4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2013 by Delphix. All rights reserved.
25 * Copyright (c) 2013, 2014, Nexenta Systems, Inc. All rights reserved.
26 * Copyright (c) 2013 Martin Matuska <mm@FreeBSD.org>. All rights reserved.
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);
1633 nvsize = *(uint64_t *)db->db_data;
1634 dmu_buf_rele(db, FTAG);
1636 packed = kmem_alloc(nvsize, KM_SLEEP);
1637 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1640 error = nvlist_unpack(packed, nvsize, value, 0);
1641 kmem_free(packed, nvsize);
1647 * Checks to see if the given vdev could not be opened, in which case we post a
1648 * sysevent to notify the autoreplace code that the device has been removed.
1651 spa_check_removed(vdev_t *vd)
1653 for (int c = 0; c < vd->vdev_children; c++)
1654 spa_check_removed(vd->vdev_child[c]);
1656 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) &&
1658 zfs_post_autoreplace(vd->vdev_spa, vd);
1659 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1664 * Validate the current config against the MOS config
1667 spa_config_valid(spa_t *spa, nvlist_t *config)
1669 vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1672 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1674 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1675 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1677 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1680 * If we're doing a normal import, then build up any additional
1681 * diagnostic information about missing devices in this config.
1682 * We'll pass this up to the user for further processing.
1684 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1685 nvlist_t **child, *nv;
1688 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1690 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1692 for (int c = 0; c < rvd->vdev_children; c++) {
1693 vdev_t *tvd = rvd->vdev_child[c];
1694 vdev_t *mtvd = mrvd->vdev_child[c];
1696 if (tvd->vdev_ops == &vdev_missing_ops &&
1697 mtvd->vdev_ops != &vdev_missing_ops &&
1699 child[idx++] = vdev_config_generate(spa, mtvd,
1704 VERIFY(nvlist_add_nvlist_array(nv,
1705 ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1706 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1707 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1709 for (int i = 0; i < idx; i++)
1710 nvlist_free(child[i]);
1713 kmem_free(child, rvd->vdev_children * sizeof (char **));
1717 * Compare the root vdev tree with the information we have
1718 * from the MOS config (mrvd). Check each top-level vdev
1719 * with the corresponding MOS config top-level (mtvd).
1721 for (int c = 0; c < rvd->vdev_children; c++) {
1722 vdev_t *tvd = rvd->vdev_child[c];
1723 vdev_t *mtvd = mrvd->vdev_child[c];
1726 * Resolve any "missing" vdevs in the current configuration.
1727 * If we find that the MOS config has more accurate information
1728 * about the top-level vdev then use that vdev instead.
1730 if (tvd->vdev_ops == &vdev_missing_ops &&
1731 mtvd->vdev_ops != &vdev_missing_ops) {
1733 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1737 * Device specific actions.
1739 if (mtvd->vdev_islog) {
1740 spa_set_log_state(spa, SPA_LOG_CLEAR);
1743 * XXX - once we have 'readonly' pool
1744 * support we should be able to handle
1745 * missing data devices by transitioning
1746 * the pool to readonly.
1752 * Swap the missing vdev with the data we were
1753 * able to obtain from the MOS config.
1755 vdev_remove_child(rvd, tvd);
1756 vdev_remove_child(mrvd, mtvd);
1758 vdev_add_child(rvd, mtvd);
1759 vdev_add_child(mrvd, tvd);
1761 spa_config_exit(spa, SCL_ALL, FTAG);
1763 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1766 } else if (mtvd->vdev_islog) {
1768 * Load the slog device's state from the MOS config
1769 * since it's possible that the label does not
1770 * contain the most up-to-date information.
1772 vdev_load_log_state(tvd, mtvd);
1777 spa_config_exit(spa, SCL_ALL, FTAG);
1780 * Ensure we were able to validate the config.
1782 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1786 * Check for missing log devices
1789 spa_check_logs(spa_t *spa)
1791 boolean_t rv = B_FALSE;
1793 switch (spa->spa_log_state) {
1794 case SPA_LOG_MISSING:
1795 /* need to recheck in case slog has been restored */
1796 case SPA_LOG_UNKNOWN:
1797 rv = (dmu_objset_find(spa->spa_name, zil_check_log_chain,
1798 NULL, DS_FIND_CHILDREN) != 0);
1800 spa_set_log_state(spa, SPA_LOG_MISSING);
1807 spa_passivate_log(spa_t *spa)
1809 vdev_t *rvd = spa->spa_root_vdev;
1810 boolean_t slog_found = B_FALSE;
1812 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1814 if (!spa_has_slogs(spa))
1817 for (int c = 0; c < rvd->vdev_children; c++) {
1818 vdev_t *tvd = rvd->vdev_child[c];
1819 metaslab_group_t *mg = tvd->vdev_mg;
1821 if (tvd->vdev_islog) {
1822 metaslab_group_passivate(mg);
1823 slog_found = B_TRUE;
1827 return (slog_found);
1831 spa_activate_log(spa_t *spa)
1833 vdev_t *rvd = spa->spa_root_vdev;
1835 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1837 for (int c = 0; c < rvd->vdev_children; c++) {
1838 vdev_t *tvd = rvd->vdev_child[c];
1839 metaslab_group_t *mg = tvd->vdev_mg;
1841 if (tvd->vdev_islog)
1842 metaslab_group_activate(mg);
1847 spa_offline_log(spa_t *spa)
1851 error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1852 NULL, DS_FIND_CHILDREN);
1855 * We successfully offlined the log device, sync out the
1856 * current txg so that the "stubby" block can be removed
1859 txg_wait_synced(spa->spa_dsl_pool, 0);
1865 spa_aux_check_removed(spa_aux_vdev_t *sav)
1869 for (i = 0; i < sav->sav_count; i++)
1870 spa_check_removed(sav->sav_vdevs[i]);
1874 spa_claim_notify(zio_t *zio)
1876 spa_t *spa = zio->io_spa;
1881 mutex_enter(&spa->spa_props_lock); /* any mutex will do */
1882 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1883 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1884 mutex_exit(&spa->spa_props_lock);
1887 typedef struct spa_load_error {
1888 uint64_t sle_meta_count;
1889 uint64_t sle_data_count;
1893 spa_load_verify_done(zio_t *zio)
1895 blkptr_t *bp = zio->io_bp;
1896 spa_load_error_t *sle = zio->io_private;
1897 dmu_object_type_t type = BP_GET_TYPE(bp);
1898 int error = zio->io_error;
1899 spa_t *spa = zio->io_spa;
1902 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1903 type != DMU_OT_INTENT_LOG)
1904 atomic_inc_64(&sle->sle_meta_count);
1906 atomic_inc_64(&sle->sle_data_count);
1908 zio_data_buf_free(zio->io_data, zio->io_size);
1910 mutex_enter(&spa->spa_scrub_lock);
1911 spa->spa_scrub_inflight--;
1912 cv_broadcast(&spa->spa_scrub_io_cv);
1913 mutex_exit(&spa->spa_scrub_lock);
1917 * Maximum number of concurrent scrub i/os to create while verifying
1918 * a pool while importing it.
1920 int spa_load_verify_maxinflight = 10000;
1921 boolean_t spa_load_verify_metadata = B_TRUE;
1922 boolean_t spa_load_verify_data = B_TRUE;
1924 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_maxinflight, CTLFLAG_RWTUN,
1925 &spa_load_verify_maxinflight, 0,
1926 "Maximum number of concurrent scrub I/Os to create while verifying a "
1927 "pool while importing it");
1929 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_metadata, CTLFLAG_RWTUN,
1930 &spa_load_verify_metadata, 0,
1931 "Check metadata on import?");
1933 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_data, CTLFLAG_RWTUN,
1934 &spa_load_verify_data, 0,
1935 "Check user data on import?");
1939 spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1940 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
1942 if (BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp))
1945 * Note: normally this routine will not be called if
1946 * spa_load_verify_metadata is not set. However, it may be useful
1947 * to manually set the flag after the traversal has begun.
1949 if (!spa_load_verify_metadata)
1951 if (BP_GET_BUFC_TYPE(bp) == ARC_BUFC_DATA && !spa_load_verify_data)
1955 size_t size = BP_GET_PSIZE(bp);
1956 void *data = zio_data_buf_alloc(size);
1958 mutex_enter(&spa->spa_scrub_lock);
1959 while (spa->spa_scrub_inflight >= spa_load_verify_maxinflight)
1960 cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
1961 spa->spa_scrub_inflight++;
1962 mutex_exit(&spa->spa_scrub_lock);
1964 zio_nowait(zio_read(rio, spa, bp, data, size,
1965 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1966 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1967 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1972 spa_load_verify(spa_t *spa)
1975 spa_load_error_t sle = { 0 };
1976 zpool_rewind_policy_t policy;
1977 boolean_t verify_ok = B_FALSE;
1980 zpool_get_rewind_policy(spa->spa_config, &policy);
1982 if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1985 rio = zio_root(spa, NULL, &sle,
1986 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1988 if (spa_load_verify_metadata) {
1989 error = traverse_pool(spa, spa->spa_verify_min_txg,
1990 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA,
1991 spa_load_verify_cb, rio);
1994 (void) zio_wait(rio);
1996 spa->spa_load_meta_errors = sle.sle_meta_count;
1997 spa->spa_load_data_errors = sle.sle_data_count;
1999 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
2000 sle.sle_data_count <= policy.zrp_maxdata) {
2004 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
2005 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
2007 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
2008 VERIFY(nvlist_add_uint64(spa->spa_load_info,
2009 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
2010 VERIFY(nvlist_add_int64(spa->spa_load_info,
2011 ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
2012 VERIFY(nvlist_add_uint64(spa->spa_load_info,
2013 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
2015 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
2019 if (error != ENXIO && error != EIO)
2020 error = SET_ERROR(EIO);
2024 return (verify_ok ? 0 : EIO);
2028 * Find a value in the pool props object.
2031 spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
2033 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
2034 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
2038 * Find a value in the pool directory object.
2041 spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
2043 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
2044 name, sizeof (uint64_t), 1, val));
2048 spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
2050 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
2055 * Fix up config after a partly-completed split. This is done with the
2056 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
2057 * pool have that entry in their config, but only the splitting one contains
2058 * a list of all the guids of the vdevs that are being split off.
2060 * This function determines what to do with that list: either rejoin
2061 * all the disks to the pool, or complete the splitting process. To attempt
2062 * the rejoin, each disk that is offlined is marked online again, and
2063 * we do a reopen() call. If the vdev label for every disk that was
2064 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
2065 * then we call vdev_split() on each disk, and complete the split.
2067 * Otherwise we leave the config alone, with all the vdevs in place in
2068 * the original pool.
2071 spa_try_repair(spa_t *spa, nvlist_t *config)
2078 boolean_t attempt_reopen;
2080 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
2083 /* check that the config is complete */
2084 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
2085 &glist, &gcount) != 0)
2088 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
2090 /* attempt to online all the vdevs & validate */
2091 attempt_reopen = B_TRUE;
2092 for (i = 0; i < gcount; i++) {
2093 if (glist[i] == 0) /* vdev is hole */
2096 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
2097 if (vd[i] == NULL) {
2099 * Don't bother attempting to reopen the disks;
2100 * just do the split.
2102 attempt_reopen = B_FALSE;
2104 /* attempt to re-online it */
2105 vd[i]->vdev_offline = B_FALSE;
2109 if (attempt_reopen) {
2110 vdev_reopen(spa->spa_root_vdev);
2112 /* check each device to see what state it's in */
2113 for (extracted = 0, i = 0; i < gcount; i++) {
2114 if (vd[i] != NULL &&
2115 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
2122 * If every disk has been moved to the new pool, or if we never
2123 * even attempted to look at them, then we split them off for
2126 if (!attempt_reopen || gcount == extracted) {
2127 for (i = 0; i < gcount; i++)
2130 vdev_reopen(spa->spa_root_vdev);
2133 kmem_free(vd, gcount * sizeof (vdev_t *));
2137 spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
2138 boolean_t mosconfig)
2140 nvlist_t *config = spa->spa_config;
2141 char *ereport = FM_EREPORT_ZFS_POOL;
2147 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
2148 return (SET_ERROR(EINVAL));
2150 ASSERT(spa->spa_comment == NULL);
2151 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
2152 spa->spa_comment = spa_strdup(comment);
2155 * Versioning wasn't explicitly added to the label until later, so if
2156 * it's not present treat it as the initial version.
2158 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
2159 &spa->spa_ubsync.ub_version) != 0)
2160 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
2162 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
2163 &spa->spa_config_txg);
2165 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
2166 spa_guid_exists(pool_guid, 0)) {
2167 error = SET_ERROR(EEXIST);
2169 spa->spa_config_guid = pool_guid;
2171 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
2173 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
2177 nvlist_free(spa->spa_load_info);
2178 spa->spa_load_info = fnvlist_alloc();
2180 gethrestime(&spa->spa_loaded_ts);
2181 error = spa_load_impl(spa, pool_guid, config, state, type,
2182 mosconfig, &ereport);
2186 * Don't count references from objsets that are already closed
2187 * and are making their way through the eviction process.
2189 spa_evicting_os_wait(spa);
2190 spa->spa_minref = refcount_count(&spa->spa_refcount);
2192 if (error != EEXIST) {
2193 spa->spa_loaded_ts.tv_sec = 0;
2194 spa->spa_loaded_ts.tv_nsec = 0;
2196 if (error != EBADF) {
2197 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
2200 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2207 * Load an existing storage pool, using the pool's builtin spa_config as a
2208 * source of configuration information.
2211 spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
2212 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
2216 nvlist_t *nvroot = NULL;
2219 uberblock_t *ub = &spa->spa_uberblock;
2220 uint64_t children, config_cache_txg = spa->spa_config_txg;
2221 int orig_mode = spa->spa_mode;
2224 boolean_t missing_feat_write = B_FALSE;
2227 * If this is an untrusted config, access the pool in read-only mode.
2228 * This prevents things like resilvering recently removed devices.
2231 spa->spa_mode = FREAD;
2233 ASSERT(MUTEX_HELD(&spa_namespace_lock));
2235 spa->spa_load_state = state;
2237 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
2238 return (SET_ERROR(EINVAL));
2240 parse = (type == SPA_IMPORT_EXISTING ?
2241 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2244 * Create "The Godfather" zio to hold all async IOs
2246 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
2248 for (int i = 0; i < max_ncpus; i++) {
2249 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
2250 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
2251 ZIO_FLAG_GODFATHER);
2255 * Parse the configuration into a vdev tree. We explicitly set the
2256 * value that will be returned by spa_version() since parsing the
2257 * configuration requires knowing the version number.
2259 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2260 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
2261 spa_config_exit(spa, SCL_ALL, FTAG);
2266 ASSERT(spa->spa_root_vdev == rvd);
2267 ASSERT3U(spa->spa_min_ashift, >=, SPA_MINBLOCKSHIFT);
2268 ASSERT3U(spa->spa_max_ashift, <=, SPA_MAXBLOCKSHIFT);
2270 if (type != SPA_IMPORT_ASSEMBLE) {
2271 ASSERT(spa_guid(spa) == pool_guid);
2275 * Try to open all vdevs, loading each label in the process.
2277 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2278 error = vdev_open(rvd);
2279 spa_config_exit(spa, SCL_ALL, FTAG);
2284 * We need to validate the vdev labels against the configuration that
2285 * we have in hand, which is dependent on the setting of mosconfig. If
2286 * mosconfig is true then we're validating the vdev labels based on
2287 * that config. Otherwise, we're validating against the cached config
2288 * (zpool.cache) that was read when we loaded the zfs module, and then
2289 * later we will recursively call spa_load() and validate against
2292 * If we're assembling a new pool that's been split off from an
2293 * existing pool, the labels haven't yet been updated so we skip
2294 * validation for now.
