5348 * will disappear once the config is regenerated.
5350 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5351 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
5352 glist, children) == 0);
5353 kmem_free(glist, children * sizeof (uint64_t));
5355 mutex_enter(&spa->spa_props_lock);
5356 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
5358 mutex_exit(&spa->spa_props_lock);
5359 spa->spa_config_splitting = nvl;
5360 vdev_config_dirty(spa->spa_root_vdev);
5362 /* configure and create the new pool */
5363 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
5364 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
5365 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
5366 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5367 spa_version(spa)) == 0);
5368 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
5369 spa->spa_config_txg) == 0);
5370 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
5371 spa_generate_guid(NULL)) == 0);
5372 (void) nvlist_lookup_string(props,
5373 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
5375 /* add the new pool to the namespace */
5376 newspa = spa_add(newname, config, altroot);
5377 newspa->spa_config_txg = spa->spa_config_txg;
5378 spa_set_log_state(newspa, SPA_LOG_CLEAR);
5380 /* release the spa config lock, retaining the namespace lock */
5381 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5383 if (zio_injection_enabled)
5384 zio_handle_panic_injection(spa, FTAG, 1);
5386 spa_activate(newspa, spa_mode_global);
5387 spa_async_suspend(newspa);
5390 /* mark that we are creating new spa by splitting */
5391 newspa->spa_splitting_newspa = B_TRUE;
5393 /* create the new pool from the disks of the original pool */
5394 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
5396 newspa->spa_splitting_newspa = B_FALSE;
5401 /* if that worked, generate a real config for the new pool */
5402 if (newspa->spa_root_vdev != NULL) {
5403 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
5404 NV_UNIQUE_NAME, KM_SLEEP) == 0);
5405 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
5406 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
5407 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
5412 if (props != NULL) {
5413 spa_configfile_set(newspa, props, B_FALSE);
5414 error = spa_prop_set(newspa, props);
5419 /* flush everything */
5420 txg = spa_vdev_config_enter(newspa);
5421 vdev_config_dirty(newspa->spa_root_vdev);
5422 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
5424 if (zio_injection_enabled)
5425 zio_handle_panic_injection(spa, FTAG, 2);
5427 spa_async_resume(newspa);
5429 /* finally, update the original pool's config */
5430 txg = spa_vdev_config_enter(spa);
5431 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
5432 error = dmu_tx_assign(tx, TXG_WAIT);
5435 for (c = 0; c < children; c++) {
5436 if (vml[c] != NULL) {
5439 spa_history_log_internal(spa, "detach", tx,
5440 "vdev=%s", vml[c]->vdev_path);
5444 vdev_config_dirty(spa->spa_root_vdev);
5445 spa->spa_config_splitting = NULL;
5449 (void) spa_vdev_exit(spa, NULL, txg, 0);
5451 if (zio_injection_enabled)
5452 zio_handle_panic_injection(spa, FTAG, 3);
5454 /* split is complete; log a history record */
5455 spa_history_log_internal(newspa, "split", NULL,
5456 "from pool %s", spa_name(spa));
5458 kmem_free(vml, children * sizeof (vdev_t *));
5460 /* if we're not going to mount the filesystems in userland, export */
5462 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
5469 spa_deactivate(newspa);
5472 txg = spa_vdev_config_enter(spa);
5474 /* re-online all offlined disks */
5475 for (c = 0; c < children; c++) {
5477 vml[c]->vdev_offline = B_FALSE;
5479 vdev_reopen(spa->spa_root_vdev);
5481 nvlist_free(spa->spa_config_splitting);
5482 spa->spa_config_splitting = NULL;
5483 (void) spa_vdev_exit(spa, NULL, txg, error);
5485 kmem_free(vml, children * sizeof (vdev_t *));
5490 spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
5492 for (int i = 0; i < count; i++) {
5495 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
5498 if (guid == target_guid)
5506 spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
5507 nvlist_t *dev_to_remove)
5509 nvlist_t **newdev = NULL;
5512 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
5514 for (int i = 0, j = 0; i < count; i++) {
5515 if (dev[i] == dev_to_remove)
5517 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
5520 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
5521 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
5523 for (int i = 0; i < count - 1; i++)
5524 nvlist_free(newdev[i]);
5527 kmem_free(newdev, (count - 1) * sizeof (void *));
5531 * Evacuate the device.
5534 spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
5539 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5540 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5541 ASSERT(vd == vd->vdev_top);
5544 * Evacuate the device. We don't hold the config lock as writer
5545 * since we need to do I/O but we do keep the
5546 * spa_namespace_lock held. Once this completes the device
5547 * should no longer have any blocks allocated on it.
5549 if (vd->vdev_islog) {
5550 if (vd->vdev_stat.vs_alloc != 0)
5551 error = spa_offline_log(spa);
5553 error = SET_ERROR(ENOTSUP);
5560 * The evacuation succeeded. Remove any remaining MOS metadata
5561 * associated with this vdev, and wait for these changes to sync.
5563 ASSERT0(vd->vdev_stat.vs_alloc);
5564 txg = spa_vdev_config_enter(spa);
5565 vd->vdev_removing = B_TRUE;
5566 vdev_dirty_leaves(vd, VDD_DTL, txg);
5567 vdev_config_dirty(vd);
5568 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5574 * Complete the removal by cleaning up the namespace.
