4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2013 by Delphix. All rights reserved.
25 * Copyright (c) 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.
31 * SPA: Storage Pool Allocator
33 * This file contains all the routines used when modifying on-disk SPA state.
34 * This includes opening, importing, destroying, exporting a pool, and syncing a
38 #include <sys/zfs_context.h>
39 #include <sys/fm/fs/zfs.h>
40 #include <sys/spa_impl.h>
42 #include <sys/zio_checksum.h>
44 #include <sys/dmu_tx.h>
48 #include <sys/vdev_impl.h>
49 #include <sys/metaslab.h>
50 #include <sys/metaslab_impl.h>
51 #include <sys/uberblock_impl.h>
54 #include <sys/dmu_traverse.h>
55 #include <sys/dmu_objset.h>
56 #include <sys/unique.h>
57 #include <sys/dsl_pool.h>
58 #include <sys/dsl_dataset.h>
59 #include <sys/dsl_dir.h>
60 #include <sys/dsl_prop.h>
61 #include <sys/dsl_synctask.h>
62 #include <sys/fs/zfs.h>
64 #include <sys/callb.h>
65 #include <sys/spa_boot.h>
66 #include <sys/zfs_ioctl.h>
67 #include <sys/dsl_scan.h>
68 #include <sys/dmu_send.h>
69 #include <sys/dsl_destroy.h>
70 #include <sys/dsl_userhold.h>
71 #include <sys/zfeature.h>
73 #include <sys/trim_map.h>
76 #include <sys/callb.h>
77 #include <sys/cpupart.h>
82 #include "zfs_comutil.h"
84 /* Check hostid on import? */
85 static int check_hostid = 1;
87 SYSCTL_DECL(_vfs_zfs);
88 TUNABLE_INT("vfs.zfs.check_hostid", &check_hostid);
89 SYSCTL_INT(_vfs_zfs, OID_AUTO, check_hostid, CTLFLAG_RW, &check_hostid, 0,
90 "Check hostid on import?");
93 * The interval, in seconds, at which failed configuration cache file writes
96 static int zfs_ccw_retry_interval = 300;
98 typedef enum zti_modes {
99 ZTI_MODE_FIXED, /* value is # of threads (min 1) */
100 ZTI_MODE_BATCH, /* cpu-intensive; value is ignored */
101 ZTI_MODE_NULL, /* don't create a taskq */
105 #define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) }
106 #define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 }
107 #define ZTI_NULL { ZTI_MODE_NULL, 0, 0 }
109 #define ZTI_N(n) ZTI_P(n, 1)
110 #define ZTI_ONE ZTI_N(1)
112 typedef struct zio_taskq_info {
113 zti_modes_t zti_mode;
118 static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
119 "issue", "issue_high", "intr", "intr_high"
123 * This table defines the taskq settings for each ZFS I/O type. When
124 * initializing a pool, we use this table to create an appropriately sized
125 * taskq. Some operations are low volume and therefore have a small, static
126 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE
127 * macros. Other operations process a large amount of data; the ZTI_BATCH
128 * macro causes us to create a taskq oriented for throughput. Some operations
129 * are so high frequency and short-lived that the taskq itself can become a a
130 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an
131 * additional degree of parallelism specified by the number of threads per-
132 * taskq and the number of taskqs; when dispatching an event in this case, the
133 * particular taskq is chosen at random.
135 * The different taskq priorities are to handle the different contexts (issue
136 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that
137 * need to be handled with minimum delay.
139 const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
140 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
141 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* NULL */
142 { ZTI_N(8), ZTI_NULL, ZTI_BATCH, ZTI_NULL }, /* READ */
143 { ZTI_BATCH, ZTI_N(5), ZTI_N(8), ZTI_N(5) }, /* WRITE */
144 { ZTI_P(12, 8), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FREE */
145 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* CLAIM */
146 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* IOCTL */
149 static void spa_sync_version(void *arg, dmu_tx_t *tx);
150 static void spa_sync_props(void *arg, dmu_tx_t *tx);
151 static boolean_t spa_has_active_shared_spare(spa_t *spa);
152 static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
153 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
155 static void spa_vdev_resilver_done(spa_t *spa);
157 uint_t zio_taskq_batch_pct = 75; /* 1 thread per cpu in pset */
159 id_t zio_taskq_psrset_bind = PS_NONE;
162 boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */
164 uint_t zio_taskq_basedc = 80; /* base duty cycle */
166 boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */
167 extern int zfs_sync_pass_deferred_free;
170 extern void spa_deadman(void *arg);
174 * This (illegal) pool name is used when temporarily importing a spa_t in order
175 * to get the vdev stats associated with the imported devices.
177 #define TRYIMPORT_NAME "$import"
180 * ==========================================================================
181 * SPA properties routines
182 * ==========================================================================
186 * Add a (source=src, propname=propval) list to an nvlist.
189 spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
190 uint64_t intval, zprop_source_t src)
192 const char *propname = zpool_prop_to_name(prop);
195 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
196 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
199 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
201 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
203 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
204 nvlist_free(propval);
208 * Get property values from the spa configuration.
211 spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
213 vdev_t *rvd = spa->spa_root_vdev;
214 dsl_pool_t *pool = spa->spa_dsl_pool;
215 uint64_t size, alloc, cap, version;
216 zprop_source_t src = ZPROP_SRC_NONE;
217 spa_config_dirent_t *dp;
218 metaslab_class_t *mc = spa_normal_class(spa);
220 ASSERT(MUTEX_HELD(&spa->spa_props_lock));
223 alloc = metaslab_class_get_alloc(spa_normal_class(spa));
224 size = metaslab_class_get_space(spa_normal_class(spa));
225 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
226 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
227 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
228 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
231 spa_prop_add_list(*nvp, ZPOOL_PROP_FRAGMENTATION, NULL,
232 metaslab_class_fragmentation(mc), src);
233 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL,
234 metaslab_class_expandable_space(mc), src);
235 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
236 (spa_mode(spa) == FREAD), src);
238 cap = (size == 0) ? 0 : (alloc * 100 / size);
239 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
241 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
242 ddt_get_pool_dedup_ratio(spa), src);
244 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
245 rvd->vdev_state, src);
247 version = spa_version(spa);
248 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
249 src = ZPROP_SRC_DEFAULT;
251 src = ZPROP_SRC_LOCAL;
252 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
257 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
258 * when opening pools before this version freedir will be NULL.
260 if (pool->dp_free_dir != NULL) {
261 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
262 dsl_dir_phys(pool->dp_free_dir)->dd_used_bytes,
265 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
269 if (pool->dp_leak_dir != NULL) {
270 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, NULL,
271 dsl_dir_phys(pool->dp_leak_dir)->dd_used_bytes,
274 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED,
279 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
281 if (spa->spa_comment != NULL) {
282 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
286 if (spa->spa_root != NULL)
287 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
290 if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) {
291 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL,
292 MIN(zfs_max_recordsize, SPA_MAXBLOCKSIZE), ZPROP_SRC_NONE);
294 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL,
295 SPA_OLD_MAXBLOCKSIZE, ZPROP_SRC_NONE);
298 if ((dp = list_head(&spa->spa_config_list)) != NULL) {
299 if (dp->scd_path == NULL) {
300 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
301 "none", 0, ZPROP_SRC_LOCAL);
302 } else if (strcmp(dp->scd_path, spa_config_path) != 0) {
303 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
304 dp->scd_path, 0, ZPROP_SRC_LOCAL);
310 * Get zpool property values.
313 spa_prop_get(spa_t *spa, nvlist_t **nvp)
315 objset_t *mos = spa->spa_meta_objset;
320 VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
322 mutex_enter(&spa->spa_props_lock);
325 * Get properties from the spa config.
327 spa_prop_get_config(spa, nvp);
329 /* If no pool property object, no more prop to get. */
330 if (mos == NULL || spa->spa_pool_props_object == 0) {
331 mutex_exit(&spa->spa_props_lock);
336 * Get properties from the MOS pool property object.
338 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
339 (err = zap_cursor_retrieve(&zc, &za)) == 0;
340 zap_cursor_advance(&zc)) {
343 zprop_source_t src = ZPROP_SRC_DEFAULT;
346 if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
349 switch (za.za_integer_length) {
351 /* integer property */
352 if (za.za_first_integer !=
353 zpool_prop_default_numeric(prop))
354 src = ZPROP_SRC_LOCAL;
356 if (prop == ZPOOL_PROP_BOOTFS) {
358 dsl_dataset_t *ds = NULL;
360 dp = spa_get_dsl(spa);
361 dsl_pool_config_enter(dp, FTAG);
362 if (err = dsl_dataset_hold_obj(dp,
363 za.za_first_integer, FTAG, &ds)) {
364 dsl_pool_config_exit(dp, FTAG);
369 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
371 dsl_dataset_name(ds, strval);
372 dsl_dataset_rele(ds, FTAG);
373 dsl_pool_config_exit(dp, FTAG);
376 intval = za.za_first_integer;
379 spa_prop_add_list(*nvp, prop, strval, intval, src);
383 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
388 /* string property */
389 strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
390 err = zap_lookup(mos, spa->spa_pool_props_object,
391 za.za_name, 1, za.za_num_integers, strval);
393 kmem_free(strval, za.za_num_integers);
396 spa_prop_add_list(*nvp, prop, strval, 0, src);
397 kmem_free(strval, za.za_num_integers);
404 zap_cursor_fini(&zc);
405 mutex_exit(&spa->spa_props_lock);
407 if (err && err != ENOENT) {
417 * Validate the given pool properties nvlist and modify the list
418 * for the property values to be set.
421 spa_prop_validate(spa_t *spa, nvlist_t *props)
424 int error = 0, reset_bootfs = 0;
426 boolean_t has_feature = B_FALSE;
429 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
431 char *strval, *slash, *check, *fname;
432 const char *propname = nvpair_name(elem);
433 zpool_prop_t prop = zpool_name_to_prop(propname);
437 if (!zpool_prop_feature(propname)) {
438 error = SET_ERROR(EINVAL);
443 * Sanitize the input.
445 if (nvpair_type(elem) != DATA_TYPE_UINT64) {
446 error = SET_ERROR(EINVAL);
450 if (nvpair_value_uint64(elem, &intval) != 0) {
451 error = SET_ERROR(EINVAL);
456 error = SET_ERROR(EINVAL);
460 fname = strchr(propname, '@') + 1;
461 if (zfeature_lookup_name(fname, NULL) != 0) {
462 error = SET_ERROR(EINVAL);
466 has_feature = B_TRUE;
469 case ZPOOL_PROP_VERSION:
470 error = nvpair_value_uint64(elem, &intval);
472 (intval < spa_version(spa) ||
473 intval > SPA_VERSION_BEFORE_FEATURES ||
475 error = SET_ERROR(EINVAL);
478 case ZPOOL_PROP_DELEGATION:
479 case ZPOOL_PROP_AUTOREPLACE:
480 case ZPOOL_PROP_LISTSNAPS:
481 case ZPOOL_PROP_AUTOEXPAND:
482 error = nvpair_value_uint64(elem, &intval);
483 if (!error && intval > 1)
484 error = SET_ERROR(EINVAL);
487 case ZPOOL_PROP_BOOTFS:
489 * If the pool version is less than SPA_VERSION_BOOTFS,
490 * or the pool is still being created (version == 0),
491 * the bootfs property cannot be set.
493 if (spa_version(spa) < SPA_VERSION_BOOTFS) {
494 error = SET_ERROR(ENOTSUP);
499 * Make sure the vdev config is bootable
501 if (!vdev_is_bootable(spa->spa_root_vdev)) {
502 error = SET_ERROR(ENOTSUP);
508 error = nvpair_value_string(elem, &strval);
514 if (strval == NULL || strval[0] == '\0') {
515 objnum = zpool_prop_default_numeric(
520 if (error = dmu_objset_hold(strval, FTAG, &os))
524 * Must be ZPL, and its property settings
525 * must be supported by GRUB (compression
526 * is not gzip, and large blocks are not used).
529 if (dmu_objset_type(os) != DMU_OST_ZFS) {
530 error = SET_ERROR(ENOTSUP);
532 dsl_prop_get_int_ds(dmu_objset_ds(os),
533 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
535 !BOOTFS_COMPRESS_VALID(propval)) {
536 error = SET_ERROR(ENOTSUP);
538 dsl_prop_get_int_ds(dmu_objset_ds(os),
539 zfs_prop_to_name(ZFS_PROP_RECORDSIZE),
541 propval > SPA_OLD_MAXBLOCKSIZE) {
542 error = SET_ERROR(ENOTSUP);
544 objnum = dmu_objset_id(os);
546 dmu_objset_rele(os, FTAG);
550 case ZPOOL_PROP_FAILUREMODE:
551 error = nvpair_value_uint64(elem, &intval);
552 if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
553 intval > ZIO_FAILURE_MODE_PANIC))
554 error = SET_ERROR(EINVAL);
557 * This is a special case which only occurs when
558 * the pool has completely failed. This allows
559 * the user to change the in-core failmode property
560 * without syncing it out to disk (I/Os might
561 * currently be blocked). We do this by returning
562 * EIO to the caller (spa_prop_set) to trick it
563 * into thinking we encountered a property validation
566 if (!error && spa_suspended(spa)) {
567 spa->spa_failmode = intval;
568 error = SET_ERROR(EIO);
572 case ZPOOL_PROP_CACHEFILE:
573 if ((error = nvpair_value_string(elem, &strval)) != 0)
576 if (strval[0] == '\0')
579 if (strcmp(strval, "none") == 0)
582 if (strval[0] != '/') {
583 error = SET_ERROR(EINVAL);
587 slash = strrchr(strval, '/');
588 ASSERT(slash != NULL);
590 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
591 strcmp(slash, "/..") == 0)
592 error = SET_ERROR(EINVAL);
595 case ZPOOL_PROP_COMMENT:
596 if ((error = nvpair_value_string(elem, &strval)) != 0)
598 for (check = strval; *check != '\0'; check++) {
600 * The kernel doesn't have an easy isprint()
601 * check. For this kernel check, we merely
602 * check ASCII apart from DEL. Fix this if
603 * there is an easy-to-use kernel isprint().
605 if (*check >= 0x7f) {
606 error = SET_ERROR(EINVAL);
611 if (strlen(strval) > ZPROP_MAX_COMMENT)
615 case ZPOOL_PROP_DEDUPDITTO:
616 if (spa_version(spa) < SPA_VERSION_DEDUP)
617 error = SET_ERROR(ENOTSUP);
619 error = nvpair_value_uint64(elem, &intval);
621 intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
622 error = SET_ERROR(EINVAL);
630 if (!error && reset_bootfs) {
631 error = nvlist_remove(props,
632 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
635 error = nvlist_add_uint64(props,
636 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
644 spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
647 spa_config_dirent_t *dp;
649 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
653 dp = kmem_alloc(sizeof (spa_config_dirent_t),
656 if (cachefile[0] == '\0')
657 dp->scd_path = spa_strdup(spa_config_path);
658 else if (strcmp(cachefile, "none") == 0)
661 dp->scd_path = spa_strdup(cachefile);
663 list_insert_head(&spa->spa_config_list, dp);
665 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
669 spa_prop_set(spa_t *spa, nvlist_t *nvp)
672 nvpair_t *elem = NULL;
673 boolean_t need_sync = B_FALSE;
675 if ((error = spa_prop_validate(spa, nvp)) != 0)
678 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
679 zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem));
681 if (prop == ZPOOL_PROP_CACHEFILE ||
682 prop == ZPOOL_PROP_ALTROOT ||
683 prop == ZPOOL_PROP_READONLY)
686 if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) {
689 if (prop == ZPOOL_PROP_VERSION) {
690 VERIFY(nvpair_value_uint64(elem, &ver) == 0);
692 ASSERT(zpool_prop_feature(nvpair_name(elem)));
693 ver = SPA_VERSION_FEATURES;
697 /* Save time if the version is already set. */
698 if (ver == spa_version(spa))
702 * In addition to the pool directory object, we might
703 * create the pool properties object, the features for
704 * read object, the features for write object, or the
705 * feature descriptions object.
707 error = dsl_sync_task(spa->spa_name, NULL,
708 spa_sync_version, &ver,
709 6, ZFS_SPACE_CHECK_RESERVED);
720 return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props,
721 nvp, 6, ZFS_SPACE_CHECK_RESERVED));
728 * If the bootfs property value is dsobj, clear it.
731 spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
733 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
734 VERIFY(zap_remove(spa->spa_meta_objset,
735 spa->spa_pool_props_object,
736 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
743 spa_change_guid_check(void *arg, dmu_tx_t *tx)
745 uint64_t *newguid = arg;
746 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
747 vdev_t *rvd = spa->spa_root_vdev;
750 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
751 vdev_state = rvd->vdev_state;
752 spa_config_exit(spa, SCL_STATE, FTAG);
754 if (vdev_state != VDEV_STATE_HEALTHY)
755 return (SET_ERROR(ENXIO));
757 ASSERT3U(spa_guid(spa), !=, *newguid);
763 spa_change_guid_sync(void *arg, dmu_tx_t *tx)
765 uint64_t *newguid = arg;
766 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
768 vdev_t *rvd = spa->spa_root_vdev;
770 oldguid = spa_guid(spa);
772 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
773 rvd->vdev_guid = *newguid;
774 rvd->vdev_guid_sum += (*newguid - oldguid);
775 vdev_config_dirty(rvd);
776 spa_config_exit(spa, SCL_STATE, FTAG);
778 spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu",
783 * Change the GUID for the pool. This is done so that we can later
784 * re-import a pool built from a clone of our own vdevs. We will modify
785 * the root vdev's guid, our own pool guid, and then mark all of our
786 * vdevs dirty. Note that we must make sure that all our vdevs are
787 * online when we do this, or else any vdevs that weren't present
788 * would be orphaned from our pool. We are also going to issue a
789 * sysevent to update any watchers.
792 spa_change_guid(spa_t *spa)
797 mutex_enter(&spa->spa_vdev_top_lock);
798 mutex_enter(&spa_namespace_lock);
799 guid = spa_generate_guid(NULL);
801 error = dsl_sync_task(spa->spa_name, spa_change_guid_check,
802 spa_change_guid_sync, &guid, 5, ZFS_SPACE_CHECK_RESERVED);
805 spa_config_sync(spa, B_FALSE, B_TRUE);
806 spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID);
809 mutex_exit(&spa_namespace_lock);
810 mutex_exit(&spa->spa_vdev_top_lock);
816 * ==========================================================================
817 * SPA state manipulation (open/create/destroy/import/export)
818 * ==========================================================================
822 spa_error_entry_compare(const void *a, const void *b)
824 spa_error_entry_t *sa = (spa_error_entry_t *)a;
825 spa_error_entry_t *sb = (spa_error_entry_t *)b;
828 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
829 sizeof (zbookmark_phys_t));
840 * Utility function which retrieves copies of the current logs and
841 * re-initializes them in the process.
844 spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
846 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
848 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
849 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
851 avl_create(&spa->spa_errlist_scrub,
852 spa_error_entry_compare, sizeof (spa_error_entry_t),
853 offsetof(spa_error_entry_t, se_avl));
854 avl_create(&spa->spa_errlist_last,
855 spa_error_entry_compare, sizeof (spa_error_entry_t),
856 offsetof(spa_error_entry_t, se_avl));
860 spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
862 const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
863 enum zti_modes mode = ztip->zti_mode;
864 uint_t value = ztip->zti_value;
865 uint_t count = ztip->zti_count;
866 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
869 boolean_t batch = B_FALSE;
871 if (mode == ZTI_MODE_NULL) {
873 tqs->stqs_taskq = NULL;
877 ASSERT3U(count, >, 0);
879 tqs->stqs_count = count;
880 tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP);
884 ASSERT3U(value, >=, 1);
885 value = MAX(value, 1);
890 flags |= TASKQ_THREADS_CPU_PCT;
891 value = zio_taskq_batch_pct;
895 panic("unrecognized mode for %s_%s taskq (%u:%u) in "
897 zio_type_name[t], zio_taskq_types[q], mode, value);
901 for (uint_t i = 0; i < count; i++) {
905 (void) snprintf(name, sizeof (name), "%s_%s_%u",
906 zio_type_name[t], zio_taskq_types[q], i);
908 (void) snprintf(name, sizeof (name), "%s_%s",
909 zio_type_name[t], zio_taskq_types[q]);
913 if (zio_taskq_sysdc && spa->spa_proc != &p0) {
915 flags |= TASKQ_DC_BATCH;
917 tq = taskq_create_sysdc(name, value, 50, INT_MAX,
918 spa->spa_proc, zio_taskq_basedc, flags);
921 pri_t pri = maxclsyspri;
923 * The write issue taskq can be extremely CPU
924 * intensive. Run it at slightly lower priority
925 * than the other taskqs.
