4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2011, 2014 by Delphix. All rights reserved.
25 * Copyright (c) 2013, 2014, Nexenta Systems, Inc. All rights reserved.
26 * Copyright (c) 2013 Martin Matuska <mm@FreeBSD.org>. All rights reserved.
30 * SPA: Storage Pool Allocator
32 * This file contains all the routines used when modifying on-disk SPA state.
33 * This includes opening, importing, destroying, exporting a pool, and syncing a
37 #include <sys/zfs_context.h>
38 #include <sys/fm/fs/zfs.h>
39 #include <sys/spa_impl.h>
41 #include <sys/zio_checksum.h>
43 #include <sys/dmu_tx.h>
47 #include <sys/vdev_impl.h>
48 #include <sys/metaslab.h>
49 #include <sys/metaslab_impl.h>
50 #include <sys/uberblock_impl.h>
53 #include <sys/dmu_traverse.h>
54 #include <sys/dmu_objset.h>
55 #include <sys/unique.h>
56 #include <sys/dsl_pool.h>
57 #include <sys/dsl_dataset.h>
58 #include <sys/dsl_dir.h>
59 #include <sys/dsl_prop.h>
60 #include <sys/dsl_synctask.h>
61 #include <sys/fs/zfs.h>
63 #include <sys/callb.h>
64 #include <sys/spa_boot.h>
65 #include <sys/zfs_ioctl.h>
66 #include <sys/dsl_scan.h>
67 #include <sys/dmu_send.h>
68 #include <sys/dsl_destroy.h>
69 #include <sys/dsl_userhold.h>
70 #include <sys/zfeature.h>
72 #include <sys/trim_map.h>
75 #include <sys/callb.h>
76 #include <sys/cpupart.h>
81 #include "zfs_comutil.h"
83 /* Check hostid on import? */
84 static int check_hostid = 1;
87 * The interval, in seconds, at which failed configuration cache file writes
90 static int zfs_ccw_retry_interval = 300;
92 SYSCTL_DECL(_vfs_zfs);
93 SYSCTL_INT(_vfs_zfs, OID_AUTO, check_hostid, CTLFLAG_RWTUN, &check_hostid, 0,
94 "Check hostid on import?");
95 TUNABLE_INT("vfs.zfs.ccw_retry_interval", &zfs_ccw_retry_interval);
96 SYSCTL_INT(_vfs_zfs, OID_AUTO, ccw_retry_interval, CTLFLAG_RW,
97 &zfs_ccw_retry_interval, 0,
98 "Configuration cache file write, retry after failure, interval (seconds)");
100 typedef enum zti_modes {
101 ZTI_MODE_FIXED, /* value is # of threads (min 1) */
102 ZTI_MODE_BATCH, /* cpu-intensive; value is ignored */
103 ZTI_MODE_NULL, /* don't create a taskq */
107 #define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) }
108 #define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 }
109 #define ZTI_NULL { ZTI_MODE_NULL, 0, 0 }
111 #define ZTI_N(n) ZTI_P(n, 1)
112 #define ZTI_ONE ZTI_N(1)
114 typedef struct zio_taskq_info {
115 zti_modes_t zti_mode;
120 static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
121 "issue", "issue_high", "intr", "intr_high"
125 * This table defines the taskq settings for each ZFS I/O type. When
126 * initializing a pool, we use this table to create an appropriately sized
127 * taskq. Some operations are low volume and therefore have a small, static
128 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE
129 * macros. Other operations process a large amount of data; the ZTI_BATCH
130 * macro causes us to create a taskq oriented for throughput. Some operations
131 * are so high frequency and short-lived that the taskq itself can become a a
132 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an
133 * additional degree of parallelism specified by the number of threads per-
134 * taskq and the number of taskqs; when dispatching an event in this case, the
135 * particular taskq is chosen at random.
137 * The different taskq priorities are to handle the different contexts (issue
138 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that
139 * need to be handled with minimum delay.
141 const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
142 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
143 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* NULL */
144 { ZTI_N(8), ZTI_NULL, ZTI_P(12, 8), ZTI_NULL }, /* READ */
145 { ZTI_BATCH, ZTI_N(5), ZTI_N(8), ZTI_N(5) }, /* WRITE */
146 { ZTI_P(12, 8), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FREE */
147 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* CLAIM */
148 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* IOCTL */
151 static void spa_sync_version(void *arg, dmu_tx_t *tx);
152 static void spa_sync_props(void *arg, dmu_tx_t *tx);
153 static boolean_t spa_has_active_shared_spare(spa_t *spa);
154 static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
155 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
157 static void spa_vdev_resilver_done(spa_t *spa);
159 uint_t zio_taskq_batch_pct = 75; /* 1 thread per cpu in pset */
161 id_t zio_taskq_psrset_bind = PS_NONE;
164 boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */
166 uint_t zio_taskq_basedc = 80; /* base duty cycle */
168 boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */
169 extern int zfs_sync_pass_deferred_free;
172 extern void spa_deadman(void *arg);
176 * This (illegal) pool name is used when temporarily importing a spa_t in order
177 * to get the vdev stats associated with the imported devices.
179 #define TRYIMPORT_NAME "$import"
182 * ==========================================================================
183 * SPA properties routines
184 * ==========================================================================
188 * Add a (source=src, propname=propval) list to an nvlist.
191 spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
192 uint64_t intval, zprop_source_t src)
194 const char *propname = zpool_prop_to_name(prop);
197 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
198 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
201 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
203 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
205 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
206 nvlist_free(propval);
210 * Get property values from the spa configuration.
213 spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
215 vdev_t *rvd = spa->spa_root_vdev;
216 dsl_pool_t *pool = spa->spa_dsl_pool;
217 uint64_t size, alloc, cap, version;
218 zprop_source_t src = ZPROP_SRC_NONE;
219 spa_config_dirent_t *dp;
220 metaslab_class_t *mc = spa_normal_class(spa);
222 ASSERT(MUTEX_HELD(&spa->spa_props_lock));
225 alloc = metaslab_class_get_alloc(spa_normal_class(spa));
226 size = metaslab_class_get_space(spa_normal_class(spa));
227 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
228 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
229 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
230 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
233 spa_prop_add_list(*nvp, ZPOOL_PROP_FRAGMENTATION, NULL,
234 metaslab_class_fragmentation(mc), src);
235 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL,
236 metaslab_class_expandable_space(mc), src);
237 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
238 (spa_mode(spa) == FREAD), src);
240 cap = (size == 0) ? 0 : (alloc * 100 / size);
241 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
243 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
244 ddt_get_pool_dedup_ratio(spa), src);
246 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
247 rvd->vdev_state, src);
249 version = spa_version(spa);
250 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
251 src = ZPROP_SRC_DEFAULT;
253 src = ZPROP_SRC_LOCAL;
254 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
259 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
260 * when opening pools before this version freedir will be NULL.
262 if (pool->dp_free_dir != NULL) {
263 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
264 pool->dp_free_dir->dd_phys->dd_used_bytes, src);
266 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
270 if (pool->dp_leak_dir != NULL) {
271 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, NULL,
272 pool->dp_leak_dir->dd_phys->dd_used_bytes, src);
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_state_dirty_list, sizeof (vdev_t),
1132 offsetof(vdev_t, vdev_state_dirty_node));
1134 txg_list_create(&spa->spa_vdev_txg_list,
1135 offsetof(struct vdev, vdev_txg_node));
1137 avl_create(&spa->spa_errlist_scrub,
1138 spa_error_entry_compare, sizeof (spa_error_entry_t),
1139 offsetof(spa_error_entry_t, se_avl));
1140 avl_create(&spa->spa_errlist_last,
1141 spa_error_entry_compare, sizeof (spa_error_entry_t),
1142 offsetof(spa_error_entry_t, se_avl));
1146 * Opposite of spa_activate().
1149 spa_deactivate(spa_t *spa)
1151 ASSERT(spa->spa_sync_on == B_FALSE);
1152 ASSERT(spa->spa_dsl_pool == NULL);
1153 ASSERT(spa->spa_root_vdev == NULL);
1154 ASSERT(spa->spa_async_zio_root == NULL);
1155 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
1158 * Stop TRIM thread in case spa_unload() wasn't called directly
1159 * before spa_deactivate().
1161 trim_thread_destroy(spa);
1163 txg_list_destroy(&spa->spa_vdev_txg_list);
1165 list_destroy(&spa->spa_config_dirty_list);
1166 list_destroy(&spa->spa_state_dirty_list);
1168 for (int t = 0; t < ZIO_TYPES; t++) {
1169 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1170 spa_taskqs_fini(spa, t, q);
1174 metaslab_class_destroy(spa->spa_normal_class);
1175 spa->spa_normal_class = NULL;
1177 metaslab_class_destroy(spa->spa_log_class);
1178 spa->spa_log_class = NULL;
1181 * If this was part of an import or the open otherwise failed, we may
1182 * still have errors left in the queues. Empty them just in case.
1184 spa_errlog_drain(spa);
1186 avl_destroy(&spa->spa_errlist_scrub);
1187 avl_destroy(&spa->spa_errlist_last);
1189 spa->spa_state = POOL_STATE_UNINITIALIZED;
1191 mutex_enter(&spa->spa_proc_lock);
1192 if (spa->spa_proc_state != SPA_PROC_NONE) {
1193 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1194 spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1195 cv_broadcast(&spa->spa_proc_cv);
1196 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1197 ASSERT(spa->spa_proc != &p0);
1198 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1200 ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1201 spa->spa_proc_state = SPA_PROC_NONE;
1203 ASSERT(spa->spa_proc == &p0);
1204 mutex_exit(&spa->spa_proc_lock);
1208 * We want to make sure spa_thread() has actually exited the ZFS
1209 * module, so that the module can't be unloaded out from underneath
1212 if (spa->spa_did != 0) {
1213 thread_join(spa->spa_did);
1216 #endif /* SPA_PROCESS */
1220 * Verify a pool configuration, and construct the vdev tree appropriately. This
1221 * will create all the necessary vdevs in the appropriate layout, with each vdev
1222 * in the CLOSED state. This will prep the pool before open/creation/import.
1223 * All vdev validation is done by the vdev_alloc() routine.
1226 spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1227 uint_t id, int atype)
1233 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1236 if ((*vdp)->vdev_ops->vdev_op_leaf)
1239 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1242 if (error == ENOENT)
1248 return (SET_ERROR(EINVAL));
1251 for (int c = 0; c < children; c++) {
1253 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1261 ASSERT(*vdp != NULL);
1267 * Opposite of spa_load().
1270 spa_unload(spa_t *spa)
1274 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1279 trim_thread_destroy(spa);
1284 spa_async_suspend(spa);
1289 if (spa->spa_sync_on) {
1290 txg_sync_stop(spa->spa_dsl_pool);
1291 spa->spa_sync_on = B_FALSE;
1295 * Wait for any outstanding async I/O to complete.
1297 if (spa->spa_async_zio_root != NULL) {
1298 for (int i = 0; i < max_ncpus; i++)
1299 (void) zio_wait(spa->spa_async_zio_root[i]);
1300 kmem_free(spa->spa_async_zio_root, max_ncpus * sizeof (void *));
1301 spa->spa_async_zio_root = NULL;
1304 bpobj_close(&spa->spa_deferred_bpobj);
1306 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1311 if (spa->spa_root_vdev)
1312 vdev_free(spa->spa_root_vdev);
1313 ASSERT(spa->spa_root_vdev == NULL);
1316 * Close the dsl pool.
1318 if (spa->spa_dsl_pool) {
1319 dsl_pool_close(spa->spa_dsl_pool);
1320 spa->spa_dsl_pool = NULL;
1321 spa->spa_meta_objset = NULL;
1328 * Drop and purge level 2 cache
1330 spa_l2cache_drop(spa);
1332 for (i = 0; i < spa->spa_spares.sav_count; i++)
1333 vdev_free(spa->spa_spares.sav_vdevs[i]);
1334 if (spa->spa_spares.sav_vdevs) {
1335 kmem_free(spa->spa_spares.sav_vdevs,
1336 spa->spa_spares.sav_count * sizeof (void *));
1337 spa->spa_spares.sav_vdevs = NULL;
1339 if (spa->spa_spares.sav_config) {
1340 nvlist_free(spa->spa_spares.sav_config);
1341 spa->spa_spares.sav_config = NULL;
1343 spa->spa_spares.sav_count = 0;
1345 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1346 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1347 vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1349 if (spa->spa_l2cache.sav_vdevs) {
1350 kmem_free(spa->spa_l2cache.sav_vdevs,
1351 spa->spa_l2cache.sav_count * sizeof (void *));
1352 spa->spa_l2cache.sav_vdevs = NULL;
1354 if (spa->spa_l2cache.sav_config) {
1355 nvlist_free(spa->spa_l2cache.sav_config);
1356 spa->spa_l2cache.sav_config = NULL;
1358 spa->spa_l2cache.sav_count = 0;
1360 spa->spa_async_suspended = 0;
1362 if (spa->spa_comment != NULL) {
1363 spa_strfree(spa->spa_comment);
1364 spa->spa_comment = NULL;
1367 spa_config_exit(spa, SCL_ALL, FTAG);
1371 * Load (or re-load) the current list of vdevs describing the active spares for
1372 * this pool. When this is called, we have some form of basic information in
1373 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1374 * then re-generate a more complete list including status information.
1377 spa_load_spares(spa_t *spa)
1384 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1387 * First, close and free any existing spare vdevs.
1389 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1390 vd = spa->spa_spares.sav_vdevs[i];
1392 /* Undo the call to spa_activate() below */
1393 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1394 B_FALSE)) != NULL && tvd->vdev_isspare)
1395 spa_spare_remove(tvd);
1400 if (spa->spa_spares.sav_vdevs)
1401 kmem_free(spa->spa_spares.sav_vdevs,
1402 spa->spa_spares.sav_count * sizeof (void *));
1404 if (spa->spa_spares.sav_config == NULL)
1407 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1408 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1410 spa->spa_spares.sav_count = (int)nspares;
1411 spa->spa_spares.sav_vdevs = NULL;
1417 * Construct the array of vdevs, opening them to get status in the
1418 * process. For each spare, there is potentially two different vdev_t
1419 * structures associated with it: one in the list of spares (used only
1420 * for basic validation purposes) and one in the active vdev
1421 * configuration (if it's spared in). During this phase we open and
1422 * validate each vdev on the spare list. If the vdev also exists in the
1423 * active configuration, then we also mark this vdev as an active spare.
1425 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1427 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1428 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1429 VDEV_ALLOC_SPARE) == 0);
1432 spa->spa_spares.sav_vdevs[i] = vd;
1434 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1435 B_FALSE)) != NULL) {
1436 if (!tvd->vdev_isspare)
1440 * We only mark the spare active if we were successfully
1441 * able to load the vdev. Otherwise, importing a pool
1442 * with a bad active spare would result in strange
1443 * behavior, because multiple pool would think the spare
1444 * is actively in use.
1446 * There is a vulnerability here to an equally bizarre
1447 * circumstance, where a dead active spare is later
1448 * brought back to life (onlined or otherwise). Given
1449 * the rarity of this scenario, and the extra complexity
1450 * it adds, we ignore the possibility.
1452 if (!vdev_is_dead(tvd))
1453 spa_spare_activate(tvd);
1457 vd->vdev_aux = &spa->spa_spares;
1459 if (vdev_open(vd) != 0)
1462 if (vdev_validate_aux(vd) == 0)
1467 * Recompute the stashed list of spares, with status information
1470 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1471 DATA_TYPE_NVLIST_ARRAY) == 0);
1473 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1475 for (i = 0; i < spa->spa_spares.sav_count; i++)
1476 spares[i] = vdev_config_generate(spa,
1477 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1478 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1479 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1480 for (i = 0; i < spa->spa_spares.sav_count; i++)
1481 nvlist_free(spares[i]);
1482 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1486 * Load (or re-load) the current list of vdevs describing the active l2cache for
1487 * this pool. When this is called, we have some form of basic information in
1488 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1489 * then re-generate a more complete list including status information.
1490 * Devices which are already active have their details maintained, and are
1494 spa_load_l2cache(spa_t *spa)
1498 int i, j, oldnvdevs;
1500 vdev_t *vd, **oldvdevs, **newvdevs;
1501 spa_aux_vdev_t *sav = &spa->spa_l2cache;
1503 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1505 if (sav->sav_config != NULL) {
1506 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1507 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1508 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1514 oldvdevs = sav->sav_vdevs;
1515 oldnvdevs = sav->sav_count;
1516 sav->sav_vdevs = NULL;
1520 * Process new nvlist of vdevs.
1522 for (i = 0; i < nl2cache; i++) {
1523 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1527 for (j = 0; j < oldnvdevs; j++) {
1529 if (vd != NULL && guid == vd->vdev_guid) {
1531 * Retain previous vdev for add/remove ops.
1539 if (newvdevs[i] == NULL) {
1543 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1544 VDEV_ALLOC_L2CACHE) == 0);
1549 * Commit this vdev as an l2cache device,
1550 * even if it fails to open.
1552 spa_l2cache_add(vd);
1557 spa_l2cache_activate(vd);
1559 if (vdev_open(vd) != 0)
1562 (void) vdev_validate_aux(vd);
1564 if (!vdev_is_dead(vd))
1565 l2arc_add_vdev(spa, vd);
1570 * Purge vdevs that were dropped
1572 for (i = 0; i < oldnvdevs; i++) {
1577 ASSERT(vd->vdev_isl2cache);
1579 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1580 pool != 0ULL && l2arc_vdev_present(vd))
1581 l2arc_remove_vdev(vd);
1582 vdev_clear_stats(vd);
1588 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1590 if (sav->sav_config == NULL)
1593 sav->sav_vdevs = newvdevs;
1594 sav->sav_count = (int)nl2cache;
1597 * Recompute the stashed list of l2cache devices, with status
1598 * information this time.
