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 2011 Nexenta Systems, Inc. All rights reserved.
25 * Copyright (c) 2012 by Delphix. All rights reserved.
29 * This file contains all the routines used when modifying on-disk SPA state.
30 * This includes opening, importing, destroying, exporting a pool, and syncing a
34 #include <sys/zfs_context.h>
35 #include <sys/fm/fs/zfs.h>
36 #include <sys/spa_impl.h>
38 #include <sys/zio_checksum.h>
40 #include <sys/dmu_tx.h>
44 #include <sys/vdev_impl.h>
45 #include <sys/vdev_disk.h>
46 #include <sys/metaslab.h>
47 #include <sys/metaslab_impl.h>
48 #include <sys/uberblock_impl.h>
51 #include <sys/dmu_traverse.h>
52 #include <sys/dmu_objset.h>
53 #include <sys/unique.h>
54 #include <sys/dsl_pool.h>
55 #include <sys/dsl_dataset.h>
56 #include <sys/dsl_dir.h>
57 #include <sys/dsl_prop.h>
58 #include <sys/dsl_synctask.h>
59 #include <sys/fs/zfs.h>
61 #include <sys/callb.h>
62 #include <sys/systeminfo.h>
63 #include <sys/spa_boot.h>
64 #include <sys/zfs_ioctl.h>
65 #include <sys/dsl_scan.h>
66 #include <sys/zfeature.h>
69 #include <sys/bootprops.h>
70 #include <sys/callb.h>
71 #include <sys/cpupart.h>
73 #include <sys/sysdc.h>
78 #include "zfs_comutil.h"
80 typedef enum zti_modes {
81 ZTI_MODE_FIXED, /* value is # of threads (min 1) */
82 ZTI_MODE_ONLINE_PERCENT, /* value is % of online CPUs */
83 ZTI_MODE_BATCH, /* cpu-intensive; value is ignored */
84 ZTI_MODE_NULL, /* don't create a taskq */
88 #define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) }
89 #define ZTI_PCT(n) { ZTI_MODE_ONLINE_PERCENT, (n), 1 }
90 #define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 }
91 #define ZTI_NULL { ZTI_MODE_NULL, 0, 0 }
93 #define ZTI_N(n) ZTI_P(n, 1)
94 #define ZTI_ONE ZTI_N(1)
96 typedef struct zio_taskq_info {
102 static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
103 "iss", "iss_h", "int", "int_h"
107 * This table defines the taskq settings for each ZFS I/O type. When
108 * initializing a pool, we use this table to create an appropriately sized
109 * taskq. Some operations are low volume and therefore have a small, static
110 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE
111 * macros. Other operations process a large amount of data; the ZTI_BATCH
112 * macro causes us to create a taskq oriented for throughput. Some operations
113 * are so high frequency and short-lived that the taskq itself can become a a
114 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an
115 * additional degree of parallelism specified by the number of threads per-
116 * taskq and the number of taskqs; when dispatching an event in this case, the
117 * particular taskq is chosen at random.
119 * The different taskq priorities are to handle the different contexts (issue
120 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that
121 * need to be handled with minimum delay.
123 const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
124 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
125 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* NULL */
126 { ZTI_N(8), ZTI_NULL, ZTI_BATCH, ZTI_NULL }, /* READ */
127 { ZTI_BATCH, ZTI_N(5), ZTI_N(16), ZTI_N(5) }, /* WRITE */
128 { ZTI_P(4, 8), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FREE */
129 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* CLAIM */
130 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* IOCTL */
133 static dsl_syncfunc_t spa_sync_version;
134 static dsl_syncfunc_t spa_sync_props;
135 static dsl_checkfunc_t spa_change_guid_check;
136 static dsl_syncfunc_t spa_change_guid_sync;
137 static boolean_t spa_has_active_shared_spare(spa_t *spa);
138 static inline int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
139 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
141 static void spa_vdev_resilver_done(spa_t *spa);
143 uint_t zio_taskq_batch_pct = 100; /* 1 thread per cpu in pset */
144 id_t zio_taskq_psrset_bind = PS_NONE;
145 boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */
146 uint_t zio_taskq_basedc = 80; /* base duty cycle */
148 boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */
151 * This (illegal) pool name is used when temporarily importing a spa_t in order
152 * to get the vdev stats associated with the imported devices.
154 #define TRYIMPORT_NAME "$import"
157 * ==========================================================================
158 * SPA properties routines
159 * ==========================================================================
163 * Add a (source=src, propname=propval) list to an nvlist.
166 spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
167 uint64_t intval, zprop_source_t src)
169 const char *propname = zpool_prop_to_name(prop);
172 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_PUSHPAGE) == 0);
173 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
176 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
178 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
180 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
181 nvlist_free(propval);
185 * Get property values from the spa configuration.
188 spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
190 vdev_t *rvd = spa->spa_root_vdev;
191 dsl_pool_t *pool = spa->spa_dsl_pool;
195 uint64_t cap, version;
196 zprop_source_t src = ZPROP_SRC_NONE;
197 spa_config_dirent_t *dp;
200 ASSERT(MUTEX_HELD(&spa->spa_props_lock));
203 alloc = metaslab_class_get_alloc(spa_normal_class(spa));
204 size = metaslab_class_get_space(spa_normal_class(spa));
205 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
206 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
207 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
208 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
212 for (c = 0; c < rvd->vdev_children; c++) {
213 vdev_t *tvd = rvd->vdev_child[c];
214 space += tvd->vdev_max_asize - tvd->vdev_asize;
216 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, space,
219 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
220 (spa_mode(spa) == FREAD), src);
222 cap = (size == 0) ? 0 : (alloc * 100 / size);
223 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
225 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
226 ddt_get_pool_dedup_ratio(spa), src);
228 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
229 rvd->vdev_state, src);
231 version = spa_version(spa);
232 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
233 src = ZPROP_SRC_DEFAULT;
235 src = ZPROP_SRC_LOCAL;
236 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
240 dsl_dir_t *freedir = pool->dp_free_dir;
243 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
244 * when opening pools before this version freedir will be NULL.
246 if (freedir != NULL) {
247 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
248 freedir->dd_phys->dd_used_bytes, src);
250 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
255 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
257 if (spa->spa_comment != NULL) {
258 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
262 if (spa->spa_root != NULL)
263 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
266 if ((dp = list_head(&spa->spa_config_list)) != NULL) {
267 if (dp->scd_path == NULL) {
268 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
269 "none", 0, ZPROP_SRC_LOCAL);
270 } else if (strcmp(dp->scd_path, spa_config_path) != 0) {
271 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
272 dp->scd_path, 0, ZPROP_SRC_LOCAL);
278 * Get zpool property values.
281 spa_prop_get(spa_t *spa, nvlist_t **nvp)
283 objset_t *mos = spa->spa_meta_objset;
288 err = nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_PUSHPAGE);
292 mutex_enter(&spa->spa_props_lock);
295 * Get properties from the spa config.
297 spa_prop_get_config(spa, nvp);
299 /* If no pool property object, no more prop to get. */
300 if (mos == NULL || spa->spa_pool_props_object == 0) {
301 mutex_exit(&spa->spa_props_lock);
306 * Get properties from the MOS pool property object.
308 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
309 (err = zap_cursor_retrieve(&zc, &za)) == 0;
310 zap_cursor_advance(&zc)) {
313 zprop_source_t src = ZPROP_SRC_DEFAULT;
316 if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
319 switch (za.za_integer_length) {
321 /* integer property */
322 if (za.za_first_integer !=
323 zpool_prop_default_numeric(prop))
324 src = ZPROP_SRC_LOCAL;
326 if (prop == ZPOOL_PROP_BOOTFS) {
328 dsl_dataset_t *ds = NULL;
330 dp = spa_get_dsl(spa);
331 rw_enter(&dp->dp_config_rwlock, RW_READER);
332 if ((err = dsl_dataset_hold_obj(dp,
333 za.za_first_integer, FTAG, &ds))) {
334 rw_exit(&dp->dp_config_rwlock);
339 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
341 dsl_dataset_name(ds, strval);
342 dsl_dataset_rele(ds, FTAG);
343 rw_exit(&dp->dp_config_rwlock);
346 intval = za.za_first_integer;
349 spa_prop_add_list(*nvp, prop, strval, intval, src);
353 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
358 /* string property */
359 strval = kmem_alloc(za.za_num_integers, KM_PUSHPAGE);
360 err = zap_lookup(mos, spa->spa_pool_props_object,
361 za.za_name, 1, za.za_num_integers, strval);
363 kmem_free(strval, za.za_num_integers);
366 spa_prop_add_list(*nvp, prop, strval, 0, src);
367 kmem_free(strval, za.za_num_integers);
374 zap_cursor_fini(&zc);
375 mutex_exit(&spa->spa_props_lock);
377 if (err && err != ENOENT) {
387 * Validate the given pool properties nvlist and modify the list
388 * for the property values to be set.
391 spa_prop_validate(spa_t *spa, nvlist_t *props)
394 int error = 0, reset_bootfs = 0;
396 boolean_t has_feature = B_FALSE;
399 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
401 char *strval, *slash, *check, *fname;
402 const char *propname = nvpair_name(elem);
403 zpool_prop_t prop = zpool_name_to_prop(propname);
407 if (!zpool_prop_feature(propname)) {
413 * Sanitize the input.
415 if (nvpair_type(elem) != DATA_TYPE_UINT64) {
420 if (nvpair_value_uint64(elem, &intval) != 0) {
430 fname = strchr(propname, '@') + 1;
431 if (zfeature_lookup_name(fname, NULL) != 0) {
436 has_feature = B_TRUE;
439 case ZPOOL_PROP_VERSION:
440 error = nvpair_value_uint64(elem, &intval);
442 (intval < spa_version(spa) ||
443 intval > SPA_VERSION_BEFORE_FEATURES ||
448 case ZPOOL_PROP_DELEGATION:
449 case ZPOOL_PROP_AUTOREPLACE:
450 case ZPOOL_PROP_LISTSNAPS:
451 case ZPOOL_PROP_AUTOEXPAND:
452 error = nvpair_value_uint64(elem, &intval);
453 if (!error && intval > 1)
457 case ZPOOL_PROP_BOOTFS:
459 * If the pool version is less than SPA_VERSION_BOOTFS,
460 * or the pool is still being created (version == 0),
461 * the bootfs property cannot be set.
463 if (spa_version(spa) < SPA_VERSION_BOOTFS) {
469 * Make sure the vdev config is bootable
471 if (!vdev_is_bootable(spa->spa_root_vdev)) {
478 error = nvpair_value_string(elem, &strval);
484 if (strval == NULL || strval[0] == '\0') {
485 objnum = zpool_prop_default_numeric(
490 if ((error = dmu_objset_hold(strval,FTAG,&os)))
493 /* Must be ZPL and not gzip compressed. */
495 if (dmu_objset_type(os) != DMU_OST_ZFS) {
497 } else if ((error = dsl_prop_get_integer(strval,
498 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
499 &compress, NULL)) == 0 &&
500 !BOOTFS_COMPRESS_VALID(compress)) {
503 objnum = dmu_objset_id(os);
505 dmu_objset_rele(os, FTAG);
509 case ZPOOL_PROP_FAILUREMODE:
510 error = nvpair_value_uint64(elem, &intval);
511 if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
512 intval > ZIO_FAILURE_MODE_PANIC))
516 * This is a special case which only occurs when
517 * the pool has completely failed. This allows
518 * the user to change the in-core failmode property
519 * without syncing it out to disk (I/Os might
520 * currently be blocked). We do this by returning
521 * EIO to the caller (spa_prop_set) to trick it
522 * into thinking we encountered a property validation
525 if (!error && spa_suspended(spa)) {
526 spa->spa_failmode = intval;
531 case ZPOOL_PROP_CACHEFILE:
532 if ((error = nvpair_value_string(elem, &strval)) != 0)
535 if (strval[0] == '\0')
538 if (strcmp(strval, "none") == 0)
541 if (strval[0] != '/') {
546 slash = strrchr(strval, '/');
547 ASSERT(slash != NULL);
549 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
550 strcmp(slash, "/..") == 0)
554 case ZPOOL_PROP_COMMENT:
555 if ((error = nvpair_value_string(elem, &strval)) != 0)
557 for (check = strval; *check != '\0'; check++) {
558 if (!isprint(*check)) {
564 if (strlen(strval) > ZPROP_MAX_COMMENT)
568 case ZPOOL_PROP_DEDUPDITTO:
569 if (spa_version(spa) < SPA_VERSION_DEDUP)
572 error = nvpair_value_uint64(elem, &intval);
574 intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
586 if (!error && reset_bootfs) {
587 error = nvlist_remove(props,
588 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
591 error = nvlist_add_uint64(props,
592 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
600 spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
603 spa_config_dirent_t *dp;
605 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
609 dp = kmem_alloc(sizeof (spa_config_dirent_t),
612 if (cachefile[0] == '\0')
613 dp->scd_path = spa_strdup(spa_config_path);
614 else if (strcmp(cachefile, "none") == 0)
617 dp->scd_path = spa_strdup(cachefile);
619 list_insert_head(&spa->spa_config_list, dp);
621 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
625 spa_prop_set(spa_t *spa, nvlist_t *nvp)
628 nvpair_t *elem = NULL;
629 boolean_t need_sync = B_FALSE;
631 if ((error = spa_prop_validate(spa, nvp)) != 0)
634 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
635 zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem));
637 if (prop == ZPOOL_PROP_CACHEFILE ||
638 prop == ZPOOL_PROP_ALTROOT ||
639 prop == ZPOOL_PROP_READONLY)
642 if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) {
645 if (prop == ZPOOL_PROP_VERSION) {
646 VERIFY(nvpair_value_uint64(elem, &ver) == 0);
648 ASSERT(zpool_prop_feature(nvpair_name(elem)));
649 ver = SPA_VERSION_FEATURES;
653 /* Save time if the version is already set. */
654 if (ver == spa_version(spa))
658 * In addition to the pool directory object, we might
659 * create the pool properties object, the features for
660 * read object, the features for write object, or the
661 * feature descriptions object.
663 error = dsl_sync_task_do(spa_get_dsl(spa), NULL,
664 spa_sync_version, spa, &ver, 6);
675 return (dsl_sync_task_do(spa_get_dsl(spa), NULL, spa_sync_props,
683 * If the bootfs property value is dsobj, clear it.
686 spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
688 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
689 VERIFY(zap_remove(spa->spa_meta_objset,
690 spa->spa_pool_props_object,
691 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
698 spa_change_guid_check(void *arg1, void *arg2, dmu_tx_t *tx)
701 vdev_t *rvd = spa->spa_root_vdev;
703 ASSERTV(uint64_t *newguid = arg2);
705 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
706 vdev_state = rvd->vdev_state;
707 spa_config_exit(spa, SCL_STATE, FTAG);
709 if (vdev_state != VDEV_STATE_HEALTHY)
712 ASSERT3U(spa_guid(spa), !=, *newguid);
718 spa_change_guid_sync(void *arg1, void *arg2, dmu_tx_t *tx)
721 uint64_t *newguid = arg2;
723 vdev_t *rvd = spa->spa_root_vdev;
725 oldguid = spa_guid(spa);
727 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
728 rvd->vdev_guid = *newguid;
729 rvd->vdev_guid_sum += (*newguid - oldguid);
730 vdev_config_dirty(rvd);
731 spa_config_exit(spa, SCL_STATE, FTAG);
733 spa_history_log_internal(LOG_POOL_GUID_CHANGE, spa, tx,
734 "old=%lld new=%lld", oldguid, *newguid);
738 * Change the GUID for the pool. This is done so that we can later
739 * re-import a pool built from a clone of our own vdevs. We will modify
740 * the root vdev's guid, our own pool guid, and then mark all of our
741 * vdevs dirty. Note that we must make sure that all our vdevs are
742 * online when we do this, or else any vdevs that weren't present
743 * would be orphaned from our pool. We are also going to issue a
744 * sysevent to update any watchers.
747 spa_change_guid(spa_t *spa)
752 mutex_enter(&spa_namespace_lock);
753 guid = spa_generate_guid(NULL);
755 error = dsl_sync_task_do(spa_get_dsl(spa), spa_change_guid_check,
756 spa_change_guid_sync, spa, &guid, 5);
759 spa_config_sync(spa, B_FALSE, B_TRUE);
760 spa_event_notify(spa, NULL, FM_EREPORT_ZFS_POOL_REGUID);
763 mutex_exit(&spa_namespace_lock);
769 * ==========================================================================
770 * SPA state manipulation (open/create/destroy/import/export)
771 * ==========================================================================
775 spa_error_entry_compare(const void *a, const void *b)
777 spa_error_entry_t *sa = (spa_error_entry_t *)a;
778 spa_error_entry_t *sb = (spa_error_entry_t *)b;
781 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
782 sizeof (zbookmark_t));
793 * Utility function which retrieves copies of the current logs and
794 * re-initializes them in the process.
797 spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
799 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
801 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
802 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
804 avl_create(&spa->spa_errlist_scrub,
805 spa_error_entry_compare, sizeof (spa_error_entry_t),
806 offsetof(spa_error_entry_t, se_avl));
807 avl_create(&spa->spa_errlist_last,
808 spa_error_entry_compare, sizeof (spa_error_entry_t),
809 offsetof(spa_error_entry_t, se_avl));
813 spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
815 const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
816 enum zti_modes mode = ztip->zti_mode;
817 uint_t value = ztip->zti_value;
818 uint_t count = ztip->zti_count;
819 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
822 boolean_t batch = B_FALSE;
824 if (mode == ZTI_MODE_NULL) {
826 tqs->stqs_taskq = NULL;
830 ASSERT3U(count, >, 0);
832 tqs->stqs_count = count;
833 tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP);
835 for (i = 0; i < count; i++) {
840 ASSERT3U(value, >=, 1);
841 value = MAX(value, 1);
846 flags |= TASKQ_THREADS_CPU_PCT;
847 value = zio_taskq_batch_pct;
850 case ZTI_MODE_ONLINE_PERCENT:
851 flags |= TASKQ_THREADS_CPU_PCT;
855 panic("unrecognized mode for %s_%s taskq (%u:%u) in "
857 zio_type_name[t], zio_taskq_types[q], mode, value);
862 (void) snprintf(name, sizeof (name), "%s_%s_%u",
863 zio_type_name[t], zio_taskq_types[q], i);
865 (void) snprintf(name, sizeof (name), "%s_%s",
866 zio_type_name[t], zio_taskq_types[q]);
869 if (zio_taskq_sysdc && spa->spa_proc != &p0) {
871 flags |= TASKQ_DC_BATCH;
873 tq = taskq_create_sysdc(name, value, 50, INT_MAX,
874 spa->spa_proc, zio_taskq_basedc, flags);
876 tq = taskq_create_proc(name, value, maxclsyspri, 50,
877 INT_MAX, spa->spa_proc, flags);
880 tqs->stqs_taskq[i] = tq;
885 spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
887 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
890 if (tqs->stqs_taskq == NULL) {
891 ASSERT3U(tqs->stqs_count, ==, 0);
895 for (i = 0; i < tqs->stqs_count; i++) {
896 ASSERT3P(tqs->stqs_taskq[i], !=, NULL);
897 taskq_destroy(tqs->stqs_taskq[i]);
900 kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *));
901 tqs->stqs_taskq = NULL;
905 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
906 * Note that a type may have multiple discrete taskqs to avoid lock contention
907 * on the taskq itself. In that case we choose which taskq at random by using
908 * the low bits of gethrtime().