927 if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE)
930 tq = taskq_create_proc(name, value, pri, 50,
931 INT_MAX, spa->spa_proc, flags);
936 tqs->stqs_taskq[i] = tq;
941 spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
943 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
945 if (tqs->stqs_taskq == NULL) {
946 ASSERT0(tqs->stqs_count);
950 for (uint_t i = 0; i < tqs->stqs_count; i++) {
951 ASSERT3P(tqs->stqs_taskq[i], !=, NULL);
952 taskq_destroy(tqs->stqs_taskq[i]);
955 kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *));
956 tqs->stqs_taskq = NULL;
960 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
961 * Note that a type may have multiple discrete taskqs to avoid lock contention
962 * on the taskq itself. In that case we choose which taskq at random by using
963 * the low bits of gethrtime().
966 spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q,
967 task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent)
969 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
972 ASSERT3P(tqs->stqs_taskq, !=, NULL);
973 ASSERT3U(tqs->stqs_count, !=, 0);
975 if (tqs->stqs_count == 1) {
976 tq = tqs->stqs_taskq[0];
979 tq = tqs->stqs_taskq[cpu_ticks() % tqs->stqs_count];
981 tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count];
985 taskq_dispatch_ent(tq, func, arg, flags, ent);
989 spa_create_zio_taskqs(spa_t *spa)
991 for (int t = 0; t < ZIO_TYPES; t++) {
992 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
993 spa_taskqs_init(spa, t, q);
1001 spa_thread(void *arg)
1003 callb_cpr_t cprinfo;
1006 user_t *pu = PTOU(curproc);
1008 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
1011 ASSERT(curproc != &p0);
1012 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
1013 "zpool-%s", spa->spa_name);
1014 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
1017 /* bind this thread to the requested psrset */
1018 if (zio_taskq_psrset_bind != PS_NONE) {
1020 mutex_enter(&cpu_lock);
1021 mutex_enter(&pidlock);
1022 mutex_enter(&curproc->p_lock);
1024 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
1025 0, NULL, NULL) == 0) {
1026 curthread->t_bind_pset = zio_taskq_psrset_bind;
1029 "Couldn't bind process for zfs pool \"%s\" to "
1030 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
1033 mutex_exit(&curproc->p_lock);
1034 mutex_exit(&pidlock);
1035 mutex_exit(&cpu_lock);
1041 if (zio_taskq_sysdc) {
1042 sysdc_thread_enter(curthread, 100, 0);
1046 spa->spa_proc = curproc;
1047 spa->spa_did = curthread->t_did;
1049 spa_create_zio_taskqs(spa);
1051 mutex_enter(&spa->spa_proc_lock);
1052 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
1054 spa->spa_proc_state = SPA_PROC_ACTIVE;
1055 cv_broadcast(&spa->spa_proc_cv);
1057 CALLB_CPR_SAFE_BEGIN(&cprinfo);
1058 while (spa->spa_proc_state == SPA_PROC_ACTIVE)
1059 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1060 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
1062 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
1063 spa->spa_proc_state = SPA_PROC_GONE;
1064 spa->spa_proc = &p0;
1065 cv_broadcast(&spa->spa_proc_cv);
1066 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */
1068 mutex_enter(&curproc->p_lock);
1071 #endif /* SPA_PROCESS */
1075 * Activate an uninitialized pool.
1078 spa_activate(spa_t *spa, int mode)
1080 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
1082 spa->spa_state = POOL_STATE_ACTIVE;
1083 spa->spa_mode = mode;
1085 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
1086 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
1088 /* Try to create a covering process */
1089 mutex_enter(&spa->spa_proc_lock);
1090 ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
1091 ASSERT(spa->spa_proc == &p0);
1095 /* Only create a process if we're going to be around a while. */
1096 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
1097 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
1099 spa->spa_proc_state = SPA_PROC_CREATED;
1100 while (spa->spa_proc_state == SPA_PROC_CREATED) {
1101 cv_wait(&spa->spa_proc_cv,
1102 &spa->spa_proc_lock);
1104 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1105 ASSERT(spa->spa_proc != &p0);
1106 ASSERT(spa->spa_did != 0);
1110 "Couldn't create process for zfs pool \"%s\"\n",
1115 #endif /* SPA_PROCESS */
1116 mutex_exit(&spa->spa_proc_lock);
1118 /* If we didn't create a process, we need to create our taskqs. */
1119 ASSERT(spa->spa_proc == &p0);
1120 if (spa->spa_proc == &p0) {
1121 spa_create_zio_taskqs(spa);
1125 * Start TRIM thread.
1127 trim_thread_create(spa);
1129 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
1130 offsetof(vdev_t, vdev_config_dirty_node));
1131 list_create(&spa->spa_evicting_os_list, sizeof (objset_t),
1132 offsetof(objset_t, os_evicting_node));
1133 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
1134 offsetof(vdev_t, vdev_state_dirty_node));
1136 txg_list_create(&spa->spa_vdev_txg_list,
1137 offsetof(struct vdev, vdev_txg_node));
1139 avl_create(&spa->spa_errlist_scrub,
1140 spa_error_entry_compare, sizeof (spa_error_entry_t),
1141 offsetof(spa_error_entry_t, se_avl));
1142 avl_create(&spa->spa_errlist_last,
1143 spa_error_entry_compare, sizeof (spa_error_entry_t),
1144 offsetof(spa_error_entry_t, se_avl));
1148 * Opposite of spa_activate().
1151 spa_deactivate(spa_t *spa)
1153 ASSERT(spa->spa_sync_on == B_FALSE);
1154 ASSERT(spa->spa_dsl_pool == NULL);
1155 ASSERT(spa->spa_root_vdev == NULL);
1156 ASSERT(spa->spa_async_zio_root == NULL);
1157 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
1160 * Stop TRIM thread in case spa_unload() wasn't called directly
1161 * before spa_deactivate().
1163 trim_thread_destroy(spa);
1165 spa_evicting_os_wait(spa);
1167 txg_list_destroy(&spa->spa_vdev_txg_list);
1169 list_destroy(&spa->spa_config_dirty_list);
1170 list_destroy(&spa->spa_evicting_os_list);
1171 list_destroy(&spa->spa_state_dirty_list);
1173 for (int t = 0; t < ZIO_TYPES; t++) {
1174 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1175 spa_taskqs_fini(spa, t, q);
1179 metaslab_class_destroy(spa->spa_normal_class);
1180 spa->spa_normal_class = NULL;
1182 metaslab_class_destroy(spa->spa_log_class);
1183 spa->spa_log_class = NULL;
1186 * If this was part of an import or the open otherwise failed, we may
1187 * still have errors left in the queues. Empty them just in case.
1189 spa_errlog_drain(spa);
1191 avl_destroy(&spa->spa_errlist_scrub);
1192 avl_destroy(&spa->spa_errlist_last);
1194 spa->spa_state = POOL_STATE_UNINITIALIZED;
1196 mutex_enter(&spa->spa_proc_lock);
1197 if (spa->spa_proc_state != SPA_PROC_NONE) {
1198 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1199 spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1200 cv_broadcast(&spa->spa_proc_cv);
1201 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1202 ASSERT(spa->spa_proc != &p0);
1203 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1205 ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1206 spa->spa_proc_state = SPA_PROC_NONE;
1208 ASSERT(spa->spa_proc == &p0);
1209 mutex_exit(&spa->spa_proc_lock);
1213 * We want to make sure spa_thread() has actually exited the ZFS
1214 * module, so that the module can't be unloaded out from underneath
1217 if (spa->spa_did != 0) {
1218 thread_join(spa->spa_did);
1221 #endif /* SPA_PROCESS */
1225 * Verify a pool configuration, and construct the vdev tree appropriately. This
1226 * will create all the necessary vdevs in the appropriate layout, with each vdev
1227 * in the CLOSED state. This will prep the pool before open/creation/import.
1228 * All vdev validation is done by the vdev_alloc() routine.
1231 spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1232 uint_t id, int atype)
1238 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1241 if ((*vdp)->vdev_ops->vdev_op_leaf)
1244 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1247 if (error == ENOENT)
1253 return (SET_ERROR(EINVAL));
1256 for (int c = 0; c < children; c++) {
1258 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1266 ASSERT(*vdp != NULL);
1272 * Opposite of spa_load().
1275 spa_unload(spa_t *spa)
1279 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1284 trim_thread_destroy(spa);
1289 spa_async_suspend(spa);
1294 if (spa->spa_sync_on) {
1295 txg_sync_stop(spa->spa_dsl_pool);
1296 spa->spa_sync_on = B_FALSE;
1300 * Wait for any outstanding async I/O to complete.
1302 if (spa->spa_async_zio_root != NULL) {
1303 for (int i = 0; i < max_ncpus; i++)
1304 (void) zio_wait(spa->spa_async_zio_root[i]);
1305 kmem_free(spa->spa_async_zio_root, max_ncpus * sizeof (void *));
1306 spa->spa_async_zio_root = NULL;
1309 bpobj_close(&spa->spa_deferred_bpobj);
1311 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1316 if (spa->spa_root_vdev)
1317 vdev_free(spa->spa_root_vdev);
1318 ASSERT(spa->spa_root_vdev == NULL);
1321 * Close the dsl pool.
1323 if (spa->spa_dsl_pool) {
1324 dsl_pool_close(spa->spa_dsl_pool);
1325 spa->spa_dsl_pool = NULL;
1326 spa->spa_meta_objset = NULL;
1333 * Drop and purge level 2 cache
1335 spa_l2cache_drop(spa);
1337 for (i = 0; i < spa->spa_spares.sav_count; i++)
1338 vdev_free(spa->spa_spares.sav_vdevs[i]);
1339 if (spa->spa_spares.sav_vdevs) {
1340 kmem_free(spa->spa_spares.sav_vdevs,
1341 spa->spa_spares.sav_count * sizeof (void *));
1342 spa->spa_spares.sav_vdevs = NULL;
1344 if (spa->spa_spares.sav_config) {
1345 nvlist_free(spa->spa_spares.sav_config);
1346 spa->spa_spares.sav_config = NULL;
1348 spa->spa_spares.sav_count = 0;
1350 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1351 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1352 vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1354 if (spa->spa_l2cache.sav_vdevs) {
1355 kmem_free(spa->spa_l2cache.sav_vdevs,
1356 spa->spa_l2cache.sav_count * sizeof (void *));
1357 spa->spa_l2cache.sav_vdevs = NULL;
1359 if (spa->spa_l2cache.sav_config) {
1360 nvlist_free(spa->spa_l2cache.sav_config);
1361 spa->spa_l2cache.sav_config = NULL;
1363 spa->spa_l2cache.sav_count = 0;
1365 spa->spa_async_suspended = 0;
1367 if (spa->spa_comment != NULL) {
1368 spa_strfree(spa->spa_comment);
1369 spa->spa_comment = NULL;
1372 spa_config_exit(spa, SCL_ALL, FTAG);
1376 * Load (or re-load) the current list of vdevs describing the active spares for
1377 * this pool. When this is called, we have some form of basic information in
1378 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1379 * then re-generate a more complete list including status information.
1382 spa_load_spares(spa_t *spa)
1389 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1392 * First, close and free any existing spare vdevs.
1394 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1395 vd = spa->spa_spares.sav_vdevs[i];
1397 /* Undo the call to spa_activate() below */
1398 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1399 B_FALSE)) != NULL && tvd->vdev_isspare)
1400 spa_spare_remove(tvd);
1405 if (spa->spa_spares.sav_vdevs)
1406 kmem_free(spa->spa_spares.sav_vdevs,
1407 spa->spa_spares.sav_count * sizeof (void *));
1409 if (spa->spa_spares.sav_config == NULL)
1412 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1413 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1415 spa->spa_spares.sav_count = (int)nspares;
1416 spa->spa_spares.sav_vdevs = NULL;
1422 * Construct the array of vdevs, opening them to get status in the
1423 * process. For each spare, there is potentially two different vdev_t
1424 * structures associated with it: one in the list of spares (used only
1425 * for basic validation purposes) and one in the active vdev
1426 * configuration (if it's spared in). During this phase we open and
1427 * validate each vdev on the spare list. If the vdev also exists in the
1428 * active configuration, then we also mark this vdev as an active spare.
1430 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1432 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1433 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1434 VDEV_ALLOC_SPARE) == 0);
1437 spa->spa_spares.sav_vdevs[i] = vd;
1439 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1440 B_FALSE)) != NULL) {
1441 if (!tvd->vdev_isspare)
1445 * We only mark the spare active if we were successfully
1446 * able to load the vdev. Otherwise, importing a pool
1447 * with a bad active spare would result in strange
1448 * behavior, because multiple pool would think the spare
1449 * is actively in use.
1451 * There is a vulnerability here to an equally bizarre
1452 * circumstance, where a dead active spare is later
1453 * brought back to life (onlined or otherwise). Given
1454 * the rarity of this scenario, and the extra complexity
1455 * it adds, we ignore the possibility.
1457 if (!vdev_is_dead(tvd))
1458 spa_spare_activate(tvd);
1462 vd->vdev_aux = &spa->spa_spares;
1464 if (vdev_open(vd) != 0)
1467 if (vdev_validate_aux(vd) == 0)
1472 * Recompute the stashed list of spares, with status information
1475 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1476 DATA_TYPE_NVLIST_ARRAY) == 0);
1478 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1480 for (i = 0; i < spa->spa_spares.sav_count; i++)
1481 spares[i] = vdev_config_generate(spa,
1482 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1483 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1484 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1485 for (i = 0; i < spa->spa_spares.sav_count; i++)
1486 nvlist_free(spares[i]);
1487 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1491 * Load (or re-load) the current list of vdevs describing the active l2cache for
1492 * this pool. When this is called, we have some form of basic information in
1493 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1494 * then re-generate a more complete list including status information.
1495 * Devices which are already active have their details maintained, and are
1499 spa_load_l2cache(spa_t *spa)
1503 int i, j, oldnvdevs;
1505 vdev_t *vd, **oldvdevs, **newvdevs;
1506 spa_aux_vdev_t *sav = &spa->spa_l2cache;
1508 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1510 if (sav->sav_config != NULL) {
1511 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1512 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1513 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1519 oldvdevs = sav->sav_vdevs;
1520 oldnvdevs = sav->sav_count;
1521 sav->sav_vdevs = NULL;
1525 * Process new nvlist of vdevs.
1527 for (i = 0; i < nl2cache; i++) {
1528 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1532 for (j = 0; j < oldnvdevs; j++) {
1534 if (vd != NULL && guid == vd->vdev_guid) {
1536 * Retain previous vdev for add/remove ops.
1544 if (newvdevs[i] == NULL) {
1548 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1549 VDEV_ALLOC_L2CACHE) == 0);
1554 * Commit this vdev as an l2cache device,
1555 * even if it fails to open.
1557 spa_l2cache_add(vd);
1562 spa_l2cache_activate(vd);
1564 if (vdev_open(vd) != 0)
1567 (void) vdev_validate_aux(vd);
1569 if (!vdev_is_dead(vd))
1570 l2arc_add_vdev(spa, vd);
1575 * Purge vdevs that were dropped
1577 for (i = 0; i < oldnvdevs; i++) {
1582 ASSERT(vd->vdev_isl2cache);
1584 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1585 pool != 0ULL && l2arc_vdev_present(vd))
1586 l2arc_remove_vdev(vd);
1587 vdev_clear_stats(vd);
1593 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1595 if (sav->sav_config == NULL)
1598 sav->sav_vdevs = newvdevs;
1599 sav->sav_count = (int)nl2cache;
1602 * Recompute the stashed list of l2cache devices, with status
1603 * information this time.
1605 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1606 DATA_TYPE_NVLIST_ARRAY) == 0);
1608 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1609 for (i = 0; i < sav->sav_count; i++)
1610 l2cache[i] = vdev_config_generate(spa,
1611 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1612 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1613 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1615 for (i = 0; i < sav->sav_count; i++)
1616 nvlist_free(l2cache[i]);
1618 kmem_free(l2cache, sav->sav_count * sizeof (void *));
1622 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1625 char *packed = NULL;
1630 error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db);
1634 nvsize = *(uint64_t *)db->db_data;
1635 dmu_buf_rele(db, FTAG);
1637 packed = kmem_alloc(nvsize, KM_SLEEP);
1638 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1641 error = nvlist_unpack(packed, nvsize, value, 0);
1642 kmem_free(packed, nvsize);
1648 * Checks to see if the given vdev could not be opened, in which case we post a
1649 * sysevent to notify the autoreplace code that the device has been removed.
1652 spa_check_removed(vdev_t *vd)
1654 for (int c = 0; c < vd->vdev_children; c++)
1655 spa_check_removed(vd->vdev_child[c]);
1657 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) &&
1659 zfs_post_autoreplace(vd->vdev_spa, vd);
1660 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1665 * Validate the current config against the MOS config
1668 spa_config_valid(spa_t *spa, nvlist_t *config)
1670 vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1673 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1675 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1676 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1678 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1681 * If we're doing a normal import, then build up any additional
1682 * diagnostic information about missing devices in this config.
1683 * We'll pass this up to the user for further processing.
1685 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1686 nvlist_t **child, *nv;
1689 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1691 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1693 for (int c = 0; c < rvd->vdev_children; c++) {
1694 vdev_t *tvd = rvd->vdev_child[c];
1695 vdev_t *mtvd = mrvd->vdev_child[c];
1697 if (tvd->vdev_ops == &vdev_missing_ops &&
1698 mtvd->vdev_ops != &vdev_missing_ops &&
1700 child[idx++] = vdev_config_generate(spa, mtvd,
1705 VERIFY(nvlist_add_nvlist_array(nv,
1706 ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1707 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1708 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1710 for (int i = 0; i < idx; i++)
1711 nvlist_free(child[i]);
1714 kmem_free(child, rvd->vdev_children * sizeof (char **));
1718 * Compare the root vdev tree with the information we have
1719 * from the MOS config (mrvd). Check each top-level vdev
1720 * with the corresponding MOS config top-level (mtvd).
1722 for (int c = 0; c < rvd->vdev_children; c++) {
1723 vdev_t *tvd = rvd->vdev_child[c];
1724 vdev_t *mtvd = mrvd->vdev_child[c];
1727 * Resolve any "missing" vdevs in the current configuration.
1728 * If we find that the MOS config has more accurate information
1729 * about the top-level vdev then use that vdev instead.
1731 if (tvd->vdev_ops == &vdev_missing_ops &&
1732 mtvd->vdev_ops != &vdev_missing_ops) {
1734 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1738 * Device specific actions.
1740 if (mtvd->vdev_islog) {
1741 spa_set_log_state(spa, SPA_LOG_CLEAR);
1744 * XXX - once we have 'readonly' pool
1745 * support we should be able to handle
1746 * missing data devices by transitioning
1747 * the pool to readonly.
1753 * Swap the missing vdev with the data we were
1754 * able to obtain from the MOS config.
1756 vdev_remove_child(rvd, tvd);
1757 vdev_remove_child(mrvd, mtvd);
1759 vdev_add_child(rvd, mtvd);
1760 vdev_add_child(mrvd, tvd);
1762 spa_config_exit(spa, SCL_ALL, FTAG);
1764 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1767 } else if (mtvd->vdev_islog) {
1769 * Load the slog device's state from the MOS config
1770 * since it's possible that the label does not
1771 * contain the most up-to-date information.
1773 vdev_load_log_state(tvd, mtvd);
1778 spa_config_exit(spa, SCL_ALL, FTAG);
1781 * Ensure we were able to validate the config.