1600 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1601 DATA_TYPE_NVLIST_ARRAY) == 0);
1603 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1604 for (i = 0; i < sav->sav_count; i++)
1605 l2cache[i] = vdev_config_generate(spa,
1606 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1607 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1608 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1610 for (i = 0; i < sav->sav_count; i++)
1611 nvlist_free(l2cache[i]);
1613 kmem_free(l2cache, sav->sav_count * sizeof (void *));
1617 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1620 char *packed = NULL;
1625 error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db);
1628 nvsize = *(uint64_t *)db->db_data;
1629 dmu_buf_rele(db, FTAG);
1631 packed = kmem_alloc(nvsize, KM_SLEEP);
1632 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1635 error = nvlist_unpack(packed, nvsize, value, 0);
1636 kmem_free(packed, nvsize);
1642 * Checks to see if the given vdev could not be opened, in which case we post a
1643 * sysevent to notify the autoreplace code that the device has been removed.
1646 spa_check_removed(vdev_t *vd)
1648 for (int c = 0; c < vd->vdev_children; c++)
1649 spa_check_removed(vd->vdev_child[c]);
1651 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) &&
1653 zfs_post_autoreplace(vd->vdev_spa, vd);
1654 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1659 * Validate the current config against the MOS config
1662 spa_config_valid(spa_t *spa, nvlist_t *config)
1664 vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1667 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1669 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1670 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1672 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1675 * If we're doing a normal import, then build up any additional
1676 * diagnostic information about missing devices in this config.
1677 * We'll pass this up to the user for further processing.
1679 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1680 nvlist_t **child, *nv;
1683 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1685 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1687 for (int c = 0; c < rvd->vdev_children; c++) {
1688 vdev_t *tvd = rvd->vdev_child[c];
1689 vdev_t *mtvd = mrvd->vdev_child[c];
1691 if (tvd->vdev_ops == &vdev_missing_ops &&
1692 mtvd->vdev_ops != &vdev_missing_ops &&
1694 child[idx++] = vdev_config_generate(spa, mtvd,
1699 VERIFY(nvlist_add_nvlist_array(nv,
1700 ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1701 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1702 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1704 for (int i = 0; i < idx; i++)
1705 nvlist_free(child[i]);
1708 kmem_free(child, rvd->vdev_children * sizeof (char **));
1712 * Compare the root vdev tree with the information we have
1713 * from the MOS config (mrvd). Check each top-level vdev
1714 * with the corresponding MOS config top-level (mtvd).
1716 for (int c = 0; c < rvd->vdev_children; c++) {
1717 vdev_t *tvd = rvd->vdev_child[c];
1718 vdev_t *mtvd = mrvd->vdev_child[c];
1721 * Resolve any "missing" vdevs in the current configuration.
1722 * If we find that the MOS config has more accurate information
1723 * about the top-level vdev then use that vdev instead.
1725 if (tvd->vdev_ops == &vdev_missing_ops &&
1726 mtvd->vdev_ops != &vdev_missing_ops) {
1728 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1732 * Device specific actions.
1734 if (mtvd->vdev_islog) {
1735 spa_set_log_state(spa, SPA_LOG_CLEAR);
1738 * XXX - once we have 'readonly' pool
1739 * support we should be able to handle
1740 * missing data devices by transitioning
1741 * the pool to readonly.
1747 * Swap the missing vdev with the data we were
1748 * able to obtain from the MOS config.
1750 vdev_remove_child(rvd, tvd);
1751 vdev_remove_child(mrvd, mtvd);
1753 vdev_add_child(rvd, mtvd);
1754 vdev_add_child(mrvd, tvd);
1756 spa_config_exit(spa, SCL_ALL, FTAG);
1758 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1761 } else if (mtvd->vdev_islog) {
1763 * Load the slog device's state from the MOS config
1764 * since it's possible that the label does not
1765 * contain the most up-to-date information.
1767 vdev_load_log_state(tvd, mtvd);
1772 spa_config_exit(spa, SCL_ALL, FTAG);
1775 * Ensure we were able to validate the config.
1777 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1781 * Check for missing log devices
1784 spa_check_logs(spa_t *spa)
1786 boolean_t rv = B_FALSE;
1788 switch (spa->spa_log_state) {
1789 case SPA_LOG_MISSING:
1790 /* need to recheck in case slog has been restored */
1791 case SPA_LOG_UNKNOWN:
1792 rv = (dmu_objset_find(spa->spa_name, zil_check_log_chain,
1793 NULL, DS_FIND_CHILDREN) != 0);
1795 spa_set_log_state(spa, SPA_LOG_MISSING);
1802 spa_passivate_log(spa_t *spa)
1804 vdev_t *rvd = spa->spa_root_vdev;
1805 boolean_t slog_found = B_FALSE;
1807 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1809 if (!spa_has_slogs(spa))
1812 for (int c = 0; c < rvd->vdev_children; c++) {
1813 vdev_t *tvd = rvd->vdev_child[c];
1814 metaslab_group_t *mg = tvd->vdev_mg;
1816 if (tvd->vdev_islog) {
1817 metaslab_group_passivate(mg);
1818 slog_found = B_TRUE;
1822 return (slog_found);
1826 spa_activate_log(spa_t *spa)
1828 vdev_t *rvd = spa->spa_root_vdev;
1830 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1832 for (int c = 0; c < rvd->vdev_children; c++) {
1833 vdev_t *tvd = rvd->vdev_child[c];
1834 metaslab_group_t *mg = tvd->vdev_mg;
1836 if (tvd->vdev_islog)
1837 metaslab_group_activate(mg);
1842 spa_offline_log(spa_t *spa)
1846 error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1847 NULL, DS_FIND_CHILDREN);
1850 * We successfully offlined the log device, sync out the
1851 * current txg so that the "stubby" block can be removed
1854 txg_wait_synced(spa->spa_dsl_pool, 0);
1860 spa_aux_check_removed(spa_aux_vdev_t *sav)
1864 for (i = 0; i < sav->sav_count; i++)
1865 spa_check_removed(sav->sav_vdevs[i]);
1869 spa_claim_notify(zio_t *zio)
1871 spa_t *spa = zio->io_spa;
1876 mutex_enter(&spa->spa_props_lock); /* any mutex will do */
1877 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1878 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1879 mutex_exit(&spa->spa_props_lock);
1882 typedef struct spa_load_error {
1883 uint64_t sle_meta_count;
1884 uint64_t sle_data_count;
1888 spa_load_verify_done(zio_t *zio)
1890 blkptr_t *bp = zio->io_bp;
1891 spa_load_error_t *sle = zio->io_private;
1892 dmu_object_type_t type = BP_GET_TYPE(bp);
1893 int error = zio->io_error;
1894 spa_t *spa = zio->io_spa;
1897 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1898 type != DMU_OT_INTENT_LOG)
1899 atomic_inc_64(&sle->sle_meta_count);
1901 atomic_inc_64(&sle->sle_data_count);
1903 zio_data_buf_free(zio->io_data, zio->io_size);
1905 mutex_enter(&spa->spa_scrub_lock);
1906 spa->spa_scrub_inflight--;
1907 cv_broadcast(&spa->spa_scrub_io_cv);
1908 mutex_exit(&spa->spa_scrub_lock);
1912 * Maximum number of concurrent scrub i/os to create while verifying
1913 * a pool while importing it.
1915 int spa_load_verify_maxinflight = 10000;
1916 boolean_t spa_load_verify_metadata = B_TRUE;
1917 boolean_t spa_load_verify_data = B_TRUE;
1919 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_maxinflight, CTLFLAG_RWTUN,
1920 &spa_load_verify_maxinflight, 0,
1921 "Maximum number of concurrent scrub I/Os to create while verifying a "
1922 "pool while importing it");
1924 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_metadata, CTLFLAG_RWTUN,
1925 &spa_load_verify_metadata, 0,
1926 "Check metadata on import?");
1928 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_data, CTLFLAG_RWTUN,
1929 &spa_load_verify_data, 0,
1930 "Check user data on import?");
1934 spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1935 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
1937 if (BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp))
1940 * Note: normally this routine will not be called if
1941 * spa_load_verify_metadata is not set. However, it may be useful
1942 * to manually set the flag after the traversal has begun.
1944 if (!spa_load_verify_metadata)
1946 if (BP_GET_BUFC_TYPE(bp) == ARC_BUFC_DATA && !spa_load_verify_data)
1950 size_t size = BP_GET_PSIZE(bp);
1951 void *data = zio_data_buf_alloc(size);
1953 mutex_enter(&spa->spa_scrub_lock);
1954 while (spa->spa_scrub_inflight >= spa_load_verify_maxinflight)
1955 cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
1956 spa->spa_scrub_inflight++;
1957 mutex_exit(&spa->spa_scrub_lock);
1959 zio_nowait(zio_read(rio, spa, bp, data, size,
1960 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1961 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1962 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1967 spa_load_verify(spa_t *spa)
1970 spa_load_error_t sle = { 0 };
1971 zpool_rewind_policy_t policy;
1972 boolean_t verify_ok = B_FALSE;
1975 zpool_get_rewind_policy(spa->spa_config, &policy);
1977 if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1980 rio = zio_root(spa, NULL, &sle,
1981 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1983 if (spa_load_verify_metadata) {
1984 error = traverse_pool(spa, spa->spa_verify_min_txg,
1985 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA,
1986 spa_load_verify_cb, rio);
1989 (void) zio_wait(rio);
1991 spa->spa_load_meta_errors = sle.sle_meta_count;
1992 spa->spa_load_data_errors = sle.sle_data_count;
1994 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1995 sle.sle_data_count <= policy.zrp_maxdata) {
1999 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
2000 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
2002 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
2003 VERIFY(nvlist_add_uint64(spa->spa_load_info,
2004 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
2005 VERIFY(nvlist_add_int64(spa->spa_load_info,
2006 ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
2007 VERIFY(nvlist_add_uint64(spa->spa_load_info,
2008 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
2010 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
2014 if (error != ENXIO && error != EIO)
2015 error = SET_ERROR(EIO);
2019 return (verify_ok ? 0 : EIO);
2023 * Find a value in the pool props object.
2026 spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
2028 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
2029 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
2033 * Find a value in the pool directory object.
2036 spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
2038 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
2039 name, sizeof (uint64_t), 1, val));
2043 spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
2045 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
2050 * Fix up config after a partly-completed split. This is done with the
2051 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
2052 * pool have that entry in their config, but only the splitting one contains
2053 * a list of all the guids of the vdevs that are being split off.
2055 * This function determines what to do with that list: either rejoin
2056 * all the disks to the pool, or complete the splitting process. To attempt
2057 * the rejoin, each disk that is offlined is marked online again, and
2058 * we do a reopen() call. If the vdev label for every disk that was
2059 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
2060 * then we call vdev_split() on each disk, and complete the split.
2062 * Otherwise we leave the config alone, with all the vdevs in place in
2063 * the original pool.
2066 spa_try_repair(spa_t *spa, nvlist_t *config)
2073 boolean_t attempt_reopen;
2075 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
2078 /* check that the config is complete */
2079 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
2080 &glist, &gcount) != 0)
2083 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
2085 /* attempt to online all the vdevs & validate */
2086 attempt_reopen = B_TRUE;
2087 for (i = 0; i < gcount; i++) {
2088 if (glist[i] == 0) /* vdev is hole */
2091 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
2092 if (vd[i] == NULL) {
2094 * Don't bother attempting to reopen the disks;
2095 * just do the split.
2097 attempt_reopen = B_FALSE;
2099 /* attempt to re-online it */
2100 vd[i]->vdev_offline = B_FALSE;
2104 if (attempt_reopen) {
2105 vdev_reopen(spa->spa_root_vdev);
2107 /* check each device to see what state it's in */
2108 for (extracted = 0, i = 0; i < gcount; i++) {
2109 if (vd[i] != NULL &&
2110 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
2117 * If every disk has been moved to the new pool, or if we never
2118 * even attempted to look at them, then we split them off for
2121 if (!attempt_reopen || gcount == extracted) {
2122 for (i = 0; i < gcount; i++)
2125 vdev_reopen(spa->spa_root_vdev);
2128 kmem_free(vd, gcount * sizeof (vdev_t *));
2132 spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
2133 boolean_t mosconfig)
2135 nvlist_t *config = spa->spa_config;
2136 char *ereport = FM_EREPORT_ZFS_POOL;
2142 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
2143 return (SET_ERROR(EINVAL));
2145 ASSERT(spa->spa_comment == NULL);
2146 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
2147 spa->spa_comment = spa_strdup(comment);
2150 * Versioning wasn't explicitly added to the label until later, so if
2151 * it's not present treat it as the initial version.
2153 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
2154 &spa->spa_ubsync.ub_version) != 0)
2155 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
2157 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
2158 &spa->spa_config_txg);
2160 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
2161 spa_guid_exists(pool_guid, 0)) {
2162 error = SET_ERROR(EEXIST);
2164 spa->spa_config_guid = pool_guid;
2166 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
2168 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
2172 nvlist_free(spa->spa_load_info);
2173 spa->spa_load_info = fnvlist_alloc();
2175 gethrestime(&spa->spa_loaded_ts);
2176 error = spa_load_impl(spa, pool_guid, config, state, type,
2177 mosconfig, &ereport);
2180 spa->spa_minref = refcount_count(&spa->spa_refcount);
2182 if (error != EEXIST) {
2183 spa->spa_loaded_ts.tv_sec = 0;
2184 spa->spa_loaded_ts.tv_nsec = 0;
2186 if (error != EBADF) {
2187 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
2190 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2197 * Load an existing storage pool, using the pool's builtin spa_config as a
2198 * source of configuration information.
2201 spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
2202 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
2206 nvlist_t *nvroot = NULL;
2209 uberblock_t *ub = &spa->spa_uberblock;
2210 uint64_t children, config_cache_txg = spa->spa_config_txg;
2211 int orig_mode = spa->spa_mode;
2214 boolean_t missing_feat_write = B_FALSE;
2217 * If this is an untrusted config, access the pool in read-only mode.
2218 * This prevents things like resilvering recently removed devices.
2221 spa->spa_mode = FREAD;
2223 ASSERT(MUTEX_HELD(&spa_namespace_lock));
2225 spa->spa_load_state = state;
2227 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
2228 return (SET_ERROR(EINVAL));
2230 parse = (type == SPA_IMPORT_EXISTING ?
2231 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2234 * Create "The Godfather" zio to hold all async IOs
2236 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
2238 for (int i = 0; i < max_ncpus; i++) {
2239 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
2240 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
2241 ZIO_FLAG_GODFATHER);
2245 * Parse the configuration into a vdev tree. We explicitly set the
2246 * value that will be returned by spa_version() since parsing the
2247 * configuration requires knowing the version number.
2249 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2250 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
2251 spa_config_exit(spa, SCL_ALL, FTAG);
2256 ASSERT(spa->spa_root_vdev == rvd);
2258 if (type != SPA_IMPORT_ASSEMBLE) {
2259 ASSERT(spa_guid(spa) == pool_guid);
2263 * Try to open all vdevs, loading each label in the process.
2265 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2266 error = vdev_open(rvd);
2267 spa_config_exit(spa, SCL_ALL, FTAG);
2272 * We need to validate the vdev labels against the configuration that
2273 * we have in hand, which is dependent on the setting of mosconfig. If
2274 * mosconfig is true then we're validating the vdev labels based on
2275 * that config. Otherwise, we're validating against the cached config
2276 * (zpool.cache) that was read when we loaded the zfs module, and then
2277 * later we will recursively call spa_load() and validate against
2280 * If we're assembling a new pool that's been split off from an
2281 * existing pool, the labels haven't yet been updated so we skip
2282 * validation for now.
2284 if (type != SPA_IMPORT_ASSEMBLE) {
2285 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2286 error = vdev_validate(rvd, mosconfig);
2287 spa_config_exit(spa, SCL_ALL, FTAG);
2292 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2293 return (SET_ERROR(ENXIO));
2297 * Find the best uberblock.
2299 vdev_uberblock_load(rvd, ub, &label);
2302 * If we weren't able to find a single valid uberblock, return failure.
2304 if (ub->ub_txg == 0) {
2306 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2310 * If the pool has an unsupported version we can't open it.
2312 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2314 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2317 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2321 * If we weren't able to find what's necessary for reading the
2322 * MOS in the label, return failure.
2324 if (label == NULL || nvlist_lookup_nvlist(label,
2325 ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) {
2327 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2332 * Update our in-core representation with the definitive values
2335 nvlist_free(spa->spa_label_features);
2336 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2342 * Look through entries in the label nvlist's features_for_read. If
2343 * there is a feature listed there which we don't understand then we
2344 * cannot open a pool.
2346 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2347 nvlist_t *unsup_feat;
2349 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2352 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2354 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2355 if (!zfeature_is_supported(nvpair_name(nvp))) {
2356 VERIFY(nvlist_add_string(unsup_feat,
2357 nvpair_name(nvp), "") == 0);
2361 if (!nvlist_empty(unsup_feat)) {
2362 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2363 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2364 nvlist_free(unsup_feat);
2365 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2369 nvlist_free(unsup_feat);
2373 * If the vdev guid sum doesn't match the uberblock, we have an
2374 * incomplete configuration. We first check to see if the pool
2375 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2376 * If it is, defer the vdev_guid_sum check till later so we
2377 * can handle missing vdevs.
2379 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
2380 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
2381 rvd->vdev_guid_sum != ub->ub_guid_sum)
2382 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2384 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2385 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2386 spa_try_repair(spa, config);
2387 spa_config_exit(spa, SCL_ALL, FTAG);
2388 nvlist_free(spa->spa_config_splitting);
2389 spa->spa_config_splitting = NULL;
2393 * Initialize internal SPA structures.