5577 spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
5579 vdev_t *rvd = spa->spa_root_vdev;
5580 uint64_t id = vd->vdev_id;
5581 boolean_t last_vdev = (id == (rvd->vdev_children - 1));
5583 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5584 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5585 ASSERT(vd == vd->vdev_top);
5588 * Only remove any devices which are empty.
5590 if (vd->vdev_stat.vs_alloc != 0)
5593 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5595 if (list_link_active(&vd->vdev_state_dirty_node))
5596 vdev_state_clean(vd);
5597 if (list_link_active(&vd->vdev_config_dirty_node))
5598 vdev_config_clean(vd);
5603 vdev_compact_children(rvd);
5605 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
5606 vdev_add_child(rvd, vd);
5608 vdev_config_dirty(rvd);
5611 * Reassess the health of our root vdev.
5617 * Remove a device from the pool -
5619 * Removing a device from the vdev namespace requires several steps
5620 * and can take a significant amount of time. As a result we use
5621 * the spa_vdev_config_[enter/exit] functions which allow us to
5622 * grab and release the spa_config_lock while still holding the namespace
5623 * lock. During each step the configuration is synced out.
5625 * Currently, this supports removing only hot spares, slogs, and level 2 ARC
5629 spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
5632 metaslab_group_t *mg;
5633 nvlist_t **spares, **l2cache, *nv;
5635 uint_t nspares, nl2cache;
5637 boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
5639 ASSERT(spa_writeable(spa));
5642 txg = spa_vdev_enter(spa);
5644 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5646 if (spa->spa_spares.sav_vdevs != NULL &&
5647 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5648 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
5649 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
5651 * Only remove the hot spare if it's not currently in use
5654 if (vd == NULL || unspare) {
5655 spa_vdev_remove_aux(spa->spa_spares.sav_config,
5656 ZPOOL_CONFIG_SPARES, spares, nspares, nv);
5657 spa_load_spares(spa);
5658 spa->spa_spares.sav_sync = B_TRUE;
5660 error = SET_ERROR(EBUSY);
5662 } else if (spa->spa_l2cache.sav_vdevs != NULL &&
5663 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5664 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
5665 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
5667 * Cache devices can always be removed.
5669 spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
5670 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
5671 spa_load_l2cache(spa);
5672 spa->spa_l2cache.sav_sync = B_TRUE;
5673 } else if (vd != NULL && vd->vdev_islog) {
5675 ASSERT(vd == vd->vdev_top);
5680 * Stop allocating from this vdev.
5682 metaslab_group_passivate(mg);
5685 * Wait for the youngest allocations and frees to sync,
5686 * and then wait for the deferral of those frees to finish.
5688 spa_vdev_config_exit(spa, NULL,
5689 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
5692 * Attempt to evacuate the vdev.
5694 error = spa_vdev_remove_evacuate(spa, vd);
5696 txg = spa_vdev_config_enter(spa);
5699 * If we couldn't evacuate the vdev, unwind.
5702 metaslab_group_activate(mg);
5703 return (spa_vdev_exit(spa, NULL, txg, error));
5707 * Clean up the vdev namespace.
5709 spa_vdev_remove_from_namespace(spa, vd);
5711 } else if (vd != NULL) {
5713 * Normal vdevs cannot be removed (yet).
5715 error = SET_ERROR(ENOTSUP);
5718 * There is no vdev of any kind with the specified guid.
5720 error = SET_ERROR(ENOENT);
5724 return (spa_vdev_exit(spa, NULL, txg, error));
5730 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5731 * currently spared, so we can detach it.
5734 spa_vdev_resilver_done_hunt(vdev_t *vd)
5736 vdev_t *newvd, *oldvd;
5738 for (int c = 0; c < vd->vdev_children; c++) {
5739 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
5745 * Check for a completed replacement. We always consider the first
5746 * vdev in the list to be the oldest vdev, and the last one to be
5747 * the newest (see spa_vdev_attach() for how that works). In
5748 * the case where the newest vdev is faulted, we will not automatically
5749 * remove it after a resilver completes. This is OK as it will require
5750 * user intervention to determine which disk the admin wishes to keep.
5752 if (vd->vdev_ops == &vdev_replacing_ops) {
5753 ASSERT(vd->vdev_children > 1);
5755 newvd = vd->vdev_child[vd->vdev_children - 1];
5756 oldvd = vd->vdev_child[0];
5758 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
5759 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5760 !vdev_dtl_required(oldvd))
5765 * Check for a completed resilver with the 'unspare' flag set.
5767 if (vd->vdev_ops == &vdev_spare_ops) {
5768 vdev_t *first = vd->vdev_child[0];
5769 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
5771 if (last->vdev_unspare) {
5774 } else if (first->vdev_unspare) {
5781 if (oldvd != NULL &&
5782 vdev_dtl_empty(newvd, DTL_MISSING) &&
5783 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5784 !vdev_dtl_required(oldvd))
5788 * If there are more than two spares attached to a disk,
5789 * and those spares are not required, then we want to
5790 * attempt to free them up now so that they can be used
5791 * by other pools. Once we're back down to a single
5792 * disk+spare, we stop removing them.