2296 if (type != SPA_IMPORT_ASSEMBLE) {
2297 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2298 error = vdev_validate(rvd, mosconfig);
2299 spa_config_exit(spa, SCL_ALL, FTAG);
2304 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2305 return (SET_ERROR(ENXIO));
2309 * Find the best uberblock.
2311 vdev_uberblock_load(rvd, ub, &label);
2314 * If we weren't able to find a single valid uberblock, return failure.
2316 if (ub->ub_txg == 0) {
2318 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2322 * If the pool has an unsupported version we can't open it.
2324 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2326 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2329 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2333 * If we weren't able to find what's necessary for reading the
2334 * MOS in the label, return failure.
2336 if (label == NULL || nvlist_lookup_nvlist(label,
2337 ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) {
2339 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2344 * Update our in-core representation with the definitive values
2347 nvlist_free(spa->spa_label_features);
2348 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2354 * Look through entries in the label nvlist's features_for_read. If
2355 * there is a feature listed there which we don't understand then we
2356 * cannot open a pool.
2358 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2359 nvlist_t *unsup_feat;
2361 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2364 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2366 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2367 if (!zfeature_is_supported(nvpair_name(nvp))) {
2368 VERIFY(nvlist_add_string(unsup_feat,
2369 nvpair_name(nvp), "") == 0);
2373 if (!nvlist_empty(unsup_feat)) {
2374 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2375 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2376 nvlist_free(unsup_feat);
2377 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2381 nvlist_free(unsup_feat);
2385 * If the vdev guid sum doesn't match the uberblock, we have an
2386 * incomplete configuration. We first check to see if the pool
2387 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2388 * If it is, defer the vdev_guid_sum check till later so we
2389 * can handle missing vdevs.
2391 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
2392 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
2393 rvd->vdev_guid_sum != ub->ub_guid_sum)
2394 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2396 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2397 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2398 spa_try_repair(spa, config);
2399 spa_config_exit(spa, SCL_ALL, FTAG);
2400 nvlist_free(spa->spa_config_splitting);
2401 spa->spa_config_splitting = NULL;
2405 * Initialize internal SPA structures.
2407 spa->spa_state = POOL_STATE_ACTIVE;
2408 spa->spa_ubsync = spa->spa_uberblock;
2409 spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2410 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2411 spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2412 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2413 spa->spa_claim_max_txg = spa->spa_first_txg;
2414 spa->spa_prev_software_version = ub->ub_software_version;
2416 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2418 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2419 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2421 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2422 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2424 if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2425 boolean_t missing_feat_read = B_FALSE;
2426 nvlist_t *unsup_feat, *enabled_feat;
2428 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2429 &spa->spa_feat_for_read_obj) != 0) {
2430 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2433 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2434 &spa->spa_feat_for_write_obj) != 0) {
2435 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2438 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2439 &spa->spa_feat_desc_obj) != 0) {
2440 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2443 enabled_feat = fnvlist_alloc();
2444 unsup_feat = fnvlist_alloc();
2446 if (!spa_features_check(spa, B_FALSE,
2447 unsup_feat, enabled_feat))
2448 missing_feat_read = B_TRUE;
2450 if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) {
2451 if (!spa_features_check(spa, B_TRUE,
2452 unsup_feat, enabled_feat)) {
2453 missing_feat_write = B_TRUE;
2457 fnvlist_add_nvlist(spa->spa_load_info,
2458 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
2460 if (!nvlist_empty(unsup_feat)) {
2461 fnvlist_add_nvlist(spa->spa_load_info,
2462 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
2465 fnvlist_free(enabled_feat);
2466 fnvlist_free(unsup_feat);
2468 if (!missing_feat_read) {
2469 fnvlist_add_boolean(spa->spa_load_info,
2470 ZPOOL_CONFIG_CAN_RDONLY);
2474 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2475 * twofold: to determine whether the pool is available for
2476 * import in read-write mode and (if it is not) whether the
2477 * pool is available for import in read-only mode. If the pool
2478 * is available for import in read-write mode, it is displayed
2479 * as available in userland; if it is not available for import
2480 * in read-only mode, it is displayed as unavailable in
2481 * userland. If the pool is available for import in read-only
2482 * mode but not read-write mode, it is displayed as unavailable
2483 * in userland with a special note that the pool is actually
2484 * available for open in read-only mode.
2486 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2487 * missing a feature for write, we must first determine whether
2488 * the pool can be opened read-only before returning to
2489 * userland in order to know whether to display the
2490 * abovementioned note.
2492 if (missing_feat_read || (missing_feat_write &&
2493 spa_writeable(spa))) {
2494 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2499 * Load refcounts for ZFS features from disk into an in-memory
2500 * cache during SPA initialization.
2502 for (spa_feature_t i = 0; i < SPA_FEATURES; i++) {
2505 error = feature_get_refcount_from_disk(spa,
2506 &spa_feature_table[i], &refcount);
2508 spa->spa_feat_refcount_cache[i] = refcount;
2509 } else if (error == ENOTSUP) {
2510 spa->spa_feat_refcount_cache[i] =
2511 SPA_FEATURE_DISABLED;
2513 return (spa_vdev_err(rvd,
2514 VDEV_AUX_CORRUPT_DATA, EIO));
2519 if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) {
2520 if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG,
2521 &spa->spa_feat_enabled_txg_obj) != 0)
2522 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2525 spa->spa_is_initializing = B_TRUE;
2526 error = dsl_pool_open(spa->spa_dsl_pool);
2527 spa->spa_is_initializing = B_FALSE;
2529 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2533 nvlist_t *policy = NULL, *nvconfig;
2535 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2536 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2538 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2539 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2541 unsigned long myhostid = 0;
2543 VERIFY(nvlist_lookup_string(nvconfig,
2544 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2547 myhostid = zone_get_hostid(NULL);
2550 * We're emulating the system's hostid in userland, so
2551 * we can't use zone_get_hostid().
2553 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2554 #endif /* _KERNEL */
2555 if (check_hostid && hostid != 0 && myhostid != 0 &&
2556 hostid != myhostid) {
2557 nvlist_free(nvconfig);
2558 cmn_err(CE_WARN, "pool '%s' could not be "
2559 "loaded as it was last accessed by "
2560 "another system (host: %s hostid: 0x%lx). "
2561 "See: http://illumos.org/msg/ZFS-8000-EY",
2562 spa_name(spa), hostname,
2563 (unsigned long)hostid);
2564 return (SET_ERROR(EBADF));
2567 if (nvlist_lookup_nvlist(spa->spa_config,
2568 ZPOOL_REWIND_POLICY, &policy) == 0)
2569 VERIFY(nvlist_add_nvlist(nvconfig,
2570 ZPOOL_REWIND_POLICY, policy) == 0);
2572 spa_config_set(spa, nvconfig);
2574 spa_deactivate(spa);
2575 spa_activate(spa, orig_mode);
2577 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2580 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2581 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2582 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2584 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2587 * Load the bit that tells us to use the new accounting function
2588 * (raid-z deflation). If we have an older pool, this will not
2591 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2592 if (error != 0 && error != ENOENT)
2593 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2595 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2596 &spa->spa_creation_version);
2597 if (error != 0 && error != ENOENT)
2598 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2601 * Load the persistent error log. If we have an older pool, this will
2604 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2605 if (error != 0 && error != ENOENT)
2606 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2608 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2609 &spa->spa_errlog_scrub);
2610 if (error != 0 && error != ENOENT)
2611 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2614 * Load the history object. If we have an older pool, this
2615 * will not be present.
2617 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2618 if (error != 0 && error != ENOENT)
2619 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2622 * If we're assembling the pool from the split-off vdevs of
2623 * an existing pool, we don't want to attach the spares & cache
2628 * Load any hot spares for this pool.
2630 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2631 if (error != 0 && error != ENOENT)
2632 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2633 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2634 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2635 if (load_nvlist(spa, spa->spa_spares.sav_object,
2636 &spa->spa_spares.sav_config) != 0)
2637 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2639 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2640 spa_load_spares(spa);
2641 spa_config_exit(spa, SCL_ALL, FTAG);
2642 } else if (error == 0) {
2643 spa->spa_spares.sav_sync = B_TRUE;
2647 * Load any level 2 ARC devices for this pool.
2649 error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2650 &spa->spa_l2cache.sav_object);
2651 if (error != 0 && error != ENOENT)
2652 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2653 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2654 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2655 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2656 &spa->spa_l2cache.sav_config) != 0)
2657 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2659 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2660 spa_load_l2cache(spa);
2661 spa_config_exit(spa, SCL_ALL, FTAG);
2662 } else if (error == 0) {
2663 spa->spa_l2cache.sav_sync = B_TRUE;
2666 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2668 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2669 if (error && error != ENOENT)
2670 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2673 uint64_t autoreplace;
2675 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2676 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2677 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2678 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2679 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2680 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2681 &spa->spa_dedup_ditto);
2683 spa->spa_autoreplace = (autoreplace != 0);
2687 * If the 'autoreplace' property is set, then post a resource notifying
2688 * the ZFS DE that it should not issue any faults for unopenable
2689 * devices. We also iterate over the vdevs, and post a sysevent for any
2690 * unopenable vdevs so that the normal autoreplace handler can take
2693 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2694 spa_check_removed(spa->spa_root_vdev);
2696 * For the import case, this is done in spa_import(), because
2697 * at this point we're using the spare definitions from
2698 * the MOS config, not necessarily from the userland config.
2700 if (state != SPA_LOAD_IMPORT) {
2701 spa_aux_check_removed(&spa->spa_spares);
2702 spa_aux_check_removed(&spa->spa_l2cache);
2707 * Load the vdev state for all toplevel vdevs.
2712 * Propagate the leaf DTLs we just loaded all the way up the tree.
2714 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2715 vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2716 spa_config_exit(spa, SCL_ALL, FTAG);
2719 * Load the DDTs (dedup tables).
2721 error = ddt_load(spa);
2723 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2725 spa_update_dspace(spa);
2728 * Validate the config, using the MOS config to fill in any
2729 * information which might be missing. If we fail to validate
2730 * the config then declare the pool unfit for use. If we're
2731 * assembling a pool from a split, the log is not transferred
2734 if (type != SPA_IMPORT_ASSEMBLE) {
2737 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2738 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2740 if (!spa_config_valid(spa, nvconfig)) {
2741 nvlist_free(nvconfig);
2742 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2745 nvlist_free(nvconfig);
2748 * Now that we've validated the config, check the state of the
2749 * root vdev. If it can't be opened, it indicates one or
2750 * more toplevel vdevs are faulted.
2752 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2753 return (SET_ERROR(ENXIO));
2755 if (spa_writeable(spa) && spa_check_logs(spa)) {
2756 *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2757 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2761 if (missing_feat_write) {
2762 ASSERT(state == SPA_LOAD_TRYIMPORT);
2765 * At this point, we know that we can open the pool in
2766 * read-only mode but not read-write mode. We now have enough
2767 * information and can return to userland.
2769 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
2773 * We've successfully opened the pool, verify that we're ready
2774 * to start pushing transactions.
2776 if (state != SPA_LOAD_TRYIMPORT) {
2777 if (error = spa_load_verify(spa))
2778 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2782 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2783 spa->spa_load_max_txg == UINT64_MAX)) {
2785 int need_update = B_FALSE;
2787 ASSERT(state != SPA_LOAD_TRYIMPORT);
2790 * Claim log blocks that haven't been committed yet.
2791 * This must all happen in a single txg.
2792 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2793 * invoked from zil_claim_log_block()'s i/o done callback.
2794 * Price of rollback is that we abandon the log.
2796 spa->spa_claiming = B_TRUE;
2798 tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2799 spa_first_txg(spa));
2800 (void) dmu_objset_find(spa_name(spa),
2801 zil_claim, tx, DS_FIND_CHILDREN);
2804 spa->spa_claiming = B_FALSE;
2806 spa_set_log_state(spa, SPA_LOG_GOOD);
2807 spa->spa_sync_on = B_TRUE;
2808 txg_sync_start(spa->spa_dsl_pool);
2811 * Wait for all claims to sync. We sync up to the highest
2812 * claimed log block birth time so that claimed log blocks
2813 * don't appear to be from the future. spa_claim_max_txg
2814 * will have been set for us by either zil_check_log_chain()
2815 * (invoked from spa_check_logs()) or zil_claim() above.
2817 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2820 * If the config cache is stale, or we have uninitialized
2821 * metaslabs (see spa_vdev_add()), then update the config.
2823 * If this is a verbatim import, trust the current
2824 * in-core spa_config and update the disk labels.
2826 if (config_cache_txg != spa->spa_config_txg ||
2827 state == SPA_LOAD_IMPORT ||
2828 state == SPA_LOAD_RECOVER ||
2829 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2830 need_update = B_TRUE;
2832 for (int c = 0; c < rvd->vdev_children; c++)
2833 if (rvd->vdev_child[c]->vdev_ms_array == 0)
2834 need_update = B_TRUE;
2837 * Update the config cache asychronously in case we're the
2838 * root pool, in which case the config cache isn't writable yet.
2841 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2844 * Check all DTLs to see if anything needs resilvering.
2846 if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2847 vdev_resilver_needed(rvd, NULL, NULL))
2848 spa_async_request(spa, SPA_ASYNC_RESILVER);
2851 * Log the fact that we booted up (so that we can detect if
2852 * we rebooted in the middle of an operation).
2854 spa_history_log_version(spa, "open");
2857 * Delete any inconsistent datasets.
2859 (void) dmu_objset_find(spa_name(spa),
2860 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2863 * Clean up any stale temporary dataset userrefs.
2865 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2872 spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2874 int mode = spa->spa_mode;
2877 spa_deactivate(spa);
2879 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg - 1;
2881 spa_activate(spa, mode);
2882 spa_async_suspend(spa);
2884 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2888 * If spa_load() fails this function will try loading prior txg's. If
2889 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2890 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2891 * function will not rewind the pool and will return the same error as
2895 spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2896 uint64_t max_request, int rewind_flags)
2898 nvlist_t *loadinfo = NULL;
2899 nvlist_t *config = NULL;
2900 int load_error, rewind_error;
2901 uint64_t safe_rewind_txg;
2904 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2905 spa->spa_load_max_txg = spa->spa_load_txg;
2906 spa_set_log_state(spa, SPA_LOG_CLEAR);
2908 spa->spa_load_max_txg = max_request;
2909 if (max_request != UINT64_MAX)
2910 spa->spa_extreme_rewind = B_TRUE;
2913 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2915 if (load_error == 0)
2918 if (spa->spa_root_vdev != NULL)
2919 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2921 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2922 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2924 if (rewind_flags & ZPOOL_NEVER_REWIND) {
2925 nvlist_free(config);
2926 return (load_error);
2929 if (state == SPA_LOAD_RECOVER) {
2930 /* Price of rolling back is discarding txgs, including log */
2931 spa_set_log_state(spa, SPA_LOG_CLEAR);
2934 * If we aren't rolling back save the load info from our first
2935 * import attempt so that we can restore it after attempting
2938 loadinfo = spa->spa_load_info;
2939 spa->spa_load_info = fnvlist_alloc();
2942 spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2943 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2944 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2945 TXG_INITIAL : safe_rewind_txg;
2948 * Continue as long as we're finding errors, we're still within
2949 * the acceptable rewind range, and we're still finding uberblocks
2951 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2952 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2953 if (spa->spa_load_max_txg < safe_rewind_txg)
2954 spa->spa_extreme_rewind = B_TRUE;
2955 rewind_error = spa_load_retry(spa, state, mosconfig);
2958 spa->spa_extreme_rewind = B_FALSE;
2959 spa->spa_load_max_txg = UINT64_MAX;
2961 if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2962 spa_config_set(spa, config);
2964 if (state == SPA_LOAD_RECOVER) {
2965 ASSERT3P(loadinfo, ==, NULL);
2966 return (rewind_error);
2968 /* Store the rewind info as part of the initial load info */
2969 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
2970 spa->spa_load_info);
2972 /* Restore the initial load info */
2973 fnvlist_free(spa->spa_load_info);
2974 spa->spa_load_info = loadinfo;
2976 return (load_error);
2983 * The import case is identical to an open except that the configuration is sent
2984 * down from userland, instead of grabbed from the configuration cache. For the
2985 * case of an open, the pool configuration will exist in the
2986 * POOL_STATE_UNINITIALIZED state.