2395 spa->spa_state = POOL_STATE_ACTIVE;
2396 spa->spa_ubsync = spa->spa_uberblock;
2397 spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2398 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2399 spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2400 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2401 spa->spa_claim_max_txg = spa->spa_first_txg;
2402 spa->spa_prev_software_version = ub->ub_software_version;
2404 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2406 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2407 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2409 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2410 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2412 if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2413 boolean_t missing_feat_read = B_FALSE;
2414 nvlist_t *unsup_feat, *enabled_feat;
2416 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2417 &spa->spa_feat_for_read_obj) != 0) {
2418 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2421 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2422 &spa->spa_feat_for_write_obj) != 0) {
2423 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2426 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2427 &spa->spa_feat_desc_obj) != 0) {
2428 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2431 enabled_feat = fnvlist_alloc();
2432 unsup_feat = fnvlist_alloc();
2434 if (!spa_features_check(spa, B_FALSE,
2435 unsup_feat, enabled_feat))
2436 missing_feat_read = B_TRUE;
2438 if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) {
2439 if (!spa_features_check(spa, B_TRUE,
2440 unsup_feat, enabled_feat)) {
2441 missing_feat_write = B_TRUE;
2445 fnvlist_add_nvlist(spa->spa_load_info,
2446 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
2448 if (!nvlist_empty(unsup_feat)) {
2449 fnvlist_add_nvlist(spa->spa_load_info,
2450 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
2453 fnvlist_free(enabled_feat);
2454 fnvlist_free(unsup_feat);
2456 if (!missing_feat_read) {
2457 fnvlist_add_boolean(spa->spa_load_info,
2458 ZPOOL_CONFIG_CAN_RDONLY);
2462 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2463 * twofold: to determine whether the pool is available for
2464 * import in read-write mode and (if it is not) whether the
2465 * pool is available for import in read-only mode. If the pool
2466 * is available for import in read-write mode, it is displayed
2467 * as available in userland; if it is not available for import
2468 * in read-only mode, it is displayed as unavailable in
2469 * userland. If the pool is available for import in read-only
2470 * mode but not read-write mode, it is displayed as unavailable
2471 * in userland with a special note that the pool is actually
2472 * available for open in read-only mode.
2474 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2475 * missing a feature for write, we must first determine whether
2476 * the pool can be opened read-only before returning to
2477 * userland in order to know whether to display the
2478 * abovementioned note.
2480 if (missing_feat_read || (missing_feat_write &&
2481 spa_writeable(spa))) {
2482 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2487 * Load refcounts for ZFS features from disk into an in-memory
2488 * cache during SPA initialization.
2490 for (spa_feature_t i = 0; i < SPA_FEATURES; i++) {
2493 error = feature_get_refcount_from_disk(spa,
2494 &spa_feature_table[i], &refcount);
2496 spa->spa_feat_refcount_cache[i] = refcount;
2497 } else if (error == ENOTSUP) {
2498 spa->spa_feat_refcount_cache[i] =
2499 SPA_FEATURE_DISABLED;
2501 return (spa_vdev_err(rvd,
2502 VDEV_AUX_CORRUPT_DATA, EIO));
2507 if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) {
2508 if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG,
2509 &spa->spa_feat_enabled_txg_obj) != 0)
2510 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2513 spa->spa_is_initializing = B_TRUE;
2514 error = dsl_pool_open(spa->spa_dsl_pool);
2515 spa->spa_is_initializing = B_FALSE;
2517 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2521 nvlist_t *policy = NULL, *nvconfig;
2523 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2524 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2526 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2527 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2529 unsigned long myhostid = 0;
2531 VERIFY(nvlist_lookup_string(nvconfig,
2532 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2535 myhostid = zone_get_hostid(NULL);
2538 * We're emulating the system's hostid in userland, so
2539 * we can't use zone_get_hostid().
2541 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2542 #endif /* _KERNEL */
2543 if (check_hostid && hostid != 0 && myhostid != 0 &&
2544 hostid != myhostid) {
2545 nvlist_free(nvconfig);
2546 cmn_err(CE_WARN, "pool '%s' could not be "
2547 "loaded as it was last accessed by "
2548 "another system (host: %s hostid: 0x%lx). "
2549 "See: http://illumos.org/msg/ZFS-8000-EY",
2550 spa_name(spa), hostname,
2551 (unsigned long)hostid);
2552 return (SET_ERROR(EBADF));
2555 if (nvlist_lookup_nvlist(spa->spa_config,
2556 ZPOOL_REWIND_POLICY, &policy) == 0)
2557 VERIFY(nvlist_add_nvlist(nvconfig,
2558 ZPOOL_REWIND_POLICY, policy) == 0);
2560 spa_config_set(spa, nvconfig);
2562 spa_deactivate(spa);
2563 spa_activate(spa, orig_mode);
2565 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2568 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2569 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2570 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2572 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2575 * Load the bit that tells us to use the new accounting function
2576 * (raid-z deflation). If we have an older pool, this will not
2579 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2580 if (error != 0 && error != ENOENT)
2581 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2583 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2584 &spa->spa_creation_version);
2585 if (error != 0 && error != ENOENT)
2586 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2589 * Load the persistent error log. If we have an older pool, this will
2592 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2593 if (error != 0 && error != ENOENT)
2594 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2596 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2597 &spa->spa_errlog_scrub);
2598 if (error != 0 && error != ENOENT)
2599 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2602 * Load the history object. If we have an older pool, this
2603 * will not be present.
2605 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2606 if (error != 0 && error != ENOENT)
2607 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2610 * If we're assembling the pool from the split-off vdevs of
2611 * an existing pool, we don't want to attach the spares & cache
2616 * Load any hot spares for this pool.
2618 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2619 if (error != 0 && error != ENOENT)
2620 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2621 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2622 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2623 if (load_nvlist(spa, spa->spa_spares.sav_object,
2624 &spa->spa_spares.sav_config) != 0)
2625 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2627 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2628 spa_load_spares(spa);
2629 spa_config_exit(spa, SCL_ALL, FTAG);
2630 } else if (error == 0) {
2631 spa->spa_spares.sav_sync = B_TRUE;
2635 * Load any level 2 ARC devices for this pool.
2637 error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2638 &spa->spa_l2cache.sav_object);
2639 if (error != 0 && error != ENOENT)
2640 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2641 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2642 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2643 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2644 &spa->spa_l2cache.sav_config) != 0)
2645 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2647 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2648 spa_load_l2cache(spa);
2649 spa_config_exit(spa, SCL_ALL, FTAG);
2650 } else if (error == 0) {
2651 spa->spa_l2cache.sav_sync = B_TRUE;
2654 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2656 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2657 if (error && error != ENOENT)
2658 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2661 uint64_t autoreplace;
2663 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2664 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2665 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2666 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2667 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2668 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2669 &spa->spa_dedup_ditto);
2671 spa->spa_autoreplace = (autoreplace != 0);
2675 * If the 'autoreplace' property is set, then post a resource notifying
2676 * the ZFS DE that it should not issue any faults for unopenable
2677 * devices. We also iterate over the vdevs, and post a sysevent for any
2678 * unopenable vdevs so that the normal autoreplace handler can take
2681 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2682 spa_check_removed(spa->spa_root_vdev);
2684 * For the import case, this is done in spa_import(), because
2685 * at this point we're using the spare definitions from
2686 * the MOS config, not necessarily from the userland config.
2688 if (state != SPA_LOAD_IMPORT) {
2689 spa_aux_check_removed(&spa->spa_spares);
2690 spa_aux_check_removed(&spa->spa_l2cache);
2695 * Load the vdev state for all toplevel vdevs.
2700 * Propagate the leaf DTLs we just loaded all the way up the tree.
2702 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2703 vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2704 spa_config_exit(spa, SCL_ALL, FTAG);
2707 * Load the DDTs (dedup tables).
2709 error = ddt_load(spa);
2711 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2713 spa_update_dspace(spa);
2716 * Validate the config, using the MOS config to fill in any
2717 * information which might be missing. If we fail to validate
2718 * the config then declare the pool unfit for use. If we're
2719 * assembling a pool from a split, the log is not transferred
2722 if (type != SPA_IMPORT_ASSEMBLE) {
2725 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2726 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2728 if (!spa_config_valid(spa, nvconfig)) {
2729 nvlist_free(nvconfig);
2730 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2733 nvlist_free(nvconfig);
2736 * Now that we've validated the config, check the state of the
2737 * root vdev. If it can't be opened, it indicates one or
2738 * more toplevel vdevs are faulted.
2740 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2741 return (SET_ERROR(ENXIO));
2743 if (spa_check_logs(spa)) {
2744 *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2745 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2749 if (missing_feat_write) {
2750 ASSERT(state == SPA_LOAD_TRYIMPORT);
2753 * At this point, we know that we can open the pool in
2754 * read-only mode but not read-write mode. We now have enough
2755 * information and can return to userland.
2757 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
2761 * We've successfully opened the pool, verify that we're ready
2762 * to start pushing transactions.
2764 if (state != SPA_LOAD_TRYIMPORT) {
2765 if (error = spa_load_verify(spa))
2766 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2770 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2771 spa->spa_load_max_txg == UINT64_MAX)) {
2773 int need_update = B_FALSE;
2775 ASSERT(state != SPA_LOAD_TRYIMPORT);
2778 * Claim log blocks that haven't been committed yet.
2779 * This must all happen in a single txg.
2780 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2781 * invoked from zil_claim_log_block()'s i/o done callback.
2782 * Price of rollback is that we abandon the log.
2784 spa->spa_claiming = B_TRUE;
2786 tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2787 spa_first_txg(spa));
2788 (void) dmu_objset_find(spa_name(spa),
2789 zil_claim, tx, DS_FIND_CHILDREN);
2792 spa->spa_claiming = B_FALSE;
2794 spa_set_log_state(spa, SPA_LOG_GOOD);
2795 spa->spa_sync_on = B_TRUE;
2796 txg_sync_start(spa->spa_dsl_pool);
2799 * Wait for all claims to sync. We sync up to the highest
2800 * claimed log block birth time so that claimed log blocks
2801 * don't appear to be from the future. spa_claim_max_txg
2802 * will have been set for us by either zil_check_log_chain()
2803 * (invoked from spa_check_logs()) or zil_claim() above.
2805 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2808 * If the config cache is stale, or we have uninitialized
2809 * metaslabs (see spa_vdev_add()), then update the config.
2811 * If this is a verbatim import, trust the current
2812 * in-core spa_config and update the disk labels.
2814 if (config_cache_txg != spa->spa_config_txg ||
2815 state == SPA_LOAD_IMPORT ||
2816 state == SPA_LOAD_RECOVER ||
2817 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2818 need_update = B_TRUE;
2820 for (int c = 0; c < rvd->vdev_children; c++)
2821 if (rvd->vdev_child[c]->vdev_ms_array == 0)
2822 need_update = B_TRUE;
2825 * Update the config cache asychronously in case we're the
2826 * root pool, in which case the config cache isn't writable yet.
2829 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2832 * Check all DTLs to see if anything needs resilvering.
2834 if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2835 vdev_resilver_needed(rvd, NULL, NULL))
2836 spa_async_request(spa, SPA_ASYNC_RESILVER);
2839 * Log the fact that we booted up (so that we can detect if
2840 * we rebooted in the middle of an operation).
2842 spa_history_log_version(spa, "open");
2845 * Delete any inconsistent datasets.
2847 (void) dmu_objset_find(spa_name(spa),
2848 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2851 * Clean up any stale temporary dataset userrefs.
2853 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2860 spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2862 int mode = spa->spa_mode;
2865 spa_deactivate(spa);
2867 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg - 1;
2869 spa_activate(spa, mode);
2870 spa_async_suspend(spa);
2872 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2876 * If spa_load() fails this function will try loading prior txg's. If
2877 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2878 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2879 * function will not rewind the pool and will return the same error as
2883 spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2884 uint64_t max_request, int rewind_flags)
2886 nvlist_t *loadinfo = NULL;
2887 nvlist_t *config = NULL;
2888 int load_error, rewind_error;
2889 uint64_t safe_rewind_txg;
2892 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2893 spa->spa_load_max_txg = spa->spa_load_txg;
2894 spa_set_log_state(spa, SPA_LOG_CLEAR);
2896 spa->spa_load_max_txg = max_request;
2897 if (max_request != UINT64_MAX)
2898 spa->spa_extreme_rewind = B_TRUE;
2901 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2903 if (load_error == 0)
2906 if (spa->spa_root_vdev != NULL)
2907 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2909 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2910 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2912 if (rewind_flags & ZPOOL_NEVER_REWIND) {
2913 nvlist_free(config);
2914 return (load_error);
2917 if (state == SPA_LOAD_RECOVER) {
2918 /* Price of rolling back is discarding txgs, including log */
2919 spa_set_log_state(spa, SPA_LOG_CLEAR);
2922 * If we aren't rolling back save the load info from our first
2923 * import attempt so that we can restore it after attempting
2926 loadinfo = spa->spa_load_info;
2927 spa->spa_load_info = fnvlist_alloc();
2930 spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2931 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2932 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2933 TXG_INITIAL : safe_rewind_txg;
2936 * Continue as long as we're finding errors, we're still within
2937 * the acceptable rewind range, and we're still finding uberblocks
2939 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2940 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2941 if (spa->spa_load_max_txg < safe_rewind_txg)
2942 spa->spa_extreme_rewind = B_TRUE;
2943 rewind_error = spa_load_retry(spa, state, mosconfig);
2946 spa->spa_extreme_rewind = B_FALSE;
2947 spa->spa_load_max_txg = UINT64_MAX;
2949 if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2950 spa_config_set(spa, config);
2952 if (state == SPA_LOAD_RECOVER) {
2953 ASSERT3P(loadinfo, ==, NULL);
2954 return (rewind_error);
2956 /* Store the rewind info as part of the initial load info */
2957 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
2958 spa->spa_load_info);
2960 /* Restore the initial load info */
2961 fnvlist_free(spa->spa_load_info);
2962 spa->spa_load_info = loadinfo;
2964 return (load_error);
2971 * The import case is identical to an open except that the configuration is sent
2972 * down from userland, instead of grabbed from the configuration cache. For the
2973 * case of an open, the pool configuration will exist in the
2974 * POOL_STATE_UNINITIALIZED state.
2976 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2977 * the same time open the pool, without having to keep around the spa_t in some
2981 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2985 spa_load_state_t state = SPA_LOAD_OPEN;
2987 int locked = B_FALSE;
2988 int firstopen = B_FALSE;
2993 * As disgusting as this is, we need to support recursive calls to this
2994 * function because dsl_dir_open() is called during spa_load(), and ends
2995 * up calling spa_open() again. The real fix is to figure out how to
2996 * avoid dsl_dir_open() calling this in the first place.
2998 if (mutex_owner(&spa_namespace_lock) != curthread) {
2999 mutex_enter(&spa_namespace_lock);
3003 if ((spa = spa_lookup(pool)) == NULL) {
3005 mutex_exit(&spa_namespace_lock);
3006 return (SET_ERROR(ENOENT));
3009 if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
3010 zpool_rewind_policy_t policy;
3014 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
3016 if (policy.zrp_request & ZPOOL_DO_REWIND)
3017 state = SPA_LOAD_RECOVER;
3019 spa_activate(spa, spa_mode_global);
3021 if (state != SPA_LOAD_RECOVER)
3022 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3024 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
3025 policy.zrp_request);
3027 if (error == EBADF) {
3029 * If vdev_validate() returns failure (indicated by
3030 * EBADF), it indicates that one of the vdevs indicates
3031 * that the pool has been exported or destroyed. If
3032 * this is the case, the config cache is out of sync and
3033 * we should remove the pool from the namespace.
3036 spa_deactivate(spa);
3037 spa_config_sync(spa, B_TRUE, B_TRUE);
3040 mutex_exit(&spa_namespace_lock);
3041 return (SET_ERROR(ENOENT));
3046 * We can't open the pool, but we still have useful
3047 * information: the state of each vdev after the
3048 * attempted vdev_open(). Return this to the user.
3050 if (config != NULL && spa->spa_config) {
3051 VERIFY(nvlist_dup(spa->spa_config, config,
3053 VERIFY(nvlist_add_nvlist(*config,
3054 ZPOOL_CONFIG_LOAD_INFO,
3055 spa->spa_load_info) == 0);
3058 spa_deactivate(spa);
3059 spa->spa_last_open_failed = error;
3061 mutex_exit(&spa_namespace_lock);
3067 spa_open_ref(spa, tag);
3070 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
3073 * If we've recovered the pool, pass back any information we
3074 * gathered while doing the load.
3076 if (state == SPA_LOAD_RECOVER) {
3077 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
3078 spa->spa_load_info) == 0);
3082 spa->spa_last_open_failed = 0;
3083 spa->spa_last_ubsync_txg = 0;
3084 spa->spa_load_txg = 0;
3085 mutex_exit(&spa_namespace_lock);
3089 zvol_create_minors(spa->spa_name);
3100 spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
3103 return (spa_open_common(name, spapp, tag, policy, config));
3107 spa_open(const char *name, spa_t **spapp, void *tag)
3109 return (spa_open_common(name, spapp, tag, NULL, NULL));
3113 * Lookup the given spa_t, incrementing the inject count in the process,
3114 * preventing it from being exported or destroyed.
3117 spa_inject_addref(char *name)
3121 mutex_enter(&spa_namespace_lock);
3122 if ((spa = spa_lookup(name)) == NULL) {
3123 mutex_exit(&spa_namespace_lock);
3126 spa->spa_inject_ref++;
3127 mutex_exit(&spa_namespace_lock);
3133 spa_inject_delref(spa_t *spa)
3135 mutex_enter(&spa_namespace_lock);
3136 spa->spa_inject_ref--;
3137 mutex_exit(&spa_namespace_lock);
3141 * Add spares device information to the nvlist.
3144 spa_add_spares(spa_t *spa, nvlist_t *config)
3154 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3156 if (spa->spa_spares.sav_count == 0)
3159 VERIFY(nvlist_lookup_nvlist(config,
3160 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3161 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
3162 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3164 VERIFY(nvlist_add_nvlist_array(nvroot,
3165 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3166 VERIFY(nvlist_lookup_nvlist_array(nvroot,
3167 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3170 * Go through and find any spares which have since been
3171 * repurposed as an active spare. If this is the case, update
3172 * their status appropriately.