5794 if (vd->vdev_children > 2) {
5795 newvd = vd->vdev_child[1];
5797 if (newvd->vdev_isspare && last->vdev_isspare &&
5798 vdev_dtl_empty(last, DTL_MISSING) &&
5799 vdev_dtl_empty(last, DTL_OUTAGE) &&
5800 !vdev_dtl_required(newvd))
5809 spa_vdev_resilver_done(spa_t *spa)
5811 vdev_t *vd, *pvd, *ppvd;
5812 uint64_t guid, sguid, pguid, ppguid;
5814 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5816 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
5817 pvd = vd->vdev_parent;
5818 ppvd = pvd->vdev_parent;
5819 guid = vd->vdev_guid;
5820 pguid = pvd->vdev_guid;
5821 ppguid = ppvd->vdev_guid;
5824 * If we have just finished replacing a hot spared device, then
5825 * we need to detach the parent's first child (the original hot
5828 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
5829 ppvd->vdev_children == 2) {
5830 ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
5831 sguid = ppvd->vdev_child[1]->vdev_guid;
5833 ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd));
5835 spa_config_exit(spa, SCL_ALL, FTAG);
5836 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
5838 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
5840 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5843 spa_config_exit(spa, SCL_ALL, FTAG);
5847 * Update the stored path or FRU for this vdev.
5850 spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
5854 boolean_t sync = B_FALSE;
5856 ASSERT(spa_writeable(spa));
5858 spa_vdev_state_enter(spa, SCL_ALL);
5860 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
5861 return (spa_vdev_state_exit(spa, NULL, ENOENT));
5863 if (!vd->vdev_ops->vdev_op_leaf)
5864 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
5867 if (strcmp(value, vd->vdev_path) != 0) {
5868 spa_strfree(vd->vdev_path);
5869 vd->vdev_path = spa_strdup(value);
5873 if (vd->vdev_fru == NULL) {
5874 vd->vdev_fru = spa_strdup(value);
5876 } else if (strcmp(value, vd->vdev_fru) != 0) {
5877 spa_strfree(vd->vdev_fru);
5878 vd->vdev_fru = spa_strdup(value);
5883 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
5887 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
5889 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5893 spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5895 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5899 * ==========================================================================
5901 * ==========================================================================
5905 spa_scan_stop(spa_t *spa)
5907 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5908 if (dsl_scan_resilvering(spa->spa_dsl_pool))
5909 return (SET_ERROR(EBUSY));
5910 return (dsl_scan_cancel(spa->spa_dsl_pool));
5914 spa_scan(spa_t *spa, pool_scan_func_t func)
5916 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5918 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5919 return (SET_ERROR(ENOTSUP));
5922 * If a resilver was requested, but there is no DTL on a
5923 * writeable leaf device, we have nothing to do.
5925 if (func == POOL_SCAN_RESILVER &&
5926 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5927 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5931 return (dsl_scan(spa->spa_dsl_pool, func));
5935 * ==========================================================================
5936 * SPA async task processing
5937 * ==========================================================================
5941 spa_async_remove(spa_t *spa, vdev_t *vd)
5943 if (vd->vdev_remove_wanted) {
5944 vd->vdev_remove_wanted = B_FALSE;
5945 vd->vdev_delayed_close = B_FALSE;
5946 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5949 * We want to clear the stats, but we don't want to do a full
5950 * vdev_clear() as that will cause us to throw away
5951 * degraded/faulted state as well as attempt to reopen the
5952 * device, all of which is a waste.
5954 vd->vdev_stat.vs_read_errors = 0;
5955 vd->vdev_stat.vs_write_errors = 0;
5956 vd->vdev_stat.vs_checksum_errors = 0;
5958 vdev_state_dirty(vd->vdev_top);
5959 /* Tell userspace that the vdev is gone. */
5960 zfs_post_remove(spa, vd);
5963 for (int c = 0; c < vd->vdev_children; c++)
5964 spa_async_remove(spa, vd->vdev_child[c]);
5968 spa_async_probe(spa_t *spa, vdev_t *vd)
5970 if (vd->vdev_probe_wanted) {
5971 vd->vdev_probe_wanted = B_FALSE;
5972 vdev_reopen(vd); /* vdev_open() does the actual probe */
5975 for (int c = 0; c < vd->vdev_children; c++)
5976 spa_async_probe(spa, vd->vdev_child[c]);
5980 spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5986 if (!spa->spa_autoexpand)
5989 for (int c = 0; c < vd->vdev_children; c++) {
5990 vdev_t *cvd = vd->vdev_child[c];
5991 spa_async_autoexpand(spa, cvd);
5994 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5997 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
5998 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
6000 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
6001 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
6003 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
6004 ESC_ZFS_VDEV_AUTOEXPAND, attr, &eid, DDI_SLEEP);
6007 kmem_free(physpath, MAXPATHLEN);
6011 spa_async_thread(void *arg)
6016 ASSERT(spa->spa_sync_on);
6018 mutex_enter(&spa->spa_async_lock);
6019 tasks = spa->spa_async_tasks;
6020 spa->spa_async_tasks &= SPA_ASYNC_REMOVE;
6021 mutex_exit(&spa->spa_async_lock);
6024 * See if the config needs to be updated.
6026 if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
6027 uint64_t old_space, new_space;
6029 mutex_enter(&spa_namespace_lock);
6030 old_space = metaslab_class_get_space(spa_normal_class(spa));
6031 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
6032 new_space = metaslab_class_get_space(spa_normal_class(spa));
6033 mutex_exit(&spa_namespace_lock);
6036 * If the pool grew as a result of the config update,
6037 * then log an internal history event.