2988 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2989 * the same time open the pool, without having to keep around the spa_t in some
2993 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2997 spa_load_state_t state = SPA_LOAD_OPEN;
2999 int locked = B_FALSE;
3000 int firstopen = B_FALSE;
3005 * As disgusting as this is, we need to support recursive calls to this
3006 * function because dsl_dir_open() is called during spa_load(), and ends
3007 * up calling spa_open() again. The real fix is to figure out how to
3008 * avoid dsl_dir_open() calling this in the first place.
3010 if (mutex_owner(&spa_namespace_lock) != curthread) {
3011 mutex_enter(&spa_namespace_lock);
3015 if ((spa = spa_lookup(pool)) == NULL) {
3017 mutex_exit(&spa_namespace_lock);
3018 return (SET_ERROR(ENOENT));
3021 if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
3022 zpool_rewind_policy_t policy;
3026 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
3028 if (policy.zrp_request & ZPOOL_DO_REWIND)
3029 state = SPA_LOAD_RECOVER;
3031 spa_activate(spa, spa_mode_global);
3033 if (state != SPA_LOAD_RECOVER)
3034 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3036 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
3037 policy.zrp_request);
3039 if (error == EBADF) {
3041 * If vdev_validate() returns failure (indicated by
3042 * EBADF), it indicates that one of the vdevs indicates
3043 * that the pool has been exported or destroyed. If
3044 * this is the case, the config cache is out of sync and
3045 * we should remove the pool from the namespace.
3048 spa_deactivate(spa);
3049 spa_config_sync(spa, B_TRUE, B_TRUE);
3052 mutex_exit(&spa_namespace_lock);
3053 return (SET_ERROR(ENOENT));
3058 * We can't open the pool, but we still have useful
3059 * information: the state of each vdev after the
3060 * attempted vdev_open(). Return this to the user.
3062 if (config != NULL && spa->spa_config) {
3063 VERIFY(nvlist_dup(spa->spa_config, config,
3065 VERIFY(nvlist_add_nvlist(*config,
3066 ZPOOL_CONFIG_LOAD_INFO,
3067 spa->spa_load_info) == 0);
3070 spa_deactivate(spa);
3071 spa->spa_last_open_failed = error;
3073 mutex_exit(&spa_namespace_lock);
3079 spa_open_ref(spa, tag);
3082 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
3085 * If we've recovered the pool, pass back any information we
3086 * gathered while doing the load.
3088 if (state == SPA_LOAD_RECOVER) {
3089 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
3090 spa->spa_load_info) == 0);
3094 spa->spa_last_open_failed = 0;
3095 spa->spa_last_ubsync_txg = 0;
3096 spa->spa_load_txg = 0;
3097 mutex_exit(&spa_namespace_lock);
3101 zvol_create_minors(spa->spa_name);
3112 spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
3115 return (spa_open_common(name, spapp, tag, policy, config));
3119 spa_open(const char *name, spa_t **spapp, void *tag)
3121 return (spa_open_common(name, spapp, tag, NULL, NULL));
3125 * Lookup the given spa_t, incrementing the inject count in the process,
3126 * preventing it from being exported or destroyed.
3129 spa_inject_addref(char *name)
3133 mutex_enter(&spa_namespace_lock);
3134 if ((spa = spa_lookup(name)) == NULL) {
3135 mutex_exit(&spa_namespace_lock);
3138 spa->spa_inject_ref++;
3139 mutex_exit(&spa_namespace_lock);
3145 spa_inject_delref(spa_t *spa)
3147 mutex_enter(&spa_namespace_lock);
3148 spa->spa_inject_ref--;
3149 mutex_exit(&spa_namespace_lock);
3153 * Add spares device information to the nvlist.
3156 spa_add_spares(spa_t *spa, nvlist_t *config)
3166 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3168 if (spa->spa_spares.sav_count == 0)
3171 VERIFY(nvlist_lookup_nvlist(config,
3172 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3173 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
3174 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3176 VERIFY(nvlist_add_nvlist_array(nvroot,
3177 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3178 VERIFY(nvlist_lookup_nvlist_array(nvroot,
3179 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3182 * Go through and find any spares which have since been
3183 * repurposed as an active spare. If this is the case, update
3184 * their status appropriately.
3186 for (i = 0; i < nspares; i++) {
3187 VERIFY(nvlist_lookup_uint64(spares[i],
3188 ZPOOL_CONFIG_GUID, &guid) == 0);
3189 if (spa_spare_exists(guid, &pool, NULL) &&
3191 VERIFY(nvlist_lookup_uint64_array(
3192 spares[i], ZPOOL_CONFIG_VDEV_STATS,
3193 (uint64_t **)&vs, &vsc) == 0);
3194 vs->vs_state = VDEV_STATE_CANT_OPEN;
3195 vs->vs_aux = VDEV_AUX_SPARED;
3202 * Add l2cache device information to the nvlist, including vdev stats.
3205 spa_add_l2cache(spa_t *spa, nvlist_t *config)
3208 uint_t i, j, nl2cache;
3215 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3217 if (spa->spa_l2cache.sav_count == 0)
3220 VERIFY(nvlist_lookup_nvlist(config,
3221 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3222 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
3223 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3224 if (nl2cache != 0) {
3225 VERIFY(nvlist_add_nvlist_array(nvroot,
3226 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3227 VERIFY(nvlist_lookup_nvlist_array(nvroot,
3228 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3231 * Update level 2 cache device stats.
3234 for (i = 0; i < nl2cache; i++) {
3235 VERIFY(nvlist_lookup_uint64(l2cache[i],
3236 ZPOOL_CONFIG_GUID, &guid) == 0);
3239 for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
3241 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
3242 vd = spa->spa_l2cache.sav_vdevs[j];
3248 VERIFY(nvlist_lookup_uint64_array(l2cache[i],
3249 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
3251 vdev_get_stats(vd, vs);
3257 spa_add_feature_stats(spa_t *spa, nvlist_t *config)
3263 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3264 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3266 /* We may be unable to read features if pool is suspended. */
3267 if (spa_suspended(spa))
3270 if (spa->spa_feat_for_read_obj != 0) {
3271 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3272 spa->spa_feat_for_read_obj);
3273 zap_cursor_retrieve(&zc, &za) == 0;
3274 zap_cursor_advance(&zc)) {
3275 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3276 za.za_num_integers == 1);
3277 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3278 za.za_first_integer));
3280 zap_cursor_fini(&zc);
3283 if (spa->spa_feat_for_write_obj != 0) {
3284 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3285 spa->spa_feat_for_write_obj);
3286 zap_cursor_retrieve(&zc, &za) == 0;
3287 zap_cursor_advance(&zc)) {
3288 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3289 za.za_num_integers == 1);
3290 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3291 za.za_first_integer));
3293 zap_cursor_fini(&zc);
3297 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
3299 nvlist_free(features);
3303 spa_get_stats(const char *name, nvlist_t **config,
3304 char *altroot, size_t buflen)
3310 error = spa_open_common(name, &spa, FTAG, NULL, config);
3314 * This still leaves a window of inconsistency where the spares
3315 * or l2cache devices could change and the config would be
3316 * self-inconsistent.
3318 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3320 if (*config != NULL) {
3321 uint64_t loadtimes[2];
3323 loadtimes[0] = spa->spa_loaded_ts.tv_sec;
3324 loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
3325 VERIFY(nvlist_add_uint64_array(*config,
3326 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
3328 VERIFY(nvlist_add_uint64(*config,
3329 ZPOOL_CONFIG_ERRCOUNT,
3330 spa_get_errlog_size(spa)) == 0);
3332 if (spa_suspended(spa))
3333 VERIFY(nvlist_add_uint64(*config,
3334 ZPOOL_CONFIG_SUSPENDED,
3335 spa->spa_failmode) == 0);
3337 spa_add_spares(spa, *config);
3338 spa_add_l2cache(spa, *config);
3339 spa_add_feature_stats(spa, *config);
3344 * We want to get the alternate root even for faulted pools, so we cheat
3345 * and call spa_lookup() directly.
3349 mutex_enter(&spa_namespace_lock);
3350 spa = spa_lookup(name);
3352 spa_altroot(spa, altroot, buflen);
3356 mutex_exit(&spa_namespace_lock);
3358 spa_altroot(spa, altroot, buflen);
3363 spa_config_exit(spa, SCL_CONFIG, FTAG);
3364 spa_close(spa, FTAG);
3371 * Validate that the auxiliary device array is well formed. We must have an
3372 * array of nvlists, each which describes a valid leaf vdev. If this is an
3373 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3374 * specified, as long as they are well-formed.
3377 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
3378 spa_aux_vdev_t *sav, const char *config, uint64_t version,
3379 vdev_labeltype_t label)
3386 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3389 * It's acceptable to have no devs specified.
3391 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
3395 return (SET_ERROR(EINVAL));
3398 * Make sure the pool is formatted with a version that supports this
3401 if (spa_version(spa) < version)
3402 return (SET_ERROR(ENOTSUP));
3405 * Set the pending device list so we correctly handle device in-use
3408 sav->sav_pending = dev;
3409 sav->sav_npending = ndev;
3411 for (i = 0; i < ndev; i++) {
3412 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
3416 if (!vd->vdev_ops->vdev_op_leaf) {
3418 error = SET_ERROR(EINVAL);
3423 * The L2ARC currently only supports disk devices in
3424 * kernel context. For user-level testing, we allow it.
3427 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
3428 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
3429 error = SET_ERROR(ENOTBLK);
3436 if ((error = vdev_open(vd)) == 0 &&
3437 (error = vdev_label_init(vd, crtxg, label)) == 0) {
3438 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
3439 vd->vdev_guid) == 0);
3445 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
3452 sav->sav_pending = NULL;
3453 sav->sav_npending = 0;
3458 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
3462 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3464 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3465 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
3466 VDEV_LABEL_SPARE)) != 0) {
3470 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3471 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
3472 VDEV_LABEL_L2CACHE));
3476 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
3481 if (sav->sav_config != NULL) {
3487 * Generate new dev list by concatentating with the
3490 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
3491 &olddevs, &oldndevs) == 0);
3493 newdevs = kmem_alloc(sizeof (void *) *
3494 (ndevs + oldndevs), KM_SLEEP);
3495 for (i = 0; i < oldndevs; i++)
3496 VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
3498 for (i = 0; i < ndevs; i++)
3499 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
3502 VERIFY(nvlist_remove(sav->sav_config, config,
3503 DATA_TYPE_NVLIST_ARRAY) == 0);
3505 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3506 config, newdevs, ndevs + oldndevs) == 0);
3507 for (i = 0; i < oldndevs + ndevs; i++)
3508 nvlist_free(newdevs[i]);
3509 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
3512 * Generate a new dev list.
3514 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
3516 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
3522 * Stop and drop level 2 ARC devices
3525 spa_l2cache_drop(spa_t *spa)
3529 spa_aux_vdev_t *sav = &spa->spa_l2cache;
3531 for (i = 0; i < sav->sav_count; i++) {
3534 vd = sav->sav_vdevs[i];
3537 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
3538 pool != 0ULL && l2arc_vdev_present(vd))
3539 l2arc_remove_vdev(vd);
3547 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
3551 char *altroot = NULL;
3556 uint64_t txg = TXG_INITIAL;
3557 nvlist_t **spares, **l2cache;
3558 uint_t nspares, nl2cache;
3559 uint64_t version, obj;
3560 boolean_t has_features;
3563 * If this pool already exists, return failure.
3565 mutex_enter(&spa_namespace_lock);
3566 if (spa_lookup(pool) != NULL) {
3567 mutex_exit(&spa_namespace_lock);
3568 return (SET_ERROR(EEXIST));
3572 * Allocate a new spa_t structure.
3574 (void) nvlist_lookup_string(props,
3575 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3576 spa = spa_add(pool, NULL, altroot);
3577 spa_activate(spa, spa_mode_global);
3579 if (props && (error = spa_prop_validate(spa, props))) {
3580 spa_deactivate(spa);
3582 mutex_exit(&spa_namespace_lock);
3586 has_features = B_FALSE;
3587 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
3588 elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
3589 if (zpool_prop_feature(nvpair_name(elem)))
3590 has_features = B_TRUE;
3593 if (has_features || nvlist_lookup_uint64(props,
3594 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
3595 version = SPA_VERSION;
3597 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
3599 spa->spa_first_txg = txg;
3600 spa->spa_uberblock.ub_txg = txg - 1;
3601 spa->spa_uberblock.ub_version = version;
3602 spa->spa_ubsync = spa->spa_uberblock;
3605 * Create "The Godfather" zio to hold all async IOs
3607 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
3609 for (int i = 0; i < max_ncpus; i++) {
3610 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
3611 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
3612 ZIO_FLAG_GODFATHER);
3616 * Create the root vdev.
3618 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3620 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3622 ASSERT(error != 0 || rvd != NULL);
3623 ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3625 if (error == 0 && !zfs_allocatable_devs(nvroot))
3626 error = SET_ERROR(EINVAL);
3629 (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3630 (error = spa_validate_aux(spa, nvroot, txg,
3631 VDEV_ALLOC_ADD)) == 0) {
3632 for (int c = 0; c < rvd->vdev_children; c++) {
3633 vdev_ashift_optimize(rvd->vdev_child[c]);
3634 vdev_metaslab_set_size(rvd->vdev_child[c]);
3635 vdev_expand(rvd->vdev_child[c], txg);
3639 spa_config_exit(spa, SCL_ALL, FTAG);
3643 spa_deactivate(spa);
3645 mutex_exit(&spa_namespace_lock);
3650 * Get the list of spares, if specified.
3652 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3653 &spares, &nspares) == 0) {
3654 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3656 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3657 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3658 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3659 spa_load_spares(spa);
3660 spa_config_exit(spa, SCL_ALL, FTAG);
3661 spa->spa_spares.sav_sync = B_TRUE;
3665 * Get the list of level 2 cache devices, if specified.
3667 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3668 &l2cache, &nl2cache) == 0) {
3669 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3670 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3671 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3672 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3673 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3674 spa_load_l2cache(spa);
3675 spa_config_exit(spa, SCL_ALL, FTAG);
3676 spa->spa_l2cache.sav_sync = B_TRUE;
3679 spa->spa_is_initializing = B_TRUE;
3680 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3681 spa->spa_meta_objset = dp->dp_meta_objset;
3682 spa->spa_is_initializing = B_FALSE;
3685 * Create DDTs (dedup tables).
3689 spa_update_dspace(spa);
3691 tx = dmu_tx_create_assigned(dp, txg);
3694 * Create the pool config object.
3696 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3697 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3698 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3700 if (zap_add(spa->spa_meta_objset,
3701 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3702 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3703 cmn_err(CE_PANIC, "failed to add pool config");
3706 if (spa_version(spa) >= SPA_VERSION_FEATURES)
3707 spa_feature_create_zap_objects(spa, tx);
3709 if (zap_add(spa->spa_meta_objset,
3710 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3711 sizeof (uint64_t), 1, &version, tx) != 0) {
3712 cmn_err(CE_PANIC, "failed to add pool version");
3715 /* Newly created pools with the right version are always deflated. */
3716 if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3717 spa->spa_deflate = TRUE;
3718 if (zap_add(spa->spa_meta_objset,
3719 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3720 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3721 cmn_err(CE_PANIC, "failed to add deflate");
3726 * Create the deferred-free bpobj. Turn off compression
3727 * because sync-to-convergence takes longer if the blocksize
3730 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3731 dmu_object_set_compress(spa->spa_meta_objset, obj,
3732 ZIO_COMPRESS_OFF, tx);
3733 if (zap_add(spa->spa_meta_objset,
3734 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3735 sizeof (uint64_t), 1, &obj, tx) != 0) {
3736 cmn_err(CE_PANIC, "failed to add bpobj");
3738 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3739 spa->spa_meta_objset, obj));
3742 * Create the pool's history object.
