4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2013 by Delphix. All rights reserved.
25 * Copyright (c) 2015, Nexenta Systems, Inc. All rights reserved.
26 * Copyright (c) 2013 Martin Matuska <mm@FreeBSD.org>. All rights reserved.
27 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
28 * Copyright 2013 Saso Kiselkov. All rights reserved.
32 * SPA: Storage Pool Allocator
34 * This file contains all the routines used when modifying on-disk SPA state.
35 * This includes opening, importing, destroying, exporting a pool, and syncing a
39 #include <sys/zfs_context.h>
40 #include <sys/fm/fs/zfs.h>
41 #include <sys/spa_impl.h>
43 #include <sys/zio_checksum.h>
45 #include <sys/dmu_tx.h>
49 #include <sys/vdev_impl.h>
50 #include <sys/metaslab.h>
51 #include <sys/metaslab_impl.h>
52 #include <sys/uberblock_impl.h>
55 #include <sys/dmu_traverse.h>
56 #include <sys/dmu_objset.h>
57 #include <sys/unique.h>
58 #include <sys/dsl_pool.h>
59 #include <sys/dsl_dataset.h>
60 #include <sys/dsl_dir.h>
61 #include <sys/dsl_prop.h>
62 #include <sys/dsl_synctask.h>
63 #include <sys/fs/zfs.h>
65 #include <sys/callb.h>
66 #include <sys/spa_boot.h>
67 #include <sys/zfs_ioctl.h>
68 #include <sys/dsl_scan.h>
69 #include <sys/dmu_send.h>
70 #include <sys/dsl_destroy.h>
71 #include <sys/dsl_userhold.h>
72 #include <sys/zfeature.h>
74 #include <sys/trim_map.h>
77 #include <sys/callb.h>
78 #include <sys/cpupart.h>
83 #include "zfs_comutil.h"
85 /* Check hostid on import? */
86 static int check_hostid = 1;
88 SYSCTL_DECL(_vfs_zfs);
89 TUNABLE_INT("vfs.zfs.check_hostid", &check_hostid);
90 SYSCTL_INT(_vfs_zfs, OID_AUTO, check_hostid, CTLFLAG_RW, &check_hostid, 0,
91 "Check hostid on import?");
94 * The interval, in seconds, at which failed configuration cache file writes
97 static int zfs_ccw_retry_interval = 300;
99 typedef enum zti_modes {
100 ZTI_MODE_FIXED, /* value is # of threads (min 1) */
101 ZTI_MODE_BATCH, /* cpu-intensive; value is ignored */
102 ZTI_MODE_NULL, /* don't create a taskq */
106 #define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) }
107 #define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 }
108 #define ZTI_NULL { ZTI_MODE_NULL, 0, 0 }
110 #define ZTI_N(n) ZTI_P(n, 1)
111 #define ZTI_ONE ZTI_N(1)
113 typedef struct zio_taskq_info {
114 zti_modes_t zti_mode;
119 static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
120 "issue", "issue_high", "intr", "intr_high"
124 * This table defines the taskq settings for each ZFS I/O type. When
125 * initializing a pool, we use this table to create an appropriately sized
126 * taskq. Some operations are low volume and therefore have a small, static
127 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE
128 * macros. Other operations process a large amount of data; the ZTI_BATCH
129 * macro causes us to create a taskq oriented for throughput. Some operations
130 * are so high frequency and short-lived that the taskq itself can become a a
131 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an
132 * additional degree of parallelism specified by the number of threads per-
133 * taskq and the number of taskqs; when dispatching an event in this case, the
134 * particular taskq is chosen at random.
136 * The different taskq priorities are to handle the different contexts (issue
137 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that
138 * need to be handled with minimum delay.
140 const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
141 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
142 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* NULL */
143 { ZTI_N(8), ZTI_NULL, ZTI_BATCH, ZTI_NULL }, /* READ */
144 { ZTI_BATCH, ZTI_N(5), ZTI_N(8), ZTI_N(5) }, /* WRITE */
145 { ZTI_P(12, 8), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FREE */
146 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* CLAIM */
147 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* IOCTL */
150 static void spa_sync_version(void *arg, dmu_tx_t *tx);
151 static void spa_sync_props(void *arg, dmu_tx_t *tx);
152 static boolean_t spa_has_active_shared_spare(spa_t *spa);
153 static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
154 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
156 static void spa_vdev_resilver_done(spa_t *spa);
158 uint_t zio_taskq_batch_pct = 75; /* 1 thread per cpu in pset */
160 id_t zio_taskq_psrset_bind = PS_NONE;
163 boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */
165 uint_t zio_taskq_basedc = 80; /* base duty cycle */
167 boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */
168 extern int zfs_sync_pass_deferred_free;
171 extern void spa_deadman(void *arg);
175 * This (illegal) pool name is used when temporarily importing a spa_t in order
176 * to get the vdev stats associated with the imported devices.
178 #define TRYIMPORT_NAME "$import"
181 * ==========================================================================
182 * SPA properties routines
183 * ==========================================================================
187 * Add a (source=src, propname=propval) list to an nvlist.
190 spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
191 uint64_t intval, zprop_source_t src)
193 const char *propname = zpool_prop_to_name(prop);
196 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
197 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
200 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
202 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
204 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
205 nvlist_free(propval);
209 * Get property values from the spa configuration.
212 spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
214 vdev_t *rvd = spa->spa_root_vdev;
215 dsl_pool_t *pool = spa->spa_dsl_pool;
216 uint64_t size, alloc, cap, version;
217 zprop_source_t src = ZPROP_SRC_NONE;
218 spa_config_dirent_t *dp;
219 metaslab_class_t *mc = spa_normal_class(spa);
221 ASSERT(MUTEX_HELD(&spa->spa_props_lock));
224 alloc = metaslab_class_get_alloc(spa_normal_class(spa));
225 size = metaslab_class_get_space(spa_normal_class(spa));
226 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
227 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
228 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
229 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
232 spa_prop_add_list(*nvp, ZPOOL_PROP_FRAGMENTATION, NULL,
233 metaslab_class_fragmentation(mc), src);
234 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL,
235 metaslab_class_expandable_space(mc), src);
236 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
237 (spa_mode(spa) == FREAD), src);
239 cap = (size == 0) ? 0 : (alloc * 100 / size);
240 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
242 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
243 ddt_get_pool_dedup_ratio(spa), src);
245 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
246 rvd->vdev_state, src);
248 version = spa_version(spa);
249 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
250 src = ZPROP_SRC_DEFAULT;
252 src = ZPROP_SRC_LOCAL;
253 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
258 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
259 * when opening pools before this version freedir will be NULL.
261 if (pool->dp_free_dir != NULL) {
262 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
263 dsl_dir_phys(pool->dp_free_dir)->dd_used_bytes,
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 dsl_dir_phys(pool->dp_leak_dir)->dd_used_bytes,
275 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED,
280 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
282 if (spa->spa_comment != NULL) {
283 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
287 if (spa->spa_root != NULL)
288 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
291 if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) {
292 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL,
293 MIN(zfs_max_recordsize, SPA_MAXBLOCKSIZE), ZPROP_SRC_NONE);
295 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL,
296 SPA_OLD_MAXBLOCKSIZE, ZPROP_SRC_NONE);
299 if ((dp = list_head(&spa->spa_config_list)) != NULL) {
300 if (dp->scd_path == NULL) {
301 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
302 "none", 0, ZPROP_SRC_LOCAL);
303 } else if (strcmp(dp->scd_path, spa_config_path) != 0) {
304 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
305 dp->scd_path, 0, ZPROP_SRC_LOCAL);
311 * Get zpool property values.
314 spa_prop_get(spa_t *spa, nvlist_t **nvp)
316 objset_t *mos = spa->spa_meta_objset;
321 VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
323 mutex_enter(&spa->spa_props_lock);
326 * Get properties from the spa config.
328 spa_prop_get_config(spa, nvp);
330 /* If no pool property object, no more prop to get. */
331 if (mos == NULL || spa->spa_pool_props_object == 0) {
332 mutex_exit(&spa->spa_props_lock);
337 * Get properties from the MOS pool property object.
339 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
340 (err = zap_cursor_retrieve(&zc, &za)) == 0;
341 zap_cursor_advance(&zc)) {
344 zprop_source_t src = ZPROP_SRC_DEFAULT;
347 if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
350 switch (za.za_integer_length) {
352 /* integer property */
353 if (za.za_first_integer !=
354 zpool_prop_default_numeric(prop))
355 src = ZPROP_SRC_LOCAL;
357 if (prop == ZPOOL_PROP_BOOTFS) {
359 dsl_dataset_t *ds = NULL;
361 dp = spa_get_dsl(spa);
362 dsl_pool_config_enter(dp, FTAG);
363 if (err = dsl_dataset_hold_obj(dp,
364 za.za_first_integer, FTAG, &ds)) {
365 dsl_pool_config_exit(dp, FTAG);
370 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
372 dsl_dataset_name(ds, strval);
373 dsl_dataset_rele(ds, FTAG);
374 dsl_pool_config_exit(dp, FTAG);
377 intval = za.za_first_integer;
380 spa_prop_add_list(*nvp, prop, strval, intval, src);
384 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
389 /* string property */
390 strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
391 err = zap_lookup(mos, spa->spa_pool_props_object,
392 za.za_name, 1, za.za_num_integers, strval);
394 kmem_free(strval, za.za_num_integers);
397 spa_prop_add_list(*nvp, prop, strval, 0, src);
398 kmem_free(strval, za.za_num_integers);
405 zap_cursor_fini(&zc);
406 mutex_exit(&spa->spa_props_lock);
408 if (err && err != ENOENT) {
418 * Validate the given pool properties nvlist and modify the list
419 * for the property values to be set.
422 spa_prop_validate(spa_t *spa, nvlist_t *props)
425 int error = 0, reset_bootfs = 0;
427 boolean_t has_feature = B_FALSE;
430 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
432 char *strval, *slash, *check, *fname;
433 const char *propname = nvpair_name(elem);
434 zpool_prop_t prop = zpool_name_to_prop(propname);
438 if (!zpool_prop_feature(propname)) {
439 error = SET_ERROR(EINVAL);
444 * Sanitize the input.
446 if (nvpair_type(elem) != DATA_TYPE_UINT64) {
447 error = SET_ERROR(EINVAL);
451 if (nvpair_value_uint64(elem, &intval) != 0) {
452 error = SET_ERROR(EINVAL);
457 error = SET_ERROR(EINVAL);
461 fname = strchr(propname, '@') + 1;
462 if (zfeature_lookup_name(fname, NULL) != 0) {
463 error = SET_ERROR(EINVAL);
467 has_feature = B_TRUE;
470 case ZPOOL_PROP_VERSION:
471 error = nvpair_value_uint64(elem, &intval);
473 (intval < spa_version(spa) ||
474 intval > SPA_VERSION_BEFORE_FEATURES ||
476 error = SET_ERROR(EINVAL);
479 case ZPOOL_PROP_DELEGATION:
480 case ZPOOL_PROP_AUTOREPLACE:
481 case ZPOOL_PROP_LISTSNAPS:
482 case ZPOOL_PROP_AUTOEXPAND:
483 error = nvpair_value_uint64(elem, &intval);
484 if (!error && intval > 1)
485 error = SET_ERROR(EINVAL);
488 case ZPOOL_PROP_BOOTFS:
490 * If the pool version is less than SPA_VERSION_BOOTFS,
491 * or the pool is still being created (version == 0),
492 * the bootfs property cannot be set.
494 if (spa_version(spa) < SPA_VERSION_BOOTFS) {
495 error = SET_ERROR(ENOTSUP);
500 * Make sure the vdev config is bootable
502 if (!vdev_is_bootable(spa->spa_root_vdev)) {
503 error = SET_ERROR(ENOTSUP);
509 error = nvpair_value_string(elem, &strval);
515 if (strval == NULL || strval[0] == '\0') {
516 objnum = zpool_prop_default_numeric(
521 if (error = dmu_objset_hold(strval, FTAG, &os))
525 * Must be ZPL, and its property settings
526 * must be supported by GRUB (compression
527 * is not gzip, and large blocks are not used).
530 if (dmu_objset_type(os) != DMU_OST_ZFS) {
531 error = SET_ERROR(ENOTSUP);
533 dsl_prop_get_int_ds(dmu_objset_ds(os),
534 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
536 !BOOTFS_COMPRESS_VALID(propval)) {
537 error = SET_ERROR(ENOTSUP);
539 dsl_prop_get_int_ds(dmu_objset_ds(os),
540 zfs_prop_to_name(ZFS_PROP_RECORDSIZE),
542 propval > SPA_OLD_MAXBLOCKSIZE) {
543 error = SET_ERROR(ENOTSUP);
545 objnum = dmu_objset_id(os);
547 dmu_objset_rele(os, FTAG);
551 case ZPOOL_PROP_FAILUREMODE:
552 error = nvpair_value_uint64(elem, &intval);
553 if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
554 intval > ZIO_FAILURE_MODE_PANIC))
555 error = SET_ERROR(EINVAL);
558 * This is a special case which only occurs when
559 * the pool has completely failed. This allows
560 * the user to change the in-core failmode property
561 * without syncing it out to disk (I/Os might
562 * currently be blocked). We do this by returning
563 * EIO to the caller (spa_prop_set) to trick it
564 * into thinking we encountered a property validation
567 if (!error && spa_suspended(spa)) {
568 spa->spa_failmode = intval;
569 error = SET_ERROR(EIO);
573 case ZPOOL_PROP_CACHEFILE:
574 if ((error = nvpair_value_string(elem, &strval)) != 0)
577 if (strval[0] == '\0')
580 if (strcmp(strval, "none") == 0)
583 if (strval[0] != '/') {
584 error = SET_ERROR(EINVAL);
588 slash = strrchr(strval, '/');
589 ASSERT(slash != NULL);
591 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
592 strcmp(slash, "/..") == 0)
593 error = SET_ERROR(EINVAL);
596 case ZPOOL_PROP_COMMENT:
597 if ((error = nvpair_value_string(elem, &strval)) != 0)
599 for (check = strval; *check != '\0'; check++) {
601 * The kernel doesn't have an easy isprint()
602 * check. For this kernel check, we merely
603 * check ASCII apart from DEL. Fix this if
604 * there is an easy-to-use kernel isprint().
606 if (*check >= 0x7f) {
607 error = SET_ERROR(EINVAL);
612 if (strlen(strval) > ZPROP_MAX_COMMENT)
616 case ZPOOL_PROP_DEDUPDITTO:
617 if (spa_version(spa) < SPA_VERSION_DEDUP)
618 error = SET_ERROR(ENOTSUP);
620 error = nvpair_value_uint64(elem, &intval);
622 intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
623 error = SET_ERROR(EINVAL);
631 if (!error && reset_bootfs) {
632 error = nvlist_remove(props,
633 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
636 error = nvlist_add_uint64(props,
637 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
645 spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
648 spa_config_dirent_t *dp;
650 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
654 dp = kmem_alloc(sizeof (spa_config_dirent_t),
657 if (cachefile[0] == '\0')
658 dp->scd_path = spa_strdup(spa_config_path);
659 else if (strcmp(cachefile, "none") == 0)
662 dp->scd_path = spa_strdup(cachefile);
664 list_insert_head(&spa->spa_config_list, dp);
666 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
670 spa_prop_set(spa_t *spa, nvlist_t *nvp)
673 nvpair_t *elem = NULL;
674 boolean_t need_sync = B_FALSE;
676 if ((error = spa_prop_validate(spa, nvp)) != 0)
679 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
680 zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem));
682 if (prop == ZPOOL_PROP_CACHEFILE ||
683 prop == ZPOOL_PROP_ALTROOT ||
684 prop == ZPOOL_PROP_READONLY)
687 if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) {
690 if (prop == ZPOOL_PROP_VERSION) {
691 VERIFY(nvpair_value_uint64(elem, &ver) == 0);
693 ASSERT(zpool_prop_feature(nvpair_name(elem)));
694 ver = SPA_VERSION_FEATURES;
698 /* Save time if the version is already set. */
699 if (ver == spa_version(spa))
703 * In addition to the pool directory object, we might
704 * create the pool properties object, the features for
705 * read object, the features for write object, or the
706 * feature descriptions object.
708 error = dsl_sync_task(spa->spa_name, NULL,
709 spa_sync_version, &ver,
710 6, ZFS_SPACE_CHECK_RESERVED);
721 return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props,
722 nvp, 6, ZFS_SPACE_CHECK_RESERVED));
729 * If the bootfs property value is dsobj, clear it.
732 spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
734 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
735 VERIFY(zap_remove(spa->spa_meta_objset,
736 spa->spa_pool_props_object,
737 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
744 spa_change_guid_check(void *arg, dmu_tx_t *tx)
746 uint64_t *newguid = arg;
747 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
748 vdev_t *rvd = spa->spa_root_vdev;
751 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
752 vdev_state = rvd->vdev_state;
753 spa_config_exit(spa, SCL_STATE, FTAG);
755 if (vdev_state != VDEV_STATE_HEALTHY)
756 return (SET_ERROR(ENXIO));
758 ASSERT3U(spa_guid(spa), !=, *newguid);
764 spa_change_guid_sync(void *arg, dmu_tx_t *tx)
766 uint64_t *newguid = arg;
767 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
769 vdev_t *rvd = spa->spa_root_vdev;
771 oldguid = spa_guid(spa);
773 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
774 rvd->vdev_guid = *newguid;
775 rvd->vdev_guid_sum += (*newguid - oldguid);
776 vdev_config_dirty(rvd);
777 spa_config_exit(spa, SCL_STATE, FTAG);
779 spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu",
784 * Change the GUID for the pool. This is done so that we can later
785 * re-import a pool built from a clone of our own vdevs. We will modify
786 * the root vdev's guid, our own pool guid, and then mark all of our
787 * vdevs dirty. Note that we must make sure that all our vdevs are
788 * online when we do this, or else any vdevs that weren't present
789 * would be orphaned from our pool. We are also going to issue a
790 * sysevent to update any watchers.
793 spa_change_guid(spa_t *spa)
798 mutex_enter(&spa->spa_vdev_top_lock);
799 mutex_enter(&spa_namespace_lock);
800 guid = spa_generate_guid(NULL);
802 error = dsl_sync_task(spa->spa_name, spa_change_guid_check,
803 spa_change_guid_sync, &guid, 5, ZFS_SPACE_CHECK_RESERVED);
806 spa_config_sync(spa, B_FALSE, B_TRUE);
807 spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID);
810 mutex_exit(&spa_namespace_lock);
811 mutex_exit(&spa->spa_vdev_top_lock);
817 * ==========================================================================
818 * SPA state manipulation (open/create/destroy/import/export)
819 * ==========================================================================
823 spa_error_entry_compare(const void *a, const void *b)
825 spa_error_entry_t *sa = (spa_error_entry_t *)a;
826 spa_error_entry_t *sb = (spa_error_entry_t *)b;
829 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
830 sizeof (zbookmark_phys_t));
841 * Utility function which retrieves copies of the current logs and
842 * re-initializes them in the process.
845 spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
847 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
849 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
850 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
852 avl_create(&spa->spa_errlist_scrub,
853 spa_error_entry_compare, sizeof (spa_error_entry_t),
854 offsetof(spa_error_entry_t, se_avl));
855 avl_create(&spa->spa_errlist_last,
856 spa_error_entry_compare, sizeof (spa_error_entry_t),
857 offsetof(spa_error_entry_t, se_avl));
861 spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
863 const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
864 enum zti_modes mode = ztip->zti_mode;
865 uint_t value = ztip->zti_value;
866 uint_t count = ztip->zti_count;
867 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
870 boolean_t batch = B_FALSE;
872 if (mode == ZTI_MODE_NULL) {
874 tqs->stqs_taskq = NULL;
878 ASSERT3U(count, >, 0);
880 tqs->stqs_count = count;
881 tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP);
885 ASSERT3U(value, >=, 1);
886 value = MAX(value, 1);
891 flags |= TASKQ_THREADS_CPU_PCT;
892 value = zio_taskq_batch_pct;
896 panic("unrecognized mode for %s_%s taskq (%u:%u) in "
898 zio_type_name[t], zio_taskq_types[q], mode, value);
902 for (uint_t i = 0; i < count; i++) {
906 (void) snprintf(name, sizeof (name), "%s_%s_%u",
907 zio_type_name[t], zio_taskq_types[q], i);
909 (void) snprintf(name, sizeof (name), "%s_%s",
910 zio_type_name[t], zio_taskq_types[q]);
914 if (zio_taskq_sysdc && spa->spa_proc != &p0) {
916 flags |= TASKQ_DC_BATCH;
918 tq = taskq_create_sysdc(name, value, 50, INT_MAX,
919 spa->spa_proc, zio_taskq_basedc, flags);
922 pri_t pri = maxclsyspri;
924 * The write issue taskq can be extremely CPU
925 * intensive. Run it at slightly lower priority
926 * than the other taskqs.