911 spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q,
912 task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent)
914 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
917 ASSERT3P(tqs->stqs_taskq, !=, NULL);
918 ASSERT3U(tqs->stqs_count, !=, 0);
920 if (tqs->stqs_count == 1) {
921 tq = tqs->stqs_taskq[0];
923 tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count];
926 taskq_dispatch_ent(tq, func, arg, flags, ent);
930 * Same as spa_taskq_dispatch_ent() but block on the task until completion.
933 spa_taskq_dispatch_sync(spa_t *spa, zio_type_t t, zio_taskq_type_t q,
934 task_func_t *func, void *arg, uint_t flags)
936 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
940 ASSERT3P(tqs->stqs_taskq, !=, NULL);
941 ASSERT3U(tqs->stqs_count, !=, 0);
943 if (tqs->stqs_count == 1) {
944 tq = tqs->stqs_taskq[0];
946 tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count];
949 id = taskq_dispatch(tq, func, arg, flags);
951 taskq_wait_id(tq, id);
955 spa_create_zio_taskqs(spa_t *spa)
959 for (t = 0; t < ZIO_TYPES; t++) {
960 for (q = 0; q < ZIO_TASKQ_TYPES; q++) {
961 spa_taskqs_init(spa, t, q);
966 #if defined(_KERNEL) && defined(HAVE_SPA_THREAD)
968 spa_thread(void *arg)
973 user_t *pu = PTOU(curproc);
975 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
978 ASSERT(curproc != &p0);
979 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
980 "zpool-%s", spa->spa_name);
981 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
983 /* bind this thread to the requested psrset */
984 if (zio_taskq_psrset_bind != PS_NONE) {
986 mutex_enter(&cpu_lock);
987 mutex_enter(&pidlock);
988 mutex_enter(&curproc->p_lock);
990 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
991 0, NULL, NULL) == 0) {
992 curthread->t_bind_pset = zio_taskq_psrset_bind;
995 "Couldn't bind process for zfs pool \"%s\" to "
996 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
999 mutex_exit(&curproc->p_lock);
1000 mutex_exit(&pidlock);
1001 mutex_exit(&cpu_lock);
1005 if (zio_taskq_sysdc) {
1006 sysdc_thread_enter(curthread, 100, 0);
1009 spa->spa_proc = curproc;
1010 spa->spa_did = curthread->t_did;
1012 spa_create_zio_taskqs(spa);
1014 mutex_enter(&spa->spa_proc_lock);
1015 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
1017 spa->spa_proc_state = SPA_PROC_ACTIVE;
1018 cv_broadcast(&spa->spa_proc_cv);
1020 CALLB_CPR_SAFE_BEGIN(&cprinfo);
1021 while (spa->spa_proc_state == SPA_PROC_ACTIVE)
1022 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1023 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
1025 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
1026 spa->spa_proc_state = SPA_PROC_GONE;
1027 spa->spa_proc = &p0;
1028 cv_broadcast(&spa->spa_proc_cv);
1029 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */
1031 mutex_enter(&curproc->p_lock);
1037 * Activate an uninitialized pool.
1040 spa_activate(spa_t *spa, int mode)
1042 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
1044 spa->spa_state = POOL_STATE_ACTIVE;
1045 spa->spa_mode = mode;
1047 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
1048 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
1050 /* Try to create a covering process */
1051 mutex_enter(&spa->spa_proc_lock);
1052 ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
1053 ASSERT(spa->spa_proc == &p0);
1056 #ifdef HAVE_SPA_THREAD
1057 /* Only create a process if we're going to be around a while. */
1058 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
1059 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
1061 spa->spa_proc_state = SPA_PROC_CREATED;
1062 while (spa->spa_proc_state == SPA_PROC_CREATED) {
1063 cv_wait(&spa->spa_proc_cv,
1064 &spa->spa_proc_lock);
1066 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1067 ASSERT(spa->spa_proc != &p0);
1068 ASSERT(spa->spa_did != 0);
1072 "Couldn't create process for zfs pool \"%s\"\n",
1077 #endif /* HAVE_SPA_THREAD */
1078 mutex_exit(&spa->spa_proc_lock);
1080 /* If we didn't create a process, we need to create our taskqs. */
1081 if (spa->spa_proc == &p0) {
1082 spa_create_zio_taskqs(spa);
1085 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
1086 offsetof(vdev_t, vdev_config_dirty_node));
1087 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
1088 offsetof(vdev_t, vdev_state_dirty_node));
1090 txg_list_create(&spa->spa_vdev_txg_list,
1091 offsetof(struct vdev, vdev_txg_node));
1093 avl_create(&spa->spa_errlist_scrub,
1094 spa_error_entry_compare, sizeof (spa_error_entry_t),
1095 offsetof(spa_error_entry_t, se_avl));
1096 avl_create(&spa->spa_errlist_last,
1097 spa_error_entry_compare, sizeof (spa_error_entry_t),
1098 offsetof(spa_error_entry_t, se_avl));
1102 * Opposite of spa_activate().
1105 spa_deactivate(spa_t *spa)
1109 ASSERT(spa->spa_sync_on == B_FALSE);
1110 ASSERT(spa->spa_dsl_pool == NULL);
1111 ASSERT(spa->spa_root_vdev == NULL);
1112 ASSERT(spa->spa_async_zio_root == NULL);
1113 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
1115 txg_list_destroy(&spa->spa_vdev_txg_list);
1117 list_destroy(&spa->spa_config_dirty_list);
1118 list_destroy(&spa->spa_state_dirty_list);
1120 taskq_cancel_id(system_taskq, spa->spa_deadman_tqid);
1122 for (t = 0; t < ZIO_TYPES; t++) {
1123 for (q = 0; q < ZIO_TASKQ_TYPES; q++) {
1124 spa_taskqs_fini(spa, t, q);
1128 metaslab_class_destroy(spa->spa_normal_class);
1129 spa->spa_normal_class = NULL;
1131 metaslab_class_destroy(spa->spa_log_class);
1132 spa->spa_log_class = NULL;
1135 * If this was part of an import or the open otherwise failed, we may
1136 * still have errors left in the queues. Empty them just in case.
1138 spa_errlog_drain(spa);
1140 avl_destroy(&spa->spa_errlist_scrub);
1141 avl_destroy(&spa->spa_errlist_last);
1143 spa->spa_state = POOL_STATE_UNINITIALIZED;
1145 mutex_enter(&spa->spa_proc_lock);
1146 if (spa->spa_proc_state != SPA_PROC_NONE) {
1147 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1148 spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1149 cv_broadcast(&spa->spa_proc_cv);
1150 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1151 ASSERT(spa->spa_proc != &p0);
1152 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1154 ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1155 spa->spa_proc_state = SPA_PROC_NONE;
1157 ASSERT(spa->spa_proc == &p0);
1158 mutex_exit(&spa->spa_proc_lock);
1161 * We want to make sure spa_thread() has actually exited the ZFS
1162 * module, so that the module can't be unloaded out from underneath
1165 if (spa->spa_did != 0) {
1166 thread_join(spa->spa_did);
1172 * Verify a pool configuration, and construct the vdev tree appropriately. This
1173 * will create all the necessary vdevs in the appropriate layout, with each vdev
1174 * in the CLOSED state. This will prep the pool before open/creation/import.
1175 * All vdev validation is done by the vdev_alloc() routine.
1178 spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1179 uint_t id, int atype)
1186 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1189 if ((*vdp)->vdev_ops->vdev_op_leaf)
1192 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1195 if (error == ENOENT)
1204 for (c = 0; c < children; c++) {
1206 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1214 ASSERT(*vdp != NULL);
1220 * Opposite of spa_load().
1223 spa_unload(spa_t *spa)
1227 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1232 spa_async_suspend(spa);
1237 if (spa->spa_sync_on) {
1238 txg_sync_stop(spa->spa_dsl_pool);
1239 spa->spa_sync_on = B_FALSE;
1243 * Wait for any outstanding async I/O to complete.
1245 if (spa->spa_async_zio_root != NULL) {
1246 (void) zio_wait(spa->spa_async_zio_root);
1247 spa->spa_async_zio_root = NULL;
1250 bpobj_close(&spa->spa_deferred_bpobj);
1253 * Close the dsl pool.
1255 if (spa->spa_dsl_pool) {
1256 dsl_pool_close(spa->spa_dsl_pool);
1257 spa->spa_dsl_pool = NULL;
1258 spa->spa_meta_objset = NULL;
1263 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1266 * Drop and purge level 2 cache
1268 spa_l2cache_drop(spa);
1273 if (spa->spa_root_vdev)
1274 vdev_free(spa->spa_root_vdev);
1275 ASSERT(spa->spa_root_vdev == NULL);
1277 for (i = 0; i < spa->spa_spares.sav_count; i++)
1278 vdev_free(spa->spa_spares.sav_vdevs[i]);
1279 if (spa->spa_spares.sav_vdevs) {
1280 kmem_free(spa->spa_spares.sav_vdevs,
1281 spa->spa_spares.sav_count * sizeof (void *));
1282 spa->spa_spares.sav_vdevs = NULL;
1284 if (spa->spa_spares.sav_config) {
1285 nvlist_free(spa->spa_spares.sav_config);
1286 spa->spa_spares.sav_config = NULL;
1288 spa->spa_spares.sav_count = 0;
1290 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1291 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1292 vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1294 if (spa->spa_l2cache.sav_vdevs) {
1295 kmem_free(spa->spa_l2cache.sav_vdevs,
1296 spa->spa_l2cache.sav_count * sizeof (void *));
1297 spa->spa_l2cache.sav_vdevs = NULL;
1299 if (spa->spa_l2cache.sav_config) {
1300 nvlist_free(spa->spa_l2cache.sav_config);
1301 spa->spa_l2cache.sav_config = NULL;
1303 spa->spa_l2cache.sav_count = 0;
1305 spa->spa_async_suspended = 0;
1307 if (spa->spa_comment != NULL) {
1308 spa_strfree(spa->spa_comment);
1309 spa->spa_comment = NULL;
1312 spa_config_exit(spa, SCL_ALL, FTAG);
1316 * Load (or re-load) the current list of vdevs describing the active spares for
1317 * this pool. When this is called, we have some form of basic information in
1318 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1319 * then re-generate a more complete list including status information.
1322 spa_load_spares(spa_t *spa)
1329 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1332 * First, close and free any existing spare vdevs.
1334 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1335 vd = spa->spa_spares.sav_vdevs[i];
1337 /* Undo the call to spa_activate() below */
1338 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1339 B_FALSE)) != NULL && tvd->vdev_isspare)
1340 spa_spare_remove(tvd);
1345 if (spa->spa_spares.sav_vdevs)
1346 kmem_free(spa->spa_spares.sav_vdevs,
1347 spa->spa_spares.sav_count * sizeof (void *));
1349 if (spa->spa_spares.sav_config == NULL)
1352 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1353 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1355 spa->spa_spares.sav_count = (int)nspares;
1356 spa->spa_spares.sav_vdevs = NULL;
1362 * Construct the array of vdevs, opening them to get status in the
1363 * process. For each spare, there is potentially two different vdev_t
1364 * structures associated with it: one in the list of spares (used only
1365 * for basic validation purposes) and one in the active vdev
1366 * configuration (if it's spared in). During this phase we open and
1367 * validate each vdev on the spare list. If the vdev also exists in the
1368 * active configuration, then we also mark this vdev as an active spare.
1370 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1372 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1373 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1374 VDEV_ALLOC_SPARE) == 0);
1377 spa->spa_spares.sav_vdevs[i] = vd;
1379 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1380 B_FALSE)) != NULL) {
1381 if (!tvd->vdev_isspare)
1385 * We only mark the spare active if we were successfully
1386 * able to load the vdev. Otherwise, importing a pool
1387 * with a bad active spare would result in strange
1388 * behavior, because multiple pool would think the spare
1389 * is actively in use.
1391 * There is a vulnerability here to an equally bizarre
1392 * circumstance, where a dead active spare is later
1393 * brought back to life (onlined or otherwise). Given
1394 * the rarity of this scenario, and the extra complexity
1395 * it adds, we ignore the possibility.
1397 if (!vdev_is_dead(tvd))
1398 spa_spare_activate(tvd);
1402 vd->vdev_aux = &spa->spa_spares;
1404 if (vdev_open(vd) != 0)
1407 if (vdev_validate_aux(vd) == 0)
1412 * Recompute the stashed list of spares, with status information
1415 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1416 DATA_TYPE_NVLIST_ARRAY) == 0);
1418 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1420 for (i = 0; i < spa->spa_spares.sav_count; i++)
1421 spares[i] = vdev_config_generate(spa,
1422 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1423 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1424 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1425 for (i = 0; i < spa->spa_spares.sav_count; i++)
1426 nvlist_free(spares[i]);
1427 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1431 * Load (or re-load) the current list of vdevs describing the active l2cache for
1432 * this pool. When this is called, we have some form of basic information in
1433 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1434 * then re-generate a more complete list including status information.
1435 * Devices which are already active have their details maintained, and are
1439 spa_load_l2cache(spa_t *spa)
1443 int i, j, oldnvdevs;
1445 vdev_t *vd, **oldvdevs, **newvdevs = NULL;
1446 spa_aux_vdev_t *sav = &spa->spa_l2cache;
1448 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1450 if (sav->sav_config != NULL) {
1451 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1452 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1453 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_PUSHPAGE);
1458 oldvdevs = sav->sav_vdevs;
1459 oldnvdevs = sav->sav_count;
1460 sav->sav_vdevs = NULL;
1464 * Process new nvlist of vdevs.
1466 for (i = 0; i < nl2cache; i++) {
1467 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1471 for (j = 0; j < oldnvdevs; j++) {
1473 if (vd != NULL && guid == vd->vdev_guid) {
1475 * Retain previous vdev for add/remove ops.
1483 if (newvdevs[i] == NULL) {
1487 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1488 VDEV_ALLOC_L2CACHE) == 0);
1493 * Commit this vdev as an l2cache device,
1494 * even if it fails to open.
1496 spa_l2cache_add(vd);
1501 spa_l2cache_activate(vd);
1503 if (vdev_open(vd) != 0)
1506 (void) vdev_validate_aux(vd);
1508 if (!vdev_is_dead(vd))
1509 l2arc_add_vdev(spa, vd);
1514 * Purge vdevs that were dropped
1516 for (i = 0; i < oldnvdevs; i++) {
1521 ASSERT(vd->vdev_isl2cache);
1523 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1524 pool != 0ULL && l2arc_vdev_present(vd))
1525 l2arc_remove_vdev(vd);
1526 vdev_clear_stats(vd);
1532 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1534 if (sav->sav_config == NULL)
1537 sav->sav_vdevs = newvdevs;
1538 sav->sav_count = (int)nl2cache;
1541 * Recompute the stashed list of l2cache devices, with status
1542 * information this time.
1544 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1545 DATA_TYPE_NVLIST_ARRAY) == 0);
1547 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_PUSHPAGE);
1548 for (i = 0; i < sav->sav_count; i++)
1549 l2cache[i] = vdev_config_generate(spa,
1550 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1551 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1552 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1554 for (i = 0; i < sav->sav_count; i++)
1555 nvlist_free(l2cache[i]);
1557 kmem_free(l2cache, sav->sav_count * sizeof (void *));
1561 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1564 char *packed = NULL;
1569 error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db);
1573 nvsize = *(uint64_t *)db->db_data;
1574 dmu_buf_rele(db, FTAG);
1576 packed = kmem_alloc(nvsize, KM_PUSHPAGE | KM_NODEBUG);
1577 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1580 error = nvlist_unpack(packed, nvsize, value, 0);
1581 kmem_free(packed, nvsize);
1587 * Checks to see if the given vdev could not be opened, in which case we post a
1588 * sysevent to notify the autoreplace code that the device has been removed.
1591 spa_check_removed(vdev_t *vd)
1595 for (c = 0; c < vd->vdev_children; c++)
1596 spa_check_removed(vd->vdev_child[c]);
1598 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd)) {
1599 zfs_ereport_post(FM_EREPORT_RESOURCE_AUTOREPLACE,
1600 vd->vdev_spa, vd, NULL, 0, 0);
1601 spa_event_notify(vd->vdev_spa, vd, FM_EREPORT_ZFS_DEVICE_CHECK);
1606 * Validate the current config against the MOS config
1609 spa_config_valid(spa_t *spa, nvlist_t *config)
1611 vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1615 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1617 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1618 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1620 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1623 * If we're doing a normal import, then build up any additional
1624 * diagnostic information about missing devices in this config.
1625 * We'll pass this up to the user for further processing.
1627 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1628 nvlist_t **child, *nv;
1631 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1633 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_PUSHPAGE) == 0);
1635 for (c = 0; c < rvd->vdev_children; c++) {
1636 vdev_t *tvd = rvd->vdev_child[c];
1637 vdev_t *mtvd = mrvd->vdev_child[c];
1639 if (tvd->vdev_ops == &vdev_missing_ops &&
1640 mtvd->vdev_ops != &vdev_missing_ops &&
1642 child[idx++] = vdev_config_generate(spa, mtvd,
1647 VERIFY(nvlist_add_nvlist_array(nv,
1648 ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1649 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1650 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1652 for (i = 0; i < idx; i++)
1653 nvlist_free(child[i]);
1656 kmem_free(child, rvd->vdev_children * sizeof (char **));
1660 * Compare the root vdev tree with the information we have
1661 * from the MOS config (mrvd). Check each top-level vdev
1662 * with the corresponding MOS config top-level (mtvd).
1664 for (c = 0; c < rvd->vdev_children; c++) {
1665 vdev_t *tvd = rvd->vdev_child[c];
1666 vdev_t *mtvd = mrvd->vdev_child[c];
1669 * Resolve any "missing" vdevs in the current configuration.
1670 * If we find that the MOS config has more accurate information
1671 * about the top-level vdev then use that vdev instead.
1673 if (tvd->vdev_ops == &vdev_missing_ops &&
1674 mtvd->vdev_ops != &vdev_missing_ops) {
1676 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1680 * Device specific actions.
1682 if (mtvd->vdev_islog) {
1683 spa_set_log_state(spa, SPA_LOG_CLEAR);
1686 * XXX - once we have 'readonly' pool
1687 * support we should be able to handle
1688 * missing data devices by transitioning
1689 * the pool to readonly.