3744 if (version >= SPA_VERSION_ZPOOL_HISTORY)
3745 spa_history_create_obj(spa, tx);
3748 * Set pool properties.
3750 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3751 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3752 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3753 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3755 if (props != NULL) {
3756 spa_configfile_set(spa, props, B_FALSE);
3757 spa_sync_props(props, tx);
3762 spa->spa_sync_on = B_TRUE;
3763 txg_sync_start(spa->spa_dsl_pool);
3766 * We explicitly wait for the first transaction to complete so that our
3767 * bean counters are appropriately updated.
3769 txg_wait_synced(spa->spa_dsl_pool, txg);
3771 spa_config_sync(spa, B_FALSE, B_TRUE);
3773 spa_history_log_version(spa, "create");
3776 * Don't count references from objsets that are already closed
3777 * and are making their way through the eviction process.
3779 spa_evicting_os_wait(spa);
3780 spa->spa_minref = refcount_count(&spa->spa_refcount);
3782 mutex_exit(&spa_namespace_lock);
3790 * Get the root pool information from the root disk, then import the root pool
3791 * during the system boot up time.
3793 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3796 spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3799 nvlist_t *nvtop, *nvroot;
3802 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3806 * Add this top-level vdev to the child array.
3808 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3810 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3812 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3815 * Put this pool's top-level vdevs into a root vdev.
3817 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3818 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3819 VDEV_TYPE_ROOT) == 0);
3820 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3821 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3822 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3826 * Replace the existing vdev_tree with the new root vdev in
3827 * this pool's configuration (remove the old, add the new).
3829 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3830 nvlist_free(nvroot);
3835 * Walk the vdev tree and see if we can find a device with "better"
3836 * configuration. A configuration is "better" if the label on that
3837 * device has a more recent txg.
3840 spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3842 for (int c = 0; c < vd->vdev_children; c++)
3843 spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3845 if (vd->vdev_ops->vdev_op_leaf) {
3849 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3853 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3857 * Do we have a better boot device?
3859 if (label_txg > *txg) {
3868 * Import a root pool.
3870 * For x86. devpath_list will consist of devid and/or physpath name of
3871 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3872 * The GRUB "findroot" command will return the vdev we should boot.
3874 * For Sparc, devpath_list consists the physpath name of the booting device
3875 * no matter the rootpool is a single device pool or a mirrored pool.
3877 * "/pci@1f,0/ide@d/disk@0,0:a"
3880 spa_import_rootpool(char *devpath, char *devid)
3883 vdev_t *rvd, *bvd, *avd = NULL;
3884 nvlist_t *config, *nvtop;
3890 * Read the label from the boot device and generate a configuration.
3892 config = spa_generate_rootconf(devpath, devid, &guid);
3893 #if defined(_OBP) && defined(_KERNEL)
3894 if (config == NULL) {
3895 if (strstr(devpath, "/iscsi/ssd") != NULL) {
3897 get_iscsi_bootpath_phy(devpath);
3898 config = spa_generate_rootconf(devpath, devid, &guid);
3902 if (config == NULL) {
3903 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
3905 return (SET_ERROR(EIO));
3908 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3910 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3912 mutex_enter(&spa_namespace_lock);
3913 if ((spa = spa_lookup(pname)) != NULL) {
3915 * Remove the existing root pool from the namespace so that we
3916 * can replace it with the correct config we just read in.
3921 spa = spa_add(pname, config, NULL);
3922 spa->spa_is_root = B_TRUE;
3923 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3926 * Build up a vdev tree based on the boot device's label config.
3928 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3930 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3931 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3932 VDEV_ALLOC_ROOTPOOL);
3933 spa_config_exit(spa, SCL_ALL, FTAG);
3935 mutex_exit(&spa_namespace_lock);
3936 nvlist_free(config);
3937 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3943 * Get the boot vdev.
3945 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3946 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3947 (u_longlong_t)guid);
3948 error = SET_ERROR(ENOENT);
3953 * Determine if there is a better boot device.
3956 spa_alt_rootvdev(rvd, &avd, &txg);
3958 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3959 "try booting from '%s'", avd->vdev_path);
3960 error = SET_ERROR(EINVAL);
3965 * If the boot device is part of a spare vdev then ensure that
3966 * we're booting off the active spare.
3968 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3969 !bvd->vdev_isspare) {
3970 cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3971 "try booting from '%s'",
3973 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3974 error = SET_ERROR(EINVAL);
3980 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3982 spa_config_exit(spa, SCL_ALL, FTAG);
3983 mutex_exit(&spa_namespace_lock);
3985 nvlist_free(config);
3991 extern int vdev_geom_read_pool_label(const char *name, nvlist_t ***configs,
3995 spa_generate_rootconf(const char *name)
3997 nvlist_t **configs, **tops;
3999 nvlist_t *best_cfg, *nvtop, *nvroot;
4008 if (vdev_geom_read_pool_label(name, &configs, &count) != 0)
4011 ASSERT3U(count, !=, 0);
4013 for (i = 0; i < count; i++) {
4016 VERIFY(nvlist_lookup_uint64(configs[i], ZPOOL_CONFIG_POOL_TXG,
4018 if (txg > best_txg) {
4020 best_cfg = configs[i];
4025 * Multi-vdev root pool configuration discovery is not supported yet.
4028 nvlist_lookup_uint64(best_cfg, ZPOOL_CONFIG_VDEV_CHILDREN, &nchildren);
4030 nvlist_lookup_uint64_array(best_cfg, ZPOOL_CONFIG_HOLE_ARRAY,
4033 tops = kmem_zalloc(nchildren * sizeof(void *), KM_SLEEP);
4034 for (i = 0; i < nchildren; i++) {
4037 if (configs[i] == NULL)
4039 VERIFY(nvlist_lookup_nvlist(configs[i], ZPOOL_CONFIG_VDEV_TREE,
4041 nvlist_dup(nvtop, &tops[i], KM_SLEEP);
4043 for (i = 0; holes != NULL && i < nholes; i++) {
4046 if (tops[holes[i]] != NULL)
4048 nvlist_alloc(&tops[holes[i]], NV_UNIQUE_NAME, KM_SLEEP);
4049 VERIFY(nvlist_add_string(tops[holes[i]], ZPOOL_CONFIG_TYPE,
4050 VDEV_TYPE_HOLE) == 0);
4051 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_ID,
4053 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_GUID,
4056 for (i = 0; i < nchildren; i++) {
4057 if (tops[i] != NULL)
4059 nvlist_alloc(&tops[i], NV_UNIQUE_NAME, KM_SLEEP);
4060 VERIFY(nvlist_add_string(tops[i], ZPOOL_CONFIG_TYPE,
4061 VDEV_TYPE_MISSING) == 0);
4062 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_ID,
4064 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_GUID,
4069 * Create pool config based on the best vdev config.
4071 nvlist_dup(best_cfg, &config, KM_SLEEP);
4074 * Put this pool's top-level vdevs into a root vdev.
4076 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4078 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4079 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
4080 VDEV_TYPE_ROOT) == 0);
4081 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
4082 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
4083 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
4084 tops, nchildren) == 0);
4087 * Replace the existing vdev_tree with the new root vdev in
4088 * this pool's configuration (remove the old, add the new).
4090 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
4093 * Drop vdev config elements that should not be present at pool level.
4095 nvlist_remove(config, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64);
4096 nvlist_remove(config, ZPOOL_CONFIG_TOP_GUID, DATA_TYPE_UINT64);
4098 for (i = 0; i < count; i++)
4099 nvlist_free(configs[i]);
4100 kmem_free(configs, count * sizeof(void *));
4101 for (i = 0; i < nchildren; i++)
4102 nvlist_free(tops[i]);
4103 kmem_free(tops, nchildren * sizeof(void *));
4104 nvlist_free(nvroot);
4109 spa_import_rootpool(const char *name)
4112 vdev_t *rvd, *bvd, *avd = NULL;
4113 nvlist_t *config, *nvtop;
4119 * Read the label from the boot device and generate a configuration.
4121 config = spa_generate_rootconf(name);
4123 mutex_enter(&spa_namespace_lock);
4124 if (config != NULL) {
4125 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
4126 &pname) == 0 && strcmp(name, pname) == 0);
4127 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg)
4130 if ((spa = spa_lookup(pname)) != NULL) {
4132 * Remove the existing root pool from the namespace so
4133 * that we can replace it with the correct config
4138 spa = spa_add(pname, config, NULL);
4141 * Set spa_ubsync.ub_version as it can be used in vdev_alloc()
4142 * via spa_version().
4144 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
4145 &spa->spa_ubsync.ub_version) != 0)
4146 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
4147 } else if ((spa = spa_lookup(name)) == NULL) {
4148 mutex_exit(&spa_namespace_lock);
4149 nvlist_free(config);
4150 cmn_err(CE_NOTE, "Cannot find the pool label for '%s'",
4154 VERIFY(nvlist_dup(spa->spa_config, &config, KM_SLEEP) == 0);
4156 spa->spa_is_root = B_TRUE;
4157 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
4160 * Build up a vdev tree based on the boot device's label config.
4162 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4164 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4165 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
4166 VDEV_ALLOC_ROOTPOOL);
4167 spa_config_exit(spa, SCL_ALL, FTAG);
4169 mutex_exit(&spa_namespace_lock);
4170 nvlist_free(config);
4171 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
4176 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4178 spa_config_exit(spa, SCL_ALL, FTAG);
4179 mutex_exit(&spa_namespace_lock);
4181 nvlist_free(config);
4189 * Import a non-root pool into the system.
4192 spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
4195 char *altroot = NULL;
4196 spa_load_state_t state = SPA_LOAD_IMPORT;
4197 zpool_rewind_policy_t policy;
4198 uint64_t mode = spa_mode_global;
4199 uint64_t readonly = B_FALSE;
4202 nvlist_t **spares, **l2cache;
4203 uint_t nspares, nl2cache;
4206 * If a pool with this name exists, return failure.
4208 mutex_enter(&spa_namespace_lock);
4209 if (spa_lookup(pool) != NULL) {
4210 mutex_exit(&spa_namespace_lock);
4211 return (SET_ERROR(EEXIST));
4215 * Create and initialize the spa structure.
4217 (void) nvlist_lookup_string(props,
4218 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4219 (void) nvlist_lookup_uint64(props,
4220 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
4223 spa = spa_add(pool, config, altroot);
4224 spa->spa_import_flags = flags;
4227 * Verbatim import - Take a pool and insert it into the namespace
4228 * as if it had been loaded at boot.
4230 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
4232 spa_configfile_set(spa, props, B_FALSE);
4234 spa_config_sync(spa, B_FALSE, B_TRUE);
4236 mutex_exit(&spa_namespace_lock);
4240 spa_activate(spa, mode);
4243 * Don't start async tasks until we know everything is healthy.
4245 spa_async_suspend(spa);
4247 zpool_get_rewind_policy(config, &policy);
4248 if (policy.zrp_request & ZPOOL_DO_REWIND)
4249 state = SPA_LOAD_RECOVER;
4252 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
4253 * because the user-supplied config is actually the one to trust when
4256 if (state != SPA_LOAD_RECOVER)
4257 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
4259 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
4260 policy.zrp_request);
4263 * Propagate anything learned while loading the pool and pass it
4264 * back to caller (i.e. rewind info, missing devices, etc).
4266 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4267 spa->spa_load_info) == 0);
4269 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4271 * Toss any existing sparelist, as it doesn't have any validity
4272 * anymore, and conflicts with spa_has_spare().
4274 if (spa->spa_spares.sav_config) {
4275 nvlist_free(spa->spa_spares.sav_config);
4276 spa->spa_spares.sav_config = NULL;
4277 spa_load_spares(spa);
4279 if (spa->spa_l2cache.sav_config) {
4280 nvlist_free(spa->spa_l2cache.sav_config);
4281 spa->spa_l2cache.sav_config = NULL;
4282 spa_load_l2cache(spa);
4285 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4288 error = spa_validate_aux(spa, nvroot, -1ULL,
4291 error = spa_validate_aux(spa, nvroot, -1ULL,
4292 VDEV_ALLOC_L2CACHE);
4293 spa_config_exit(spa, SCL_ALL, FTAG);
4296 spa_configfile_set(spa, props, B_FALSE);
4298 if (error != 0 || (props && spa_writeable(spa) &&
4299 (error = spa_prop_set(spa, props)))) {
4301 spa_deactivate(spa);
4303 mutex_exit(&spa_namespace_lock);
4307 spa_async_resume(spa);
4310 * Override any spares and level 2 cache devices as specified by
4311 * the user, as these may have correct device names/devids, etc.
4313 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
4314 &spares, &nspares) == 0) {
4315 if (spa->spa_spares.sav_config)
4316 VERIFY(nvlist_remove(spa->spa_spares.sav_config,
4317 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
4319 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
4320 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4321 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
4322 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
4323 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4324 spa_load_spares(spa);
4325 spa_config_exit(spa, SCL_ALL, FTAG);
4326 spa->spa_spares.sav_sync = B_TRUE;
4328 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
4329 &l2cache, &nl2cache) == 0) {
4330 if (spa->spa_l2cache.sav_config)
4331 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
4332 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
4334 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
4335 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4336 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
4337 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
4338 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4339 spa_load_l2cache(spa);
4340 spa_config_exit(spa, SCL_ALL, FTAG);
4341 spa->spa_l2cache.sav_sync = B_TRUE;
4345 * Check for any removed devices.
4347 if (spa->spa_autoreplace) {
4348 spa_aux_check_removed(&spa->spa_spares);
4349 spa_aux_check_removed(&spa->spa_l2cache);
4352 if (spa_writeable(spa)) {
4354 * Update the config cache to include the newly-imported pool.
4356 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4360 * It's possible that the pool was expanded while it was exported.
4361 * We kick off an async task to handle this for us.
4363 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
4365 mutex_exit(&spa_namespace_lock);
4366 spa_history_log_version(spa, "import");
4370 zvol_create_minors(pool);
4377 spa_tryimport(nvlist_t *tryconfig)
4379 nvlist_t *config = NULL;
4385 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
4388 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
4392 * Create and initialize the spa structure.
4394 mutex_enter(&spa_namespace_lock);
4395 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
4396 spa_activate(spa, FREAD);
4399 * Pass off the heavy lifting to spa_load().
4400 * Pass TRUE for mosconfig because the user-supplied config
4401 * is actually the one to trust when doing an import.
4403 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
4406 * If 'tryconfig' was at least parsable, return the current config.
4408 if (spa->spa_root_vdev != NULL) {
4409 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4410 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
4412 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4414 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
4415 spa->spa_uberblock.ub_timestamp) == 0);
4416 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4417 spa->spa_load_info) == 0);
4420 * If the bootfs property exists on this pool then we
4421 * copy it out so that external consumers can tell which
4422 * pools are bootable.
4424 if ((!error || error == EEXIST) && spa->spa_bootfs) {
4425 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4428 * We have to play games with the name since the
4429 * pool was opened as TRYIMPORT_NAME.
4431 if (dsl_dsobj_to_dsname(spa_name(spa),
4432 spa->spa_bootfs, tmpname) == 0) {
4434 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4436 cp = strchr(tmpname, '/');
4438 (void) strlcpy(dsname, tmpname,
4441 (void) snprintf(dsname, MAXPATHLEN,
4442 "%s/%s", poolname, ++cp);
4444 VERIFY(nvlist_add_string(config,
4445 ZPOOL_CONFIG_BOOTFS, dsname) == 0);
4446 kmem_free(dsname, MAXPATHLEN);
4448 kmem_free(tmpname, MAXPATHLEN);
4452 * Add the list of hot spares and level 2 cache devices.
