4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2013 by Delphix. All rights reserved.
25 * Copyright 2013 Nexenta Systems, Inc. All rights reserved.
26 * Copyright (c) 2013 Martin Matuska <mm@FreeBSD.org>. All rights reserved.
30 * SPA: Storage Pool Allocator
32 * This file contains all the routines used when modifying on-disk SPA state.
33 * This includes opening, importing, destroying, exporting a pool, and syncing a
37 #include <sys/zfs_context.h>
38 #include <sys/fm/fs/zfs.h>
39 #include <sys/spa_impl.h>
41 #include <sys/zio_checksum.h>
43 #include <sys/dmu_tx.h>
47 #include <sys/vdev_impl.h>
48 #include <sys/metaslab.h>
49 #include <sys/metaslab_impl.h>
50 #include <sys/uberblock_impl.h>
53 #include <sys/dmu_traverse.h>
54 #include <sys/dmu_objset.h>
55 #include <sys/unique.h>
56 #include <sys/dsl_pool.h>
57 #include <sys/dsl_dataset.h>
58 #include <sys/dsl_dir.h>
59 #include <sys/dsl_prop.h>
60 #include <sys/dsl_synctask.h>
61 #include <sys/fs/zfs.h>
63 #include <sys/callb.h>
64 #include <sys/spa_boot.h>
65 #include <sys/zfs_ioctl.h>
66 #include <sys/dsl_scan.h>
67 #include <sys/dmu_send.h>
68 #include <sys/dsl_destroy.h>
69 #include <sys/dsl_userhold.h>
70 #include <sys/zfeature.h>
72 #include <sys/trim_map.h>
75 #include <sys/callb.h>
76 #include <sys/cpupart.h>
81 #include "zfs_comutil.h"
83 /* Check hostid on import? */
84 static int check_hostid = 1;
86 SYSCTL_DECL(_vfs_zfs);
87 TUNABLE_INT("vfs.zfs.check_hostid", &check_hostid);
88 SYSCTL_INT(_vfs_zfs, OID_AUTO, check_hostid, CTLFLAG_RW, &check_hostid, 0,
89 "Check hostid on import?");
92 * The interval, in seconds, at which failed configuration cache file writes
95 static int zfs_ccw_retry_interval = 300;
97 typedef enum zti_modes {
98 zti_mode_fixed, /* value is # of threads (min 1) */
99 zti_mode_online_percent, /* value is % of online CPUs */
100 zti_mode_batch, /* cpu-intensive; value is ignored */
101 zti_mode_null, /* don't create a taskq */
105 #define ZTI_FIX(n) { zti_mode_fixed, (n) }
106 #define ZTI_PCT(n) { zti_mode_online_percent, (n) }
107 #define ZTI_BATCH { zti_mode_batch, 0 }
108 #define ZTI_NULL { zti_mode_null, 0 }
110 #define ZTI_ONE ZTI_FIX(1)
112 typedef struct zio_taskq_info {
113 enum zti_modes zti_mode;
117 static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
118 "issue", "issue_high", "intr", "intr_high"
122 * Define the taskq threads for the following I/O types:
123 * NULL, READ, WRITE, FREE, CLAIM, and IOCTL
125 const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
126 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
127 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL },
128 { ZTI_FIX(8), ZTI_NULL, ZTI_BATCH, ZTI_NULL },
129 { ZTI_BATCH, ZTI_FIX(5), ZTI_FIX(8), ZTI_FIX(5) },
130 { ZTI_FIX(100), ZTI_NULL, ZTI_ONE, ZTI_NULL },
131 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL },
132 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL },
135 static void spa_sync_version(void *arg, dmu_tx_t *tx);
136 static void spa_sync_props(void *arg, dmu_tx_t *tx);
137 static boolean_t spa_has_active_shared_spare(spa_t *spa);
138 static 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 */
145 id_t zio_taskq_psrset_bind = PS_NONE;
148 boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */
150 uint_t zio_taskq_basedc = 80; /* base duty cycle */
152 boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */
153 extern int zfs_sync_pass_deferred_free;
156 extern void spa_deadman(void *arg);
160 * This (illegal) pool name is used when temporarily importing a spa_t in order
161 * to get the vdev stats associated with the imported devices.