6039 if (new_space != old_space) {
6040 spa_history_log_internal(spa, "vdev online", NULL,
6041 "pool '%s' size: %llu(+%llu)",
6042 spa_name(spa), new_space, new_space - old_space);
6046 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
6047 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
6048 spa_async_autoexpand(spa, spa->spa_root_vdev);
6049 spa_config_exit(spa, SCL_CONFIG, FTAG);
6053 * See if any devices need to be probed.
6055 if (tasks & SPA_ASYNC_PROBE) {
6056 spa_vdev_state_enter(spa, SCL_NONE);
6057 spa_async_probe(spa, spa->spa_root_vdev);
6058 (void) spa_vdev_state_exit(spa, NULL, 0);
6062 * If any devices are done replacing, detach them.
6064 if (tasks & SPA_ASYNC_RESILVER_DONE)
6065 spa_vdev_resilver_done(spa);
6068 * Kick off a resilver.
6070 if (tasks & SPA_ASYNC_RESILVER)
6071 dsl_resilver_restart(spa->spa_dsl_pool, 0);
6074 * Let the world know that we're done.
6076 mutex_enter(&spa->spa_async_lock);
6077 spa->spa_async_thread = NULL;
6078 cv_broadcast(&spa->spa_async_cv);
6079 mutex_exit(&spa->spa_async_lock);
6084 spa_async_thread_vd(void *arg)
6089 ASSERT(spa->spa_sync_on);
6091 mutex_enter(&spa->spa_async_lock);
6092 tasks = spa->spa_async_tasks;
6094 spa->spa_async_tasks &= ~SPA_ASYNC_REMOVE;
6095 mutex_exit(&spa->spa_async_lock);
6098 * See if any devices need to be marked REMOVED.
6100 if (tasks & SPA_ASYNC_REMOVE) {
6101 spa_vdev_state_enter(spa, SCL_NONE);
6102 spa_async_remove(spa, spa->spa_root_vdev);
6103 for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
6104 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
6105 for (int i = 0; i < spa->spa_spares.sav_count; i++)
6106 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
6107 (void) spa_vdev_state_exit(spa, NULL, 0);
6111 * Let the world know that we're done.
6113 mutex_enter(&spa->spa_async_lock);
6114 tasks = spa->spa_async_tasks;
6115 if ((tasks & SPA_ASYNC_REMOVE) != 0)
6117 spa->spa_async_thread_vd = NULL;
6118 cv_broadcast(&spa->spa_async_cv);
6119 mutex_exit(&spa->spa_async_lock);
6124 spa_async_suspend(spa_t *spa)
6126 mutex_enter(&spa->spa_async_lock);
6127 spa->spa_async_suspended++;
6128 while (spa->spa_async_thread != NULL &&
6129 spa->spa_async_thread_vd != NULL)
6130 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
6131 mutex_exit(&spa->spa_async_lock);
6135 spa_async_resume(spa_t *spa)
6137 mutex_enter(&spa->spa_async_lock);
6138 ASSERT(spa->spa_async_suspended != 0);
6139 spa->spa_async_suspended--;
6140 mutex_exit(&spa->spa_async_lock);
6144 spa_async_tasks_pending(spa_t *spa)
6146 uint_t non_config_tasks;
6148 boolean_t config_task_suspended;
6150 non_config_tasks = spa->spa_async_tasks & ~(SPA_ASYNC_CONFIG_UPDATE |
6152 config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE;
6153 if (spa->spa_ccw_fail_time == 0) {
6154 config_task_suspended = B_FALSE;
6156 config_task_suspended =
6157 (gethrtime() - spa->spa_ccw_fail_time) <
6158 (zfs_ccw_retry_interval * NANOSEC);
6161 return (non_config_tasks || (config_task && !config_task_suspended));
6165 spa_async_dispatch(spa_t *spa)
6167 mutex_enter(&spa->spa_async_lock);
6168 if (spa_async_tasks_pending(spa) &&
6169 !spa->spa_async_suspended &&
6170 spa->spa_async_thread == NULL &&
6172 spa->spa_async_thread = thread_create(NULL, 0,
6173 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
6174 mutex_exit(&spa->spa_async_lock);
6178 spa_async_dispatch_vd(spa_t *spa)
6180 mutex_enter(&spa->spa_async_lock);
6181 if ((spa->spa_async_tasks & SPA_ASYNC_REMOVE) != 0 &&
6182 !spa->spa_async_suspended &&
6183 spa->spa_async_thread_vd == NULL &&
6185 spa->spa_async_thread_vd = thread_create(NULL, 0,
6186 spa_async_thread_vd, spa, 0, &p0, TS_RUN, maxclsyspri);
6187 mutex_exit(&spa->spa_async_lock);
6191 spa_async_request(spa_t *spa, int task)
6193 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
6194 mutex_enter(&spa->spa_async_lock);
6195 spa->spa_async_tasks |= task;
6196 mutex_exit(&spa->spa_async_lock);
6197 spa_async_dispatch_vd(spa);
6201 * ==========================================================================
6202 * SPA syncing routines
6203 * ==========================================================================
6207 bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
6210 bpobj_enqueue(bpo, bp, tx);
6215 spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
6219 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
6220 BP_GET_PSIZE(bp), zio->io_flags));
6225 * Note: this simple function is not inlined to make it easier to dtrace the
6226 * amount of time spent syncing frees.