4454 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4455 spa_add_spares(spa, config);
4456 spa_add_l2cache(spa, config);
4457 spa_config_exit(spa, SCL_CONFIG, FTAG);
4461 spa_deactivate(spa);
4463 mutex_exit(&spa_namespace_lock);
4469 * Pool export/destroy
4471 * The act of destroying or exporting a pool is very simple. We make sure there
4472 * is no more pending I/O and any references to the pool are gone. Then, we
4473 * update the pool state and sync all the labels to disk, removing the
4474 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4475 * we don't sync the labels or remove the configuration cache.
4478 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
4479 boolean_t force, boolean_t hardforce)
4486 if (!(spa_mode_global & FWRITE))
4487 return (SET_ERROR(EROFS));
4489 mutex_enter(&spa_namespace_lock);
4490 if ((spa = spa_lookup(pool)) == NULL) {
4491 mutex_exit(&spa_namespace_lock);
4492 return (SET_ERROR(ENOENT));
4496 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4497 * reacquire the namespace lock, and see if we can export.
4499 spa_open_ref(spa, FTAG);
4500 mutex_exit(&spa_namespace_lock);
4501 spa_async_suspend(spa);
4502 mutex_enter(&spa_namespace_lock);
4503 spa_close(spa, FTAG);
4506 * The pool will be in core if it's openable,
4507 * in which case we can modify its state.
4509 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
4511 * Objsets may be open only because they're dirty, so we
4512 * have to force it to sync before checking spa_refcnt.
4514 txg_wait_synced(spa->spa_dsl_pool, 0);
4515 spa_evicting_os_wait(spa);
4518 * A pool cannot be exported or destroyed if there are active
4519 * references. If we are resetting a pool, allow references by
4520 * fault injection handlers.
4522 if (!spa_refcount_zero(spa) ||
4523 (spa->spa_inject_ref != 0 &&
4524 new_state != POOL_STATE_UNINITIALIZED)) {
4525 spa_async_resume(spa);
4526 mutex_exit(&spa_namespace_lock);
4527 return (SET_ERROR(EBUSY));
4531 * A pool cannot be exported if it has an active shared spare.
4532 * This is to prevent other pools stealing the active spare
4533 * from an exported pool. At user's own will, such pool can
4534 * be forcedly exported.
4536 if (!force && new_state == POOL_STATE_EXPORTED &&
4537 spa_has_active_shared_spare(spa)) {
4538 spa_async_resume(spa);
4539 mutex_exit(&spa_namespace_lock);
4540 return (SET_ERROR(EXDEV));
4544 * We want this to be reflected on every label,
4545 * so mark them all dirty. spa_unload() will do the
4546 * final sync that pushes these changes out.
4548 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
4549 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4550 spa->spa_state = new_state;
4551 spa->spa_final_txg = spa_last_synced_txg(spa) +
4553 vdev_config_dirty(spa->spa_root_vdev);
4554 spa_config_exit(spa, SCL_ALL, FTAG);
4558 spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
4560 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4562 spa_deactivate(spa);
4565 if (oldconfig && spa->spa_config)
4566 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
4568 if (new_state != POOL_STATE_UNINITIALIZED) {
4570 spa_config_sync(spa, B_TRUE, B_TRUE);
4573 mutex_exit(&spa_namespace_lock);
4579 * Destroy a storage pool.
4582 spa_destroy(char *pool)
4584 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
4589 * Export a storage pool.
4592 spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
4593 boolean_t hardforce)
4595 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
4600 * Similar to spa_export(), this unloads the spa_t without actually removing it
4601 * from the namespace in any way.
4604 spa_reset(char *pool)
4606 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
4611 * ==========================================================================
4612 * Device manipulation
4613 * ==========================================================================
4617 * Add a device to a storage pool.
4620 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
4624 vdev_t *rvd = spa->spa_root_vdev;
4626 nvlist_t **spares, **l2cache;
4627 uint_t nspares, nl2cache;
4629 ASSERT(spa_writeable(spa));
4631 txg = spa_vdev_enter(spa);
4633 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
4634 VDEV_ALLOC_ADD)) != 0)
4635 return (spa_vdev_exit(spa, NULL, txg, error));
4637 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */
4639 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
4643 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
4647 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
4648 return (spa_vdev_exit(spa, vd, txg, EINVAL));
4650 if (vd->vdev_children != 0 &&
4651 (error = vdev_create(vd, txg, B_FALSE)) != 0)
4652 return (spa_vdev_exit(spa, vd, txg, error));
4655 * We must validate the spares and l2cache devices after checking the
4656 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
4658 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
4659 return (spa_vdev_exit(spa, vd, txg, error));
4662 * Transfer each new top-level vdev from vd to rvd.
4664 for (int c = 0; c < vd->vdev_children; c++) {
4667 * Set the vdev id to the first hole, if one exists.
4669 for (id = 0; id < rvd->vdev_children; id++) {
4670 if (rvd->vdev_child[id]->vdev_ishole) {
4671 vdev_free(rvd->vdev_child[id]);
4675 tvd = vd->vdev_child[c];
4676 vdev_remove_child(vd, tvd);
4678 vdev_add_child(rvd, tvd);
4679 vdev_config_dirty(tvd);
4683 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
4684 ZPOOL_CONFIG_SPARES);
4685 spa_load_spares(spa);
4686 spa->spa_spares.sav_sync = B_TRUE;
4689 if (nl2cache != 0) {
4690 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
4691 ZPOOL_CONFIG_L2CACHE);
4692 spa_load_l2cache(spa);
4693 spa->spa_l2cache.sav_sync = B_TRUE;
4697 * We have to be careful when adding new vdevs to an existing pool.
4698 * If other threads start allocating from these vdevs before we
4699 * sync the config cache, and we lose power, then upon reboot we may
4700 * fail to open the pool because there are DVAs that the config cache
4701 * can't translate. Therefore, we first add the vdevs without
4702 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4703 * and then let spa_config_update() initialize the new metaslabs.
4705 * spa_load() checks for added-but-not-initialized vdevs, so that
4706 * if we lose power at any point in this sequence, the remaining
4707 * steps will be completed the next time we load the pool.
4709 (void) spa_vdev_exit(spa, vd, txg, 0);
4711 mutex_enter(&spa_namespace_lock);
4712 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4713 mutex_exit(&spa_namespace_lock);
4719 * Attach a device to a mirror. The arguments are the path to any device
4720 * in the mirror, and the nvroot for the new device. If the path specifies
4721 * a device that is not mirrored, we automatically insert the mirror vdev.
4723 * If 'replacing' is specified, the new device is intended to replace the
4724 * existing device; in this case the two devices are made into their own
4725 * mirror using the 'replacing' vdev, which is functionally identical to
4726 * the mirror vdev (it actually reuses all the same ops) but has a few
4727 * extra rules: you can't attach to it after it's been created, and upon
4728 * completion of resilvering, the first disk (the one being replaced)
4729 * is automatically detached.
4732 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
4734 uint64_t txg, dtl_max_txg;
4735 vdev_t *rvd = spa->spa_root_vdev;
4736 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
4738 char *oldvdpath, *newvdpath;
4742 ASSERT(spa_writeable(spa));
4744 txg = spa_vdev_enter(spa);
4746 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
4749 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4751 if (!oldvd->vdev_ops->vdev_op_leaf)
4752 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4754 pvd = oldvd->vdev_parent;
4756 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
4757 VDEV_ALLOC_ATTACH)) != 0)
4758 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4760 if (newrootvd->vdev_children != 1)
4761 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4763 newvd = newrootvd->vdev_child[0];
4765 if (!newvd->vdev_ops->vdev_op_leaf)
4766 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4768 if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
4769 return (spa_vdev_exit(spa, newrootvd, txg, error));
4772 * Spares can't replace logs
4774 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
4775 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4779 * For attach, the only allowable parent is a mirror or the root
4782 if (pvd->vdev_ops != &vdev_mirror_ops &&
4783 pvd->vdev_ops != &vdev_root_ops)
4784 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4786 pvops = &vdev_mirror_ops;
4789 * Active hot spares can only be replaced by inactive hot
4792 if (pvd->vdev_ops == &vdev_spare_ops &&
4793 oldvd->vdev_isspare &&
4794 !spa_has_spare(spa, newvd->vdev_guid))
4795 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4798 * If the source is a hot spare, and the parent isn't already a
4799 * spare, then we want to create a new hot spare. Otherwise, we
4800 * want to create a replacing vdev. The user is not allowed to
4801 * attach to a spared vdev child unless the 'isspare' state is
4802 * the same (spare replaces spare, non-spare replaces
4805 if (pvd->vdev_ops == &vdev_replacing_ops &&
4806 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
4807 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4808 } else if (pvd->vdev_ops == &vdev_spare_ops &&
4809 newvd->vdev_isspare != oldvd->vdev_isspare) {
4810 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4813 if (newvd->vdev_isspare)
4814 pvops = &vdev_spare_ops;
4816 pvops = &vdev_replacing_ops;
4820 * Make sure the new device is big enough.
4822 if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
4823 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
4826 * The new device cannot have a higher alignment requirement
4827 * than the top-level vdev.
4829 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4830 return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4833 * If this is an in-place replacement, update oldvd's path and devid
4834 * to make it distinguishable from newvd, and unopenable from now on.
4836 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4837 spa_strfree(oldvd->vdev_path);
4838 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4840 (void) sprintf(oldvd->vdev_path, "%s/%s",
4841 newvd->vdev_path, "old");
4842 if (oldvd->vdev_devid != NULL) {
4843 spa_strfree(oldvd->vdev_devid);
4844 oldvd->vdev_devid = NULL;
4848 /* mark the device being resilvered */
4849 newvd->vdev_resilver_txg = txg;
4852 * If the parent is not a mirror, or if we're replacing, insert the new
4853 * mirror/replacing/spare vdev above oldvd.
4855 if (pvd->vdev_ops != pvops)
4856 pvd = vdev_add_parent(oldvd, pvops);
4858 ASSERT(pvd->vdev_top->vdev_parent == rvd);
4859 ASSERT(pvd->vdev_ops == pvops);
4860 ASSERT(oldvd->vdev_parent == pvd);
4863 * Extract the new device from its root and add it to pvd.
4865 vdev_remove_child(newrootvd, newvd);
4866 newvd->vdev_id = pvd->vdev_children;
4867 newvd->vdev_crtxg = oldvd->vdev_crtxg;
4868 vdev_add_child(pvd, newvd);
4870 tvd = newvd->vdev_top;
4871 ASSERT(pvd->vdev_top == tvd);
4872 ASSERT(tvd->vdev_parent == rvd);
4874 vdev_config_dirty(tvd);
4877 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4878 * for any dmu_sync-ed blocks. It will propagate upward when
4879 * spa_vdev_exit() calls vdev_dtl_reassess().
4881 dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4883 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4884 dtl_max_txg - TXG_INITIAL);
4886 if (newvd->vdev_isspare) {
4887 spa_spare_activate(newvd);
4888 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
4891 oldvdpath = spa_strdup(oldvd->vdev_path);
4892 newvdpath = spa_strdup(newvd->vdev_path);
4893 newvd_isspare = newvd->vdev_isspare;
4896 * Mark newvd's DTL dirty in this txg.
4898 vdev_dirty(tvd, VDD_DTL, newvd, txg);
4901 * Schedule the resilver to restart in the future. We do this to
4902 * ensure that dmu_sync-ed blocks have been stitched into the
4903 * respective datasets.
4905 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4910 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4912 spa_history_log_internal(spa, "vdev attach", NULL,
4913 "%s vdev=%s %s vdev=%s",
4914 replacing && newvd_isspare ? "spare in" :
4915 replacing ? "replace" : "attach", newvdpath,
4916 replacing ? "for" : "to", oldvdpath);
4918 spa_strfree(oldvdpath);
4919 spa_strfree(newvdpath);
4921 if (spa->spa_bootfs)
4922 spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
4928 * Detach a device from a mirror or replacing vdev.
4930 * If 'replace_done' is specified, only detach if the parent
4931 * is a replacing vdev.
4934 spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4938 vdev_t *rvd = spa->spa_root_vdev;
4939 vdev_t *vd, *pvd, *cvd, *tvd;
4940 boolean_t unspare = B_FALSE;
4941 uint64_t unspare_guid = 0;
4944 ASSERT(spa_writeable(spa));
4946 txg = spa_vdev_enter(spa);
4948 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4951 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4953 if (!vd->vdev_ops->vdev_op_leaf)
4954 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4956 pvd = vd->vdev_parent;
4959 * If the parent/child relationship is not as expected, don't do it.
4960 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4961 * vdev that's replacing B with C. The user's intent in replacing
4962 * is to go from M(A,B) to M(A,C). If the user decides to cancel
4963 * the replace by detaching C, the expected behavior is to end up
4964 * M(A,B). But suppose that right after deciding to detach C,
4965 * the replacement of B completes. We would have M(A,C), and then
4966 * ask to detach C, which would leave us with just A -- not what
4967 * the user wanted. To prevent this, we make sure that the
4968 * parent/child relationship hasn't changed -- in this example,
4969 * that C's parent is still the replacing vdev R.
4971 if (pvd->vdev_guid != pguid && pguid != 0)
4972 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4975 * Only 'replacing' or 'spare' vdevs can be replaced.
4977 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4978 pvd->vdev_ops != &vdev_spare_ops)
4979 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4981 ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4982 spa_version(spa) >= SPA_VERSION_SPARES);
4985 * Only mirror, replacing, and spare vdevs support detach.
4987 if (pvd->vdev_ops != &vdev_replacing_ops &&
4988 pvd->vdev_ops != &vdev_mirror_ops &&
4989 pvd->vdev_ops != &vdev_spare_ops)
4990 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4993 * If this device has the only valid copy of some data,
4994 * we cannot safely detach it.
4996 if (vdev_dtl_required(vd))
4997 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4999 ASSERT(pvd->vdev_children >= 2);
5002 * If we are detaching the second disk from a replacing vdev, then
5003 * check to see if we changed the original vdev's path to have "/old"
5004 * at the end in spa_vdev_attach(). If so, undo that change now.
5006 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
5007 vd->vdev_path != NULL) {
5008 size_t len = strlen(vd->vdev_path);
5010 for (int c = 0; c < pvd->vdev_children; c++) {
5011 cvd = pvd->vdev_child[c];
5013 if (cvd == vd || cvd->vdev_path == NULL)
5016 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
5017 strcmp(cvd->vdev_path + len, "/old") == 0) {
5018 spa_strfree(cvd->vdev_path);
5019 cvd->vdev_path = spa_strdup(vd->vdev_path);
5026 * If we are detaching the original disk from a spare, then it implies
5027 * that the spare should become a real disk, and be removed from the
5028 * active spare list for the pool.
5030 if (pvd->vdev_ops == &vdev_spare_ops &&
5032 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
5036 * Erase the disk labels so the disk can be used for other things.
5037 * This must be done after all other error cases are handled,
5038 * but before we disembowel vd (so we can still do I/O to it).
5039 * But if we can't do it, don't treat the error as fatal --
5040 * it may be that the unwritability of the disk is the reason
5041 * it's being detached!
5043 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5046 * Remove vd from its parent and compact the parent's children.
5048 vdev_remove_child(pvd, vd);
5049 vdev_compact_children(pvd);
5052 * Remember one of the remaining children so we can get tvd below.
5054 cvd = pvd->vdev_child[pvd->vdev_children - 1];
5057 * If we need to remove the remaining child from the list of hot spares,
5058 * do it now, marking the vdev as no longer a spare in the process.