1695 * Swap the missing vdev with the data we were
1696 * able to obtain from the MOS config.
1698 vdev_remove_child(rvd, tvd);
1699 vdev_remove_child(mrvd, mtvd);
1701 vdev_add_child(rvd, mtvd);
1702 vdev_add_child(mrvd, tvd);
1704 spa_config_exit(spa, SCL_ALL, FTAG);
1706 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1709 } else if (mtvd->vdev_islog) {
1711 * Load the slog device's state from the MOS config
1712 * since it's possible that the label does not
1713 * contain the most up-to-date information.
1715 vdev_load_log_state(tvd, mtvd);
1720 spa_config_exit(spa, SCL_ALL, FTAG);
1723 * Ensure we were able to validate the config.
1725 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1729 * Check for missing log devices
1732 spa_check_logs(spa_t *spa)
1734 switch (spa->spa_log_state) {
1737 case SPA_LOG_MISSING:
1738 /* need to recheck in case slog has been restored */
1739 case SPA_LOG_UNKNOWN:
1740 if (dmu_objset_find(spa->spa_name, zil_check_log_chain, NULL,
1741 DS_FIND_CHILDREN)) {
1742 spa_set_log_state(spa, SPA_LOG_MISSING);
1751 spa_passivate_log(spa_t *spa)
1753 vdev_t *rvd = spa->spa_root_vdev;
1754 boolean_t slog_found = B_FALSE;
1757 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1759 if (!spa_has_slogs(spa))
1762 for (c = 0; c < rvd->vdev_children; c++) {
1763 vdev_t *tvd = rvd->vdev_child[c];
1764 metaslab_group_t *mg = tvd->vdev_mg;
1766 if (tvd->vdev_islog) {
1767 metaslab_group_passivate(mg);
1768 slog_found = B_TRUE;
1772 return (slog_found);
1776 spa_activate_log(spa_t *spa)
1778 vdev_t *rvd = spa->spa_root_vdev;
1781 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1783 for (c = 0; c < rvd->vdev_children; c++) {
1784 vdev_t *tvd = rvd->vdev_child[c];
1785 metaslab_group_t *mg = tvd->vdev_mg;
1787 if (tvd->vdev_islog)
1788 metaslab_group_activate(mg);
1793 spa_offline_log(spa_t *spa)
1797 if ((error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1798 NULL, DS_FIND_CHILDREN)) == 0) {
1801 * We successfully offlined the log device, sync out the
1802 * current txg so that the "stubby" block can be removed
1805 txg_wait_synced(spa->spa_dsl_pool, 0);
1811 spa_aux_check_removed(spa_aux_vdev_t *sav)
1815 for (i = 0; i < sav->sav_count; i++)
1816 spa_check_removed(sav->sav_vdevs[i]);
1820 spa_claim_notify(zio_t *zio)
1822 spa_t *spa = zio->io_spa;
1827 mutex_enter(&spa->spa_props_lock); /* any mutex will do */
1828 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1829 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1830 mutex_exit(&spa->spa_props_lock);
1833 typedef struct spa_load_error {
1834 uint64_t sle_meta_count;
1835 uint64_t sle_data_count;
1839 spa_load_verify_done(zio_t *zio)
1841 blkptr_t *bp = zio->io_bp;
1842 spa_load_error_t *sle = zio->io_private;
1843 dmu_object_type_t type = BP_GET_TYPE(bp);
1844 int error = zio->io_error;
1847 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1848 type != DMU_OT_INTENT_LOG)
1849 atomic_add_64(&sle->sle_meta_count, 1);
1851 atomic_add_64(&sle->sle_data_count, 1);
1853 zio_data_buf_free(zio->io_data, zio->io_size);
1858 spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1859 arc_buf_t *pbuf, const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1863 size_t size = BP_GET_PSIZE(bp);
1864 void *data = zio_data_buf_alloc(size);
1866 zio_nowait(zio_read(rio, spa, bp, data, size,
1867 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1868 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1869 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1875 spa_load_verify(spa_t *spa)
1878 spa_load_error_t sle = { 0 };
1879 zpool_rewind_policy_t policy;
1880 boolean_t verify_ok = B_FALSE;
1883 zpool_get_rewind_policy(spa->spa_config, &policy);
1885 if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1888 rio = zio_root(spa, NULL, &sle,
1889 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1891 error = traverse_pool(spa, spa->spa_verify_min_txg,
1892 TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio);
1894 (void) zio_wait(rio);
1896 spa->spa_load_meta_errors = sle.sle_meta_count;
1897 spa->spa_load_data_errors = sle.sle_data_count;
1899 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1900 sle.sle_data_count <= policy.zrp_maxdata) {
1904 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1905 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1907 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1908 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1909 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1910 VERIFY(nvlist_add_int64(spa->spa_load_info,
1911 ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1912 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1913 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1915 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1919 if (error != ENXIO && error != EIO)
1924 return (verify_ok ? 0 : EIO);
1928 * Find a value in the pool props object.
1931 spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1933 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1934 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1938 * Find a value in the pool directory object.
1941 spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1943 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1944 name, sizeof (uint64_t), 1, val));
1948 spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1950 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1955 * Fix up config after a partly-completed split. This is done with the
1956 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
1957 * pool have that entry in their config, but only the splitting one contains
1958 * a list of all the guids of the vdevs that are being split off.
1960 * This function determines what to do with that list: either rejoin
1961 * all the disks to the pool, or complete the splitting process. To attempt
1962 * the rejoin, each disk that is offlined is marked online again, and
1963 * we do a reopen() call. If the vdev label for every disk that was
1964 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1965 * then we call vdev_split() on each disk, and complete the split.
1967 * Otherwise we leave the config alone, with all the vdevs in place in
1968 * the original pool.
1971 spa_try_repair(spa_t *spa, nvlist_t *config)
1978 boolean_t attempt_reopen;
1980 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
1983 /* check that the config is complete */
1984 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
1985 &glist, &gcount) != 0)
1988 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_PUSHPAGE);
1990 /* attempt to online all the vdevs & validate */
1991 attempt_reopen = B_TRUE;
1992 for (i = 0; i < gcount; i++) {
1993 if (glist[i] == 0) /* vdev is hole */
1996 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
1997 if (vd[i] == NULL) {
1999 * Don't bother attempting to reopen the disks;
2000 * just do the split.
2002 attempt_reopen = B_FALSE;
2004 /* attempt to re-online it */
2005 vd[i]->vdev_offline = B_FALSE;
2009 if (attempt_reopen) {
2010 vdev_reopen(spa->spa_root_vdev);
2012 /* check each device to see what state it's in */
2013 for (extracted = 0, i = 0; i < gcount; i++) {
2014 if (vd[i] != NULL &&
2015 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
2022 * If every disk has been moved to the new pool, or if we never
2023 * even attempted to look at them, then we split them off for
2026 if (!attempt_reopen || gcount == extracted) {
2027 for (i = 0; i < gcount; i++)
2030 vdev_reopen(spa->spa_root_vdev);
2033 kmem_free(vd, gcount * sizeof (vdev_t *));
2037 spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
2038 boolean_t mosconfig)
2040 nvlist_t *config = spa->spa_config;
2041 char *ereport = FM_EREPORT_ZFS_POOL;
2047 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
2050 ASSERT(spa->spa_comment == NULL);
2051 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
2052 spa->spa_comment = spa_strdup(comment);
2055 * Versioning wasn't explicitly added to the label until later, so if
2056 * it's not present treat it as the initial version.
2058 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
2059 &spa->spa_ubsync.ub_version) != 0)
2060 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
2062 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
2063 &spa->spa_config_txg);
2065 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
2066 spa_guid_exists(pool_guid, 0)) {
2069 spa->spa_config_guid = pool_guid;
2071 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
2073 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
2077 nvlist_free(spa->spa_load_info);
2078 spa->spa_load_info = fnvlist_alloc();
2080 gethrestime(&spa->spa_loaded_ts);
2081 error = spa_load_impl(spa, pool_guid, config, state, type,
2082 mosconfig, &ereport);
2085 spa->spa_minref = refcount_count(&spa->spa_refcount);
2087 if (error != EEXIST) {
2088 spa->spa_loaded_ts.tv_sec = 0;
2089 spa->spa_loaded_ts.tv_nsec = 0;
2091 if (error != EBADF) {
2092 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
2095 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2102 * Load an existing storage pool, using the pool's builtin spa_config as a
2103 * source of configuration information.
2105 __attribute__((always_inline))
2107 spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
2108 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
2112 nvlist_t *nvroot = NULL;
2115 uberblock_t *ub = &spa->spa_uberblock;
2116 uint64_t children, config_cache_txg = spa->spa_config_txg;
2117 int orig_mode = spa->spa_mode;
2120 boolean_t missing_feat_write = B_FALSE;
2123 * If this is an untrusted config, access the pool in read-only mode.
2124 * This prevents things like resilvering recently removed devices.
2127 spa->spa_mode = FREAD;
2129 ASSERT(MUTEX_HELD(&spa_namespace_lock));
2131 spa->spa_load_state = state;
2133 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
2136 parse = (type == SPA_IMPORT_EXISTING ?
2137 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2140 * Create "The Godfather" zio to hold all async IOs
2142 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2143 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2146 * Parse the configuration into a vdev tree. We explicitly set the
2147 * value that will be returned by spa_version() since parsing the
2148 * configuration requires knowing the version number.
2150 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2151 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
2152 spa_config_exit(spa, SCL_ALL, FTAG);
2157 ASSERT(spa->spa_root_vdev == rvd);
2159 if (type != SPA_IMPORT_ASSEMBLE) {
2160 ASSERT(spa_guid(spa) == pool_guid);
2164 * Try to open all vdevs, loading each label in the process.
2166 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2167 error = vdev_open(rvd);
2168 spa_config_exit(spa, SCL_ALL, FTAG);
2173 * We need to validate the vdev labels against the configuration that
2174 * we have in hand, which is dependent on the setting of mosconfig. If
2175 * mosconfig is true then we're validating the vdev labels based on
2176 * that config. Otherwise, we're validating against the cached config
2177 * (zpool.cache) that was read when we loaded the zfs module, and then
2178 * later we will recursively call spa_load() and validate against
2181 * If we're assembling a new pool that's been split off from an
2182 * existing pool, the labels haven't yet been updated so we skip
2183 * validation for now.
2185 if (type != SPA_IMPORT_ASSEMBLE) {
2186 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2187 error = vdev_validate(rvd, mosconfig);
2188 spa_config_exit(spa, SCL_ALL, FTAG);
2193 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2198 * Find the best uberblock.
2200 vdev_uberblock_load(rvd, ub, &label);
2203 * If we weren't able to find a single valid uberblock, return failure.
2205 if (ub->ub_txg == 0) {
2207 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2211 * If the pool has an unsupported version we can't open it.
2213 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2215 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2218 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2222 * If we weren't able to find what's necessary for reading the
2223 * MOS in the label, return failure.
2225 if (label == NULL || nvlist_lookup_nvlist(label,
2226 ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) {
2228 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2233 * Update our in-core representation with the definitive values
2236 nvlist_free(spa->spa_label_features);
2237 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2243 * Look through entries in the label nvlist's features_for_read. If
2244 * there is a feature listed there which we don't understand then we
2245 * cannot open a pool.
2247 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2248 nvlist_t *unsup_feat;
2251 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2254 for (nvp = nvlist_next_nvpair(spa->spa_label_features, NULL);
2256 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2257 if (!zfeature_is_supported(nvpair_name(nvp))) {
2258 VERIFY(nvlist_add_string(unsup_feat,
2259 nvpair_name(nvp), "") == 0);
2263 if (!nvlist_empty(unsup_feat)) {
2264 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2265 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2266 nvlist_free(unsup_feat);
2267 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2271 nvlist_free(unsup_feat);
2275 * If the vdev guid sum doesn't match the uberblock, we have an
2276 * incomplete configuration. We first check to see if the pool
2277 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2278 * If it is, defer the vdev_guid_sum check till later so we
2279 * can handle missing vdevs.
2281 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
2282 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
2283 rvd->vdev_guid_sum != ub->ub_guid_sum)
2284 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2286 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2287 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2288 spa_try_repair(spa, config);
2289 spa_config_exit(spa, SCL_ALL, FTAG);
2290 nvlist_free(spa->spa_config_splitting);
2291 spa->spa_config_splitting = NULL;
2295 * Initialize internal SPA structures.
2297 spa->spa_state = POOL_STATE_ACTIVE;
2298 spa->spa_ubsync = spa->spa_uberblock;
2299 spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2300 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2301 spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2302 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2303 spa->spa_claim_max_txg = spa->spa_first_txg;
2304 spa->spa_prev_software_version = ub->ub_software_version;
2306 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2308 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2309 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2311 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2312 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2314 if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2315 boolean_t missing_feat_read = B_FALSE;
2316 nvlist_t *unsup_feat, *enabled_feat;
2318 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2319 &spa->spa_feat_for_read_obj) != 0) {
2320 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2323 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2324 &spa->spa_feat_for_write_obj) != 0) {
2325 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2328 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2329 &spa->spa_feat_desc_obj) != 0) {
2330 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2333 enabled_feat = fnvlist_alloc();
2334 unsup_feat = fnvlist_alloc();
2336 if (!feature_is_supported(spa->spa_meta_objset,
2337 spa->spa_feat_for_read_obj, spa->spa_feat_desc_obj,
2338 unsup_feat, enabled_feat))
2339 missing_feat_read = B_TRUE;
2341 if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) {
2342 if (!feature_is_supported(spa->spa_meta_objset,
2343 spa->spa_feat_for_write_obj, spa->spa_feat_desc_obj,
2344 unsup_feat, enabled_feat)) {
2345 missing_feat_write = B_TRUE;
2349 fnvlist_add_nvlist(spa->spa_load_info,
2350 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
2352 if (!nvlist_empty(unsup_feat)) {
2353 fnvlist_add_nvlist(spa->spa_load_info,
2354 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
2357 fnvlist_free(enabled_feat);
2358 fnvlist_free(unsup_feat);
2360 if (!missing_feat_read) {
2361 fnvlist_add_boolean(spa->spa_load_info,
2362 ZPOOL_CONFIG_CAN_RDONLY);
2366 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2367 * twofold: to determine whether the pool is available for
2368 * import in read-write mode and (if it is not) whether the
2369 * pool is available for import in read-only mode. If the pool
2370 * is available for import in read-write mode, it is displayed
2371 * as available in userland; if it is not available for import
2372 * in read-only mode, it is displayed as unavailable in
2373 * userland. If the pool is available for import in read-only
2374 * mode but not read-write mode, it is displayed as unavailable
2375 * in userland with a special note that the pool is actually
2376 * available for open in read-only mode.
2378 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2379 * missing a feature for write, we must first determine whether
2380 * the pool can be opened read-only before returning to
2381 * userland in order to know whether to display the
2382 * abovementioned note.
2384 if (missing_feat_read || (missing_feat_write &&
2385 spa_writeable(spa))) {
2386 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2391 spa->spa_is_initializing = B_TRUE;
2392 error = dsl_pool_open(spa->spa_dsl_pool);
2393 spa->spa_is_initializing = B_FALSE;
2395 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2399 nvlist_t *policy = NULL, *nvconfig;
2401 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2402 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2404 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2405 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2407 unsigned long myhostid = 0;
2409 VERIFY(nvlist_lookup_string(nvconfig,
2410 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2413 myhostid = zone_get_hostid(NULL);
2416 * We're emulating the system's hostid in userland, so
2417 * we can't use zone_get_hostid().
2419 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2420 #endif /* _KERNEL */
2421 if (hostid != 0 && myhostid != 0 &&
2422 hostid != myhostid) {
2423 nvlist_free(nvconfig);
2424 cmn_err(CE_WARN, "pool '%s' could not be "
2425 "loaded as it was last accessed by "
2426 "another system (host: %s hostid: 0x%lx). "
2427 "See: http://zfsonlinux.org/msg/ZFS-8000-EY",
2428 spa_name(spa), hostname,
2429 (unsigned long)hostid);
2433 if (nvlist_lookup_nvlist(spa->spa_config,
2434 ZPOOL_REWIND_POLICY, &policy) == 0)
2435 VERIFY(nvlist_add_nvlist(nvconfig,
2436 ZPOOL_REWIND_POLICY, policy) == 0);
2438 spa_config_set(spa, nvconfig);
2440 spa_deactivate(spa);
2441 spa_activate(spa, orig_mode);
2443 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2446 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2447 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2448 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2450 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2453 * Load the bit that tells us to use the new accounting function
2454 * (raid-z deflation). If we have an older pool, this will not
2457 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2458 if (error != 0 && error != ENOENT)
2459 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2461 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2462 &spa->spa_creation_version);
2463 if (error != 0 && error != ENOENT)
2464 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2467 * Load the persistent error log. If we have an older pool, this will
2470 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2471 if (error != 0 && error != ENOENT)
2472 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2474 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2475 &spa->spa_errlog_scrub);
2476 if (error != 0 && error != ENOENT)
2477 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2480 * Load the history object. If we have an older pool, this
2481 * will not be present.
2483 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2484 if (error != 0 && error != ENOENT)
2485 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2488 * If we're assembling the pool from the split-off vdevs of
2489 * an existing pool, we don't want to attach the spares & cache
2494 * Load any hot spares for this pool.
2496 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2497 if (error != 0 && error != ENOENT)
2498 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2499 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2500 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2501 if (load_nvlist(spa, spa->spa_spares.sav_object,
2502 &spa->spa_spares.sav_config) != 0)
2503 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2505 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2506 spa_load_spares(spa);
2507 spa_config_exit(spa, SCL_ALL, FTAG);
2508 } else if (error == 0) {
2509 spa->spa_spares.sav_sync = B_TRUE;
2513 * Load any level 2 ARC devices for this pool.
2515 error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2516 &spa->spa_l2cache.sav_object);
2517 if (error != 0 && error != ENOENT)
2518 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2519 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2520 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2521 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2522 &spa->spa_l2cache.sav_config) != 0)
2523 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2525 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2526 spa_load_l2cache(spa);
2527 spa_config_exit(spa, SCL_ALL, FTAG);
2528 } else if (error == 0) {
2529 spa->spa_l2cache.sav_sync = B_TRUE;
2532 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2534 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2535 if (error && error != ENOENT)
2536 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2539 uint64_t autoreplace;
2541 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2542 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2543 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2544 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2545 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2546 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2547 &spa->spa_dedup_ditto);
2549 spa->spa_autoreplace = (autoreplace != 0);
2553 * If the 'autoreplace' property is set, then post a resource notifying
2554 * the ZFS DE that it should not issue any faults for unopenable
2555 * devices. We also iterate over the vdevs, and post a sysevent for any
2556 * unopenable vdevs so that the normal autoreplace handler can take
2559 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2560 spa_check_removed(spa->spa_root_vdev);
2562 * For the import case, this is done in spa_import(), because
2563 * at this point we're using the spare definitions from
2564 * the MOS config, not necessarily from the userland config.