1783 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1787 * Check for missing log devices
1790 spa_check_logs(spa_t *spa)
1792 boolean_t rv = B_FALSE;
1793 dsl_pool_t *dp = spa_get_dsl(spa);
1795 switch (spa->spa_log_state) {
1796 case SPA_LOG_MISSING:
1797 /* need to recheck in case slog has been restored */
1798 case SPA_LOG_UNKNOWN:
1799 rv = (dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
1800 zil_check_log_chain, NULL, DS_FIND_CHILDREN) != 0);
1802 spa_set_log_state(spa, SPA_LOG_MISSING);
1809 spa_passivate_log(spa_t *spa)
1811 vdev_t *rvd = spa->spa_root_vdev;
1812 boolean_t slog_found = B_FALSE;
1814 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1816 if (!spa_has_slogs(spa))
1819 for (int c = 0; c < rvd->vdev_children; c++) {
1820 vdev_t *tvd = rvd->vdev_child[c];
1821 metaslab_group_t *mg = tvd->vdev_mg;
1823 if (tvd->vdev_islog) {
1824 metaslab_group_passivate(mg);
1825 slog_found = B_TRUE;
1829 return (slog_found);
1833 spa_activate_log(spa_t *spa)
1835 vdev_t *rvd = spa->spa_root_vdev;
1837 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1839 for (int c = 0; c < rvd->vdev_children; c++) {
1840 vdev_t *tvd = rvd->vdev_child[c];
1841 metaslab_group_t *mg = tvd->vdev_mg;
1843 if (tvd->vdev_islog)
1844 metaslab_group_activate(mg);
1849 spa_offline_log(spa_t *spa)
1853 error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1854 NULL, DS_FIND_CHILDREN);
1857 * We successfully offlined the log device, sync out the
1858 * current txg so that the "stubby" block can be removed
1861 txg_wait_synced(spa->spa_dsl_pool, 0);
1867 spa_aux_check_removed(spa_aux_vdev_t *sav)
1871 for (i = 0; i < sav->sav_count; i++)
1872 spa_check_removed(sav->sav_vdevs[i]);
1876 spa_claim_notify(zio_t *zio)
1878 spa_t *spa = zio->io_spa;
1883 mutex_enter(&spa->spa_props_lock); /* any mutex will do */
1884 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1885 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1886 mutex_exit(&spa->spa_props_lock);
1889 typedef struct spa_load_error {
1890 uint64_t sle_meta_count;
1891 uint64_t sle_data_count;
1895 spa_load_verify_done(zio_t *zio)
1897 blkptr_t *bp = zio->io_bp;
1898 spa_load_error_t *sle = zio->io_private;
1899 dmu_object_type_t type = BP_GET_TYPE(bp);
1900 int error = zio->io_error;
1901 spa_t *spa = zio->io_spa;
1904 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1905 type != DMU_OT_INTENT_LOG)
1906 atomic_inc_64(&sle->sle_meta_count);
1908 atomic_inc_64(&sle->sle_data_count);
1910 zio_data_buf_free(zio->io_data, zio->io_size);
1912 mutex_enter(&spa->spa_scrub_lock);
1913 spa->spa_scrub_inflight--;
1914 cv_broadcast(&spa->spa_scrub_io_cv);
1915 mutex_exit(&spa->spa_scrub_lock);
1919 * Maximum number of concurrent scrub i/os to create while verifying
1920 * a pool while importing it.
1922 int spa_load_verify_maxinflight = 10000;
1923 boolean_t spa_load_verify_metadata = B_TRUE;
1924 boolean_t spa_load_verify_data = B_TRUE;
1926 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_maxinflight, CTLFLAG_RWTUN,
1927 &spa_load_verify_maxinflight, 0,
1928 "Maximum number of concurrent scrub I/Os to create while verifying a "
1929 "pool while importing it");
1931 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_metadata, CTLFLAG_RWTUN,
1932 &spa_load_verify_metadata, 0,
1933 "Check metadata on import?");
1935 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_data, CTLFLAG_RWTUN,
1936 &spa_load_verify_data, 0,
1937 "Check user data on import?");
1941 spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1942 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
1944 if (bp == NULL || BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp))
1947 * Note: normally this routine will not be called if
1948 * spa_load_verify_metadata is not set. However, it may be useful
1949 * to manually set the flag after the traversal has begun.
1951 if (!spa_load_verify_metadata)
1953 if (BP_GET_BUFC_TYPE(bp) == ARC_BUFC_DATA && !spa_load_verify_data)
1957 size_t size = BP_GET_PSIZE(bp);
1958 void *data = zio_data_buf_alloc(size);
1960 mutex_enter(&spa->spa_scrub_lock);
1961 while (spa->spa_scrub_inflight >= spa_load_verify_maxinflight)
1962 cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
1963 spa->spa_scrub_inflight++;
1964 mutex_exit(&spa->spa_scrub_lock);
1966 zio_nowait(zio_read(rio, spa, bp, data, size,
1967 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1968 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1969 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1974 spa_load_verify(spa_t *spa)
1977 spa_load_error_t sle = { 0 };
1978 zpool_rewind_policy_t policy;
1979 boolean_t verify_ok = B_FALSE;
1982 zpool_get_rewind_policy(spa->spa_config, &policy);
1984 if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1987 rio = zio_root(spa, NULL, &sle,
1988 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1990 if (spa_load_verify_metadata) {
1991 error = traverse_pool(spa, spa->spa_verify_min_txg,
1992 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA,
1993 spa_load_verify_cb, rio);
1996 (void) zio_wait(rio);
1998 spa->spa_load_meta_errors = sle.sle_meta_count;
1999 spa->spa_load_data_errors = sle.sle_data_count;
2001 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
2002 sle.sle_data_count <= policy.zrp_maxdata) {
2006 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
2007 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
2009 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
2010 VERIFY(nvlist_add_uint64(spa->spa_load_info,
2011 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
2012 VERIFY(nvlist_add_int64(spa->spa_load_info,
2013 ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
2014 VERIFY(nvlist_add_uint64(spa->spa_load_info,
2015 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
2017 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
2021 if (error != ENXIO && error != EIO)
2022 error = SET_ERROR(EIO);
2026 return (verify_ok ? 0 : EIO);
2030 * Find a value in the pool props object.
2033 spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
2035 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
2036 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
2040 * Find a value in the pool directory object.
2043 spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
2045 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
2046 name, sizeof (uint64_t), 1, val));
2050 spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
2052 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
2057 * Fix up config after a partly-completed split. This is done with the
2058 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
2059 * pool have that entry in their config, but only the splitting one contains
2060 * a list of all the guids of the vdevs that are being split off.
2062 * This function determines what to do with that list: either rejoin
2063 * all the disks to the pool, or complete the splitting process. To attempt
2064 * the rejoin, each disk that is offlined is marked online again, and
2065 * we do a reopen() call. If the vdev label for every disk that was
2066 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
2067 * then we call vdev_split() on each disk, and complete the split.
2069 * Otherwise we leave the config alone, with all the vdevs in place in
2070 * the original pool.
2073 spa_try_repair(spa_t *spa, nvlist_t *config)
2080 boolean_t attempt_reopen;
2082 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
2085 /* check that the config is complete */
2086 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
2087 &glist, &gcount) != 0)
2090 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
2092 /* attempt to online all the vdevs & validate */
2093 attempt_reopen = B_TRUE;
2094 for (i = 0; i < gcount; i++) {
2095 if (glist[i] == 0) /* vdev is hole */
2098 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
2099 if (vd[i] == NULL) {
2101 * Don't bother attempting to reopen the disks;
2102 * just do the split.
2104 attempt_reopen = B_FALSE;
2106 /* attempt to re-online it */
2107 vd[i]->vdev_offline = B_FALSE;
2111 if (attempt_reopen) {
2112 vdev_reopen(spa->spa_root_vdev);
2114 /* check each device to see what state it's in */
2115 for (extracted = 0, i = 0; i < gcount; i++) {
2116 if (vd[i] != NULL &&
2117 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
2124 * If every disk has been moved to the new pool, or if we never
2125 * even attempted to look at them, then we split them off for
2128 if (!attempt_reopen || gcount == extracted) {
2129 for (i = 0; i < gcount; i++)
2132 vdev_reopen(spa->spa_root_vdev);
2135 kmem_free(vd, gcount * sizeof (vdev_t *));
2139 spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
2140 boolean_t mosconfig)
2142 nvlist_t *config = spa->spa_config;
2143 char *ereport = FM_EREPORT_ZFS_POOL;
2149 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
2150 return (SET_ERROR(EINVAL));
2152 ASSERT(spa->spa_comment == NULL);
2153 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
2154 spa->spa_comment = spa_strdup(comment);
2157 * Versioning wasn't explicitly added to the label until later, so if
2158 * it's not present treat it as the initial version.
2160 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
2161 &spa->spa_ubsync.ub_version) != 0)
2162 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
2164 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
2165 &spa->spa_config_txg);
2167 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
2168 spa_guid_exists(pool_guid, 0)) {
2169 error = SET_ERROR(EEXIST);
2171 spa->spa_config_guid = pool_guid;
2173 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
2175 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
2179 nvlist_free(spa->spa_load_info);
2180 spa->spa_load_info = fnvlist_alloc();
2182 gethrestime(&spa->spa_loaded_ts);
2183 error = spa_load_impl(spa, pool_guid, config, state, type,
2184 mosconfig, &ereport);
2188 * Don't count references from objsets that are already closed
2189 * and are making their way through the eviction process.
2191 spa_evicting_os_wait(spa);
2192 spa->spa_minref = refcount_count(&spa->spa_refcount);
2194 if (error != EEXIST) {
2195 spa->spa_loaded_ts.tv_sec = 0;
2196 spa->spa_loaded_ts.tv_nsec = 0;
2198 if (error != EBADF) {
2199 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
2202 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2209 * Load an existing storage pool, using the pool's builtin spa_config as a
2210 * source of configuration information.
2213 spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
2214 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
2218 nvlist_t *nvroot = NULL;
2221 uberblock_t *ub = &spa->spa_uberblock;
2222 uint64_t children, config_cache_txg = spa->spa_config_txg;
2223 int orig_mode = spa->spa_mode;
2226 boolean_t missing_feat_write = B_FALSE;
2229 * If this is an untrusted config, access the pool in read-only mode.
2230 * This prevents things like resilvering recently removed devices.
2233 spa->spa_mode = FREAD;
2235 ASSERT(MUTEX_HELD(&spa_namespace_lock));
2237 spa->spa_load_state = state;
2239 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
2240 return (SET_ERROR(EINVAL));
2242 parse = (type == SPA_IMPORT_EXISTING ?
2243 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2246 * Create "The Godfather" zio to hold all async IOs
2248 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
2250 for (int i = 0; i < max_ncpus; i++) {
2251 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
2252 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
2253 ZIO_FLAG_GODFATHER);
2257 * Parse the configuration into a vdev tree. We explicitly set the
2258 * value that will be returned by spa_version() since parsing the
2259 * configuration requires knowing the version number.
2261 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2262 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
2263 spa_config_exit(spa, SCL_ALL, FTAG);
2268 ASSERT(spa->spa_root_vdev == rvd);
2269 ASSERT3U(spa->spa_min_ashift, >=, SPA_MINBLOCKSHIFT);
2270 ASSERT3U(spa->spa_max_ashift, <=, SPA_MAXBLOCKSHIFT);
2272 if (type != SPA_IMPORT_ASSEMBLE) {
2273 ASSERT(spa_guid(spa) == pool_guid);
2277 * Try to open all vdevs, loading each label in the process.
2279 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2280 error = vdev_open(rvd);
2281 spa_config_exit(spa, SCL_ALL, FTAG);
2286 * We need to validate the vdev labels against the configuration that
2287 * we have in hand, which is dependent on the setting of mosconfig. If
2288 * mosconfig is true then we're validating the vdev labels based on
2289 * that config. Otherwise, we're validating against the cached config
2290 * (zpool.cache) that was read when we loaded the zfs module, and then
2291 * later we will recursively call spa_load() and validate against
2294 * If we're assembling a new pool that's been split off from an
2295 * existing pool, the labels haven't yet been updated so we skip
2296 * validation for now.
2298 if (type != SPA_IMPORT_ASSEMBLE) {
2299 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2300 error = vdev_validate(rvd, mosconfig);
2301 spa_config_exit(spa, SCL_ALL, FTAG);
2306 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2307 return (SET_ERROR(ENXIO));
2311 * Find the best uberblock.
2313 vdev_uberblock_load(rvd, ub, &label);
2316 * If we weren't able to find a single valid uberblock, return failure.
2318 if (ub->ub_txg == 0) {
2320 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2324 * If the pool has an unsupported version we can't open it.
2326 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2328 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2331 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2335 * If we weren't able to find what's necessary for reading the
2336 * MOS in the label, return failure.
2338 if (label == NULL || nvlist_lookup_nvlist(label,
2339 ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) {
2341 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2346 * Update our in-core representation with the definitive values
2349 nvlist_free(spa->spa_label_features);
2350 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2356 * Look through entries in the label nvlist's features_for_read. If
2357 * there is a feature listed there which we don't understand then we
2358 * cannot open a pool.
2360 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2361 nvlist_t *unsup_feat;
2363 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2366 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2368 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2369 if (!zfeature_is_supported(nvpair_name(nvp))) {
2370 VERIFY(nvlist_add_string(unsup_feat,
2371 nvpair_name(nvp), "") == 0);
2375 if (!nvlist_empty(unsup_feat)) {
2376 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2377 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2378 nvlist_free(unsup_feat);
2379 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2383 nvlist_free(unsup_feat);
2387 * If the vdev guid sum doesn't match the uberblock, we have an
2388 * incomplete configuration. We first check to see if the pool
2389 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2390 * If it is, defer the vdev_guid_sum check till later so we
2391 * can handle missing vdevs.
2393 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
2394 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
2395 rvd->vdev_guid_sum != ub->ub_guid_sum)
2396 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2398 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2399 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2400 spa_try_repair(spa, config);
2401 spa_config_exit(spa, SCL_ALL, FTAG);
2402 nvlist_free(spa->spa_config_splitting);
2403 spa->spa_config_splitting = NULL;
2407 * Initialize internal SPA structures.
2409 spa->spa_state = POOL_STATE_ACTIVE;
2410 spa->spa_ubsync = spa->spa_uberblock;
2411 spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2412 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2413 spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2414 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2415 spa->spa_claim_max_txg = spa->spa_first_txg;
2416 spa->spa_prev_software_version = ub->ub_software_version;
2418 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2420 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2421 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2423 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2424 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2426 if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2427 boolean_t missing_feat_read = B_FALSE;
2428 nvlist_t *unsup_feat, *enabled_feat;
2430 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2431 &spa->spa_feat_for_read_obj) != 0) {
2432 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2435 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2436 &spa->spa_feat_for_write_obj) != 0) {
2437 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2440 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2441 &spa->spa_feat_desc_obj) != 0) {
2442 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2445 enabled_feat = fnvlist_alloc();
2446 unsup_feat = fnvlist_alloc();
2448 if (!spa_features_check(spa, B_FALSE,
2449 unsup_feat, enabled_feat))
2450 missing_feat_read = B_TRUE;
2452 if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) {
2453 if (!spa_features_check(spa, B_TRUE,
2454 unsup_feat, enabled_feat)) {
2455 missing_feat_write = B_TRUE;
2459 fnvlist_add_nvlist(spa->spa_load_info,
2460 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
2462 if (!nvlist_empty(unsup_feat)) {
2463 fnvlist_add_nvlist(spa->spa_load_info,
2464 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
2467 fnvlist_free(enabled_feat);
2468 fnvlist_free(unsup_feat);
2470 if (!missing_feat_read) {
2471 fnvlist_add_boolean(spa->spa_load_info,
2472 ZPOOL_CONFIG_CAN_RDONLY);
2476 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2477 * twofold: to determine whether the pool is available for
2478 * import in read-write mode and (if it is not) whether the
2479 * pool is available for import in read-only mode. If the pool
2480 * is available for import in read-write mode, it is displayed
2481 * as available in userland; if it is not available for import
2482 * in read-only mode, it is displayed as unavailable in
2483 * userland. If the pool is available for import in read-only
2484 * mode but not read-write mode, it is displayed as unavailable
2485 * in userland with a special note that the pool is actually
2486 * available for open in read-only mode.
2488 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2489 * missing a feature for write, we must first determine whether
2490 * the pool can be opened read-only before returning to
2491 * userland in order to know whether to display the
2492 * abovementioned note.
2494 if (missing_feat_read || (missing_feat_write &&
2495 spa_writeable(spa))) {
2496 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2501 * Load refcounts for ZFS features from disk into an in-memory
2502 * cache during SPA initialization.
2504 for (spa_feature_t i = 0; i < SPA_FEATURES; i++) {
2507 error = feature_get_refcount_from_disk(spa,
2508 &spa_feature_table[i], &refcount);
2510 spa->spa_feat_refcount_cache[i] = refcount;
2511 } else if (error == ENOTSUP) {
2512 spa->spa_feat_refcount_cache[i] =
2513 SPA_FEATURE_DISABLED;
2515 return (spa_vdev_err(rvd,
2516 VDEV_AUX_CORRUPT_DATA, EIO));
2521 if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) {
2522 if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG,
2523 &spa->spa_feat_enabled_txg_obj) != 0)
2524 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2527 spa->spa_is_initializing = B_TRUE;
2528 error = dsl_pool_open(spa->spa_dsl_pool);
2529 spa->spa_is_initializing = B_FALSE;
2531 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2535 nvlist_t *policy = NULL, *nvconfig;
2537 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2538 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2540 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2541 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2543 unsigned long myhostid = 0;
2545 VERIFY(nvlist_lookup_string(nvconfig,
2546 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2549 myhostid = zone_get_hostid(NULL);
2552 * We're emulating the system's hostid in userland, so
2553 * we can't use zone_get_hostid().
2555 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2556 #endif /* _KERNEL */
2557 if (check_hostid && hostid != 0 && myhostid != 0 &&
2558 hostid != myhostid) {
2559 nvlist_free(nvconfig);
2560 cmn_err(CE_WARN, "pool '%s' could not be "
2561 "loaded as it was last accessed by "
2562 "another system (host: %s hostid: 0x%lx). "
2563 "See: http://illumos.org/msg/ZFS-8000-EY",
2564 spa_name(spa), hostname,
2565 (unsigned long)hostid);
2566 return (SET_ERROR(EBADF));
2569 if (nvlist_lookup_nvlist(spa->spa_config,
2570 ZPOOL_REWIND_POLICY, &policy) == 0)
2571 VERIFY(nvlist_add_nvlist(nvconfig,
2572 ZPOOL_REWIND_POLICY, policy) == 0);
2574 spa_config_set(spa, nvconfig);
2576 spa_deactivate(spa);
2577 spa_activate(spa, orig_mode);
2579 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2582 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2583 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2584 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2586 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2589 * Load the bit that tells us to use the new accounting function
2590 * (raid-z deflation). If we have an older pool, this will not
2593 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2594 if (error != 0 && error != ENOENT)
2595 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2597 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2598 &spa->spa_creation_version);
2599 if (error != 0 && error != ENOENT)
2600 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2603 * Load the persistent error log. If we have an older pool, this will
2606 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2607 if (error != 0 && error != ENOENT)
2608 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2610 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2611 &spa->spa_errlog_scrub);
2612 if (error != 0 && error != ENOENT)
2613 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2616 * Load the history object. If we have an older pool, this
2617 * will not be present.
2619 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2620 if (error != 0 && error != ENOENT)
2621 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2624 * If we're assembling the pool from the split-off vdevs of
2625 * an existing pool, we don't want to attach the spares & cache
2630 * Load any hot spares for this pool.
2632 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2633 if (error != 0 && error != ENOENT)
2634 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2635 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2636 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2637 if (load_nvlist(spa, spa->spa_spares.sav_object,
2638 &spa->spa_spares.sav_config) != 0)
2639 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2641 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2642 spa_load_spares(spa);
2643 spa_config_exit(spa, SCL_ALL, FTAG);
2644 } else if (error == 0) {
2645 spa->spa_spares.sav_sync = B_TRUE;
2649 * Load any level 2 ARC devices for this pool.
2651 error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2652 &spa->spa_l2cache.sav_object);
2653 if (error != 0 && error != ENOENT)
2654 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2655 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2656 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2657 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2658 &spa->spa_l2cache.sav_config) != 0)
2659 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2661 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2662 spa_load_l2cache(spa);
2663 spa_config_exit(spa, SCL_ALL, FTAG);
2664 } else if (error == 0) {
2665 spa->spa_l2cache.sav_sync = B_TRUE;
2668 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2670 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2671 if (error && error != ENOENT)
2672 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2675 uint64_t autoreplace;
2677 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2678 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2679 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2680 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2681 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2682 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2683 &spa->spa_dedup_ditto);
2685 spa->spa_autoreplace = (autoreplace != 0);
2689 * If the 'autoreplace' property is set, then post a resource notifying
2690 * the ZFS DE that it should not issue any faults for unopenable
2691 * devices. We also iterate over the vdevs, and post a sysevent for any
2692 * unopenable vdevs so that the normal autoreplace handler can take
2695 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2696 spa_check_removed(spa->spa_root_vdev);
2698 * For the import case, this is done in spa_import(), because
2699 * at this point we're using the spare definitions from
2700 * the MOS config, not necessarily from the userland config.