928 if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE)
931 tq = taskq_create_proc(name, value, pri, 50,
932 INT_MAX, spa->spa_proc, flags);
937 tqs->stqs_taskq[i] = tq;
942 spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
944 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
946 if (tqs->stqs_taskq == NULL) {
947 ASSERT0(tqs->stqs_count);
951 for (uint_t i = 0; i < tqs->stqs_count; i++) {
952 ASSERT3P(tqs->stqs_taskq[i], !=, NULL);
953 taskq_destroy(tqs->stqs_taskq[i]);
956 kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *));
957 tqs->stqs_taskq = NULL;
961 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
962 * Note that a type may have multiple discrete taskqs to avoid lock contention
963 * on the taskq itself. In that case we choose which taskq at random by using
964 * the low bits of gethrtime().
967 spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q,
968 task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent)
970 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
973 ASSERT3P(tqs->stqs_taskq, !=, NULL);
974 ASSERT3U(tqs->stqs_count, !=, 0);
976 if (tqs->stqs_count == 1) {
977 tq = tqs->stqs_taskq[0];
980 tq = tqs->stqs_taskq[cpu_ticks() % tqs->stqs_count];
982 tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count];
986 taskq_dispatch_ent(tq, func, arg, flags, ent);
990 spa_create_zio_taskqs(spa_t *spa)
992 for (int t = 0; t < ZIO_TYPES; t++) {
993 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
994 spa_taskqs_init(spa, t, q);
1002 spa_thread(void *arg)
1004 callb_cpr_t cprinfo;
1007 user_t *pu = PTOU(curproc);
1009 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
1012 ASSERT(curproc != &p0);
1013 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
1014 "zpool-%s", spa->spa_name);
1015 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
1018 /* bind this thread to the requested psrset */
1019 if (zio_taskq_psrset_bind != PS_NONE) {
1021 mutex_enter(&cpu_lock);
1022 mutex_enter(&pidlock);
1023 mutex_enter(&curproc->p_lock);
1025 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
1026 0, NULL, NULL) == 0) {
1027 curthread->t_bind_pset = zio_taskq_psrset_bind;
1030 "Couldn't bind process for zfs pool \"%s\" to "
1031 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
1034 mutex_exit(&curproc->p_lock);
1035 mutex_exit(&pidlock);
1036 mutex_exit(&cpu_lock);
1042 if (zio_taskq_sysdc) {
1043 sysdc_thread_enter(curthread, 100, 0);
1047 spa->spa_proc = curproc;
1048 spa->spa_did = curthread->t_did;
1050 spa_create_zio_taskqs(spa);
1052 mutex_enter(&spa->spa_proc_lock);
1053 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
1055 spa->spa_proc_state = SPA_PROC_ACTIVE;
1056 cv_broadcast(&spa->spa_proc_cv);
1058 CALLB_CPR_SAFE_BEGIN(&cprinfo);
1059 while (spa->spa_proc_state == SPA_PROC_ACTIVE)
1060 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1061 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
1063 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
1064 spa->spa_proc_state = SPA_PROC_GONE;
1065 spa->spa_proc = &p0;
1066 cv_broadcast(&spa->spa_proc_cv);
1067 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */
1069 mutex_enter(&curproc->p_lock);
1072 #endif /* SPA_PROCESS */
1076 * Activate an uninitialized pool.
1079 spa_activate(spa_t *spa, int mode)
1081 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
1083 spa->spa_state = POOL_STATE_ACTIVE;
1084 spa->spa_mode = mode;
1086 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
1087 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
1089 /* Try to create a covering process */
1090 mutex_enter(&spa->spa_proc_lock);
1091 ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
1092 ASSERT(spa->spa_proc == &p0);
1096 /* Only create a process if we're going to be around a while. */
1097 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
1098 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
1100 spa->spa_proc_state = SPA_PROC_CREATED;
1101 while (spa->spa_proc_state == SPA_PROC_CREATED) {
1102 cv_wait(&spa->spa_proc_cv,
1103 &spa->spa_proc_lock);
1105 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1106 ASSERT(spa->spa_proc != &p0);
1107 ASSERT(spa->spa_did != 0);
1111 "Couldn't create process for zfs pool \"%s\"\n",
1116 #endif /* SPA_PROCESS */
1117 mutex_exit(&spa->spa_proc_lock);
1119 /* If we didn't create a process, we need to create our taskqs. */
1120 ASSERT(spa->spa_proc == &p0);
1121 if (spa->spa_proc == &p0) {
1122 spa_create_zio_taskqs(spa);
1126 * Start TRIM thread.
1128 trim_thread_create(spa);
1130 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
1131 offsetof(vdev_t, vdev_config_dirty_node));
1132 list_create(&spa->spa_evicting_os_list, sizeof (objset_t),
1133 offsetof(objset_t, os_evicting_node));
1134 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
1135 offsetof(vdev_t, vdev_state_dirty_node));
1137 txg_list_create(&spa->spa_vdev_txg_list,
1138 offsetof(struct vdev, vdev_txg_node));
1140 avl_create(&spa->spa_errlist_scrub,
1141 spa_error_entry_compare, sizeof (spa_error_entry_t),
1142 offsetof(spa_error_entry_t, se_avl));
1143 avl_create(&spa->spa_errlist_last,
1144 spa_error_entry_compare, sizeof (spa_error_entry_t),
1145 offsetof(spa_error_entry_t, se_avl));
1149 * Opposite of spa_activate().
1152 spa_deactivate(spa_t *spa)
1154 ASSERT(spa->spa_sync_on == B_FALSE);
1155 ASSERT(spa->spa_dsl_pool == NULL);
1156 ASSERT(spa->spa_root_vdev == NULL);
1157 ASSERT(spa->spa_async_zio_root == NULL);
1158 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
1161 * Stop TRIM thread in case spa_unload() wasn't called directly
1162 * before spa_deactivate().
1164 trim_thread_destroy(spa);
1166 spa_evicting_os_wait(spa);
1168 txg_list_destroy(&spa->spa_vdev_txg_list);
1170 list_destroy(&spa->spa_config_dirty_list);
1171 list_destroy(&spa->spa_evicting_os_list);
1172 list_destroy(&spa->spa_state_dirty_list);
1174 for (int t = 0; t < ZIO_TYPES; t++) {
1175 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1176 spa_taskqs_fini(spa, t, q);
1180 metaslab_class_destroy(spa->spa_normal_class);
1181 spa->spa_normal_class = NULL;
1183 metaslab_class_destroy(spa->spa_log_class);
1184 spa->spa_log_class = NULL;
1187 * If this was part of an import or the open otherwise failed, we may
1188 * still have errors left in the queues. Empty them just in case.
1190 spa_errlog_drain(spa);
1192 avl_destroy(&spa->spa_errlist_scrub);
1193 avl_destroy(&spa->spa_errlist_last);
1195 spa->spa_state = POOL_STATE_UNINITIALIZED;
1197 mutex_enter(&spa->spa_proc_lock);
1198 if (spa->spa_proc_state != SPA_PROC_NONE) {
1199 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1200 spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1201 cv_broadcast(&spa->spa_proc_cv);
1202 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1203 ASSERT(spa->spa_proc != &p0);
1204 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1206 ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1207 spa->spa_proc_state = SPA_PROC_NONE;
1209 ASSERT(spa->spa_proc == &p0);
1210 mutex_exit(&spa->spa_proc_lock);
1214 * We want to make sure spa_thread() has actually exited the ZFS
1215 * module, so that the module can't be unloaded out from underneath
1218 if (spa->spa_did != 0) {
1219 thread_join(spa->spa_did);
1222 #endif /* SPA_PROCESS */
1226 * Verify a pool configuration, and construct the vdev tree appropriately. This
1227 * will create all the necessary vdevs in the appropriate layout, with each vdev
1228 * in the CLOSED state. This will prep the pool before open/creation/import.
1229 * All vdev validation is done by the vdev_alloc() routine.
1232 spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1233 uint_t id, int atype)
1239 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1242 if ((*vdp)->vdev_ops->vdev_op_leaf)
1245 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1248 if (error == ENOENT)
1254 return (SET_ERROR(EINVAL));
1257 for (int c = 0; c < children; c++) {
1259 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1267 ASSERT(*vdp != NULL);
1273 * Opposite of spa_load().
1276 spa_unload(spa_t *spa)
1280 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1285 trim_thread_destroy(spa);
1290 spa_async_suspend(spa);
1295 if (spa->spa_sync_on) {
1296 txg_sync_stop(spa->spa_dsl_pool);
1297 spa->spa_sync_on = B_FALSE;
1301 * Wait for any outstanding async I/O to complete.
1303 if (spa->spa_async_zio_root != NULL) {
1304 for (int i = 0; i < max_ncpus; i++)
1305 (void) zio_wait(spa->spa_async_zio_root[i]);
1306 kmem_free(spa->spa_async_zio_root, max_ncpus * sizeof (void *));
1307 spa->spa_async_zio_root = NULL;
1310 bpobj_close(&spa->spa_deferred_bpobj);
1312 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1317 if (spa->spa_root_vdev)
1318 vdev_free(spa->spa_root_vdev);
1319 ASSERT(spa->spa_root_vdev == NULL);
1322 * Close the dsl pool.
1324 if (spa->spa_dsl_pool) {
1325 dsl_pool_close(spa->spa_dsl_pool);
1326 spa->spa_dsl_pool = NULL;
1327 spa->spa_meta_objset = NULL;
1334 * Drop and purge level 2 cache
1336 spa_l2cache_drop(spa);
1338 for (i = 0; i < spa->spa_spares.sav_count; i++)
1339 vdev_free(spa->spa_spares.sav_vdevs[i]);
1340 if (spa->spa_spares.sav_vdevs) {
1341 kmem_free(spa->spa_spares.sav_vdevs,
1342 spa->spa_spares.sav_count * sizeof (void *));
1343 spa->spa_spares.sav_vdevs = NULL;
1345 if (spa->spa_spares.sav_config) {
1346 nvlist_free(spa->spa_spares.sav_config);
1347 spa->spa_spares.sav_config = NULL;
1349 spa->spa_spares.sav_count = 0;
1351 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1352 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1353 vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1355 if (spa->spa_l2cache.sav_vdevs) {
1356 kmem_free(spa->spa_l2cache.sav_vdevs,
1357 spa->spa_l2cache.sav_count * sizeof (void *));
1358 spa->spa_l2cache.sav_vdevs = NULL;
1360 if (spa->spa_l2cache.sav_config) {
1361 nvlist_free(spa->spa_l2cache.sav_config);
1362 spa->spa_l2cache.sav_config = NULL;
1364 spa->spa_l2cache.sav_count = 0;
1366 spa->spa_async_suspended = 0;
1368 if (spa->spa_comment != NULL) {
1369 spa_strfree(spa->spa_comment);
1370 spa->spa_comment = NULL;
1373 spa_config_exit(spa, SCL_ALL, FTAG);
1377 * Load (or re-load) the current list of vdevs describing the active spares for
1378 * this pool. When this is called, we have some form of basic information in
1379 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1380 * then re-generate a more complete list including status information.
1383 spa_load_spares(spa_t *spa)
1390 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1393 * First, close and free any existing spare vdevs.
1395 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1396 vd = spa->spa_spares.sav_vdevs[i];
1398 /* Undo the call to spa_activate() below */
1399 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1400 B_FALSE)) != NULL && tvd->vdev_isspare)
1401 spa_spare_remove(tvd);
1406 if (spa->spa_spares.sav_vdevs)
1407 kmem_free(spa->spa_spares.sav_vdevs,
1408 spa->spa_spares.sav_count * sizeof (void *));
1410 if (spa->spa_spares.sav_config == NULL)
1413 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1414 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1416 spa->spa_spares.sav_count = (int)nspares;
1417 spa->spa_spares.sav_vdevs = NULL;
1423 * Construct the array of vdevs, opening them to get status in the
1424 * process. For each spare, there is potentially two different vdev_t
1425 * structures associated with it: one in the list of spares (used only
1426 * for basic validation purposes) and one in the active vdev
1427 * configuration (if it's spared in). During this phase we open and
1428 * validate each vdev on the spare list. If the vdev also exists in the
1429 * active configuration, then we also mark this vdev as an active spare.
1431 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1433 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1434 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1435 VDEV_ALLOC_SPARE) == 0);
1438 spa->spa_spares.sav_vdevs[i] = vd;
1440 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1441 B_FALSE)) != NULL) {
1442 if (!tvd->vdev_isspare)
1446 * We only mark the spare active if we were successfully
1447 * able to load the vdev. Otherwise, importing a pool
1448 * with a bad active spare would result in strange
1449 * behavior, because multiple pool would think the spare
1450 * is actively in use.
1452 * There is a vulnerability here to an equally bizarre
1453 * circumstance, where a dead active spare is later
1454 * brought back to life (onlined or otherwise). Given
1455 * the rarity of this scenario, and the extra complexity
1456 * it adds, we ignore the possibility.
1458 if (!vdev_is_dead(tvd))
1459 spa_spare_activate(tvd);
1463 vd->vdev_aux = &spa->spa_spares;
1465 if (vdev_open(vd) != 0)
1468 if (vdev_validate_aux(vd) == 0)
1473 * Recompute the stashed list of spares, with status information
1476 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1477 DATA_TYPE_NVLIST_ARRAY) == 0);
1479 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1481 for (i = 0; i < spa->spa_spares.sav_count; i++)
1482 spares[i] = vdev_config_generate(spa,
1483 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1484 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1485 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1486 for (i = 0; i < spa->spa_spares.sav_count; i++)
1487 nvlist_free(spares[i]);
1488 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1492 * Load (or re-load) the current list of vdevs describing the active l2cache for
1493 * this pool. When this is called, we have some form of basic information in
1494 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1495 * then re-generate a more complete list including status information.
1496 * Devices which are already active have their details maintained, and are
1500 spa_load_l2cache(spa_t *spa)
1504 int i, j, oldnvdevs;
1506 vdev_t *vd, **oldvdevs, **newvdevs;
1507 spa_aux_vdev_t *sav = &spa->spa_l2cache;
1509 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1511 if (sav->sav_config != NULL) {
1512 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1513 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1514 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1520 oldvdevs = sav->sav_vdevs;
1521 oldnvdevs = sav->sav_count;
1522 sav->sav_vdevs = NULL;
1526 * Process new nvlist of vdevs.
1528 for (i = 0; i < nl2cache; i++) {
1529 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1533 for (j = 0; j < oldnvdevs; j++) {
1535 if (vd != NULL && guid == vd->vdev_guid) {
1537 * Retain previous vdev for add/remove ops.
1545 if (newvdevs[i] == NULL) {
1549 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1550 VDEV_ALLOC_L2CACHE) == 0);
1555 * Commit this vdev as an l2cache device,
1556 * even if it fails to open.
1558 spa_l2cache_add(vd);
1563 spa_l2cache_activate(vd);
1565 if (vdev_open(vd) != 0)
1568 (void) vdev_validate_aux(vd);
1570 if (!vdev_is_dead(vd))
1571 l2arc_add_vdev(spa, vd);
1576 * Purge vdevs that were dropped
1578 for (i = 0; i < oldnvdevs; i++) {
1583 ASSERT(vd->vdev_isl2cache);
1585 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1586 pool != 0ULL && l2arc_vdev_present(vd))
1587 l2arc_remove_vdev(vd);
1588 vdev_clear_stats(vd);
1594 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1596 if (sav->sav_config == NULL)
1599 sav->sav_vdevs = newvdevs;
1600 sav->sav_count = (int)nl2cache;
1603 * Recompute the stashed list of l2cache devices, with status
1604 * information this time.
1606 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1607 DATA_TYPE_NVLIST_ARRAY) == 0);
1609 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1610 for (i = 0; i < sav->sav_count; i++)
1611 l2cache[i] = vdev_config_generate(spa,
1612 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1613 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1614 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1616 for (i = 0; i < sav->sav_count; i++)
1617 nvlist_free(l2cache[i]);
1619 kmem_free(l2cache, sav->sav_count * sizeof (void *));
1623 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1626 char *packed = NULL;
1631 error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db);
1635 nvsize = *(uint64_t *)db->db_data;
1636 dmu_buf_rele(db, FTAG);
1638 packed = kmem_alloc(nvsize, KM_SLEEP);
1639 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1642 error = nvlist_unpack(packed, nvsize, value, 0);
1643 kmem_free(packed, nvsize);
1649 * Checks to see if the given vdev could not be opened, in which case we post a
1650 * sysevent to notify the autoreplace code that the device has been removed.
1653 spa_check_removed(vdev_t *vd)
1655 for (int c = 0; c < vd->vdev_children; c++)
1656 spa_check_removed(vd->vdev_child[c]);
1658 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) &&
1660 zfs_post_autoreplace(vd->vdev_spa, vd);
1661 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1666 * Validate the current config against the MOS config
1669 spa_config_valid(spa_t *spa, nvlist_t *config)
1671 vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1674 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1676 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1677 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1679 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1682 * If we're doing a normal import, then build up any additional
1683 * diagnostic information about missing devices in this config.
1684 * We'll pass this up to the user for further processing.
1686 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1687 nvlist_t **child, *nv;
1690 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1692 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1694 for (int c = 0; c < rvd->vdev_children; c++) {
1695 vdev_t *tvd = rvd->vdev_child[c];
1696 vdev_t *mtvd = mrvd->vdev_child[c];
1698 if (tvd->vdev_ops == &vdev_missing_ops &&
1699 mtvd->vdev_ops != &vdev_missing_ops &&
1701 child[idx++] = vdev_config_generate(spa, mtvd,
1706 VERIFY(nvlist_add_nvlist_array(nv,
1707 ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1708 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1709 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1711 for (int i = 0; i < idx; i++)
1712 nvlist_free(child[i]);
1715 kmem_free(child, rvd->vdev_children * sizeof (char **));
1719 * Compare the root vdev tree with the information we have
1720 * from the MOS config (mrvd). Check each top-level vdev
1721 * with the corresponding MOS config top-level (mtvd).
1723 for (int c = 0; c < rvd->vdev_children; c++) {
1724 vdev_t *tvd = rvd->vdev_child[c];
1725 vdev_t *mtvd = mrvd->vdev_child[c];
1728 * Resolve any "missing" vdevs in the current configuration.
1729 * If we find that the MOS config has more accurate information
1730 * about the top-level vdev then use that vdev instead.
1732 if (tvd->vdev_ops == &vdev_missing_ops &&
1733 mtvd->vdev_ops != &vdev_missing_ops) {
1735 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1739 * Device specific actions.
1741 if (mtvd->vdev_islog) {
1742 spa_set_log_state(spa, SPA_LOG_CLEAR);
1745 * XXX - once we have 'readonly' pool
1746 * support we should be able to handle
1747 * missing data devices by transitioning
1748 * the pool to readonly.
1754 * Swap the missing vdev with the data we were
1755 * able to obtain from the MOS config.
1757 vdev_remove_child(rvd, tvd);
1758 vdev_remove_child(mrvd, mtvd);
1760 vdev_add_child(rvd, mtvd);
1761 vdev_add_child(mrvd, tvd);
1763 spa_config_exit(spa, SCL_ALL, FTAG);
1765 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1768 } else if (mtvd->vdev_islog) {
1770 * Load the slog device's state from the MOS config
1771 * since it's possible that the label does not
1772 * contain the most up-to-date information.
1774 vdev_load_log_state(tvd, mtvd);
1779 spa_config_exit(spa, SCL_ALL, FTAG);
1782 * Ensure we were able to validate the config.