2566 if (state != SPA_LOAD_IMPORT) {
2567 spa_aux_check_removed(&spa->spa_spares);
2568 spa_aux_check_removed(&spa->spa_l2cache);
2573 * Load the vdev state for all toplevel vdevs.
2578 * Propagate the leaf DTLs we just loaded all the way up the tree.
2580 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2581 vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2582 spa_config_exit(spa, SCL_ALL, FTAG);
2585 * Load the DDTs (dedup tables).
2587 error = ddt_load(spa);
2589 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2591 spa_update_dspace(spa);
2594 * Validate the config, using the MOS config to fill in any
2595 * information which might be missing. If we fail to validate
2596 * the config then declare the pool unfit for use. If we're
2597 * assembling a pool from a split, the log is not transferred
2600 if (type != SPA_IMPORT_ASSEMBLE) {
2603 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2604 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2606 if (!spa_config_valid(spa, nvconfig)) {
2607 nvlist_free(nvconfig);
2608 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2611 nvlist_free(nvconfig);
2614 * Now that we've validated the config, check the state of the
2615 * root vdev. If it can't be opened, it indicates one or
2616 * more toplevel vdevs are faulted.
2618 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2621 if (spa_check_logs(spa)) {
2622 *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2623 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2627 if (missing_feat_write) {
2628 ASSERT(state == SPA_LOAD_TRYIMPORT);
2631 * At this point, we know that we can open the pool in
2632 * read-only mode but not read-write mode. We now have enough
2633 * information and can return to userland.
2635 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
2639 * We've successfully opened the pool, verify that we're ready
2640 * to start pushing transactions.
2642 if (state != SPA_LOAD_TRYIMPORT) {
2643 if ((error = spa_load_verify(spa)))
2644 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2648 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2649 spa->spa_load_max_txg == UINT64_MAX)) {
2651 int need_update = B_FALSE;
2654 ASSERT(state != SPA_LOAD_TRYIMPORT);
2657 * Claim log blocks that haven't been committed yet.
2658 * This must all happen in a single txg.
2659 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2660 * invoked from zil_claim_log_block()'s i/o done callback.
2661 * Price of rollback is that we abandon the log.
2663 spa->spa_claiming = B_TRUE;
2665 tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2666 spa_first_txg(spa));
2667 (void) dmu_objset_find(spa_name(spa),
2668 zil_claim, tx, DS_FIND_CHILDREN);
2671 spa->spa_claiming = B_FALSE;
2673 spa_set_log_state(spa, SPA_LOG_GOOD);
2674 spa->spa_sync_on = B_TRUE;
2675 txg_sync_start(spa->spa_dsl_pool);
2678 * Wait for all claims to sync. We sync up to the highest
2679 * claimed log block birth time so that claimed log blocks
2680 * don't appear to be from the future. spa_claim_max_txg
2681 * will have been set for us by either zil_check_log_chain()
2682 * (invoked from spa_check_logs()) or zil_claim() above.
2684 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2687 * If the config cache is stale, or we have uninitialized
2688 * metaslabs (see spa_vdev_add()), then update the config.
2690 * If this is a verbatim import, trust the current
2691 * in-core spa_config and update the disk labels.
2693 if (config_cache_txg != spa->spa_config_txg ||
2694 state == SPA_LOAD_IMPORT ||
2695 state == SPA_LOAD_RECOVER ||
2696 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2697 need_update = B_TRUE;
2699 for (c = 0; c < rvd->vdev_children; c++)
2700 if (rvd->vdev_child[c]->vdev_ms_array == 0)
2701 need_update = B_TRUE;
2704 * Update the config cache asychronously in case we're the
2705 * root pool, in which case the config cache isn't writable yet.
2708 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2711 * Check all DTLs to see if anything needs resilvering.
2713 if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2714 vdev_resilver_needed(rvd, NULL, NULL))
2715 spa_async_request(spa, SPA_ASYNC_RESILVER);
2718 * Delete any inconsistent datasets.
2720 (void) dmu_objset_find(spa_name(spa),
2721 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2724 * Clean up any stale temporary dataset userrefs.
2726 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2733 spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2735 int mode = spa->spa_mode;
2738 spa_deactivate(spa);
2740 spa->spa_load_max_txg--;
2742 spa_activate(spa, mode);
2743 spa_async_suspend(spa);
2745 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2749 * If spa_load() fails this function will try loading prior txg's. If
2750 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2751 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2752 * function will not rewind the pool and will return the same error as
2756 spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2757 uint64_t max_request, int rewind_flags)
2759 nvlist_t *loadinfo = NULL;
2760 nvlist_t *config = NULL;
2761 int load_error, rewind_error;
2762 uint64_t safe_rewind_txg;
2765 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2766 spa->spa_load_max_txg = spa->spa_load_txg;
2767 spa_set_log_state(spa, SPA_LOG_CLEAR);
2769 spa->spa_load_max_txg = max_request;
2772 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2774 if (load_error == 0)
2777 if (spa->spa_root_vdev != NULL)
2778 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2780 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2781 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2783 if (rewind_flags & ZPOOL_NEVER_REWIND) {
2784 nvlist_free(config);
2785 return (load_error);
2788 if (state == SPA_LOAD_RECOVER) {
2789 /* Price of rolling back is discarding txgs, including log */
2790 spa_set_log_state(spa, SPA_LOG_CLEAR);
2793 * If we aren't rolling back save the load info from our first
2794 * import attempt so that we can restore it after attempting
2797 loadinfo = spa->spa_load_info;
2798 spa->spa_load_info = fnvlist_alloc();
2801 spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2802 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2803 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2804 TXG_INITIAL : safe_rewind_txg;
2807 * Continue as long as we're finding errors, we're still within
2808 * the acceptable rewind range, and we're still finding uberblocks
2810 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2811 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2812 if (spa->spa_load_max_txg < safe_rewind_txg)
2813 spa->spa_extreme_rewind = B_TRUE;
2814 rewind_error = spa_load_retry(spa, state, mosconfig);
2817 spa->spa_extreme_rewind = B_FALSE;
2818 spa->spa_load_max_txg = UINT64_MAX;
2820 if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2821 spa_config_set(spa, config);
2823 if (state == SPA_LOAD_RECOVER) {
2824 ASSERT3P(loadinfo, ==, NULL);
2825 return (rewind_error);
2827 /* Store the rewind info as part of the initial load info */
2828 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
2829 spa->spa_load_info);
2831 /* Restore the initial load info */
2832 fnvlist_free(spa->spa_load_info);
2833 spa->spa_load_info = loadinfo;
2835 return (load_error);
2842 * The import case is identical to an open except that the configuration is sent
2843 * down from userland, instead of grabbed from the configuration cache. For the
2844 * case of an open, the pool configuration will exist in the
2845 * POOL_STATE_UNINITIALIZED state.
2847 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2848 * the same time open the pool, without having to keep around the spa_t in some
2852 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2856 spa_load_state_t state = SPA_LOAD_OPEN;
2858 int locked = B_FALSE;
2863 * As disgusting as this is, we need to support recursive calls to this
2864 * function because dsl_dir_open() is called during spa_load(), and ends
2865 * up calling spa_open() again. The real fix is to figure out how to
2866 * avoid dsl_dir_open() calling this in the first place.
2868 if (mutex_owner(&spa_namespace_lock) != curthread) {
2869 mutex_enter(&spa_namespace_lock);
2873 if ((spa = spa_lookup(pool)) == NULL) {
2875 mutex_exit(&spa_namespace_lock);
2879 if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2880 zpool_rewind_policy_t policy;
2882 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2884 if (policy.zrp_request & ZPOOL_DO_REWIND)
2885 state = SPA_LOAD_RECOVER;
2887 spa_activate(spa, spa_mode_global);
2889 if (state != SPA_LOAD_RECOVER)
2890 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2892 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2893 policy.zrp_request);
2895 if (error == EBADF) {
2897 * If vdev_validate() returns failure (indicated by
2898 * EBADF), it indicates that one of the vdevs indicates
2899 * that the pool has been exported or destroyed. If
2900 * this is the case, the config cache is out of sync and
2901 * we should remove the pool from the namespace.
2904 spa_deactivate(spa);
2905 spa_config_sync(spa, B_TRUE, B_TRUE);
2908 mutex_exit(&spa_namespace_lock);
2914 * We can't open the pool, but we still have useful
2915 * information: the state of each vdev after the
2916 * attempted vdev_open(). Return this to the user.
2918 if (config != NULL && spa->spa_config) {
2919 VERIFY(nvlist_dup(spa->spa_config, config,
2921 VERIFY(nvlist_add_nvlist(*config,
2922 ZPOOL_CONFIG_LOAD_INFO,
2923 spa->spa_load_info) == 0);
2926 spa_deactivate(spa);
2927 spa->spa_last_open_failed = error;
2929 mutex_exit(&spa_namespace_lock);
2935 spa_open_ref(spa, tag);
2938 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2941 * If we've recovered the pool, pass back any information we
2942 * gathered while doing the load.
2944 if (state == SPA_LOAD_RECOVER) {
2945 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
2946 spa->spa_load_info) == 0);
2950 spa->spa_last_open_failed = 0;
2951 spa->spa_last_ubsync_txg = 0;
2952 spa->spa_load_txg = 0;
2953 mutex_exit(&spa_namespace_lock);
2962 spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
2965 return (spa_open_common(name, spapp, tag, policy, config));
2969 spa_open(const char *name, spa_t **spapp, void *tag)
2971 return (spa_open_common(name, spapp, tag, NULL, NULL));
2975 * Lookup the given spa_t, incrementing the inject count in the process,
2976 * preventing it from being exported or destroyed.
2979 spa_inject_addref(char *name)
2983 mutex_enter(&spa_namespace_lock);
2984 if ((spa = spa_lookup(name)) == NULL) {
2985 mutex_exit(&spa_namespace_lock);
2988 spa->spa_inject_ref++;
2989 mutex_exit(&spa_namespace_lock);
2995 spa_inject_delref(spa_t *spa)
2997 mutex_enter(&spa_namespace_lock);
2998 spa->spa_inject_ref--;
2999 mutex_exit(&spa_namespace_lock);
3003 * Add spares device information to the nvlist.
3006 spa_add_spares(spa_t *spa, nvlist_t *config)
3016 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3018 if (spa->spa_spares.sav_count == 0)
3021 VERIFY(nvlist_lookup_nvlist(config,
3022 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3023 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
3024 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3026 VERIFY(nvlist_add_nvlist_array(nvroot,
3027 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3028 VERIFY(nvlist_lookup_nvlist_array(nvroot,
3029 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3032 * Go through and find any spares which have since been
3033 * repurposed as an active spare. If this is the case, update
3034 * their status appropriately.
3036 for (i = 0; i < nspares; i++) {
3037 VERIFY(nvlist_lookup_uint64(spares[i],
3038 ZPOOL_CONFIG_GUID, &guid) == 0);
3039 if (spa_spare_exists(guid, &pool, NULL) &&
3041 VERIFY(nvlist_lookup_uint64_array(
3042 spares[i], ZPOOL_CONFIG_VDEV_STATS,
3043 (uint64_t **)&vs, &vsc) == 0);
3044 vs->vs_state = VDEV_STATE_CANT_OPEN;
3045 vs->vs_aux = VDEV_AUX_SPARED;
3052 * Add l2cache device information to the nvlist, including vdev stats.
3055 spa_add_l2cache(spa_t *spa, nvlist_t *config)
3058 uint_t i, j, nl2cache;
3065 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3067 if (spa->spa_l2cache.sav_count == 0)
3070 VERIFY(nvlist_lookup_nvlist(config,
3071 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3072 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
3073 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3074 if (nl2cache != 0) {
3075 VERIFY(nvlist_add_nvlist_array(nvroot,
3076 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3077 VERIFY(nvlist_lookup_nvlist_array(nvroot,
3078 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3081 * Update level 2 cache device stats.
3084 for (i = 0; i < nl2cache; i++) {
3085 VERIFY(nvlist_lookup_uint64(l2cache[i],
3086 ZPOOL_CONFIG_GUID, &guid) == 0);
3089 for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
3091 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
3092 vd = spa->spa_l2cache.sav_vdevs[j];
3098 VERIFY(nvlist_lookup_uint64_array(l2cache[i],
3099 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
3101 vdev_get_stats(vd, vs);
3107 spa_add_feature_stats(spa_t *spa, nvlist_t *config)
3113 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3114 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3116 if (spa->spa_feat_for_read_obj != 0) {
3117 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3118 spa->spa_feat_for_read_obj);
3119 zap_cursor_retrieve(&zc, &za) == 0;
3120 zap_cursor_advance(&zc)) {
3121 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3122 za.za_num_integers == 1);
3123 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3124 za.za_first_integer));
3126 zap_cursor_fini(&zc);
3129 if (spa->spa_feat_for_write_obj != 0) {
3130 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3131 spa->spa_feat_for_write_obj);
3132 zap_cursor_retrieve(&zc, &za) == 0;
3133 zap_cursor_advance(&zc)) {
3134 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3135 za.za_num_integers == 1);
3136 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3137 za.za_first_integer));
3139 zap_cursor_fini(&zc);
3142 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
3144 nvlist_free(features);
3148 spa_get_stats(const char *name, nvlist_t **config,
3149 char *altroot, size_t buflen)
3155 error = spa_open_common(name, &spa, FTAG, NULL, config);
3159 * This still leaves a window of inconsistency where the spares
3160 * or l2cache devices could change and the config would be
3161 * self-inconsistent.
3163 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3165 if (*config != NULL) {
3166 uint64_t loadtimes[2];
3168 loadtimes[0] = spa->spa_loaded_ts.tv_sec;
3169 loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
3170 VERIFY(nvlist_add_uint64_array(*config,
3171 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
3173 VERIFY(nvlist_add_uint64(*config,
3174 ZPOOL_CONFIG_ERRCOUNT,
3175 spa_get_errlog_size(spa)) == 0);
3177 if (spa_suspended(spa))
3178 VERIFY(nvlist_add_uint64(*config,
3179 ZPOOL_CONFIG_SUSPENDED,
3180 spa->spa_failmode) == 0);
3182 spa_add_spares(spa, *config);
3183 spa_add_l2cache(spa, *config);
3184 spa_add_feature_stats(spa, *config);
3189 * We want to get the alternate root even for faulted pools, so we cheat
3190 * and call spa_lookup() directly.
3194 mutex_enter(&spa_namespace_lock);
3195 spa = spa_lookup(name);
3197 spa_altroot(spa, altroot, buflen);
3201 mutex_exit(&spa_namespace_lock);
3203 spa_altroot(spa, altroot, buflen);
3208 spa_config_exit(spa, SCL_CONFIG, FTAG);
3209 spa_close(spa, FTAG);
3216 * Validate that the auxiliary device array is well formed. We must have an
3217 * array of nvlists, each which describes a valid leaf vdev. If this is an
3218 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3219 * specified, as long as they are well-formed.
3222 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
3223 spa_aux_vdev_t *sav, const char *config, uint64_t version,
3224 vdev_labeltype_t label)
3231 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3234 * It's acceptable to have no devs specified.
3236 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
3243 * Make sure the pool is formatted with a version that supports this
3246 if (spa_version(spa) < version)
3250 * Set the pending device list so we correctly handle device in-use
3253 sav->sav_pending = dev;
3254 sav->sav_npending = ndev;
3256 for (i = 0; i < ndev; i++) {
3257 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
3261 if (!vd->vdev_ops->vdev_op_leaf) {
3268 * The L2ARC currently only supports disk devices in
3269 * kernel context. For user-level testing, we allow it.
3272 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
3273 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
3281 if ((error = vdev_open(vd)) == 0 &&
3282 (error = vdev_label_init(vd, crtxg, label)) == 0) {
3283 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
3284 vd->vdev_guid) == 0);
3290 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
3297 sav->sav_pending = NULL;
3298 sav->sav_npending = 0;
3303 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
3307 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3309 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3310 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
3311 VDEV_LABEL_SPARE)) != 0) {
3315 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3316 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
3317 VDEV_LABEL_L2CACHE));
3321 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
3326 if (sav->sav_config != NULL) {
3332 * Generate new dev list by concatentating with the
3335 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
3336 &olddevs, &oldndevs) == 0);
3338 newdevs = kmem_alloc(sizeof (void *) *
3339 (ndevs + oldndevs), KM_PUSHPAGE);
3340 for (i = 0; i < oldndevs; i++)
3341 VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
3343 for (i = 0; i < ndevs; i++)
3344 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
3347 VERIFY(nvlist_remove(sav->sav_config, config,
3348 DATA_TYPE_NVLIST_ARRAY) == 0);
3350 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3351 config, newdevs, ndevs + oldndevs) == 0);
3352 for (i = 0; i < oldndevs + ndevs; i++)
3353 nvlist_free(newdevs[i]);
3354 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
3357 * Generate a new dev list.
3359 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
3361 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
3367 * Stop and drop level 2 ARC devices
3370 spa_l2cache_drop(spa_t *spa)
3374 spa_aux_vdev_t *sav = &spa->spa_l2cache;
3376 for (i = 0; i < sav->sav_count; i++) {
3379 vd = sav->sav_vdevs[i];
3382 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
3383 pool != 0ULL && l2arc_vdev_present(vd))
3384 l2arc_remove_vdev(vd);
3392 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
3393 const char *history_str, nvlist_t *zplprops)
3396 char *altroot = NULL;
3401 uint64_t txg = TXG_INITIAL;
3402 nvlist_t **spares, **l2cache;
3403 uint_t nspares, nl2cache;
3404 uint64_t version, obj;
3405 boolean_t has_features;
3410 * If this pool already exists, return failure.
3412 mutex_enter(&spa_namespace_lock);
3413 if (spa_lookup(pool) != NULL) {
3414 mutex_exit(&spa_namespace_lock);
3419 * Allocate a new spa_t structure.
3421 (void) nvlist_lookup_string(props,
3422 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3423 spa = spa_add(pool, NULL, altroot);
3424 spa_activate(spa, spa_mode_global);
3426 if (props && (error = spa_prop_validate(spa, props))) {
3427 spa_deactivate(spa);
3429 mutex_exit(&spa_namespace_lock);
3433 has_features = B_FALSE;
3434 for (elem = nvlist_next_nvpair(props, NULL);
3435 elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
3436 if (zpool_prop_feature(nvpair_name(elem)))
3437 has_features = B_TRUE;
3440 if (has_features || nvlist_lookup_uint64(props,
3441 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
3442 version = SPA_VERSION;
3444 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
3446 spa->spa_first_txg = txg;
3447 spa->spa_uberblock.ub_txg = txg - 1;
3448 spa->spa_uberblock.ub_version = version;
3449 spa->spa_ubsync = spa->spa_uberblock;
3452 * Create "The Godfather" zio to hold all async IOs
3454 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
3455 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
3458 * Create the root vdev.