2702 if (state != SPA_LOAD_IMPORT) {
2703 spa_aux_check_removed(&spa->spa_spares);
2704 spa_aux_check_removed(&spa->spa_l2cache);
2709 * Load the vdev state for all toplevel vdevs.
2714 * Propagate the leaf DTLs we just loaded all the way up the tree.
2716 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2717 vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2718 spa_config_exit(spa, SCL_ALL, FTAG);
2721 * Load the DDTs (dedup tables).
2723 error = ddt_load(spa);
2725 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2727 spa_update_dspace(spa);
2730 * Validate the config, using the MOS config to fill in any
2731 * information which might be missing. If we fail to validate
2732 * the config then declare the pool unfit for use. If we're
2733 * assembling a pool from a split, the log is not transferred
2736 if (type != SPA_IMPORT_ASSEMBLE) {
2739 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2740 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2742 if (!spa_config_valid(spa, nvconfig)) {
2743 nvlist_free(nvconfig);
2744 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2747 nvlist_free(nvconfig);
2750 * Now that we've validated the config, check the state of the
2751 * root vdev. If it can't be opened, it indicates one or
2752 * more toplevel vdevs are faulted.
2754 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2755 return (SET_ERROR(ENXIO));
2757 if (spa_writeable(spa) && spa_check_logs(spa)) {
2758 *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2759 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2763 if (missing_feat_write) {
2764 ASSERT(state == SPA_LOAD_TRYIMPORT);
2767 * At this point, we know that we can open the pool in
2768 * read-only mode but not read-write mode. We now have enough
2769 * information and can return to userland.
2771 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
2775 * We've successfully opened the pool, verify that we're ready
2776 * to start pushing transactions.
2778 if (state != SPA_LOAD_TRYIMPORT) {
2779 if (error = spa_load_verify(spa))
2780 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2784 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2785 spa->spa_load_max_txg == UINT64_MAX)) {
2787 int need_update = B_FALSE;
2788 dsl_pool_t *dp = spa_get_dsl(spa);
2790 ASSERT(state != SPA_LOAD_TRYIMPORT);
2793 * Claim log blocks that haven't been committed yet.
2794 * This must all happen in a single txg.
2795 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2796 * invoked from zil_claim_log_block()'s i/o done callback.
2797 * Price of rollback is that we abandon the log.
2799 spa->spa_claiming = B_TRUE;
2801 tx = dmu_tx_create_assigned(dp, spa_first_txg(spa));
2802 (void) dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
2803 zil_claim, tx, DS_FIND_CHILDREN);
2806 spa->spa_claiming = B_FALSE;
2808 spa_set_log_state(spa, SPA_LOG_GOOD);
2809 spa->spa_sync_on = B_TRUE;
2810 txg_sync_start(spa->spa_dsl_pool);
2813 * Wait for all claims to sync. We sync up to the highest
2814 * claimed log block birth time so that claimed log blocks
2815 * don't appear to be from the future. spa_claim_max_txg
2816 * will have been set for us by either zil_check_log_chain()
2817 * (invoked from spa_check_logs()) or zil_claim() above.
2819 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2822 * If the config cache is stale, or we have uninitialized
2823 * metaslabs (see spa_vdev_add()), then update the config.
2825 * If this is a verbatim import, trust the current
2826 * in-core spa_config and update the disk labels.
2828 if (config_cache_txg != spa->spa_config_txg ||
2829 state == SPA_LOAD_IMPORT ||
2830 state == SPA_LOAD_RECOVER ||
2831 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2832 need_update = B_TRUE;
2834 for (int c = 0; c < rvd->vdev_children; c++)
2835 if (rvd->vdev_child[c]->vdev_ms_array == 0)
2836 need_update = B_TRUE;
2839 * Update the config cache asychronously in case we're the
2840 * root pool, in which case the config cache isn't writable yet.
2843 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2846 * Check all DTLs to see if anything needs resilvering.
2848 if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2849 vdev_resilver_needed(rvd, NULL, NULL))
2850 spa_async_request(spa, SPA_ASYNC_RESILVER);
2853 * Log the fact that we booted up (so that we can detect if
2854 * we rebooted in the middle of an operation).
2856 spa_history_log_version(spa, "open");
2859 * Delete any inconsistent datasets.
2861 (void) dmu_objset_find(spa_name(spa),
2862 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2865 * Clean up any stale temporary dataset userrefs.
2867 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2874 spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2876 int mode = spa->spa_mode;
2879 spa_deactivate(spa);
2881 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg - 1;
2883 spa_activate(spa, mode);
2884 spa_async_suspend(spa);
2886 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2890 * If spa_load() fails this function will try loading prior txg's. If
2891 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2892 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2893 * function will not rewind the pool and will return the same error as
2897 spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2898 uint64_t max_request, int rewind_flags)
2900 nvlist_t *loadinfo = NULL;
2901 nvlist_t *config = NULL;
2902 int load_error, rewind_error;
2903 uint64_t safe_rewind_txg;
2906 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2907 spa->spa_load_max_txg = spa->spa_load_txg;
2908 spa_set_log_state(spa, SPA_LOG_CLEAR);
2910 spa->spa_load_max_txg = max_request;
2911 if (max_request != UINT64_MAX)
2912 spa->spa_extreme_rewind = B_TRUE;
2915 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2917 if (load_error == 0)
2920 if (spa->spa_root_vdev != NULL)
2921 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2923 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2924 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2926 if (rewind_flags & ZPOOL_NEVER_REWIND) {
2927 nvlist_free(config);
2928 return (load_error);
2931 if (state == SPA_LOAD_RECOVER) {
2932 /* Price of rolling back is discarding txgs, including log */
2933 spa_set_log_state(spa, SPA_LOG_CLEAR);
2936 * If we aren't rolling back save the load info from our first
2937 * import attempt so that we can restore it after attempting
2940 loadinfo = spa->spa_load_info;
2941 spa->spa_load_info = fnvlist_alloc();
2944 spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2945 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2946 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2947 TXG_INITIAL : safe_rewind_txg;
2950 * Continue as long as we're finding errors, we're still within
2951 * the acceptable rewind range, and we're still finding uberblocks
2953 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2954 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2955 if (spa->spa_load_max_txg < safe_rewind_txg)
2956 spa->spa_extreme_rewind = B_TRUE;
2957 rewind_error = spa_load_retry(spa, state, mosconfig);
2960 spa->spa_extreme_rewind = B_FALSE;
2961 spa->spa_load_max_txg = UINT64_MAX;
2963 if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2964 spa_config_set(spa, config);
2966 if (state == SPA_LOAD_RECOVER) {
2967 ASSERT3P(loadinfo, ==, NULL);
2968 return (rewind_error);
2970 /* Store the rewind info as part of the initial load info */
2971 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
2972 spa->spa_load_info);
2974 /* Restore the initial load info */
2975 fnvlist_free(spa->spa_load_info);
2976 spa->spa_load_info = loadinfo;
2978 return (load_error);
2985 * The import case is identical to an open except that the configuration is sent
2986 * down from userland, instead of grabbed from the configuration cache. For the
2987 * case of an open, the pool configuration will exist in the
2988 * POOL_STATE_UNINITIALIZED state.
2990 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2991 * the same time open the pool, without having to keep around the spa_t in some
2995 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2999 spa_load_state_t state = SPA_LOAD_OPEN;
3001 int locked = B_FALSE;
3002 int firstopen = B_FALSE;
3007 * As disgusting as this is, we need to support recursive calls to this
3008 * function because dsl_dir_open() is called during spa_load(), and ends
3009 * up calling spa_open() again. The real fix is to figure out how to
3010 * avoid dsl_dir_open() calling this in the first place.
3012 if (mutex_owner(&spa_namespace_lock) != curthread) {
3013 mutex_enter(&spa_namespace_lock);
3017 if ((spa = spa_lookup(pool)) == NULL) {
3019 mutex_exit(&spa_namespace_lock);
3020 return (SET_ERROR(ENOENT));
3023 if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
3024 zpool_rewind_policy_t policy;
3028 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
3030 if (policy.zrp_request & ZPOOL_DO_REWIND)
3031 state = SPA_LOAD_RECOVER;
3033 spa_activate(spa, spa_mode_global);
3035 if (state != SPA_LOAD_RECOVER)
3036 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3038 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
3039 policy.zrp_request);
3041 if (error == EBADF) {
3043 * If vdev_validate() returns failure (indicated by
3044 * EBADF), it indicates that one of the vdevs indicates
3045 * that the pool has been exported or destroyed. If
3046 * this is the case, the config cache is out of sync and
3047 * we should remove the pool from the namespace.
3050 spa_deactivate(spa);
3051 spa_config_sync(spa, B_TRUE, B_TRUE);
3054 mutex_exit(&spa_namespace_lock);
3055 return (SET_ERROR(ENOENT));
3060 * We can't open the pool, but we still have useful
3061 * information: the state of each vdev after the
3062 * attempted vdev_open(). Return this to the user.
3064 if (config != NULL && spa->spa_config) {
3065 VERIFY(nvlist_dup(spa->spa_config, config,
3067 VERIFY(nvlist_add_nvlist(*config,
3068 ZPOOL_CONFIG_LOAD_INFO,
3069 spa->spa_load_info) == 0);
3072 spa_deactivate(spa);
3073 spa->spa_last_open_failed = error;
3075 mutex_exit(&spa_namespace_lock);
3081 spa_open_ref(spa, tag);
3084 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
3087 * If we've recovered the pool, pass back any information we
3088 * gathered while doing the load.
3090 if (state == SPA_LOAD_RECOVER) {
3091 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
3092 spa->spa_load_info) == 0);
3096 spa->spa_last_open_failed = 0;
3097 spa->spa_last_ubsync_txg = 0;
3098 spa->spa_load_txg = 0;
3099 mutex_exit(&spa_namespace_lock);
3103 zvol_create_minors(spa->spa_name);
3114 spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
3117 return (spa_open_common(name, spapp, tag, policy, config));
3121 spa_open(const char *name, spa_t **spapp, void *tag)
3123 return (spa_open_common(name, spapp, tag, NULL, NULL));
3127 * Lookup the given spa_t, incrementing the inject count in the process,
3128 * preventing it from being exported or destroyed.
3131 spa_inject_addref(char *name)
3135 mutex_enter(&spa_namespace_lock);
3136 if ((spa = spa_lookup(name)) == NULL) {
3137 mutex_exit(&spa_namespace_lock);
3140 spa->spa_inject_ref++;
3141 mutex_exit(&spa_namespace_lock);
3147 spa_inject_delref(spa_t *spa)
3149 mutex_enter(&spa_namespace_lock);
3150 spa->spa_inject_ref--;
3151 mutex_exit(&spa_namespace_lock);
3155 * Add spares device information to the nvlist.
3158 spa_add_spares(spa_t *spa, nvlist_t *config)
3168 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3170 if (spa->spa_spares.sav_count == 0)
3173 VERIFY(nvlist_lookup_nvlist(config,
3174 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3175 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
3176 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3178 VERIFY(nvlist_add_nvlist_array(nvroot,
3179 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3180 VERIFY(nvlist_lookup_nvlist_array(nvroot,
3181 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3184 * Go through and find any spares which have since been
3185 * repurposed as an active spare. If this is the case, update
3186 * their status appropriately.
3188 for (i = 0; i < nspares; i++) {
3189 VERIFY(nvlist_lookup_uint64(spares[i],
3190 ZPOOL_CONFIG_GUID, &guid) == 0);
3191 if (spa_spare_exists(guid, &pool, NULL) &&
3193 VERIFY(nvlist_lookup_uint64_array(
3194 spares[i], ZPOOL_CONFIG_VDEV_STATS,
3195 (uint64_t **)&vs, &vsc) == 0);
3196 vs->vs_state = VDEV_STATE_CANT_OPEN;
3197 vs->vs_aux = VDEV_AUX_SPARED;
3204 * Add l2cache device information to the nvlist, including vdev stats.
3207 spa_add_l2cache(spa_t *spa, nvlist_t *config)
3210 uint_t i, j, nl2cache;
3217 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3219 if (spa->spa_l2cache.sav_count == 0)
3222 VERIFY(nvlist_lookup_nvlist(config,
3223 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3224 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
3225 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3226 if (nl2cache != 0) {
3227 VERIFY(nvlist_add_nvlist_array(nvroot,
3228 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3229 VERIFY(nvlist_lookup_nvlist_array(nvroot,
3230 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3233 * Update level 2 cache device stats.
3236 for (i = 0; i < nl2cache; i++) {
3237 VERIFY(nvlist_lookup_uint64(l2cache[i],
3238 ZPOOL_CONFIG_GUID, &guid) == 0);
3241 for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
3243 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
3244 vd = spa->spa_l2cache.sav_vdevs[j];
3250 VERIFY(nvlist_lookup_uint64_array(l2cache[i],
3251 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
3253 vdev_get_stats(vd, vs);
3259 spa_add_feature_stats(spa_t *spa, nvlist_t *config)
3265 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3266 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3268 /* We may be unable to read features if pool is suspended. */
3269 if (spa_suspended(spa))
3272 if (spa->spa_feat_for_read_obj != 0) {
3273 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3274 spa->spa_feat_for_read_obj);
3275 zap_cursor_retrieve(&zc, &za) == 0;
3276 zap_cursor_advance(&zc)) {
3277 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3278 za.za_num_integers == 1);
3279 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3280 za.za_first_integer));
3282 zap_cursor_fini(&zc);
3285 if (spa->spa_feat_for_write_obj != 0) {
3286 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3287 spa->spa_feat_for_write_obj);
3288 zap_cursor_retrieve(&zc, &za) == 0;
3289 zap_cursor_advance(&zc)) {
3290 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3291 za.za_num_integers == 1);
3292 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3293 za.za_first_integer));
3295 zap_cursor_fini(&zc);
3299 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
3301 nvlist_free(features);
3305 spa_get_stats(const char *name, nvlist_t **config,
3306 char *altroot, size_t buflen)
3312 error = spa_open_common(name, &spa, FTAG, NULL, config);
3316 * This still leaves a window of inconsistency where the spares
3317 * or l2cache devices could change and the config would be
3318 * self-inconsistent.
3320 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3322 if (*config != NULL) {
3323 uint64_t loadtimes[2];
3325 loadtimes[0] = spa->spa_loaded_ts.tv_sec;
3326 loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
3327 VERIFY(nvlist_add_uint64_array(*config,
3328 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
3330 VERIFY(nvlist_add_uint64(*config,
3331 ZPOOL_CONFIG_ERRCOUNT,
3332 spa_get_errlog_size(spa)) == 0);
3334 if (spa_suspended(spa))
3335 VERIFY(nvlist_add_uint64(*config,
3336 ZPOOL_CONFIG_SUSPENDED,
3337 spa->spa_failmode) == 0);
3339 spa_add_spares(spa, *config);
3340 spa_add_l2cache(spa, *config);
3341 spa_add_feature_stats(spa, *config);
3346 * We want to get the alternate root even for faulted pools, so we cheat
3347 * and call spa_lookup() directly.
3351 mutex_enter(&spa_namespace_lock);
3352 spa = spa_lookup(name);
3354 spa_altroot(spa, altroot, buflen);
3358 mutex_exit(&spa_namespace_lock);
3360 spa_altroot(spa, altroot, buflen);
3365 spa_config_exit(spa, SCL_CONFIG, FTAG);
3366 spa_close(spa, FTAG);
3373 * Validate that the auxiliary device array is well formed. We must have an
3374 * array of nvlists, each which describes a valid leaf vdev. If this is an
3375 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3376 * specified, as long as they are well-formed.
3379 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
3380 spa_aux_vdev_t *sav, const char *config, uint64_t version,
3381 vdev_labeltype_t label)
3388 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3391 * It's acceptable to have no devs specified.
3393 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
3397 return (SET_ERROR(EINVAL));
3400 * Make sure the pool is formatted with a version that supports this
3403 if (spa_version(spa) < version)
3404 return (SET_ERROR(ENOTSUP));
3407 * Set the pending device list so we correctly handle device in-use
3410 sav->sav_pending = dev;
3411 sav->sav_npending = ndev;
3413 for (i = 0; i < ndev; i++) {
3414 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
3418 if (!vd->vdev_ops->vdev_op_leaf) {
3420 error = SET_ERROR(EINVAL);
3425 * The L2ARC currently only supports disk devices in
3426 * kernel context. For user-level testing, we allow it.
3429 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
3430 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
3431 error = SET_ERROR(ENOTBLK);
3438 if ((error = vdev_open(vd)) == 0 &&
3439 (error = vdev_label_init(vd, crtxg, label)) == 0) {
3440 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
3441 vd->vdev_guid) == 0);
3447 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
3454 sav->sav_pending = NULL;
3455 sav->sav_npending = 0;
3460 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
3464 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3466 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3467 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
3468 VDEV_LABEL_SPARE)) != 0) {
3472 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3473 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
3474 VDEV_LABEL_L2CACHE));
3478 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
3483 if (sav->sav_config != NULL) {
3489 * Generate new dev list by concatentating with the
3492 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
3493 &olddevs, &oldndevs) == 0);
3495 newdevs = kmem_alloc(sizeof (void *) *
3496 (ndevs + oldndevs), KM_SLEEP);
3497 for (i = 0; i < oldndevs; i++)
3498 VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
3500 for (i = 0; i < ndevs; i++)
3501 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
3504 VERIFY(nvlist_remove(sav->sav_config, config,
3505 DATA_TYPE_NVLIST_ARRAY) == 0);
3507 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3508 config, newdevs, ndevs + oldndevs) == 0);
3509 for (i = 0; i < oldndevs + ndevs; i++)
3510 nvlist_free(newdevs[i]);
3511 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
3514 * Generate a new dev list.
3516 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
3518 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
3524 * Stop and drop level 2 ARC devices
3527 spa_l2cache_drop(spa_t *spa)
3531 spa_aux_vdev_t *sav = &spa->spa_l2cache;
3533 for (i = 0; i < sav->sav_count; i++) {
3536 vd = sav->sav_vdevs[i];
3539 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
3540 pool != 0ULL && l2arc_vdev_present(vd))
3541 l2arc_remove_vdev(vd);
3549 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
3553 char *altroot = NULL;
3558 uint64_t txg = TXG_INITIAL;
3559 nvlist_t **spares, **l2cache;
3560 uint_t nspares, nl2cache;
3561 uint64_t version, obj;
3562 boolean_t has_features;
3565 * If this pool already exists, return failure.
3567 mutex_enter(&spa_namespace_lock);
3568 if (spa_lookup(pool) != NULL) {
3569 mutex_exit(&spa_namespace_lock);
3570 return (SET_ERROR(EEXIST));
3574 * Allocate a new spa_t structure.
3576 (void) nvlist_lookup_string(props,
3577 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3578 spa = spa_add(pool, NULL, altroot);
3579 spa_activate(spa, spa_mode_global);
3581 if (props && (error = spa_prop_validate(spa, props))) {
3582 spa_deactivate(spa);
3584 mutex_exit(&spa_namespace_lock);
3588 has_features = B_FALSE;
3589 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
3590 elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
3591 if (zpool_prop_feature(nvpair_name(elem)))
3592 has_features = B_TRUE;
3595 if (has_features || nvlist_lookup_uint64(props,
3596 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
3597 version = SPA_VERSION;
3599 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
3601 spa->spa_first_txg = txg;
3602 spa->spa_uberblock.ub_txg = txg - 1;
3603 spa->spa_uberblock.ub_version = version;
3604 spa->spa_ubsync = spa->spa_uberblock;
3607 * Create "The Godfather" zio to hold all async IOs
3609 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
3611 for (int i = 0; i < max_ncpus; i++) {
3612 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
3613 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
3614 ZIO_FLAG_GODFATHER);
3618 * Create the root vdev.