5059 * We must do this before vdev_remove_parent(), because that can
5060 * change the GUID if it creates a new toplevel GUID. For a similar
5061 * reason, we must remove the spare now, in the same txg as the detach;
5062 * otherwise someone could attach a new sibling, change the GUID, and
5063 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
5066 ASSERT(cvd->vdev_isspare);
5067 spa_spare_remove(cvd);
5068 unspare_guid = cvd->vdev_guid;
5069 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
5070 cvd->vdev_unspare = B_TRUE;
5074 * If the parent mirror/replacing vdev only has one child,
5075 * the parent is no longer needed. Remove it from the tree.
5077 if (pvd->vdev_children == 1) {
5078 if (pvd->vdev_ops == &vdev_spare_ops)
5079 cvd->vdev_unspare = B_FALSE;
5080 vdev_remove_parent(cvd);
5085 * We don't set tvd until now because the parent we just removed
5086 * may have been the previous top-level vdev.
5088 tvd = cvd->vdev_top;
5089 ASSERT(tvd->vdev_parent == rvd);
5092 * Reevaluate the parent vdev state.
5094 vdev_propagate_state(cvd);
5097 * If the 'autoexpand' property is set on the pool then automatically
5098 * try to expand the size of the pool. For example if the device we
5099 * just detached was smaller than the others, it may be possible to
5100 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
5101 * first so that we can obtain the updated sizes of the leaf vdevs.
5103 if (spa->spa_autoexpand) {
5105 vdev_expand(tvd, txg);
5108 vdev_config_dirty(tvd);
5111 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
5112 * vd->vdev_detached is set and free vd's DTL object in syncing context.
5113 * But first make sure we're not on any *other* txg's DTL list, to
5114 * prevent vd from being accessed after it's freed.
5116 vdpath = spa_strdup(vd->vdev_path);
5117 for (int t = 0; t < TXG_SIZE; t++)
5118 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
5119 vd->vdev_detached = B_TRUE;
5120 vdev_dirty(tvd, VDD_DTL, vd, txg);
5122 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
5124 /* hang on to the spa before we release the lock */
5125 spa_open_ref(spa, FTAG);
5127 error = spa_vdev_exit(spa, vd, txg, 0);
5129 spa_history_log_internal(spa, "detach", NULL,
5131 spa_strfree(vdpath);
5134 * If this was the removal of the original device in a hot spare vdev,
5135 * then we want to go through and remove the device from the hot spare
5136 * list of every other pool.
5139 spa_t *altspa = NULL;
5141 mutex_enter(&spa_namespace_lock);
5142 while ((altspa = spa_next(altspa)) != NULL) {
5143 if (altspa->spa_state != POOL_STATE_ACTIVE ||
5147 spa_open_ref(altspa, FTAG);
5148 mutex_exit(&spa_namespace_lock);
5149 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
5150 mutex_enter(&spa_namespace_lock);
5151 spa_close(altspa, FTAG);
5153 mutex_exit(&spa_namespace_lock);
5155 /* search the rest of the vdevs for spares to remove */
5156 spa_vdev_resilver_done(spa);
5159 /* all done with the spa; OK to release */
5160 mutex_enter(&spa_namespace_lock);
5161 spa_close(spa, FTAG);
5162 mutex_exit(&spa_namespace_lock);
5168 * Split a set of devices from their mirrors, and create a new pool from them.
5171 spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
5172 nvlist_t *props, boolean_t exp)
5175 uint64_t txg, *glist;
5177 uint_t c, children, lastlog;
5178 nvlist_t **child, *nvl, *tmp;
5180 char *altroot = NULL;
5181 vdev_t *rvd, **vml = NULL; /* vdev modify list */
5182 boolean_t activate_slog;
5184 ASSERT(spa_writeable(spa));
5186 txg = spa_vdev_enter(spa);
5188 /* clear the log and flush everything up to now */
5189 activate_slog = spa_passivate_log(spa);
5190 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5191 error = spa_offline_log(spa);
5192 txg = spa_vdev_config_enter(spa);
5195 spa_activate_log(spa);
5198 return (spa_vdev_exit(spa, NULL, txg, error));
5200 /* check new spa name before going any further */
5201 if (spa_lookup(newname) != NULL)
5202 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
5205 * scan through all the children to ensure they're all mirrors
5207 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
5208 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
5210 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5212 /* first, check to ensure we've got the right child count */
5213 rvd = spa->spa_root_vdev;
5215 for (c = 0; c < rvd->vdev_children; c++) {
5216 vdev_t *vd = rvd->vdev_child[c];
5218 /* don't count the holes & logs as children */
5219 if (vd->vdev_islog || vd->vdev_ishole) {
5227 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
5228 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5230 /* next, ensure no spare or cache devices are part of the split */
5231 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
5232 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
5233 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5235 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
5236 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
5238 /* then, loop over each vdev and validate it */
5239 for (c = 0; c < children; c++) {
5240 uint64_t is_hole = 0;
5242 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
5246 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
5247 spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
5250 error = SET_ERROR(EINVAL);
5255 /* which disk is going to be split? */
5256 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
5258 error = SET_ERROR(EINVAL);
5262 /* look it up in the spa */
5263 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
5264 if (vml[c] == NULL) {
5265 error = SET_ERROR(ENODEV);
5269 /* make sure there's nothing stopping the split */
5270 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
5271 vml[c]->vdev_islog ||
5272 vml[c]->vdev_ishole ||
5273 vml[c]->vdev_isspare ||
5274 vml[c]->vdev_isl2cache ||
5275 !vdev_writeable(vml[c]) ||
5276 vml[c]->vdev_children != 0 ||
5277 vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
5278 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
5279 error = SET_ERROR(EINVAL);
5283 if (vdev_dtl_required(vml[c])) {
5284 error = SET_ERROR(EBUSY);
5288 /* we need certain info from the top level */
5289 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
5290 vml[c]->vdev_top->vdev_ms_array) == 0);
5291 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
5292 vml[c]->vdev_top->vdev_ms_shift) == 0);
5293 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
5294 vml[c]->vdev_top->vdev_asize) == 0);
5295 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
5296 vml[c]->vdev_top->vdev_ashift) == 0);
5300 kmem_free(vml, children * sizeof (vdev_t *));
5301 kmem_free(glist, children * sizeof (uint64_t));
5302 return (spa_vdev_exit(spa, NULL, txg, error));
5305 /* stop writers from using the disks */
5306 for (c = 0; c < children; c++) {
5308 vml[c]->vdev_offline = B_TRUE;
5310 vdev_reopen(spa->spa_root_vdev);
5313 * Temporarily record the splitting vdevs in the spa config. This
5314 * will disappear once the config is regenerated.
5316 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5317 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
5318 glist, children) == 0);
5319 kmem_free(glist, children * sizeof (uint64_t));
5321 mutex_enter(&spa->spa_props_lock);
5322 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
5324 mutex_exit(&spa->spa_props_lock);
5325 spa->spa_config_splitting = nvl;
5326 vdev_config_dirty(spa->spa_root_vdev);
5328 /* configure and create the new pool */
5329 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
5330 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
5331 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
5332 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5333 spa_version(spa)) == 0);
5334 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
5335 spa->spa_config_txg) == 0);
5336 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
5337 spa_generate_guid(NULL)) == 0);
5338 (void) nvlist_lookup_string(props,
5339 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
5341 /* add the new pool to the namespace */
5342 newspa = spa_add(newname, config, altroot);
5343 newspa->spa_config_txg = spa->spa_config_txg;
5344 spa_set_log_state(newspa, SPA_LOG_CLEAR);
5346 /* release the spa config lock, retaining the namespace lock */
5347 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5349 if (zio_injection_enabled)
5350 zio_handle_panic_injection(spa, FTAG, 1);
5352 spa_activate(newspa, spa_mode_global);
5353 spa_async_suspend(newspa);
5356 /* mark that we are creating new spa by splitting */
5357 newspa->spa_splitting_newspa = B_TRUE;
5359 /* create the new pool from the disks of the original pool */
5360 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
5362 newspa->spa_splitting_newspa = B_FALSE;
5367 /* if that worked, generate a real config for the new pool */
5368 if (newspa->spa_root_vdev != NULL) {
5369 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
5370 NV_UNIQUE_NAME, KM_SLEEP) == 0);
5371 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
5372 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
5373 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
5378 if (props != NULL) {
5379 spa_configfile_set(newspa, props, B_FALSE);
5380 error = spa_prop_set(newspa, props);
5385 /* flush everything */
5386 txg = spa_vdev_config_enter(newspa);
5387 vdev_config_dirty(newspa->spa_root_vdev);
5388 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
5390 if (zio_injection_enabled)
5391 zio_handle_panic_injection(spa, FTAG, 2);
5393 spa_async_resume(newspa);
5395 /* finally, update the original pool's config */
5396 txg = spa_vdev_config_enter(spa);
5397 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
5398 error = dmu_tx_assign(tx, TXG_WAIT);
5401 for (c = 0; c < children; c++) {
5402 if (vml[c] != NULL) {
5405 spa_history_log_internal(spa, "detach", tx,
5406 "vdev=%s", vml[c]->vdev_path);
5410 vdev_config_dirty(spa->spa_root_vdev);
5411 spa->spa_config_splitting = NULL;
5415 (void) spa_vdev_exit(spa, NULL, txg, 0);
5417 if (zio_injection_enabled)
5418 zio_handle_panic_injection(spa, FTAG, 3);
5420 /* split is complete; log a history record */
5421 spa_history_log_internal(newspa, "split", NULL,
5422 "from pool %s", spa_name(spa));
5424 kmem_free(vml, children * sizeof (vdev_t *));
5426 /* if we're not going to mount the filesystems in userland, export */
5428 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
5435 spa_deactivate(newspa);
5438 txg = spa_vdev_config_enter(spa);
5440 /* re-online all offlined disks */
5441 for (c = 0; c < children; c++) {
5443 vml[c]->vdev_offline = B_FALSE;
5445 vdev_reopen(spa->spa_root_vdev);
5447 nvlist_free(spa->spa_config_splitting);
5448 spa->spa_config_splitting = NULL;
5449 (void) spa_vdev_exit(spa, NULL, txg, error);
5451 kmem_free(vml, children * sizeof (vdev_t *));
5456 spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
5458 for (int i = 0; i < count; i++) {
5461 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
5464 if (guid == target_guid)
5472 spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
5473 nvlist_t *dev_to_remove)
5475 nvlist_t **newdev = NULL;
5478 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
5480 for (int i = 0, j = 0; i < count; i++) {
5481 if (dev[i] == dev_to_remove)
5483 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
5486 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
5487 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
5489 for (int i = 0; i < count - 1; i++)
5490 nvlist_free(newdev[i]);
5493 kmem_free(newdev, (count - 1) * sizeof (void *));
5497 * Evacuate the device.
5500 spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
5505 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5506 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5507 ASSERT(vd == vd->vdev_top);
5510 * Evacuate the device. We don't hold the config lock as writer
5511 * since we need to do I/O but we do keep the
5512 * spa_namespace_lock held. Once this completes the device
5513 * should no longer have any blocks allocated on it.
5515 if (vd->vdev_islog) {
5516 if (vd->vdev_stat.vs_alloc != 0)
5517 error = spa_offline_log(spa);
5519 error = SET_ERROR(ENOTSUP);
5526 * The evacuation succeeded. Remove any remaining MOS metadata
5527 * associated with this vdev, and wait for these changes to sync.
5529 ASSERT0(vd->vdev_stat.vs_alloc);
5530 txg = spa_vdev_config_enter(spa);
5531 vd->vdev_removing = B_TRUE;
5532 vdev_dirty_leaves(vd, VDD_DTL, txg);
5533 vdev_config_dirty(vd);
5534 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5540 * Complete the removal by cleaning up the namespace.
5543 spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
5545 vdev_t *rvd = spa->spa_root_vdev;
5546 uint64_t id = vd->vdev_id;
5547 boolean_t last_vdev = (id == (rvd->vdev_children - 1));
5549 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5550 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5551 ASSERT(vd == vd->vdev_top);
5554 * Only remove any devices which are empty.
5556 if (vd->vdev_stat.vs_alloc != 0)
5559 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5561 if (list_link_active(&vd->vdev_state_dirty_node))
5562 vdev_state_clean(vd);
5563 if (list_link_active(&vd->vdev_config_dirty_node))
5564 vdev_config_clean(vd);
5569 vdev_compact_children(rvd);
5571 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
5572 vdev_add_child(rvd, vd);
5574 vdev_config_dirty(rvd);
5577 * Reassess the health of our root vdev.
5583 * Remove a device from the pool -
5585 * Removing a device from the vdev namespace requires several steps
5586 * and can take a significant amount of time. As a result we use
5587 * the spa_vdev_config_[enter/exit] functions which allow us to
5588 * grab and release the spa_config_lock while still holding the namespace
5589 * lock. During each step the configuration is synced out.
5591 * Currently, this supports removing only hot spares, slogs, and level 2 ARC
5595 spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
5598 metaslab_group_t *mg;
5599 nvlist_t **spares, **l2cache, *nv;
5601 uint_t nspares, nl2cache;
5603 boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
5605 ASSERT(spa_writeable(spa));
5608 txg = spa_vdev_enter(spa);
5610 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5612 if (spa->spa_spares.sav_vdevs != NULL &&
5613 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5614 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
5615 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
5617 * Only remove the hot spare if it's not currently in use
5620 if (vd == NULL || unspare) {
5621 spa_vdev_remove_aux(spa->spa_spares.sav_config,
5622 ZPOOL_CONFIG_SPARES, spares, nspares, nv);
5623 spa_load_spares(spa);
5624 spa->spa_spares.sav_sync = B_TRUE;
5626 error = SET_ERROR(EBUSY);
5628 } else if (spa->spa_l2cache.sav_vdevs != NULL &&
5629 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5630 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
5631 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
5633 * Cache devices can always be removed.
5635 spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
5636 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
5637 spa_load_l2cache(spa);
5638 spa->spa_l2cache.sav_sync = B_TRUE;
5639 } else if (vd != NULL && vd->vdev_islog) {
5641 ASSERT(vd == vd->vdev_top);
5646 * Stop allocating from this vdev.
5648 metaslab_group_passivate(mg);
5651 * Wait for the youngest allocations and frees to sync,
5652 * and then wait for the deferral of those frees to finish.
5654 spa_vdev_config_exit(spa, NULL,
5655 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
5658 * Attempt to evacuate the vdev.
5660 error = spa_vdev_remove_evacuate(spa, vd);
5662 txg = spa_vdev_config_enter(spa);
5665 * If we couldn't evacuate the vdev, unwind.
5668 metaslab_group_activate(mg);
5669 return (spa_vdev_exit(spa, NULL, txg, error));
5673 * Clean up the vdev namespace.
5675 spa_vdev_remove_from_namespace(spa, vd);
5677 } else if (vd != NULL) {
5679 * Normal vdevs cannot be removed (yet).
5681 error = SET_ERROR(ENOTSUP);
5684 * There is no vdev of any kind with the specified guid.
5686 error = SET_ERROR(ENOENT);
5690 return (spa_vdev_exit(spa, NULL, txg, error));
5696 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5697 * currently spared, so we can detach it.
5700 spa_vdev_resilver_done_hunt(vdev_t *vd)
5702 vdev_t *newvd, *oldvd;
5704 for (int c = 0; c < vd->vdev_children; c++) {
5705 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
5711 * Check for a completed replacement. We always consider the first
5712 * vdev in the list to be the oldest vdev, and the last one to be
5713 * the newest (see spa_vdev_attach() for how that works). In
5714 * the case where the newest vdev is faulted, we will not automatically
5715 * remove it after a resilver completes. This is OK as it will require
5716 * user intervention to determine which disk the admin wishes to keep.
5718 if (vd->vdev_ops == &vdev_replacing_ops) {
5719 ASSERT(vd->vdev_children > 1);
5721 newvd = vd->vdev_child[vd->vdev_children - 1];
5722 oldvd = vd->vdev_child[0];
5724 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
5725 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5726 !vdev_dtl_required(oldvd))
5731 * Check for a completed resilver with the 'unspare' flag set.