1784 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1788 * Check for missing log devices
1791 spa_check_logs(spa_t *spa)
1793 boolean_t rv = B_FALSE;
1794 dsl_pool_t *dp = spa_get_dsl(spa);
1796 switch (spa->spa_log_state) {
1797 case SPA_LOG_MISSING:
1798 /* need to recheck in case slog has been restored */
1799 case SPA_LOG_UNKNOWN:
1800 rv = (dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
1801 zil_check_log_chain, NULL, DS_FIND_CHILDREN) != 0);
1803 spa_set_log_state(spa, SPA_LOG_MISSING);
1810 spa_passivate_log(spa_t *spa)
1812 vdev_t *rvd = spa->spa_root_vdev;
1813 boolean_t slog_found = B_FALSE;
1815 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1817 if (!spa_has_slogs(spa))
1820 for (int c = 0; c < rvd->vdev_children; c++) {
1821 vdev_t *tvd = rvd->vdev_child[c];
1822 metaslab_group_t *mg = tvd->vdev_mg;
1824 if (tvd->vdev_islog) {
1825 metaslab_group_passivate(mg);
1826 slog_found = B_TRUE;
1830 return (slog_found);
1834 spa_activate_log(spa_t *spa)
1836 vdev_t *rvd = spa->spa_root_vdev;
1838 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1840 for (int c = 0; c < rvd->vdev_children; c++) {
1841 vdev_t *tvd = rvd->vdev_child[c];
1842 metaslab_group_t *mg = tvd->vdev_mg;
1844 if (tvd->vdev_islog)
1845 metaslab_group_activate(mg);
1850 spa_offline_log(spa_t *spa)
1854 error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1855 NULL, DS_FIND_CHILDREN);
1858 * We successfully offlined the log device, sync out the
1859 * current txg so that the "stubby" block can be removed
1862 txg_wait_synced(spa->spa_dsl_pool, 0);
1868 spa_aux_check_removed(spa_aux_vdev_t *sav)
1872 for (i = 0; i < sav->sav_count; i++)
1873 spa_check_removed(sav->sav_vdevs[i]);
1877 spa_claim_notify(zio_t *zio)
1879 spa_t *spa = zio->io_spa;
1884 mutex_enter(&spa->spa_props_lock); /* any mutex will do */
1885 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1886 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1887 mutex_exit(&spa->spa_props_lock);
1890 typedef struct spa_load_error {
1891 uint64_t sle_meta_count;
1892 uint64_t sle_data_count;
1896 spa_load_verify_done(zio_t *zio)
1898 blkptr_t *bp = zio->io_bp;
1899 spa_load_error_t *sle = zio->io_private;
1900 dmu_object_type_t type = BP_GET_TYPE(bp);
1901 int error = zio->io_error;
1902 spa_t *spa = zio->io_spa;
1905 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1906 type != DMU_OT_INTENT_LOG)
1907 atomic_inc_64(&sle->sle_meta_count);
1909 atomic_inc_64(&sle->sle_data_count);
1911 zio_data_buf_free(zio->io_data, zio->io_size);
1913 mutex_enter(&spa->spa_scrub_lock);
1914 spa->spa_scrub_inflight--;
1915 cv_broadcast(&spa->spa_scrub_io_cv);
1916 mutex_exit(&spa->spa_scrub_lock);
1920 * Maximum number of concurrent scrub i/os to create while verifying
1921 * a pool while importing it.
1923 int spa_load_verify_maxinflight = 10000;
1924 boolean_t spa_load_verify_metadata = B_TRUE;
1925 boolean_t spa_load_verify_data = B_TRUE;
1927 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_maxinflight, CTLFLAG_RWTUN,
1928 &spa_load_verify_maxinflight, 0,
1929 "Maximum number of concurrent scrub I/Os to create while verifying a "
1930 "pool while importing it");
1932 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_metadata, CTLFLAG_RWTUN,
1933 &spa_load_verify_metadata, 0,
1934 "Check metadata on import?");
1936 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_data, CTLFLAG_RWTUN,
1937 &spa_load_verify_data, 0,
1938 "Check user data on import?");
1942 spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1943 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
1945 if (bp == NULL || BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp))
1948 * Note: normally this routine will not be called if
1949 * spa_load_verify_metadata is not set. However, it may be useful
1950 * to manually set the flag after the traversal has begun.
1952 if (!spa_load_verify_metadata)
1954 if (BP_GET_BUFC_TYPE(bp) == ARC_BUFC_DATA && !spa_load_verify_data)
1958 size_t size = BP_GET_PSIZE(bp);
1959 void *data = zio_data_buf_alloc(size);
1961 mutex_enter(&spa->spa_scrub_lock);
1962 while (spa->spa_scrub_inflight >= spa_load_verify_maxinflight)
1963 cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
1964 spa->spa_scrub_inflight++;
1965 mutex_exit(&spa->spa_scrub_lock);
1967 zio_nowait(zio_read(rio, spa, bp, data, size,
1968 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1969 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1970 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1975 spa_load_verify(spa_t *spa)
1978 spa_load_error_t sle = { 0 };
1979 zpool_rewind_policy_t policy;
1980 boolean_t verify_ok = B_FALSE;
1983 zpool_get_rewind_policy(spa->spa_config, &policy);
1985 if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1988 rio = zio_root(spa, NULL, &sle,
1989 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1991 if (spa_load_verify_metadata) {
1992 error = traverse_pool(spa, spa->spa_verify_min_txg,
1993 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA,
1994 spa_load_verify_cb, rio);
1997 (void) zio_wait(rio);
1999 spa->spa_load_meta_errors = sle.sle_meta_count;
2000 spa->spa_load_data_errors = sle.sle_data_count;
2002 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
2003 sle.sle_data_count <= policy.zrp_maxdata) {
2007 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
2008 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
2010 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
2011 VERIFY(nvlist_add_uint64(spa->spa_load_info,
2012 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
2013 VERIFY(nvlist_add_int64(spa->spa_load_info,
2014 ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
2015 VERIFY(nvlist_add_uint64(spa->spa_load_info,
2016 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
2018 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
2022 if (error != ENXIO && error != EIO)
2023 error = SET_ERROR(EIO);
2027 return (verify_ok ? 0 : EIO);
2031 * Find a value in the pool props object.
2034 spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
2036 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
2037 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
2041 * Find a value in the pool directory object.
2044 spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
2046 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
2047 name, sizeof (uint64_t), 1, val));
2051 spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
2053 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
2058 * Fix up config after a partly-completed split. This is done with the
2059 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
2060 * pool have that entry in their config, but only the splitting one contains
2061 * a list of all the guids of the vdevs that are being split off.
2063 * This function determines what to do with that list: either rejoin
2064 * all the disks to the pool, or complete the splitting process. To attempt
2065 * the rejoin, each disk that is offlined is marked online again, and
2066 * we do a reopen() call. If the vdev label for every disk that was
2067 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
2068 * then we call vdev_split() on each disk, and complete the split.
2070 * Otherwise we leave the config alone, with all the vdevs in place in
2071 * the original pool.
2074 spa_try_repair(spa_t *spa, nvlist_t *config)
2081 boolean_t attempt_reopen;
2083 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
2086 /* check that the config is complete */
2087 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
2088 &glist, &gcount) != 0)
2091 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
2093 /* attempt to online all the vdevs & validate */
2094 attempt_reopen = B_TRUE;
2095 for (i = 0; i < gcount; i++) {
2096 if (glist[i] == 0) /* vdev is hole */
2099 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
2100 if (vd[i] == NULL) {
2102 * Don't bother attempting to reopen the disks;
2103 * just do the split.
2105 attempt_reopen = B_FALSE;
2107 /* attempt to re-online it */
2108 vd[i]->vdev_offline = B_FALSE;
2112 if (attempt_reopen) {
2113 vdev_reopen(spa->spa_root_vdev);
2115 /* check each device to see what state it's in */
2116 for (extracted = 0, i = 0; i < gcount; i++) {
2117 if (vd[i] != NULL &&
2118 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
2125 * If every disk has been moved to the new pool, or if we never
2126 * even attempted to look at them, then we split them off for
2129 if (!attempt_reopen || gcount == extracted) {
2130 for (i = 0; i < gcount; i++)
2133 vdev_reopen(spa->spa_root_vdev);
2136 kmem_free(vd, gcount * sizeof (vdev_t *));
2140 spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
2141 boolean_t mosconfig)
2143 nvlist_t *config = spa->spa_config;
2144 char *ereport = FM_EREPORT_ZFS_POOL;
2150 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
2151 return (SET_ERROR(EINVAL));
2153 ASSERT(spa->spa_comment == NULL);
2154 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
2155 spa->spa_comment = spa_strdup(comment);
2158 * Versioning wasn't explicitly added to the label until later, so if
2159 * it's not present treat it as the initial version.
2161 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
2162 &spa->spa_ubsync.ub_version) != 0)
2163 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
2165 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
2166 &spa->spa_config_txg);
2168 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
2169 spa_guid_exists(pool_guid, 0)) {
2170 error = SET_ERROR(EEXIST);
2172 spa->spa_config_guid = pool_guid;
2174 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
2176 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
2180 nvlist_free(spa->spa_load_info);
2181 spa->spa_load_info = fnvlist_alloc();
2183 gethrestime(&spa->spa_loaded_ts);
2184 error = spa_load_impl(spa, pool_guid, config, state, type,
2185 mosconfig, &ereport);
2189 * Don't count references from objsets that are already closed
2190 * and are making their way through the eviction process.
2192 spa_evicting_os_wait(spa);
2193 spa->spa_minref = refcount_count(&spa->spa_refcount);
2195 if (error != EEXIST) {
2196 spa->spa_loaded_ts.tv_sec = 0;
2197 spa->spa_loaded_ts.tv_nsec = 0;
2199 if (error != EBADF) {
2200 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
2203 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2210 * Load an existing storage pool, using the pool's builtin spa_config as a
2211 * source of configuration information.
2214 spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
2215 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
2219 nvlist_t *nvroot = NULL;
2222 uberblock_t *ub = &spa->spa_uberblock;
2223 uint64_t children, config_cache_txg = spa->spa_config_txg;
2224 int orig_mode = spa->spa_mode;
2227 boolean_t missing_feat_write = B_FALSE;
2230 * If this is an untrusted config, access the pool in read-only mode.
2231 * This prevents things like resilvering recently removed devices.
2234 spa->spa_mode = FREAD;
2236 ASSERT(MUTEX_HELD(&spa_namespace_lock));
2238 spa->spa_load_state = state;
2240 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
2241 return (SET_ERROR(EINVAL));
2243 parse = (type == SPA_IMPORT_EXISTING ?
2244 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2247 * Create "The Godfather" zio to hold all async IOs
2249 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
2251 for (int i = 0; i < max_ncpus; i++) {
2252 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
2253 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
2254 ZIO_FLAG_GODFATHER);
2258 * Parse the configuration into a vdev tree. We explicitly set the
2259 * value that will be returned by spa_version() since parsing the
2260 * configuration requires knowing the version number.
2262 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2263 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
2264 spa_config_exit(spa, SCL_ALL, FTAG);
2269 ASSERT(spa->spa_root_vdev == rvd);
2270 ASSERT3U(spa->spa_min_ashift, >=, SPA_MINBLOCKSHIFT);
2271 ASSERT3U(spa->spa_max_ashift, <=, SPA_MAXBLOCKSHIFT);
2273 if (type != SPA_IMPORT_ASSEMBLE) {
2274 ASSERT(spa_guid(spa) == pool_guid);
2278 * Try to open all vdevs, loading each label in the process.
2280 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2281 error = vdev_open(rvd);
2282 spa_config_exit(spa, SCL_ALL, FTAG);
2287 * We need to validate the vdev labels against the configuration that
2288 * we have in hand, which is dependent on the setting of mosconfig. If
2289 * mosconfig is true then we're validating the vdev labels based on
2290 * that config. Otherwise, we're validating against the cached config
2291 * (zpool.cache) that was read when we loaded the zfs module, and then
2292 * later we will recursively call spa_load() and validate against
2295 * If we're assembling a new pool that's been split off from an
2296 * existing pool, the labels haven't yet been updated so we skip
2297 * validation for now.
2299 if (type != SPA_IMPORT_ASSEMBLE) {
2300 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2301 error = vdev_validate(rvd, mosconfig);
2302 spa_config_exit(spa, SCL_ALL, FTAG);
2307 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2308 return (SET_ERROR(ENXIO));
2312 * Find the best uberblock.
2314 vdev_uberblock_load(rvd, ub, &label);
2317 * If we weren't able to find a single valid uberblock, return failure.
2319 if (ub->ub_txg == 0) {
2321 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2325 * If the pool has an unsupported version we can't open it.
2327 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2329 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2332 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2336 * If we weren't able to find what's necessary for reading the
2337 * MOS in the label, return failure.
2339 if (label == NULL || nvlist_lookup_nvlist(label,
2340 ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) {
2342 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2347 * Update our in-core representation with the definitive values
2350 nvlist_free(spa->spa_label_features);
2351 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2357 * Look through entries in the label nvlist's features_for_read. If
2358 * there is a feature listed there which we don't understand then we
2359 * cannot open a pool.
2361 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2362 nvlist_t *unsup_feat;
2364 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2367 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2369 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2370 if (!zfeature_is_supported(nvpair_name(nvp))) {
2371 VERIFY(nvlist_add_string(unsup_feat,
2372 nvpair_name(nvp), "") == 0);
2376 if (!nvlist_empty(unsup_feat)) {
2377 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2378 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2379 nvlist_free(unsup_feat);
2380 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2384 nvlist_free(unsup_feat);
2388 * If the vdev guid sum doesn't match the uberblock, we have an
2389 * incomplete configuration. We first check to see if the pool
2390 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2391 * If it is, defer the vdev_guid_sum check till later so we
2392 * can handle missing vdevs.
2394 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
2395 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
2396 rvd->vdev_guid_sum != ub->ub_guid_sum)
2397 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2399 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2400 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2401 spa_try_repair(spa, config);
2402 spa_config_exit(spa, SCL_ALL, FTAG);
2403 nvlist_free(spa->spa_config_splitting);
2404 spa->spa_config_splitting = NULL;
2408 * Initialize internal SPA structures.
2410 spa->spa_state = POOL_STATE_ACTIVE;
2411 spa->spa_ubsync = spa->spa_uberblock;
2412 spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2413 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2414 spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2415 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2416 spa->spa_claim_max_txg = spa->spa_first_txg;
2417 spa->spa_prev_software_version = ub->ub_software_version;
2419 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2421 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2422 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2424 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2425 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2427 if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2428 boolean_t missing_feat_read = B_FALSE;
2429 nvlist_t *unsup_feat, *enabled_feat;
2431 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2432 &spa->spa_feat_for_read_obj) != 0) {
2433 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2436 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2437 &spa->spa_feat_for_write_obj) != 0) {
2438 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2441 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2442 &spa->spa_feat_desc_obj) != 0) {
2443 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2446 enabled_feat = fnvlist_alloc();
2447 unsup_feat = fnvlist_alloc();
2449 if (!spa_features_check(spa, B_FALSE,
2450 unsup_feat, enabled_feat))
2451 missing_feat_read = B_TRUE;
2453 if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) {
2454 if (!spa_features_check(spa, B_TRUE,
2455 unsup_feat, enabled_feat)) {
2456 missing_feat_write = B_TRUE;
2460 fnvlist_add_nvlist(spa->spa_load_info,
2461 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
2463 if (!nvlist_empty(unsup_feat)) {
2464 fnvlist_add_nvlist(spa->spa_load_info,
2465 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
2468 fnvlist_free(enabled_feat);
2469 fnvlist_free(unsup_feat);
2471 if (!missing_feat_read) {
2472 fnvlist_add_boolean(spa->spa_load_info,
2473 ZPOOL_CONFIG_CAN_RDONLY);
2477 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2478 * twofold: to determine whether the pool is available for
2479 * import in read-write mode and (if it is not) whether the
2480 * pool is available for import in read-only mode. If the pool
2481 * is available for import in read-write mode, it is displayed
2482 * as available in userland; if it is not available for import
2483 * in read-only mode, it is displayed as unavailable in
2484 * userland. If the pool is available for import in read-only
2485 * mode but not read-write mode, it is displayed as unavailable
2486 * in userland with a special note that the pool is actually
2487 * available for open in read-only mode.
2489 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2490 * missing a feature for write, we must first determine whether
2491 * the pool can be opened read-only before returning to
2492 * userland in order to know whether to display the
2493 * abovementioned note.
2495 if (missing_feat_read || (missing_feat_write &&
2496 spa_writeable(spa))) {
2497 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2502 * Load refcounts for ZFS features from disk into an in-memory
2503 * cache during SPA initialization.
2505 for (spa_feature_t i = 0; i < SPA_FEATURES; i++) {
2508 error = feature_get_refcount_from_disk(spa,
2509 &spa_feature_table[i], &refcount);
2511 spa->spa_feat_refcount_cache[i] = refcount;
2512 } else if (error == ENOTSUP) {
2513 spa->spa_feat_refcount_cache[i] =
2514 SPA_FEATURE_DISABLED;
2516 return (spa_vdev_err(rvd,
2517 VDEV_AUX_CORRUPT_DATA, EIO));
2522 if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) {
2523 if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG,
2524 &spa->spa_feat_enabled_txg_obj) != 0)
2525 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2528 spa->spa_is_initializing = B_TRUE;
2529 error = dsl_pool_open(spa->spa_dsl_pool);
2530 spa->spa_is_initializing = B_FALSE;
2532 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2536 nvlist_t *policy = NULL, *nvconfig;
2538 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2539 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2541 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2542 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2544 unsigned long myhostid = 0;
2546 VERIFY(nvlist_lookup_string(nvconfig,
2547 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2550 myhostid = zone_get_hostid(NULL);
2553 * We're emulating the system's hostid in userland, so
2554 * we can't use zone_get_hostid().
2556 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2557 #endif /* _KERNEL */
2558 if (check_hostid && hostid != 0 && myhostid != 0 &&
2559 hostid != myhostid) {
2560 nvlist_free(nvconfig);
2561 cmn_err(CE_WARN, "pool '%s' could not be "
2562 "loaded as it was last accessed by "
2563 "another system (host: %s hostid: 0x%lx). "
2564 "See: http://illumos.org/msg/ZFS-8000-EY",
2565 spa_name(spa), hostname,
2566 (unsigned long)hostid);
2567 return (SET_ERROR(EBADF));
2570 if (nvlist_lookup_nvlist(spa->spa_config,
2571 ZPOOL_REWIND_POLICY, &policy) == 0)
2572 VERIFY(nvlist_add_nvlist(nvconfig,
2573 ZPOOL_REWIND_POLICY, policy) == 0);
2575 spa_config_set(spa, nvconfig);
2577 spa_deactivate(spa);
2578 spa_activate(spa, orig_mode);
2580 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2583 /* Grab the secret checksum salt from the MOS. */
2584 error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
2585 DMU_POOL_CHECKSUM_SALT, 1,
2586 sizeof (spa->spa_cksum_salt.zcs_bytes),
2587 spa->spa_cksum_salt.zcs_bytes);
2588 if (error == ENOENT) {
2589 /* Generate a new salt for subsequent use */
2590 (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes,
2591 sizeof (spa->spa_cksum_salt.zcs_bytes));
2592 } else if (error != 0) {
2593 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2596 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2597 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2598 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2600 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2603 * Load the bit that tells us to use the new accounting function
2604 * (raid-z deflation). If we have an older pool, this will not
2607 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2608 if (error != 0 && error != ENOENT)
2609 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2611 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2612 &spa->spa_creation_version);
2613 if (error != 0 && error != ENOENT)
2614 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2617 * Load the persistent error log. If we have an older pool, this will
2620 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2621 if (error != 0 && error != ENOENT)
2622 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2624 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2625 &spa->spa_errlog_scrub);
2626 if (error != 0 && error != ENOENT)
2627 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2630 * Load the history object. If we have an older pool, this
2631 * will not be present.
2633 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2634 if (error != 0 && error != ENOENT)
2635 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2638 * If we're assembling the pool from the split-off vdevs of
2639 * an existing pool, we don't want to attach the spares & cache
2644 * Load any hot spares for this pool.
2646 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2647 if (error != 0 && error != ENOENT)
2648 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2649 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2650 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2651 if (load_nvlist(spa, spa->spa_spares.sav_object,
2652 &spa->spa_spares.sav_config) != 0)
2653 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2655 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2656 spa_load_spares(spa);
2657 spa_config_exit(spa, SCL_ALL, FTAG);
2658 } else if (error == 0) {
2659 spa->spa_spares.sav_sync = B_TRUE;
2663 * Load any level 2 ARC devices for this pool.
2665 error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2666 &spa->spa_l2cache.sav_object);
2667 if (error != 0 && error != ENOENT)
2668 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2669 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2670 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2671 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2672 &spa->spa_l2cache.sav_config) != 0)
2673 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2675 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2676 spa_load_l2cache(spa);
2677 spa_config_exit(spa, SCL_ALL, FTAG);
2678 } else if (error == 0) {
2679 spa->spa_l2cache.sav_sync = B_TRUE;
2682 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2684 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2685 if (error && error != ENOENT)
2686 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2689 uint64_t autoreplace;
2691 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2692 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2693 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2694 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2695 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2696 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2697 &spa->spa_dedup_ditto);
2699 spa->spa_autoreplace = (autoreplace != 0);
2703 * If the 'autoreplace' property is set, then post a resource notifying
2704 * the ZFS DE that it should not issue any faults for unopenable
2705 * devices. We also iterate over the vdevs, and post a sysevent for any
2706 * unopenable vdevs so that the normal autoreplace handler can take
2709 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2710 spa_check_removed(spa->spa_root_vdev);
2712 * For the import case, this is done in spa_import(), because
2713 * at this point we're using the spare definitions from
2714 * the MOS config, not necessarily from the userland config.