6229 spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx)
6231 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6232 bplist_iterate(bpl, spa_free_sync_cb, zio, tx);
6233 VERIFY(zio_wait(zio) == 0);
6237 * Note: this simple function is not inlined to make it easier to dtrace the
6238 * amount of time spent syncing deferred frees.
6241 spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx)
6243 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6244 VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj,
6245 spa_free_sync_cb, zio, tx), ==, 0);
6246 VERIFY0(zio_wait(zio));
6251 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
6253 char *packed = NULL;
6258 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
6261 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
6262 * information. This avoids the dmu_buf_will_dirty() path and
6263 * saves us a pre-read to get data we don't actually care about.
6265 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
6266 packed = kmem_alloc(bufsize, KM_SLEEP);
6268 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
6270 bzero(packed + nvsize, bufsize - nvsize);
6272 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
6274 kmem_free(packed, bufsize);
6276 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
6277 dmu_buf_will_dirty(db, tx);
6278 *(uint64_t *)db->db_data = nvsize;
6279 dmu_buf_rele(db, FTAG);
6283 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
6284 const char *config, const char *entry)
6294 * Update the MOS nvlist describing the list of available devices.
6295 * spa_validate_aux() will have already made sure this nvlist is
6296 * valid and the vdevs are labeled appropriately.
6298 if (sav->sav_object == 0) {
6299 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
6300 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
6301 sizeof (uint64_t), tx);
6302 VERIFY(zap_update(spa->spa_meta_objset,
6303 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
6304 &sav->sav_object, tx) == 0);
6307 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
6308 if (sav->sav_count == 0) {
6309 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
6311 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
6312 for (i = 0; i < sav->sav_count; i++)
6313 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
6314 B_FALSE, VDEV_CONFIG_L2CACHE);
6315 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
6316 sav->sav_count) == 0);
6317 for (i = 0; i < sav->sav_count; i++)
6318 nvlist_free(list[i]);
6319 kmem_free(list, sav->sav_count * sizeof (void *));
6322 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
6323 nvlist_free(nvroot);
6325 sav->sav_sync = B_FALSE;
6329 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
6333 if (list_is_empty(&spa->spa_config_dirty_list))
6336 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6338 config = spa_config_generate(spa, spa->spa_root_vdev,
6339 dmu_tx_get_txg(tx), B_FALSE);
6342 * If we're upgrading the spa version then make sure that
6343 * the config object gets updated with the correct version.
6345 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
6346 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
6347 spa->spa_uberblock.ub_version);
6349 spa_config_exit(spa, SCL_STATE, FTAG);
6351 nvlist_free(spa->spa_config_syncing);
6352 spa->spa_config_syncing = config;
6354 spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
6358 spa_sync_version(void *arg, dmu_tx_t *tx)
6360 uint64_t *versionp = arg;
6361 uint64_t version = *versionp;
6362 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
6365 * Setting the version is special cased when first creating the pool.
6367 ASSERT(tx->tx_txg != TXG_INITIAL);
6369 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
6370 ASSERT(version >= spa_version(spa));
6372 spa->spa_uberblock.ub_version = version;
6373 vdev_config_dirty(spa->spa_root_vdev);
6374 spa_history_log_internal(spa, "set", tx, "version=%lld", version);
6378 * Set zpool properties.
6381 spa_sync_props(void *arg, dmu_tx_t *tx)
6383 nvlist_t *nvp = arg;
6384 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
6385 objset_t *mos = spa->spa_meta_objset;
6386 nvpair_t *elem = NULL;
6388 mutex_enter(&spa->spa_props_lock);
6390 while ((elem = nvlist_next_nvpair(nvp, elem))) {
6392 char *strval, *fname;
6394 const char *propname;
6395 zprop_type_t proptype;
6398 switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
6401 * We checked this earlier in spa_prop_validate().
6403 ASSERT(zpool_prop_feature(nvpair_name(elem)));
6405 fname = strchr(nvpair_name(elem), '@') + 1;
6406 VERIFY0(zfeature_lookup_name(fname, &fid));
6408 spa_feature_enable(spa, fid, tx);
6409 spa_history_log_internal(spa, "set", tx,
6410 "%s=enabled", nvpair_name(elem));
6413 case ZPOOL_PROP_VERSION:
6414 intval = fnvpair_value_uint64(elem);
6416 * The version is synced seperatly before other
6417 * properties and should be correct by now.
6419 ASSERT3U(spa_version(spa), >=, intval);
6422 case ZPOOL_PROP_ALTROOT:
6424 * 'altroot' is a non-persistent property. It should
6425 * have been set temporarily at creation or import time.
6427 ASSERT(spa->spa_root != NULL);
6430 case ZPOOL_PROP_READONLY:
6431 case ZPOOL_PROP_CACHEFILE:
6433 * 'readonly' and 'cachefile' are also non-persisitent
6437 case ZPOOL_PROP_COMMENT:
6438 strval = fnvpair_value_string(elem);
6439 if (spa->spa_comment != NULL)
6440 spa_strfree(spa->spa_comment);
6441 spa->spa_comment = spa_strdup(strval);
6443 * We need to dirty the configuration on all the vdevs
6444 * so that their labels get updated. It's unnecessary
6445 * to do this for pool creation since the vdev's
6446 * configuratoin has already been dirtied.