3460 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3462 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3464 ASSERT(error != 0 || rvd != NULL);
3465 ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3467 if (error == 0 && !zfs_allocatable_devs(nvroot))
3471 (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3472 (error = spa_validate_aux(spa, nvroot, txg,
3473 VDEV_ALLOC_ADD)) == 0) {
3474 for (c = 0; c < rvd->vdev_children; c++) {
3475 vdev_metaslab_set_size(rvd->vdev_child[c]);
3476 vdev_expand(rvd->vdev_child[c], txg);
3480 spa_config_exit(spa, SCL_ALL, FTAG);
3484 spa_deactivate(spa);
3486 mutex_exit(&spa_namespace_lock);
3491 * Get the list of spares, if specified.
3493 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3494 &spares, &nspares) == 0) {
3495 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3497 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3498 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3499 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3500 spa_load_spares(spa);
3501 spa_config_exit(spa, SCL_ALL, FTAG);
3502 spa->spa_spares.sav_sync = B_TRUE;
3506 * Get the list of level 2 cache devices, if specified.
3508 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3509 &l2cache, &nl2cache) == 0) {
3510 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3511 NV_UNIQUE_NAME, KM_PUSHPAGE) == 0);
3512 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3513 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3514 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3515 spa_load_l2cache(spa);
3516 spa_config_exit(spa, SCL_ALL, FTAG);
3517 spa->spa_l2cache.sav_sync = B_TRUE;
3520 spa->spa_is_initializing = B_TRUE;
3521 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3522 spa->spa_meta_objset = dp->dp_meta_objset;
3523 spa->spa_is_initializing = B_FALSE;
3526 * Create DDTs (dedup tables).
3530 spa_update_dspace(spa);
3532 tx = dmu_tx_create_assigned(dp, txg);
3535 * Create the pool config object.
3537 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3538 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3539 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3541 if (zap_add(spa->spa_meta_objset,
3542 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3543 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3544 cmn_err(CE_PANIC, "failed to add pool config");
3547 if (spa_version(spa) >= SPA_VERSION_FEATURES)
3548 spa_feature_create_zap_objects(spa, tx);
3550 if (zap_add(spa->spa_meta_objset,
3551 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3552 sizeof (uint64_t), 1, &version, tx) != 0) {
3553 cmn_err(CE_PANIC, "failed to add pool version");
3556 /* Newly created pools with the right version are always deflated. */
3557 if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3558 spa->spa_deflate = TRUE;
3559 if (zap_add(spa->spa_meta_objset,
3560 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3561 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3562 cmn_err(CE_PANIC, "failed to add deflate");
3567 * Create the deferred-free bpobj. Turn off compression
3568 * because sync-to-convergence takes longer if the blocksize
3571 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3572 dmu_object_set_compress(spa->spa_meta_objset, obj,
3573 ZIO_COMPRESS_OFF, tx);
3574 if (zap_add(spa->spa_meta_objset,
3575 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3576 sizeof (uint64_t), 1, &obj, tx) != 0) {
3577 cmn_err(CE_PANIC, "failed to add bpobj");
3579 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3580 spa->spa_meta_objset, obj));
3583 * Create the pool's history object.
3585 if (version >= SPA_VERSION_ZPOOL_HISTORY)
3586 spa_history_create_obj(spa, tx);
3589 * Set pool properties.
3591 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3592 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3593 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3594 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3596 if (props != NULL) {
3597 spa_configfile_set(spa, props, B_FALSE);
3598 spa_sync_props(spa, props, tx);
3603 spa->spa_sync_on = B_TRUE;
3604 txg_sync_start(spa->spa_dsl_pool);
3607 * We explicitly wait for the first transaction to complete so that our
3608 * bean counters are appropriately updated.
3610 txg_wait_synced(spa->spa_dsl_pool, txg);
3612 spa_config_sync(spa, B_FALSE, B_TRUE);
3614 if (version >= SPA_VERSION_ZPOOL_HISTORY && history_str != NULL)
3615 (void) spa_history_log(spa, history_str, LOG_CMD_POOL_CREATE);
3616 spa_history_log_version(spa, LOG_POOL_CREATE);
3618 spa->spa_minref = refcount_count(&spa->spa_refcount);
3620 mutex_exit(&spa_namespace_lock);
3627 * Get the root pool information from the root disk, then import the root pool
3628 * during the system boot up time.
3630 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3633 spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3636 nvlist_t *nvtop, *nvroot;
3639 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3643 * Add this top-level vdev to the child array.
3645 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3647 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3649 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3652 * Put this pool's top-level vdevs into a root vdev.
3654 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_PUSHPAGE) == 0);
3655 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3656 VDEV_TYPE_ROOT) == 0);
3657 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3658 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3659 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3663 * Replace the existing vdev_tree with the new root vdev in
3664 * this pool's configuration (remove the old, add the new).
3666 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3667 nvlist_free(nvroot);
3672 * Walk the vdev tree and see if we can find a device with "better"
3673 * configuration. A configuration is "better" if the label on that
3674 * device has a more recent txg.
3677 spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3681 for (c = 0; c < vd->vdev_children; c++)
3682 spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3684 if (vd->vdev_ops->vdev_op_leaf) {
3688 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3692 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3696 * Do we have a better boot device?
3698 if (label_txg > *txg) {
3707 * Import a root pool.
3709 * For x86. devpath_list will consist of devid and/or physpath name of
3710 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3711 * The GRUB "findroot" command will return the vdev we should boot.
3713 * For Sparc, devpath_list consists the physpath name of the booting device
3714 * no matter the rootpool is a single device pool or a mirrored pool.
3716 * "/pci@1f,0/ide@d/disk@0,0:a"
3719 spa_import_rootpool(char *devpath, char *devid)
3722 vdev_t *rvd, *bvd, *avd = NULL;
3723 nvlist_t *config, *nvtop;
3729 * Read the label from the boot device and generate a configuration.
3731 config = spa_generate_rootconf(devpath, devid, &guid);
3732 #if defined(_OBP) && defined(_KERNEL)
3733 if (config == NULL) {
3734 if (strstr(devpath, "/iscsi/ssd") != NULL) {
3736 get_iscsi_bootpath_phy(devpath);
3737 config = spa_generate_rootconf(devpath, devid, &guid);
3741 if (config == NULL) {
3742 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
3747 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3749 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3751 mutex_enter(&spa_namespace_lock);
3752 if ((spa = spa_lookup(pname)) != NULL) {
3754 * Remove the existing root pool from the namespace so that we
3755 * can replace it with the correct config we just read in.
3760 spa = spa_add(pname, config, NULL);
3761 spa->spa_is_root = B_TRUE;
3762 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3765 * Build up a vdev tree based on the boot device's label config.
3767 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3769 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3770 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3771 VDEV_ALLOC_ROOTPOOL);
3772 spa_config_exit(spa, SCL_ALL, FTAG);
3774 mutex_exit(&spa_namespace_lock);
3775 nvlist_free(config);
3776 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3782 * Get the boot vdev.
3784 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3785 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3786 (u_longlong_t)guid);
3792 * Determine if there is a better boot device.
3795 spa_alt_rootvdev(rvd, &avd, &txg);
3797 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3798 "try booting from '%s'", avd->vdev_path);
3804 * If the boot device is part of a spare vdev then ensure that
3805 * we're booting off the active spare.
3807 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3808 !bvd->vdev_isspare) {
3809 cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3810 "try booting from '%s'",
3812 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3818 spa_history_log_version(spa, LOG_POOL_IMPORT);
3820 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3822 spa_config_exit(spa, SCL_ALL, FTAG);
3823 mutex_exit(&spa_namespace_lock);
3825 nvlist_free(config);
3832 * Import a non-root pool into the system.
3835 spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
3838 char *altroot = NULL;
3839 spa_load_state_t state = SPA_LOAD_IMPORT;
3840 zpool_rewind_policy_t policy;
3841 uint64_t mode = spa_mode_global;
3842 uint64_t readonly = B_FALSE;
3845 nvlist_t **spares, **l2cache;
3846 uint_t nspares, nl2cache;
3849 * If a pool with this name exists, return failure.
3851 mutex_enter(&spa_namespace_lock);
3852 if (spa_lookup(pool) != NULL) {
3853 mutex_exit(&spa_namespace_lock);
3858 * Create and initialize the spa structure.
3860 (void) nvlist_lookup_string(props,
3861 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3862 (void) nvlist_lookup_uint64(props,
3863 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
3866 spa = spa_add(pool, config, altroot);
3867 spa->spa_import_flags = flags;
3870 * Verbatim import - Take a pool and insert it into the namespace
3871 * as if it had been loaded at boot.
3873 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
3875 spa_configfile_set(spa, props, B_FALSE);
3877 spa_config_sync(spa, B_FALSE, B_TRUE);
3879 mutex_exit(&spa_namespace_lock);
3880 spa_history_log_version(spa, LOG_POOL_IMPORT);
3885 spa_activate(spa, mode);
3888 * Don't start async tasks until we know everything is healthy.
3890 spa_async_suspend(spa);
3892 zpool_get_rewind_policy(config, &policy);
3893 if (policy.zrp_request & ZPOOL_DO_REWIND)
3894 state = SPA_LOAD_RECOVER;
3897 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
3898 * because the user-supplied config is actually the one to trust when
3901 if (state != SPA_LOAD_RECOVER)
3902 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3904 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
3905 policy.zrp_request);
3908 * Propagate anything learned while loading the pool and pass it
3909 * back to caller (i.e. rewind info, missing devices, etc).
3911 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
3912 spa->spa_load_info) == 0);
3914 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3916 * Toss any existing sparelist, as it doesn't have any validity
3917 * anymore, and conflicts with spa_has_spare().
3919 if (spa->spa_spares.sav_config) {
3920 nvlist_free(spa->spa_spares.sav_config);
3921 spa->spa_spares.sav_config = NULL;
3922 spa_load_spares(spa);
3924 if (spa->spa_l2cache.sav_config) {
3925 nvlist_free(spa->spa_l2cache.sav_config);
3926 spa->spa_l2cache.sav_config = NULL;
3927 spa_load_l2cache(spa);
3930 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3933 error = spa_validate_aux(spa, nvroot, -1ULL,
3936 error = spa_validate_aux(spa, nvroot, -1ULL,
3937 VDEV_ALLOC_L2CACHE);
3938 spa_config_exit(spa, SCL_ALL, FTAG);
3941 spa_configfile_set(spa, props, B_FALSE);
3943 if (error != 0 || (props && spa_writeable(spa) &&
3944 (error = spa_prop_set(spa, props)))) {
3946 spa_deactivate(spa);
3948 mutex_exit(&spa_namespace_lock);
3952 spa_async_resume(spa);
3955 * Override any spares and level 2 cache devices as specified by
3956 * the user, as these may have correct device names/devids, etc.
3958 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3959 &spares, &nspares) == 0) {
3960 if (spa->spa_spares.sav_config)
3961 VERIFY(nvlist_remove(spa->spa_spares.sav_config,
3962 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
3964 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
3965 NV_UNIQUE_NAME, KM_PUSHPAGE) == 0);
3966 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3967 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3968 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3969 spa_load_spares(spa);
3970 spa_config_exit(spa, SCL_ALL, FTAG);
3971 spa->spa_spares.sav_sync = B_TRUE;
3973 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3974 &l2cache, &nl2cache) == 0) {
3975 if (spa->spa_l2cache.sav_config)
3976 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
3977 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
3979 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3980 NV_UNIQUE_NAME, KM_PUSHPAGE) == 0);
3981 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3982 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3983 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3984 spa_load_l2cache(spa);
3985 spa_config_exit(spa, SCL_ALL, FTAG);
3986 spa->spa_l2cache.sav_sync = B_TRUE;
3990 * Check for any removed devices.
3992 if (spa->spa_autoreplace) {
3993 spa_aux_check_removed(&spa->spa_spares);
3994 spa_aux_check_removed(&spa->spa_l2cache);
3997 if (spa_writeable(spa)) {
3999 * Update the config cache to include the newly-imported pool.
4001 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4005 * It's possible that the pool was expanded while it was exported.
4006 * We kick off an async task to handle this for us.
4008 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
4010 mutex_exit(&spa_namespace_lock);
4011 spa_history_log_version(spa, LOG_POOL_IMPORT);
4017 spa_tryimport(nvlist_t *tryconfig)
4019 nvlist_t *config = NULL;
4025 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
4028 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
4032 * Create and initialize the spa structure.
4034 mutex_enter(&spa_namespace_lock);
4035 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
4036 spa_activate(spa, FREAD);
4039 * Pass off the heavy lifting to spa_load().
4040 * Pass TRUE for mosconfig because the user-supplied config
4041 * is actually the one to trust when doing an import.
4043 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
4046 * If 'tryconfig' was at least parsable, return the current config.
4048 if (spa->spa_root_vdev != NULL) {
4049 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4050 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
4052 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4054 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
4055 spa->spa_uberblock.ub_timestamp) == 0);
4056 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4057 spa->spa_load_info) == 0);
4060 * If the bootfs property exists on this pool then we
4061 * copy it out so that external consumers can tell which
4062 * pools are bootable.
4064 if ((!error || error == EEXIST) && spa->spa_bootfs) {
4065 char *tmpname = kmem_alloc(MAXPATHLEN, KM_PUSHPAGE);
4068 * We have to play games with the name since the
4069 * pool was opened as TRYIMPORT_NAME.
4071 if (dsl_dsobj_to_dsname(spa_name(spa),
4072 spa->spa_bootfs, tmpname) == 0) {
4074 char *dsname = kmem_alloc(MAXPATHLEN, KM_PUSHPAGE);
4076 cp = strchr(tmpname, '/');
4078 (void) strlcpy(dsname, tmpname,
4081 (void) snprintf(dsname, MAXPATHLEN,
4082 "%s/%s", poolname, ++cp);
4084 VERIFY(nvlist_add_string(config,
4085 ZPOOL_CONFIG_BOOTFS, dsname) == 0);
4086 kmem_free(dsname, MAXPATHLEN);
4088 kmem_free(tmpname, MAXPATHLEN);
4092 * Add the list of hot spares and level 2 cache devices.
4094 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4095 spa_add_spares(spa, config);
4096 spa_add_l2cache(spa, config);
4097 spa_config_exit(spa, SCL_CONFIG, FTAG);
4101 spa_deactivate(spa);
4103 mutex_exit(&spa_namespace_lock);
4109 * Pool export/destroy
4111 * The act of destroying or exporting a pool is very simple. We make sure there
4112 * is no more pending I/O and any references to the pool are gone. Then, we
4113 * update the pool state and sync all the labels to disk, removing the
4114 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4115 * we don't sync the labels or remove the configuration cache.
4118 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
4119 boolean_t force, boolean_t hardforce)
4126 if (!(spa_mode_global & FWRITE))
4129 mutex_enter(&spa_namespace_lock);
4130 if ((spa = spa_lookup(pool)) == NULL) {
4131 mutex_exit(&spa_namespace_lock);
4136 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4137 * reacquire the namespace lock, and see if we can export.
4139 spa_open_ref(spa, FTAG);
4140 mutex_exit(&spa_namespace_lock);
4141 spa_async_suspend(spa);
4142 mutex_enter(&spa_namespace_lock);
4143 spa_close(spa, FTAG);
4146 * The pool will be in core if it's openable,
4147 * in which case we can modify its state.
4149 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
4151 * Objsets may be open only because they're dirty, so we
4152 * have to force it to sync before checking spa_refcnt.
4154 txg_wait_synced(spa->spa_dsl_pool, 0);
4157 * A pool cannot be exported or destroyed if there are active
4158 * references. If we are resetting a pool, allow references by
4159 * fault injection handlers.
4161 if (!spa_refcount_zero(spa) ||
4162 (spa->spa_inject_ref != 0 &&
4163 new_state != POOL_STATE_UNINITIALIZED)) {
4164 spa_async_resume(spa);
4165 mutex_exit(&spa_namespace_lock);
4170 * A pool cannot be exported if it has an active shared spare.
4171 * This is to prevent other pools stealing the active spare
4172 * from an exported pool. At user's own will, such pool can
4173 * be forcedly exported.
4175 if (!force && new_state == POOL_STATE_EXPORTED &&
4176 spa_has_active_shared_spare(spa)) {
4177 spa_async_resume(spa);
4178 mutex_exit(&spa_namespace_lock);
4183 * We want this to be reflected on every label,
4184 * so mark them all dirty. spa_unload() will do the
4185 * final sync that pushes these changes out.
4187 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
4188 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4189 spa->spa_state = new_state;
4190 spa->spa_final_txg = spa_last_synced_txg(spa) +
4192 vdev_config_dirty(spa->spa_root_vdev);
4193 spa_config_exit(spa, SCL_ALL, FTAG);
4197 spa_event_notify(spa, NULL, FM_EREPORT_ZFS_POOL_DESTROY);
4199 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4201 spa_deactivate(spa);
4204 if (oldconfig && spa->spa_config)
4205 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
4207 if (new_state != POOL_STATE_UNINITIALIZED) {
4209 spa_config_sync(spa, B_TRUE, B_TRUE);
4212 mutex_exit(&spa_namespace_lock);
4218 * Destroy a storage pool.
4221 spa_destroy(char *pool)
4223 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
4228 * Export a storage pool.
4231 spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
4232 boolean_t hardforce)
4234 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
4239 * Similar to spa_export(), this unloads the spa_t without actually removing it
4240 * from the namespace in any way.
4243 spa_reset(char *pool)
4245 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
4250 * ==========================================================================
4251 * Device manipulation
4252 * ==========================================================================
4256 * Add a device to a storage pool.
4259 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
4263 vdev_t *rvd = spa->spa_root_vdev;
4265 nvlist_t **spares, **l2cache;
4266 uint_t nspares, nl2cache;
4269 ASSERT(spa_writeable(spa));
4271 txg = spa_vdev_enter(spa);
4273 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
4274 VDEV_ALLOC_ADD)) != 0)
4275 return (spa_vdev_exit(spa, NULL, txg, error));
4277 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */
4279 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
4283 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
4287 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
4288 return (spa_vdev_exit(spa, vd, txg, EINVAL));
4290 if (vd->vdev_children != 0 &&
4291 (error = vdev_create(vd, txg, B_FALSE)) != 0)
4292 return (spa_vdev_exit(spa, vd, txg, error));
4295 * We must validate the spares and l2cache devices after checking the
4296 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
4298 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
4299 return (spa_vdev_exit(spa, vd, txg, error));
4302 * Transfer each new top-level vdev from vd to rvd.
4304 for (c = 0; c < vd->vdev_children; c++) {
4307 * Set the vdev id to the first hole, if one exists.