3174 for (i = 0; i < nspares; i++) {
3175 VERIFY(nvlist_lookup_uint64(spares[i],
3176 ZPOOL_CONFIG_GUID, &guid) == 0);
3177 if (spa_spare_exists(guid, &pool, NULL) &&
3179 VERIFY(nvlist_lookup_uint64_array(
3180 spares[i], ZPOOL_CONFIG_VDEV_STATS,
3181 (uint64_t **)&vs, &vsc) == 0);
3182 vs->vs_state = VDEV_STATE_CANT_OPEN;
3183 vs->vs_aux = VDEV_AUX_SPARED;
3190 * Add l2cache device information to the nvlist, including vdev stats.
3193 spa_add_l2cache(spa_t *spa, nvlist_t *config)
3196 uint_t i, j, nl2cache;
3203 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3205 if (spa->spa_l2cache.sav_count == 0)
3208 VERIFY(nvlist_lookup_nvlist(config,
3209 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3210 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
3211 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3212 if (nl2cache != 0) {
3213 VERIFY(nvlist_add_nvlist_array(nvroot,
3214 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3215 VERIFY(nvlist_lookup_nvlist_array(nvroot,
3216 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3219 * Update level 2 cache device stats.
3222 for (i = 0; i < nl2cache; i++) {
3223 VERIFY(nvlist_lookup_uint64(l2cache[i],
3224 ZPOOL_CONFIG_GUID, &guid) == 0);
3227 for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
3229 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
3230 vd = spa->spa_l2cache.sav_vdevs[j];
3236 VERIFY(nvlist_lookup_uint64_array(l2cache[i],
3237 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
3239 vdev_get_stats(vd, vs);
3245 spa_add_feature_stats(spa_t *spa, nvlist_t *config)
3251 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3252 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3254 /* We may be unable to read features if pool is suspended. */
3255 if (spa_suspended(spa))
3258 if (spa->spa_feat_for_read_obj != 0) {
3259 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3260 spa->spa_feat_for_read_obj);
3261 zap_cursor_retrieve(&zc, &za) == 0;
3262 zap_cursor_advance(&zc)) {
3263 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3264 za.za_num_integers == 1);
3265 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3266 za.za_first_integer));
3268 zap_cursor_fini(&zc);
3271 if (spa->spa_feat_for_write_obj != 0) {
3272 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3273 spa->spa_feat_for_write_obj);
3274 zap_cursor_retrieve(&zc, &za) == 0;
3275 zap_cursor_advance(&zc)) {
3276 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3277 za.za_num_integers == 1);
3278 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3279 za.za_first_integer));
3281 zap_cursor_fini(&zc);
3285 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
3287 nvlist_free(features);
3291 spa_get_stats(const char *name, nvlist_t **config,
3292 char *altroot, size_t buflen)
3298 error = spa_open_common(name, &spa, FTAG, NULL, config);
3302 * This still leaves a window of inconsistency where the spares
3303 * or l2cache devices could change and the config would be
3304 * self-inconsistent.
3306 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3308 if (*config != NULL) {
3309 uint64_t loadtimes[2];
3311 loadtimes[0] = spa->spa_loaded_ts.tv_sec;
3312 loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
3313 VERIFY(nvlist_add_uint64_array(*config,
3314 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
3316 VERIFY(nvlist_add_uint64(*config,
3317 ZPOOL_CONFIG_ERRCOUNT,
3318 spa_get_errlog_size(spa)) == 0);
3320 if (spa_suspended(spa))
3321 VERIFY(nvlist_add_uint64(*config,
3322 ZPOOL_CONFIG_SUSPENDED,
3323 spa->spa_failmode) == 0);
3325 spa_add_spares(spa, *config);
3326 spa_add_l2cache(spa, *config);
3327 spa_add_feature_stats(spa, *config);
3332 * We want to get the alternate root even for faulted pools, so we cheat
3333 * and call spa_lookup() directly.
3337 mutex_enter(&spa_namespace_lock);
3338 spa = spa_lookup(name);
3340 spa_altroot(spa, altroot, buflen);
3344 mutex_exit(&spa_namespace_lock);
3346 spa_altroot(spa, altroot, buflen);
3351 spa_config_exit(spa, SCL_CONFIG, FTAG);
3352 spa_close(spa, FTAG);
3359 * Validate that the auxiliary device array is well formed. We must have an
3360 * array of nvlists, each which describes a valid leaf vdev. If this is an
3361 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3362 * specified, as long as they are well-formed.
3365 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
3366 spa_aux_vdev_t *sav, const char *config, uint64_t version,
3367 vdev_labeltype_t label)
3374 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3377 * It's acceptable to have no devs specified.
3379 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
3383 return (SET_ERROR(EINVAL));
3386 * Make sure the pool is formatted with a version that supports this
3389 if (spa_version(spa) < version)
3390 return (SET_ERROR(ENOTSUP));
3393 * Set the pending device list so we correctly handle device in-use
3396 sav->sav_pending = dev;
3397 sav->sav_npending = ndev;
3399 for (i = 0; i < ndev; i++) {
3400 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
3404 if (!vd->vdev_ops->vdev_op_leaf) {
3406 error = SET_ERROR(EINVAL);
3411 * The L2ARC currently only supports disk devices in
3412 * kernel context. For user-level testing, we allow it.
3415 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
3416 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
3417 error = SET_ERROR(ENOTBLK);
3424 if ((error = vdev_open(vd)) == 0 &&
3425 (error = vdev_label_init(vd, crtxg, label)) == 0) {
3426 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
3427 vd->vdev_guid) == 0);
3433 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
3440 sav->sav_pending = NULL;
3441 sav->sav_npending = 0;
3446 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
3450 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3452 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3453 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
3454 VDEV_LABEL_SPARE)) != 0) {
3458 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3459 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
3460 VDEV_LABEL_L2CACHE));
3464 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
3469 if (sav->sav_config != NULL) {
3475 * Generate new dev list by concatentating with the
3478 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
3479 &olddevs, &oldndevs) == 0);
3481 newdevs = kmem_alloc(sizeof (void *) *
3482 (ndevs + oldndevs), KM_SLEEP);
3483 for (i = 0; i < oldndevs; i++)
3484 VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
3486 for (i = 0; i < ndevs; i++)
3487 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
3490 VERIFY(nvlist_remove(sav->sav_config, config,
3491 DATA_TYPE_NVLIST_ARRAY) == 0);
3493 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3494 config, newdevs, ndevs + oldndevs) == 0);
3495 for (i = 0; i < oldndevs + ndevs; i++)
3496 nvlist_free(newdevs[i]);
3497 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
3500 * Generate a new dev list.
3502 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
3504 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
3510 * Stop and drop level 2 ARC devices
3513 spa_l2cache_drop(spa_t *spa)
3517 spa_aux_vdev_t *sav = &spa->spa_l2cache;
3519 for (i = 0; i < sav->sav_count; i++) {
3522 vd = sav->sav_vdevs[i];
3525 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
3526 pool != 0ULL && l2arc_vdev_present(vd))
3527 l2arc_remove_vdev(vd);
3535 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
3539 char *altroot = NULL;
3544 uint64_t txg = TXG_INITIAL;
3545 nvlist_t **spares, **l2cache;
3546 uint_t nspares, nl2cache;
3547 uint64_t version, obj;
3548 boolean_t has_features;
3551 * If this pool already exists, return failure.
3553 mutex_enter(&spa_namespace_lock);
3554 if (spa_lookup(pool) != NULL) {
3555 mutex_exit(&spa_namespace_lock);
3556 return (SET_ERROR(EEXIST));
3560 * Allocate a new spa_t structure.
3562 (void) nvlist_lookup_string(props,
3563 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3564 spa = spa_add(pool, NULL, altroot);
3565 spa_activate(spa, spa_mode_global);
3567 if (props && (error = spa_prop_validate(spa, props))) {
3568 spa_deactivate(spa);
3570 mutex_exit(&spa_namespace_lock);
3574 has_features = B_FALSE;
3575 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
3576 elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
3577 if (zpool_prop_feature(nvpair_name(elem)))
3578 has_features = B_TRUE;
3581 if (has_features || nvlist_lookup_uint64(props,
3582 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
3583 version = SPA_VERSION;
3585 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
3587 spa->spa_first_txg = txg;
3588 spa->spa_uberblock.ub_txg = txg - 1;
3589 spa->spa_uberblock.ub_version = version;
3590 spa->spa_ubsync = spa->spa_uberblock;
3593 * Create "The Godfather" zio to hold all async IOs
3595 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
3597 for (int i = 0; i < max_ncpus; i++) {
3598 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
3599 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
3600 ZIO_FLAG_GODFATHER);
3604 * Create the root vdev.
3606 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3608 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3610 ASSERT(error != 0 || rvd != NULL);
3611 ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3613 if (error == 0 && !zfs_allocatable_devs(nvroot))
3614 error = SET_ERROR(EINVAL);
3617 (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3618 (error = spa_validate_aux(spa, nvroot, txg,
3619 VDEV_ALLOC_ADD)) == 0) {
3620 for (int c = 0; c < rvd->vdev_children; c++) {
3621 vdev_ashift_optimize(rvd->vdev_child[c]);
3622 vdev_metaslab_set_size(rvd->vdev_child[c]);
3623 vdev_expand(rvd->vdev_child[c], txg);
3627 spa_config_exit(spa, SCL_ALL, FTAG);
3631 spa_deactivate(spa);
3633 mutex_exit(&spa_namespace_lock);
3638 * Get the list of spares, if specified.
3640 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3641 &spares, &nspares) == 0) {
3642 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3644 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3645 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3646 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3647 spa_load_spares(spa);
3648 spa_config_exit(spa, SCL_ALL, FTAG);
3649 spa->spa_spares.sav_sync = B_TRUE;
3653 * Get the list of level 2 cache devices, if specified.
3655 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3656 &l2cache, &nl2cache) == 0) {
3657 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3658 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3659 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3660 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3661 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3662 spa_load_l2cache(spa);
3663 spa_config_exit(spa, SCL_ALL, FTAG);
3664 spa->spa_l2cache.sav_sync = B_TRUE;
3667 spa->spa_is_initializing = B_TRUE;
3668 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3669 spa->spa_meta_objset = dp->dp_meta_objset;
3670 spa->spa_is_initializing = B_FALSE;
3673 * Create DDTs (dedup tables).
3677 spa_update_dspace(spa);
3679 tx = dmu_tx_create_assigned(dp, txg);
3682 * Create the pool config object.
3684 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3685 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3686 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3688 if (zap_add(spa->spa_meta_objset,
3689 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3690 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3691 cmn_err(CE_PANIC, "failed to add pool config");
3694 if (spa_version(spa) >= SPA_VERSION_FEATURES)
3695 spa_feature_create_zap_objects(spa, tx);
3697 if (zap_add(spa->spa_meta_objset,
3698 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3699 sizeof (uint64_t), 1, &version, tx) != 0) {
3700 cmn_err(CE_PANIC, "failed to add pool version");
3703 /* Newly created pools with the right version are always deflated. */
3704 if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3705 spa->spa_deflate = TRUE;
3706 if (zap_add(spa->spa_meta_objset,
3707 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3708 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3709 cmn_err(CE_PANIC, "failed to add deflate");
3714 * Create the deferred-free bpobj. Turn off compression
3715 * because sync-to-convergence takes longer if the blocksize
3718 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3719 dmu_object_set_compress(spa->spa_meta_objset, obj,
3720 ZIO_COMPRESS_OFF, tx);
3721 if (zap_add(spa->spa_meta_objset,
3722 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3723 sizeof (uint64_t), 1, &obj, tx) != 0) {
3724 cmn_err(CE_PANIC, "failed to add bpobj");
3726 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3727 spa->spa_meta_objset, obj));
3730 * Create the pool's history object.
3732 if (version >= SPA_VERSION_ZPOOL_HISTORY)
3733 spa_history_create_obj(spa, tx);
3736 * Set pool properties.
3738 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3739 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3740 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3741 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3743 if (props != NULL) {
3744 spa_configfile_set(spa, props, B_FALSE);
3745 spa_sync_props(props, tx);
3750 spa->spa_sync_on = B_TRUE;
3751 txg_sync_start(spa->spa_dsl_pool);
3754 * We explicitly wait for the first transaction to complete so that our
3755 * bean counters are appropriately updated.
3757 txg_wait_synced(spa->spa_dsl_pool, txg);
3759 spa_config_sync(spa, B_FALSE, B_TRUE);
3761 spa_history_log_version(spa, "create");
3763 spa->spa_minref = refcount_count(&spa->spa_refcount);
3765 mutex_exit(&spa_namespace_lock);
3773 * Get the root pool information from the root disk, then import the root pool
3774 * during the system boot up time.
3776 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3779 spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3782 nvlist_t *nvtop, *nvroot;
3785 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3789 * Add this top-level vdev to the child array.
3791 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3793 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3795 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3798 * Put this pool's top-level vdevs into a root vdev.
3800 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3801 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3802 VDEV_TYPE_ROOT) == 0);
3803 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3804 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3805 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3809 * Replace the existing vdev_tree with the new root vdev in
3810 * this pool's configuration (remove the old, add the new).
3812 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3813 nvlist_free(nvroot);
3818 * Walk the vdev tree and see if we can find a device with "better"
3819 * configuration. A configuration is "better" if the label on that
3820 * device has a more recent txg.
3823 spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3825 for (int c = 0; c < vd->vdev_children; c++)
3826 spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3828 if (vd->vdev_ops->vdev_op_leaf) {
3832 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3836 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3840 * Do we have a better boot device?
3842 if (label_txg > *txg) {
3851 * Import a root pool.
3853 * For x86. devpath_list will consist of devid and/or physpath name of
3854 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3855 * The GRUB "findroot" command will return the vdev we should boot.
3857 * For Sparc, devpath_list consists the physpath name of the booting device
3858 * no matter the rootpool is a single device pool or a mirrored pool.
3860 * "/pci@1f,0/ide@d/disk@0,0:a"
3863 spa_import_rootpool(char *devpath, char *devid)
3866 vdev_t *rvd, *bvd, *avd = NULL;
3867 nvlist_t *config, *nvtop;
3873 * Read the label from the boot device and generate a configuration.
3875 config = spa_generate_rootconf(devpath, devid, &guid);
3876 #if defined(_OBP) && defined(_KERNEL)
3877 if (config == NULL) {
3878 if (strstr(devpath, "/iscsi/ssd") != NULL) {
3880 get_iscsi_bootpath_phy(devpath);
3881 config = spa_generate_rootconf(devpath, devid, &guid);
3885 if (config == NULL) {
3886 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
3888 return (SET_ERROR(EIO));
3891 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3893 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3895 mutex_enter(&spa_namespace_lock);
3896 if ((spa = spa_lookup(pname)) != NULL) {
3898 * Remove the existing root pool from the namespace so that we
3899 * can replace it with the correct config we just read in.
3904 spa = spa_add(pname, config, NULL);
3905 spa->spa_is_root = B_TRUE;
3906 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3909 * Build up a vdev tree based on the boot device's label config.
3911 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3913 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3914 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3915 VDEV_ALLOC_ROOTPOOL);
3916 spa_config_exit(spa, SCL_ALL, FTAG);
3918 mutex_exit(&spa_namespace_lock);
3919 nvlist_free(config);
3920 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3926 * Get the boot vdev.
3928 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3929 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3930 (u_longlong_t)guid);
3931 error = SET_ERROR(ENOENT);
3936 * Determine if there is a better boot device.
3939 spa_alt_rootvdev(rvd, &avd, &txg);
3941 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3942 "try booting from '%s'", avd->vdev_path);
3943 error = SET_ERROR(EINVAL);
3948 * If the boot device is part of a spare vdev then ensure that
3949 * we're booting off the active spare.
3951 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3952 !bvd->vdev_isspare) {
3953 cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3954 "try booting from '%s'",
3956 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3957 error = SET_ERROR(EINVAL);
3963 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3965 spa_config_exit(spa, SCL_ALL, FTAG);
3966 mutex_exit(&spa_namespace_lock);
3968 nvlist_free(config);
3974 extern int vdev_geom_read_pool_label(const char *name, nvlist_t ***configs,
3978 spa_generate_rootconf(const char *name)
3980 nvlist_t **configs, **tops;
3982 nvlist_t *best_cfg, *nvtop, *nvroot;
3991 if (vdev_geom_read_pool_label(name, &configs, &count) != 0)
3994 ASSERT3U(count, !=, 0);
3996 for (i = 0; i < count; i++) {
3999 VERIFY(nvlist_lookup_uint64(configs[i], ZPOOL_CONFIG_POOL_TXG,
4001 if (txg > best_txg) {
4003 best_cfg = configs[i];
4008 nvlist_lookup_uint64(best_cfg, ZPOOL_CONFIG_VDEV_CHILDREN, &nchildren);
4010 nvlist_lookup_uint64_array(best_cfg, ZPOOL_CONFIG_HOLE_ARRAY,
4013 tops = kmem_zalloc(nchildren * sizeof(void *), KM_SLEEP);
4014 for (i = 0; i < nchildren; i++) {
4017 if (configs[i] == NULL)
4019 VERIFY(nvlist_lookup_nvlist(configs[i], ZPOOL_CONFIG_VDEV_TREE,
4021 nvlist_dup(nvtop, &tops[i], KM_SLEEP);
4023 for (i = 0; holes != NULL && i < nholes; i++) {
4026 if (tops[holes[i]] != NULL)
4028 nvlist_alloc(&tops[holes[i]], NV_UNIQUE_NAME, KM_SLEEP);
4029 VERIFY(nvlist_add_string(tops[holes[i]], ZPOOL_CONFIG_TYPE,
4030 VDEV_TYPE_HOLE) == 0);
4031 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_ID,
4033 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_GUID,
4036 for (i = 0; i < nchildren; i++) {
4037 if (tops[i] != NULL)
4039 nvlist_alloc(&tops[i], NV_UNIQUE_NAME, KM_SLEEP);
4040 VERIFY(nvlist_add_string(tops[i], ZPOOL_CONFIG_TYPE,
4041 VDEV_TYPE_MISSING) == 0);
4042 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_ID,
4044 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_GUID,
4049 * Create pool config based on the best vdev config.