6448 if (tx->tx_txg != TXG_INITIAL)
6449 vdev_config_dirty(spa->spa_root_vdev);
6450 spa_history_log_internal(spa, "set", tx,
6451 "%s=%s", nvpair_name(elem), strval);
6455 * Set pool property values in the poolprops mos object.
6457 if (spa->spa_pool_props_object == 0) {
6458 spa->spa_pool_props_object =
6459 zap_create_link(mos, DMU_OT_POOL_PROPS,
6460 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
6464 /* normalize the property name */
6465 propname = zpool_prop_to_name(prop);
6466 proptype = zpool_prop_get_type(prop);
6468 if (nvpair_type(elem) == DATA_TYPE_STRING) {
6469 ASSERT(proptype == PROP_TYPE_STRING);
6470 strval = fnvpair_value_string(elem);
6471 VERIFY0(zap_update(mos,
6472 spa->spa_pool_props_object, propname,
6473 1, strlen(strval) + 1, strval, tx));
6474 spa_history_log_internal(spa, "set", tx,
6475 "%s=%s", nvpair_name(elem), strval);
6476 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
6477 intval = fnvpair_value_uint64(elem);
6479 if (proptype == PROP_TYPE_INDEX) {
6481 VERIFY0(zpool_prop_index_to_string(
6482 prop, intval, &unused));
6484 VERIFY0(zap_update(mos,
6485 spa->spa_pool_props_object, propname,
6486 8, 1, &intval, tx));
6487 spa_history_log_internal(spa, "set", tx,
6488 "%s=%lld", nvpair_name(elem), intval);
6490 ASSERT(0); /* not allowed */
6494 case ZPOOL_PROP_DELEGATION:
6495 spa->spa_delegation = intval;
6497 case ZPOOL_PROP_BOOTFS:
6498 spa->spa_bootfs = intval;
6500 case ZPOOL_PROP_FAILUREMODE:
6501 spa->spa_failmode = intval;
6503 case ZPOOL_PROP_AUTOEXPAND:
6504 spa->spa_autoexpand = intval;
6505 if (tx->tx_txg != TXG_INITIAL)
6506 spa_async_request(spa,
6507 SPA_ASYNC_AUTOEXPAND);
6509 case ZPOOL_PROP_DEDUPDITTO:
6510 spa->spa_dedup_ditto = intval;
6519 mutex_exit(&spa->spa_props_lock);
6523 * Perform one-time upgrade on-disk changes. spa_version() does not
6524 * reflect the new version this txg, so there must be no changes this
6525 * txg to anything that the upgrade code depends on after it executes.
6526 * Therefore this must be called after dsl_pool_sync() does the sync
6530 spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
6532 dsl_pool_t *dp = spa->spa_dsl_pool;
6534 ASSERT(spa->spa_sync_pass == 1);
6536 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
6538 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
6539 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
6540 dsl_pool_create_origin(dp, tx);
6542 /* Keeping the origin open increases spa_minref */
6543 spa->spa_minref += 3;
6546 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
6547 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
6548 dsl_pool_upgrade_clones(dp, tx);
6551 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
6552 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
6553 dsl_pool_upgrade_dir_clones(dp, tx);
6555 /* Keeping the freedir open increases spa_minref */
6556 spa->spa_minref += 3;
6559 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES &&
6560 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
6561 spa_feature_create_zap_objects(spa, tx);
6565 * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable
6566 * when possibility to use lz4 compression for metadata was added
6567 * Old pools that have this feature enabled must be upgraded to have
6568 * this feature active
6570 if (spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
6571 boolean_t lz4_en = spa_feature_is_enabled(spa,
6572 SPA_FEATURE_LZ4_COMPRESS);
6573 boolean_t lz4_ac = spa_feature_is_active(spa,
6574 SPA_FEATURE_LZ4_COMPRESS);
6576 if (lz4_en && !lz4_ac)
6577 spa_feature_incr(spa, SPA_FEATURE_LZ4_COMPRESS, tx);
6581 * If we haven't written the salt, do so now. Note that the
6582 * feature may not be activated yet, but that's fine since
6583 * the presence of this ZAP entry is backwards compatible.
6585 if (zap_contains(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
6586 DMU_POOL_CHECKSUM_SALT) == ENOENT) {
6587 VERIFY0(zap_add(spa->spa_meta_objset,
6588 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CHECKSUM_SALT, 1,
6589 sizeof (spa->spa_cksum_salt.zcs_bytes),
6590 spa->spa_cksum_salt.zcs_bytes, tx));
6593 rrw_exit(&dp->dp_config_rwlock, FTAG);
6597 * Sync the specified transaction group. New blocks may be dirtied as
6598 * part of the process, so we iterate until it converges.
6601 spa_sync(spa_t *spa, uint64_t txg)
6603 dsl_pool_t *dp = spa->spa_dsl_pool;
6604 objset_t *mos = spa->spa_meta_objset;
6605 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
6606 vdev_t *rvd = spa->spa_root_vdev;
6611 VERIFY(spa_writeable(spa));
6614 * Lock out configuration changes.
6616 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
6618 spa->spa_syncing_txg = txg;
6619 spa->spa_sync_pass = 0;
6622 * If there are any pending vdev state changes, convert them
6623 * into config changes that go out with this transaction group.