2716 if (state != SPA_LOAD_IMPORT) {
2717 spa_aux_check_removed(&spa->spa_spares);
2718 spa_aux_check_removed(&spa->spa_l2cache);
2723 * Load the vdev state for all toplevel vdevs.
2728 * Propagate the leaf DTLs we just loaded all the way up the tree.
2730 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2731 vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2732 spa_config_exit(spa, SCL_ALL, FTAG);
2735 * Load the DDTs (dedup tables).
2737 error = ddt_load(spa);
2739 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2741 spa_update_dspace(spa);
2744 * Validate the config, using the MOS config to fill in any
2745 * information which might be missing. If we fail to validate
2746 * the config then declare the pool unfit for use. If we're
2747 * assembling a pool from a split, the log is not transferred
2750 if (type != SPA_IMPORT_ASSEMBLE) {
2753 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2754 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2756 if (!spa_config_valid(spa, nvconfig)) {
2757 nvlist_free(nvconfig);
2758 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2761 nvlist_free(nvconfig);
2764 * Now that we've validated the config, check the state of the
2765 * root vdev. If it can't be opened, it indicates one or
2766 * more toplevel vdevs are faulted.
2768 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2769 return (SET_ERROR(ENXIO));
2771 if (spa_writeable(spa) && spa_check_logs(spa)) {
2772 *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2773 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2777 if (missing_feat_write) {
2778 ASSERT(state == SPA_LOAD_TRYIMPORT);
2781 * At this point, we know that we can open the pool in
2782 * read-only mode but not read-write mode. We now have enough
2783 * information and can return to userland.
2785 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
2789 * We've successfully opened the pool, verify that we're ready
2790 * to start pushing transactions.
2792 if (state != SPA_LOAD_TRYIMPORT) {
2793 if (error = spa_load_verify(spa))
2794 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2798 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2799 spa->spa_load_max_txg == UINT64_MAX)) {
2801 int need_update = B_FALSE;
2802 dsl_pool_t *dp = spa_get_dsl(spa);
2804 ASSERT(state != SPA_LOAD_TRYIMPORT);
2807 * Claim log blocks that haven't been committed yet.
2808 * This must all happen in a single txg.
2809 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2810 * invoked from zil_claim_log_block()'s i/o done callback.
2811 * Price of rollback is that we abandon the log.
2813 spa->spa_claiming = B_TRUE;
2815 tx = dmu_tx_create_assigned(dp, spa_first_txg(spa));
2816 (void) dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
2817 zil_claim, tx, DS_FIND_CHILDREN);
2820 spa->spa_claiming = B_FALSE;
2822 spa_set_log_state(spa, SPA_LOG_GOOD);
2823 spa->spa_sync_on = B_TRUE;
2824 txg_sync_start(spa->spa_dsl_pool);
2827 * Wait for all claims to sync. We sync up to the highest
2828 * claimed log block birth time so that claimed log blocks
2829 * don't appear to be from the future. spa_claim_max_txg
2830 * will have been set for us by either zil_check_log_chain()
2831 * (invoked from spa_check_logs()) or zil_claim() above.
2833 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2836 * If the config cache is stale, or we have uninitialized
2837 * metaslabs (see spa_vdev_add()), then update the config.
2839 * If this is a verbatim import, trust the current
2840 * in-core spa_config and update the disk labels.
2842 if (config_cache_txg != spa->spa_config_txg ||
2843 state == SPA_LOAD_IMPORT ||
2844 state == SPA_LOAD_RECOVER ||
2845 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2846 need_update = B_TRUE;
2848 for (int c = 0; c < rvd->vdev_children; c++)
2849 if (rvd->vdev_child[c]->vdev_ms_array == 0)
2850 need_update = B_TRUE;
2853 * Update the config cache asychronously in case we're the
2854 * root pool, in which case the config cache isn't writable yet.
2857 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2860 * Check all DTLs to see if anything needs resilvering.
2862 if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2863 vdev_resilver_needed(rvd, NULL, NULL))
2864 spa_async_request(spa, SPA_ASYNC_RESILVER);
2867 * Log the fact that we booted up (so that we can detect if
2868 * we rebooted in the middle of an operation).
2870 spa_history_log_version(spa, "open");
2873 * Delete any inconsistent datasets.
2875 (void) dmu_objset_find(spa_name(spa),
2876 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2879 * Clean up any stale temporary dataset userrefs.
2881 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2888 spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2890 int mode = spa->spa_mode;
2893 spa_deactivate(spa);
2895 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg - 1;
2897 spa_activate(spa, mode);
2898 spa_async_suspend(spa);
2900 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2904 * If spa_load() fails this function will try loading prior txg's. If
2905 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2906 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2907 * function will not rewind the pool and will return the same error as
2911 spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2912 uint64_t max_request, int rewind_flags)
2914 nvlist_t *loadinfo = NULL;
2915 nvlist_t *config = NULL;
2916 int load_error, rewind_error;
2917 uint64_t safe_rewind_txg;
2920 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2921 spa->spa_load_max_txg = spa->spa_load_txg;
2922 spa_set_log_state(spa, SPA_LOG_CLEAR);
2924 spa->spa_load_max_txg = max_request;
2925 if (max_request != UINT64_MAX)
2926 spa->spa_extreme_rewind = B_TRUE;
2929 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2931 if (load_error == 0)
2934 if (spa->spa_root_vdev != NULL)
2935 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2937 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2938 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2940 if (rewind_flags & ZPOOL_NEVER_REWIND) {
2941 nvlist_free(config);
2942 return (load_error);
2945 if (state == SPA_LOAD_RECOVER) {
2946 /* Price of rolling back is discarding txgs, including log */
2947 spa_set_log_state(spa, SPA_LOG_CLEAR);
2950 * If we aren't rolling back save the load info from our first
2951 * import attempt so that we can restore it after attempting
2954 loadinfo = spa->spa_load_info;
2955 spa->spa_load_info = fnvlist_alloc();
2958 spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2959 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2960 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2961 TXG_INITIAL : safe_rewind_txg;
2964 * Continue as long as we're finding errors, we're still within
2965 * the acceptable rewind range, and we're still finding uberblocks
2967 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2968 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2969 if (spa->spa_load_max_txg < safe_rewind_txg)
2970 spa->spa_extreme_rewind = B_TRUE;
2971 rewind_error = spa_load_retry(spa, state, mosconfig);
2974 spa->spa_extreme_rewind = B_FALSE;
2975 spa->spa_load_max_txg = UINT64_MAX;
2977 if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2978 spa_config_set(spa, config);
2980 if (state == SPA_LOAD_RECOVER) {
2981 ASSERT3P(loadinfo, ==, NULL);
2982 return (rewind_error);
2984 /* Store the rewind info as part of the initial load info */
2985 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
2986 spa->spa_load_info);
2988 /* Restore the initial load info */
2989 fnvlist_free(spa->spa_load_info);
2990 spa->spa_load_info = loadinfo;
2992 return (load_error);
2999 * The import case is identical to an open except that the configuration is sent
3000 * down from userland, instead of grabbed from the configuration cache. For the
3001 * case of an open, the pool configuration will exist in the
3002 * POOL_STATE_UNINITIALIZED state.
3004 * The stats information (gen/count/ustats) is used to gather vdev statistics at
3005 * the same time open the pool, without having to keep around the spa_t in some
3009 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
3013 spa_load_state_t state = SPA_LOAD_OPEN;
3015 int locked = B_FALSE;
3016 int firstopen = B_FALSE;
3021 * As disgusting as this is, we need to support recursive calls to this
3022 * function because dsl_dir_open() is called during spa_load(), and ends
3023 * up calling spa_open() again. The real fix is to figure out how to
3024 * avoid dsl_dir_open() calling this in the first place.
3026 if (mutex_owner(&spa_namespace_lock) != curthread) {
3027 mutex_enter(&spa_namespace_lock);
3031 if ((spa = spa_lookup(pool)) == NULL) {
3033 mutex_exit(&spa_namespace_lock);
3034 return (SET_ERROR(ENOENT));
3037 if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
3038 zpool_rewind_policy_t policy;
3042 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
3044 if (policy.zrp_request & ZPOOL_DO_REWIND)
3045 state = SPA_LOAD_RECOVER;
3047 spa_activate(spa, spa_mode_global);
3049 if (state != SPA_LOAD_RECOVER)
3050 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3052 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
3053 policy.zrp_request);
3055 if (error == EBADF) {
3057 * If vdev_validate() returns failure (indicated by
3058 * EBADF), it indicates that one of the vdevs indicates
3059 * that the pool has been exported or destroyed. If
3060 * this is the case, the config cache is out of sync and
3061 * we should remove the pool from the namespace.
3064 spa_deactivate(spa);
3065 spa_config_sync(spa, B_TRUE, B_TRUE);
3068 mutex_exit(&spa_namespace_lock);
3069 return (SET_ERROR(ENOENT));
3074 * We can't open the pool, but we still have useful
3075 * information: the state of each vdev after the
3076 * attempted vdev_open(). Return this to the user.
3078 if (config != NULL && spa->spa_config) {
3079 VERIFY(nvlist_dup(spa->spa_config, config,
3081 VERIFY(nvlist_add_nvlist(*config,
3082 ZPOOL_CONFIG_LOAD_INFO,
3083 spa->spa_load_info) == 0);
3086 spa_deactivate(spa);
3087 spa->spa_last_open_failed = error;
3089 mutex_exit(&spa_namespace_lock);
3095 spa_open_ref(spa, tag);
3098 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
3101 * If we've recovered the pool, pass back any information we
3102 * gathered while doing the load.
3104 if (state == SPA_LOAD_RECOVER) {
3105 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
3106 spa->spa_load_info) == 0);
3110 spa->spa_last_open_failed = 0;
3111 spa->spa_last_ubsync_txg = 0;
3112 spa->spa_load_txg = 0;
3113 mutex_exit(&spa_namespace_lock);
3117 zvol_create_minors(spa->spa_name);
3128 spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
3131 return (spa_open_common(name, spapp, tag, policy, config));
3135 spa_open(const char *name, spa_t **spapp, void *tag)
3137 return (spa_open_common(name, spapp, tag, NULL, NULL));
3141 * Lookup the given spa_t, incrementing the inject count in the process,
3142 * preventing it from being exported or destroyed.
3145 spa_inject_addref(char *name)
3149 mutex_enter(&spa_namespace_lock);
3150 if ((spa = spa_lookup(name)) == NULL) {
3151 mutex_exit(&spa_namespace_lock);
3154 spa->spa_inject_ref++;
3155 mutex_exit(&spa_namespace_lock);
3161 spa_inject_delref(spa_t *spa)
3163 mutex_enter(&spa_namespace_lock);
3164 spa->spa_inject_ref--;
3165 mutex_exit(&spa_namespace_lock);
3169 * Add spares device information to the nvlist.
3172 spa_add_spares(spa_t *spa, nvlist_t *config)
3182 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3184 if (spa->spa_spares.sav_count == 0)
3187 VERIFY(nvlist_lookup_nvlist(config,
3188 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3189 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
3190 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3192 VERIFY(nvlist_add_nvlist_array(nvroot,
3193 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3194 VERIFY(nvlist_lookup_nvlist_array(nvroot,
3195 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3198 * Go through and find any spares which have since been
3199 * repurposed as an active spare. If this is the case, update
3200 * their status appropriately.
3202 for (i = 0; i < nspares; i++) {
3203 VERIFY(nvlist_lookup_uint64(spares[i],
3204 ZPOOL_CONFIG_GUID, &guid) == 0);
3205 if (spa_spare_exists(guid, &pool, NULL) &&
3207 VERIFY(nvlist_lookup_uint64_array(
3208 spares[i], ZPOOL_CONFIG_VDEV_STATS,
3209 (uint64_t **)&vs, &vsc) == 0);
3210 vs->vs_state = VDEV_STATE_CANT_OPEN;
3211 vs->vs_aux = VDEV_AUX_SPARED;
3218 * Add l2cache device information to the nvlist, including vdev stats.
3221 spa_add_l2cache(spa_t *spa, nvlist_t *config)
3224 uint_t i, j, nl2cache;
3231 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3233 if (spa->spa_l2cache.sav_count == 0)
3236 VERIFY(nvlist_lookup_nvlist(config,
3237 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3238 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
3239 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3240 if (nl2cache != 0) {
3241 VERIFY(nvlist_add_nvlist_array(nvroot,
3242 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3243 VERIFY(nvlist_lookup_nvlist_array(nvroot,
3244 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3247 * Update level 2 cache device stats.
3250 for (i = 0; i < nl2cache; i++) {
3251 VERIFY(nvlist_lookup_uint64(l2cache[i],
3252 ZPOOL_CONFIG_GUID, &guid) == 0);
3255 for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
3257 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
3258 vd = spa->spa_l2cache.sav_vdevs[j];
3264 VERIFY(nvlist_lookup_uint64_array(l2cache[i],
3265 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
3267 vdev_get_stats(vd, vs);
3273 spa_add_feature_stats(spa_t *spa, nvlist_t *config)
3279 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3280 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3282 /* We may be unable to read features if pool is suspended. */
3283 if (spa_suspended(spa))
3286 if (spa->spa_feat_for_read_obj != 0) {
3287 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3288 spa->spa_feat_for_read_obj);
3289 zap_cursor_retrieve(&zc, &za) == 0;
3290 zap_cursor_advance(&zc)) {
3291 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3292 za.za_num_integers == 1);
3293 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3294 za.za_first_integer));
3296 zap_cursor_fini(&zc);
3299 if (spa->spa_feat_for_write_obj != 0) {
3300 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3301 spa->spa_feat_for_write_obj);
3302 zap_cursor_retrieve(&zc, &za) == 0;
3303 zap_cursor_advance(&zc)) {
3304 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3305 za.za_num_integers == 1);
3306 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3307 za.za_first_integer));
3309 zap_cursor_fini(&zc);
3313 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
3315 nvlist_free(features);
3319 spa_get_stats(const char *name, nvlist_t **config,
3320 char *altroot, size_t buflen)
3326 error = spa_open_common(name, &spa, FTAG, NULL, config);
3330 * This still leaves a window of inconsistency where the spares
3331 * or l2cache devices could change and the config would be
3332 * self-inconsistent.
3334 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3336 if (*config != NULL) {
3337 uint64_t loadtimes[2];
3339 loadtimes[0] = spa->spa_loaded_ts.tv_sec;
3340 loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
3341 VERIFY(nvlist_add_uint64_array(*config,
3342 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
3344 VERIFY(nvlist_add_uint64(*config,
3345 ZPOOL_CONFIG_ERRCOUNT,
3346 spa_get_errlog_size(spa)) == 0);
3348 if (spa_suspended(spa))
3349 VERIFY(nvlist_add_uint64(*config,
3350 ZPOOL_CONFIG_SUSPENDED,
3351 spa->spa_failmode) == 0);
3353 spa_add_spares(spa, *config);
3354 spa_add_l2cache(spa, *config);
3355 spa_add_feature_stats(spa, *config);
3360 * We want to get the alternate root even for faulted pools, so we cheat
3361 * and call spa_lookup() directly.
3365 mutex_enter(&spa_namespace_lock);
3366 spa = spa_lookup(name);
3368 spa_altroot(spa, altroot, buflen);
3372 mutex_exit(&spa_namespace_lock);
3374 spa_altroot(spa, altroot, buflen);
3379 spa_config_exit(spa, SCL_CONFIG, FTAG);
3380 spa_close(spa, FTAG);
3387 * Validate that the auxiliary device array is well formed. We must have an
3388 * array of nvlists, each which describes a valid leaf vdev. If this is an
3389 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3390 * specified, as long as they are well-formed.
3393 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
3394 spa_aux_vdev_t *sav, const char *config, uint64_t version,
3395 vdev_labeltype_t label)
3402 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3405 * It's acceptable to have no devs specified.
3407 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
3411 return (SET_ERROR(EINVAL));
3414 * Make sure the pool is formatted with a version that supports this
3417 if (spa_version(spa) < version)
3418 return (SET_ERROR(ENOTSUP));
3421 * Set the pending device list so we correctly handle device in-use
3424 sav->sav_pending = dev;
3425 sav->sav_npending = ndev;
3427 for (i = 0; i < ndev; i++) {
3428 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
3432 if (!vd->vdev_ops->vdev_op_leaf) {
3434 error = SET_ERROR(EINVAL);
3439 * The L2ARC currently only supports disk devices in
3440 * kernel context. For user-level testing, we allow it.
3443 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
3444 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
3445 error = SET_ERROR(ENOTBLK);
3452 if ((error = vdev_open(vd)) == 0 &&
3453 (error = vdev_label_init(vd, crtxg, label)) == 0) {
3454 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
3455 vd->vdev_guid) == 0);
3461 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
3468 sav->sav_pending = NULL;
3469 sav->sav_npending = 0;
3474 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
3478 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3480 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3481 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
3482 VDEV_LABEL_SPARE)) != 0) {
3486 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3487 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
3488 VDEV_LABEL_L2CACHE));
3492 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
3497 if (sav->sav_config != NULL) {
3503 * Generate new dev list by concatentating with the
3506 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
3507 &olddevs, &oldndevs) == 0);
3509 newdevs = kmem_alloc(sizeof (void *) *
3510 (ndevs + oldndevs), KM_SLEEP);
3511 for (i = 0; i < oldndevs; i++)
3512 VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
3514 for (i = 0; i < ndevs; i++)
3515 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
3518 VERIFY(nvlist_remove(sav->sav_config, config,
3519 DATA_TYPE_NVLIST_ARRAY) == 0);
3521 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3522 config, newdevs, ndevs + oldndevs) == 0);
3523 for (i = 0; i < oldndevs + ndevs; i++)
3524 nvlist_free(newdevs[i]);
3525 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
3528 * Generate a new dev list.
3530 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
3532 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
3538 * Stop and drop level 2 ARC devices
3541 spa_l2cache_drop(spa_t *spa)
3545 spa_aux_vdev_t *sav = &spa->spa_l2cache;
3547 for (i = 0; i < sav->sav_count; i++) {
3550 vd = sav->sav_vdevs[i];
3553 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
3554 pool != 0ULL && l2arc_vdev_present(vd))
3555 l2arc_remove_vdev(vd);
3563 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
3567 char *altroot = NULL;
3572 uint64_t txg = TXG_INITIAL;
3573 nvlist_t **spares, **l2cache;
3574 uint_t nspares, nl2cache;
3575 uint64_t version, obj;
3576 boolean_t has_features;
3579 * If this pool already exists, return failure.
3581 mutex_enter(&spa_namespace_lock);
3582 if (spa_lookup(pool) != NULL) {
3583 mutex_exit(&spa_namespace_lock);
3584 return (SET_ERROR(EEXIST));
3588 * Allocate a new spa_t structure.
3590 (void) nvlist_lookup_string(props,
3591 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3592 spa = spa_add(pool, NULL, altroot);
3593 spa_activate(spa, spa_mode_global);
3595 if (props && (error = spa_prop_validate(spa, props))) {
3596 spa_deactivate(spa);
3598 mutex_exit(&spa_namespace_lock);
3602 has_features = B_FALSE;
3603 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
3604 elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
3605 if (zpool_prop_feature(nvpair_name(elem)))
3606 has_features = B_TRUE;
3609 if (has_features || nvlist_lookup_uint64(props,
3610 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
3611 version = SPA_VERSION;
3613 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
3615 spa->spa_first_txg = txg;
3616 spa->spa_uberblock.ub_txg = txg - 1;
3617 spa->spa_uberblock.ub_version = version;
3618 spa->spa_ubsync = spa->spa_uberblock;
3621 * Create "The Godfather" zio to hold all async IOs
3623 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
3625 for (int i = 0; i < max_ncpus; i++) {
3626 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
3627 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
3628 ZIO_FLAG_GODFATHER);
3632 * Create the root vdev.