163 #define TRYIMPORT_NAME "$import"
166 * ==========================================================================
167 * SPA properties routines
168 * ==========================================================================
172 * Add a (source=src, propname=propval) list to an nvlist.
175 spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
176 uint64_t intval, zprop_source_t src)
178 const char *propname = zpool_prop_to_name(prop);
181 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
182 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
185 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
187 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
189 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
190 nvlist_free(propval);
194 * Get property values from the spa configuration.
197 spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
199 vdev_t *rvd = spa->spa_root_vdev;
200 dsl_pool_t *pool = spa->spa_dsl_pool;
204 uint64_t cap, version;
205 zprop_source_t src = ZPROP_SRC_NONE;
206 spa_config_dirent_t *dp;
208 ASSERT(MUTEX_HELD(&spa->spa_props_lock));
211 alloc = metaslab_class_get_alloc(spa_normal_class(spa));
212 size = metaslab_class_get_space(spa_normal_class(spa));
213 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
214 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
215 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
216 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
220 for (int c = 0; c < rvd->vdev_children; c++) {
221 vdev_t *tvd = rvd->vdev_child[c];
222 space += tvd->vdev_max_asize - tvd->vdev_asize;
224 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, space,
227 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
228 (spa_mode(spa) == FREAD), src);
230 cap = (size == 0) ? 0 : (alloc * 100 / size);
231 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
233 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
234 ddt_get_pool_dedup_ratio(spa), src);
236 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
237 rvd->vdev_state, src);
239 version = spa_version(spa);
240 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
241 src = ZPROP_SRC_DEFAULT;
243 src = ZPROP_SRC_LOCAL;
244 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
248 dsl_dir_t *freedir = pool->dp_free_dir;
251 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
252 * when opening pools before this version freedir will be NULL.
254 if (freedir != NULL) {
255 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
256 freedir->dd_phys->dd_used_bytes, src);
258 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
263 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
265 if (spa->spa_comment != NULL) {
266 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
270 if (spa->spa_root != NULL)
271 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
274 if ((dp = list_head(&spa->spa_config_list)) != NULL) {
275 if (dp->scd_path == NULL) {
276 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
277 "none", 0, ZPROP_SRC_LOCAL);
278 } else if (strcmp(dp->scd_path, spa_config_path) != 0) {
279 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
280 dp->scd_path, 0, ZPROP_SRC_LOCAL);
286 * Get zpool property values.
289 spa_prop_get(spa_t *spa, nvlist_t **nvp)
291 objset_t *mos = spa->spa_meta_objset;
296 VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
298 mutex_enter(&spa->spa_props_lock);
301 * Get properties from the spa config.
303 spa_prop_get_config(spa, nvp);
305 /* If no pool property object, no more prop to get. */
306 if (mos == NULL || spa->spa_pool_props_object == 0) {
307 mutex_exit(&spa->spa_props_lock);
312 * Get properties from the MOS pool property object.