5733 if (vd->vdev_ops == &vdev_spare_ops) {
5734 vdev_t *first = vd->vdev_child[0];
5735 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
5737 if (last->vdev_unspare) {
5740 } else if (first->vdev_unspare) {
5747 if (oldvd != NULL &&
5748 vdev_dtl_empty(newvd, DTL_MISSING) &&
5749 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5750 !vdev_dtl_required(oldvd))
5754 * If there are more than two spares attached to a disk,
5755 * and those spares are not required, then we want to
5756 * attempt to free them up now so that they can be used
5757 * by other pools. Once we're back down to a single
5758 * disk+spare, we stop removing them.
5760 if (vd->vdev_children > 2) {
5761 newvd = vd->vdev_child[1];
5763 if (newvd->vdev_isspare && last->vdev_isspare &&
5764 vdev_dtl_empty(last, DTL_MISSING) &&
5765 vdev_dtl_empty(last, DTL_OUTAGE) &&
5766 !vdev_dtl_required(newvd))
5775 spa_vdev_resilver_done(spa_t *spa)
5777 vdev_t *vd, *pvd, *ppvd;
5778 uint64_t guid, sguid, pguid, ppguid;
5780 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5782 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
5783 pvd = vd->vdev_parent;
5784 ppvd = pvd->vdev_parent;
5785 guid = vd->vdev_guid;
5786 pguid = pvd->vdev_guid;
5787 ppguid = ppvd->vdev_guid;
5790 * If we have just finished replacing a hot spared device, then
5791 * we need to detach the parent's first child (the original hot
5794 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
5795 ppvd->vdev_children == 2) {
5796 ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
5797 sguid = ppvd->vdev_child[1]->vdev_guid;
5799 ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd));
5801 spa_config_exit(spa, SCL_ALL, FTAG);
5802 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
5804 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
5806 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5809 spa_config_exit(spa, SCL_ALL, FTAG);
5813 * Update the stored path or FRU for this vdev.
5816 spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
5820 boolean_t sync = B_FALSE;
5822 ASSERT(spa_writeable(spa));
5824 spa_vdev_state_enter(spa, SCL_ALL);
5826 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
5827 return (spa_vdev_state_exit(spa, NULL, ENOENT));
5829 if (!vd->vdev_ops->vdev_op_leaf)
5830 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
5833 if (strcmp(value, vd->vdev_path) != 0) {
5834 spa_strfree(vd->vdev_path);
5835 vd->vdev_path = spa_strdup(value);
5839 if (vd->vdev_fru == NULL) {
5840 vd->vdev_fru = spa_strdup(value);
5842 } else if (strcmp(value, vd->vdev_fru) != 0) {
5843 spa_strfree(vd->vdev_fru);
5844 vd->vdev_fru = spa_strdup(value);
5849 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
5853 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
5855 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5859 spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5861 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5865 * ==========================================================================
5867 * ==========================================================================
5871 spa_scan_stop(spa_t *spa)
5873 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5874 if (dsl_scan_resilvering(spa->spa_dsl_pool))
5875 return (SET_ERROR(EBUSY));
5876 return (dsl_scan_cancel(spa->spa_dsl_pool));
5880 spa_scan(spa_t *spa, pool_scan_func_t func)
5882 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5884 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5885 return (SET_ERROR(ENOTSUP));
5888 * If a resilver was requested, but there is no DTL on a
5889 * writeable leaf device, we have nothing to do.
5891 if (func == POOL_SCAN_RESILVER &&
5892 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5893 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5897 return (dsl_scan(spa->spa_dsl_pool, func));
5901 * ==========================================================================
5902 * SPA async task processing
5903 * ==========================================================================
5907 spa_async_remove(spa_t *spa, vdev_t *vd)
5909 if (vd->vdev_remove_wanted) {
5910 vd->vdev_remove_wanted = B_FALSE;
5911 vd->vdev_delayed_close = B_FALSE;
5912 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5915 * We want to clear the stats, but we don't want to do a full
5916 * vdev_clear() as that will cause us to throw away
5917 * degraded/faulted state as well as attempt to reopen the
5918 * device, all of which is a waste.
5920 vd->vdev_stat.vs_read_errors = 0;
5921 vd->vdev_stat.vs_write_errors = 0;
5922 vd->vdev_stat.vs_checksum_errors = 0;
5924 vdev_state_dirty(vd->vdev_top);
5927 for (int c = 0; c < vd->vdev_children; c++)
5928 spa_async_remove(spa, vd->vdev_child[c]);
5932 spa_async_probe(spa_t *spa, vdev_t *vd)
5934 if (vd->vdev_probe_wanted) {
5935 vd->vdev_probe_wanted = B_FALSE;
5936 vdev_reopen(vd); /* vdev_open() does the actual probe */
5939 for (int c = 0; c < vd->vdev_children; c++)
5940 spa_async_probe(spa, vd->vdev_child[c]);
5944 spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5950 if (!spa->spa_autoexpand)
5953 for (int c = 0; c < vd->vdev_children; c++) {
5954 vdev_t *cvd = vd->vdev_child[c];
5955 spa_async_autoexpand(spa, cvd);
5958 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5961 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
5962 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
5964 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5965 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
5967 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
5968 ESC_ZFS_VDEV_AUTOEXPAND, attr, &eid, DDI_SLEEP);
5971 kmem_free(physpath, MAXPATHLEN);
5975 spa_async_thread(void *arg)
5980 ASSERT(spa->spa_sync_on);
5982 mutex_enter(&spa->spa_async_lock);
5983 tasks = spa->spa_async_tasks;
5984 spa->spa_async_tasks &= SPA_ASYNC_REMOVE;
5985 mutex_exit(&spa->spa_async_lock);
5988 * See if the config needs to be updated.
5990 if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5991 uint64_t old_space, new_space;
5993 mutex_enter(&spa_namespace_lock);
5994 old_space = metaslab_class_get_space(spa_normal_class(spa));
5995 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5996 new_space = metaslab_class_get_space(spa_normal_class(spa));
5997 mutex_exit(&spa_namespace_lock);
6000 * If the pool grew as a result of the config update,
6001 * then log an internal history event.
6003 if (new_space != old_space) {
6004 spa_history_log_internal(spa, "vdev online", NULL,
6005 "pool '%s' size: %llu(+%llu)",
6006 spa_name(spa), new_space, new_space - old_space);
6010 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
6011 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
6012 spa_async_autoexpand(spa, spa->spa_root_vdev);
6013 spa_config_exit(spa, SCL_CONFIG, FTAG);
6017 * See if any devices need to be probed.
6019 if (tasks & SPA_ASYNC_PROBE) {
6020 spa_vdev_state_enter(spa, SCL_NONE);
6021 spa_async_probe(spa, spa->spa_root_vdev);
6022 (void) spa_vdev_state_exit(spa, NULL, 0);
6026 * If any devices are done replacing, detach them.
6028 if (tasks & SPA_ASYNC_RESILVER_DONE)
6029 spa_vdev_resilver_done(spa);
6032 * Kick off a resilver.
6034 if (tasks & SPA_ASYNC_RESILVER)
6035 dsl_resilver_restart(spa->spa_dsl_pool, 0);
6038 * Let the world know that we're done.
6040 mutex_enter(&spa->spa_async_lock);
6041 spa->spa_async_thread = NULL;
6042 cv_broadcast(&spa->spa_async_cv);
6043 mutex_exit(&spa->spa_async_lock);
6048 spa_async_thread_vd(void *arg)
6053 ASSERT(spa->spa_sync_on);
6055 mutex_enter(&spa->spa_async_lock);
6056 tasks = spa->spa_async_tasks;
6058 spa->spa_async_tasks &= ~SPA_ASYNC_REMOVE;
6059 mutex_exit(&spa->spa_async_lock);
6062 * See if any devices need to be marked REMOVED.
6064 if (tasks & SPA_ASYNC_REMOVE) {
6065 spa_vdev_state_enter(spa, SCL_NONE);
6066 spa_async_remove(spa, spa->spa_root_vdev);
6067 for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
6068 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
6069 for (int i = 0; i < spa->spa_spares.sav_count; i++)
6070 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
6071 (void) spa_vdev_state_exit(spa, NULL, 0);
6075 * Let the world know that we're done.
6077 mutex_enter(&spa->spa_async_lock);
6078 tasks = spa->spa_async_tasks;
6079 if ((tasks & SPA_ASYNC_REMOVE) != 0)
6081 spa->spa_async_thread_vd = NULL;
6082 cv_broadcast(&spa->spa_async_cv);
6083 mutex_exit(&spa->spa_async_lock);
6088 spa_async_suspend(spa_t *spa)
6090 mutex_enter(&spa->spa_async_lock);
6091 spa->spa_async_suspended++;
6092 while (spa->spa_async_thread != NULL &&
6093 spa->spa_async_thread_vd != NULL)
6094 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
6095 mutex_exit(&spa->spa_async_lock);
6099 spa_async_resume(spa_t *spa)
6101 mutex_enter(&spa->spa_async_lock);
6102 ASSERT(spa->spa_async_suspended != 0);
6103 spa->spa_async_suspended--;
6104 mutex_exit(&spa->spa_async_lock);
6108 spa_async_tasks_pending(spa_t *spa)
6110 uint_t non_config_tasks;
6112 boolean_t config_task_suspended;
6114 non_config_tasks = spa->spa_async_tasks & ~(SPA_ASYNC_CONFIG_UPDATE |
6116 config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE;
6117 if (spa->spa_ccw_fail_time == 0) {
6118 config_task_suspended = B_FALSE;
6120 config_task_suspended =
6121 (gethrtime() - spa->spa_ccw_fail_time) <
6122 (zfs_ccw_retry_interval * NANOSEC);
6125 return (non_config_tasks || (config_task && !config_task_suspended));
6129 spa_async_dispatch(spa_t *spa)
6131 mutex_enter(&spa->spa_async_lock);
6132 if (spa_async_tasks_pending(spa) &&
6133 !spa->spa_async_suspended &&
6134 spa->spa_async_thread == NULL &&
6136 spa->spa_async_thread = thread_create(NULL, 0,
6137 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
6138 mutex_exit(&spa->spa_async_lock);
6142 spa_async_dispatch_vd(spa_t *spa)
6144 mutex_enter(&spa->spa_async_lock);
6145 if ((spa->spa_async_tasks & SPA_ASYNC_REMOVE) != 0 &&
6146 !spa->spa_async_suspended &&
6147 spa->spa_async_thread_vd == NULL &&
6149 spa->spa_async_thread_vd = thread_create(NULL, 0,
6150 spa_async_thread_vd, spa, 0, &p0, TS_RUN, maxclsyspri);
6151 mutex_exit(&spa->spa_async_lock);
6155 spa_async_request(spa_t *spa, int task)
6157 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
6158 mutex_enter(&spa->spa_async_lock);
6159 spa->spa_async_tasks |= task;
6160 mutex_exit(&spa->spa_async_lock);
6161 spa_async_dispatch_vd(spa);
6165 * ==========================================================================
6166 * SPA syncing routines
6167 * ==========================================================================
6171 bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
6174 bpobj_enqueue(bpo, bp, tx);
6179 spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
6183 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
6184 BP_GET_PSIZE(bp), zio->io_flags));
6189 * Note: this simple function is not inlined to make it easier to dtrace the
6190 * amount of time spent syncing frees.
6193 spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx)
6195 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6196 bplist_iterate(bpl, spa_free_sync_cb, zio, tx);
6197 VERIFY(zio_wait(zio) == 0);
6201 * Note: this simple function is not inlined to make it easier to dtrace the
6202 * amount of time spent syncing deferred frees.
6205 spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx)
6207 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6208 VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj,
6209 spa_free_sync_cb, zio, tx), ==, 0);
6210 VERIFY0(zio_wait(zio));
6215 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
6217 char *packed = NULL;
6222 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
6225 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
6226 * information. This avoids the dmu_buf_will_dirty() path and
6227 * saves us a pre-read to get data we don't actually care about.
6229 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
6230 packed = kmem_alloc(bufsize, KM_SLEEP);
6232 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
6234 bzero(packed + nvsize, bufsize - nvsize);
6236 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
6238 kmem_free(packed, bufsize);
6240 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
6241 dmu_buf_will_dirty(db, tx);
6242 *(uint64_t *)db->db_data = nvsize;
6243 dmu_buf_rele(db, FTAG);
6247 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
6248 const char *config, const char *entry)
6258 * Update the MOS nvlist describing the list of available devices.
6259 * spa_validate_aux() will have already made sure this nvlist is
6260 * valid and the vdevs are labeled appropriately.
6262 if (sav->sav_object == 0) {
6263 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
6264 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
6265 sizeof (uint64_t), tx);
6266 VERIFY(zap_update(spa->spa_meta_objset,
6267 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
6268 &sav->sav_object, tx) == 0);
6271 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
6272 if (sav->sav_count == 0) {
6273 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
6275 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
6276 for (i = 0; i < sav->sav_count; i++)
6277 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
6278 B_FALSE, VDEV_CONFIG_L2CACHE);
6279 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
6280 sav->sav_count) == 0);
6281 for (i = 0; i < sav->sav_count; i++)
6282 nvlist_free(list[i]);
6283 kmem_free(list, sav->sav_count * sizeof (void *));
6286 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
6287 nvlist_free(nvroot);
6289 sav->sav_sync = B_FALSE;
6293 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
6297 if (list_is_empty(&spa->spa_config_dirty_list))
6300 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6302 config = spa_config_generate(spa, spa->spa_root_vdev,
6303 dmu_tx_get_txg(tx), B_FALSE);
6306 * If we're upgrading the spa version then make sure that
6307 * the config object gets updated with the correct version.
6309 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
6310 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
6311 spa->spa_uberblock.ub_version);
6313 spa_config_exit(spa, SCL_STATE, FTAG);
6315 if (spa->spa_config_syncing)
6316 nvlist_free(spa->spa_config_syncing);
6317 spa->spa_config_syncing = config;
6319 spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
6323 spa_sync_version(void *arg, dmu_tx_t *tx)
6325 uint64_t *versionp = arg;
6326 uint64_t version = *versionp;
6327 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
6330 * Setting the version is special cased when first creating the pool.
6332 ASSERT(tx->tx_txg != TXG_INITIAL);
6334 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
6335 ASSERT(version >= spa_version(spa));
6337 spa->spa_uberblock.ub_version = version;
6338 vdev_config_dirty(spa->spa_root_vdev);
6339 spa_history_log_internal(spa, "set", tx, "version=%lld", version);
6343 * Set zpool properties.
6346 spa_sync_props(void *arg, dmu_tx_t *tx)
6348 nvlist_t *nvp = arg;
6349 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
6350 objset_t *mos = spa->spa_meta_objset;
6351 nvpair_t *elem = NULL;
6353 mutex_enter(&spa->spa_props_lock);
6355 while ((elem = nvlist_next_nvpair(nvp, elem))) {
6357 char *strval, *fname;
6359 const char *propname;
6360 zprop_type_t proptype;
6363 switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
6366 * We checked this earlier in spa_prop_validate().
6368 ASSERT(zpool_prop_feature(nvpair_name(elem)));
6370 fname = strchr(nvpair_name(elem), '@') + 1;
6371 VERIFY0(zfeature_lookup_name(fname, &fid));
6373 spa_feature_enable(spa, fid, tx);
6374 spa_history_log_internal(spa, "set", tx,
6375 "%s=enabled", nvpair_name(elem));
6378 case ZPOOL_PROP_VERSION:
6379 intval = fnvpair_value_uint64(elem);
6381 * The version is synced seperatly before other
6382 * properties and should be correct by now.
6384 ASSERT3U(spa_version(spa), >=, intval);
6387 case ZPOOL_PROP_ALTROOT:
6389 * 'altroot' is a non-persistent property. It should
6390 * have been set temporarily at creation or import time.