3620 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3622 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3624 ASSERT(error != 0 || rvd != NULL);
3625 ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3627 if (error == 0 && !zfs_allocatable_devs(nvroot))
3628 error = SET_ERROR(EINVAL);
3631 (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3632 (error = spa_validate_aux(spa, nvroot, txg,
3633 VDEV_ALLOC_ADD)) == 0) {
3634 for (int c = 0; c < rvd->vdev_children; c++) {
3635 vdev_ashift_optimize(rvd->vdev_child[c]);
3636 vdev_metaslab_set_size(rvd->vdev_child[c]);
3637 vdev_expand(rvd->vdev_child[c], txg);
3641 spa_config_exit(spa, SCL_ALL, FTAG);
3645 spa_deactivate(spa);
3647 mutex_exit(&spa_namespace_lock);
3652 * Get the list of spares, if specified.
3654 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3655 &spares, &nspares) == 0) {
3656 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3658 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3659 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3660 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3661 spa_load_spares(spa);
3662 spa_config_exit(spa, SCL_ALL, FTAG);
3663 spa->spa_spares.sav_sync = B_TRUE;
3667 * Get the list of level 2 cache devices, if specified.
3669 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3670 &l2cache, &nl2cache) == 0) {
3671 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3672 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3673 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3674 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3675 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3676 spa_load_l2cache(spa);
3677 spa_config_exit(spa, SCL_ALL, FTAG);
3678 spa->spa_l2cache.sav_sync = B_TRUE;
3681 spa->spa_is_initializing = B_TRUE;
3682 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3683 spa->spa_meta_objset = dp->dp_meta_objset;
3684 spa->spa_is_initializing = B_FALSE;
3687 * Create DDTs (dedup tables).
3691 spa_update_dspace(spa);
3693 tx = dmu_tx_create_assigned(dp, txg);
3696 * Create the pool config object.
3698 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3699 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3700 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3702 if (zap_add(spa->spa_meta_objset,
3703 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3704 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3705 cmn_err(CE_PANIC, "failed to add pool config");
3708 if (spa_version(spa) >= SPA_VERSION_FEATURES)
3709 spa_feature_create_zap_objects(spa, tx);
3711 if (zap_add(spa->spa_meta_objset,
3712 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3713 sizeof (uint64_t), 1, &version, tx) != 0) {
3714 cmn_err(CE_PANIC, "failed to add pool version");
3717 /* Newly created pools with the right version are always deflated. */
3718 if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3719 spa->spa_deflate = TRUE;
3720 if (zap_add(spa->spa_meta_objset,
3721 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3722 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3723 cmn_err(CE_PANIC, "failed to add deflate");
3728 * Create the deferred-free bpobj. Turn off compression
3729 * because sync-to-convergence takes longer if the blocksize
3732 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3733 dmu_object_set_compress(spa->spa_meta_objset, obj,
3734 ZIO_COMPRESS_OFF, tx);
3735 if (zap_add(spa->spa_meta_objset,
3736 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3737 sizeof (uint64_t), 1, &obj, tx) != 0) {
3738 cmn_err(CE_PANIC, "failed to add bpobj");
3740 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3741 spa->spa_meta_objset, obj));
3744 * Create the pool's history object.
3746 if (version >= SPA_VERSION_ZPOOL_HISTORY)
3747 spa_history_create_obj(spa, tx);
3750 * Set pool properties.
3752 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3753 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3754 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3755 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3757 if (props != NULL) {
3758 spa_configfile_set(spa, props, B_FALSE);
3759 spa_sync_props(props, tx);
3764 spa->spa_sync_on = B_TRUE;
3765 txg_sync_start(spa->spa_dsl_pool);
3768 * We explicitly wait for the first transaction to complete so that our
3769 * bean counters are appropriately updated.
3771 txg_wait_synced(spa->spa_dsl_pool, txg);
3773 spa_config_sync(spa, B_FALSE, B_TRUE);
3774 spa_event_notify(spa, NULL, ESC_ZFS_POOL_CREATE);
3776 spa_history_log_version(spa, "create");
3779 * Don't count references from objsets that are already closed
3780 * and are making their way through the eviction process.
3782 spa_evicting_os_wait(spa);
3783 spa->spa_minref = refcount_count(&spa->spa_refcount);
3785 mutex_exit(&spa_namespace_lock);
3793 * Get the root pool information from the root disk, then import the root pool
3794 * during the system boot up time.
3796 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3799 spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3802 nvlist_t *nvtop, *nvroot;
3805 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3809 * Add this top-level vdev to the child array.
3811 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3813 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3815 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3818 * Put this pool's top-level vdevs into a root vdev.
3820 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3821 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3822 VDEV_TYPE_ROOT) == 0);
3823 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3824 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3825 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3829 * Replace the existing vdev_tree with the new root vdev in
3830 * this pool's configuration (remove the old, add the new).
3832 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3833 nvlist_free(nvroot);
3838 * Walk the vdev tree and see if we can find a device with "better"
3839 * configuration. A configuration is "better" if the label on that
3840 * device has a more recent txg.
3843 spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3845 for (int c = 0; c < vd->vdev_children; c++)
3846 spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3848 if (vd->vdev_ops->vdev_op_leaf) {
3852 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3856 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3860 * Do we have a better boot device?
3862 if (label_txg > *txg) {
3871 * Import a root pool.
3873 * For x86. devpath_list will consist of devid and/or physpath name of
3874 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3875 * The GRUB "findroot" command will return the vdev we should boot.
3877 * For Sparc, devpath_list consists the physpath name of the booting device
3878 * no matter the rootpool is a single device pool or a mirrored pool.
3880 * "/pci@1f,0/ide@d/disk@0,0:a"
3883 spa_import_rootpool(char *devpath, char *devid)
3886 vdev_t *rvd, *bvd, *avd = NULL;
3887 nvlist_t *config, *nvtop;
3893 * Read the label from the boot device and generate a configuration.
3895 config = spa_generate_rootconf(devpath, devid, &guid);
3896 #if defined(_OBP) && defined(_KERNEL)
3897 if (config == NULL) {
3898 if (strstr(devpath, "/iscsi/ssd") != NULL) {
3900 get_iscsi_bootpath_phy(devpath);
3901 config = spa_generate_rootconf(devpath, devid, &guid);
3905 if (config == NULL) {
3906 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
3908 return (SET_ERROR(EIO));
3911 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3913 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3915 mutex_enter(&spa_namespace_lock);
3916 if ((spa = spa_lookup(pname)) != NULL) {
3918 * Remove the existing root pool from the namespace so that we
3919 * can replace it with the correct config we just read in.
3924 spa = spa_add(pname, config, NULL);
3925 spa->spa_is_root = B_TRUE;
3926 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3929 * Build up a vdev tree based on the boot device's label config.
3931 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3933 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3934 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3935 VDEV_ALLOC_ROOTPOOL);
3936 spa_config_exit(spa, SCL_ALL, FTAG);
3938 mutex_exit(&spa_namespace_lock);
3939 nvlist_free(config);
3940 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3946 * Get the boot vdev.
3948 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3949 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3950 (u_longlong_t)guid);
3951 error = SET_ERROR(ENOENT);
3956 * Determine if there is a better boot device.
3959 spa_alt_rootvdev(rvd, &avd, &txg);
3961 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3962 "try booting from '%s'", avd->vdev_path);
3963 error = SET_ERROR(EINVAL);
3968 * If the boot device is part of a spare vdev then ensure that
3969 * we're booting off the active spare.
3971 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3972 !bvd->vdev_isspare) {
3973 cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3974 "try booting from '%s'",
3976 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3977 error = SET_ERROR(EINVAL);
3983 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3985 spa_config_exit(spa, SCL_ALL, FTAG);
3986 mutex_exit(&spa_namespace_lock);
3988 nvlist_free(config);
3994 extern int vdev_geom_read_pool_label(const char *name, nvlist_t ***configs,
3998 spa_generate_rootconf(const char *name)
4000 nvlist_t **configs, **tops;
4002 nvlist_t *best_cfg, *nvtop, *nvroot;
4011 if (vdev_geom_read_pool_label(name, &configs, &count) != 0)
4014 ASSERT3U(count, !=, 0);
4016 for (i = 0; i < count; i++) {
4019 VERIFY(nvlist_lookup_uint64(configs[i], ZPOOL_CONFIG_POOL_TXG,
4021 if (txg > best_txg) {
4023 best_cfg = configs[i];
4028 * Multi-vdev root pool configuration discovery is not supported yet.
4031 nvlist_lookup_uint64(best_cfg, ZPOOL_CONFIG_VDEV_CHILDREN, &nchildren);
4033 nvlist_lookup_uint64_array(best_cfg, ZPOOL_CONFIG_HOLE_ARRAY,
4036 tops = kmem_zalloc(nchildren * sizeof(void *), KM_SLEEP);
4037 for (i = 0; i < nchildren; i++) {
4040 if (configs[i] == NULL)
4042 VERIFY(nvlist_lookup_nvlist(configs[i], ZPOOL_CONFIG_VDEV_TREE,
4044 nvlist_dup(nvtop, &tops[i], KM_SLEEP);
4046 for (i = 0; holes != NULL && i < nholes; i++) {
4049 if (tops[holes[i]] != NULL)
4051 nvlist_alloc(&tops[holes[i]], NV_UNIQUE_NAME, KM_SLEEP);
4052 VERIFY(nvlist_add_string(tops[holes[i]], ZPOOL_CONFIG_TYPE,
4053 VDEV_TYPE_HOLE) == 0);
4054 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_ID,
4056 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_GUID,
4059 for (i = 0; i < nchildren; i++) {
4060 if (tops[i] != NULL)
4062 nvlist_alloc(&tops[i], NV_UNIQUE_NAME, KM_SLEEP);
4063 VERIFY(nvlist_add_string(tops[i], ZPOOL_CONFIG_TYPE,
4064 VDEV_TYPE_MISSING) == 0);
4065 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_ID,
4067 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_GUID,
4072 * Create pool config based on the best vdev config.
4074 nvlist_dup(best_cfg, &config, KM_SLEEP);
4077 * Put this pool's top-level vdevs into a root vdev.
4079 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4081 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4082 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
4083 VDEV_TYPE_ROOT) == 0);
4084 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
4085 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
4086 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
4087 tops, nchildren) == 0);
4090 * Replace the existing vdev_tree with the new root vdev in
4091 * this pool's configuration (remove the old, add the new).
4093 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
4096 * Drop vdev config elements that should not be present at pool level.
4098 nvlist_remove(config, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64);
4099 nvlist_remove(config, ZPOOL_CONFIG_TOP_GUID, DATA_TYPE_UINT64);
4101 for (i = 0; i < count; i++)
4102 nvlist_free(configs[i]);
4103 kmem_free(configs, count * sizeof(void *));
4104 for (i = 0; i < nchildren; i++)
4105 nvlist_free(tops[i]);
4106 kmem_free(tops, nchildren * sizeof(void *));
4107 nvlist_free(nvroot);
4112 spa_import_rootpool(const char *name)
4115 vdev_t *rvd, *bvd, *avd = NULL;
4116 nvlist_t *config, *nvtop;
4122 * Read the label from the boot device and generate a configuration.
4124 config = spa_generate_rootconf(name);
4126 mutex_enter(&spa_namespace_lock);
4127 if (config != NULL) {
4128 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
4129 &pname) == 0 && strcmp(name, pname) == 0);
4130 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg)
4133 if ((spa = spa_lookup(pname)) != NULL) {
4135 * Remove the existing root pool from the namespace so
4136 * that we can replace it with the correct config
4141 spa = spa_add(pname, config, NULL);
4144 * Set spa_ubsync.ub_version as it can be used in vdev_alloc()
4145 * via spa_version().
4147 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
4148 &spa->spa_ubsync.ub_version) != 0)
4149 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
4150 } else if ((spa = spa_lookup(name)) == NULL) {
4151 mutex_exit(&spa_namespace_lock);
4152 nvlist_free(config);
4153 cmn_err(CE_NOTE, "Cannot find the pool label for '%s'",
4157 VERIFY(nvlist_dup(spa->spa_config, &config, KM_SLEEP) == 0);
4159 spa->spa_is_root = B_TRUE;
4160 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
4163 * Build up a vdev tree based on the boot device's label config.
4165 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4167 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4168 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
4169 VDEV_ALLOC_ROOTPOOL);
4170 spa_config_exit(spa, SCL_ALL, FTAG);
4172 mutex_exit(&spa_namespace_lock);
4173 nvlist_free(config);
4174 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
4179 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4181 spa_config_exit(spa, SCL_ALL, FTAG);
4182 mutex_exit(&spa_namespace_lock);
4184 nvlist_free(config);
4192 * Import a non-root pool into the system.
4195 spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
4198 char *altroot = NULL;
4199 spa_load_state_t state = SPA_LOAD_IMPORT;
4200 zpool_rewind_policy_t policy;
4201 uint64_t mode = spa_mode_global;
4202 uint64_t readonly = B_FALSE;
4205 nvlist_t **spares, **l2cache;
4206 uint_t nspares, nl2cache;
4209 * If a pool with this name exists, return failure.
4211 mutex_enter(&spa_namespace_lock);
4212 if (spa_lookup(pool) != NULL) {
4213 mutex_exit(&spa_namespace_lock);
4214 return (SET_ERROR(EEXIST));
4218 * Create and initialize the spa structure.
4220 (void) nvlist_lookup_string(props,
4221 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4222 (void) nvlist_lookup_uint64(props,
4223 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
4226 spa = spa_add(pool, config, altroot);
4227 spa->spa_import_flags = flags;
4230 * Verbatim import - Take a pool and insert it into the namespace
4231 * as if it had been loaded at boot.
4233 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
4235 spa_configfile_set(spa, props, B_FALSE);
4237 spa_config_sync(spa, B_FALSE, B_TRUE);
4238 spa_event_notify(spa, NULL, ESC_ZFS_POOL_IMPORT);
4240 mutex_exit(&spa_namespace_lock);
4244 spa_activate(spa, mode);
4247 * Don't start async tasks until we know everything is healthy.
4249 spa_async_suspend(spa);
4251 zpool_get_rewind_policy(config, &policy);
4252 if (policy.zrp_request & ZPOOL_DO_REWIND)
4253 state = SPA_LOAD_RECOVER;
4256 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
4257 * because the user-supplied config is actually the one to trust when
4260 if (state != SPA_LOAD_RECOVER)
4261 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
4263 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
4264 policy.zrp_request);
4267 * Propagate anything learned while loading the pool and pass it
4268 * back to caller (i.e. rewind info, missing devices, etc).
4270 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4271 spa->spa_load_info) == 0);
4273 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4275 * Toss any existing sparelist, as it doesn't have any validity
4276 * anymore, and conflicts with spa_has_spare().
4278 if (spa->spa_spares.sav_config) {
4279 nvlist_free(spa->spa_spares.sav_config);
4280 spa->spa_spares.sav_config = NULL;
4281 spa_load_spares(spa);
4283 if (spa->spa_l2cache.sav_config) {
4284 nvlist_free(spa->spa_l2cache.sav_config);
4285 spa->spa_l2cache.sav_config = NULL;
4286 spa_load_l2cache(spa);
4289 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4292 error = spa_validate_aux(spa, nvroot, -1ULL,
4295 error = spa_validate_aux(spa, nvroot, -1ULL,
4296 VDEV_ALLOC_L2CACHE);
4297 spa_config_exit(spa, SCL_ALL, FTAG);
4300 spa_configfile_set(spa, props, B_FALSE);
4302 if (error != 0 || (props && spa_writeable(spa) &&
4303 (error = spa_prop_set(spa, props)))) {
4305 spa_deactivate(spa);
4307 mutex_exit(&spa_namespace_lock);
4311 spa_async_resume(spa);
4314 * Override any spares and level 2 cache devices as specified by
4315 * the user, as these may have correct device names/devids, etc.
4317 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
4318 &spares, &nspares) == 0) {
4319 if (spa->spa_spares.sav_config)
4320 VERIFY(nvlist_remove(spa->spa_spares.sav_config,
4321 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
4323 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
4324 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4325 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
4326 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
4327 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4328 spa_load_spares(spa);
4329 spa_config_exit(spa, SCL_ALL, FTAG);
4330 spa->spa_spares.sav_sync = B_TRUE;
4332 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
4333 &l2cache, &nl2cache) == 0) {
4334 if (spa->spa_l2cache.sav_config)
4335 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
4336 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
4338 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
4339 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4340 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
4341 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
4342 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4343 spa_load_l2cache(spa);
4344 spa_config_exit(spa, SCL_ALL, FTAG);
4345 spa->spa_l2cache.sav_sync = B_TRUE;
4349 * Check for any removed devices.
4351 if (spa->spa_autoreplace) {
4352 spa_aux_check_removed(&spa->spa_spares);
4353 spa_aux_check_removed(&spa->spa_l2cache);
4356 if (spa_writeable(spa)) {
4358 * Update the config cache to include the newly-imported pool.
4360 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4364 * It's possible that the pool was expanded while it was exported.
4365 * We kick off an async task to handle this for us.
4367 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
4369 spa_history_log_version(spa, "import");
4371 spa_event_notify(spa, NULL, ESC_ZFS_POOL_IMPORT);
4373 mutex_exit(&spa_namespace_lock);
4377 zvol_create_minors(pool);
4384 spa_tryimport(nvlist_t *tryconfig)
4386 nvlist_t *config = NULL;
4392 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
4395 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
4399 * Create and initialize the spa structure.
4401 mutex_enter(&spa_namespace_lock);
4402 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
4403 spa_activate(spa, FREAD);
4406 * Pass off the heavy lifting to spa_load().
4407 * Pass TRUE for mosconfig because the user-supplied config
4408 * is actually the one to trust when doing an import.
4410 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
4413 * If 'tryconfig' was at least parsable, return the current config.
4415 if (spa->spa_root_vdev != NULL) {
4416 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4417 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
4419 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4421 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
4422 spa->spa_uberblock.ub_timestamp) == 0);
4423 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4424 spa->spa_load_info) == 0);
4427 * If the bootfs property exists on this pool then we
4428 * copy it out so that external consumers can tell which
4429 * pools are bootable.
4431 if ((!error || error == EEXIST) && spa->spa_bootfs) {
4432 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4435 * We have to play games with the name since the
4436 * pool was opened as TRYIMPORT_NAME.
4438 if (dsl_dsobj_to_dsname(spa_name(spa),
4439 spa->spa_bootfs, tmpname) == 0) {
4441 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4443 cp = strchr(tmpname, '/');
4445 (void) strlcpy(dsname, tmpname,
4448 (void) snprintf(dsname, MAXPATHLEN,
4449 "%s/%s", poolname, ++cp);
4451 VERIFY(nvlist_add_string(config,
4452 ZPOOL_CONFIG_BOOTFS, dsname) == 0);
4453 kmem_free(dsname, MAXPATHLEN);
4455 kmem_free(tmpname, MAXPATHLEN);
4459 * Add the list of hot spares and level 2 cache devices.
4461 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4462 spa_add_spares(spa, config);
4463 spa_add_l2cache(spa, config);
4464 spa_config_exit(spa, SCL_CONFIG, FTAG);
4468 spa_deactivate(spa);
4470 mutex_exit(&spa_namespace_lock);
4476 * Pool export/destroy
4478 * The act of destroying or exporting a pool is very simple. We make sure there
4479 * is no more pending I/O and any references to the pool are gone. Then, we
4480 * update the pool state and sync all the labels to disk, removing the
4481 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4482 * we don't sync the labels or remove the configuration cache.
4485 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
4486 boolean_t force, boolean_t hardforce)
4493 if (!(spa_mode_global & FWRITE))
4494 return (SET_ERROR(EROFS));
4496 mutex_enter(&spa_namespace_lock);
4497 if ((spa = spa_lookup(pool)) == NULL) {
4498 mutex_exit(&spa_namespace_lock);
4499 return (SET_ERROR(ENOENT));
4503 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4504 * reacquire the namespace lock, and see if we can export.
4506 spa_open_ref(spa, FTAG);
4507 mutex_exit(&spa_namespace_lock);
4508 spa_async_suspend(spa);
4509 mutex_enter(&spa_namespace_lock);
4510 spa_close(spa, FTAG);
4513 * The pool will be in core if it's openable,
4514 * in which case we can modify its state.
4516 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
4518 * Objsets may be open only because they're dirty, so we
4519 * have to force it to sync before checking spa_refcnt.
4521 txg_wait_synced(spa->spa_dsl_pool, 0);
4522 spa_evicting_os_wait(spa);
4525 * A pool cannot be exported or destroyed if there are active
4526 * references. If we are resetting a pool, allow references by
4527 * fault injection handlers.
4529 if (!spa_refcount_zero(spa) ||
4530 (spa->spa_inject_ref != 0 &&
4531 new_state != POOL_STATE_UNINITIALIZED)) {
4532 spa_async_resume(spa);
4533 mutex_exit(&spa_namespace_lock);
4534 return (SET_ERROR(EBUSY));
4538 * A pool cannot be exported if it has an active shared spare.