314 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
315 (err = zap_cursor_retrieve(&zc, &za)) == 0;
316 zap_cursor_advance(&zc)) {
319 zprop_source_t src = ZPROP_SRC_DEFAULT;
322 if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
325 switch (za.za_integer_length) {
327 /* integer property */
328 if (za.za_first_integer !=
329 zpool_prop_default_numeric(prop))
330 src = ZPROP_SRC_LOCAL;
332 if (prop == ZPOOL_PROP_BOOTFS) {
334 dsl_dataset_t *ds = NULL;
336 dp = spa_get_dsl(spa);
337 dsl_pool_config_enter(dp, FTAG);
338 if (err = dsl_dataset_hold_obj(dp,
339 za.za_first_integer, FTAG, &ds)) {
340 dsl_pool_config_exit(dp, FTAG);
345 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
347 dsl_dataset_name(ds, strval);
348 dsl_dataset_rele(ds, FTAG);
349 dsl_pool_config_exit(dp, FTAG);
352 intval = za.za_first_integer;
355 spa_prop_add_list(*nvp, prop, strval, intval, src);
359 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
364 /* string property */
365 strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
366 err = zap_lookup(mos, spa->spa_pool_props_object,
367 za.za_name, 1, za.za_num_integers, strval);
369 kmem_free(strval, za.za_num_integers);
372 spa_prop_add_list(*nvp, prop, strval, 0, src);
373 kmem_free(strval, za.za_num_integers);
380 zap_cursor_fini(&zc);
381 mutex_exit(&spa->spa_props_lock);
383 if (err && err != ENOENT) {
393 * Validate the given pool properties nvlist and modify the list
394 * for the property values to be set.
397 spa_prop_validate(spa_t *spa, nvlist_t *props)
400 int error = 0, reset_bootfs = 0;
402 boolean_t has_feature = B_FALSE;
405 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
407 char *strval, *slash, *check, *fname;
408 const char *propname = nvpair_name(elem);
409 zpool_prop_t prop = zpool_name_to_prop(propname);
413 if (!zpool_prop_feature(propname)) {
414 error = SET_ERROR(EINVAL);
419 * Sanitize the input.
421 if (nvpair_type(elem) != DATA_TYPE_UINT64) {
422 error = SET_ERROR(EINVAL);
426 if (nvpair_value_uint64(elem, &intval) != 0) {
427 error = SET_ERROR(EINVAL);
432 error = SET_ERROR(EINVAL);
436 fname = strchr(propname, '@') + 1;
437 if (zfeature_lookup_name(fname, NULL) != 0) {
438 error = SET_ERROR(EINVAL);
442 has_feature = B_TRUE;
445 case ZPOOL_PROP_VERSION:
446 error = nvpair_value_uint64(elem, &intval);
448 (intval < spa_version(spa) ||
449 intval > SPA_VERSION_BEFORE_FEATURES ||
451 error = SET_ERROR(EINVAL);
454 case ZPOOL_PROP_DELEGATION:
455 case ZPOOL_PROP_AUTOREPLACE:
456 case ZPOOL_PROP_LISTSNAPS:
457 case ZPOOL_PROP_AUTOEXPAND:
458 error = nvpair_value_uint64(elem, &intval);
459 if (!error && intval > 1)
460 error = SET_ERROR(EINVAL);
463 case ZPOOL_PROP_BOOTFS:
465 * If the pool version is less than SPA_VERSION_BOOTFS,
466 * or the pool is still being created (version == 0),
467 * the bootfs property cannot be set.
469 if (spa_version(spa) < SPA_VERSION_BOOTFS) {
470 error = SET_ERROR(ENOTSUP);
475 * Make sure the vdev config is bootable
477 if (!vdev_is_bootable(spa->spa_root_vdev)) {
478 error = SET_ERROR(ENOTSUP);
484 error = nvpair_value_string(elem, &strval);
490 if (strval == NULL || strval[0] == '\0') {
491 objnum = zpool_prop_default_numeric(
496 if (error = dmu_objset_hold(strval, FTAG, &os))
499 /* Must be ZPL and not gzip compressed. */
501 if (dmu_objset_type(os) != DMU_OST_ZFS) {
502 error = SET_ERROR(ENOTSUP);
504 dsl_prop_get_int_ds(dmu_objset_ds(os),
505 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
507 !BOOTFS_COMPRESS_VALID(compress)) {
508 error = SET_ERROR(ENOTSUP);
510 objnum = dmu_objset_id(os);