4051 nvlist_dup(best_cfg, &config, KM_SLEEP);
4054 * Put this pool's top-level vdevs into a root vdev.
4056 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4058 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4059 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
4060 VDEV_TYPE_ROOT) == 0);
4061 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
4062 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
4063 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
4064 tops, nchildren) == 0);
4067 * Replace the existing vdev_tree with the new root vdev in
4068 * this pool's configuration (remove the old, add the new).
4070 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
4073 * Drop vdev config elements that should not be present at pool level.
4075 nvlist_remove(config, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64);
4076 nvlist_remove(config, ZPOOL_CONFIG_TOP_GUID, DATA_TYPE_UINT64);
4078 for (i = 0; i < count; i++)
4079 nvlist_free(configs[i]);
4080 kmem_free(configs, count * sizeof(void *));
4081 for (i = 0; i < nchildren; i++)
4082 nvlist_free(tops[i]);
4083 kmem_free(tops, nchildren * sizeof(void *));
4084 nvlist_free(nvroot);
4089 spa_import_rootpool(const char *name)
4092 vdev_t *rvd, *bvd, *avd = NULL;
4093 nvlist_t *config, *nvtop;
4099 * Read the label from the boot device and generate a configuration.
4101 config = spa_generate_rootconf(name);
4103 mutex_enter(&spa_namespace_lock);
4104 if (config != NULL) {
4105 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
4106 &pname) == 0 && strcmp(name, pname) == 0);
4107 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg)
4110 if ((spa = spa_lookup(pname)) != NULL) {
4112 * Remove the existing root pool from the namespace so
4113 * that we can replace it with the correct config
4118 spa = spa_add(pname, config, NULL);
4121 * Set spa_ubsync.ub_version as it can be used in vdev_alloc()
4122 * via spa_version().
4124 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
4125 &spa->spa_ubsync.ub_version) != 0)
4126 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
4127 } else if ((spa = spa_lookup(name)) == NULL) {
4128 cmn_err(CE_NOTE, "Cannot find the pool label for '%s'",
4132 VERIFY(nvlist_dup(spa->spa_config, &config, KM_SLEEP) == 0);
4134 spa->spa_is_root = B_TRUE;
4135 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
4138 * Build up a vdev tree based on the boot device's label config.
4140 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4142 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4143 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
4144 VDEV_ALLOC_ROOTPOOL);
4145 spa_config_exit(spa, SCL_ALL, FTAG);
4147 mutex_exit(&spa_namespace_lock);
4148 nvlist_free(config);
4149 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
4154 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4156 spa_config_exit(spa, SCL_ALL, FTAG);
4157 mutex_exit(&spa_namespace_lock);
4159 nvlist_free(config);
4167 * Import a non-root pool into the system.
4170 spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
4173 char *altroot = NULL;
4174 spa_load_state_t state = SPA_LOAD_IMPORT;
4175 zpool_rewind_policy_t policy;
4176 uint64_t mode = spa_mode_global;
4177 uint64_t readonly = B_FALSE;
4180 nvlist_t **spares, **l2cache;
4181 uint_t nspares, nl2cache;
4184 * If a pool with this name exists, return failure.
4186 mutex_enter(&spa_namespace_lock);
4187 if (spa_lookup(pool) != NULL) {
4188 mutex_exit(&spa_namespace_lock);
4189 return (SET_ERROR(EEXIST));
4193 * Create and initialize the spa structure.
4195 (void) nvlist_lookup_string(props,
4196 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4197 (void) nvlist_lookup_uint64(props,
4198 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
4201 spa = spa_add(pool, config, altroot);
4202 spa->spa_import_flags = flags;
4205 * Verbatim import - Take a pool and insert it into the namespace
4206 * as if it had been loaded at boot.
4208 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
4210 spa_configfile_set(spa, props, B_FALSE);
4212 spa_config_sync(spa, B_FALSE, B_TRUE);
4214 mutex_exit(&spa_namespace_lock);
4218 spa_activate(spa, mode);
4221 * Don't start async tasks until we know everything is healthy.
4223 spa_async_suspend(spa);
4225 zpool_get_rewind_policy(config, &policy);
4226 if (policy.zrp_request & ZPOOL_DO_REWIND)
4227 state = SPA_LOAD_RECOVER;
4230 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
4231 * because the user-supplied config is actually the one to trust when
4234 if (state != SPA_LOAD_RECOVER)
4235 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
4237 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
4238 policy.zrp_request);
4241 * Propagate anything learned while loading the pool and pass it
4242 * back to caller (i.e. rewind info, missing devices, etc).
4244 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4245 spa->spa_load_info) == 0);
4247 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4249 * Toss any existing sparelist, as it doesn't have any validity
4250 * anymore, and conflicts with spa_has_spare().
4252 if (spa->spa_spares.sav_config) {
4253 nvlist_free(spa->spa_spares.sav_config);
4254 spa->spa_spares.sav_config = NULL;
4255 spa_load_spares(spa);
4257 if (spa->spa_l2cache.sav_config) {
4258 nvlist_free(spa->spa_l2cache.sav_config);
4259 spa->spa_l2cache.sav_config = NULL;
4260 spa_load_l2cache(spa);
4263 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4266 error = spa_validate_aux(spa, nvroot, -1ULL,
4269 error = spa_validate_aux(spa, nvroot, -1ULL,
4270 VDEV_ALLOC_L2CACHE);
4271 spa_config_exit(spa, SCL_ALL, FTAG);
4274 spa_configfile_set(spa, props, B_FALSE);
4276 if (error != 0 || (props && spa_writeable(spa) &&
4277 (error = spa_prop_set(spa, props)))) {
4279 spa_deactivate(spa);
4281 mutex_exit(&spa_namespace_lock);
4285 spa_async_resume(spa);
4288 * Override any spares and level 2 cache devices as specified by
4289 * the user, as these may have correct device names/devids, etc.
4291 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
4292 &spares, &nspares) == 0) {
4293 if (spa->spa_spares.sav_config)
4294 VERIFY(nvlist_remove(spa->spa_spares.sav_config,
4295 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
4297 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
4298 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4299 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
4300 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
4301 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4302 spa_load_spares(spa);
4303 spa_config_exit(spa, SCL_ALL, FTAG);
4304 spa->spa_spares.sav_sync = B_TRUE;
4306 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
4307 &l2cache, &nl2cache) == 0) {
4308 if (spa->spa_l2cache.sav_config)
4309 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
4310 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
4312 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
4313 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4314 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
4315 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
4316 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4317 spa_load_l2cache(spa);
4318 spa_config_exit(spa, SCL_ALL, FTAG);
4319 spa->spa_l2cache.sav_sync = B_TRUE;
4323 * Check for any removed devices.
4325 if (spa->spa_autoreplace) {
4326 spa_aux_check_removed(&spa->spa_spares);
4327 spa_aux_check_removed(&spa->spa_l2cache);
4330 if (spa_writeable(spa)) {
4332 * Update the config cache to include the newly-imported pool.
4334 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4338 * It's possible that the pool was expanded while it was exported.
4339 * We kick off an async task to handle this for us.
4341 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
4343 mutex_exit(&spa_namespace_lock);
4344 spa_history_log_version(spa, "import");
4348 zvol_create_minors(pool);
4355 spa_tryimport(nvlist_t *tryconfig)
4357 nvlist_t *config = NULL;
4363 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
4366 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
4370 * Create and initialize the spa structure.
4372 mutex_enter(&spa_namespace_lock);
4373 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
4374 spa_activate(spa, FREAD);
4377 * Pass off the heavy lifting to spa_load().
4378 * Pass TRUE for mosconfig because the user-supplied config
4379 * is actually the one to trust when doing an import.
4381 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
4384 * If 'tryconfig' was at least parsable, return the current config.
4386 if (spa->spa_root_vdev != NULL) {
4387 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4388 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
4390 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4392 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
4393 spa->spa_uberblock.ub_timestamp) == 0);
4394 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4395 spa->spa_load_info) == 0);
4398 * If the bootfs property exists on this pool then we
4399 * copy it out so that external consumers can tell which
4400 * pools are bootable.
4402 if ((!error || error == EEXIST) && spa->spa_bootfs) {
4403 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4406 * We have to play games with the name since the
4407 * pool was opened as TRYIMPORT_NAME.
4409 if (dsl_dsobj_to_dsname(spa_name(spa),
4410 spa->spa_bootfs, tmpname) == 0) {
4412 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4414 cp = strchr(tmpname, '/');
4416 (void) strlcpy(dsname, tmpname,
4419 (void) snprintf(dsname, MAXPATHLEN,
4420 "%s/%s", poolname, ++cp);
4422 VERIFY(nvlist_add_string(config,
4423 ZPOOL_CONFIG_BOOTFS, dsname) == 0);
4424 kmem_free(dsname, MAXPATHLEN);
4426 kmem_free(tmpname, MAXPATHLEN);
4430 * Add the list of hot spares and level 2 cache devices.
4432 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4433 spa_add_spares(spa, config);
4434 spa_add_l2cache(spa, config);
4435 spa_config_exit(spa, SCL_CONFIG, FTAG);
4439 spa_deactivate(spa);
4441 mutex_exit(&spa_namespace_lock);
4447 * Pool export/destroy
4449 * The act of destroying or exporting a pool is very simple. We make sure there
4450 * is no more pending I/O and any references to the pool are gone. Then, we
4451 * update the pool state and sync all the labels to disk, removing the
4452 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4453 * we don't sync the labels or remove the configuration cache.
4456 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
4457 boolean_t force, boolean_t hardforce)
4464 if (!(spa_mode_global & FWRITE))
4465 return (SET_ERROR(EROFS));
4467 mutex_enter(&spa_namespace_lock);
4468 if ((spa = spa_lookup(pool)) == NULL) {
4469 mutex_exit(&spa_namespace_lock);
4470 return (SET_ERROR(ENOENT));
4474 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4475 * reacquire the namespace lock, and see if we can export.
4477 spa_open_ref(spa, FTAG);
4478 mutex_exit(&spa_namespace_lock);
4479 spa_async_suspend(spa);
4480 mutex_enter(&spa_namespace_lock);
4481 spa_close(spa, FTAG);
4484 * The pool will be in core if it's openable,
4485 * in which case we can modify its state.
4487 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
4489 * Objsets may be open only because they're dirty, so we
4490 * have to force it to sync before checking spa_refcnt.
4492 txg_wait_synced(spa->spa_dsl_pool, 0);
4495 * A pool cannot be exported or destroyed if there are active
4496 * references. If we are resetting a pool, allow references by
4497 * fault injection handlers.
4499 if (!spa_refcount_zero(spa) ||
4500 (spa->spa_inject_ref != 0 &&
4501 new_state != POOL_STATE_UNINITIALIZED)) {
4502 spa_async_resume(spa);
4503 mutex_exit(&spa_namespace_lock);
4504 return (SET_ERROR(EBUSY));
4508 * A pool cannot be exported if it has an active shared spare.
4509 * This is to prevent other pools stealing the active spare
4510 * from an exported pool. At user's own will, such pool can
4511 * be forcedly exported.
4513 if (!force && new_state == POOL_STATE_EXPORTED &&
4514 spa_has_active_shared_spare(spa)) {
4515 spa_async_resume(spa);
4516 mutex_exit(&spa_namespace_lock);
4517 return (SET_ERROR(EXDEV));
4521 * We want this to be reflected on every label,
4522 * so mark them all dirty. spa_unload() will do the
4523 * final sync that pushes these changes out.
4525 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
4526 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4527 spa->spa_state = new_state;
4528 spa->spa_final_txg = spa_last_synced_txg(spa) +
4530 vdev_config_dirty(spa->spa_root_vdev);
4531 spa_config_exit(spa, SCL_ALL, FTAG);
4535 spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
4537 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4539 spa_deactivate(spa);
4542 if (oldconfig && spa->spa_config)
4543 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
4545 if (new_state != POOL_STATE_UNINITIALIZED) {
4547 spa_config_sync(spa, B_TRUE, B_TRUE);
4550 mutex_exit(&spa_namespace_lock);
4556 * Destroy a storage pool.
4559 spa_destroy(char *pool)
4561 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
4566 * Export a storage pool.
4569 spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
4570 boolean_t hardforce)
4572 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
4577 * Similar to spa_export(), this unloads the spa_t without actually removing it
4578 * from the namespace in any way.
4581 spa_reset(char *pool)
4583 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
4588 * ==========================================================================
4589 * Device manipulation
4590 * ==========================================================================
4594 * Add a device to a storage pool.
4597 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
4601 vdev_t *rvd = spa->spa_root_vdev;
4603 nvlist_t **spares, **l2cache;
4604 uint_t nspares, nl2cache;
4606 ASSERT(spa_writeable(spa));
4608 txg = spa_vdev_enter(spa);
4610 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
4611 VDEV_ALLOC_ADD)) != 0)
4612 return (spa_vdev_exit(spa, NULL, txg, error));
4614 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */
4616 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
4620 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
4624 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
4625 return (spa_vdev_exit(spa, vd, txg, EINVAL));
4627 if (vd->vdev_children != 0 &&
4628 (error = vdev_create(vd, txg, B_FALSE)) != 0)
4629 return (spa_vdev_exit(spa, vd, txg, error));
4632 * We must validate the spares and l2cache devices after checking the
4633 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
4635 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
4636 return (spa_vdev_exit(spa, vd, txg, error));
4639 * Transfer each new top-level vdev from vd to rvd.
4641 for (int c = 0; c < vd->vdev_children; c++) {
4644 * Set the vdev id to the first hole, if one exists.
4646 for (id = 0; id < rvd->vdev_children; id++) {
4647 if (rvd->vdev_child[id]->vdev_ishole) {
4648 vdev_free(rvd->vdev_child[id]);
4652 tvd = vd->vdev_child[c];
4653 vdev_remove_child(vd, tvd);
4655 vdev_add_child(rvd, tvd);
4656 vdev_config_dirty(tvd);
4660 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
4661 ZPOOL_CONFIG_SPARES);
4662 spa_load_spares(spa);
4663 spa->spa_spares.sav_sync = B_TRUE;
4666 if (nl2cache != 0) {
4667 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
4668 ZPOOL_CONFIG_L2CACHE);
4669 spa_load_l2cache(spa);
4670 spa->spa_l2cache.sav_sync = B_TRUE;
4674 * We have to be careful when adding new vdevs to an existing pool.
4675 * If other threads start allocating from these vdevs before we
4676 * sync the config cache, and we lose power, then upon reboot we may
4677 * fail to open the pool because there are DVAs that the config cache
4678 * can't translate. Therefore, we first add the vdevs without
4679 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4680 * and then let spa_config_update() initialize the new metaslabs.
4682 * spa_load() checks for added-but-not-initialized vdevs, so that
4683 * if we lose power at any point in this sequence, the remaining
4684 * steps will be completed the next time we load the pool.
4686 (void) spa_vdev_exit(spa, vd, txg, 0);
4688 mutex_enter(&spa_namespace_lock);
4689 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4690 mutex_exit(&spa_namespace_lock);
4696 * Attach a device to a mirror. The arguments are the path to any device
4697 * in the mirror, and the nvroot for the new device. If the path specifies
4698 * a device that is not mirrored, we automatically insert the mirror vdev.
4700 * If 'replacing' is specified, the new device is intended to replace the
4701 * existing device; in this case the two devices are made into their own
4702 * mirror using the 'replacing' vdev, which is functionally identical to
4703 * the mirror vdev (it actually reuses all the same ops) but has a few
4704 * extra rules: you can't attach to it after it's been created, and upon
4705 * completion of resilvering, the first disk (the one being replaced)
4706 * is automatically detached.
4709 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
4711 uint64_t txg, dtl_max_txg;
4712 vdev_t *rvd = spa->spa_root_vdev;
4713 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
4715 char *oldvdpath, *newvdpath;
4719 ASSERT(spa_writeable(spa));
4721 txg = spa_vdev_enter(spa);
4723 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
4726 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4728 if (!oldvd->vdev_ops->vdev_op_leaf)
4729 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4731 pvd = oldvd->vdev_parent;
4733 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
4734 VDEV_ALLOC_ATTACH)) != 0)
4735 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4737 if (newrootvd->vdev_children != 1)
4738 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4740 newvd = newrootvd->vdev_child[0];
4742 if (!newvd->vdev_ops->vdev_op_leaf)
4743 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4745 if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
4746 return (spa_vdev_exit(spa, newrootvd, txg, error));
4749 * Spares can't replace logs
4751 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
4752 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4756 * For attach, the only allowable parent is a mirror or the root
4759 if (pvd->vdev_ops != &vdev_mirror_ops &&
4760 pvd->vdev_ops != &vdev_root_ops)
4761 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4763 pvops = &vdev_mirror_ops;
4766 * Active hot spares can only be replaced by inactive hot
4769 if (pvd->vdev_ops == &vdev_spare_ops &&
4770 oldvd->vdev_isspare &&
4771 !spa_has_spare(spa, newvd->vdev_guid))
4772 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4775 * If the source is a hot spare, and the parent isn't already a
4776 * spare, then we want to create a new hot spare. Otherwise, we
4777 * want to create a replacing vdev. The user is not allowed to
4778 * attach to a spared vdev child unless the 'isspare' state is
4779 * the same (spare replaces spare, non-spare replaces
4782 if (pvd->vdev_ops == &vdev_replacing_ops &&
4783 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
4784 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4785 } else if (pvd->vdev_ops == &vdev_spare_ops &&
4786 newvd->vdev_isspare != oldvd->vdev_isspare) {
4787 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4790 if (newvd->vdev_isspare)
4791 pvops = &vdev_spare_ops;
4793 pvops = &vdev_replacing_ops;
4797 * Make sure the new device is big enough.