4309 for (id = 0; id < rvd->vdev_children; id++) {
4310 if (rvd->vdev_child[id]->vdev_ishole) {
4311 vdev_free(rvd->vdev_child[id]);
4315 tvd = vd->vdev_child[c];
4316 vdev_remove_child(vd, tvd);
4318 vdev_add_child(rvd, tvd);
4319 vdev_config_dirty(tvd);
4323 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
4324 ZPOOL_CONFIG_SPARES);
4325 spa_load_spares(spa);
4326 spa->spa_spares.sav_sync = B_TRUE;
4329 if (nl2cache != 0) {
4330 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
4331 ZPOOL_CONFIG_L2CACHE);
4332 spa_load_l2cache(spa);
4333 spa->spa_l2cache.sav_sync = B_TRUE;
4337 * We have to be careful when adding new vdevs to an existing pool.
4338 * If other threads start allocating from these vdevs before we
4339 * sync the config cache, and we lose power, then upon reboot we may
4340 * fail to open the pool because there are DVAs that the config cache
4341 * can't translate. Therefore, we first add the vdevs without
4342 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4343 * and then let spa_config_update() initialize the new metaslabs.
4345 * spa_load() checks for added-but-not-initialized vdevs, so that
4346 * if we lose power at any point in this sequence, the remaining
4347 * steps will be completed the next time we load the pool.
4349 (void) spa_vdev_exit(spa, vd, txg, 0);
4351 mutex_enter(&spa_namespace_lock);
4352 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4353 mutex_exit(&spa_namespace_lock);
4359 * Attach a device to a mirror. The arguments are the path to any device
4360 * in the mirror, and the nvroot for the new device. If the path specifies
4361 * a device that is not mirrored, we automatically insert the mirror vdev.
4363 * If 'replacing' is specified, the new device is intended to replace the
4364 * existing device; in this case the two devices are made into their own
4365 * mirror using the 'replacing' vdev, which is functionally identical to
4366 * the mirror vdev (it actually reuses all the same ops) but has a few
4367 * extra rules: you can't attach to it after it's been created, and upon
4368 * completion of resilvering, the first disk (the one being replaced)
4369 * is automatically detached.
4372 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
4374 uint64_t txg, dtl_max_txg;
4375 ASSERTV(vdev_t *rvd = spa->spa_root_vdev;)
4376 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
4378 char *oldvdpath, *newvdpath;
4382 ASSERT(spa_writeable(spa));
4384 txg = spa_vdev_enter(spa);
4386 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
4389 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4391 if (!oldvd->vdev_ops->vdev_op_leaf)
4392 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4394 pvd = oldvd->vdev_parent;
4396 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
4397 VDEV_ALLOC_ATTACH)) != 0)
4398 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4400 if (newrootvd->vdev_children != 1)
4401 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4403 newvd = newrootvd->vdev_child[0];
4405 if (!newvd->vdev_ops->vdev_op_leaf)
4406 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4408 if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
4409 return (spa_vdev_exit(spa, newrootvd, txg, error));
4412 * Spares can't replace logs
4414 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
4415 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4419 * For attach, the only allowable parent is a mirror or the root
4422 if (pvd->vdev_ops != &vdev_mirror_ops &&
4423 pvd->vdev_ops != &vdev_root_ops)
4424 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4426 pvops = &vdev_mirror_ops;
4429 * Active hot spares can only be replaced by inactive hot
4432 if (pvd->vdev_ops == &vdev_spare_ops &&
4433 oldvd->vdev_isspare &&
4434 !spa_has_spare(spa, newvd->vdev_guid))
4435 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4438 * If the source is a hot spare, and the parent isn't already a
4439 * spare, then we want to create a new hot spare. Otherwise, we
4440 * want to create a replacing vdev. The user is not allowed to
4441 * attach to a spared vdev child unless the 'isspare' state is
4442 * the same (spare replaces spare, non-spare replaces
4445 if (pvd->vdev_ops == &vdev_replacing_ops &&
4446 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
4447 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4448 } else if (pvd->vdev_ops == &vdev_spare_ops &&
4449 newvd->vdev_isspare != oldvd->vdev_isspare) {
4450 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4453 if (newvd->vdev_isspare)
4454 pvops = &vdev_spare_ops;
4456 pvops = &vdev_replacing_ops;
4460 * Make sure the new device is big enough.
4462 if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
4463 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
4466 * The new device cannot have a higher alignment requirement
4467 * than the top-level vdev.
4469 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4470 return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4473 * If this is an in-place replacement, update oldvd's path and devid
4474 * to make it distinguishable from newvd, and unopenable from now on.
4476 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4477 spa_strfree(oldvd->vdev_path);
4478 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4480 (void) sprintf(oldvd->vdev_path, "%s/%s",
4481 newvd->vdev_path, "old");
4482 if (oldvd->vdev_devid != NULL) {
4483 spa_strfree(oldvd->vdev_devid);
4484 oldvd->vdev_devid = NULL;
4488 /* mark the device being resilvered */
4489 newvd->vdev_resilvering = B_TRUE;
4492 * If the parent is not a mirror, or if we're replacing, insert the new
4493 * mirror/replacing/spare vdev above oldvd.
4495 if (pvd->vdev_ops != pvops)
4496 pvd = vdev_add_parent(oldvd, pvops);
4498 ASSERT(pvd->vdev_top->vdev_parent == rvd);
4499 ASSERT(pvd->vdev_ops == pvops);
4500 ASSERT(oldvd->vdev_parent == pvd);
4503 * Extract the new device from its root and add it to pvd.
4505 vdev_remove_child(newrootvd, newvd);
4506 newvd->vdev_id = pvd->vdev_children;
4507 newvd->vdev_crtxg = oldvd->vdev_crtxg;
4508 vdev_add_child(pvd, newvd);
4510 tvd = newvd->vdev_top;
4511 ASSERT(pvd->vdev_top == tvd);
4512 ASSERT(tvd->vdev_parent == rvd);
4514 vdev_config_dirty(tvd);
4517 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4518 * for any dmu_sync-ed blocks. It will propagate upward when
4519 * spa_vdev_exit() calls vdev_dtl_reassess().
4521 dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4523 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4524 dtl_max_txg - TXG_INITIAL);
4526 if (newvd->vdev_isspare) {
4527 spa_spare_activate(newvd);
4528 spa_event_notify(spa, newvd, FM_EREPORT_ZFS_DEVICE_SPARE);
4531 oldvdpath = spa_strdup(oldvd->vdev_path);
4532 newvdpath = spa_strdup(newvd->vdev_path);
4533 newvd_isspare = newvd->vdev_isspare;
4536 * Mark newvd's DTL dirty in this txg.
4538 vdev_dirty(tvd, VDD_DTL, newvd, txg);
4541 * Restart the resilver
4543 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4548 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4550 spa_history_log_internal(LOG_POOL_VDEV_ATTACH, spa, NULL,
4551 "%s vdev=%s %s vdev=%s",
4552 replacing && newvd_isspare ? "spare in" :
4553 replacing ? "replace" : "attach", newvdpath,
4554 replacing ? "for" : "to", oldvdpath);
4556 spa_strfree(oldvdpath);
4557 spa_strfree(newvdpath);
4559 if (spa->spa_bootfs)
4560 spa_event_notify(spa, newvd, FM_EREPORT_ZFS_BOOTFS_VDEV_ATTACH);
4566 * Detach a device from a mirror or replacing vdev.
4567 * If 'replace_done' is specified, only detach if the parent
4568 * is a replacing vdev.
4571 spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4575 ASSERTV(vdev_t *rvd = spa->spa_root_vdev;)
4576 vdev_t *vd, *pvd, *cvd, *tvd;
4577 boolean_t unspare = B_FALSE;
4578 uint64_t unspare_guid = 0;
4582 ASSERT(spa_writeable(spa));
4584 txg = spa_vdev_enter(spa);
4586 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4589 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4591 if (!vd->vdev_ops->vdev_op_leaf)
4592 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4594 pvd = vd->vdev_parent;
4597 * If the parent/child relationship is not as expected, don't do it.
4598 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4599 * vdev that's replacing B with C. The user's intent in replacing
4600 * is to go from M(A,B) to M(A,C). If the user decides to cancel
4601 * the replace by detaching C, the expected behavior is to end up
4602 * M(A,B). But suppose that right after deciding to detach C,
4603 * the replacement of B completes. We would have M(A,C), and then
4604 * ask to detach C, which would leave us with just A -- not what
4605 * the user wanted. To prevent this, we make sure that the
4606 * parent/child relationship hasn't changed -- in this example,
4607 * that C's parent is still the replacing vdev R.
4609 if (pvd->vdev_guid != pguid && pguid != 0)
4610 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4613 * Only 'replacing' or 'spare' vdevs can be replaced.
4615 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4616 pvd->vdev_ops != &vdev_spare_ops)
4617 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4619 ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4620 spa_version(spa) >= SPA_VERSION_SPARES);
4623 * Only mirror, replacing, and spare vdevs support detach.
4625 if (pvd->vdev_ops != &vdev_replacing_ops &&
4626 pvd->vdev_ops != &vdev_mirror_ops &&
4627 pvd->vdev_ops != &vdev_spare_ops)
4628 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4631 * If this device has the only valid copy of some data,
4632 * we cannot safely detach it.
4634 if (vdev_dtl_required(vd))
4635 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4637 ASSERT(pvd->vdev_children >= 2);
4640 * If we are detaching the second disk from a replacing vdev, then
4641 * check to see if we changed the original vdev's path to have "/old"
4642 * at the end in spa_vdev_attach(). If so, undo that change now.
4644 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4645 vd->vdev_path != NULL) {
4646 size_t len = strlen(vd->vdev_path);
4648 for (c = 0; c < pvd->vdev_children; c++) {
4649 cvd = pvd->vdev_child[c];
4651 if (cvd == vd || cvd->vdev_path == NULL)
4654 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4655 strcmp(cvd->vdev_path + len, "/old") == 0) {
4656 spa_strfree(cvd->vdev_path);
4657 cvd->vdev_path = spa_strdup(vd->vdev_path);
4664 * If we are detaching the original disk from a spare, then it implies
4665 * that the spare should become a real disk, and be removed from the
4666 * active spare list for the pool.
4668 if (pvd->vdev_ops == &vdev_spare_ops &&
4670 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4674 * Erase the disk labels so the disk can be used for other things.
4675 * This must be done after all other error cases are handled,
4676 * but before we disembowel vd (so we can still do I/O to it).
4677 * But if we can't do it, don't treat the error as fatal --
4678 * it may be that the unwritability of the disk is the reason
4679 * it's being detached!
4681 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4684 * Remove vd from its parent and compact the parent's children.
4686 vdev_remove_child(pvd, vd);
4687 vdev_compact_children(pvd);
4690 * Remember one of the remaining children so we can get tvd below.
4692 cvd = pvd->vdev_child[pvd->vdev_children - 1];
4695 * If we need to remove the remaining child from the list of hot spares,
4696 * do it now, marking the vdev as no longer a spare in the process.
4697 * We must do this before vdev_remove_parent(), because that can
4698 * change the GUID if it creates a new toplevel GUID. For a similar
4699 * reason, we must remove the spare now, in the same txg as the detach;
4700 * otherwise someone could attach a new sibling, change the GUID, and
4701 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4704 ASSERT(cvd->vdev_isspare);
4705 spa_spare_remove(cvd);
4706 unspare_guid = cvd->vdev_guid;
4707 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4708 cvd->vdev_unspare = B_TRUE;
4712 * If the parent mirror/replacing vdev only has one child,
4713 * the parent is no longer needed. Remove it from the tree.
4715 if (pvd->vdev_children == 1) {
4716 if (pvd->vdev_ops == &vdev_spare_ops)
4717 cvd->vdev_unspare = B_FALSE;
4718 vdev_remove_parent(cvd);
4719 cvd->vdev_resilvering = B_FALSE;
4724 * We don't set tvd until now because the parent we just removed
4725 * may have been the previous top-level vdev.
4727 tvd = cvd->vdev_top;
4728 ASSERT(tvd->vdev_parent == rvd);
4731 * Reevaluate the parent vdev state.
4733 vdev_propagate_state(cvd);
4736 * If the 'autoexpand' property is set on the pool then automatically
4737 * try to expand the size of the pool. For example if the device we
4738 * just detached was smaller than the others, it may be possible to
4739 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4740 * first so that we can obtain the updated sizes of the leaf vdevs.
4742 if (spa->spa_autoexpand) {
4744 vdev_expand(tvd, txg);
4747 vdev_config_dirty(tvd);
4750 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
4751 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4752 * But first make sure we're not on any *other* txg's DTL list, to
4753 * prevent vd from being accessed after it's freed.
4755 vdpath = spa_strdup(vd->vdev_path);
4756 for (t = 0; t < TXG_SIZE; t++)
4757 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4758 vd->vdev_detached = B_TRUE;
4759 vdev_dirty(tvd, VDD_DTL, vd, txg);
4761 spa_event_notify(spa, vd, FM_EREPORT_ZFS_DEVICE_REMOVE);
4763 /* hang on to the spa before we release the lock */
4764 spa_open_ref(spa, FTAG);
4766 error = spa_vdev_exit(spa, vd, txg, 0);
4768 spa_history_log_internal(LOG_POOL_VDEV_DETACH, spa, NULL,
4770 spa_strfree(vdpath);
4773 * If this was the removal of the original device in a hot spare vdev,
4774 * then we want to go through and remove the device from the hot spare
4775 * list of every other pool.
4778 spa_t *altspa = NULL;
4780 mutex_enter(&spa_namespace_lock);
4781 while ((altspa = spa_next(altspa)) != NULL) {
4782 if (altspa->spa_state != POOL_STATE_ACTIVE ||
4786 spa_open_ref(altspa, FTAG);
4787 mutex_exit(&spa_namespace_lock);
4788 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
4789 mutex_enter(&spa_namespace_lock);
4790 spa_close(altspa, FTAG);
4792 mutex_exit(&spa_namespace_lock);
4794 /* search the rest of the vdevs for spares to remove */
4795 spa_vdev_resilver_done(spa);
4798 /* all done with the spa; OK to release */
4799 mutex_enter(&spa_namespace_lock);
4800 spa_close(spa, FTAG);
4801 mutex_exit(&spa_namespace_lock);
4807 * Split a set of devices from their mirrors, and create a new pool from them.
4810 spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
4811 nvlist_t *props, boolean_t exp)
4814 uint64_t txg, *glist;
4816 uint_t c, children, lastlog;
4817 nvlist_t **child, *nvl, *tmp;
4819 char *altroot = NULL;
4820 vdev_t *rvd, **vml = NULL; /* vdev modify list */
4821 boolean_t activate_slog;
4823 ASSERT(spa_writeable(spa));
4825 txg = spa_vdev_enter(spa);
4827 /* clear the log and flush everything up to now */
4828 activate_slog = spa_passivate_log(spa);
4829 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4830 error = spa_offline_log(spa);
4831 txg = spa_vdev_config_enter(spa);
4834 spa_activate_log(spa);
4837 return (spa_vdev_exit(spa, NULL, txg, error));
4839 /* check new spa name before going any further */
4840 if (spa_lookup(newname) != NULL)
4841 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
4844 * scan through all the children to ensure they're all mirrors
4846 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
4847 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
4849 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4851 /* first, check to ensure we've got the right child count */
4852 rvd = spa->spa_root_vdev;
4854 for (c = 0; c < rvd->vdev_children; c++) {
4855 vdev_t *vd = rvd->vdev_child[c];
4857 /* don't count the holes & logs as children */
4858 if (vd->vdev_islog || vd->vdev_ishole) {
4866 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
4867 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4869 /* next, ensure no spare or cache devices are part of the split */
4870 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
4871 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
4872 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4874 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_PUSHPAGE);
4875 glist = kmem_zalloc(children * sizeof (uint64_t), KM_PUSHPAGE);
4877 /* then, loop over each vdev and validate it */
4878 for (c = 0; c < children; c++) {
4879 uint64_t is_hole = 0;
4881 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
4885 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
4886 spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
4894 /* which disk is going to be split? */
4895 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
4901 /* look it up in the spa */
4902 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
4903 if (vml[c] == NULL) {
4908 /* make sure there's nothing stopping the split */
4909 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
4910 vml[c]->vdev_islog ||
4911 vml[c]->vdev_ishole ||
4912 vml[c]->vdev_isspare ||
4913 vml[c]->vdev_isl2cache ||
4914 !vdev_writeable(vml[c]) ||
4915 vml[c]->vdev_children != 0 ||
4916 vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
4917 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
4922 if (vdev_dtl_required(vml[c])) {
4927 /* we need certain info from the top level */
4928 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
4929 vml[c]->vdev_top->vdev_ms_array) == 0);
4930 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
4931 vml[c]->vdev_top->vdev_ms_shift) == 0);
4932 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
4933 vml[c]->vdev_top->vdev_asize) == 0);
4934 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
4935 vml[c]->vdev_top->vdev_ashift) == 0);
4939 kmem_free(vml, children * sizeof (vdev_t *));
4940 kmem_free(glist, children * sizeof (uint64_t));
4941 return (spa_vdev_exit(spa, NULL, txg, error));
4944 /* stop writers from using the disks */
4945 for (c = 0; c < children; c++) {
4947 vml[c]->vdev_offline = B_TRUE;
4949 vdev_reopen(spa->spa_root_vdev);
4952 * Temporarily record the splitting vdevs in the spa config. This
4953 * will disappear once the config is regenerated.