512 dmu_objset_rele(os, FTAG);
516 case ZPOOL_PROP_FAILUREMODE:
517 error = nvpair_value_uint64(elem, &intval);
518 if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
519 intval > ZIO_FAILURE_MODE_PANIC))
520 error = SET_ERROR(EINVAL);
523 * This is a special case which only occurs when
524 * the pool has completely failed. This allows
525 * the user to change the in-core failmode property
526 * without syncing it out to disk (I/Os might
527 * currently be blocked). We do this by returning
528 * EIO to the caller (spa_prop_set) to trick it
529 * into thinking we encountered a property validation
532 if (!error && spa_suspended(spa)) {
533 spa->spa_failmode = intval;
534 error = SET_ERROR(EIO);
538 case ZPOOL_PROP_CACHEFILE:
539 if ((error = nvpair_value_string(elem, &strval)) != 0)
542 if (strval[0] == '\0')
545 if (strcmp(strval, "none") == 0)
548 if (strval[0] != '/') {
549 error = SET_ERROR(EINVAL);
553 slash = strrchr(strval, '/');
554 ASSERT(slash != NULL);
556 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
557 strcmp(slash, "/..") == 0)
558 error = SET_ERROR(EINVAL);
561 case ZPOOL_PROP_COMMENT:
562 if ((error = nvpair_value_string(elem, &strval)) != 0)
564 for (check = strval; *check != '\0'; check++) {
566 * The kernel doesn't have an easy isprint()
567 * check. For this kernel check, we merely
568 * check ASCII apart from DEL. Fix this if
569 * there is an easy-to-use kernel isprint().
571 if (*check >= 0x7f) {
572 error = SET_ERROR(EINVAL);
577 if (strlen(strval) > ZPROP_MAX_COMMENT)
581 case ZPOOL_PROP_DEDUPDITTO:
582 if (spa_version(spa) < SPA_VERSION_DEDUP)
583 error = SET_ERROR(ENOTSUP);
585 error = nvpair_value_uint64(elem, &intval);
587 intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
588 error = SET_ERROR(EINVAL);
596 if (!error && reset_bootfs) {
597 error = nvlist_remove(props,
598 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
601 error = nvlist_add_uint64(props,
602 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
610 spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
613 spa_config_dirent_t *dp;
615 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
619 dp = kmem_alloc(sizeof (spa_config_dirent_t),
622 if (cachefile[0] == '\0')
623 dp->scd_path = spa_strdup(spa_config_path);
624 else if (strcmp(cachefile, "none") == 0)
627 dp->scd_path = spa_strdup(cachefile);
629 list_insert_head(&spa->spa_config_list, dp);
631 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
635 spa_prop_set(spa_t *spa, nvlist_t *nvp)
638 nvpair_t *elem = NULL;
639 boolean_t need_sync = B_FALSE;
641 if ((error = spa_prop_validate(spa, nvp)) != 0)
644 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
645 zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem));
647 if (prop == ZPOOL_PROP_CACHEFILE ||
648 prop == ZPOOL_PROP_ALTROOT ||
649 prop == ZPOOL_PROP_READONLY)
652 if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) {
655 if (prop == ZPOOL_PROP_VERSION) {
656 VERIFY(nvpair_value_uint64(elem, &ver) == 0);
658 ASSERT(zpool_prop_feature(nvpair_name(elem)));
659 ver = SPA_VERSION_FEATURES;
663 /* Save time if the version is already set. */
664 if (ver == spa_version(spa))
668 * In addition to the pool directory object, we might
669 * create the pool properties object, the features for
670 * read object, the features for write object, or the
671 * feature descriptions object.
673 error = dsl_sync_task(spa->spa_name, NULL,
674 spa_sync_version, &ver, 6);
685 return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props,
693 * If the bootfs property value is dsobj, clear it.