4539 * This is to prevent other pools stealing the active spare
4540 * from an exported pool. At user's own will, such pool can
4541 * be forcedly exported.
4543 if (!force && new_state == POOL_STATE_EXPORTED &&
4544 spa_has_active_shared_spare(spa)) {
4545 spa_async_resume(spa);
4546 mutex_exit(&spa_namespace_lock);
4547 return (SET_ERROR(EXDEV));
4551 * We want this to be reflected on every label,
4552 * so mark them all dirty. spa_unload() will do the
4553 * final sync that pushes these changes out.
4555 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
4556 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4557 spa->spa_state = new_state;
4558 spa->spa_final_txg = spa_last_synced_txg(spa) +
4560 vdev_config_dirty(spa->spa_root_vdev);
4561 spa_config_exit(spa, SCL_ALL, FTAG);
4565 spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
4567 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4569 spa_deactivate(spa);
4572 if (oldconfig && spa->spa_config)
4573 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
4575 if (new_state != POOL_STATE_UNINITIALIZED) {
4577 spa_config_sync(spa, B_TRUE, B_TRUE);
4580 mutex_exit(&spa_namespace_lock);
4586 * Destroy a storage pool.
4589 spa_destroy(char *pool)
4591 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
4596 * Export a storage pool.
4599 spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
4600 boolean_t hardforce)
4602 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
4607 * Similar to spa_export(), this unloads the spa_t without actually removing it
4608 * from the namespace in any way.
4611 spa_reset(char *pool)
4613 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
4618 * ==========================================================================
4619 * Device manipulation
4620 * ==========================================================================
4624 * Add a device to a storage pool.
4627 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
4631 vdev_t *rvd = spa->spa_root_vdev;
4633 nvlist_t **spares, **l2cache;
4634 uint_t nspares, nl2cache;
4636 ASSERT(spa_writeable(spa));
4638 txg = spa_vdev_enter(spa);
4640 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
4641 VDEV_ALLOC_ADD)) != 0)
4642 return (spa_vdev_exit(spa, NULL, txg, error));
4644 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */
4646 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
4650 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
4654 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
4655 return (spa_vdev_exit(spa, vd, txg, EINVAL));
4657 if (vd->vdev_children != 0 &&
4658 (error = vdev_create(vd, txg, B_FALSE)) != 0)
4659 return (spa_vdev_exit(spa, vd, txg, error));
4662 * We must validate the spares and l2cache devices after checking the
4663 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
4665 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
4666 return (spa_vdev_exit(spa, vd, txg, error));
4669 * Transfer each new top-level vdev from vd to rvd.
4671 for (int c = 0; c < vd->vdev_children; c++) {
4674 * Set the vdev id to the first hole, if one exists.
4676 for (id = 0; id < rvd->vdev_children; id++) {
4677 if (rvd->vdev_child[id]->vdev_ishole) {
4678 vdev_free(rvd->vdev_child[id]);
4682 tvd = vd->vdev_child[c];
4683 vdev_remove_child(vd, tvd);
4685 vdev_add_child(rvd, tvd);
4686 vdev_config_dirty(tvd);
4690 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
4691 ZPOOL_CONFIG_SPARES);
4692 spa_load_spares(spa);
4693 spa->spa_spares.sav_sync = B_TRUE;
4696 if (nl2cache != 0) {
4697 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
4698 ZPOOL_CONFIG_L2CACHE);
4699 spa_load_l2cache(spa);
4700 spa->spa_l2cache.sav_sync = B_TRUE;
4704 * We have to be careful when adding new vdevs to an existing pool.
4705 * If other threads start allocating from these vdevs before we
4706 * sync the config cache, and we lose power, then upon reboot we may
4707 * fail to open the pool because there are DVAs that the config cache
4708 * can't translate. Therefore, we first add the vdevs without
4709 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4710 * and then let spa_config_update() initialize the new metaslabs.
4712 * spa_load() checks for added-but-not-initialized vdevs, so that
4713 * if we lose power at any point in this sequence, the remaining
4714 * steps will be completed the next time we load the pool.
4716 (void) spa_vdev_exit(spa, vd, txg, 0);
4718 mutex_enter(&spa_namespace_lock);
4719 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4720 spa_event_notify(spa, NULL, ESC_ZFS_VDEV_ADD);
4721 mutex_exit(&spa_namespace_lock);
4727 * Attach a device to a mirror. The arguments are the path to any device
4728 * in the mirror, and the nvroot for the new device. If the path specifies
4729 * a device that is not mirrored, we automatically insert the mirror vdev.
4731 * If 'replacing' is specified, the new device is intended to replace the
4732 * existing device; in this case the two devices are made into their own
4733 * mirror using the 'replacing' vdev, which is functionally identical to
4734 * the mirror vdev (it actually reuses all the same ops) but has a few
4735 * extra rules: you can't attach to it after it's been created, and upon
4736 * completion of resilvering, the first disk (the one being replaced)
4737 * is automatically detached.
4740 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
4742 uint64_t txg, dtl_max_txg;
4743 vdev_t *rvd = spa->spa_root_vdev;
4744 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
4746 char *oldvdpath, *newvdpath;
4750 ASSERT(spa_writeable(spa));
4752 txg = spa_vdev_enter(spa);
4754 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
4757 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4759 if (!oldvd->vdev_ops->vdev_op_leaf)
4760 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4762 pvd = oldvd->vdev_parent;
4764 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
4765 VDEV_ALLOC_ATTACH)) != 0)
4766 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4768 if (newrootvd->vdev_children != 1)
4769 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4771 newvd = newrootvd->vdev_child[0];
4773 if (!newvd->vdev_ops->vdev_op_leaf)
4774 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4776 if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
4777 return (spa_vdev_exit(spa, newrootvd, txg, error));
4780 * Spares can't replace logs
4782 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
4783 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4787 * For attach, the only allowable parent is a mirror or the root
4790 if (pvd->vdev_ops != &vdev_mirror_ops &&
4791 pvd->vdev_ops != &vdev_root_ops)
4792 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4794 pvops = &vdev_mirror_ops;
4797 * Active hot spares can only be replaced by inactive hot
4800 if (pvd->vdev_ops == &vdev_spare_ops &&
4801 oldvd->vdev_isspare &&
4802 !spa_has_spare(spa, newvd->vdev_guid))
4803 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4806 * If the source is a hot spare, and the parent isn't already a
4807 * spare, then we want to create a new hot spare. Otherwise, we
4808 * want to create a replacing vdev. The user is not allowed to
4809 * attach to a spared vdev child unless the 'isspare' state is
4810 * the same (spare replaces spare, non-spare replaces
4813 if (pvd->vdev_ops == &vdev_replacing_ops &&
4814 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
4815 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4816 } else if (pvd->vdev_ops == &vdev_spare_ops &&
4817 newvd->vdev_isspare != oldvd->vdev_isspare) {
4818 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4821 if (newvd->vdev_isspare)
4822 pvops = &vdev_spare_ops;
4824 pvops = &vdev_replacing_ops;
4828 * Make sure the new device is big enough.
4830 if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
4831 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
4834 * The new device cannot have a higher alignment requirement
4835 * than the top-level vdev.
4837 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4838 return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4841 * If this is an in-place replacement, update oldvd's path and devid
4842 * to make it distinguishable from newvd, and unopenable from now on.
4844 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4845 spa_strfree(oldvd->vdev_path);
4846 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4848 (void) sprintf(oldvd->vdev_path, "%s/%s",
4849 newvd->vdev_path, "old");
4850 if (oldvd->vdev_devid != NULL) {
4851 spa_strfree(oldvd->vdev_devid);
4852 oldvd->vdev_devid = NULL;
4856 /* mark the device being resilvered */
4857 newvd->vdev_resilver_txg = txg;
4860 * If the parent is not a mirror, or if we're replacing, insert the new
4861 * mirror/replacing/spare vdev above oldvd.
4863 if (pvd->vdev_ops != pvops)
4864 pvd = vdev_add_parent(oldvd, pvops);
4866 ASSERT(pvd->vdev_top->vdev_parent == rvd);
4867 ASSERT(pvd->vdev_ops == pvops);
4868 ASSERT(oldvd->vdev_parent == pvd);
4871 * Extract the new device from its root and add it to pvd.
4873 vdev_remove_child(newrootvd, newvd);
4874 newvd->vdev_id = pvd->vdev_children;
4875 newvd->vdev_crtxg = oldvd->vdev_crtxg;
4876 vdev_add_child(pvd, newvd);
4878 tvd = newvd->vdev_top;
4879 ASSERT(pvd->vdev_top == tvd);
4880 ASSERT(tvd->vdev_parent == rvd);
4882 vdev_config_dirty(tvd);
4885 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4886 * for any dmu_sync-ed blocks. It will propagate upward when
4887 * spa_vdev_exit() calls vdev_dtl_reassess().
4889 dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4891 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4892 dtl_max_txg - TXG_INITIAL);
4894 if (newvd->vdev_isspare) {
4895 spa_spare_activate(newvd);
4896 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
4899 oldvdpath = spa_strdup(oldvd->vdev_path);
4900 newvdpath = spa_strdup(newvd->vdev_path);
4901 newvd_isspare = newvd->vdev_isspare;
4904 * Mark newvd's DTL dirty in this txg.
4906 vdev_dirty(tvd, VDD_DTL, newvd, txg);
4909 * Schedule the resilver to restart in the future. We do this to
4910 * ensure that dmu_sync-ed blocks have been stitched into the
4911 * respective datasets.
4913 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4915 if (spa->spa_bootfs)
4916 spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
4918 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_ATTACH);
4923 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4925 spa_history_log_internal(spa, "vdev attach", NULL,
4926 "%s vdev=%s %s vdev=%s",
4927 replacing && newvd_isspare ? "spare in" :
4928 replacing ? "replace" : "attach", newvdpath,
4929 replacing ? "for" : "to", oldvdpath);
4931 spa_strfree(oldvdpath);
4932 spa_strfree(newvdpath);
4938 * Detach a device from a mirror or replacing vdev.
4940 * If 'replace_done' is specified, only detach if the parent
4941 * is a replacing vdev.
4944 spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4948 vdev_t *rvd = spa->spa_root_vdev;
4949 vdev_t *vd, *pvd, *cvd, *tvd;
4950 boolean_t unspare = B_FALSE;
4951 uint64_t unspare_guid = 0;
4954 ASSERT(spa_writeable(spa));
4956 txg = spa_vdev_enter(spa);
4958 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4961 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4963 if (!vd->vdev_ops->vdev_op_leaf)
4964 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4966 pvd = vd->vdev_parent;
4969 * If the parent/child relationship is not as expected, don't do it.
4970 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4971 * vdev that's replacing B with C. The user's intent in replacing
4972 * is to go from M(A,B) to M(A,C). If the user decides to cancel
4973 * the replace by detaching C, the expected behavior is to end up
4974 * M(A,B). But suppose that right after deciding to detach C,
4975 * the replacement of B completes. We would have M(A,C), and then
4976 * ask to detach C, which would leave us with just A -- not what
4977 * the user wanted. To prevent this, we make sure that the
4978 * parent/child relationship hasn't changed -- in this example,
4979 * that C's parent is still the replacing vdev R.
4981 if (pvd->vdev_guid != pguid && pguid != 0)
4982 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4985 * Only 'replacing' or 'spare' vdevs can be replaced.
4987 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4988 pvd->vdev_ops != &vdev_spare_ops)
4989 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4991 ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4992 spa_version(spa) >= SPA_VERSION_SPARES);
4995 * Only mirror, replacing, and spare vdevs support detach.
4997 if (pvd->vdev_ops != &vdev_replacing_ops &&
4998 pvd->vdev_ops != &vdev_mirror_ops &&
4999 pvd->vdev_ops != &vdev_spare_ops)
5000 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
5003 * If this device has the only valid copy of some data,
5004 * we cannot safely detach it.
5006 if (vdev_dtl_required(vd))
5007 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
5009 ASSERT(pvd->vdev_children >= 2);
5012 * If we are detaching the second disk from a replacing vdev, then
5013 * check to see if we changed the original vdev's path to have "/old"
5014 * at the end in spa_vdev_attach(). If so, undo that change now.
5016 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
5017 vd->vdev_path != NULL) {
5018 size_t len = strlen(vd->vdev_path);
5020 for (int c = 0; c < pvd->vdev_children; c++) {
5021 cvd = pvd->vdev_child[c];
5023 if (cvd == vd || cvd->vdev_path == NULL)
5026 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
5027 strcmp(cvd->vdev_path + len, "/old") == 0) {
5028 spa_strfree(cvd->vdev_path);
5029 cvd->vdev_path = spa_strdup(vd->vdev_path);
5036 * If we are detaching the original disk from a spare, then it implies
5037 * that the spare should become a real disk, and be removed from the
5038 * active spare list for the pool.
5040 if (pvd->vdev_ops == &vdev_spare_ops &&
5042 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
5046 * Erase the disk labels so the disk can be used for other things.
5047 * This must be done after all other error cases are handled,
5048 * but before we disembowel vd (so we can still do I/O to it).
5049 * But if we can't do it, don't treat the error as fatal --
5050 * it may be that the unwritability of the disk is the reason
5051 * it's being detached!
5053 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5056 * Remove vd from its parent and compact the parent's children.
5058 vdev_remove_child(pvd, vd);
5059 vdev_compact_children(pvd);
5062 * Remember one of the remaining children so we can get tvd below.
5064 cvd = pvd->vdev_child[pvd->vdev_children - 1];
5067 * If we need to remove the remaining child from the list of hot spares,
5068 * do it now, marking the vdev as no longer a spare in the process.
5069 * We must do this before vdev_remove_parent(), because that can
5070 * change the GUID if it creates a new toplevel GUID. For a similar
5071 * reason, we must remove the spare now, in the same txg as the detach;
5072 * otherwise someone could attach a new sibling, change the GUID, and
5073 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
5076 ASSERT(cvd->vdev_isspare);
5077 spa_spare_remove(cvd);
5078 unspare_guid = cvd->vdev_guid;
5079 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
5080 cvd->vdev_unspare = B_TRUE;
5084 * If the parent mirror/replacing vdev only has one child,
5085 * the parent is no longer needed. Remove it from the tree.
5087 if (pvd->vdev_children == 1) {
5088 if (pvd->vdev_ops == &vdev_spare_ops)
5089 cvd->vdev_unspare = B_FALSE;
5090 vdev_remove_parent(cvd);
5095 * We don't set tvd until now because the parent we just removed
5096 * may have been the previous top-level vdev.
5098 tvd = cvd->vdev_top;
5099 ASSERT(tvd->vdev_parent == rvd);
5102 * Reevaluate the parent vdev state.
5104 vdev_propagate_state(cvd);
5107 * If the 'autoexpand' property is set on the pool then automatically
5108 * try to expand the size of the pool. For example if the device we
5109 * just detached was smaller than the others, it may be possible to
5110 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
5111 * first so that we can obtain the updated sizes of the leaf vdevs.
5113 if (spa->spa_autoexpand) {
5115 vdev_expand(tvd, txg);
5118 vdev_config_dirty(tvd);
5121 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
5122 * vd->vdev_detached is set and free vd's DTL object in syncing context.
5123 * But first make sure we're not on any *other* txg's DTL list, to
5124 * prevent vd from being accessed after it's freed.
5126 vdpath = spa_strdup(vd->vdev_path);
5127 for (int t = 0; t < TXG_SIZE; t++)
5128 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
5129 vd->vdev_detached = B_TRUE;
5130 vdev_dirty(tvd, VDD_DTL, vd, txg);
5132 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
5134 /* hang on to the spa before we release the lock */
5135 spa_open_ref(spa, FTAG);
5137 error = spa_vdev_exit(spa, vd, txg, 0);
5139 spa_history_log_internal(spa, "detach", NULL,
5141 spa_strfree(vdpath);
5144 * If this was the removal of the original device in a hot spare vdev,
5145 * then we want to go through and remove the device from the hot spare
5146 * list of every other pool.
5149 spa_t *altspa = NULL;
5151 mutex_enter(&spa_namespace_lock);
5152 while ((altspa = spa_next(altspa)) != NULL) {
5153 if (altspa->spa_state != POOL_STATE_ACTIVE ||
5157 spa_open_ref(altspa, FTAG);
5158 mutex_exit(&spa_namespace_lock);
5159 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
5160 mutex_enter(&spa_namespace_lock);
5161 spa_close(altspa, FTAG);
5163 mutex_exit(&spa_namespace_lock);
5165 /* search the rest of the vdevs for spares to remove */
5166 spa_vdev_resilver_done(spa);
5169 /* all done with the spa; OK to release */
5170 mutex_enter(&spa_namespace_lock);
5171 spa_close(spa, FTAG);
5172 mutex_exit(&spa_namespace_lock);
5178 * Split a set of devices from their mirrors, and create a new pool from them.
5181 spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
5182 nvlist_t *props, boolean_t exp)
5185 uint64_t txg, *glist;
5187 uint_t c, children, lastlog;
5188 nvlist_t **child, *nvl, *tmp;
5190 char *altroot = NULL;
5191 vdev_t *rvd, **vml = NULL; /* vdev modify list */
5192 boolean_t activate_slog;
5194 ASSERT(spa_writeable(spa));
5196 txg = spa_vdev_enter(spa);
5198 /* clear the log and flush everything up to now */
5199 activate_slog = spa_passivate_log(spa);
5200 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5201 error = spa_offline_log(spa);
5202 txg = spa_vdev_config_enter(spa);
5205 spa_activate_log(spa);
5208 return (spa_vdev_exit(spa, NULL, txg, error));
5210 /* check new spa name before going any further */
5211 if (spa_lookup(newname) != NULL)
5212 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
5215 * scan through all the children to ensure they're all mirrors
5217 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
5218 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
5220 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5222 /* first, check to ensure we've got the right child count */
5223 rvd = spa->spa_root_vdev;
5225 for (c = 0; c < rvd->vdev_children; c++) {
5226 vdev_t *vd = rvd->vdev_child[c];
5228 /* don't count the holes & logs as children */
5229 if (vd->vdev_islog || vd->vdev_ishole) {
5237 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
5238 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5240 /* next, ensure no spare or cache devices are part of the split */
5241 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
5242 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
5243 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5245 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
5246 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
5248 /* then, loop over each vdev and validate it */
5249 for (c = 0; c < children; c++) {
5250 uint64_t is_hole = 0;
5252 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
5256 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
5257 spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
5260 error = SET_ERROR(EINVAL);
5265 /* which disk is going to be split? */
5266 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
5268 error = SET_ERROR(EINVAL);
5272 /* look it up in the spa */
5273 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
5274 if (vml[c] == NULL) {
5275 error = SET_ERROR(ENODEV);
5279 /* make sure there's nothing stopping the split */
5280 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
5281 vml[c]->vdev_islog ||
5282 vml[c]->vdev_ishole ||
5283 vml[c]->vdev_isspare ||
5284 vml[c]->vdev_isl2cache ||
5285 !vdev_writeable(vml[c]) ||
5286 vml[c]->vdev_children != 0 ||
5287 vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
5288 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
5289 error = SET_ERROR(EINVAL);
5293 if (vdev_dtl_required(vml[c])) {
5294 error = SET_ERROR(EBUSY);
5298 /* we need certain info from the top level */
5299 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
5300 vml[c]->vdev_top->vdev_ms_array) == 0);
5301 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
5302 vml[c]->vdev_top->vdev_ms_shift) == 0);
5303 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
5304 vml[c]->vdev_top->vdev_asize) == 0);
5305 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
5306 vml[c]->vdev_top->vdev_ashift) == 0);
5310 kmem_free(vml, children * sizeof (vdev_t *));
5311 kmem_free(glist, children * sizeof (uint64_t));
5312 return (spa_vdev_exit(spa, NULL, txg, error));
5315 /* stop writers from using the disks */
5316 for (c = 0; c < children; c++) {
5318 vml[c]->vdev_offline = B_TRUE;
5320 vdev_reopen(spa->spa_root_vdev);
5323 * Temporarily record the splitting vdevs in the spa config. This
5324 * will disappear once the config is regenerated.