4955 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_PUSHPAGE) == 0);
4956 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
4957 glist, children) == 0);
4958 kmem_free(glist, children * sizeof (uint64_t));
4960 mutex_enter(&spa->spa_props_lock);
4961 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
4963 mutex_exit(&spa->spa_props_lock);
4964 spa->spa_config_splitting = nvl;
4965 vdev_config_dirty(spa->spa_root_vdev);
4967 /* configure and create the new pool */
4968 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
4969 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4970 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
4971 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
4972 spa_version(spa)) == 0);
4973 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
4974 spa->spa_config_txg) == 0);
4975 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4976 spa_generate_guid(NULL)) == 0);
4977 (void) nvlist_lookup_string(props,
4978 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4980 /* add the new pool to the namespace */
4981 newspa = spa_add(newname, config, altroot);
4982 newspa->spa_config_txg = spa->spa_config_txg;
4983 spa_set_log_state(newspa, SPA_LOG_CLEAR);
4985 /* release the spa config lock, retaining the namespace lock */
4986 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4988 if (zio_injection_enabled)
4989 zio_handle_panic_injection(spa, FTAG, 1);
4991 spa_activate(newspa, spa_mode_global);
4992 spa_async_suspend(newspa);
4994 /* create the new pool from the disks of the original pool */
4995 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
4999 /* if that worked, generate a real config for the new pool */
5000 if (newspa->spa_root_vdev != NULL) {
5001 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
5002 NV_UNIQUE_NAME, KM_PUSHPAGE) == 0);
5003 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
5004 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
5005 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
5010 if (props != NULL) {
5011 spa_configfile_set(newspa, props, B_FALSE);
5012 error = spa_prop_set(newspa, props);
5017 /* flush everything */
5018 txg = spa_vdev_config_enter(newspa);
5019 vdev_config_dirty(newspa->spa_root_vdev);
5020 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
5022 if (zio_injection_enabled)
5023 zio_handle_panic_injection(spa, FTAG, 2);
5025 spa_async_resume(newspa);
5027 /* finally, update the original pool's config */
5028 txg = spa_vdev_config_enter(spa);
5029 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
5030 error = dmu_tx_assign(tx, TXG_WAIT);
5033 for (c = 0; c < children; c++) {
5034 if (vml[c] != NULL) {
5037 spa_history_log_internal(LOG_POOL_VDEV_DETACH,
5043 vdev_config_dirty(spa->spa_root_vdev);
5044 spa->spa_config_splitting = NULL;
5048 (void) spa_vdev_exit(spa, NULL, txg, 0);
5050 if (zio_injection_enabled)
5051 zio_handle_panic_injection(spa, FTAG, 3);
5053 /* split is complete; log a history record */
5054 spa_history_log_internal(LOG_POOL_SPLIT, newspa, NULL,
5055 "split new pool %s from pool %s", newname, spa_name(spa));
5057 kmem_free(vml, children * sizeof (vdev_t *));
5059 /* if we're not going to mount the filesystems in userland, export */
5061 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
5068 spa_deactivate(newspa);
5071 txg = spa_vdev_config_enter(spa);
5073 /* re-online all offlined disks */
5074 for (c = 0; c < children; c++) {
5076 vml[c]->vdev_offline = B_FALSE;
5078 vdev_reopen(spa->spa_root_vdev);
5080 nvlist_free(spa->spa_config_splitting);
5081 spa->spa_config_splitting = NULL;
5082 (void) spa_vdev_exit(spa, NULL, txg, error);
5084 kmem_free(vml, children * sizeof (vdev_t *));
5089 spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
5093 for (i = 0; i < count; i++) {
5096 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
5099 if (guid == target_guid)
5107 spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
5108 nvlist_t *dev_to_remove)
5110 nvlist_t **newdev = NULL;
5114 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_PUSHPAGE);
5116 for (i = 0, j = 0; i < count; i++) {
5117 if (dev[i] == dev_to_remove)
5119 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_PUSHPAGE) == 0);
5122 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
5123 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
5125 for (i = 0; i < count - 1; i++)
5126 nvlist_free(newdev[i]);
5129 kmem_free(newdev, (count - 1) * sizeof (void *));
5133 * Evacuate the device.
5136 spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
5141 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5142 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5143 ASSERT(vd == vd->vdev_top);
5146 * Evacuate the device. We don't hold the config lock as writer
5147 * since we need to do I/O but we do keep the
5148 * spa_namespace_lock held. Once this completes the device
5149 * should no longer have any blocks allocated on it.
5151 if (vd->vdev_islog) {
5152 if (vd->vdev_stat.vs_alloc != 0)
5153 error = spa_offline_log(spa);
5162 * The evacuation succeeded. Remove any remaining MOS metadata
5163 * associated with this vdev, and wait for these changes to sync.
5165 ASSERT0(vd->vdev_stat.vs_alloc);
5166 txg = spa_vdev_config_enter(spa);
5167 vd->vdev_removing = B_TRUE;
5168 vdev_dirty(vd, 0, NULL, txg);
5169 vdev_config_dirty(vd);
5170 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5176 * Complete the removal by cleaning up the namespace.
5179 spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
5181 vdev_t *rvd = spa->spa_root_vdev;
5182 uint64_t id = vd->vdev_id;
5183 boolean_t last_vdev = (id == (rvd->vdev_children - 1));
5185 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5186 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5187 ASSERT(vd == vd->vdev_top);
5190 * Only remove any devices which are empty.
5192 if (vd->vdev_stat.vs_alloc != 0)
5195 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5197 if (list_link_active(&vd->vdev_state_dirty_node))
5198 vdev_state_clean(vd);
5199 if (list_link_active(&vd->vdev_config_dirty_node))
5200 vdev_config_clean(vd);
5205 vdev_compact_children(rvd);
5207 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
5208 vdev_add_child(rvd, vd);
5210 vdev_config_dirty(rvd);
5213 * Reassess the health of our root vdev.
5219 * Remove a device from the pool -
5221 * Removing a device from the vdev namespace requires several steps
5222 * and can take a significant amount of time. As a result we use
5223 * the spa_vdev_config_[enter/exit] functions which allow us to
5224 * grab and release the spa_config_lock while still holding the namespace
5225 * lock. During each step the configuration is synced out.
5229 * Remove a device from the pool. Currently, this supports removing only hot
5230 * spares, slogs, and level 2 ARC devices.
5233 spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
5236 metaslab_group_t *mg;
5237 nvlist_t **spares, **l2cache, *nv;
5239 uint_t nspares, nl2cache;
5241 boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
5243 ASSERT(spa_writeable(spa));
5246 txg = spa_vdev_enter(spa);
5248 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5250 if (spa->spa_spares.sav_vdevs != NULL &&
5251 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5252 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
5253 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
5255 * Only remove the hot spare if it's not currently in use
5258 if (vd == NULL || unspare) {
5259 spa_vdev_remove_aux(spa->spa_spares.sav_config,
5260 ZPOOL_CONFIG_SPARES, spares, nspares, nv);
5261 spa_load_spares(spa);
5262 spa->spa_spares.sav_sync = B_TRUE;
5266 } else if (spa->spa_l2cache.sav_vdevs != NULL &&
5267 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5268 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
5269 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
5271 * Cache devices can always be removed.
5273 spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
5274 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
5275 spa_load_l2cache(spa);
5276 spa->spa_l2cache.sav_sync = B_TRUE;
5277 } else if (vd != NULL && vd->vdev_islog) {
5279 ASSERT(vd == vd->vdev_top);
5282 * XXX - Once we have bp-rewrite this should
5283 * become the common case.
5289 * Stop allocating from this vdev.
5291 metaslab_group_passivate(mg);
5294 * Wait for the youngest allocations and frees to sync,
5295 * and then wait for the deferral of those frees to finish.
5297 spa_vdev_config_exit(spa, NULL,
5298 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
5301 * Attempt to evacuate the vdev.
5303 error = spa_vdev_remove_evacuate(spa, vd);
5305 txg = spa_vdev_config_enter(spa);
5308 * If we couldn't evacuate the vdev, unwind.
5311 metaslab_group_activate(mg);
5312 return (spa_vdev_exit(spa, NULL, txg, error));
5316 * Clean up the vdev namespace.
5318 spa_vdev_remove_from_namespace(spa, vd);
5320 } else if (vd != NULL) {
5322 * Normal vdevs cannot be removed (yet).
5327 * There is no vdev of any kind with the specified guid.
5333 return (spa_vdev_exit(spa, NULL, txg, error));
5339 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5340 * current spared, so we can detach it.
5343 spa_vdev_resilver_done_hunt(vdev_t *vd)
5345 vdev_t *newvd, *oldvd;
5348 for (c = 0; c < vd->vdev_children; c++) {
5349 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
5355 * Check for a completed replacement. We always consider the first
5356 * vdev in the list to be the oldest vdev, and the last one to be
5357 * the newest (see spa_vdev_attach() for how that works). In
5358 * the case where the newest vdev is faulted, we will not automatically
5359 * remove it after a resilver completes. This is OK as it will require
5360 * user intervention to determine which disk the admin wishes to keep.
5362 if (vd->vdev_ops == &vdev_replacing_ops) {
5363 ASSERT(vd->vdev_children > 1);
5365 newvd = vd->vdev_child[vd->vdev_children - 1];
5366 oldvd = vd->vdev_child[0];
5368 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
5369 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5370 !vdev_dtl_required(oldvd))
5375 * Check for a completed resilver with the 'unspare' flag set.
5377 if (vd->vdev_ops == &vdev_spare_ops) {
5378 vdev_t *first = vd->vdev_child[0];
5379 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
5381 if (last->vdev_unspare) {
5384 } else if (first->vdev_unspare) {
5391 if (oldvd != NULL &&
5392 vdev_dtl_empty(newvd, DTL_MISSING) &&
5393 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5394 !vdev_dtl_required(oldvd))
5398 * If there are more than two spares attached to a disk,
5399 * and those spares are not required, then we want to
5400 * attempt to free them up now so that they can be used
5401 * by other pools. Once we're back down to a single
5402 * disk+spare, we stop removing them.
5404 if (vd->vdev_children > 2) {
5405 newvd = vd->vdev_child[1];
5407 if (newvd->vdev_isspare && last->vdev_isspare &&
5408 vdev_dtl_empty(last, DTL_MISSING) &&
5409 vdev_dtl_empty(last, DTL_OUTAGE) &&
5410 !vdev_dtl_required(newvd))
5419 spa_vdev_resilver_done(spa_t *spa)
5421 vdev_t *vd, *pvd, *ppvd;
5422 uint64_t guid, sguid, pguid, ppguid;
5424 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5426 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
5427 pvd = vd->vdev_parent;
5428 ppvd = pvd->vdev_parent;
5429 guid = vd->vdev_guid;
5430 pguid = pvd->vdev_guid;
5431 ppguid = ppvd->vdev_guid;
5434 * If we have just finished replacing a hot spared device, then
5435 * we need to detach the parent's first child (the original hot
5438 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
5439 ppvd->vdev_children == 2) {
5440 ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
5441 sguid = ppvd->vdev_child[1]->vdev_guid;
5443 spa_config_exit(spa, SCL_ALL, FTAG);
5444 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
5446 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
5448 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5451 spa_config_exit(spa, SCL_ALL, FTAG);
5455 * Update the stored path or FRU for this vdev.
5458 spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
5462 boolean_t sync = B_FALSE;
5464 ASSERT(spa_writeable(spa));
5466 spa_vdev_state_enter(spa, SCL_ALL);
5468 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
5469 return (spa_vdev_state_exit(spa, NULL, ENOENT));
5471 if (!vd->vdev_ops->vdev_op_leaf)
5472 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
5475 if (strcmp(value, vd->vdev_path) != 0) {
5476 spa_strfree(vd->vdev_path);
5477 vd->vdev_path = spa_strdup(value);
5481 if (vd->vdev_fru == NULL) {
5482 vd->vdev_fru = spa_strdup(value);
5484 } else if (strcmp(value, vd->vdev_fru) != 0) {
5485 spa_strfree(vd->vdev_fru);
5486 vd->vdev_fru = spa_strdup(value);
5491 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
5495 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
5497 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5501 spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5503 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5507 * ==========================================================================
5509 * ==========================================================================
5513 spa_scan_stop(spa_t *spa)
5515 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5516 if (dsl_scan_resilvering(spa->spa_dsl_pool))
5518 return (dsl_scan_cancel(spa->spa_dsl_pool));
5522 spa_scan(spa_t *spa, pool_scan_func_t func)
5524 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5526 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5530 * If a resilver was requested, but there is no DTL on a
5531 * writeable leaf device, we have nothing to do.
5533 if (func == POOL_SCAN_RESILVER &&
5534 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5535 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5539 return (dsl_scan(spa->spa_dsl_pool, func));
5543 * ==========================================================================
5544 * SPA async task processing
5545 * ==========================================================================
5549 spa_async_remove(spa_t *spa, vdev_t *vd)
5553 if (vd->vdev_remove_wanted) {
5554 vd->vdev_remove_wanted = B_FALSE;
5555 vd->vdev_delayed_close = B_FALSE;
5556 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5559 * We want to clear the stats, but we don't want to do a full
5560 * vdev_clear() as that will cause us to throw away
5561 * degraded/faulted state as well as attempt to reopen the
5562 * device, all of which is a waste.
5564 vd->vdev_stat.vs_read_errors = 0;
5565 vd->vdev_stat.vs_write_errors = 0;
5566 vd->vdev_stat.vs_checksum_errors = 0;
5568 vdev_state_dirty(vd->vdev_top);
5571 for (c = 0; c < vd->vdev_children; c++)
5572 spa_async_remove(spa, vd->vdev_child[c]);
5576 spa_async_probe(spa_t *spa, vdev_t *vd)
5580 if (vd->vdev_probe_wanted) {
5581 vd->vdev_probe_wanted = B_FALSE;
5582 vdev_reopen(vd); /* vdev_open() does the actual probe */
5585 for (c = 0; c < vd->vdev_children; c++)
5586 spa_async_probe(spa, vd->vdev_child[c]);
5590 spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5594 if (!spa->spa_autoexpand)
5597 for (c = 0; c < vd->vdev_children; c++) {
5598 vdev_t *cvd = vd->vdev_child[c];
5599 spa_async_autoexpand(spa, cvd);
5602 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5605 spa_event_notify(vd->vdev_spa, vd, FM_EREPORT_ZFS_DEVICE_AUTOEXPAND);
5609 spa_async_thread(spa_t *spa)
5613 ASSERT(spa->spa_sync_on);
5615 mutex_enter(&spa->spa_async_lock);
5616 tasks = spa->spa_async_tasks;
5617 spa->spa_async_tasks = 0;
5618 mutex_exit(&spa->spa_async_lock);
5621 * See if the config needs to be updated.
5623 if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5624 uint64_t old_space, new_space;
5626 mutex_enter(&spa_namespace_lock);
5627 old_space = metaslab_class_get_space(spa_normal_class(spa));
5628 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5629 new_space = metaslab_class_get_space(spa_normal_class(spa));
5630 mutex_exit(&spa_namespace_lock);
5633 * If the pool grew as a result of the config update,
5634 * then log an internal history event.
5636 if (new_space != old_space) {
5637 spa_history_log_internal(LOG_POOL_VDEV_ONLINE,
5639 "pool '%s' size: %llu(+%llu)",
5640 spa_name(spa), new_space, new_space - old_space);
5645 * See if any devices need to be marked REMOVED.
5647 if (tasks & SPA_ASYNC_REMOVE) {
5648 spa_vdev_state_enter(spa, SCL_NONE);
5649 spa_async_remove(spa, spa->spa_root_vdev);
5650 for (i = 0; i < spa->spa_l2cache.sav_count; i++)
5651 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5652 for (i = 0; i < spa->spa_spares.sav_count; i++)
5653 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5654 (void) spa_vdev_state_exit(spa, NULL, 0);
5657 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5658 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5659 spa_async_autoexpand(spa, spa->spa_root_vdev);
5660 spa_config_exit(spa, SCL_CONFIG, FTAG);
5664 * See if any devices need to be probed.
5666 if (tasks & SPA_ASYNC_PROBE) {
5667 spa_vdev_state_enter(spa, SCL_NONE);
5668 spa_async_probe(spa, spa->spa_root_vdev);
5669 (void) spa_vdev_state_exit(spa, NULL, 0);
5673 * If any devices are done replacing, detach them.
5675 if (tasks & SPA_ASYNC_RESILVER_DONE)
5676 spa_vdev_resilver_done(spa);
5679 * Kick off a resilver.
5681 if (tasks & SPA_ASYNC_RESILVER)
5682 dsl_resilver_restart(spa->spa_dsl_pool, 0);
5685 * Let the world know that we're done.
5687 mutex_enter(&spa->spa_async_lock);
5688 spa->spa_async_thread = NULL;
5689 cv_broadcast(&spa->spa_async_cv);
5690 mutex_exit(&spa->spa_async_lock);
5695 spa_async_suspend(spa_t *spa)
5697 mutex_enter(&spa->spa_async_lock);
5698 spa->spa_async_suspended++;
5699 while (spa->spa_async_thread != NULL)
5700 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5701 mutex_exit(&spa->spa_async_lock);
5705 spa_async_resume(spa_t *spa)
5707 mutex_enter(&spa->spa_async_lock);
5708 ASSERT(spa->spa_async_suspended != 0);
5709 spa->spa_async_suspended--;
5710 mutex_exit(&spa->spa_async_lock);
5714 spa_async_dispatch(spa_t *spa)
5716 mutex_enter(&spa->spa_async_lock);
5717 if (spa->spa_async_tasks && !spa->spa_async_suspended &&
5718 spa->spa_async_thread == NULL &&
5719 rootdir != NULL && !vn_is_readonly(rootdir))
5720 spa->spa_async_thread = thread_create(NULL, 0,
5721 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
5722 mutex_exit(&spa->spa_async_lock);
5726 spa_async_request(spa_t *spa, int task)
5728 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
5729 mutex_enter(&spa->spa_async_lock);
5730 spa->spa_async_tasks |= task;
5731 mutex_exit(&spa->spa_async_lock);
5735 * ==========================================================================
5736 * SPA syncing routines
5737 * ==========================================================================
5741 bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5744 bpobj_enqueue(bpo, bp, tx);
5749 spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5753 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
5759 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
5761 char *packed = NULL;
5766 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
5769 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5770 * information. This avoids the dbuf_will_dirty() path and
5771 * saves us a pre-read to get data we don't actually care about.
5773 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
5774 packed = vmem_alloc(bufsize, KM_PUSHPAGE);
5776 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
5778 bzero(packed + nvsize, bufsize - nvsize);
5780 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
5782 vmem_free(packed, bufsize);
5784 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
5785 dmu_buf_will_dirty(db, tx);
5786 *(uint64_t *)db->db_data = nvsize;
5787 dmu_buf_rele(db, FTAG);
5791 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
5792 const char *config, const char *entry)
5802 * Update the MOS nvlist describing the list of available devices.
5803 * spa_validate_aux() will have already made sure this nvlist is
5804 * valid and the vdevs are labeled appropriately.
5806 if (sav->sav_object == 0) {
5807 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
5808 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
5809 sizeof (uint64_t), tx);
5810 VERIFY(zap_update(spa->spa_meta_objset,
5811 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
5812 &sav->sav_object, tx) == 0);
5815 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_PUSHPAGE) == 0);
5816 if (sav->sav_count == 0) {
5817 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
5819 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_PUSHPAGE);
5820 for (i = 0; i < sav->sav_count; i++)
5821 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
5822 B_FALSE, VDEV_CONFIG_L2CACHE);
5823 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
5824 sav->sav_count) == 0);
5825 for (i = 0; i < sav->sav_count; i++)
5826 nvlist_free(list[i]);
5827 kmem_free(list, sav->sav_count * sizeof (void *));
5830 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
5831 nvlist_free(nvroot);
5833 sav->sav_sync = B_FALSE;
5837 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
5841 if (list_is_empty(&spa->spa_config_dirty_list))
5844 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5846 config = spa_config_generate(spa, spa->spa_root_vdev,
5847 dmu_tx_get_txg(tx), B_FALSE);
5850 * If we're upgrading the spa version then make sure that
5851 * the config object gets updated with the correct version.