6392 ASSERT(spa->spa_root != NULL);
6395 case ZPOOL_PROP_READONLY:
6396 case ZPOOL_PROP_CACHEFILE:
6398 * 'readonly' and 'cachefile' are also non-persisitent
6402 case ZPOOL_PROP_COMMENT:
6403 strval = fnvpair_value_string(elem);
6404 if (spa->spa_comment != NULL)
6405 spa_strfree(spa->spa_comment);
6406 spa->spa_comment = spa_strdup(strval);
6408 * We need to dirty the configuration on all the vdevs
6409 * so that their labels get updated. It's unnecessary
6410 * to do this for pool creation since the vdev's
6411 * configuratoin has already been dirtied.
6413 if (tx->tx_txg != TXG_INITIAL)
6414 vdev_config_dirty(spa->spa_root_vdev);
6415 spa_history_log_internal(spa, "set", tx,
6416 "%s=%s", nvpair_name(elem), strval);
6420 * Set pool property values in the poolprops mos object.
6422 if (spa->spa_pool_props_object == 0) {
6423 spa->spa_pool_props_object =
6424 zap_create_link(mos, DMU_OT_POOL_PROPS,
6425 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
6429 /* normalize the property name */
6430 propname = zpool_prop_to_name(prop);
6431 proptype = zpool_prop_get_type(prop);
6433 if (nvpair_type(elem) == DATA_TYPE_STRING) {
6434 ASSERT(proptype == PROP_TYPE_STRING);
6435 strval = fnvpair_value_string(elem);
6436 VERIFY0(zap_update(mos,
6437 spa->spa_pool_props_object, propname,
6438 1, strlen(strval) + 1, strval, tx));
6439 spa_history_log_internal(spa, "set", tx,
6440 "%s=%s", nvpair_name(elem), strval);
6441 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
6442 intval = fnvpair_value_uint64(elem);
6444 if (proptype == PROP_TYPE_INDEX) {
6446 VERIFY0(zpool_prop_index_to_string(
6447 prop, intval, &unused));
6449 VERIFY0(zap_update(mos,
6450 spa->spa_pool_props_object, propname,
6451 8, 1, &intval, tx));
6452 spa_history_log_internal(spa, "set", tx,
6453 "%s=%lld", nvpair_name(elem), intval);
6455 ASSERT(0); /* not allowed */
6459 case ZPOOL_PROP_DELEGATION:
6460 spa->spa_delegation = intval;
6462 case ZPOOL_PROP_BOOTFS:
6463 spa->spa_bootfs = intval;
6465 case ZPOOL_PROP_FAILUREMODE:
6466 spa->spa_failmode = intval;
6468 case ZPOOL_PROP_AUTOEXPAND:
6469 spa->spa_autoexpand = intval;
6470 if (tx->tx_txg != TXG_INITIAL)
6471 spa_async_request(spa,
6472 SPA_ASYNC_AUTOEXPAND);
6474 case ZPOOL_PROP_DEDUPDITTO:
6475 spa->spa_dedup_ditto = intval;
6484 mutex_exit(&spa->spa_props_lock);
6488 * Perform one-time upgrade on-disk changes. spa_version() does not
6489 * reflect the new version this txg, so there must be no changes this
6490 * txg to anything that the upgrade code depends on after it executes.
6491 * Therefore this must be called after dsl_pool_sync() does the sync
6495 spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
6497 dsl_pool_t *dp = spa->spa_dsl_pool;
6499 ASSERT(spa->spa_sync_pass == 1);
6501 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
6503 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
6504 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
6505 dsl_pool_create_origin(dp, tx);
6507 /* Keeping the origin open increases spa_minref */
6508 spa->spa_minref += 3;
6511 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
6512 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
6513 dsl_pool_upgrade_clones(dp, tx);
6516 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
6517 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
6518 dsl_pool_upgrade_dir_clones(dp, tx);
6520 /* Keeping the freedir open increases spa_minref */
6521 spa->spa_minref += 3;
6524 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES &&
6525 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
6526 spa_feature_create_zap_objects(spa, tx);
6530 * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable
6531 * when possibility to use lz4 compression for metadata was added
6532 * Old pools that have this feature enabled must be upgraded to have
6533 * this feature active
6535 if (spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
6536 boolean_t lz4_en = spa_feature_is_enabled(spa,
6537 SPA_FEATURE_LZ4_COMPRESS);
6538 boolean_t lz4_ac = spa_feature_is_active(spa,
6539 SPA_FEATURE_LZ4_COMPRESS);
6541 if (lz4_en && !lz4_ac)
6542 spa_feature_incr(spa, SPA_FEATURE_LZ4_COMPRESS, tx);
6544 rrw_exit(&dp->dp_config_rwlock, FTAG);
6548 * Sync the specified transaction group. New blocks may be dirtied as
6549 * part of the process, so we iterate until it converges.
6552 spa_sync(spa_t *spa, uint64_t txg)
6554 dsl_pool_t *dp = spa->spa_dsl_pool;
6555 objset_t *mos = spa->spa_meta_objset;
6556 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
6557 vdev_t *rvd = spa->spa_root_vdev;
6562 VERIFY(spa_writeable(spa));
6565 * Lock out configuration changes.
6567 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
6569 spa->spa_syncing_txg = txg;
6570 spa->spa_sync_pass = 0;
6573 * If there are any pending vdev state changes, convert them
6574 * into config changes that go out with this transaction group.
6576 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6577 while (list_head(&spa->spa_state_dirty_list) != NULL) {
6579 * We need the write lock here because, for aux vdevs,
6580 * calling vdev_config_dirty() modifies sav_config.
6581 * This is ugly and will become unnecessary when we
6582 * eliminate the aux vdev wart by integrating all vdevs
6583 * into the root vdev tree.
6585 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6586 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
6587 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
6588 vdev_state_clean(vd);
6589 vdev_config_dirty(vd);
6591 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6592 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
6594 spa_config_exit(spa, SCL_STATE, FTAG);
6596 tx = dmu_tx_create_assigned(dp, txg);
6598 spa->spa_sync_starttime = gethrtime();
6600 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid,
6601 spa->spa_sync_starttime + spa->spa_deadman_synctime));
6604 callout_reset(&spa->spa_deadman_cycid,
6605 hz * spa->spa_deadman_synctime / NANOSEC, spa_deadman, spa);
6610 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
6611 * set spa_deflate if we have no raid-z vdevs.
6613 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
6614 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
6617 for (i = 0; i < rvd->vdev_children; i++) {
6618 vd = rvd->vdev_child[i];
6619 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
6622 if (i == rvd->vdev_children) {
6623 spa->spa_deflate = TRUE;
6624 VERIFY(0 == zap_add(spa->spa_meta_objset,
6625 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
6626 sizeof (uint64_t), 1, &spa->spa_deflate, tx));
6631 * Iterate to convergence.
6634 int pass = ++spa->spa_sync_pass;
6636 spa_sync_config_object(spa, tx);
6637 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
6638 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
6639 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
6640 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
6641 spa_errlog_sync(spa, txg);
6642 dsl_pool_sync(dp, txg);
6644 if (pass < zfs_sync_pass_deferred_free) {
6645 spa_sync_frees(spa, free_bpl, tx);
6648 * We can not defer frees in pass 1, because
6649 * we sync the deferred frees later in pass 1.
6651 ASSERT3U(pass, >, 1);
6652 bplist_iterate(free_bpl, bpobj_enqueue_cb,
6653 &spa->spa_deferred_bpobj, tx);
6657 dsl_scan_sync(dp, tx);
6659 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
6663 spa_sync_upgrades(spa, tx);
6665 spa->spa_uberblock.ub_rootbp.blk_birth);
6667 * Note: We need to check if the MOS is dirty
6668 * because we could have marked the MOS dirty
6669 * without updating the uberblock (e.g. if we
6670 * have sync tasks but no dirty user data). We
6671 * need to check the uberblock's rootbp because
6672 * it is updated if we have synced out dirty
6673 * data (though in this case the MOS will most
6674 * likely also be dirty due to second order
6675 * effects, we don't want to rely on that here).
6677 if (spa->spa_uberblock.ub_rootbp.blk_birth < txg &&
6678 !dmu_objset_is_dirty(mos, txg)) {
6680 * Nothing changed on the first pass,
6681 * therefore this TXG is a no-op. Avoid
6682 * syncing deferred frees, so that we
6683 * can keep this TXG as a no-op.
6685 ASSERT(txg_list_empty(&dp->dp_dirty_datasets,
6687 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
6688 ASSERT(txg_list_empty(&dp->dp_sync_tasks, txg));
6691 spa_sync_deferred_frees(spa, tx);
6694 } while (dmu_objset_is_dirty(mos, txg));
6697 * Rewrite the vdev configuration (which includes the uberblock)
6698 * to commit the transaction group.
6700 * If there are no dirty vdevs, we sync the uberblock to a few
6701 * random top-level vdevs that are known to be visible in the
6702 * config cache (see spa_vdev_add() for a complete description).
6703 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
6707 * We hold SCL_STATE to prevent vdev open/close/etc.
6708 * while we're attempting to write the vdev labels.
6710 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6712 if (list_is_empty(&spa->spa_config_dirty_list)) {
6713 vdev_t *svd[SPA_DVAS_PER_BP];
6715 int children = rvd->vdev_children;
6716 int c0 = spa_get_random(children);
6718 for (int c = 0; c < children; c++) {
6719 vd = rvd->vdev_child[(c0 + c) % children];
6720 if (vd->vdev_ms_array == 0 || vd->vdev_islog)
6722 svd[svdcount++] = vd;
6723 if (svdcount == SPA_DVAS_PER_BP)
6726 error = vdev_config_sync(svd, svdcount, txg, B_FALSE);
6728 error = vdev_config_sync(svd, svdcount, txg,
6731 error = vdev_config_sync(rvd->vdev_child,
6732 rvd->vdev_children, txg, B_FALSE);
6734 error = vdev_config_sync(rvd->vdev_child,
6735 rvd->vdev_children, txg, B_TRUE);
6739 spa->spa_last_synced_guid = rvd->vdev_guid;
6741 spa_config_exit(spa, SCL_STATE, FTAG);
6745 zio_suspend(spa, NULL);
6746 zio_resume_wait(spa);
6751 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY));
6754 callout_drain(&spa->spa_deadman_cycid);
6759 * Clear the dirty config list.
6761 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
6762 vdev_config_clean(vd);
6765 * Now that the new config has synced transactionally,
6766 * let it become visible to the config cache.
6768 if (spa->spa_config_syncing != NULL) {
6769 spa_config_set(spa, spa->spa_config_syncing);
6770 spa->spa_config_txg = txg;
6771 spa->spa_config_syncing = NULL;
6774 spa->spa_ubsync = spa->spa_uberblock;
6776 dsl_pool_sync_done(dp, txg);
6779 * Update usable space statistics.
6781 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
6782 vdev_sync_done(vd, txg);
6784 spa_update_dspace(spa);
6787 * It had better be the case that we didn't dirty anything
6788 * since vdev_config_sync().
6790 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
6791 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
6792 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
6794 spa->spa_sync_pass = 0;
6796 spa_config_exit(spa, SCL_CONFIG, FTAG);
6798 spa_handle_ignored_writes(spa);
6801 * If any async tasks have been requested, kick them off.
6803 spa_async_dispatch(spa);
6804 spa_async_dispatch_vd(spa);
6808 * Sync all pools. We don't want to hold the namespace lock across these
6809 * operations, so we take a reference on the spa_t and drop the lock during the
6813 spa_sync_allpools(void)
6816 mutex_enter(&spa_namespace_lock);
6817 while ((spa = spa_next(spa)) != NULL) {
6818 if (spa_state(spa) != POOL_STATE_ACTIVE ||
6819 !spa_writeable(spa) || spa_suspended(spa))
6821 spa_open_ref(spa, FTAG);
6822 mutex_exit(&spa_namespace_lock);
6823 txg_wait_synced(spa_get_dsl(spa), 0);
6824 mutex_enter(&spa_namespace_lock);
6825 spa_close(spa, FTAG);
6827 mutex_exit(&spa_namespace_lock);
6831 * ==========================================================================
6832 * Miscellaneous routines
6833 * ==========================================================================
6837 * Remove all pools in the system.
6845 * Remove all cached state. All pools should be closed now,
6846 * so every spa in the AVL tree should be unreferenced.
6848 mutex_enter(&spa_namespace_lock);
6849 while ((spa = spa_next(NULL)) != NULL) {
6851 * Stop async tasks. The async thread may need to detach
6852 * a device that's been replaced, which requires grabbing
6853 * spa_namespace_lock, so we must drop it here.
6855 spa_open_ref(spa, FTAG);
6856 mutex_exit(&spa_namespace_lock);
6857 spa_async_suspend(spa);
6858 mutex_enter(&spa_namespace_lock);
6859 spa_close(spa, FTAG);
6861 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
6863 spa_deactivate(spa);
6867 mutex_exit(&spa_namespace_lock);
6871 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
6876 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
6880 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
6881 vd = spa->spa_l2cache.sav_vdevs[i];
6882 if (vd->vdev_guid == guid)
6886 for (i = 0; i < spa->spa_spares.sav_count; i++) {
6887 vd = spa->spa_spares.sav_vdevs[i];
6888 if (vd->vdev_guid == guid)
6897 spa_upgrade(spa_t *spa, uint64_t version)
6899 ASSERT(spa_writeable(spa));
6901 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6904 * This should only be called for a non-faulted pool, and since a
6905 * future version would result in an unopenable pool, this shouldn't be
6908 ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version));
6909 ASSERT3U(version, >=, spa->spa_uberblock.ub_version);
6911 spa->spa_uberblock.ub_version = version;
6912 vdev_config_dirty(spa->spa_root_vdev);
6914 spa_config_exit(spa, SCL_ALL, FTAG);
6916 txg_wait_synced(spa_get_dsl(spa), 0);
6920 spa_has_spare(spa_t *spa, uint64_t guid)
6924 spa_aux_vdev_t *sav = &spa->spa_spares;
6926 for (i = 0; i < sav->sav_count; i++)
6927 if (sav->sav_vdevs[i]->vdev_guid == guid)
6930 for (i = 0; i < sav->sav_npending; i++) {
6931 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
6932 &spareguid) == 0 && spareguid == guid)
6940 * Check if a pool has an active shared spare device.
6941 * Note: reference count of an active spare is 2, as a spare and as a replace
6944 spa_has_active_shared_spare(spa_t *spa)
6948 spa_aux_vdev_t *sav = &spa->spa_spares;
6950 for (i = 0; i < sav->sav_count; i++) {
6951 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
6952 &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
6961 * Post a sysevent corresponding to the given event. The 'name' must be one of
6962 * the event definitions in sys/sysevent/eventdefs.h. The payload will be
6963 * filled in from the spa and (optionally) the vdev. This doesn't do anything
6964 * in the userland libzpool, as we don't want consumers to misinterpret ztest
6965 * or zdb as real changes.
6968 spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
6972 sysevent_attr_list_t *attr = NULL;
6973 sysevent_value_t value;
6976 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
6979 value.value_type = SE_DATA_TYPE_STRING;
6980 value.value.sv_string = spa_name(spa);
6981 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
6984 value.value_type = SE_DATA_TYPE_UINT64;
6985 value.value.sv_uint64 = spa_guid(spa);
6986 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
6990 value.value_type = SE_DATA_TYPE_UINT64;
6991 value.value.sv_uint64 = vd->vdev_guid;
6992 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
6996 if (vd->vdev_path) {
6997 value.value_type = SE_DATA_TYPE_STRING;
6998 value.value.sv_string = vd->vdev_path;
6999 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
7000 &value, SE_SLEEP) != 0)
7005 if (sysevent_attach_attributes(ev, attr) != 0)
7009 (void) log_sysevent(ev, SE_SLEEP, &eid);
7013 sysevent_free_attr(attr);