4799 if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
4800 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
4803 * The new device cannot have a higher alignment requirement
4804 * than the top-level vdev.
4806 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4807 return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4810 * If this is an in-place replacement, update oldvd's path and devid
4811 * to make it distinguishable from newvd, and unopenable from now on.
4813 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4814 spa_strfree(oldvd->vdev_path);
4815 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4817 (void) sprintf(oldvd->vdev_path, "%s/%s",
4818 newvd->vdev_path, "old");
4819 if (oldvd->vdev_devid != NULL) {
4820 spa_strfree(oldvd->vdev_devid);
4821 oldvd->vdev_devid = NULL;
4825 /* mark the device being resilvered */
4826 newvd->vdev_resilver_txg = txg;
4829 * If the parent is not a mirror, or if we're replacing, insert the new
4830 * mirror/replacing/spare vdev above oldvd.
4832 if (pvd->vdev_ops != pvops)
4833 pvd = vdev_add_parent(oldvd, pvops);
4835 ASSERT(pvd->vdev_top->vdev_parent == rvd);
4836 ASSERT(pvd->vdev_ops == pvops);
4837 ASSERT(oldvd->vdev_parent == pvd);
4840 * Extract the new device from its root and add it to pvd.
4842 vdev_remove_child(newrootvd, newvd);
4843 newvd->vdev_id = pvd->vdev_children;
4844 newvd->vdev_crtxg = oldvd->vdev_crtxg;
4845 vdev_add_child(pvd, newvd);
4847 tvd = newvd->vdev_top;
4848 ASSERT(pvd->vdev_top == tvd);
4849 ASSERT(tvd->vdev_parent == rvd);
4851 vdev_config_dirty(tvd);
4854 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4855 * for any dmu_sync-ed blocks. It will propagate upward when
4856 * spa_vdev_exit() calls vdev_dtl_reassess().
4858 dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4860 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4861 dtl_max_txg - TXG_INITIAL);
4863 if (newvd->vdev_isspare) {
4864 spa_spare_activate(newvd);
4865 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
4868 oldvdpath = spa_strdup(oldvd->vdev_path);
4869 newvdpath = spa_strdup(newvd->vdev_path);
4870 newvd_isspare = newvd->vdev_isspare;
4873 * Mark newvd's DTL dirty in this txg.
4875 vdev_dirty(tvd, VDD_DTL, newvd, txg);
4878 * Schedule the resilver to restart in the future. We do this to
4879 * ensure that dmu_sync-ed blocks have been stitched into the
4880 * respective datasets.
4882 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4887 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4889 spa_history_log_internal(spa, "vdev attach", NULL,
4890 "%s vdev=%s %s vdev=%s",
4891 replacing && newvd_isspare ? "spare in" :
4892 replacing ? "replace" : "attach", newvdpath,
4893 replacing ? "for" : "to", oldvdpath);
4895 spa_strfree(oldvdpath);
4896 spa_strfree(newvdpath);
4898 if (spa->spa_bootfs)
4899 spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
4905 * Detach a device from a mirror or replacing vdev.
4907 * If 'replace_done' is specified, only detach if the parent
4908 * is a replacing vdev.
4911 spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4915 vdev_t *rvd = spa->spa_root_vdev;
4916 vdev_t *vd, *pvd, *cvd, *tvd;
4917 boolean_t unspare = B_FALSE;
4918 uint64_t unspare_guid = 0;
4921 ASSERT(spa_writeable(spa));
4923 txg = spa_vdev_enter(spa);
4925 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4928 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4930 if (!vd->vdev_ops->vdev_op_leaf)
4931 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4933 pvd = vd->vdev_parent;
4936 * If the parent/child relationship is not as expected, don't do it.
4937 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4938 * vdev that's replacing B with C. The user's intent in replacing
4939 * is to go from M(A,B) to M(A,C). If the user decides to cancel
4940 * the replace by detaching C, the expected behavior is to end up
4941 * M(A,B). But suppose that right after deciding to detach C,
4942 * the replacement of B completes. We would have M(A,C), and then
4943 * ask to detach C, which would leave us with just A -- not what
4944 * the user wanted. To prevent this, we make sure that the
4945 * parent/child relationship hasn't changed -- in this example,
4946 * that C's parent is still the replacing vdev R.
4948 if (pvd->vdev_guid != pguid && pguid != 0)
4949 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4952 * Only 'replacing' or 'spare' vdevs can be replaced.
4954 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4955 pvd->vdev_ops != &vdev_spare_ops)
4956 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4958 ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4959 spa_version(spa) >= SPA_VERSION_SPARES);
4962 * Only mirror, replacing, and spare vdevs support detach.
4964 if (pvd->vdev_ops != &vdev_replacing_ops &&
4965 pvd->vdev_ops != &vdev_mirror_ops &&
4966 pvd->vdev_ops != &vdev_spare_ops)
4967 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4970 * If this device has the only valid copy of some data,
4971 * we cannot safely detach it.
4973 if (vdev_dtl_required(vd))
4974 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4976 ASSERT(pvd->vdev_children >= 2);
4979 * If we are detaching the second disk from a replacing vdev, then
4980 * check to see if we changed the original vdev's path to have "/old"
4981 * at the end in spa_vdev_attach(). If so, undo that change now.
4983 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4984 vd->vdev_path != NULL) {
4985 size_t len = strlen(vd->vdev_path);
4987 for (int c = 0; c < pvd->vdev_children; c++) {
4988 cvd = pvd->vdev_child[c];
4990 if (cvd == vd || cvd->vdev_path == NULL)
4993 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4994 strcmp(cvd->vdev_path + len, "/old") == 0) {
4995 spa_strfree(cvd->vdev_path);
4996 cvd->vdev_path = spa_strdup(vd->vdev_path);
5003 * If we are detaching the original disk from a spare, then it implies
5004 * that the spare should become a real disk, and be removed from the
5005 * active spare list for the pool.
5007 if (pvd->vdev_ops == &vdev_spare_ops &&
5009 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
5013 * Erase the disk labels so the disk can be used for other things.
5014 * This must be done after all other error cases are handled,
5015 * but before we disembowel vd (so we can still do I/O to it).
5016 * But if we can't do it, don't treat the error as fatal --
5017 * it may be that the unwritability of the disk is the reason
5018 * it's being detached!
5020 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5023 * Remove vd from its parent and compact the parent's children.
5025 vdev_remove_child(pvd, vd);
5026 vdev_compact_children(pvd);
5029 * Remember one of the remaining children so we can get tvd below.
5031 cvd = pvd->vdev_child[pvd->vdev_children - 1];
5034 * If we need to remove the remaining child from the list of hot spares,
5035 * do it now, marking the vdev as no longer a spare in the process.
5036 * We must do this before vdev_remove_parent(), because that can
5037 * change the GUID if it creates a new toplevel GUID. For a similar
5038 * reason, we must remove the spare now, in the same txg as the detach;
5039 * otherwise someone could attach a new sibling, change the GUID, and
5040 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
5043 ASSERT(cvd->vdev_isspare);
5044 spa_spare_remove(cvd);
5045 unspare_guid = cvd->vdev_guid;
5046 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
5047 cvd->vdev_unspare = B_TRUE;
5051 * If the parent mirror/replacing vdev only has one child,
5052 * the parent is no longer needed. Remove it from the tree.
5054 if (pvd->vdev_children == 1) {
5055 if (pvd->vdev_ops == &vdev_spare_ops)
5056 cvd->vdev_unspare = B_FALSE;
5057 vdev_remove_parent(cvd);
5062 * We don't set tvd until now because the parent we just removed
5063 * may have been the previous top-level vdev.
5065 tvd = cvd->vdev_top;
5066 ASSERT(tvd->vdev_parent == rvd);
5069 * Reevaluate the parent vdev state.
5071 vdev_propagate_state(cvd);
5074 * If the 'autoexpand' property is set on the pool then automatically
5075 * try to expand the size of the pool. For example if the device we
5076 * just detached was smaller than the others, it may be possible to
5077 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
5078 * first so that we can obtain the updated sizes of the leaf vdevs.
5080 if (spa->spa_autoexpand) {
5082 vdev_expand(tvd, txg);
5085 vdev_config_dirty(tvd);
5088 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
5089 * vd->vdev_detached is set and free vd's DTL object in syncing context.
5090 * But first make sure we're not on any *other* txg's DTL list, to
5091 * prevent vd from being accessed after it's freed.
5093 vdpath = spa_strdup(vd->vdev_path);
5094 for (int t = 0; t < TXG_SIZE; t++)
5095 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
5096 vd->vdev_detached = B_TRUE;
5097 vdev_dirty(tvd, VDD_DTL, vd, txg);
5099 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
5101 /* hang on to the spa before we release the lock */
5102 spa_open_ref(spa, FTAG);
5104 error = spa_vdev_exit(spa, vd, txg, 0);
5106 spa_history_log_internal(spa, "detach", NULL,
5108 spa_strfree(vdpath);
5111 * If this was the removal of the original device in a hot spare vdev,
5112 * then we want to go through and remove the device from the hot spare
5113 * list of every other pool.
5116 spa_t *altspa = NULL;
5118 mutex_enter(&spa_namespace_lock);
5119 while ((altspa = spa_next(altspa)) != NULL) {
5120 if (altspa->spa_state != POOL_STATE_ACTIVE ||
5124 spa_open_ref(altspa, FTAG);
5125 mutex_exit(&spa_namespace_lock);
5126 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
5127 mutex_enter(&spa_namespace_lock);
5128 spa_close(altspa, FTAG);
5130 mutex_exit(&spa_namespace_lock);
5132 /* search the rest of the vdevs for spares to remove */
5133 spa_vdev_resilver_done(spa);
5136 /* all done with the spa; OK to release */
5137 mutex_enter(&spa_namespace_lock);
5138 spa_close(spa, FTAG);
5139 mutex_exit(&spa_namespace_lock);
5145 * Split a set of devices from their mirrors, and create a new pool from them.
5148 spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
5149 nvlist_t *props, boolean_t exp)
5152 uint64_t txg, *glist;
5154 uint_t c, children, lastlog;
5155 nvlist_t **child, *nvl, *tmp;
5157 char *altroot = NULL;
5158 vdev_t *rvd, **vml = NULL; /* vdev modify list */
5159 boolean_t activate_slog;
5161 ASSERT(spa_writeable(spa));
5163 txg = spa_vdev_enter(spa);
5165 /* clear the log and flush everything up to now */
5166 activate_slog = spa_passivate_log(spa);
5167 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5168 error = spa_offline_log(spa);
5169 txg = spa_vdev_config_enter(spa);
5172 spa_activate_log(spa);
5175 return (spa_vdev_exit(spa, NULL, txg, error));
5177 /* check new spa name before going any further */
5178 if (spa_lookup(newname) != NULL)
5179 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
5182 * scan through all the children to ensure they're all mirrors
5184 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
5185 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
5187 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5189 /* first, check to ensure we've got the right child count */
5190 rvd = spa->spa_root_vdev;
5192 for (c = 0; c < rvd->vdev_children; c++) {
5193 vdev_t *vd = rvd->vdev_child[c];
5195 /* don't count the holes & logs as children */
5196 if (vd->vdev_islog || vd->vdev_ishole) {
5204 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
5205 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5207 /* next, ensure no spare or cache devices are part of the split */
5208 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
5209 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
5210 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5212 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
5213 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
5215 /* then, loop over each vdev and validate it */
5216 for (c = 0; c < children; c++) {
5217 uint64_t is_hole = 0;
5219 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
5223 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
5224 spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
5227 error = SET_ERROR(EINVAL);
5232 /* which disk is going to be split? */
5233 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
5235 error = SET_ERROR(EINVAL);
5239 /* look it up in the spa */
5240 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
5241 if (vml[c] == NULL) {
5242 error = SET_ERROR(ENODEV);
5246 /* make sure there's nothing stopping the split */
5247 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
5248 vml[c]->vdev_islog ||
5249 vml[c]->vdev_ishole ||
5250 vml[c]->vdev_isspare ||
5251 vml[c]->vdev_isl2cache ||
5252 !vdev_writeable(vml[c]) ||
5253 vml[c]->vdev_children != 0 ||
5254 vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
5255 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
5256 error = SET_ERROR(EINVAL);
5260 if (vdev_dtl_required(vml[c])) {
5261 error = SET_ERROR(EBUSY);
5265 /* we need certain info from the top level */
5266 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
5267 vml[c]->vdev_top->vdev_ms_array) == 0);
5268 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
5269 vml[c]->vdev_top->vdev_ms_shift) == 0);
5270 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
5271 vml[c]->vdev_top->vdev_asize) == 0);
5272 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
5273 vml[c]->vdev_top->vdev_ashift) == 0);
5277 kmem_free(vml, children * sizeof (vdev_t *));
5278 kmem_free(glist, children * sizeof (uint64_t));
5279 return (spa_vdev_exit(spa, NULL, txg, error));
5282 /* stop writers from using the disks */
5283 for (c = 0; c < children; c++) {
5285 vml[c]->vdev_offline = B_TRUE;
5287 vdev_reopen(spa->spa_root_vdev);
5290 * Temporarily record the splitting vdevs in the spa config. This
5291 * will disappear once the config is regenerated.
5293 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5294 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
5295 glist, children) == 0);
5296 kmem_free(glist, children * sizeof (uint64_t));
5298 mutex_enter(&spa->spa_props_lock);
5299 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
5301 mutex_exit(&spa->spa_props_lock);
5302 spa->spa_config_splitting = nvl;
5303 vdev_config_dirty(spa->spa_root_vdev);
5305 /* configure and create the new pool */
5306 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
5307 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
5308 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
5309 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5310 spa_version(spa)) == 0);
5311 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
5312 spa->spa_config_txg) == 0);
5313 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
5314 spa_generate_guid(NULL)) == 0);
5315 (void) nvlist_lookup_string(props,
5316 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
5318 /* add the new pool to the namespace */
5319 newspa = spa_add(newname, config, altroot);
5320 newspa->spa_config_txg = spa->spa_config_txg;
5321 spa_set_log_state(newspa, SPA_LOG_CLEAR);
5323 /* release the spa config lock, retaining the namespace lock */
5324 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5326 if (zio_injection_enabled)
5327 zio_handle_panic_injection(spa, FTAG, 1);
5329 spa_activate(newspa, spa_mode_global);
5330 spa_async_suspend(newspa);
5333 /* mark that we are creating new spa by splitting */
5334 newspa->spa_splitting_newspa = B_TRUE;
5336 /* create the new pool from the disks of the original pool */
5337 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
5339 newspa->spa_splitting_newspa = B_FALSE;
5344 /* if that worked, generate a real config for the new pool */
5345 if (newspa->spa_root_vdev != NULL) {
5346 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
5347 NV_UNIQUE_NAME, KM_SLEEP) == 0);
5348 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
5349 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
5350 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
5355 if (props != NULL) {
5356 spa_configfile_set(newspa, props, B_FALSE);
5357 error = spa_prop_set(newspa, props);
5362 /* flush everything */
5363 txg = spa_vdev_config_enter(newspa);
5364 vdev_config_dirty(newspa->spa_root_vdev);
5365 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
5367 if (zio_injection_enabled)
5368 zio_handle_panic_injection(spa, FTAG, 2);
5370 spa_async_resume(newspa);
5372 /* finally, update the original pool's config */
5373 txg = spa_vdev_config_enter(spa);
5374 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
5375 error = dmu_tx_assign(tx, TXG_WAIT);
5378 for (c = 0; c < children; c++) {
5379 if (vml[c] != NULL) {
5382 spa_history_log_internal(spa, "detach", tx,
5383 "vdev=%s", vml[c]->vdev_path);
5387 vdev_config_dirty(spa->spa_root_vdev);
5388 spa->spa_config_splitting = NULL;
5392 (void) spa_vdev_exit(spa, NULL, txg, 0);
5394 if (zio_injection_enabled)
5395 zio_handle_panic_injection(spa, FTAG, 3);
5397 /* split is complete; log a history record */
5398 spa_history_log_internal(newspa, "split", NULL,
5399 "from pool %s", spa_name(spa));
5401 kmem_free(vml, children * sizeof (vdev_t *));
5403 /* if we're not going to mount the filesystems in userland, export */
5405 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
5412 spa_deactivate(newspa);
5415 txg = spa_vdev_config_enter(spa);
5417 /* re-online all offlined disks */
5418 for (c = 0; c < children; c++) {
5420 vml[c]->vdev_offline = B_FALSE;
5422 vdev_reopen(spa->spa_root_vdev);
5424 nvlist_free(spa->spa_config_splitting);
5425 spa->spa_config_splitting = NULL;
5426 (void) spa_vdev_exit(spa, NULL, txg, error);
5428 kmem_free(vml, children * sizeof (vdev_t *));
5433 spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
5435 for (int i = 0; i < count; i++) {
5438 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
5441 if (guid == target_guid)
5449 spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
5450 nvlist_t *dev_to_remove)
5452 nvlist_t **newdev = NULL;
5455 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
5457 for (int i = 0, j = 0; i < count; i++) {
5458 if (dev[i] == dev_to_remove)
5460 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
5463 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
5464 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
5466 for (int i = 0; i < count - 1; i++)
5467 nvlist_free(newdev[i]);
5470 kmem_free(newdev, (count - 1) * sizeof (void *));
5474 * Evacuate the device.
5477 spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
5482 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5483 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5484 ASSERT(vd == vd->vdev_top);
5487 * Evacuate the device. We don't hold the config lock as writer
5488 * since we need to do I/O but we do keep the
5489 * spa_namespace_lock held. Once this completes the device
5490 * should no longer have any blocks allocated on it.
5492 if (vd->vdev_islog) {
5493 if (vd->vdev_stat.vs_alloc != 0)
5494 error = spa_offline_log(spa);
5496 error = SET_ERROR(ENOTSUP);
5503 * The evacuation succeeded. Remove any remaining MOS metadata
5504 * associated with this vdev, and wait for these changes to sync.