3634 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3636 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3638 ASSERT(error != 0 || rvd != NULL);
3639 ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3641 if (error == 0 && !zfs_allocatable_devs(nvroot))
3642 error = SET_ERROR(EINVAL);
3645 (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3646 (error = spa_validate_aux(spa, nvroot, txg,
3647 VDEV_ALLOC_ADD)) == 0) {
3648 for (int c = 0; c < rvd->vdev_children; c++) {
3649 vdev_ashift_optimize(rvd->vdev_child[c]);
3650 vdev_metaslab_set_size(rvd->vdev_child[c]);
3651 vdev_expand(rvd->vdev_child[c], txg);
3655 spa_config_exit(spa, SCL_ALL, FTAG);
3659 spa_deactivate(spa);
3661 mutex_exit(&spa_namespace_lock);
3666 * Get the list of spares, if specified.
3668 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3669 &spares, &nspares) == 0) {
3670 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3672 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3673 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3674 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3675 spa_load_spares(spa);
3676 spa_config_exit(spa, SCL_ALL, FTAG);
3677 spa->spa_spares.sav_sync = B_TRUE;
3681 * Get the list of level 2 cache devices, if specified.
3683 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3684 &l2cache, &nl2cache) == 0) {
3685 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3686 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3687 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3688 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3689 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3690 spa_load_l2cache(spa);
3691 spa_config_exit(spa, SCL_ALL, FTAG);
3692 spa->spa_l2cache.sav_sync = B_TRUE;
3695 spa->spa_is_initializing = B_TRUE;
3696 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3697 spa->spa_meta_objset = dp->dp_meta_objset;
3698 spa->spa_is_initializing = B_FALSE;
3701 * Create DDTs (dedup tables).
3705 spa_update_dspace(spa);
3707 tx = dmu_tx_create_assigned(dp, txg);
3710 * Create the pool config object.
3712 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3713 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3714 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3716 if (zap_add(spa->spa_meta_objset,
3717 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3718 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3719 cmn_err(CE_PANIC, "failed to add pool config");
3722 if (spa_version(spa) >= SPA_VERSION_FEATURES)
3723 spa_feature_create_zap_objects(spa, tx);
3725 if (zap_add(spa->spa_meta_objset,
3726 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3727 sizeof (uint64_t), 1, &version, tx) != 0) {
3728 cmn_err(CE_PANIC, "failed to add pool version");
3731 /* Newly created pools with the right version are always deflated. */
3732 if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3733 spa->spa_deflate = TRUE;
3734 if (zap_add(spa->spa_meta_objset,
3735 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3736 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3737 cmn_err(CE_PANIC, "failed to add deflate");
3742 * Create the deferred-free bpobj. Turn off compression
3743 * because sync-to-convergence takes longer if the blocksize
3746 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3747 dmu_object_set_compress(spa->spa_meta_objset, obj,
3748 ZIO_COMPRESS_OFF, tx);
3749 if (zap_add(spa->spa_meta_objset,
3750 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3751 sizeof (uint64_t), 1, &obj, tx) != 0) {
3752 cmn_err(CE_PANIC, "failed to add bpobj");
3754 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3755 spa->spa_meta_objset, obj));
3758 * Create the pool's history object.
3760 if (version >= SPA_VERSION_ZPOOL_HISTORY)
3761 spa_history_create_obj(spa, tx);
3764 * Generate some random noise for salted checksums to operate on.
3766 (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes,
3767 sizeof (spa->spa_cksum_salt.zcs_bytes));
3770 * Set pool properties.
3772 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3773 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3774 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3775 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3777 if (props != NULL) {
3778 spa_configfile_set(spa, props, B_FALSE);
3779 spa_sync_props(props, tx);
3784 spa->spa_sync_on = B_TRUE;
3785 txg_sync_start(spa->spa_dsl_pool);
3788 * We explicitly wait for the first transaction to complete so that our
3789 * bean counters are appropriately updated.
3791 txg_wait_synced(spa->spa_dsl_pool, txg);
3793 spa_config_sync(spa, B_FALSE, B_TRUE);
3794 spa_event_notify(spa, NULL, ESC_ZFS_POOL_CREATE);
3796 spa_history_log_version(spa, "create");
3799 * Don't count references from objsets that are already closed
3800 * and are making their way through the eviction process.
3802 spa_evicting_os_wait(spa);
3803 spa->spa_minref = refcount_count(&spa->spa_refcount);
3805 mutex_exit(&spa_namespace_lock);
3813 * Get the root pool information from the root disk, then import the root pool
3814 * during the system boot up time.
3816 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3819 spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3822 nvlist_t *nvtop, *nvroot;
3825 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3829 * Add this top-level vdev to the child array.
3831 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3833 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3835 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3838 * Put this pool's top-level vdevs into a root vdev.
3840 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3841 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3842 VDEV_TYPE_ROOT) == 0);
3843 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3844 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3845 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3849 * Replace the existing vdev_tree with the new root vdev in
3850 * this pool's configuration (remove the old, add the new).
3852 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3853 nvlist_free(nvroot);
3858 * Walk the vdev tree and see if we can find a device with "better"
3859 * configuration. A configuration is "better" if the label on that
3860 * device has a more recent txg.
3863 spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3865 for (int c = 0; c < vd->vdev_children; c++)
3866 spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3868 if (vd->vdev_ops->vdev_op_leaf) {
3872 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3876 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3880 * Do we have a better boot device?
3882 if (label_txg > *txg) {
3891 * Import a root pool.
3893 * For x86. devpath_list will consist of devid and/or physpath name of
3894 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3895 * The GRUB "findroot" command will return the vdev we should boot.
3897 * For Sparc, devpath_list consists the physpath name of the booting device
3898 * no matter the rootpool is a single device pool or a mirrored pool.
3900 * "/pci@1f,0/ide@d/disk@0,0:a"
3903 spa_import_rootpool(char *devpath, char *devid)
3906 vdev_t *rvd, *bvd, *avd = NULL;
3907 nvlist_t *config, *nvtop;
3913 * Read the label from the boot device and generate a configuration.
3915 config = spa_generate_rootconf(devpath, devid, &guid);
3916 #if defined(_OBP) && defined(_KERNEL)
3917 if (config == NULL) {
3918 if (strstr(devpath, "/iscsi/ssd") != NULL) {
3920 get_iscsi_bootpath_phy(devpath);
3921 config = spa_generate_rootconf(devpath, devid, &guid);
3925 if (config == NULL) {
3926 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
3928 return (SET_ERROR(EIO));
3931 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3933 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3935 mutex_enter(&spa_namespace_lock);
3936 if ((spa = spa_lookup(pname)) != NULL) {
3938 * Remove the existing root pool from the namespace so that we
3939 * can replace it with the correct config we just read in.
3944 spa = spa_add(pname, config, NULL);
3945 spa->spa_is_root = B_TRUE;
3946 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3949 * Build up a vdev tree based on the boot device's label config.
3951 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3953 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3954 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3955 VDEV_ALLOC_ROOTPOOL);
3956 spa_config_exit(spa, SCL_ALL, FTAG);
3958 mutex_exit(&spa_namespace_lock);
3959 nvlist_free(config);
3960 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3966 * Get the boot vdev.
3968 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3969 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3970 (u_longlong_t)guid);
3971 error = SET_ERROR(ENOENT);
3976 * Determine if there is a better boot device.
3979 spa_alt_rootvdev(rvd, &avd, &txg);
3981 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3982 "try booting from '%s'", avd->vdev_path);
3983 error = SET_ERROR(EINVAL);
3988 * If the boot device is part of a spare vdev then ensure that
3989 * we're booting off the active spare.
3991 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3992 !bvd->vdev_isspare) {
3993 cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3994 "try booting from '%s'",
3996 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3997 error = SET_ERROR(EINVAL);
4003 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4005 spa_config_exit(spa, SCL_ALL, FTAG);
4006 mutex_exit(&spa_namespace_lock);
4008 nvlist_free(config);
4014 extern int vdev_geom_read_pool_label(const char *name, nvlist_t ***configs,
4018 spa_generate_rootconf(const char *name)
4020 nvlist_t **configs, **tops;
4022 nvlist_t *best_cfg, *nvtop, *nvroot;
4031 if (vdev_geom_read_pool_label(name, &configs, &count) != 0)
4034 ASSERT3U(count, !=, 0);
4036 for (i = 0; i < count; i++) {
4039 VERIFY(nvlist_lookup_uint64(configs[i], ZPOOL_CONFIG_POOL_TXG,
4041 if (txg > best_txg) {
4043 best_cfg = configs[i];
4048 * Multi-vdev root pool configuration discovery is not supported yet.
4051 nvlist_lookup_uint64(best_cfg, ZPOOL_CONFIG_VDEV_CHILDREN, &nchildren);
4053 nvlist_lookup_uint64_array(best_cfg, ZPOOL_CONFIG_HOLE_ARRAY,
4056 tops = kmem_zalloc(nchildren * sizeof(void *), KM_SLEEP);
4057 for (i = 0; i < nchildren; i++) {
4060 if (configs[i] == NULL)
4062 VERIFY(nvlist_lookup_nvlist(configs[i], ZPOOL_CONFIG_VDEV_TREE,
4064 nvlist_dup(nvtop, &tops[i], KM_SLEEP);
4066 for (i = 0; holes != NULL && i < nholes; i++) {
4069 if (tops[holes[i]] != NULL)
4071 nvlist_alloc(&tops[holes[i]], NV_UNIQUE_NAME, KM_SLEEP);
4072 VERIFY(nvlist_add_string(tops[holes[i]], ZPOOL_CONFIG_TYPE,
4073 VDEV_TYPE_HOLE) == 0);
4074 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_ID,
4076 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_GUID,
4079 for (i = 0; i < nchildren; i++) {
4080 if (tops[i] != NULL)
4082 nvlist_alloc(&tops[i], NV_UNIQUE_NAME, KM_SLEEP);
4083 VERIFY(nvlist_add_string(tops[i], ZPOOL_CONFIG_TYPE,
4084 VDEV_TYPE_MISSING) == 0);
4085 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_ID,
4087 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_GUID,
4092 * Create pool config based on the best vdev config.
4094 nvlist_dup(best_cfg, &config, KM_SLEEP);
4097 * Put this pool's top-level vdevs into a root vdev.
4099 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4101 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4102 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
4103 VDEV_TYPE_ROOT) == 0);
4104 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
4105 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
4106 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
4107 tops, nchildren) == 0);
4110 * Replace the existing vdev_tree with the new root vdev in
4111 * this pool's configuration (remove the old, add the new).
4113 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
4116 * Drop vdev config elements that should not be present at pool level.
4118 nvlist_remove(config, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64);
4119 nvlist_remove(config, ZPOOL_CONFIG_TOP_GUID, DATA_TYPE_UINT64);
4121 for (i = 0; i < count; i++)
4122 nvlist_free(configs[i]);
4123 kmem_free(configs, count * sizeof(void *));
4124 for (i = 0; i < nchildren; i++)
4125 nvlist_free(tops[i]);
4126 kmem_free(tops, nchildren * sizeof(void *));
4127 nvlist_free(nvroot);
4132 spa_import_rootpool(const char *name)
4135 vdev_t *rvd, *bvd, *avd = NULL;
4136 nvlist_t *config, *nvtop;
4142 * Read the label from the boot device and generate a configuration.
4144 config = spa_generate_rootconf(name);
4146 mutex_enter(&spa_namespace_lock);
4147 if (config != NULL) {
4148 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
4149 &pname) == 0 && strcmp(name, pname) == 0);
4150 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg)
4153 if ((spa = spa_lookup(pname)) != NULL) {
4155 * Remove the existing root pool from the namespace so
4156 * that we can replace it with the correct config
4161 spa = spa_add(pname, config, NULL);
4164 * Set spa_ubsync.ub_version as it can be used in vdev_alloc()
4165 * via spa_version().
4167 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
4168 &spa->spa_ubsync.ub_version) != 0)
4169 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
4170 } else if ((spa = spa_lookup(name)) == NULL) {
4171 mutex_exit(&spa_namespace_lock);
4172 nvlist_free(config);
4173 cmn_err(CE_NOTE, "Cannot find the pool label for '%s'",
4177 VERIFY(nvlist_dup(spa->spa_config, &config, KM_SLEEP) == 0);
4179 spa->spa_is_root = B_TRUE;
4180 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
4183 * Build up a vdev tree based on the boot device's label config.
4185 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4187 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4188 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
4189 VDEV_ALLOC_ROOTPOOL);
4190 spa_config_exit(spa, SCL_ALL, FTAG);
4192 mutex_exit(&spa_namespace_lock);
4193 nvlist_free(config);
4194 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
4199 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4201 spa_config_exit(spa, SCL_ALL, FTAG);
4202 mutex_exit(&spa_namespace_lock);
4204 nvlist_free(config);
4212 * Import a non-root pool into the system.
4215 spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
4218 char *altroot = NULL;
4219 spa_load_state_t state = SPA_LOAD_IMPORT;
4220 zpool_rewind_policy_t policy;
4221 uint64_t mode = spa_mode_global;
4222 uint64_t readonly = B_FALSE;
4225 nvlist_t **spares, **l2cache;
4226 uint_t nspares, nl2cache;
4229 * If a pool with this name exists, return failure.
4231 mutex_enter(&spa_namespace_lock);
4232 if (spa_lookup(pool) != NULL) {
4233 mutex_exit(&spa_namespace_lock);
4234 return (SET_ERROR(EEXIST));
4238 * Create and initialize the spa structure.
4240 (void) nvlist_lookup_string(props,
4241 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4242 (void) nvlist_lookup_uint64(props,
4243 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
4246 spa = spa_add(pool, config, altroot);
4247 spa->spa_import_flags = flags;
4250 * Verbatim import - Take a pool and insert it into the namespace
4251 * as if it had been loaded at boot.
4253 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
4255 spa_configfile_set(spa, props, B_FALSE);
4257 spa_config_sync(spa, B_FALSE, B_TRUE);
4258 spa_event_notify(spa, NULL, ESC_ZFS_POOL_IMPORT);
4260 mutex_exit(&spa_namespace_lock);
4264 spa_activate(spa, mode);
4267 * Don't start async tasks until we know everything is healthy.
4269 spa_async_suspend(spa);
4271 zpool_get_rewind_policy(config, &policy);
4272 if (policy.zrp_request & ZPOOL_DO_REWIND)
4273 state = SPA_LOAD_RECOVER;
4276 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
4277 * because the user-supplied config is actually the one to trust when
4280 if (state != SPA_LOAD_RECOVER)
4281 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
4283 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
4284 policy.zrp_request);
4287 * Propagate anything learned while loading the pool and pass it
4288 * back to caller (i.e. rewind info, missing devices, etc).
4290 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4291 spa->spa_load_info) == 0);
4293 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4295 * Toss any existing sparelist, as it doesn't have any validity
4296 * anymore, and conflicts with spa_has_spare().
4298 if (spa->spa_spares.sav_config) {
4299 nvlist_free(spa->spa_spares.sav_config);
4300 spa->spa_spares.sav_config = NULL;
4301 spa_load_spares(spa);
4303 if (spa->spa_l2cache.sav_config) {
4304 nvlist_free(spa->spa_l2cache.sav_config);
4305 spa->spa_l2cache.sav_config = NULL;
4306 spa_load_l2cache(spa);
4309 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4312 error = spa_validate_aux(spa, nvroot, -1ULL,
4315 error = spa_validate_aux(spa, nvroot, -1ULL,
4316 VDEV_ALLOC_L2CACHE);
4317 spa_config_exit(spa, SCL_ALL, FTAG);
4320 spa_configfile_set(spa, props, B_FALSE);
4322 if (error != 0 || (props && spa_writeable(spa) &&
4323 (error = spa_prop_set(spa, props)))) {
4325 spa_deactivate(spa);
4327 mutex_exit(&spa_namespace_lock);
4331 spa_async_resume(spa);
4334 * Override any spares and level 2 cache devices as specified by
4335 * the user, as these may have correct device names/devids, etc.
4337 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
4338 &spares, &nspares) == 0) {
4339 if (spa->spa_spares.sav_config)
4340 VERIFY(nvlist_remove(spa->spa_spares.sav_config,
4341 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
4343 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
4344 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4345 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
4346 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
4347 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4348 spa_load_spares(spa);
4349 spa_config_exit(spa, SCL_ALL, FTAG);
4350 spa->spa_spares.sav_sync = B_TRUE;
4352 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
4353 &l2cache, &nl2cache) == 0) {
4354 if (spa->spa_l2cache.sav_config)
4355 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
4356 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
4358 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
4359 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4360 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
4361 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
4362 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4363 spa_load_l2cache(spa);
4364 spa_config_exit(spa, SCL_ALL, FTAG);
4365 spa->spa_l2cache.sav_sync = B_TRUE;
4369 * Check for any removed devices.
4371 if (spa->spa_autoreplace) {
4372 spa_aux_check_removed(&spa->spa_spares);
4373 spa_aux_check_removed(&spa->spa_l2cache);
4376 if (spa_writeable(spa)) {
4378 * Update the config cache to include the newly-imported pool.
4380 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4384 * It's possible that the pool was expanded while it was exported.
4385 * We kick off an async task to handle this for us.
4387 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
4389 spa_history_log_version(spa, "import");
4391 spa_event_notify(spa, NULL, ESC_ZFS_POOL_IMPORT);
4393 mutex_exit(&spa_namespace_lock);
4397 zvol_create_minors(pool);
4404 spa_tryimport(nvlist_t *tryconfig)
4406 nvlist_t *config = NULL;
4412 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
4415 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
4419 * Create and initialize the spa structure.
4421 mutex_enter(&spa_namespace_lock);
4422 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
4423 spa_activate(spa, FREAD);
4426 * Pass off the heavy lifting to spa_load().
4427 * Pass TRUE for mosconfig because the user-supplied config
4428 * is actually the one to trust when doing an import.
4430 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
4433 * If 'tryconfig' was at least parsable, return the current config.
4435 if (spa->spa_root_vdev != NULL) {
4436 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4437 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
4439 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4441 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
4442 spa->spa_uberblock.ub_timestamp) == 0);
4443 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4444 spa->spa_load_info) == 0);
4447 * If the bootfs property exists on this pool then we
4448 * copy it out so that external consumers can tell which
4449 * pools are bootable.
4451 if ((!error || error == EEXIST) && spa->spa_bootfs) {
4452 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4455 * We have to play games with the name since the
4456 * pool was opened as TRYIMPORT_NAME.
4458 if (dsl_dsobj_to_dsname(spa_name(spa),
4459 spa->spa_bootfs, tmpname) == 0) {
4461 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4463 cp = strchr(tmpname, '/');
4465 (void) strlcpy(dsname, tmpname,
4468 (void) snprintf(dsname, MAXPATHLEN,
4469 "%s/%s", poolname, ++cp);
4471 VERIFY(nvlist_add_string(config,
4472 ZPOOL_CONFIG_BOOTFS, dsname) == 0);
4473 kmem_free(dsname, MAXPATHLEN);
4475 kmem_free(tmpname, MAXPATHLEN);
4479 * Add the list of hot spares and level 2 cache devices.
4481 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4482 spa_add_spares(spa, config);
4483 spa_add_l2cache(spa, config);
4484 spa_config_exit(spa, SCL_CONFIG, FTAG);
4488 spa_deactivate(spa);
4490 mutex_exit(&spa_namespace_lock);
4496 * Pool export/destroy
4498 * The act of destroying or exporting a pool is very simple. We make sure there
4499 * is no more pending I/O and any references to the pool are gone. Then, we
4500 * update the pool state and sync all the labels to disk, removing the
4501 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4502 * we don't sync the labels or remove the configuration cache.
4505 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
4506 boolean_t force, boolean_t hardforce)
4513 if (!(spa_mode_global & FWRITE))
4514 return (SET_ERROR(EROFS));
4516 mutex_enter(&spa_namespace_lock);
4517 if ((spa = spa_lookup(pool)) == NULL) {
4518 mutex_exit(&spa_namespace_lock);
4519 return (SET_ERROR(ENOENT));
4523 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4524 * reacquire the namespace lock, and see if we can export.
4526 spa_open_ref(spa, FTAG);
4527 mutex_exit(&spa_namespace_lock);
4528 spa_async_suspend(spa);
4529 mutex_enter(&spa_namespace_lock);
4530 spa_close(spa, FTAG);
4533 * The pool will be in core if it's openable,
4534 * in which case we can modify its state.
4536 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
4538 * Objsets may be open only because they're dirty, so we
4539 * have to force it to sync before checking spa_refcnt.
4541 txg_wait_synced(spa->spa_dsl_pool, 0);
4542 spa_evicting_os_wait(spa);
4545 * A pool cannot be exported or destroyed if there are active
4546 * references. If we are resetting a pool, allow references by
4547 * fault injection handlers.
4549 if (!spa_refcount_zero(spa) ||
4550 (spa->spa_inject_ref != 0 &&
4551 new_state != POOL_STATE_UNINITIALIZED)) {
4552 spa_async_resume(spa);
4553 mutex_exit(&spa_namespace_lock);
4554 return (SET_ERROR(EBUSY));
4558 * A pool cannot be exported if it has an active shared spare.