5326 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5327 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
5328 glist, children) == 0);
5329 kmem_free(glist, children * sizeof (uint64_t));
5331 mutex_enter(&spa->spa_props_lock);
5332 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
5334 mutex_exit(&spa->spa_props_lock);
5335 spa->spa_config_splitting = nvl;
5336 vdev_config_dirty(spa->spa_root_vdev);
5338 /* configure and create the new pool */
5339 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
5340 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
5341 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
5342 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5343 spa_version(spa)) == 0);
5344 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
5345 spa->spa_config_txg) == 0);
5346 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
5347 spa_generate_guid(NULL)) == 0);
5348 (void) nvlist_lookup_string(props,
5349 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
5351 /* add the new pool to the namespace */
5352 newspa = spa_add(newname, config, altroot);
5353 newspa->spa_config_txg = spa->spa_config_txg;
5354 spa_set_log_state(newspa, SPA_LOG_CLEAR);
5356 /* release the spa config lock, retaining the namespace lock */
5357 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5359 if (zio_injection_enabled)
5360 zio_handle_panic_injection(spa, FTAG, 1);
5362 spa_activate(newspa, spa_mode_global);
5363 spa_async_suspend(newspa);
5366 /* mark that we are creating new spa by splitting */
5367 newspa->spa_splitting_newspa = B_TRUE;
5369 /* create the new pool from the disks of the original pool */
5370 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
5372 newspa->spa_splitting_newspa = B_FALSE;
5377 /* if that worked, generate a real config for the new pool */
5378 if (newspa->spa_root_vdev != NULL) {
5379 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
5380 NV_UNIQUE_NAME, KM_SLEEP) == 0);
5381 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
5382 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
5383 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
5388 if (props != NULL) {
5389 spa_configfile_set(newspa, props, B_FALSE);
5390 error = spa_prop_set(newspa, props);
5395 /* flush everything */
5396 txg = spa_vdev_config_enter(newspa);
5397 vdev_config_dirty(newspa->spa_root_vdev);
5398 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
5400 if (zio_injection_enabled)
5401 zio_handle_panic_injection(spa, FTAG, 2);
5403 spa_async_resume(newspa);
5405 /* finally, update the original pool's config */
5406 txg = spa_vdev_config_enter(spa);
5407 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
5408 error = dmu_tx_assign(tx, TXG_WAIT);
5411 for (c = 0; c < children; c++) {
5412 if (vml[c] != NULL) {
5415 spa_history_log_internal(spa, "detach", tx,
5416 "vdev=%s", vml[c]->vdev_path);
5420 vdev_config_dirty(spa->spa_root_vdev);
5421 spa->spa_config_splitting = NULL;
5425 (void) spa_vdev_exit(spa, NULL, txg, 0);
5427 if (zio_injection_enabled)
5428 zio_handle_panic_injection(spa, FTAG, 3);
5430 /* split is complete; log a history record */
5431 spa_history_log_internal(newspa, "split", NULL,
5432 "from pool %s", spa_name(spa));
5434 kmem_free(vml, children * sizeof (vdev_t *));
5436 /* if we're not going to mount the filesystems in userland, export */
5438 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
5445 spa_deactivate(newspa);
5448 txg = spa_vdev_config_enter(spa);
5450 /* re-online all offlined disks */
5451 for (c = 0; c < children; c++) {
5453 vml[c]->vdev_offline = B_FALSE;
5455 vdev_reopen(spa->spa_root_vdev);
5457 nvlist_free(spa->spa_config_splitting);
5458 spa->spa_config_splitting = NULL;
5459 (void) spa_vdev_exit(spa, NULL, txg, error);
5461 kmem_free(vml, children * sizeof (vdev_t *));
5466 spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
5468 for (int i = 0; i < count; i++) {
5471 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
5474 if (guid == target_guid)
5482 spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
5483 nvlist_t *dev_to_remove)
5485 nvlist_t **newdev = NULL;
5488 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
5490 for (int i = 0, j = 0; i < count; i++) {
5491 if (dev[i] == dev_to_remove)
5493 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
5496 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
5497 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
5499 for (int i = 0; i < count - 1; i++)
5500 nvlist_free(newdev[i]);
5503 kmem_free(newdev, (count - 1) * sizeof (void *));
5507 * Evacuate the device.
5510 spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
5515 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5516 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5517 ASSERT(vd == vd->vdev_top);
5520 * Evacuate the device. We don't hold the config lock as writer
5521 * since we need to do I/O but we do keep the
5522 * spa_namespace_lock held. Once this completes the device
5523 * should no longer have any blocks allocated on it.
5525 if (vd->vdev_islog) {
5526 if (vd->vdev_stat.vs_alloc != 0)
5527 error = spa_offline_log(spa);
5529 error = SET_ERROR(ENOTSUP);
5536 * The evacuation succeeded. Remove any remaining MOS metadata
5537 * associated with this vdev, and wait for these changes to sync.
5539 ASSERT0(vd->vdev_stat.vs_alloc);
5540 txg = spa_vdev_config_enter(spa);
5541 vd->vdev_removing = B_TRUE;
5542 vdev_dirty_leaves(vd, VDD_DTL, txg);
5543 vdev_config_dirty(vd);
5544 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5550 * Complete the removal by cleaning up the namespace.
5553 spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
5555 vdev_t *rvd = spa->spa_root_vdev;
5556 uint64_t id = vd->vdev_id;
5557 boolean_t last_vdev = (id == (rvd->vdev_children - 1));
5559 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5560 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5561 ASSERT(vd == vd->vdev_top);
5564 * Only remove any devices which are empty.
5566 if (vd->vdev_stat.vs_alloc != 0)
5569 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5571 if (list_link_active(&vd->vdev_state_dirty_node))
5572 vdev_state_clean(vd);
5573 if (list_link_active(&vd->vdev_config_dirty_node))
5574 vdev_config_clean(vd);
5579 vdev_compact_children(rvd);
5581 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
5582 vdev_add_child(rvd, vd);
5584 vdev_config_dirty(rvd);
5587 * Reassess the health of our root vdev.
5593 * Remove a device from the pool -
5595 * Removing a device from the vdev namespace requires several steps
5596 * and can take a significant amount of time. As a result we use
5597 * the spa_vdev_config_[enter/exit] functions which allow us to
5598 * grab and release the spa_config_lock while still holding the namespace
5599 * lock. During each step the configuration is synced out.
5601 * Currently, this supports removing only hot spares, slogs, and level 2 ARC
5605 spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
5608 metaslab_group_t *mg;
5609 nvlist_t **spares, **l2cache, *nv;
5611 uint_t nspares, nl2cache;
5613 boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
5615 ASSERT(spa_writeable(spa));
5618 txg = spa_vdev_enter(spa);
5620 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5622 if (spa->spa_spares.sav_vdevs != NULL &&
5623 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5624 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
5625 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
5627 * Only remove the hot spare if it's not currently in use
5630 if (vd == NULL || unspare) {
5631 spa_vdev_remove_aux(spa->spa_spares.sav_config,
5632 ZPOOL_CONFIG_SPARES, spares, nspares, nv);
5633 spa_load_spares(spa);
5634 spa->spa_spares.sav_sync = B_TRUE;
5636 error = SET_ERROR(EBUSY);
5638 } else if (spa->spa_l2cache.sav_vdevs != NULL &&
5639 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5640 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
5641 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
5643 * Cache devices can always be removed.
5645 spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
5646 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
5647 spa_load_l2cache(spa);
5648 spa->spa_l2cache.sav_sync = B_TRUE;
5649 } else if (vd != NULL && vd->vdev_islog) {
5651 ASSERT(vd == vd->vdev_top);
5656 * Stop allocating from this vdev.
5658 metaslab_group_passivate(mg);
5661 * Wait for the youngest allocations and frees to sync,
5662 * and then wait for the deferral of those frees to finish.
5664 spa_vdev_config_exit(spa, NULL,
5665 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
5668 * Attempt to evacuate the vdev.
5670 error = spa_vdev_remove_evacuate(spa, vd);
5672 txg = spa_vdev_config_enter(spa);
5675 * If we couldn't evacuate the vdev, unwind.
5678 metaslab_group_activate(mg);
5679 return (spa_vdev_exit(spa, NULL, txg, error));
5683 * Clean up the vdev namespace.
5685 spa_vdev_remove_from_namespace(spa, vd);
5687 } else if (vd != NULL) {
5689 * Normal vdevs cannot be removed (yet).
5691 error = SET_ERROR(ENOTSUP);
5694 * There is no vdev of any kind with the specified guid.
5696 error = SET_ERROR(ENOENT);
5700 return (spa_vdev_exit(spa, NULL, txg, error));
5706 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5707 * currently spared, so we can detach it.
5710 spa_vdev_resilver_done_hunt(vdev_t *vd)
5712 vdev_t *newvd, *oldvd;
5714 for (int c = 0; c < vd->vdev_children; c++) {
5715 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
5721 * Check for a completed replacement. We always consider the first
5722 * vdev in the list to be the oldest vdev, and the last one to be
5723 * the newest (see spa_vdev_attach() for how that works). In
5724 * the case where the newest vdev is faulted, we will not automatically
5725 * remove it after a resilver completes. This is OK as it will require
5726 * user intervention to determine which disk the admin wishes to keep.
5728 if (vd->vdev_ops == &vdev_replacing_ops) {
5729 ASSERT(vd->vdev_children > 1);
5731 newvd = vd->vdev_child[vd->vdev_children - 1];
5732 oldvd = vd->vdev_child[0];
5734 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
5735 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5736 !vdev_dtl_required(oldvd))
5741 * Check for a completed resilver with the 'unspare' flag set.
5743 if (vd->vdev_ops == &vdev_spare_ops) {
5744 vdev_t *first = vd->vdev_child[0];
5745 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
5747 if (last->vdev_unspare) {
5750 } else if (first->vdev_unspare) {
5757 if (oldvd != NULL &&
5758 vdev_dtl_empty(newvd, DTL_MISSING) &&
5759 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5760 !vdev_dtl_required(oldvd))
5764 * If there are more than two spares attached to a disk,
5765 * and those spares are not required, then we want to
5766 * attempt to free them up now so that they can be used
5767 * by other pools. Once we're back down to a single
5768 * disk+spare, we stop removing them.
5770 if (vd->vdev_children > 2) {
5771 newvd = vd->vdev_child[1];
5773 if (newvd->vdev_isspare && last->vdev_isspare &&
5774 vdev_dtl_empty(last, DTL_MISSING) &&
5775 vdev_dtl_empty(last, DTL_OUTAGE) &&
5776 !vdev_dtl_required(newvd))
5785 spa_vdev_resilver_done(spa_t *spa)
5787 vdev_t *vd, *pvd, *ppvd;
5788 uint64_t guid, sguid, pguid, ppguid;
5790 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5792 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
5793 pvd = vd->vdev_parent;
5794 ppvd = pvd->vdev_parent;
5795 guid = vd->vdev_guid;
5796 pguid = pvd->vdev_guid;
5797 ppguid = ppvd->vdev_guid;
5800 * If we have just finished replacing a hot spared device, then
5801 * we need to detach the parent's first child (the original hot
5804 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
5805 ppvd->vdev_children == 2) {
5806 ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
5807 sguid = ppvd->vdev_child[1]->vdev_guid;
5809 ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd));
5811 spa_config_exit(spa, SCL_ALL, FTAG);
5812 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
5814 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
5816 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5819 spa_config_exit(spa, SCL_ALL, FTAG);
5823 * Update the stored path or FRU for this vdev.
5826 spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
5830 boolean_t sync = B_FALSE;
5832 ASSERT(spa_writeable(spa));
5834 spa_vdev_state_enter(spa, SCL_ALL);
5836 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
5837 return (spa_vdev_state_exit(spa, NULL, ENOENT));
5839 if (!vd->vdev_ops->vdev_op_leaf)
5840 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
5843 if (strcmp(value, vd->vdev_path) != 0) {
5844 spa_strfree(vd->vdev_path);
5845 vd->vdev_path = spa_strdup(value);
5849 if (vd->vdev_fru == NULL) {
5850 vd->vdev_fru = spa_strdup(value);
5852 } else if (strcmp(value, vd->vdev_fru) != 0) {
5853 spa_strfree(vd->vdev_fru);
5854 vd->vdev_fru = spa_strdup(value);
5859 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
5863 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
5865 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5869 spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5871 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5875 * ==========================================================================
5877 * ==========================================================================
5881 spa_scan_stop(spa_t *spa)
5883 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5884 if (dsl_scan_resilvering(spa->spa_dsl_pool))
5885 return (SET_ERROR(EBUSY));
5886 return (dsl_scan_cancel(spa->spa_dsl_pool));
5890 spa_scan(spa_t *spa, pool_scan_func_t func)
5892 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5894 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5895 return (SET_ERROR(ENOTSUP));
5898 * If a resilver was requested, but there is no DTL on a
5899 * writeable leaf device, we have nothing to do.
5901 if (func == POOL_SCAN_RESILVER &&
5902 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5903 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5907 return (dsl_scan(spa->spa_dsl_pool, func));
5911 * ==========================================================================
5912 * SPA async task processing
5913 * ==========================================================================
5917 spa_async_remove(spa_t *spa, vdev_t *vd)
5919 if (vd->vdev_remove_wanted) {
5920 vd->vdev_remove_wanted = B_FALSE;
5921 vd->vdev_delayed_close = B_FALSE;
5922 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5925 * We want to clear the stats, but we don't want to do a full
5926 * vdev_clear() as that will cause us to throw away
5927 * degraded/faulted state as well as attempt to reopen the
5928 * device, all of which is a waste.
5930 vd->vdev_stat.vs_read_errors = 0;
5931 vd->vdev_stat.vs_write_errors = 0;
5932 vd->vdev_stat.vs_checksum_errors = 0;
5934 vdev_state_dirty(vd->vdev_top);
5937 for (int c = 0; c < vd->vdev_children; c++)
5938 spa_async_remove(spa, vd->vdev_child[c]);
5942 spa_async_probe(spa_t *spa, vdev_t *vd)
5944 if (vd->vdev_probe_wanted) {
5945 vd->vdev_probe_wanted = B_FALSE;
5946 vdev_reopen(vd); /* vdev_open() does the actual probe */
5949 for (int c = 0; c < vd->vdev_children; c++)
5950 spa_async_probe(spa, vd->vdev_child[c]);
5954 spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5960 if (!spa->spa_autoexpand)
5963 for (int c = 0; c < vd->vdev_children; c++) {
5964 vdev_t *cvd = vd->vdev_child[c];
5965 spa_async_autoexpand(spa, cvd);
5968 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5971 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
5972 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
5974 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5975 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
5977 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
5978 ESC_ZFS_VDEV_AUTOEXPAND, attr, &eid, DDI_SLEEP);
5981 kmem_free(physpath, MAXPATHLEN);
5985 spa_async_thread(void *arg)
5990 ASSERT(spa->spa_sync_on);
5992 mutex_enter(&spa->spa_async_lock);
5993 tasks = spa->spa_async_tasks;
5994 spa->spa_async_tasks &= SPA_ASYNC_REMOVE;
5995 mutex_exit(&spa->spa_async_lock);
5998 * See if the config needs to be updated.
6000 if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
6001 uint64_t old_space, new_space;
6003 mutex_enter(&spa_namespace_lock);
6004 old_space = metaslab_class_get_space(spa_normal_class(spa));
6005 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
6006 new_space = metaslab_class_get_space(spa_normal_class(spa));
6007 mutex_exit(&spa_namespace_lock);
6010 * If the pool grew as a result of the config update,
6011 * then log an internal history event.
6013 if (new_space != old_space) {
6014 spa_history_log_internal(spa, "vdev online", NULL,
6015 "pool '%s' size: %llu(+%llu)",
6016 spa_name(spa), new_space, new_space - old_space);
6020 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
6021 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
6022 spa_async_autoexpand(spa, spa->spa_root_vdev);
6023 spa_config_exit(spa, SCL_CONFIG, FTAG);
6027 * See if any devices need to be probed.
6029 if (tasks & SPA_ASYNC_PROBE) {
6030 spa_vdev_state_enter(spa, SCL_NONE);
6031 spa_async_probe(spa, spa->spa_root_vdev);
6032 (void) spa_vdev_state_exit(spa, NULL, 0);
6036 * If any devices are done replacing, detach them.
6038 if (tasks & SPA_ASYNC_RESILVER_DONE)
6039 spa_vdev_resilver_done(spa);
6042 * Kick off a resilver.
6044 if (tasks & SPA_ASYNC_RESILVER)
6045 dsl_resilver_restart(spa->spa_dsl_pool, 0);
6048 * Let the world know that we're done.
6050 mutex_enter(&spa->spa_async_lock);
6051 spa->spa_async_thread = NULL;
6052 cv_broadcast(&spa->spa_async_cv);
6053 mutex_exit(&spa->spa_async_lock);
6058 spa_async_thread_vd(void *arg)
6063 ASSERT(spa->spa_sync_on);
6065 mutex_enter(&spa->spa_async_lock);
6066 tasks = spa->spa_async_tasks;
6068 spa->spa_async_tasks &= ~SPA_ASYNC_REMOVE;
6069 mutex_exit(&spa->spa_async_lock);
6072 * See if any devices need to be marked REMOVED.
6074 if (tasks & SPA_ASYNC_REMOVE) {
6075 spa_vdev_state_enter(spa, SCL_NONE);
6076 spa_async_remove(spa, spa->spa_root_vdev);
6077 for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
6078 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
6079 for (int i = 0; i < spa->spa_spares.sav_count; i++)
6080 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
6081 (void) spa_vdev_state_exit(spa, NULL, 0);
6085 * Let the world know that we're done.
6087 mutex_enter(&spa->spa_async_lock);
6088 tasks = spa->spa_async_tasks;
6089 if ((tasks & SPA_ASYNC_REMOVE) != 0)
6091 spa->spa_async_thread_vd = NULL;
6092 cv_broadcast(&spa->spa_async_cv);
6093 mutex_exit(&spa->spa_async_lock);
6098 spa_async_suspend(spa_t *spa)
6100 mutex_enter(&spa->spa_async_lock);
6101 spa->spa_async_suspended++;
6102 while (spa->spa_async_thread != NULL &&
6103 spa->spa_async_thread_vd != NULL)
6104 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
6105 mutex_exit(&spa->spa_async_lock);
6109 spa_async_resume(spa_t *spa)
6111 mutex_enter(&spa->spa_async_lock);
6112 ASSERT(spa->spa_async_suspended != 0);
6113 spa->spa_async_suspended--;
6114 mutex_exit(&spa->spa_async_lock);
6118 spa_async_tasks_pending(spa_t *spa)
6120 uint_t non_config_tasks;
6122 boolean_t config_task_suspended;
6124 non_config_tasks = spa->spa_async_tasks & ~(SPA_ASYNC_CONFIG_UPDATE |
6126 config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE;
6127 if (spa->spa_ccw_fail_time == 0) {
6128 config_task_suspended = B_FALSE;
6130 config_task_suspended =
6131 (gethrtime() - spa->spa_ccw_fail_time) <
6132 (zfs_ccw_retry_interval * NANOSEC);
6135 return (non_config_tasks || (config_task && !config_task_suspended));
6139 spa_async_dispatch(spa_t *spa)
6141 mutex_enter(&spa->spa_async_lock);
6142 if (spa_async_tasks_pending(spa) &&
6143 !spa->spa_async_suspended &&
6144 spa->spa_async_thread == NULL &&
6146 spa->spa_async_thread = thread_create(NULL, 0,
6147 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
6148 mutex_exit(&spa->spa_async_lock);
6152 spa_async_dispatch_vd(spa_t *spa)
6154 mutex_enter(&spa->spa_async_lock);
6155 if ((spa->spa_async_tasks & SPA_ASYNC_REMOVE) != 0 &&
6156 !spa->spa_async_suspended &&
6157 spa->spa_async_thread_vd == NULL &&
6159 spa->spa_async_thread_vd = thread_create(NULL, 0,
6160 spa_async_thread_vd, spa, 0, &p0, TS_RUN, maxclsyspri);
6161 mutex_exit(&spa->spa_async_lock);
6165 spa_async_request(spa_t *spa, int task)
6167 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
6168 mutex_enter(&spa->spa_async_lock);
6169 spa->spa_async_tasks |= task;
6170 mutex_exit(&spa->spa_async_lock);
6171 spa_async_dispatch_vd(spa);
6175 * ==========================================================================
6176 * SPA syncing routines
6177 * ==========================================================================
6181 bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
6184 bpobj_enqueue(bpo, bp, tx);
6189 spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
6193 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
6194 BP_GET_PSIZE(bp), zio->io_flags));
6199 * Note: this simple function is not inlined to make it easier to dtrace the
6200 * amount of time spent syncing frees.