5506 ASSERT0(vd->vdev_stat.vs_alloc);
5507 txg = spa_vdev_config_enter(spa);
5508 vd->vdev_removing = B_TRUE;
5509 vdev_dirty_leaves(vd, VDD_DTL, txg);
5510 vdev_config_dirty(vd);
5511 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5517 * Complete the removal by cleaning up the namespace.
5520 spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
5522 vdev_t *rvd = spa->spa_root_vdev;
5523 uint64_t id = vd->vdev_id;
5524 boolean_t last_vdev = (id == (rvd->vdev_children - 1));
5526 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5527 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5528 ASSERT(vd == vd->vdev_top);
5531 * Only remove any devices which are empty.
5533 if (vd->vdev_stat.vs_alloc != 0)
5536 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5538 if (list_link_active(&vd->vdev_state_dirty_node))
5539 vdev_state_clean(vd);
5540 if (list_link_active(&vd->vdev_config_dirty_node))
5541 vdev_config_clean(vd);
5546 vdev_compact_children(rvd);
5548 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
5549 vdev_add_child(rvd, vd);
5551 vdev_config_dirty(rvd);
5554 * Reassess the health of our root vdev.
5560 * Remove a device from the pool -
5562 * Removing a device from the vdev namespace requires several steps
5563 * and can take a significant amount of time. As a result we use
5564 * the spa_vdev_config_[enter/exit] functions which allow us to
5565 * grab and release the spa_config_lock while still holding the namespace
5566 * lock. During each step the configuration is synced out.
5568 * Currently, this supports removing only hot spares, slogs, and level 2 ARC
5572 spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
5575 metaslab_group_t *mg;
5576 nvlist_t **spares, **l2cache, *nv;
5578 uint_t nspares, nl2cache;
5580 boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
5582 ASSERT(spa_writeable(spa));
5585 txg = spa_vdev_enter(spa);
5587 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5589 if (spa->spa_spares.sav_vdevs != NULL &&
5590 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5591 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
5592 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
5594 * Only remove the hot spare if it's not currently in use
5597 if (vd == NULL || unspare) {
5598 spa_vdev_remove_aux(spa->spa_spares.sav_config,
5599 ZPOOL_CONFIG_SPARES, spares, nspares, nv);
5600 spa_load_spares(spa);
5601 spa->spa_spares.sav_sync = B_TRUE;
5603 error = SET_ERROR(EBUSY);
5605 } else if (spa->spa_l2cache.sav_vdevs != NULL &&
5606 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5607 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
5608 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
5610 * Cache devices can always be removed.
5612 spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
5613 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
5614 spa_load_l2cache(spa);
5615 spa->spa_l2cache.sav_sync = B_TRUE;
5616 } else if (vd != NULL && vd->vdev_islog) {
5618 ASSERT(vd == vd->vdev_top);
5623 * Stop allocating from this vdev.
5625 metaslab_group_passivate(mg);
5628 * Wait for the youngest allocations and frees to sync,
5629 * and then wait for the deferral of those frees to finish.
5631 spa_vdev_config_exit(spa, NULL,
5632 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
5635 * Attempt to evacuate the vdev.
5637 error = spa_vdev_remove_evacuate(spa, vd);
5639 txg = spa_vdev_config_enter(spa);
5642 * If we couldn't evacuate the vdev, unwind.
5645 metaslab_group_activate(mg);
5646 return (spa_vdev_exit(spa, NULL, txg, error));
5650 * Clean up the vdev namespace.
5652 spa_vdev_remove_from_namespace(spa, vd);
5654 } else if (vd != NULL) {
5656 * Normal vdevs cannot be removed (yet).
5658 error = SET_ERROR(ENOTSUP);
5661 * There is no vdev of any kind with the specified guid.
5663 error = SET_ERROR(ENOENT);
5667 return (spa_vdev_exit(spa, NULL, txg, error));
5673 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5674 * currently spared, so we can detach it.
5677 spa_vdev_resilver_done_hunt(vdev_t *vd)
5679 vdev_t *newvd, *oldvd;
5681 for (int c = 0; c < vd->vdev_children; c++) {
5682 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
5688 * Check for a completed replacement. We always consider the first
5689 * vdev in the list to be the oldest vdev, and the last one to be
5690 * the newest (see spa_vdev_attach() for how that works). In
5691 * the case where the newest vdev is faulted, we will not automatically
5692 * remove it after a resilver completes. This is OK as it will require
5693 * user intervention to determine which disk the admin wishes to keep.
5695 if (vd->vdev_ops == &vdev_replacing_ops) {
5696 ASSERT(vd->vdev_children > 1);
5698 newvd = vd->vdev_child[vd->vdev_children - 1];
5699 oldvd = vd->vdev_child[0];
5701 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
5702 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5703 !vdev_dtl_required(oldvd))
5708 * Check for a completed resilver with the 'unspare' flag set.
5710 if (vd->vdev_ops == &vdev_spare_ops) {
5711 vdev_t *first = vd->vdev_child[0];
5712 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
5714 if (last->vdev_unspare) {
5717 } else if (first->vdev_unspare) {
5724 if (oldvd != NULL &&
5725 vdev_dtl_empty(newvd, DTL_MISSING) &&
5726 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5727 !vdev_dtl_required(oldvd))
5731 * If there are more than two spares attached to a disk,
5732 * and those spares are not required, then we want to
5733 * attempt to free them up now so that they can be used
5734 * by other pools. Once we're back down to a single
5735 * disk+spare, we stop removing them.
5737 if (vd->vdev_children > 2) {
5738 newvd = vd->vdev_child[1];
5740 if (newvd->vdev_isspare && last->vdev_isspare &&
5741 vdev_dtl_empty(last, DTL_MISSING) &&
5742 vdev_dtl_empty(last, DTL_OUTAGE) &&
5743 !vdev_dtl_required(newvd))
5752 spa_vdev_resilver_done(spa_t *spa)
5754 vdev_t *vd, *pvd, *ppvd;
5755 uint64_t guid, sguid, pguid, ppguid;
5757 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5759 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
5760 pvd = vd->vdev_parent;
5761 ppvd = pvd->vdev_parent;
5762 guid = vd->vdev_guid;
5763 pguid = pvd->vdev_guid;
5764 ppguid = ppvd->vdev_guid;
5767 * If we have just finished replacing a hot spared device, then
5768 * we need to detach the parent's first child (the original hot
5771 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
5772 ppvd->vdev_children == 2) {
5773 ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
5774 sguid = ppvd->vdev_child[1]->vdev_guid;
5776 ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd));
5778 spa_config_exit(spa, SCL_ALL, FTAG);
5779 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
5781 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
5783 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5786 spa_config_exit(spa, SCL_ALL, FTAG);
5790 * Update the stored path or FRU for this vdev.
5793 spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
5797 boolean_t sync = B_FALSE;
5799 ASSERT(spa_writeable(spa));
5801 spa_vdev_state_enter(spa, SCL_ALL);
5803 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
5804 return (spa_vdev_state_exit(spa, NULL, ENOENT));
5806 if (!vd->vdev_ops->vdev_op_leaf)
5807 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
5810 if (strcmp(value, vd->vdev_path) != 0) {
5811 spa_strfree(vd->vdev_path);
5812 vd->vdev_path = spa_strdup(value);
5816 if (vd->vdev_fru == NULL) {
5817 vd->vdev_fru = spa_strdup(value);
5819 } else if (strcmp(value, vd->vdev_fru) != 0) {
5820 spa_strfree(vd->vdev_fru);
5821 vd->vdev_fru = spa_strdup(value);
5826 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
5830 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
5832 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5836 spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5838 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5842 * ==========================================================================
5844 * ==========================================================================
5848 spa_scan_stop(spa_t *spa)
5850 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5851 if (dsl_scan_resilvering(spa->spa_dsl_pool))
5852 return (SET_ERROR(EBUSY));
5853 return (dsl_scan_cancel(spa->spa_dsl_pool));
5857 spa_scan(spa_t *spa, pool_scan_func_t func)
5859 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5861 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5862 return (SET_ERROR(ENOTSUP));
5865 * If a resilver was requested, but there is no DTL on a
5866 * writeable leaf device, we have nothing to do.
5868 if (func == POOL_SCAN_RESILVER &&
5869 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5870 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5874 return (dsl_scan(spa->spa_dsl_pool, func));
5878 * ==========================================================================
5879 * SPA async task processing
5880 * ==========================================================================
5884 spa_async_remove(spa_t *spa, vdev_t *vd)
5886 if (vd->vdev_remove_wanted) {
5887 vd->vdev_remove_wanted = B_FALSE;
5888 vd->vdev_delayed_close = B_FALSE;
5889 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5892 * We want to clear the stats, but we don't want to do a full
5893 * vdev_clear() as that will cause us to throw away
5894 * degraded/faulted state as well as attempt to reopen the
5895 * device, all of which is a waste.
5897 vd->vdev_stat.vs_read_errors = 0;
5898 vd->vdev_stat.vs_write_errors = 0;
5899 vd->vdev_stat.vs_checksum_errors = 0;
5901 vdev_state_dirty(vd->vdev_top);
5904 for (int c = 0; c < vd->vdev_children; c++)
5905 spa_async_remove(spa, vd->vdev_child[c]);
5909 spa_async_probe(spa_t *spa, vdev_t *vd)
5911 if (vd->vdev_probe_wanted) {
5912 vd->vdev_probe_wanted = B_FALSE;
5913 vdev_reopen(vd); /* vdev_open() does the actual probe */
5916 for (int c = 0; c < vd->vdev_children; c++)
5917 spa_async_probe(spa, vd->vdev_child[c]);
5921 spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5927 if (!spa->spa_autoexpand)
5930 for (int c = 0; c < vd->vdev_children; c++) {
5931 vdev_t *cvd = vd->vdev_child[c];
5932 spa_async_autoexpand(spa, cvd);
5935 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5938 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
5939 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
5941 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5942 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
5944 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
5945 ESC_ZFS_VDEV_AUTOEXPAND, attr, &eid, DDI_SLEEP);
5948 kmem_free(physpath, MAXPATHLEN);
5952 spa_async_thread(void *arg)
5957 ASSERT(spa->spa_sync_on);
5959 mutex_enter(&spa->spa_async_lock);
5960 tasks = spa->spa_async_tasks;
5961 spa->spa_async_tasks &= SPA_ASYNC_REMOVE;
5962 mutex_exit(&spa->spa_async_lock);
5965 * See if the config needs to be updated.
5967 if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5968 uint64_t old_space, new_space;
5970 mutex_enter(&spa_namespace_lock);
5971 old_space = metaslab_class_get_space(spa_normal_class(spa));
5972 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5973 new_space = metaslab_class_get_space(spa_normal_class(spa));
5974 mutex_exit(&spa_namespace_lock);
5977 * If the pool grew as a result of the config update,
5978 * then log an internal history event.
5980 if (new_space != old_space) {
5981 spa_history_log_internal(spa, "vdev online", NULL,
5982 "pool '%s' size: %llu(+%llu)",
5983 spa_name(spa), new_space, new_space - old_space);
5987 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5988 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5989 spa_async_autoexpand(spa, spa->spa_root_vdev);
5990 spa_config_exit(spa, SCL_CONFIG, FTAG);
5994 * See if any devices need to be probed.
5996 if (tasks & SPA_ASYNC_PROBE) {
5997 spa_vdev_state_enter(spa, SCL_NONE);
5998 spa_async_probe(spa, spa->spa_root_vdev);
5999 (void) spa_vdev_state_exit(spa, NULL, 0);
6003 * If any devices are done replacing, detach them.
6005 if (tasks & SPA_ASYNC_RESILVER_DONE)
6006 spa_vdev_resilver_done(spa);
6009 * Kick off a resilver.
6011 if (tasks & SPA_ASYNC_RESILVER)
6012 dsl_resilver_restart(spa->spa_dsl_pool, 0);
6015 * Let the world know that we're done.
6017 mutex_enter(&spa->spa_async_lock);
6018 spa->spa_async_thread = NULL;
6019 cv_broadcast(&spa->spa_async_cv);
6020 mutex_exit(&spa->spa_async_lock);
6025 spa_async_thread_vd(void *arg)
6030 ASSERT(spa->spa_sync_on);
6032 mutex_enter(&spa->spa_async_lock);
6033 tasks = spa->spa_async_tasks;
6035 spa->spa_async_tasks &= ~SPA_ASYNC_REMOVE;
6036 mutex_exit(&spa->spa_async_lock);
6039 * See if any devices need to be marked REMOVED.
6041 if (tasks & SPA_ASYNC_REMOVE) {
6042 spa_vdev_state_enter(spa, SCL_NONE);
6043 spa_async_remove(spa, spa->spa_root_vdev);
6044 for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
6045 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
6046 for (int i = 0; i < spa->spa_spares.sav_count; i++)
6047 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
6048 (void) spa_vdev_state_exit(spa, NULL, 0);
6052 * Let the world know that we're done.
6054 mutex_enter(&spa->spa_async_lock);
6055 tasks = spa->spa_async_tasks;
6056 if ((tasks & SPA_ASYNC_REMOVE) != 0)
6058 spa->spa_async_thread_vd = NULL;
6059 cv_broadcast(&spa->spa_async_cv);
6060 mutex_exit(&spa->spa_async_lock);
6065 spa_async_suspend(spa_t *spa)
6067 mutex_enter(&spa->spa_async_lock);
6068 spa->spa_async_suspended++;
6069 while (spa->spa_async_thread != NULL &&
6070 spa->spa_async_thread_vd != NULL)
6071 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
6072 mutex_exit(&spa->spa_async_lock);
6076 spa_async_resume(spa_t *spa)
6078 mutex_enter(&spa->spa_async_lock);
6079 ASSERT(spa->spa_async_suspended != 0);
6080 spa->spa_async_suspended--;
6081 mutex_exit(&spa->spa_async_lock);
6085 spa_async_tasks_pending(spa_t *spa)
6087 uint_t non_config_tasks;
6089 boolean_t config_task_suspended;
6091 non_config_tasks = spa->spa_async_tasks & ~(SPA_ASYNC_CONFIG_UPDATE |
6093 config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE;
6094 if (spa->spa_ccw_fail_time == 0) {
6095 config_task_suspended = B_FALSE;
6097 config_task_suspended =
6098 (gethrtime() - spa->spa_ccw_fail_time) <
6099 (zfs_ccw_retry_interval * NANOSEC);
6102 return (non_config_tasks || (config_task && !config_task_suspended));
6106 spa_async_dispatch(spa_t *spa)
6108 mutex_enter(&spa->spa_async_lock);
6109 if (spa_async_tasks_pending(spa) &&
6110 !spa->spa_async_suspended &&
6111 spa->spa_async_thread == NULL &&
6113 spa->spa_async_thread = thread_create(NULL, 0,
6114 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
6115 mutex_exit(&spa->spa_async_lock);
6119 spa_async_dispatch_vd(spa_t *spa)
6121 mutex_enter(&spa->spa_async_lock);
6122 if ((spa->spa_async_tasks & SPA_ASYNC_REMOVE) != 0 &&
6123 !spa->spa_async_suspended &&
6124 spa->spa_async_thread_vd == NULL &&
6126 spa->spa_async_thread_vd = thread_create(NULL, 0,
6127 spa_async_thread_vd, spa, 0, &p0, TS_RUN, maxclsyspri);
6128 mutex_exit(&spa->spa_async_lock);
6132 spa_async_request(spa_t *spa, int task)
6134 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
6135 mutex_enter(&spa->spa_async_lock);
6136 spa->spa_async_tasks |= task;
6137 mutex_exit(&spa->spa_async_lock);
6138 spa_async_dispatch_vd(spa);
6142 * ==========================================================================
6143 * SPA syncing routines
6144 * ==========================================================================
6148 bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
6151 bpobj_enqueue(bpo, bp, tx);
6156 spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
6160 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
6161 BP_GET_PSIZE(bp), zio->io_flags));
6166 * Note: this simple function is not inlined to make it easier to dtrace the
6167 * amount of time spent syncing frees.
6170 spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx)
6172 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6173 bplist_iterate(bpl, spa_free_sync_cb, zio, tx);
6174 VERIFY(zio_wait(zio) == 0);
6178 * Note: this simple function is not inlined to make it easier to dtrace the
6179 * amount of time spent syncing deferred frees.
6182 spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx)
6184 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6185 VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj,
6186 spa_free_sync_cb, zio, tx), ==, 0);
6187 VERIFY0(zio_wait(zio));
6192 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
6194 char *packed = NULL;
6199 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
6202 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
6203 * information. This avoids the dmu_buf_will_dirty() path and
6204 * saves us a pre-read to get data we don't actually care about.
6206 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
6207 packed = kmem_alloc(bufsize, KM_SLEEP);
6209 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
6211 bzero(packed + nvsize, bufsize - nvsize);
6213 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
6215 kmem_free(packed, bufsize);
6217 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
6218 dmu_buf_will_dirty(db, tx);
6219 *(uint64_t *)db->db_data = nvsize;
6220 dmu_buf_rele(db, FTAG);
6224 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
6225 const char *config, const char *entry)
6235 * Update the MOS nvlist describing the list of available devices.
6236 * spa_validate_aux() will have already made sure this nvlist is
6237 * valid and the vdevs are labeled appropriately.