4559 * This is to prevent other pools stealing the active spare
4560 * from an exported pool. At user's own will, such pool can
4561 * be forcedly exported.
4563 if (!force && new_state == POOL_STATE_EXPORTED &&
4564 spa_has_active_shared_spare(spa)) {
4565 spa_async_resume(spa);
4566 mutex_exit(&spa_namespace_lock);
4567 return (SET_ERROR(EXDEV));
4571 * We want this to be reflected on every label,
4572 * so mark them all dirty. spa_unload() will do the
4573 * final sync that pushes these changes out.
4575 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
4576 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4577 spa->spa_state = new_state;
4578 spa->spa_final_txg = spa_last_synced_txg(spa) +
4580 vdev_config_dirty(spa->spa_root_vdev);
4581 spa_config_exit(spa, SCL_ALL, FTAG);
4585 spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
4587 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4589 spa_deactivate(spa);
4592 if (oldconfig && spa->spa_config)
4593 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
4595 if (new_state != POOL_STATE_UNINITIALIZED) {
4597 spa_config_sync(spa, B_TRUE, B_TRUE);
4600 mutex_exit(&spa_namespace_lock);
4606 * Destroy a storage pool.
4609 spa_destroy(char *pool)
4611 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
4616 * Export a storage pool.
4619 spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
4620 boolean_t hardforce)
4622 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
4627 * Similar to spa_export(), this unloads the spa_t without actually removing it
4628 * from the namespace in any way.
4631 spa_reset(char *pool)
4633 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
4638 * ==========================================================================
4639 * Device manipulation
4640 * ==========================================================================
4644 * Add a device to a storage pool.
4647 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
4651 vdev_t *rvd = spa->spa_root_vdev;
4653 nvlist_t **spares, **l2cache;
4654 uint_t nspares, nl2cache;
4656 ASSERT(spa_writeable(spa));
4658 txg = spa_vdev_enter(spa);
4660 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
4661 VDEV_ALLOC_ADD)) != 0)
4662 return (spa_vdev_exit(spa, NULL, txg, error));
4664 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */
4666 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
4670 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
4674 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
4675 return (spa_vdev_exit(spa, vd, txg, EINVAL));
4677 if (vd->vdev_children != 0 &&
4678 (error = vdev_create(vd, txg, B_FALSE)) != 0)
4679 return (spa_vdev_exit(spa, vd, txg, error));
4682 * We must validate the spares and l2cache devices after checking the
4683 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
4685 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
4686 return (spa_vdev_exit(spa, vd, txg, error));
4689 * Transfer each new top-level vdev from vd to rvd.
4691 for (int c = 0; c < vd->vdev_children; c++) {
4694 * Set the vdev id to the first hole, if one exists.
4696 for (id = 0; id < rvd->vdev_children; id++) {
4697 if (rvd->vdev_child[id]->vdev_ishole) {
4698 vdev_free(rvd->vdev_child[id]);
4702 tvd = vd->vdev_child[c];
4703 vdev_remove_child(vd, tvd);
4705 vdev_add_child(rvd, tvd);
4706 vdev_config_dirty(tvd);
4710 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
4711 ZPOOL_CONFIG_SPARES);
4712 spa_load_spares(spa);
4713 spa->spa_spares.sav_sync = B_TRUE;
4716 if (nl2cache != 0) {
4717 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
4718 ZPOOL_CONFIG_L2CACHE);
4719 spa_load_l2cache(spa);
4720 spa->spa_l2cache.sav_sync = B_TRUE;
4724 * We have to be careful when adding new vdevs to an existing pool.
4725 * If other threads start allocating from these vdevs before we
4726 * sync the config cache, and we lose power, then upon reboot we may
4727 * fail to open the pool because there are DVAs that the config cache
4728 * can't translate. Therefore, we first add the vdevs without
4729 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4730 * and then let spa_config_update() initialize the new metaslabs.
4732 * spa_load() checks for added-but-not-initialized vdevs, so that
4733 * if we lose power at any point in this sequence, the remaining
4734 * steps will be completed the next time we load the pool.
4736 (void) spa_vdev_exit(spa, vd, txg, 0);
4738 mutex_enter(&spa_namespace_lock);
4739 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4740 spa_event_notify(spa, NULL, ESC_ZFS_VDEV_ADD);
4741 mutex_exit(&spa_namespace_lock);
4747 * Attach a device to a mirror. The arguments are the path to any device
4748 * in the mirror, and the nvroot for the new device. If the path specifies
4749 * a device that is not mirrored, we automatically insert the mirror vdev.
4751 * If 'replacing' is specified, the new device is intended to replace the
4752 * existing device; in this case the two devices are made into their own
4753 * mirror using the 'replacing' vdev, which is functionally identical to
4754 * the mirror vdev (it actually reuses all the same ops) but has a few
4755 * extra rules: you can't attach to it after it's been created, and upon
4756 * completion of resilvering, the first disk (the one being replaced)
4757 * is automatically detached.
4760 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
4762 uint64_t txg, dtl_max_txg;
4763 vdev_t *rvd = spa->spa_root_vdev;
4764 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
4766 char *oldvdpath, *newvdpath;
4770 ASSERT(spa_writeable(spa));
4772 txg = spa_vdev_enter(spa);
4774 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
4777 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4779 if (!oldvd->vdev_ops->vdev_op_leaf)
4780 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4782 pvd = oldvd->vdev_parent;
4784 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
4785 VDEV_ALLOC_ATTACH)) != 0)
4786 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4788 if (newrootvd->vdev_children != 1)
4789 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4791 newvd = newrootvd->vdev_child[0];
4793 if (!newvd->vdev_ops->vdev_op_leaf)
4794 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4796 if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
4797 return (spa_vdev_exit(spa, newrootvd, txg, error));
4800 * Spares can't replace logs
4802 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
4803 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4807 * For attach, the only allowable parent is a mirror or the root
4810 if (pvd->vdev_ops != &vdev_mirror_ops &&
4811 pvd->vdev_ops != &vdev_root_ops)
4812 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4814 pvops = &vdev_mirror_ops;
4817 * Active hot spares can only be replaced by inactive hot
4820 if (pvd->vdev_ops == &vdev_spare_ops &&
4821 oldvd->vdev_isspare &&
4822 !spa_has_spare(spa, newvd->vdev_guid))
4823 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4826 * If the source is a hot spare, and the parent isn't already a
4827 * spare, then we want to create a new hot spare. Otherwise, we
4828 * want to create a replacing vdev. The user is not allowed to
4829 * attach to a spared vdev child unless the 'isspare' state is
4830 * the same (spare replaces spare, non-spare replaces
4833 if (pvd->vdev_ops == &vdev_replacing_ops &&
4834 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
4835 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4836 } else if (pvd->vdev_ops == &vdev_spare_ops &&
4837 newvd->vdev_isspare != oldvd->vdev_isspare) {
4838 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4841 if (newvd->vdev_isspare)
4842 pvops = &vdev_spare_ops;
4844 pvops = &vdev_replacing_ops;
4848 * Make sure the new device is big enough.
4850 if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
4851 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
4854 * The new device cannot have a higher alignment requirement
4855 * than the top-level vdev.
4857 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4858 return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4861 * If this is an in-place replacement, update oldvd's path and devid
4862 * to make it distinguishable from newvd, and unopenable from now on.
4864 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4865 spa_strfree(oldvd->vdev_path);
4866 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4868 (void) sprintf(oldvd->vdev_path, "%s/%s",
4869 newvd->vdev_path, "old");
4870 if (oldvd->vdev_devid != NULL) {
4871 spa_strfree(oldvd->vdev_devid);
4872 oldvd->vdev_devid = NULL;
4876 /* mark the device being resilvered */
4877 newvd->vdev_resilver_txg = txg;
4880 * If the parent is not a mirror, or if we're replacing, insert the new
4881 * mirror/replacing/spare vdev above oldvd.
4883 if (pvd->vdev_ops != pvops)
4884 pvd = vdev_add_parent(oldvd, pvops);
4886 ASSERT(pvd->vdev_top->vdev_parent == rvd);
4887 ASSERT(pvd->vdev_ops == pvops);
4888 ASSERT(oldvd->vdev_parent == pvd);
4891 * Extract the new device from its root and add it to pvd.
4893 vdev_remove_child(newrootvd, newvd);
4894 newvd->vdev_id = pvd->vdev_children;
4895 newvd->vdev_crtxg = oldvd->vdev_crtxg;
4896 vdev_add_child(pvd, newvd);
4898 tvd = newvd->vdev_top;
4899 ASSERT(pvd->vdev_top == tvd);
4900 ASSERT(tvd->vdev_parent == rvd);
4902 vdev_config_dirty(tvd);
4905 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4906 * for any dmu_sync-ed blocks. It will propagate upward when
4907 * spa_vdev_exit() calls vdev_dtl_reassess().
4909 dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4911 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4912 dtl_max_txg - TXG_INITIAL);
4914 if (newvd->vdev_isspare) {
4915 spa_spare_activate(newvd);
4916 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
4919 oldvdpath = spa_strdup(oldvd->vdev_path);
4920 newvdpath = spa_strdup(newvd->vdev_path);
4921 newvd_isspare = newvd->vdev_isspare;
4924 * Mark newvd's DTL dirty in this txg.
4926 vdev_dirty(tvd, VDD_DTL, newvd, txg);
4929 * Schedule the resilver to restart in the future. We do this to
4930 * ensure that dmu_sync-ed blocks have been stitched into the
4931 * respective datasets.
4933 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4935 if (spa->spa_bootfs)
4936 spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
4938 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_ATTACH);
4943 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4945 spa_history_log_internal(spa, "vdev attach", NULL,
4946 "%s vdev=%s %s vdev=%s",
4947 replacing && newvd_isspare ? "spare in" :
4948 replacing ? "replace" : "attach", newvdpath,
4949 replacing ? "for" : "to", oldvdpath);
4951 spa_strfree(oldvdpath);
4952 spa_strfree(newvdpath);
4958 * Detach a device from a mirror or replacing vdev.
4960 * If 'replace_done' is specified, only detach if the parent
4961 * is a replacing vdev.
4964 spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4968 vdev_t *rvd = spa->spa_root_vdev;
4969 vdev_t *vd, *pvd, *cvd, *tvd;
4970 boolean_t unspare = B_FALSE;
4971 uint64_t unspare_guid = 0;
4974 ASSERT(spa_writeable(spa));
4976 txg = spa_vdev_enter(spa);
4978 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4981 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4983 if (!vd->vdev_ops->vdev_op_leaf)
4984 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4986 pvd = vd->vdev_parent;
4989 * If the parent/child relationship is not as expected, don't do it.
4990 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4991 * vdev that's replacing B with C. The user's intent in replacing
4992 * is to go from M(A,B) to M(A,C). If the user decides to cancel
4993 * the replace by detaching C, the expected behavior is to end up
4994 * M(A,B). But suppose that right after deciding to detach C,
4995 * the replacement of B completes. We would have M(A,C), and then
4996 * ask to detach C, which would leave us with just A -- not what
4997 * the user wanted. To prevent this, we make sure that the
4998 * parent/child relationship hasn't changed -- in this example,
4999 * that C's parent is still the replacing vdev R.
5001 if (pvd->vdev_guid != pguid && pguid != 0)
5002 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
5005 * Only 'replacing' or 'spare' vdevs can be replaced.
5007 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
5008 pvd->vdev_ops != &vdev_spare_ops)
5009 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
5011 ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
5012 spa_version(spa) >= SPA_VERSION_SPARES);
5015 * Only mirror, replacing, and spare vdevs support detach.
5017 if (pvd->vdev_ops != &vdev_replacing_ops &&
5018 pvd->vdev_ops != &vdev_mirror_ops &&
5019 pvd->vdev_ops != &vdev_spare_ops)
5020 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
5023 * If this device has the only valid copy of some data,
5024 * we cannot safely detach it.
5026 if (vdev_dtl_required(vd))
5027 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
5029 ASSERT(pvd->vdev_children >= 2);
5032 * If we are detaching the second disk from a replacing vdev, then
5033 * check to see if we changed the original vdev's path to have "/old"
5034 * at the end in spa_vdev_attach(). If so, undo that change now.
5036 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
5037 vd->vdev_path != NULL) {
5038 size_t len = strlen(vd->vdev_path);
5040 for (int c = 0; c < pvd->vdev_children; c++) {
5041 cvd = pvd->vdev_child[c];
5043 if (cvd == vd || cvd->vdev_path == NULL)
5046 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
5047 strcmp(cvd->vdev_path + len, "/old") == 0) {
5048 spa_strfree(cvd->vdev_path);
5049 cvd->vdev_path = spa_strdup(vd->vdev_path);
5056 * If we are detaching the original disk from a spare, then it implies
5057 * that the spare should become a real disk, and be removed from the
5058 * active spare list for the pool.
5060 if (pvd->vdev_ops == &vdev_spare_ops &&
5062 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
5066 * Erase the disk labels so the disk can be used for other things.
5067 * This must be done after all other error cases are handled,
5068 * but before we disembowel vd (so we can still do I/O to it).
5069 * But if we can't do it, don't treat the error as fatal --
5070 * it may be that the unwritability of the disk is the reason
5071 * it's being detached!
5073 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5076 * Remove vd from its parent and compact the parent's children.
5078 vdev_remove_child(pvd, vd);
5079 vdev_compact_children(pvd);
5082 * Remember one of the remaining children so we can get tvd below.
5084 cvd = pvd->vdev_child[pvd->vdev_children - 1];
5087 * If we need to remove the remaining child from the list of hot spares,
5088 * do it now, marking the vdev as no longer a spare in the process.
5089 * We must do this before vdev_remove_parent(), because that can
5090 * change the GUID if it creates a new toplevel GUID. For a similar
5091 * reason, we must remove the spare now, in the same txg as the detach;
5092 * otherwise someone could attach a new sibling, change the GUID, and
5093 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
5096 ASSERT(cvd->vdev_isspare);
5097 spa_spare_remove(cvd);
5098 unspare_guid = cvd->vdev_guid;
5099 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
5100 cvd->vdev_unspare = B_TRUE;
5104 * If the parent mirror/replacing vdev only has one child,
5105 * the parent is no longer needed. Remove it from the tree.
5107 if (pvd->vdev_children == 1) {
5108 if (pvd->vdev_ops == &vdev_spare_ops)
5109 cvd->vdev_unspare = B_FALSE;
5110 vdev_remove_parent(cvd);
5115 * We don't set tvd until now because the parent we just removed
5116 * may have been the previous top-level vdev.
5118 tvd = cvd->vdev_top;
5119 ASSERT(tvd->vdev_parent == rvd);
5122 * Reevaluate the parent vdev state.
5124 vdev_propagate_state(cvd);
5127 * If the 'autoexpand' property is set on the pool then automatically
5128 * try to expand the size of the pool. For example if the device we
5129 * just detached was smaller than the others, it may be possible to
5130 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
5131 * first so that we can obtain the updated sizes of the leaf vdevs.
5133 if (spa->spa_autoexpand) {
5135 vdev_expand(tvd, txg);
5138 vdev_config_dirty(tvd);
5141 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
5142 * vd->vdev_detached is set and free vd's DTL object in syncing context.
5143 * But first make sure we're not on any *other* txg's DTL list, to
5144 * prevent vd from being accessed after it's freed.
5146 vdpath = spa_strdup(vd->vdev_path);
5147 for (int t = 0; t < TXG_SIZE; t++)
5148 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
5149 vd->vdev_detached = B_TRUE;
5150 vdev_dirty(tvd, VDD_DTL, vd, txg);
5152 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
5154 /* hang on to the spa before we release the lock */
5155 spa_open_ref(spa, FTAG);
5157 error = spa_vdev_exit(spa, vd, txg, 0);
5159 spa_history_log_internal(spa, "detach", NULL,
5161 spa_strfree(vdpath);
5164 * If this was the removal of the original device in a hot spare vdev,
5165 * then we want to go through and remove the device from the hot spare
5166 * list of every other pool.
5169 spa_t *altspa = NULL;
5171 mutex_enter(&spa_namespace_lock);
5172 while ((altspa = spa_next(altspa)) != NULL) {
5173 if (altspa->spa_state != POOL_STATE_ACTIVE ||
5177 spa_open_ref(altspa, FTAG);
5178 mutex_exit(&spa_namespace_lock);
5179 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
5180 mutex_enter(&spa_namespace_lock);
5181 spa_close(altspa, FTAG);
5183 mutex_exit(&spa_namespace_lock);
5185 /* search the rest of the vdevs for spares to remove */
5186 spa_vdev_resilver_done(spa);
5189 /* all done with the spa; OK to release */
5190 mutex_enter(&spa_namespace_lock);
5191 spa_close(spa, FTAG);
5192 mutex_exit(&spa_namespace_lock);
5198 * Split a set of devices from their mirrors, and create a new pool from them.
5201 spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
5202 nvlist_t *props, boolean_t exp)
5205 uint64_t txg, *glist;
5207 uint_t c, children, lastlog;
5208 nvlist_t **child, *nvl, *tmp;
5210 char *altroot = NULL;
5211 vdev_t *rvd, **vml = NULL; /* vdev modify list */
5212 boolean_t activate_slog;
5214 ASSERT(spa_writeable(spa));
5216 txg = spa_vdev_enter(spa);
5218 /* clear the log and flush everything up to now */
5219 activate_slog = spa_passivate_log(spa);
5220 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5221 error = spa_offline_log(spa);
5222 txg = spa_vdev_config_enter(spa);
5225 spa_activate_log(spa);
5228 return (spa_vdev_exit(spa, NULL, txg, error));
5230 /* check new spa name before going any further */
5231 if (spa_lookup(newname) != NULL)
5232 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
5235 * scan through all the children to ensure they're all mirrors
5237 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
5238 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
5240 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5242 /* first, check to ensure we've got the right child count */
5243 rvd = spa->spa_root_vdev;
5245 for (c = 0; c < rvd->vdev_children; c++) {
5246 vdev_t *vd = rvd->vdev_child[c];
5248 /* don't count the holes & logs as children */
5249 if (vd->vdev_islog || vd->vdev_ishole) {
5257 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
5258 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5260 /* next, ensure no spare or cache devices are part of the split */
5261 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
5262 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
5263 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5265 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
5266 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
5268 /* then, loop over each vdev and validate it */
5269 for (c = 0; c < children; c++) {
5270 uint64_t is_hole = 0;
5272 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
5276 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
5277 spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
5280 error = SET_ERROR(EINVAL);
5285 /* which disk is going to be split? */
5286 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
5288 error = SET_ERROR(EINVAL);
5292 /* look it up in the spa */
5293 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
5294 if (vml[c] == NULL) {
5295 error = SET_ERROR(ENODEV);
5299 /* make sure there's nothing stopping the split */
5300 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
5301 vml[c]->vdev_islog ||
5302 vml[c]->vdev_ishole ||
5303 vml[c]->vdev_isspare ||
5304 vml[c]->vdev_isl2cache ||
5305 !vdev_writeable(vml[c]) ||
5306 vml[c]->vdev_children != 0 ||
5307 vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
5308 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
5309 error = SET_ERROR(EINVAL);
5313 if (vdev_dtl_required(vml[c])) {
5314 error = SET_ERROR(EBUSY);
5318 /* we need certain info from the top level */
5319 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
5320 vml[c]->vdev_top->vdev_ms_array) == 0);
5321 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
5322 vml[c]->vdev_top->vdev_ms_shift) == 0);
5323 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
5324 vml[c]->vdev_top->vdev_asize) == 0);
5325 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
5326 vml[c]->vdev_top->vdev_ashift) == 0);
5330 kmem_free(vml, children * sizeof (vdev_t *));
5331 kmem_free(glist, children * sizeof (uint64_t));
5332 return (spa_vdev_exit(spa, NULL, txg, error));
5335 /* stop writers from using the disks */
5336 for (c = 0; c < children; c++) {
5338 vml[c]->vdev_offline = B_TRUE;
5340 vdev_reopen(spa->spa_root_vdev);
5343 * Temporarily record the splitting vdevs in the spa config. This
5344 * will disappear once the config is regenerated.