6203 spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx)
6205 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6206 bplist_iterate(bpl, spa_free_sync_cb, zio, tx);
6207 VERIFY(zio_wait(zio) == 0);
6211 * Note: this simple function is not inlined to make it easier to dtrace the
6212 * amount of time spent syncing deferred frees.
6215 spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx)
6217 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6218 VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj,
6219 spa_free_sync_cb, zio, tx), ==, 0);
6220 VERIFY0(zio_wait(zio));
6225 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
6227 char *packed = NULL;
6232 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
6235 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
6236 * information. This avoids the dmu_buf_will_dirty() path and
6237 * saves us a pre-read to get data we don't actually care about.
6239 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
6240 packed = kmem_alloc(bufsize, KM_SLEEP);
6242 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
6244 bzero(packed + nvsize, bufsize - nvsize);
6246 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
6248 kmem_free(packed, bufsize);
6250 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
6251 dmu_buf_will_dirty(db, tx);
6252 *(uint64_t *)db->db_data = nvsize;
6253 dmu_buf_rele(db, FTAG);
6257 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
6258 const char *config, const char *entry)
6268 * Update the MOS nvlist describing the list of available devices.
6269 * spa_validate_aux() will have already made sure this nvlist is
6270 * valid and the vdevs are labeled appropriately.
6272 if (sav->sav_object == 0) {
6273 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
6274 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
6275 sizeof (uint64_t), tx);
6276 VERIFY(zap_update(spa->spa_meta_objset,
6277 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
6278 &sav->sav_object, tx) == 0);
6281 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
6282 if (sav->sav_count == 0) {
6283 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
6285 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
6286 for (i = 0; i < sav->sav_count; i++)
6287 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
6288 B_FALSE, VDEV_CONFIG_L2CACHE);
6289 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
6290 sav->sav_count) == 0);
6291 for (i = 0; i < sav->sav_count; i++)
6292 nvlist_free(list[i]);
6293 kmem_free(list, sav->sav_count * sizeof (void *));
6296 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
6297 nvlist_free(nvroot);
6299 sav->sav_sync = B_FALSE;
6303 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
6307 if (list_is_empty(&spa->spa_config_dirty_list))
6310 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6312 config = spa_config_generate(spa, spa->spa_root_vdev,
6313 dmu_tx_get_txg(tx), B_FALSE);
6316 * If we're upgrading the spa version then make sure that
6317 * the config object gets updated with the correct version.
6319 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
6320 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
6321 spa->spa_uberblock.ub_version);
6323 spa_config_exit(spa, SCL_STATE, FTAG);
6325 if (spa->spa_config_syncing)
6326 nvlist_free(spa->spa_config_syncing);
6327 spa->spa_config_syncing = config;
6329 spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
6333 spa_sync_version(void *arg, dmu_tx_t *tx)
6335 uint64_t *versionp = arg;
6336 uint64_t version = *versionp;
6337 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
6340 * Setting the version is special cased when first creating the pool.
6342 ASSERT(tx->tx_txg != TXG_INITIAL);
6344 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
6345 ASSERT(version >= spa_version(spa));
6347 spa->spa_uberblock.ub_version = version;
6348 vdev_config_dirty(spa->spa_root_vdev);
6349 spa_history_log_internal(spa, "set", tx, "version=%lld", version);
6353 * Set zpool properties.
6356 spa_sync_props(void *arg, dmu_tx_t *tx)
6358 nvlist_t *nvp = arg;
6359 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
6360 objset_t *mos = spa->spa_meta_objset;
6361 nvpair_t *elem = NULL;
6363 mutex_enter(&spa->spa_props_lock);
6365 while ((elem = nvlist_next_nvpair(nvp, elem))) {
6367 char *strval, *fname;
6369 const char *propname;
6370 zprop_type_t proptype;
6373 switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
6376 * We checked this earlier in spa_prop_validate().
6378 ASSERT(zpool_prop_feature(nvpair_name(elem)));
6380 fname = strchr(nvpair_name(elem), '@') + 1;
6381 VERIFY0(zfeature_lookup_name(fname, &fid));
6383 spa_feature_enable(spa, fid, tx);
6384 spa_history_log_internal(spa, "set", tx,
6385 "%s=enabled", nvpair_name(elem));
6388 case ZPOOL_PROP_VERSION:
6389 intval = fnvpair_value_uint64(elem);
6391 * The version is synced seperatly before other
6392 * properties and should be correct by now.
6394 ASSERT3U(spa_version(spa), >=, intval);
6397 case ZPOOL_PROP_ALTROOT:
6399 * 'altroot' is a non-persistent property. It should
6400 * have been set temporarily at creation or import time.
6402 ASSERT(spa->spa_root != NULL);
6405 case ZPOOL_PROP_READONLY:
6406 case ZPOOL_PROP_CACHEFILE:
6408 * 'readonly' and 'cachefile' are also non-persisitent
6412 case ZPOOL_PROP_COMMENT:
6413 strval = fnvpair_value_string(elem);
6414 if (spa->spa_comment != NULL)
6415 spa_strfree(spa->spa_comment);
6416 spa->spa_comment = spa_strdup(strval);
6418 * We need to dirty the configuration on all the vdevs
6419 * so that their labels get updated. It's unnecessary
6420 * to do this for pool creation since the vdev's
6421 * configuratoin has already been dirtied.
6423 if (tx->tx_txg != TXG_INITIAL)
6424 vdev_config_dirty(spa->spa_root_vdev);
6425 spa_history_log_internal(spa, "set", tx,
6426 "%s=%s", nvpair_name(elem), strval);
6430 * Set pool property values in the poolprops mos object.
6432 if (spa->spa_pool_props_object == 0) {
6433 spa->spa_pool_props_object =
6434 zap_create_link(mos, DMU_OT_POOL_PROPS,
6435 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
6439 /* normalize the property name */
6440 propname = zpool_prop_to_name(prop);
6441 proptype = zpool_prop_get_type(prop);
6443 if (nvpair_type(elem) == DATA_TYPE_STRING) {
6444 ASSERT(proptype == PROP_TYPE_STRING);
6445 strval = fnvpair_value_string(elem);
6446 VERIFY0(zap_update(mos,
6447 spa->spa_pool_props_object, propname,
6448 1, strlen(strval) + 1, strval, tx));
6449 spa_history_log_internal(spa, "set", tx,
6450 "%s=%s", nvpair_name(elem), strval);
6451 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
6452 intval = fnvpair_value_uint64(elem);
6454 if (proptype == PROP_TYPE_INDEX) {
6456 VERIFY0(zpool_prop_index_to_string(
6457 prop, intval, &unused));
6459 VERIFY0(zap_update(mos,
6460 spa->spa_pool_props_object, propname,
6461 8, 1, &intval, tx));
6462 spa_history_log_internal(spa, "set", tx,
6463 "%s=%lld", nvpair_name(elem), intval);
6465 ASSERT(0); /* not allowed */
6469 case ZPOOL_PROP_DELEGATION:
6470 spa->spa_delegation = intval;
6472 case ZPOOL_PROP_BOOTFS:
6473 spa->spa_bootfs = intval;
6475 case ZPOOL_PROP_FAILUREMODE:
6476 spa->spa_failmode = intval;
6478 case ZPOOL_PROP_AUTOEXPAND:
6479 spa->spa_autoexpand = intval;
6480 if (tx->tx_txg != TXG_INITIAL)
6481 spa_async_request(spa,
6482 SPA_ASYNC_AUTOEXPAND);
6484 case ZPOOL_PROP_DEDUPDITTO:
6485 spa->spa_dedup_ditto = intval;
6494 mutex_exit(&spa->spa_props_lock);
6498 * Perform one-time upgrade on-disk changes. spa_version() does not
6499 * reflect the new version this txg, so there must be no changes this
6500 * txg to anything that the upgrade code depends on after it executes.
6501 * Therefore this must be called after dsl_pool_sync() does the sync
6505 spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
6507 dsl_pool_t *dp = spa->spa_dsl_pool;
6509 ASSERT(spa->spa_sync_pass == 1);
6511 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
6513 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
6514 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
6515 dsl_pool_create_origin(dp, tx);
6517 /* Keeping the origin open increases spa_minref */
6518 spa->spa_minref += 3;
6521 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
6522 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
6523 dsl_pool_upgrade_clones(dp, tx);
6526 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
6527 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
6528 dsl_pool_upgrade_dir_clones(dp, tx);
6530 /* Keeping the freedir open increases spa_minref */
6531 spa->spa_minref += 3;
6534 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES &&
6535 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
6536 spa_feature_create_zap_objects(spa, tx);
6540 * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable
6541 * when possibility to use lz4 compression for metadata was added
6542 * Old pools that have this feature enabled must be upgraded to have
6543 * this feature active
6545 if (spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
6546 boolean_t lz4_en = spa_feature_is_enabled(spa,
6547 SPA_FEATURE_LZ4_COMPRESS);
6548 boolean_t lz4_ac = spa_feature_is_active(spa,
6549 SPA_FEATURE_LZ4_COMPRESS);
6551 if (lz4_en && !lz4_ac)
6552 spa_feature_incr(spa, SPA_FEATURE_LZ4_COMPRESS, tx);
6554 rrw_exit(&dp->dp_config_rwlock, FTAG);
6558 * Sync the specified transaction group. New blocks may be dirtied as
6559 * part of the process, so we iterate until it converges.
6562 spa_sync(spa_t *spa, uint64_t txg)
6564 dsl_pool_t *dp = spa->spa_dsl_pool;
6565 objset_t *mos = spa->spa_meta_objset;
6566 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
6567 vdev_t *rvd = spa->spa_root_vdev;
6572 VERIFY(spa_writeable(spa));
6575 * Lock out configuration changes.
6577 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
6579 spa->spa_syncing_txg = txg;
6580 spa->spa_sync_pass = 0;
6583 * If there are any pending vdev state changes, convert them
6584 * into config changes that go out with this transaction group.
6586 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6587 while (list_head(&spa->spa_state_dirty_list) != NULL) {
6589 * We need the write lock here because, for aux vdevs,
6590 * calling vdev_config_dirty() modifies sav_config.
6591 * This is ugly and will become unnecessary when we
6592 * eliminate the aux vdev wart by integrating all vdevs
6593 * into the root vdev tree.
6595 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6596 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
6597 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
6598 vdev_state_clean(vd);
6599 vdev_config_dirty(vd);
6601 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6602 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
6604 spa_config_exit(spa, SCL_STATE, FTAG);
6606 tx = dmu_tx_create_assigned(dp, txg);
6608 spa->spa_sync_starttime = gethrtime();
6610 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid,
6611 spa->spa_sync_starttime + spa->spa_deadman_synctime));
6614 callout_reset(&spa->spa_deadman_cycid,
6615 hz * spa->spa_deadman_synctime / NANOSEC, spa_deadman, spa);
6620 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
6621 * set spa_deflate if we have no raid-z vdevs.
6623 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
6624 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
6627 for (i = 0; i < rvd->vdev_children; i++) {
6628 vd = rvd->vdev_child[i];
6629 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
6632 if (i == rvd->vdev_children) {
6633 spa->spa_deflate = TRUE;
6634 VERIFY(0 == zap_add(spa->spa_meta_objset,
6635 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
6636 sizeof (uint64_t), 1, &spa->spa_deflate, tx));
6641 * Iterate to convergence.
6644 int pass = ++spa->spa_sync_pass;
6646 spa_sync_config_object(spa, tx);
6647 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
6648 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
6649 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
6650 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
6651 spa_errlog_sync(spa, txg);
6652 dsl_pool_sync(dp, txg);
6654 if (pass < zfs_sync_pass_deferred_free) {
6655 spa_sync_frees(spa, free_bpl, tx);
6658 * We can not defer frees in pass 1, because
6659 * we sync the deferred frees later in pass 1.
6661 ASSERT3U(pass, >, 1);
6662 bplist_iterate(free_bpl, bpobj_enqueue_cb,
6663 &spa->spa_deferred_bpobj, tx);
6667 dsl_scan_sync(dp, tx);
6669 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
6673 spa_sync_upgrades(spa, tx);
6675 spa->spa_uberblock.ub_rootbp.blk_birth);
6677 * Note: We need to check if the MOS is dirty
6678 * because we could have marked the MOS dirty
6679 * without updating the uberblock (e.g. if we
6680 * have sync tasks but no dirty user data). We
6681 * need to check the uberblock's rootbp because
6682 * it is updated if we have synced out dirty
6683 * data (though in this case the MOS will most
6684 * likely also be dirty due to second order
6685 * effects, we don't want to rely on that here).
6687 if (spa->spa_uberblock.ub_rootbp.blk_birth < txg &&
6688 !dmu_objset_is_dirty(mos, txg)) {
6690 * Nothing changed on the first pass,
6691 * therefore this TXG is a no-op. Avoid
6692 * syncing deferred frees, so that we
6693 * can keep this TXG as a no-op.
6695 ASSERT(txg_list_empty(&dp->dp_dirty_datasets,
6697 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
6698 ASSERT(txg_list_empty(&dp->dp_sync_tasks, txg));
6701 spa_sync_deferred_frees(spa, tx);
6704 } while (dmu_objset_is_dirty(mos, txg));
6707 * Rewrite the vdev configuration (which includes the uberblock)
6708 * to commit the transaction group.
6710 * If there are no dirty vdevs, we sync the uberblock to a few
6711 * random top-level vdevs that are known to be visible in the
6712 * config cache (see spa_vdev_add() for a complete description).
6713 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
6717 * We hold SCL_STATE to prevent vdev open/close/etc.
6718 * while we're attempting to write the vdev labels.
6720 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6722 if (list_is_empty(&spa->spa_config_dirty_list)) {
6723 vdev_t *svd[SPA_DVAS_PER_BP];
6725 int children = rvd->vdev_children;
6726 int c0 = spa_get_random(children);
6728 for (int c = 0; c < children; c++) {
6729 vd = rvd->vdev_child[(c0 + c) % children];
6730 if (vd->vdev_ms_array == 0 || vd->vdev_islog)
6732 svd[svdcount++] = vd;
6733 if (svdcount == SPA_DVAS_PER_BP)
6736 error = vdev_config_sync(svd, svdcount, txg, B_FALSE);
6738 error = vdev_config_sync(svd, svdcount, txg,
6741 error = vdev_config_sync(rvd->vdev_child,
6742 rvd->vdev_children, txg, B_FALSE);
6744 error = vdev_config_sync(rvd->vdev_child,
6745 rvd->vdev_children, txg, B_TRUE);
6749 spa->spa_last_synced_guid = rvd->vdev_guid;
6751 spa_config_exit(spa, SCL_STATE, FTAG);
6755 zio_suspend(spa, NULL);
6756 zio_resume_wait(spa);
6761 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY));
6764 callout_drain(&spa->spa_deadman_cycid);
6769 * Clear the dirty config list.
6771 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
6772 vdev_config_clean(vd);
6775 * Now that the new config has synced transactionally,
6776 * let it become visible to the config cache.
6778 if (spa->spa_config_syncing != NULL) {
6779 spa_config_set(spa, spa->spa_config_syncing);
6780 spa->spa_config_txg = txg;
6781 spa->spa_config_syncing = NULL;
6784 spa->spa_ubsync = spa->spa_uberblock;
6786 dsl_pool_sync_done(dp, txg);
6789 * Update usable space statistics.
6791 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
6792 vdev_sync_done(vd, txg);
6794 spa_update_dspace(spa);
6797 * It had better be the case that we didn't dirty anything
6798 * since vdev_config_sync().
6800 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
6801 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
6802 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
6804 spa->spa_sync_pass = 0;
6806 spa_config_exit(spa, SCL_CONFIG, FTAG);
6808 spa_handle_ignored_writes(spa);
6811 * If any async tasks have been requested, kick them off.
6813 spa_async_dispatch(spa);
6814 spa_async_dispatch_vd(spa);
6818 * Sync all pools. We don't want to hold the namespace lock across these
6819 * operations, so we take a reference on the spa_t and drop the lock during the
6823 spa_sync_allpools(void)
6826 mutex_enter(&spa_namespace_lock);
6827 while ((spa = spa_next(spa)) != NULL) {
6828 if (spa_state(spa) != POOL_STATE_ACTIVE ||
6829 !spa_writeable(spa) || spa_suspended(spa))
6831 spa_open_ref(spa, FTAG);
6832 mutex_exit(&spa_namespace_lock);
6833 txg_wait_synced(spa_get_dsl(spa), 0);
6834 mutex_enter(&spa_namespace_lock);
6835 spa_close(spa, FTAG);
6837 mutex_exit(&spa_namespace_lock);
6841 * ==========================================================================
6842 * Miscellaneous routines
6843 * ==========================================================================
6847 * Remove all pools in the system.
6855 * Remove all cached state. All pools should be closed now,
6856 * so every spa in the AVL tree should be unreferenced.
6858 mutex_enter(&spa_namespace_lock);
6859 while ((spa = spa_next(NULL)) != NULL) {
6861 * Stop async tasks. The async thread may need to detach
6862 * a device that's been replaced, which requires grabbing
6863 * spa_namespace_lock, so we must drop it here.
6865 spa_open_ref(spa, FTAG);
6866 mutex_exit(&spa_namespace_lock);
6867 spa_async_suspend(spa);
6868 mutex_enter(&spa_namespace_lock);
6869 spa_close(spa, FTAG);
6871 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
6873 spa_deactivate(spa);
6877 mutex_exit(&spa_namespace_lock);
6881 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
6886 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
6890 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
6891 vd = spa->spa_l2cache.sav_vdevs[i];
6892 if (vd->vdev_guid == guid)
6896 for (i = 0; i < spa->spa_spares.sav_count; i++) {
6897 vd = spa->spa_spares.sav_vdevs[i];
6898 if (vd->vdev_guid == guid)
6907 spa_upgrade(spa_t *spa, uint64_t version)
6909 ASSERT(spa_writeable(spa));
6911 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6914 * This should only be called for a non-faulted pool, and since a
6915 * future version would result in an unopenable pool, this shouldn't be
6918 ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version));
6919 ASSERT3U(version, >=, spa->spa_uberblock.ub_version);
6921 spa->spa_uberblock.ub_version = version;
6922 vdev_config_dirty(spa->spa_root_vdev);
6924 spa_config_exit(spa, SCL_ALL, FTAG);
6926 txg_wait_synced(spa_get_dsl(spa), 0);
6930 spa_has_spare(spa_t *spa, uint64_t guid)
6934 spa_aux_vdev_t *sav = &spa->spa_spares;
6936 for (i = 0; i < sav->sav_count; i++)
6937 if (sav->sav_vdevs[i]->vdev_guid == guid)
6940 for (i = 0; i < sav->sav_npending; i++) {
6941 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
6942 &spareguid) == 0 && spareguid == guid)
6950 * Check if a pool has an active shared spare device.
6951 * Note: reference count of an active spare is 2, as a spare and as a replace
6954 spa_has_active_shared_spare(spa_t *spa)
6958 spa_aux_vdev_t *sav = &spa->spa_spares;
6960 for (i = 0; i < sav->sav_count; i++) {
6961 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
6962 &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
6971 * Post a sysevent corresponding to the given event. The 'name' must be one of
6972 * the event definitions in sys/sysevent/eventdefs.h. The payload will be
6973 * filled in from the spa and (optionally) the vdev. This doesn't do anything
6974 * in the userland libzpool, as we don't want consumers to misinterpret ztest
6975 * or zdb as real changes.
6978 spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
6982 sysevent_attr_list_t *attr = NULL;
6983 sysevent_value_t value;
6986 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
6989 value.value_type = SE_DATA_TYPE_STRING;
6990 value.value.sv_string = spa_name(spa);
6991 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
6994 value.value_type = SE_DATA_TYPE_UINT64;
6995 value.value.sv_uint64 = spa_guid(spa);
6996 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
7000 value.value_type = SE_DATA_TYPE_UINT64;
7001 value.value.sv_uint64 = vd->vdev_guid;
7002 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
7006 if (vd->vdev_path) {
7007 value.value_type = SE_DATA_TYPE_STRING;
7008 value.value.sv_string = vd->vdev_path;
7009 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
7010 &value, SE_SLEEP) != 0)
7015 if (sysevent_attach_attributes(ev, attr) != 0)
7019 (void) log_sysevent(ev, SE_SLEEP, &eid);
7023 sysevent_free_attr(attr);