6239 if (sav->sav_object == 0) {
6240 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
6241 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
6242 sizeof (uint64_t), tx);
6243 VERIFY(zap_update(spa->spa_meta_objset,
6244 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
6245 &sav->sav_object, tx) == 0);
6248 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
6249 if (sav->sav_count == 0) {
6250 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
6252 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
6253 for (i = 0; i < sav->sav_count; i++)
6254 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
6255 B_FALSE, VDEV_CONFIG_L2CACHE);
6256 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
6257 sav->sav_count) == 0);
6258 for (i = 0; i < sav->sav_count; i++)
6259 nvlist_free(list[i]);
6260 kmem_free(list, sav->sav_count * sizeof (void *));
6263 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
6264 nvlist_free(nvroot);
6266 sav->sav_sync = B_FALSE;
6270 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
6274 if (list_is_empty(&spa->spa_config_dirty_list))
6277 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6279 config = spa_config_generate(spa, spa->spa_root_vdev,
6280 dmu_tx_get_txg(tx), B_FALSE);
6283 * If we're upgrading the spa version then make sure that
6284 * the config object gets updated with the correct version.
6286 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
6287 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
6288 spa->spa_uberblock.ub_version);
6290 spa_config_exit(spa, SCL_STATE, FTAG);
6292 if (spa->spa_config_syncing)
6293 nvlist_free(spa->spa_config_syncing);
6294 spa->spa_config_syncing = config;
6296 spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
6300 spa_sync_version(void *arg, dmu_tx_t *tx)
6302 uint64_t *versionp = arg;
6303 uint64_t version = *versionp;
6304 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
6307 * Setting the version is special cased when first creating the pool.
6309 ASSERT(tx->tx_txg != TXG_INITIAL);
6311 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
6312 ASSERT(version >= spa_version(spa));
6314 spa->spa_uberblock.ub_version = version;
6315 vdev_config_dirty(spa->spa_root_vdev);
6316 spa_history_log_internal(spa, "set", tx, "version=%lld", version);
6320 * Set zpool properties.
6323 spa_sync_props(void *arg, dmu_tx_t *tx)
6325 nvlist_t *nvp = arg;
6326 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
6327 objset_t *mos = spa->spa_meta_objset;
6328 nvpair_t *elem = NULL;
6330 mutex_enter(&spa->spa_props_lock);
6332 while ((elem = nvlist_next_nvpair(nvp, elem))) {
6334 char *strval, *fname;
6336 const char *propname;
6337 zprop_type_t proptype;
6340 switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
6343 * We checked this earlier in spa_prop_validate().
6345 ASSERT(zpool_prop_feature(nvpair_name(elem)));
6347 fname = strchr(nvpair_name(elem), '@') + 1;
6348 VERIFY0(zfeature_lookup_name(fname, &fid));
6350 spa_feature_enable(spa, fid, tx);
6351 spa_history_log_internal(spa, "set", tx,
6352 "%s=enabled", nvpair_name(elem));
6355 case ZPOOL_PROP_VERSION:
6356 intval = fnvpair_value_uint64(elem);
6358 * The version is synced seperatly before other
6359 * properties and should be correct by now.
6361 ASSERT3U(spa_version(spa), >=, intval);
6364 case ZPOOL_PROP_ALTROOT:
6366 * 'altroot' is a non-persistent property. It should
6367 * have been set temporarily at creation or import time.
6369 ASSERT(spa->spa_root != NULL);
6372 case ZPOOL_PROP_READONLY:
6373 case ZPOOL_PROP_CACHEFILE:
6375 * 'readonly' and 'cachefile' are also non-persisitent
6379 case ZPOOL_PROP_COMMENT:
6380 strval = fnvpair_value_string(elem);
6381 if (spa->spa_comment != NULL)
6382 spa_strfree(spa->spa_comment);
6383 spa->spa_comment = spa_strdup(strval);
6385 * We need to dirty the configuration on all the vdevs
6386 * so that their labels get updated. It's unnecessary
6387 * to do this for pool creation since the vdev's
6388 * configuratoin has already been dirtied.
6390 if (tx->tx_txg != TXG_INITIAL)
6391 vdev_config_dirty(spa->spa_root_vdev);
6392 spa_history_log_internal(spa, "set", tx,
6393 "%s=%s", nvpair_name(elem), strval);
6397 * Set pool property values in the poolprops mos object.
6399 if (spa->spa_pool_props_object == 0) {
6400 spa->spa_pool_props_object =
6401 zap_create_link(mos, DMU_OT_POOL_PROPS,
6402 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
6406 /* normalize the property name */
6407 propname = zpool_prop_to_name(prop);
6408 proptype = zpool_prop_get_type(prop);
6410 if (nvpair_type(elem) == DATA_TYPE_STRING) {
6411 ASSERT(proptype == PROP_TYPE_STRING);
6412 strval = fnvpair_value_string(elem);
6413 VERIFY0(zap_update(mos,
6414 spa->spa_pool_props_object, propname,
6415 1, strlen(strval) + 1, strval, tx));
6416 spa_history_log_internal(spa, "set", tx,
6417 "%s=%s", nvpair_name(elem), strval);
6418 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
6419 intval = fnvpair_value_uint64(elem);
6421 if (proptype == PROP_TYPE_INDEX) {
6423 VERIFY0(zpool_prop_index_to_string(
6424 prop, intval, &unused));
6426 VERIFY0(zap_update(mos,
6427 spa->spa_pool_props_object, propname,
6428 8, 1, &intval, tx));
6429 spa_history_log_internal(spa, "set", tx,
6430 "%s=%lld", nvpair_name(elem), intval);
6432 ASSERT(0); /* not allowed */
6436 case ZPOOL_PROP_DELEGATION:
6437 spa->spa_delegation = intval;
6439 case ZPOOL_PROP_BOOTFS:
6440 spa->spa_bootfs = intval;
6442 case ZPOOL_PROP_FAILUREMODE:
6443 spa->spa_failmode = intval;
6445 case ZPOOL_PROP_AUTOEXPAND:
6446 spa->spa_autoexpand = intval;
6447 if (tx->tx_txg != TXG_INITIAL)
6448 spa_async_request(spa,
6449 SPA_ASYNC_AUTOEXPAND);
6451 case ZPOOL_PROP_DEDUPDITTO:
6452 spa->spa_dedup_ditto = intval;
6461 mutex_exit(&spa->spa_props_lock);
6465 * Perform one-time upgrade on-disk changes. spa_version() does not
6466 * reflect the new version this txg, so there must be no changes this
6467 * txg to anything that the upgrade code depends on after it executes.
6468 * Therefore this must be called after dsl_pool_sync() does the sync
6472 spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
6474 dsl_pool_t *dp = spa->spa_dsl_pool;
6476 ASSERT(spa->spa_sync_pass == 1);
6478 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
6480 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
6481 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
6482 dsl_pool_create_origin(dp, tx);
6484 /* Keeping the origin open increases spa_minref */
6485 spa->spa_minref += 3;
6488 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
6489 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
6490 dsl_pool_upgrade_clones(dp, tx);
6493 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
6494 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
6495 dsl_pool_upgrade_dir_clones(dp, tx);
6497 /* Keeping the freedir open increases spa_minref */
6498 spa->spa_minref += 3;
6501 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES &&
6502 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
6503 spa_feature_create_zap_objects(spa, tx);
6507 * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable
6508 * when possibility to use lz4 compression for metadata was added
6509 * Old pools that have this feature enabled must be upgraded to have
6510 * this feature active
6512 if (spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
6513 boolean_t lz4_en = spa_feature_is_enabled(spa,
6514 SPA_FEATURE_LZ4_COMPRESS);
6515 boolean_t lz4_ac = spa_feature_is_active(spa,
6516 SPA_FEATURE_LZ4_COMPRESS);
6518 if (lz4_en && !lz4_ac)
6519 spa_feature_incr(spa, SPA_FEATURE_LZ4_COMPRESS, tx);
6521 rrw_exit(&dp->dp_config_rwlock, FTAG);
6525 * Sync the specified transaction group. New blocks may be dirtied as
6526 * part of the process, so we iterate until it converges.
6529 spa_sync(spa_t *spa, uint64_t txg)
6531 dsl_pool_t *dp = spa->spa_dsl_pool;
6532 objset_t *mos = spa->spa_meta_objset;
6533 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
6534 vdev_t *rvd = spa->spa_root_vdev;
6539 VERIFY(spa_writeable(spa));
6542 * Lock out configuration changes.
6544 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
6546 spa->spa_syncing_txg = txg;
6547 spa->spa_sync_pass = 0;
6550 * If there are any pending vdev state changes, convert them
6551 * into config changes that go out with this transaction group.
6553 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6554 while (list_head(&spa->spa_state_dirty_list) != NULL) {
6556 * We need the write lock here because, for aux vdevs,
6557 * calling vdev_config_dirty() modifies sav_config.
6558 * This is ugly and will become unnecessary when we
6559 * eliminate the aux vdev wart by integrating all vdevs
6560 * into the root vdev tree.
6562 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6563 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
6564 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
6565 vdev_state_clean(vd);
6566 vdev_config_dirty(vd);
6568 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6569 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
6571 spa_config_exit(spa, SCL_STATE, FTAG);
6573 tx = dmu_tx_create_assigned(dp, txg);
6575 spa->spa_sync_starttime = gethrtime();
6577 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid,
6578 spa->spa_sync_starttime + spa->spa_deadman_synctime));
6581 callout_reset(&spa->spa_deadman_cycid,
6582 hz * spa->spa_deadman_synctime / NANOSEC, spa_deadman, spa);
6587 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
6588 * set spa_deflate if we have no raid-z vdevs.
6590 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
6591 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
6594 for (i = 0; i < rvd->vdev_children; i++) {
6595 vd = rvd->vdev_child[i];
6596 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
6599 if (i == rvd->vdev_children) {
6600 spa->spa_deflate = TRUE;
6601 VERIFY(0 == zap_add(spa->spa_meta_objset,
6602 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
6603 sizeof (uint64_t), 1, &spa->spa_deflate, tx));
6608 * If anything has changed in this txg, or if someone is waiting
6609 * for this txg to sync (eg, spa_vdev_remove()), push the
6610 * deferred frees from the previous txg. If not, leave them
6611 * alone so that we don't generate work on an otherwise idle
6614 if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
6615 !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
6616 !txg_list_empty(&dp->dp_sync_tasks, txg) ||
6617 ((dsl_scan_active(dp->dp_scan) ||
6618 txg_sync_waiting(dp)) && !spa_shutting_down(spa))) {
6619 spa_sync_deferred_frees(spa, tx);
6623 * Iterate to convergence.
6626 int pass = ++spa->spa_sync_pass;
6628 spa_sync_config_object(spa, tx);
6629 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
6630 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
6631 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
6632 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
6633 spa_errlog_sync(spa, txg);
6634 dsl_pool_sync(dp, txg);
6636 if (pass < zfs_sync_pass_deferred_free) {
6637 spa_sync_frees(spa, free_bpl, tx);
6639 bplist_iterate(free_bpl, bpobj_enqueue_cb,
6640 &spa->spa_deferred_bpobj, tx);
6644 dsl_scan_sync(dp, tx);
6646 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
6650 spa_sync_upgrades(spa, tx);
6652 } while (dmu_objset_is_dirty(mos, txg));
6655 * Rewrite the vdev configuration (which includes the uberblock)
6656 * to commit the transaction group.
6658 * If there are no dirty vdevs, we sync the uberblock to a few
6659 * random top-level vdevs that are known to be visible in the
6660 * config cache (see spa_vdev_add() for a complete description).
6661 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
6665 * We hold SCL_STATE to prevent vdev open/close/etc.
6666 * while we're attempting to write the vdev labels.
6668 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6670 if (list_is_empty(&spa->spa_config_dirty_list)) {
6671 vdev_t *svd[SPA_DVAS_PER_BP];
6673 int children = rvd->vdev_children;
6674 int c0 = spa_get_random(children);
6676 for (int c = 0; c < children; c++) {
6677 vd = rvd->vdev_child[(c0 + c) % children];
6678 if (vd->vdev_ms_array == 0 || vd->vdev_islog)
6680 svd[svdcount++] = vd;
6681 if (svdcount == SPA_DVAS_PER_BP)
6684 error = vdev_config_sync(svd, svdcount, txg, B_FALSE);
6686 error = vdev_config_sync(svd, svdcount, txg,
6689 error = vdev_config_sync(rvd->vdev_child,
6690 rvd->vdev_children, txg, B_FALSE);
6692 error = vdev_config_sync(rvd->vdev_child,
6693 rvd->vdev_children, txg, B_TRUE);
6697 spa->spa_last_synced_guid = rvd->vdev_guid;
6699 spa_config_exit(spa, SCL_STATE, FTAG);
6703 zio_suspend(spa, NULL);
6704 zio_resume_wait(spa);
6709 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY));
6712 callout_drain(&spa->spa_deadman_cycid);
6717 * Clear the dirty config list.
6719 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
6720 vdev_config_clean(vd);
6723 * Now that the new config has synced transactionally,
6724 * let it become visible to the config cache.
6726 if (spa->spa_config_syncing != NULL) {
6727 spa_config_set(spa, spa->spa_config_syncing);
6728 spa->spa_config_txg = txg;
6729 spa->spa_config_syncing = NULL;
6732 spa->spa_ubsync = spa->spa_uberblock;
6734 dsl_pool_sync_done(dp, txg);
6737 * Update usable space statistics.
6739 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
6740 vdev_sync_done(vd, txg);
6742 spa_update_dspace(spa);
6745 * It had better be the case that we didn't dirty anything
6746 * since vdev_config_sync().
6748 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
6749 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
6750 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
6752 spa->spa_sync_pass = 0;
6754 spa_config_exit(spa, SCL_CONFIG, FTAG);
6756 spa_handle_ignored_writes(spa);
6759 * If any async tasks have been requested, kick them off.
6761 spa_async_dispatch(spa);
6762 spa_async_dispatch_vd(spa);
6766 * Sync all pools. We don't want to hold the namespace lock across these
6767 * operations, so we take a reference on the spa_t and drop the lock during the
6771 spa_sync_allpools(void)
6774 mutex_enter(&spa_namespace_lock);
6775 while ((spa = spa_next(spa)) != NULL) {
6776 if (spa_state(spa) != POOL_STATE_ACTIVE ||
6777 !spa_writeable(spa) || spa_suspended(spa))
6779 spa_open_ref(spa, FTAG);
6780 mutex_exit(&spa_namespace_lock);
6781 txg_wait_synced(spa_get_dsl(spa), 0);
6782 mutex_enter(&spa_namespace_lock);
6783 spa_close(spa, FTAG);
6785 mutex_exit(&spa_namespace_lock);
6789 * ==========================================================================
6790 * Miscellaneous routines
6791 * ==========================================================================
6795 * Remove all pools in the system.
6803 * Remove all cached state. All pools should be closed now,
6804 * so every spa in the AVL tree should be unreferenced.
6806 mutex_enter(&spa_namespace_lock);
6807 while ((spa = spa_next(NULL)) != NULL) {
6809 * Stop async tasks. The async thread may need to detach
6810 * a device that's been replaced, which requires grabbing
6811 * spa_namespace_lock, so we must drop it here.
6813 spa_open_ref(spa, FTAG);
6814 mutex_exit(&spa_namespace_lock);
6815 spa_async_suspend(spa);
6816 mutex_enter(&spa_namespace_lock);
6817 spa_close(spa, FTAG);
6819 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
6821 spa_deactivate(spa);
6825 mutex_exit(&spa_namespace_lock);
6829 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
6834 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
6838 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
6839 vd = spa->spa_l2cache.sav_vdevs[i];
6840 if (vd->vdev_guid == guid)
6844 for (i = 0; i < spa->spa_spares.sav_count; i++) {
6845 vd = spa->spa_spares.sav_vdevs[i];
6846 if (vd->vdev_guid == guid)
6855 spa_upgrade(spa_t *spa, uint64_t version)
6857 ASSERT(spa_writeable(spa));
6859 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6862 * This should only be called for a non-faulted pool, and since a
6863 * future version would result in an unopenable pool, this shouldn't be
6866 ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version));
6867 ASSERT3U(version, >=, spa->spa_uberblock.ub_version);
6869 spa->spa_uberblock.ub_version = version;
6870 vdev_config_dirty(spa->spa_root_vdev);
6872 spa_config_exit(spa, SCL_ALL, FTAG);
6874 txg_wait_synced(spa_get_dsl(spa), 0);
6878 spa_has_spare(spa_t *spa, uint64_t guid)
6882 spa_aux_vdev_t *sav = &spa->spa_spares;
6884 for (i = 0; i < sav->sav_count; i++)
6885 if (sav->sav_vdevs[i]->vdev_guid == guid)
6888 for (i = 0; i < sav->sav_npending; i++) {
6889 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
6890 &spareguid) == 0 && spareguid == guid)
6898 * Check if a pool has an active shared spare device.
6899 * Note: reference count of an active spare is 2, as a spare and as a replace
6902 spa_has_active_shared_spare(spa_t *spa)
6906 spa_aux_vdev_t *sav = &spa->spa_spares;
6908 for (i = 0; i < sav->sav_count; i++) {
6909 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
6910 &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
6919 * Post a sysevent corresponding to the given event. The 'name' must be one of
6920 * the event definitions in sys/sysevent/eventdefs.h. The payload will be
6921 * filled in from the spa and (optionally) the vdev. This doesn't do anything
6922 * in the userland libzpool, as we don't want consumers to misinterpret ztest
6923 * or zdb as real changes.
6926 spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
6930 sysevent_attr_list_t *attr = NULL;
6931 sysevent_value_t value;
6934 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
6937 value.value_type = SE_DATA_TYPE_STRING;
6938 value.value.sv_string = spa_name(spa);
6939 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
6942 value.value_type = SE_DATA_TYPE_UINT64;
6943 value.value.sv_uint64 = spa_guid(spa);
6944 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
6948 value.value_type = SE_DATA_TYPE_UINT64;
6949 value.value.sv_uint64 = vd->vdev_guid;
6950 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
6954 if (vd->vdev_path) {
6955 value.value_type = SE_DATA_TYPE_STRING;
6956 value.value.sv_string = vd->vdev_path;
6957 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
6958 &value, SE_SLEEP) != 0)
6963 if (sysevent_attach_attributes(ev, attr) != 0)
6967 (void) log_sysevent(ev, SE_SLEEP, &eid);
6971 sysevent_free_attr(attr);