5346 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5347 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
5348 glist, children) == 0);
5349 kmem_free(glist, children * sizeof (uint64_t));
5351 mutex_enter(&spa->spa_props_lock);
5352 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
5354 mutex_exit(&spa->spa_props_lock);
5355 spa->spa_config_splitting = nvl;
5356 vdev_config_dirty(spa->spa_root_vdev);
5358 /* configure and create the new pool */
5359 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
5360 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
5361 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
5362 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5363 spa_version(spa)) == 0);
5364 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
5365 spa->spa_config_txg) == 0);
5366 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
5367 spa_generate_guid(NULL)) == 0);
5368 (void) nvlist_lookup_string(props,
5369 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
5371 /* add the new pool to the namespace */
5372 newspa = spa_add(newname, config, altroot);
5373 newspa->spa_config_txg = spa->spa_config_txg;
5374 spa_set_log_state(newspa, SPA_LOG_CLEAR);
5376 /* release the spa config lock, retaining the namespace lock */
5377 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5379 if (zio_injection_enabled)
5380 zio_handle_panic_injection(spa, FTAG, 1);
5382 spa_activate(newspa, spa_mode_global);
5383 spa_async_suspend(newspa);
5386 /* mark that we are creating new spa by splitting */
5387 newspa->spa_splitting_newspa = B_TRUE;
5389 /* create the new pool from the disks of the original pool */
5390 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
5392 newspa->spa_splitting_newspa = B_FALSE;
5397 /* if that worked, generate a real config for the new pool */
5398 if (newspa->spa_root_vdev != NULL) {
5399 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
5400 NV_UNIQUE_NAME, KM_SLEEP) == 0);
5401 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
5402 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
5403 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
5408 if (props != NULL) {
5409 spa_configfile_set(newspa, props, B_FALSE);
5410 error = spa_prop_set(newspa, props);
5415 /* flush everything */
5416 txg = spa_vdev_config_enter(newspa);
5417 vdev_config_dirty(newspa->spa_root_vdev);
5418 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
5420 if (zio_injection_enabled)
5421 zio_handle_panic_injection(spa, FTAG, 2);
5423 spa_async_resume(newspa);
5425 /* finally, update the original pool's config */
5426 txg = spa_vdev_config_enter(spa);
5427 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
5428 error = dmu_tx_assign(tx, TXG_WAIT);
5431 for (c = 0; c < children; c++) {
5432 if (vml[c] != NULL) {
5435 spa_history_log_internal(spa, "detach", tx,
5436 "vdev=%s", vml[c]->vdev_path);
5440 vdev_config_dirty(spa->spa_root_vdev);
5441 spa->spa_config_splitting = NULL;
5445 (void) spa_vdev_exit(spa, NULL, txg, 0);
5447 if (zio_injection_enabled)
5448 zio_handle_panic_injection(spa, FTAG, 3);
5450 /* split is complete; log a history record */
5451 spa_history_log_internal(newspa, "split", NULL,
5452 "from pool %s", spa_name(spa));
5454 kmem_free(vml, children * sizeof (vdev_t *));
5456 /* if we're not going to mount the filesystems in userland, export */
5458 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
5465 spa_deactivate(newspa);
5468 txg = spa_vdev_config_enter(spa);
5470 /* re-online all offlined disks */
5471 for (c = 0; c < children; c++) {
5473 vml[c]->vdev_offline = B_FALSE;
5475 vdev_reopen(spa->spa_root_vdev);
5477 nvlist_free(spa->spa_config_splitting);
5478 spa->spa_config_splitting = NULL;
5479 (void) spa_vdev_exit(spa, NULL, txg, error);
5481 kmem_free(vml, children * sizeof (vdev_t *));
5486 spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
5488 for (int i = 0; i < count; i++) {
5491 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
5494 if (guid == target_guid)
5502 spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
5503 nvlist_t *dev_to_remove)
5505 nvlist_t **newdev = NULL;
5508 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
5510 for (int i = 0, j = 0; i < count; i++) {
5511 if (dev[i] == dev_to_remove)
5513 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
5516 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
5517 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
5519 for (int i = 0; i < count - 1; i++)
5520 nvlist_free(newdev[i]);
5523 kmem_free(newdev, (count - 1) * sizeof (void *));
5527 * Evacuate the device.
5530 spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
5535 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5536 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5537 ASSERT(vd == vd->vdev_top);
5540 * Evacuate the device. We don't hold the config lock as writer
5541 * since we need to do I/O but we do keep the
5542 * spa_namespace_lock held. Once this completes the device
5543 * should no longer have any blocks allocated on it.
5545 if (vd->vdev_islog) {
5546 if (vd->vdev_stat.vs_alloc != 0)
5547 error = spa_offline_log(spa);
5549 error = SET_ERROR(ENOTSUP);
5556 * The evacuation succeeded. Remove any remaining MOS metadata
5557 * associated with this vdev, and wait for these changes to sync.
5559 ASSERT0(vd->vdev_stat.vs_alloc);
5560 txg = spa_vdev_config_enter(spa);
5561 vd->vdev_removing = B_TRUE;
5562 vdev_dirty_leaves(vd, VDD_DTL, txg);
5563 vdev_config_dirty(vd);
5564 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5570 * Complete the removal by cleaning up the namespace.
5573 spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
5575 vdev_t *rvd = spa->spa_root_vdev;
5576 uint64_t id = vd->vdev_id;
5577 boolean_t last_vdev = (id == (rvd->vdev_children - 1));
5579 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5580 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5581 ASSERT(vd == vd->vdev_top);
5584 * Only remove any devices which are empty.
5586 if (vd->vdev_stat.vs_alloc != 0)
5589 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5591 if (list_link_active(&vd->vdev_state_dirty_node))
5592 vdev_state_clean(vd);
5593 if (list_link_active(&vd->vdev_config_dirty_node))
5594 vdev_config_clean(vd);
5599 vdev_compact_children(rvd);
5601 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
5602 vdev_add_child(rvd, vd);
5604 vdev_config_dirty(rvd);
5607 * Reassess the health of our root vdev.
5613 * Remove a device from the pool -
5615 * Removing a device from the vdev namespace requires several steps
5616 * and can take a significant amount of time. As a result we use
5617 * the spa_vdev_config_[enter/exit] functions which allow us to
5618 * grab and release the spa_config_lock while still holding the namespace
5619 * lock. During each step the configuration is synced out.
5621 * Currently, this supports removing only hot spares, slogs, and level 2 ARC
5625 spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
5628 metaslab_group_t *mg;
5629 nvlist_t **spares, **l2cache, *nv;
5631 uint_t nspares, nl2cache;
5633 boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
5635 ASSERT(spa_writeable(spa));
5638 txg = spa_vdev_enter(spa);
5640 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5642 if (spa->spa_spares.sav_vdevs != NULL &&
5643 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5644 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
5645 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
5647 * Only remove the hot spare if it's not currently in use
5650 if (vd == NULL || unspare) {
5651 spa_vdev_remove_aux(spa->spa_spares.sav_config,
5652 ZPOOL_CONFIG_SPARES, spares, nspares, nv);
5653 spa_load_spares(spa);
5654 spa->spa_spares.sav_sync = B_TRUE;
5656 error = SET_ERROR(EBUSY);
5658 } else if (spa->spa_l2cache.sav_vdevs != NULL &&
5659 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5660 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
5661 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
5663 * Cache devices can always be removed.
5665 spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
5666 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
5667 spa_load_l2cache(spa);
5668 spa->spa_l2cache.sav_sync = B_TRUE;
5669 } else if (vd != NULL && vd->vdev_islog) {
5671 ASSERT(vd == vd->vdev_top);
5676 * Stop allocating from this vdev.
5678 metaslab_group_passivate(mg);
5681 * Wait for the youngest allocations and frees to sync,
5682 * and then wait for the deferral of those frees to finish.
5684 spa_vdev_config_exit(spa, NULL,
5685 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
5688 * Attempt to evacuate the vdev.
5690 error = spa_vdev_remove_evacuate(spa, vd);
5692 txg = spa_vdev_config_enter(spa);
5695 * If we couldn't evacuate the vdev, unwind.
5698 metaslab_group_activate(mg);
5699 return (spa_vdev_exit(spa, NULL, txg, error));
5703 * Clean up the vdev namespace.
5705 spa_vdev_remove_from_namespace(spa, vd);
5707 } else if (vd != NULL) {
5709 * Normal vdevs cannot be removed (yet).
5711 error = SET_ERROR(ENOTSUP);
5714 * There is no vdev of any kind with the specified guid.
5716 error = SET_ERROR(ENOENT);
5720 return (spa_vdev_exit(spa, NULL, txg, error));
5726 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5727 * currently spared, so we can detach it.
5730 spa_vdev_resilver_done_hunt(vdev_t *vd)
5732 vdev_t *newvd, *oldvd;
5734 for (int c = 0; c < vd->vdev_children; c++) {
5735 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
5741 * Check for a completed replacement. We always consider the first
5742 * vdev in the list to be the oldest vdev, and the last one to be
5743 * the newest (see spa_vdev_attach() for how that works). In
5744 * the case where the newest vdev is faulted, we will not automatically
5745 * remove it after a resilver completes. This is OK as it will require
5746 * user intervention to determine which disk the admin wishes to keep.
5748 if (vd->vdev_ops == &vdev_replacing_ops) {
5749 ASSERT(vd->vdev_children > 1);
5751 newvd = vd->vdev_child[vd->vdev_children - 1];
5752 oldvd = vd->vdev_child[0];
5754 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
5755 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5756 !vdev_dtl_required(oldvd))
5761 * Check for a completed resilver with the 'unspare' flag set.
5763 if (vd->vdev_ops == &vdev_spare_ops) {
5764 vdev_t *first = vd->vdev_child[0];
5765 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
5767 if (last->vdev_unspare) {
5770 } else if (first->vdev_unspare) {
5777 if (oldvd != NULL &&
5778 vdev_dtl_empty(newvd, DTL_MISSING) &&
5779 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5780 !vdev_dtl_required(oldvd))
5784 * If there are more than two spares attached to a disk,
5785 * and those spares are not required, then we want to
5786 * attempt to free them up now so that they can be used
5787 * by other pools. Once we're back down to a single
5788 * disk+spare, we stop removing them.
5790 if (vd->vdev_children > 2) {
5791 newvd = vd->vdev_child[1];
5793 if (newvd->vdev_isspare && last->vdev_isspare &&
5794 vdev_dtl_empty(last, DTL_MISSING) &&
5795 vdev_dtl_empty(last, DTL_OUTAGE) &&
5796 !vdev_dtl_required(newvd))
5805 spa_vdev_resilver_done(spa_t *spa)
5807 vdev_t *vd, *pvd, *ppvd;
5808 uint64_t guid, sguid, pguid, ppguid;
5810 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5812 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
5813 pvd = vd->vdev_parent;
5814 ppvd = pvd->vdev_parent;
5815 guid = vd->vdev_guid;
5816 pguid = pvd->vdev_guid;
5817 ppguid = ppvd->vdev_guid;
5820 * If we have just finished replacing a hot spared device, then
5821 * we need to detach the parent's first child (the original hot
5824 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
5825 ppvd->vdev_children == 2) {
5826 ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
5827 sguid = ppvd->vdev_child[1]->vdev_guid;
5829 ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd));
5831 spa_config_exit(spa, SCL_ALL, FTAG);
5832 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
5834 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
5836 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5839 spa_config_exit(spa, SCL_ALL, FTAG);
5843 * Update the stored path or FRU for this vdev.
5846 spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
5850 boolean_t sync = B_FALSE;
5852 ASSERT(spa_writeable(spa));
5854 spa_vdev_state_enter(spa, SCL_ALL);
5856 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
5857 return (spa_vdev_state_exit(spa, NULL, ENOENT));
5859 if (!vd->vdev_ops->vdev_op_leaf)
5860 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
5863 if (strcmp(value, vd->vdev_path) != 0) {
5864 spa_strfree(vd->vdev_path);
5865 vd->vdev_path = spa_strdup(value);
5869 if (vd->vdev_fru == NULL) {
5870 vd->vdev_fru = spa_strdup(value);
5872 } else if (strcmp(value, vd->vdev_fru) != 0) {
5873 spa_strfree(vd->vdev_fru);
5874 vd->vdev_fru = spa_strdup(value);
5879 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
5883 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
5885 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5889 spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5891 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5895 * ==========================================================================
5897 * ==========================================================================
5901 spa_scan_stop(spa_t *spa)
5903 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5904 if (dsl_scan_resilvering(spa->spa_dsl_pool))
5905 return (SET_ERROR(EBUSY));
5906 return (dsl_scan_cancel(spa->spa_dsl_pool));
5910 spa_scan(spa_t *spa, pool_scan_func_t func)
5912 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5914 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5915 return (SET_ERROR(ENOTSUP));
5918 * If a resilver was requested, but there is no DTL on a
5919 * writeable leaf device, we have nothing to do.
5921 if (func == POOL_SCAN_RESILVER &&
5922 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5923 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5927 return (dsl_scan(spa->spa_dsl_pool, func));
5931 * ==========================================================================
5932 * SPA async task processing
5933 * ==========================================================================
5937 spa_async_remove(spa_t *spa, vdev_t *vd)
5939 if (vd->vdev_remove_wanted) {
5940 vd->vdev_remove_wanted = B_FALSE;
5941 vd->vdev_delayed_close = B_FALSE;
5942 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5945 * We want to clear the stats, but we don't want to do a full
5946 * vdev_clear() as that will cause us to throw away
5947 * degraded/faulted state as well as attempt to reopen the
5948 * device, all of which is a waste.
5950 vd->vdev_stat.vs_read_errors = 0;
5951 vd->vdev_stat.vs_write_errors = 0;
5952 vd->vdev_stat.vs_checksum_errors = 0;
5954 vdev_state_dirty(vd->vdev_top);
5957 for (int c = 0; c < vd->vdev_children; c++)
5958 spa_async_remove(spa, vd->vdev_child[c]);
5962 spa_async_probe(spa_t *spa, vdev_t *vd)
5964 if (vd->vdev_probe_wanted) {
5965 vd->vdev_probe_wanted = B_FALSE;
5966 vdev_reopen(vd); /* vdev_open() does the actual probe */
5969 for (int c = 0; c < vd->vdev_children; c++)
5970 spa_async_probe(spa, vd->vdev_child[c]);
5974 spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5980 if (!spa->spa_autoexpand)
5983 for (int c = 0; c < vd->vdev_children; c++) {
5984 vdev_t *cvd = vd->vdev_child[c];
5985 spa_async_autoexpand(spa, cvd);
5988 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5991 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
5992 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
5994 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5995 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
5997 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
5998 ESC_ZFS_VDEV_AUTOEXPAND, attr, &eid, DDI_SLEEP);
6001 kmem_free(physpath, MAXPATHLEN);
6005 spa_async_thread(void *arg)
6010 ASSERT(spa->spa_sync_on);
6012 mutex_enter(&spa->spa_async_lock);
6013 tasks = spa->spa_async_tasks;
6014 spa->spa_async_tasks &= SPA_ASYNC_REMOVE;
6015 mutex_exit(&spa->spa_async_lock);
6018 * See if the config needs to be updated.
6020 if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
6021 uint64_t old_space, new_space;
6023 mutex_enter(&spa_namespace_lock);
6024 old_space = metaslab_class_get_space(spa_normal_class(spa));
6025 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
6026 new_space = metaslab_class_get_space(spa_normal_class(spa));
6027 mutex_exit(&spa_namespace_lock);
6030 * If the pool grew as a result of the config update,
6031 * then log an internal history event.
6033 if (new_space != old_space) {
6034 spa_history_log_internal(spa, "vdev online", NULL,
6035 "pool '%s' size: %llu(+%llu)",
6036 spa_name(spa), new_space, new_space - old_space);
6040 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
6041 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
6042 spa_async_autoexpand(spa, spa->spa_root_vdev);
6043 spa_config_exit(spa, SCL_CONFIG, FTAG);
6047 * See if any devices need to be probed.
6049 if (tasks & SPA_ASYNC_PROBE) {
6050 spa_vdev_state_enter(spa, SCL_NONE);
6051 spa_async_probe(spa, spa->spa_root_vdev);
6052 (void) spa_vdev_state_exit(spa, NULL, 0);
6056 * If any devices are done replacing, detach them.
6058 if (tasks & SPA_ASYNC_RESILVER_DONE)
6059 spa_vdev_resilver_done(spa);
6062 * Kick off a resilver.
6064 if (tasks & SPA_ASYNC_RESILVER)
6065 dsl_resilver_restart(spa->spa_dsl_pool, 0);
6068 * Let the world know that we're done.
6070 mutex_enter(&spa->spa_async_lock);
6071 spa->spa_async_thread = NULL;
6072 cv_broadcast(&spa->spa_async_cv);
6073 mutex_exit(&spa->spa_async_lock);
6078 spa_async_thread_vd(void *arg)
6083 ASSERT(spa->spa_sync_on);
6085 mutex_enter(&spa->spa_async_lock);
6086 tasks = spa->spa_async_tasks;
6088 spa->spa_async_tasks &= ~SPA_ASYNC_REMOVE;
6089 mutex_exit(&spa->spa_async_lock);
6092 * See if any devices need to be marked REMOVED.
6094 if (tasks & SPA_ASYNC_REMOVE) {
6095 spa_vdev_state_enter(spa, SCL_NONE);
6096 spa_async_remove(spa, spa->spa_root_vdev);
6097 for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
6098 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
6099 for (int i = 0; i < spa->spa_spares.sav_count; i++)
6100 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
6101 (void) spa_vdev_state_exit(spa, NULL, 0);
6105 * Let the world know that we're done.
6107 mutex_enter(&spa->spa_async_lock);
6108 tasks = spa->spa_async_tasks;
6109 if ((tasks & SPA_ASYNC_REMOVE) != 0)
6111 spa->spa_async_thread_vd = NULL;
6112 cv_broadcast(&spa->spa_async_cv);
6113 mutex_exit(&spa->spa_async_lock);
6118 spa_async_suspend(spa_t *spa)
6120 mutex_enter(&spa->spa_async_lock);
6121 spa->spa_async_suspended++;
6122 while (spa->spa_async_thread != NULL &&
6123 spa->spa_async_thread_vd != NULL)
6124 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
6125 mutex_exit(&spa->spa_async_lock);
6129 spa_async_resume(spa_t *spa)
6131 mutex_enter(&spa->spa_async_lock);
6132 ASSERT(spa->spa_async_suspended != 0);
6133 spa->spa_async_suspended--;
6134 mutex_exit(&spa->spa_async_lock);
6138 spa_async_tasks_pending(spa_t *spa)
6140 uint_t non_config_tasks;
6142 boolean_t config_task_suspended;
6144 non_config_tasks = spa->spa_async_tasks & ~(SPA_ASYNC_CONFIG_UPDATE |
6146 config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE;
6147 if (spa->spa_ccw_fail_time == 0) {
6148 config_task_suspended = B_FALSE;
6150 config_task_suspended =
6151 (gethrtime() - spa->spa_ccw_fail_time) <
6152 (zfs_ccw_retry_interval * NANOSEC);
6155 return (non_config_tasks || (config_task && !config_task_suspended));
6159 spa_async_dispatch(spa_t *spa)
6161 mutex_enter(&spa->spa_async_lock);
6162 if (spa_async_tasks_pending(spa) &&
6163 !spa->spa_async_suspended &&
6164 spa->spa_async_thread == NULL &&
6166 spa->spa_async_thread = thread_create(NULL, 0,
6167 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
6168 mutex_exit(&spa->spa_async_lock);
6172 spa_async_dispatch_vd(spa_t *spa)
6174 mutex_enter(&spa->spa_async_lock);
6175 if ((spa->spa_async_tasks & SPA_ASYNC_REMOVE) != 0 &&
6176 !spa->spa_async_suspended &&
6177 spa->spa_async_thread_vd == NULL &&
6179 spa->spa_async_thread_vd = thread_create(NULL, 0,
6180 spa_async_thread_vd, spa, 0, &p0, TS_RUN, maxclsyspri);
6181 mutex_exit(&spa->spa_async_lock);
6185 spa_async_request(spa_t *spa, int task)
6187 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
6188 mutex_enter(&spa->spa_async_lock);
6189 spa->spa_async_tasks |= task;
6190 mutex_exit(&spa->spa_async_lock);
6191 spa_async_dispatch_vd(spa);
6195 * ==========================================================================
6196 * SPA syncing routines
6197 * ==========================================================================
6201 bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
6204 bpobj_enqueue(bpo, bp, tx);
6209 spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
6213 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
6214 BP_GET_PSIZE(bp), zio->io_flags));
6219 * Note: this simple function is not inlined to make it easier to dtrace the
6220 * amount of time spent syncing frees.
6223 spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx)
6225 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6226 bplist_iterate(bpl, spa_free_sync_cb, zio, tx);
6227 VERIFY(zio_wait(zio) == 0);
6231 * Note: this simple function is not inlined to make it easier to dtrace the
6232 * amount of time spent syncing deferred frees.