6625 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6626 while (list_head(&spa->spa_state_dirty_list) != NULL) {
6628 * We need the write lock here because, for aux vdevs,
6629 * calling vdev_config_dirty() modifies sav_config.
6630 * This is ugly and will become unnecessary when we
6631 * eliminate the aux vdev wart by integrating all vdevs
6632 * into the root vdev tree.
6634 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6635 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
6636 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
6637 vdev_state_clean(vd);
6638 vdev_config_dirty(vd);
6640 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6641 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
6643 spa_config_exit(spa, SCL_STATE, FTAG);
6645 tx = dmu_tx_create_assigned(dp, txg);
6647 spa->spa_sync_starttime = gethrtime();
6649 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid,
6650 spa->spa_sync_starttime + spa->spa_deadman_synctime));
6651 #else /* !illumos */
6653 callout_reset(&spa->spa_deadman_cycid,
6654 hz * spa->spa_deadman_synctime / NANOSEC, spa_deadman, spa);
6656 #endif /* illumos */
6659 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
6660 * set spa_deflate if we have no raid-z vdevs.
6662 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
6663 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
6666 for (i = 0; i < rvd->vdev_children; i++) {
6667 vd = rvd->vdev_child[i];
6668 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
6671 if (i == rvd->vdev_children) {
6672 spa->spa_deflate = TRUE;
6673 VERIFY(0 == zap_add(spa->spa_meta_objset,
6674 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
6675 sizeof (uint64_t), 1, &spa->spa_deflate, tx));
6680 * Iterate to convergence.
6683 int pass = ++spa->spa_sync_pass;
6685 spa_sync_config_object(spa, tx);
6686 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
6687 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
6688 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
6689 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
6690 spa_errlog_sync(spa, txg);
6691 dsl_pool_sync(dp, txg);
6693 if (pass < zfs_sync_pass_deferred_free) {
6694 spa_sync_frees(spa, free_bpl, tx);
6697 * We can not defer frees in pass 1, because
6698 * we sync the deferred frees later in pass 1.
6700 ASSERT3U(pass, >, 1);
6701 bplist_iterate(free_bpl, bpobj_enqueue_cb,
6702 &spa->spa_deferred_bpobj, tx);
6706 dsl_scan_sync(dp, tx);
6708 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
6712 spa_sync_upgrades(spa, tx);
6714 spa->spa_uberblock.ub_rootbp.blk_birth);
6716 * Note: We need to check if the MOS is dirty
6717 * because we could have marked the MOS dirty
6718 * without updating the uberblock (e.g. if we
6719 * have sync tasks but no dirty user data). We
6720 * need to check the uberblock's rootbp because
6721 * it is updated if we have synced out dirty
6722 * data (though in this case the MOS will most
6723 * likely also be dirty due to second order
6724 * effects, we don't want to rely on that here).
6726 if (spa->spa_uberblock.ub_rootbp.blk_birth < txg &&
6727 !dmu_objset_is_dirty(mos, txg)) {
6729 * Nothing changed on the first pass,
6730 * therefore this TXG is a no-op. Avoid
6731 * syncing deferred frees, so that we
6732 * can keep this TXG as a no-op.
6734 ASSERT(txg_list_empty(&dp->dp_dirty_datasets,
6736 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
6737 ASSERT(txg_list_empty(&dp->dp_sync_tasks, txg));
6740 spa_sync_deferred_frees(spa, tx);
6743 } while (dmu_objset_is_dirty(mos, txg));
6746 * Rewrite the vdev configuration (which includes the uberblock)
6747 * to commit the transaction group.
6749 * If there are no dirty vdevs, we sync the uberblock to a few
6750 * random top-level vdevs that are known to be visible in the
6751 * config cache (see spa_vdev_add() for a complete description).
6752 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
6756 * We hold SCL_STATE to prevent vdev open/close/etc.
6757 * while we're attempting to write the vdev labels.
6759 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6761 if (list_is_empty(&spa->spa_config_dirty_list)) {
6762 vdev_t *svd[SPA_DVAS_PER_BP];
6764 int children = rvd->vdev_children;
6765 int c0 = spa_get_random(children);
6767 for (int c = 0; c < children; c++) {
6768 vd = rvd->vdev_child[(c0 + c) % children];
6769 if (vd->vdev_ms_array == 0 || vd->vdev_islog)
6771 svd[svdcount++] = vd;
6772 if (svdcount == SPA_DVAS_PER_BP)
6775 error = vdev_config_sync(svd, svdcount, txg);
6777 error = vdev_config_sync(rvd->vdev_child,
6778 rvd->vdev_children, txg);
6782 spa->spa_last_synced_guid = rvd->vdev_guid;
6784 spa_config_exit(spa, SCL_STATE, FTAG);
6788 zio_suspend(spa, NULL);
6789 zio_resume_wait(spa);
6794 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY));
6795 #else /* !illumos */
6797 callout_drain(&spa->spa_deadman_cycid);
6799 #endif /* illumos */
6802 * Clear the dirty config list.
6804 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
6805 vdev_config_clean(vd);
6808 * Now that the new config has synced transactionally,
6809 * let it become visible to the config cache.
6811 if (spa->spa_config_syncing != NULL) {
6812 spa_config_set(spa, spa->spa_config_syncing);
6813 spa->spa_config_txg = txg;
6814 spa->spa_config_syncing = NULL;
6817 spa->spa_ubsync = spa->spa_uberblock;
6819 dsl_pool_sync_done(dp, txg);
6822 * Update usable space statistics.