5853 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
5854 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5855 spa->spa_uberblock.ub_version);
5857 spa_config_exit(spa, SCL_STATE, FTAG);
5859 if (spa->spa_config_syncing)
5860 nvlist_free(spa->spa_config_syncing);
5861 spa->spa_config_syncing = config;
5863 spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
5867 spa_sync_version(void *arg1, void *arg2, dmu_tx_t *tx)
5870 uint64_t version = *(uint64_t *)arg2;
5873 * Setting the version is special cased when first creating the pool.
5875 ASSERT(tx->tx_txg != TXG_INITIAL);
5877 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
5878 ASSERT(version >= spa_version(spa));
5880 spa->spa_uberblock.ub_version = version;
5881 vdev_config_dirty(spa->spa_root_vdev);
5885 * Set zpool properties.
5888 spa_sync_props(void *arg1, void *arg2, dmu_tx_t *tx)
5891 objset_t *mos = spa->spa_meta_objset;
5892 nvlist_t *nvp = arg2;
5893 nvpair_t *elem = NULL;
5895 mutex_enter(&spa->spa_props_lock);
5897 while ((elem = nvlist_next_nvpair(nvp, elem))) {
5899 char *strval, *fname;
5901 const char *propname;
5902 zprop_type_t proptype;
5903 zfeature_info_t *feature;
5905 prop = zpool_name_to_prop(nvpair_name(elem));
5906 switch ((int)prop) {
5909 * We checked this earlier in spa_prop_validate().
5911 ASSERT(zpool_prop_feature(nvpair_name(elem)));
5913 fname = strchr(nvpair_name(elem), '@') + 1;
5914 VERIFY3U(0, ==, zfeature_lookup_name(fname, &feature));
5916 spa_feature_enable(spa, feature, tx);
5919 case ZPOOL_PROP_VERSION:
5920 VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5922 * The version is synced seperatly before other
5923 * properties and should be correct by now.
5925 ASSERT3U(spa_version(spa), >=, intval);
5928 case ZPOOL_PROP_ALTROOT:
5930 * 'altroot' is a non-persistent property. It should
5931 * have been set temporarily at creation or import time.
5933 ASSERT(spa->spa_root != NULL);
5936 case ZPOOL_PROP_READONLY:
5937 case ZPOOL_PROP_CACHEFILE:
5939 * 'readonly' and 'cachefile' are also non-persisitent
5943 case ZPOOL_PROP_COMMENT:
5944 VERIFY(nvpair_value_string(elem, &strval) == 0);
5945 if (spa->spa_comment != NULL)
5946 spa_strfree(spa->spa_comment);
5947 spa->spa_comment = spa_strdup(strval);
5949 * We need to dirty the configuration on all the vdevs
5950 * so that their labels get updated. It's unnecessary
5951 * to do this for pool creation since the vdev's
5952 * configuratoin has already been dirtied.
5954 if (tx->tx_txg != TXG_INITIAL)
5955 vdev_config_dirty(spa->spa_root_vdev);
5959 * Set pool property values in the poolprops mos object.
5961 if (spa->spa_pool_props_object == 0) {
5962 spa->spa_pool_props_object =
5963 zap_create_link(mos, DMU_OT_POOL_PROPS,
5964 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
5968 /* normalize the property name */
5969 propname = zpool_prop_to_name(prop);
5970 proptype = zpool_prop_get_type(prop);
5972 if (nvpair_type(elem) == DATA_TYPE_STRING) {
5973 ASSERT(proptype == PROP_TYPE_STRING);
5974 VERIFY(nvpair_value_string(elem, &strval) == 0);
5975 VERIFY(zap_update(mos,
5976 spa->spa_pool_props_object, propname,
5977 1, strlen(strval) + 1, strval, tx) == 0);
5979 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
5980 VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5982 if (proptype == PROP_TYPE_INDEX) {
5984 VERIFY(zpool_prop_index_to_string(
5985 prop, intval, &unused) == 0);
5987 VERIFY(zap_update(mos,
5988 spa->spa_pool_props_object, propname,
5989 8, 1, &intval, tx) == 0);
5991 ASSERT(0); /* not allowed */
5995 case ZPOOL_PROP_DELEGATION:
5996 spa->spa_delegation = intval;
5998 case ZPOOL_PROP_BOOTFS:
5999 spa->spa_bootfs = intval;
6001 case ZPOOL_PROP_FAILUREMODE:
6002 spa->spa_failmode = intval;
6004 case ZPOOL_PROP_AUTOEXPAND:
6005 spa->spa_autoexpand = intval;
6006 if (tx->tx_txg != TXG_INITIAL)
6007 spa_async_request(spa,
6008 SPA_ASYNC_AUTOEXPAND);
6010 case ZPOOL_PROP_DEDUPDITTO:
6011 spa->spa_dedup_ditto = intval;
6018 /* log internal history if this is not a zpool create */
6019 if (spa_version(spa) >= SPA_VERSION_ZPOOL_HISTORY &&
6020 tx->tx_txg != TXG_INITIAL) {
6021 spa_history_log_internal(LOG_POOL_PROPSET,
6022 spa, tx, "%s %lld %s",
6023 nvpair_name(elem), intval, spa_name(spa));
6027 mutex_exit(&spa->spa_props_lock);
6031 * Perform one-time upgrade on-disk changes. spa_version() does not
6032 * reflect the new version this txg, so there must be no changes this
6033 * txg to anything that the upgrade code depends on after it executes.
6034 * Therefore this must be called after dsl_pool_sync() does the sync
6038 spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
6040 dsl_pool_t *dp = spa->spa_dsl_pool;
6042 ASSERT(spa->spa_sync_pass == 1);
6044 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
6045 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
6046 dsl_pool_create_origin(dp, tx);
6048 /* Keeping the origin open increases spa_minref */
6049 spa->spa_minref += 3;
6052 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
6053 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
6054 dsl_pool_upgrade_clones(dp, tx);
6057 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
6058 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
6059 dsl_pool_upgrade_dir_clones(dp, tx);
6061 /* Keeping the freedir open increases spa_minref */
6062 spa->spa_minref += 3;
6065 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES &&
6066 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
6067 spa_feature_create_zap_objects(spa, tx);
6072 * Sync the specified transaction group. New blocks may be dirtied as
6073 * part of the process, so we iterate until it converges.
6076 spa_sync(spa_t *spa, uint64_t txg)
6078 dsl_pool_t *dp = spa->spa_dsl_pool;
6079 objset_t *mos = spa->spa_meta_objset;
6080 bpobj_t *defer_bpo = &spa->spa_deferred_bpobj;
6081 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
6082 vdev_t *rvd = spa->spa_root_vdev;
6088 VERIFY(spa_writeable(spa));
6091 * Lock out configuration changes.
6093 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
6095 spa->spa_syncing_txg = txg;
6096 spa->spa_sync_pass = 0;
6099 * If there are any pending vdev state changes, convert them
6100 * into config changes that go out with this transaction group.
6102 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6103 while (list_head(&spa->spa_state_dirty_list) != NULL) {
6105 * We need the write lock here because, for aux vdevs,
6106 * calling vdev_config_dirty() modifies sav_config.
6107 * This is ugly and will become unnecessary when we
6108 * eliminate the aux vdev wart by integrating all vdevs
6109 * into the root vdev tree.
6111 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6112 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
6113 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
6114 vdev_state_clean(vd);
6115 vdev_config_dirty(vd);
6117 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6118 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
6120 spa_config_exit(spa, SCL_STATE, FTAG);
6122 tx = dmu_tx_create_assigned(dp, txg);
6124 spa->spa_sync_starttime = gethrtime();
6125 taskq_cancel_id(system_taskq, spa->spa_deadman_tqid);
6126 spa->spa_deadman_tqid = taskq_dispatch_delay(system_taskq,
6127 spa_deadman, spa, TQ_SLEEP, ddi_get_lbolt() +
6128 NSEC_TO_TICK(spa->spa_deadman_synctime));
6131 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
6132 * set spa_deflate if we have no raid-z vdevs.
6134 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
6135 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
6138 for (i = 0; i < rvd->vdev_children; i++) {
6139 vd = rvd->vdev_child[i];
6140 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
6143 if (i == rvd->vdev_children) {
6144 spa->spa_deflate = TRUE;
6145 VERIFY(0 == zap_add(spa->spa_meta_objset,
6146 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
6147 sizeof (uint64_t), 1, &spa->spa_deflate, tx));
6152 * If anything has changed in this txg, or if someone is waiting
6153 * for this txg to sync (eg, spa_vdev_remove()), push the
6154 * deferred frees from the previous txg. If not, leave them
6155 * alone so that we don't generate work on an otherwise idle
6158 if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
6159 !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
6160 !txg_list_empty(&dp->dp_sync_tasks, txg) ||
6161 ((dsl_scan_active(dp->dp_scan) ||
6162 txg_sync_waiting(dp)) && !spa_shutting_down(spa))) {
6163 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6164 VERIFY3U(bpobj_iterate(defer_bpo,
6165 spa_free_sync_cb, zio, tx), ==, 0);
6166 VERIFY0(zio_wait(zio));
6170 * Iterate to convergence.
6173 int pass = ++spa->spa_sync_pass;
6175 spa_sync_config_object(spa, tx);
6176 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
6177 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
6178 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
6179 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
6180 spa_errlog_sync(spa, txg);
6181 dsl_pool_sync(dp, txg);
6183 if (pass < zfs_sync_pass_deferred_free) {
6184 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6185 bplist_iterate(free_bpl, spa_free_sync_cb,
6187 VERIFY(zio_wait(zio) == 0);
6189 bplist_iterate(free_bpl, bpobj_enqueue_cb,
6194 dsl_scan_sync(dp, tx);
6196 while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)))
6200 spa_sync_upgrades(spa, tx);
6202 } while (dmu_objset_is_dirty(mos, txg));
6205 * Rewrite the vdev configuration (which includes the uberblock)
6206 * to commit the transaction group.
6208 * If there are no dirty vdevs, we sync the uberblock to a few
6209 * random top-level vdevs that are known to be visible in the
6210 * config cache (see spa_vdev_add() for a complete description).
6211 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
6215 * We hold SCL_STATE to prevent vdev open/close/etc.
6216 * while we're attempting to write the vdev labels.
6218 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6220 if (list_is_empty(&spa->spa_config_dirty_list)) {
6221 vdev_t *svd[SPA_DVAS_PER_BP];
6223 int children = rvd->vdev_children;
6224 int c0 = spa_get_random(children);
6226 for (c = 0; c < children; c++) {
6227 vd = rvd->vdev_child[(c0 + c) % children];
6228 if (vd->vdev_ms_array == 0 || vd->vdev_islog)
6230 svd[svdcount++] = vd;
6231 if (svdcount == SPA_DVAS_PER_BP)
6234 error = vdev_config_sync(svd, svdcount, txg, B_FALSE);
6236 error = vdev_config_sync(svd, svdcount, txg,
6239 error = vdev_config_sync(rvd->vdev_child,
6240 rvd->vdev_children, txg, B_FALSE);
6242 error = vdev_config_sync(rvd->vdev_child,
6243 rvd->vdev_children, txg, B_TRUE);
6247 spa->spa_last_synced_guid = rvd->vdev_guid;
6249 spa_config_exit(spa, SCL_STATE, FTAG);
6253 zio_suspend(spa, NULL);
6254 zio_resume_wait(spa);
6258 taskq_cancel_id(system_taskq, spa->spa_deadman_tqid);
6259 spa->spa_deadman_tqid = 0;
6262 * Clear the dirty config list.
6264 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
6265 vdev_config_clean(vd);
6268 * Now that the new config has synced transactionally,
6269 * let it become visible to the config cache.
6271 if (spa->spa_config_syncing != NULL) {
6272 spa_config_set(spa, spa->spa_config_syncing);
6273 spa->spa_config_txg = txg;
6274 spa->spa_config_syncing = NULL;
6277 spa->spa_ubsync = spa->spa_uberblock;
6279 dsl_pool_sync_done(dp, txg);
6282 * Update usable space statistics.
6284 while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg))))
6285 vdev_sync_done(vd, txg);
6287 spa_update_dspace(spa);
6290 * It had better be the case that we didn't dirty anything
6291 * since vdev_config_sync().
6293 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
6294 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
6295 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
6297 spa->spa_sync_pass = 0;
6299 spa_config_exit(spa, SCL_CONFIG, FTAG);
6301 spa_handle_ignored_writes(spa);
6304 * If any async tasks have been requested, kick them off.
6306 spa_async_dispatch(spa);
6310 * Sync all pools. We don't want to hold the namespace lock across these
6311 * operations, so we take a reference on the spa_t and drop the lock during the
6315 spa_sync_allpools(void)
6318 mutex_enter(&spa_namespace_lock);
6319 while ((spa = spa_next(spa)) != NULL) {
6320 if (spa_state(spa) != POOL_STATE_ACTIVE ||
6321 !spa_writeable(spa) || spa_suspended(spa))
6323 spa_open_ref(spa, FTAG);
6324 mutex_exit(&spa_namespace_lock);
6325 txg_wait_synced(spa_get_dsl(spa), 0);
6326 mutex_enter(&spa_namespace_lock);
6327 spa_close(spa, FTAG);
6329 mutex_exit(&spa_namespace_lock);
6333 * ==========================================================================
6334 * Miscellaneous routines
6335 * ==========================================================================
6339 * Remove all pools in the system.
6347 * Remove all cached state. All pools should be closed now,
6348 * so every spa in the AVL tree should be unreferenced.
6350 mutex_enter(&spa_namespace_lock);
6351 while ((spa = spa_next(NULL)) != NULL) {
6353 * Stop async tasks. The async thread may need to detach
6354 * a device that's been replaced, which requires grabbing
6355 * spa_namespace_lock, so we must drop it here.
6357 spa_open_ref(spa, FTAG);
6358 mutex_exit(&spa_namespace_lock);
6359 spa_async_suspend(spa);
6360 mutex_enter(&spa_namespace_lock);
6361 spa_close(spa, FTAG);
6363 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
6365 spa_deactivate(spa);
6369 mutex_exit(&spa_namespace_lock);
6373 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
6378 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
6382 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
6383 vd = spa->spa_l2cache.sav_vdevs[i];
6384 if (vd->vdev_guid == guid)
6388 for (i = 0; i < spa->spa_spares.sav_count; i++) {
6389 vd = spa->spa_spares.sav_vdevs[i];
6390 if (vd->vdev_guid == guid)
6399 spa_upgrade(spa_t *spa, uint64_t version)
6401 ASSERT(spa_writeable(spa));
6403 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6406 * This should only be called for a non-faulted pool, and since a
6407 * future version would result in an unopenable pool, this shouldn't be
6410 ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version));
6411 ASSERT(version >= spa->spa_uberblock.ub_version);
6413 spa->spa_uberblock.ub_version = version;
6414 vdev_config_dirty(spa->spa_root_vdev);
6416 spa_config_exit(spa, SCL_ALL, FTAG);
6418 txg_wait_synced(spa_get_dsl(spa), 0);
6422 spa_has_spare(spa_t *spa, uint64_t guid)
6426 spa_aux_vdev_t *sav = &spa->spa_spares;
6428 for (i = 0; i < sav->sav_count; i++)
6429 if (sav->sav_vdevs[i]->vdev_guid == guid)
6432 for (i = 0; i < sav->sav_npending; i++) {
6433 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
6434 &spareguid) == 0 && spareguid == guid)
6442 * Check if a pool has an active shared spare device.
6443 * Note: reference count of an active spare is 2, as a spare and as a replace
6446 spa_has_active_shared_spare(spa_t *spa)
6450 spa_aux_vdev_t *sav = &spa->spa_spares;
6452 for (i = 0; i < sav->sav_count; i++) {
6453 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
6454 &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
6463 * Post a FM_EREPORT_ZFS_* event from sys/fm/fs/zfs.h. The payload will be
6464 * filled in from the spa and (optionally) the vdev. This doesn't do anything
6465 * in the userland libzpool, as we don't want consumers to misinterpret ztest
6466 * or zdb as real changes.
6469 spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
6472 zfs_ereport_post(name, spa, vd, NULL, 0, 0);
6476 #if defined(_KERNEL) && defined(HAVE_SPL)
6477 /* state manipulation functions */
6478 EXPORT_SYMBOL(spa_open);
6479 EXPORT_SYMBOL(spa_open_rewind);
6480 EXPORT_SYMBOL(spa_get_stats);
6481 EXPORT_SYMBOL(spa_create);
6482 EXPORT_SYMBOL(spa_import_rootpool);
6483 EXPORT_SYMBOL(spa_import);
6484 EXPORT_SYMBOL(spa_tryimport);
6485 EXPORT_SYMBOL(spa_destroy);
6486 EXPORT_SYMBOL(spa_export);
6487 EXPORT_SYMBOL(spa_reset);
6488 EXPORT_SYMBOL(spa_async_request);
6489 EXPORT_SYMBOL(spa_async_suspend);
6490 EXPORT_SYMBOL(spa_async_resume);
6491 EXPORT_SYMBOL(spa_inject_addref);
6492 EXPORT_SYMBOL(spa_inject_delref);
6493 EXPORT_SYMBOL(spa_scan_stat_init);
6494 EXPORT_SYMBOL(spa_scan_get_stats);
6496 /* device maniion */
6497 EXPORT_SYMBOL(spa_vdev_add);
6498 EXPORT_SYMBOL(spa_vdev_attach);
6499 EXPORT_SYMBOL(spa_vdev_detach);
6500 EXPORT_SYMBOL(spa_vdev_remove);
6501 EXPORT_SYMBOL(spa_vdev_setpath);
6502 EXPORT_SYMBOL(spa_vdev_setfru);
6503 EXPORT_SYMBOL(spa_vdev_split_mirror);
6505 /* spare statech is global across all pools) */
6506 EXPORT_SYMBOL(spa_spare_add);
6507 EXPORT_SYMBOL(spa_spare_remove);
6508 EXPORT_SYMBOL(spa_spare_exists);
6509 EXPORT_SYMBOL(spa_spare_activate);
6511 /* L2ARC statech is global across all pools) */
6512 EXPORT_SYMBOL(spa_l2cache_add);
6513 EXPORT_SYMBOL(spa_l2cache_remove);
6514 EXPORT_SYMBOL(spa_l2cache_exists);
6515 EXPORT_SYMBOL(spa_l2cache_activate);
6516 EXPORT_SYMBOL(spa_l2cache_drop);
6519 EXPORT_SYMBOL(spa_scan);
6520 EXPORT_SYMBOL(spa_scan_stop);
6523 EXPORT_SYMBOL(spa_sync); /* only for DMU use */
6524 EXPORT_SYMBOL(spa_sync_allpools);
6527 EXPORT_SYMBOL(spa_prop_set);
6528 EXPORT_SYMBOL(spa_prop_get);
6529 EXPORT_SYMBOL(spa_prop_clear_bootfs);
6531 /* asynchronous event notification */
6532 EXPORT_SYMBOL(spa_event_notify);