6235 spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx)
6237 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6238 VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj,
6239 spa_free_sync_cb, zio, tx), ==, 0);
6240 VERIFY0(zio_wait(zio));
6245 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
6247 char *packed = NULL;
6252 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
6255 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
6256 * information. This avoids the dmu_buf_will_dirty() path and
6257 * saves us a pre-read to get data we don't actually care about.
6259 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
6260 packed = kmem_alloc(bufsize, KM_SLEEP);
6262 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
6264 bzero(packed + nvsize, bufsize - nvsize);
6266 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
6268 kmem_free(packed, bufsize);
6270 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
6271 dmu_buf_will_dirty(db, tx);
6272 *(uint64_t *)db->db_data = nvsize;
6273 dmu_buf_rele(db, FTAG);
6277 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
6278 const char *config, const char *entry)
6288 * Update the MOS nvlist describing the list of available devices.
6289 * spa_validate_aux() will have already made sure this nvlist is
6290 * valid and the vdevs are labeled appropriately.
6292 if (sav->sav_object == 0) {
6293 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
6294 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
6295 sizeof (uint64_t), tx);
6296 VERIFY(zap_update(spa->spa_meta_objset,
6297 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
6298 &sav->sav_object, tx) == 0);
6301 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
6302 if (sav->sav_count == 0) {
6303 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
6305 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
6306 for (i = 0; i < sav->sav_count; i++)
6307 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
6308 B_FALSE, VDEV_CONFIG_L2CACHE);
6309 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
6310 sav->sav_count) == 0);
6311 for (i = 0; i < sav->sav_count; i++)
6312 nvlist_free(list[i]);
6313 kmem_free(list, sav->sav_count * sizeof (void *));
6316 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
6317 nvlist_free(nvroot);
6319 sav->sav_sync = B_FALSE;
6323 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
6327 if (list_is_empty(&spa->spa_config_dirty_list))
6330 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6332 config = spa_config_generate(spa, spa->spa_root_vdev,
6333 dmu_tx_get_txg(tx), B_FALSE);
6336 * If we're upgrading the spa version then make sure that
6337 * the config object gets updated with the correct version.
6339 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
6340 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
6341 spa->spa_uberblock.ub_version);
6343 spa_config_exit(spa, SCL_STATE, FTAG);
6345 if (spa->spa_config_syncing)
6346 nvlist_free(spa->spa_config_syncing);
6347 spa->spa_config_syncing = config;
6349 spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
6353 spa_sync_version(void *arg, dmu_tx_t *tx)
6355 uint64_t *versionp = arg;
6356 uint64_t version = *versionp;
6357 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
6360 * Setting the version is special cased when first creating the pool.
6362 ASSERT(tx->tx_txg != TXG_INITIAL);
6364 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
6365 ASSERT(version >= spa_version(spa));
6367 spa->spa_uberblock.ub_version = version;
6368 vdev_config_dirty(spa->spa_root_vdev);
6369 spa_history_log_internal(spa, "set", tx, "version=%lld", version);
6373 * Set zpool properties.
6376 spa_sync_props(void *arg, dmu_tx_t *tx)
6378 nvlist_t *nvp = arg;
6379 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
6380 objset_t *mos = spa->spa_meta_objset;
6381 nvpair_t *elem = NULL;
6383 mutex_enter(&spa->spa_props_lock);
6385 while ((elem = nvlist_next_nvpair(nvp, elem))) {
6387 char *strval, *fname;
6389 const char *propname;
6390 zprop_type_t proptype;
6393 switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
6396 * We checked this earlier in spa_prop_validate().
6398 ASSERT(zpool_prop_feature(nvpair_name(elem)));
6400 fname = strchr(nvpair_name(elem), '@') + 1;
6401 VERIFY0(zfeature_lookup_name(fname, &fid));
6403 spa_feature_enable(spa, fid, tx);
6404 spa_history_log_internal(spa, "set", tx,
6405 "%s=enabled", nvpair_name(elem));
6408 case ZPOOL_PROP_VERSION:
6409 intval = fnvpair_value_uint64(elem);
6411 * The version is synced seperatly before other
6412 * properties and should be correct by now.
6414 ASSERT3U(spa_version(spa), >=, intval);
6417 case ZPOOL_PROP_ALTROOT:
6419 * 'altroot' is a non-persistent property. It should
6420 * have been set temporarily at creation or import time.
6422 ASSERT(spa->spa_root != NULL);
6425 case ZPOOL_PROP_READONLY:
6426 case ZPOOL_PROP_CACHEFILE:
6428 * 'readonly' and 'cachefile' are also non-persisitent
6432 case ZPOOL_PROP_COMMENT:
6433 strval = fnvpair_value_string(elem);
6434 if (spa->spa_comment != NULL)
6435 spa_strfree(spa->spa_comment);
6436 spa->spa_comment = spa_strdup(strval);
6438 * We need to dirty the configuration on all the vdevs
6439 * so that their labels get updated. It's unnecessary
6440 * to do this for pool creation since the vdev's
6441 * configuratoin has already been dirtied.
6443 if (tx->tx_txg != TXG_INITIAL)
6444 vdev_config_dirty(spa->spa_root_vdev);
6445 spa_history_log_internal(spa, "set", tx,
6446 "%s=%s", nvpair_name(elem), strval);
6450 * Set pool property values in the poolprops mos object.
6452 if (spa->spa_pool_props_object == 0) {
6453 spa->spa_pool_props_object =
6454 zap_create_link(mos, DMU_OT_POOL_PROPS,
6455 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
6459 /* normalize the property name */
6460 propname = zpool_prop_to_name(prop);
6461 proptype = zpool_prop_get_type(prop);
6463 if (nvpair_type(elem) == DATA_TYPE_STRING) {
6464 ASSERT(proptype == PROP_TYPE_STRING);
6465 strval = fnvpair_value_string(elem);
6466 VERIFY0(zap_update(mos,
6467 spa->spa_pool_props_object, propname,
6468 1, strlen(strval) + 1, strval, tx));
6469 spa_history_log_internal(spa, "set", tx,
6470 "%s=%s", nvpair_name(elem), strval);
6471 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
6472 intval = fnvpair_value_uint64(elem);
6474 if (proptype == PROP_TYPE_INDEX) {
6476 VERIFY0(zpool_prop_index_to_string(
6477 prop, intval, &unused));
6479 VERIFY0(zap_update(mos,
6480 spa->spa_pool_props_object, propname,
6481 8, 1, &intval, tx));
6482 spa_history_log_internal(spa, "set", tx,
6483 "%s=%lld", nvpair_name(elem), intval);
6485 ASSERT(0); /* not allowed */
6489 case ZPOOL_PROP_DELEGATION:
6490 spa->spa_delegation = intval;
6492 case ZPOOL_PROP_BOOTFS:
6493 spa->spa_bootfs = intval;
6495 case ZPOOL_PROP_FAILUREMODE:
6496 spa->spa_failmode = intval;
6498 case ZPOOL_PROP_AUTOEXPAND:
6499 spa->spa_autoexpand = intval;
6500 if (tx->tx_txg != TXG_INITIAL)
6501 spa_async_request(spa,
6502 SPA_ASYNC_AUTOEXPAND);
6504 case ZPOOL_PROP_DEDUPDITTO:
6505 spa->spa_dedup_ditto = intval;
6514 mutex_exit(&spa->spa_props_lock);
6518 * Perform one-time upgrade on-disk changes. spa_version() does not
6519 * reflect the new version this txg, so there must be no changes this
6520 * txg to anything that the upgrade code depends on after it executes.
6521 * Therefore this must be called after dsl_pool_sync() does the sync
6525 spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
6527 dsl_pool_t *dp = spa->spa_dsl_pool;
6529 ASSERT(spa->spa_sync_pass == 1);
6531 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
6533 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
6534 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
6535 dsl_pool_create_origin(dp, tx);
6537 /* Keeping the origin open increases spa_minref */
6538 spa->spa_minref += 3;
6541 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
6542 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
6543 dsl_pool_upgrade_clones(dp, tx);
6546 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
6547 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
6548 dsl_pool_upgrade_dir_clones(dp, tx);
6550 /* Keeping the freedir open increases spa_minref */
6551 spa->spa_minref += 3;
6554 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES &&
6555 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
6556 spa_feature_create_zap_objects(spa, tx);
6560 * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable
6561 * when possibility to use lz4 compression for metadata was added
6562 * Old pools that have this feature enabled must be upgraded to have
6563 * this feature active
6565 if (spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
6566 boolean_t lz4_en = spa_feature_is_enabled(spa,
6567 SPA_FEATURE_LZ4_COMPRESS);
6568 boolean_t lz4_ac = spa_feature_is_active(spa,
6569 SPA_FEATURE_LZ4_COMPRESS);
6571 if (lz4_en && !lz4_ac)
6572 spa_feature_incr(spa, SPA_FEATURE_LZ4_COMPRESS, tx);
6576 * If we haven't written the salt, do so now. Note that the
6577 * feature may not be activated yet, but that's fine since
6578 * the presence of this ZAP entry is backwards compatible.
6580 if (zap_contains(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
6581 DMU_POOL_CHECKSUM_SALT) == ENOENT) {
6582 VERIFY0(zap_add(spa->spa_meta_objset,
6583 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CHECKSUM_SALT, 1,
6584 sizeof (spa->spa_cksum_salt.zcs_bytes),
6585 spa->spa_cksum_salt.zcs_bytes, tx));
6588 rrw_exit(&dp->dp_config_rwlock, FTAG);
6592 * Sync the specified transaction group. New blocks may be dirtied as
6593 * part of the process, so we iterate until it converges.
6596 spa_sync(spa_t *spa, uint64_t txg)
6598 dsl_pool_t *dp = spa->spa_dsl_pool;
6599 objset_t *mos = spa->spa_meta_objset;
6600 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
6601 vdev_t *rvd = spa->spa_root_vdev;
6606 VERIFY(spa_writeable(spa));
6609 * Lock out configuration changes.
6611 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
6613 spa->spa_syncing_txg = txg;
6614 spa->spa_sync_pass = 0;
6617 * If there are any pending vdev state changes, convert them
6618 * into config changes that go out with this transaction group.
6620 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6621 while (list_head(&spa->spa_state_dirty_list) != NULL) {
6623 * We need the write lock here because, for aux vdevs,
6624 * calling vdev_config_dirty() modifies sav_config.
6625 * This is ugly and will become unnecessary when we
6626 * eliminate the aux vdev wart by integrating all vdevs
6627 * into the root vdev tree.
6629 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6630 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
6631 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
6632 vdev_state_clean(vd);
6633 vdev_config_dirty(vd);
6635 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6636 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
6638 spa_config_exit(spa, SCL_STATE, FTAG);
6640 tx = dmu_tx_create_assigned(dp, txg);
6642 spa->spa_sync_starttime = gethrtime();
6644 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid,
6645 spa->spa_sync_starttime + spa->spa_deadman_synctime));
6648 callout_reset(&spa->spa_deadman_cycid,
6649 hz * spa->spa_deadman_synctime / NANOSEC, spa_deadman, spa);
6654 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
6655 * set spa_deflate if we have no raid-z vdevs.
6657 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
6658 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
6661 for (i = 0; i < rvd->vdev_children; i++) {
6662 vd = rvd->vdev_child[i];
6663 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
6666 if (i == rvd->vdev_children) {
6667 spa->spa_deflate = TRUE;
6668 VERIFY(0 == zap_add(spa->spa_meta_objset,
6669 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
6670 sizeof (uint64_t), 1, &spa->spa_deflate, tx));
6675 * Iterate to convergence.
6678 int pass = ++spa->spa_sync_pass;
6680 spa_sync_config_object(spa, tx);
6681 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
6682 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
6683 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
6684 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
6685 spa_errlog_sync(spa, txg);
6686 dsl_pool_sync(dp, txg);
6688 if (pass < zfs_sync_pass_deferred_free) {
6689 spa_sync_frees(spa, free_bpl, tx);
6692 * We can not defer frees in pass 1, because
6693 * we sync the deferred frees later in pass 1.
6695 ASSERT3U(pass, >, 1);
6696 bplist_iterate(free_bpl, bpobj_enqueue_cb,
6697 &spa->spa_deferred_bpobj, tx);
6701 dsl_scan_sync(dp, tx);
6703 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
6707 spa_sync_upgrades(spa, tx);
6709 spa->spa_uberblock.ub_rootbp.blk_birth);
6711 * Note: We need to check if the MOS is dirty
6712 * because we could have marked the MOS dirty
6713 * without updating the uberblock (e.g. if we
6714 * have sync tasks but no dirty user data). We
6715 * need to check the uberblock's rootbp because
6716 * it is updated if we have synced out dirty
6717 * data (though in this case the MOS will most
6718 * likely also be dirty due to second order
6719 * effects, we don't want to rely on that here).
6721 if (spa->spa_uberblock.ub_rootbp.blk_birth < txg &&
6722 !dmu_objset_is_dirty(mos, txg)) {
6724 * Nothing changed on the first pass,
6725 * therefore this TXG is a no-op. Avoid
6726 * syncing deferred frees, so that we
6727 * can keep this TXG as a no-op.
6729 ASSERT(txg_list_empty(&dp->dp_dirty_datasets,
6731 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
6732 ASSERT(txg_list_empty(&dp->dp_sync_tasks, txg));
6735 spa_sync_deferred_frees(spa, tx);
6738 } while (dmu_objset_is_dirty(mos, txg));
6741 * Rewrite the vdev configuration (which includes the uberblock)
6742 * to commit the transaction group.
6744 * If there are no dirty vdevs, we sync the uberblock to a few
6745 * random top-level vdevs that are known to be visible in the
6746 * config cache (see spa_vdev_add() for a complete description).
6747 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
6751 * We hold SCL_STATE to prevent vdev open/close/etc.
6752 * while we're attempting to write the vdev labels.
6754 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6756 if (list_is_empty(&spa->spa_config_dirty_list)) {
6757 vdev_t *svd[SPA_DVAS_PER_BP];
6759 int children = rvd->vdev_children;
6760 int c0 = spa_get_random(children);
6762 for (int c = 0; c < children; c++) {
6763 vd = rvd->vdev_child[(c0 + c) % children];
6764 if (vd->vdev_ms_array == 0 || vd->vdev_islog)
6766 svd[svdcount++] = vd;
6767 if (svdcount == SPA_DVAS_PER_BP)
6770 error = vdev_config_sync(svd, svdcount, txg, B_FALSE);
6772 error = vdev_config_sync(svd, svdcount, txg,
6775 error = vdev_config_sync(rvd->vdev_child,
6776 rvd->vdev_children, txg, B_FALSE);
6778 error = vdev_config_sync(rvd->vdev_child,
6779 rvd->vdev_children, txg, B_TRUE);
6783 spa->spa_last_synced_guid = rvd->vdev_guid;
6785 spa_config_exit(spa, SCL_STATE, FTAG);
6789 zio_suspend(spa, NULL);
6790 zio_resume_wait(spa);
6795 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY));
6798 callout_drain(&spa->spa_deadman_cycid);
6803 * Clear the dirty config list.
6805 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
6806 vdev_config_clean(vd);
6809 * Now that the new config has synced transactionally,
6810 * let it become visible to the config cache.
6812 if (spa->spa_config_syncing != NULL) {
6813 spa_config_set(spa, spa->spa_config_syncing);
6814 spa->spa_config_txg = txg;
6815 spa->spa_config_syncing = NULL;
6818 spa->spa_ubsync = spa->spa_uberblock;
6820 dsl_pool_sync_done(dp, txg);
6823 * Update usable space statistics.
6825 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
6826 vdev_sync_done(vd, txg);
6828 spa_update_dspace(spa);
6831 * It had better be the case that we didn't dirty anything
6832 * since vdev_config_sync().
6834 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
6835 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
6836 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
6838 spa->spa_sync_pass = 0;
6840 spa_config_exit(spa, SCL_CONFIG, FTAG);
6842 spa_handle_ignored_writes(spa);
6845 * If any async tasks have been requested, kick them off.
6847 spa_async_dispatch(spa);
6848 spa_async_dispatch_vd(spa);
6852 * Sync all pools. We don't want to hold the namespace lock across these
6853 * operations, so we take a reference on the spa_t and drop the lock during the
6857 spa_sync_allpools(void)
6860 mutex_enter(&spa_namespace_lock);
6861 while ((spa = spa_next(spa)) != NULL) {
6862 if (spa_state(spa) != POOL_STATE_ACTIVE ||
6863 !spa_writeable(spa) || spa_suspended(spa))
6865 spa_open_ref(spa, FTAG);
6866 mutex_exit(&spa_namespace_lock);
6867 txg_wait_synced(spa_get_dsl(spa), 0);
6868 mutex_enter(&spa_namespace_lock);
6869 spa_close(spa, FTAG);
6871 mutex_exit(&spa_namespace_lock);
6875 * ==========================================================================
6876 * Miscellaneous routines
6877 * ==========================================================================
6881 * Remove all pools in the system.
6889 * Remove all cached state. All pools should be closed now,
6890 * so every spa in the AVL tree should be unreferenced.
6892 mutex_enter(&spa_namespace_lock);
6893 while ((spa = spa_next(NULL)) != NULL) {
6895 * Stop async tasks. The async thread may need to detach
6896 * a device that's been replaced, which requires grabbing
6897 * spa_namespace_lock, so we must drop it here.
6899 spa_open_ref(spa, FTAG);
6900 mutex_exit(&spa_namespace_lock);
6901 spa_async_suspend(spa);
6902 mutex_enter(&spa_namespace_lock);
6903 spa_close(spa, FTAG);
6905 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
6907 spa_deactivate(spa);
6911 mutex_exit(&spa_namespace_lock);
6915 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
6920 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
6924 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
6925 vd = spa->spa_l2cache.sav_vdevs[i];
6926 if (vd->vdev_guid == guid)
6930 for (i = 0; i < spa->spa_spares.sav_count; i++) {
6931 vd = spa->spa_spares.sav_vdevs[i];
6932 if (vd->vdev_guid == guid)
6941 spa_upgrade(spa_t *spa, uint64_t version)
6943 ASSERT(spa_writeable(spa));
6945 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6948 * This should only be called for a non-faulted pool, and since a
6949 * future version would result in an unopenable pool, this shouldn't be
6952 ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version));
6953 ASSERT3U(version, >=, spa->spa_uberblock.ub_version);
6955 spa->spa_uberblock.ub_version = version;
6956 vdev_config_dirty(spa->spa_root_vdev);
6958 spa_config_exit(spa, SCL_ALL, FTAG);
6960 txg_wait_synced(spa_get_dsl(spa), 0);
6964 spa_has_spare(spa_t *spa, uint64_t guid)
6968 spa_aux_vdev_t *sav = &spa->spa_spares;
6970 for (i = 0; i < sav->sav_count; i++)
6971 if (sav->sav_vdevs[i]->vdev_guid == guid)
6974 for (i = 0; i < sav->sav_npending; i++) {
6975 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
6976 &spareguid) == 0 && spareguid == guid)
6984 * Check if a pool has an active shared spare device.
6985 * Note: reference count of an active spare is 2, as a spare and as a replace
6988 spa_has_active_shared_spare(spa_t *spa)
6992 spa_aux_vdev_t *sav = &spa->spa_spares;
6994 for (i = 0; i < sav->sav_count; i++) {
6995 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
6996 &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
7005 * Post a sysevent corresponding to the given event. The 'name' must be one of
7006 * the event definitions in sys/sysevent/eventdefs.h. The payload will be
7007 * filled in from the spa and (optionally) the vdev. This doesn't do anything
7008 * in the userland libzpool, as we don't want consumers to misinterpret ztest
7009 * or zdb as real changes.
7012 spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
7016 sysevent_attr_list_t *attr = NULL;
7017 sysevent_value_t value;
7020 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
7023 value.value_type = SE_DATA_TYPE_STRING;
7024 value.value.sv_string = spa_name(spa);
7025 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
7028 value.value_type = SE_DATA_TYPE_UINT64;
7029 value.value.sv_uint64 = spa_guid(spa);
7030 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
7034 value.value_type = SE_DATA_TYPE_UINT64;
7035 value.value.sv_uint64 = vd->vdev_guid;
7036 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
7040 if (vd->vdev_path) {
7041 value.value_type = SE_DATA_TYPE_STRING;
7042 value.value.sv_string = vd->vdev_path;
7043 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
7044 &value, SE_SLEEP) != 0)
7049 if (sysevent_attach_attributes(ev, attr) != 0)
7053 (void) log_sysevent(ev, SE_SLEEP, &eid);
7057 sysevent_free_attr(attr);