6824 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
6825 vdev_sync_done(vd, txg);
6827 spa_update_dspace(spa);
6830 * It had better be the case that we didn't dirty anything
6831 * since vdev_config_sync().
6833 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
6834 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
6835 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
6837 spa->spa_sync_pass = 0;
6839 spa_config_exit(spa, SCL_CONFIG, FTAG);
6841 spa_handle_ignored_writes(spa);
6844 * If any async tasks have been requested, kick them off.
6846 spa_async_dispatch(spa);
6847 spa_async_dispatch_vd(spa);
6851 * Sync all pools. We don't want to hold the namespace lock across these
6852 * operations, so we take a reference on the spa_t and drop the lock during the
6856 spa_sync_allpools(void)
6859 mutex_enter(&spa_namespace_lock);
6860 while ((spa = spa_next(spa)) != NULL) {
6861 if (spa_state(spa) != POOL_STATE_ACTIVE ||
6862 !spa_writeable(spa) || spa_suspended(spa))
6864 spa_open_ref(spa, FTAG);
6865 mutex_exit(&spa_namespace_lock);
6866 txg_wait_synced(spa_get_dsl(spa), 0);
6867 mutex_enter(&spa_namespace_lock);
6868 spa_close(spa, FTAG);
6870 mutex_exit(&spa_namespace_lock);
6874 * ==========================================================================
6875 * Miscellaneous routines
6876 * ==========================================================================
6880 * Remove all pools in the system.
6888 * Remove all cached state. All pools should be closed now,
6889 * so every spa in the AVL tree should be unreferenced.
6891 mutex_enter(&spa_namespace_lock);
6892 while ((spa = spa_next(NULL)) != NULL) {
6894 * Stop async tasks. The async thread may need to detach
6895 * a device that's been replaced, which requires grabbing
6896 * spa_namespace_lock, so we must drop it here.
6898 spa_open_ref(spa, FTAG);
6899 mutex_exit(&spa_namespace_lock);
6900 spa_async_suspend(spa);
6901 mutex_enter(&spa_namespace_lock);
6902 spa_close(spa, FTAG);
6904 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
6906 spa_deactivate(spa);
6910 mutex_exit(&spa_namespace_lock);
6914 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
6919 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
6923 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
6924 vd = spa->spa_l2cache.sav_vdevs[i];
6925 if (vd->vdev_guid == guid)
6929 for (i = 0; i < spa->spa_spares.sav_count; i++) {
6930 vd = spa->spa_spares.sav_vdevs[i];
6931 if (vd->vdev_guid == guid)
6940 spa_upgrade(spa_t *spa, uint64_t version)
6942 ASSERT(spa_writeable(spa));
6944 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6947 * This should only be called for a non-faulted pool, and since a
6948 * future version would result in an unopenable pool, this shouldn't be
6951 ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version));
6952 ASSERT3U(version, >=, spa->spa_uberblock.ub_version);
6954 spa->spa_uberblock.ub_version = version;
6955 vdev_config_dirty(spa->spa_root_vdev);
6957 spa_config_exit(spa, SCL_ALL, FTAG);
6959 txg_wait_synced(spa_get_dsl(spa), 0);
6963 spa_has_spare(spa_t *spa, uint64_t guid)
6967 spa_aux_vdev_t *sav = &spa->spa_spares;
6969 for (i = 0; i < sav->sav_count; i++)
6970 if (sav->sav_vdevs[i]->vdev_guid == guid)
6973 for (i = 0; i < sav->sav_npending; i++) {
6974 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
6975 &spareguid) == 0 && spareguid == guid)
6983 * Check if a pool has an active shared spare device.
6984 * Note: reference count of an active spare is 2, as a spare and as a replace
6987 spa_has_active_shared_spare(spa_t *spa)
6991 spa_aux_vdev_t *sav = &spa->spa_spares;
6993 for (i = 0; i < sav->sav_count; i++) {
6994 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
6995 &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
7004 * Post a sysevent corresponding to the given event. The 'name' must be one of
7005 * the event definitions in sys/sysevent/eventdefs.h. The payload will be
7006 * filled in from the spa and (optionally) the vdev. This doesn't do anything
7007 * in the userland libzpool, as we don't want consumers to misinterpret ztest
7008 * or zdb as real changes.
7011 spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
7015 sysevent_attr_list_t *attr = NULL;
7016 sysevent_value_t value;
7019 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
7022 value.value_type = SE_DATA_TYPE_STRING;
7023 value.value.sv_string = spa_name(spa);
7024 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
7027 value.value_type = SE_DATA_TYPE_UINT64;
7028 value.value.sv_uint64 = spa_guid(spa);
7029 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
7033 value.value_type = SE_DATA_TYPE_UINT64;
7034 value.value.sv_uint64 = vd->vdev_guid;
7035 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
7039 if (vd->vdev_path) {
7040 value.value_type = SE_DATA_TYPE_STRING;
7041 value.value.sv_string = vd->vdev_path;
7042 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
7043 &value, SE_SLEEP) != 0)
7048 if (sysevent_attach_attributes(ev, attr) != 0)
7052 (void) log_sysevent(ev, SE_SLEEP, &eid);
7056 sysevent_free_attr(attr);