696 spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
698 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
699 VERIFY(zap_remove(spa->spa_meta_objset,
700 spa->spa_pool_props_object,
701 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
708 spa_change_guid_check(void *arg, dmu_tx_t *tx)
710 uint64_t *newguid = arg;
711 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
712 vdev_t *rvd = spa->spa_root_vdev;
715 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
716 vdev_state = rvd->vdev_state;
717 spa_config_exit(spa, SCL_STATE, FTAG);
719 if (vdev_state != VDEV_STATE_HEALTHY)
720 return (SET_ERROR(ENXIO));
722 ASSERT3U(spa_guid(spa), !=, *newguid);
728 spa_change_guid_sync(void *arg, dmu_tx_t *tx)
730 uint64_t *newguid = arg;
731 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
733 vdev_t *rvd = spa->spa_root_vdev;
735 oldguid = spa_guid(spa);
737 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
738 rvd->vdev_guid = *newguid;
739 rvd->vdev_guid_sum += (*newguid - oldguid);
740 vdev_config_dirty(rvd);
741 spa_config_exit(spa, SCL_STATE, FTAG);
743 spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu",
748 * Change the GUID for the pool. This is done so that we can later
749 * re-import a pool built from a clone of our own vdevs. We will modify
750 * the root vdev's guid, our own pool guid, and then mark all of our
751 * vdevs dirty. Note that we must make sure that all our vdevs are
752 * online when we do this, or else any vdevs that weren't present
753 * would be orphaned from our pool. We are also going to issue a
754 * sysevent to update any watchers.
757 spa_change_guid(spa_t *spa)
762 mutex_enter(&spa->spa_vdev_top_lock);
763 mutex_enter(&spa_namespace_lock);
764 guid = spa_generate_guid(NULL);
766 error = dsl_sync_task(spa->spa_name, spa_change_guid_check,
767 spa_change_guid_sync, &guid, 5);
770 spa_config_sync(spa, B_FALSE, B_TRUE);
771 spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID);
774 mutex_exit(&spa_namespace_lock);
775 mutex_exit(&spa->spa_vdev_top_lock);
781 * ==========================================================================
782 * SPA state manipulation (open/create/destroy/import/export)
783 * ==========================================================================
787 spa_error_entry_compare(const void *a, const void *b)
789 spa_error_entry_t *sa = (spa_error_entry_t *)a;
790 spa_error_entry_t *sb = (spa_error_entry_t *)b;
793 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
794 sizeof (zbookmark_t));
805 * Utility function which retrieves copies of the current logs and
806 * re-initializes them in the process.
809 spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
811 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
813 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
814 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
816 avl_create(&spa->spa_errlist_scrub,
817 spa_error_entry_compare, sizeof (spa_error_entry_t),
818 offsetof(spa_error_entry_t, se_avl));
819 avl_create(&spa->spa_errlist_last,
820 spa_error_entry_compare, sizeof (spa_error_entry_t),
821 offsetof(spa_error_entry_t, se_avl));
825 spa_taskq_create(spa_t *spa, const char *name, enum zti_modes mode,
829 boolean_t batch = B_FALSE;
833 return (NULL); /* no taskq needed */
836 ASSERT3U(value, >=, 1);
837 value = MAX(value, 1);
842 flags |= TASKQ_THREADS_CPU_PCT;
843 value = zio_taskq_batch_pct;
846 case zti_mode_online_percent:
847 flags |= TASKQ_THREADS_CPU_PCT;
851 panic("unrecognized mode for %s taskq (%u:%u) in "
858 if (zio_taskq_sysdc && spa->spa_proc != &p0) {
860 flags |= TASKQ_DC_BATCH;
862 return (taskq_create_sysdc(name, value, 50, INT_MAX,
863 spa->spa_proc, zio_taskq_basedc, flags));
866 return (taskq_create_proc(name, value, maxclsyspri, 50, INT_MAX,
867 spa->spa_proc, flags));
871 spa_create_zio_taskqs(spa_t *spa)
873 for (int t = 0; t < ZIO_TYPES; t++) {
874 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
875 const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
876 enum zti_modes mode = ztip->zti_mode;
877 uint_t value = ztip->zti_value;
880 (void) snprintf(name, sizeof (name),
881 "%s_%s", zio_type_name[t], zio_taskq_types[q]);
883 spa->spa_zio_taskq[t][q] =
884 spa_taskq_create(spa, name, mode, value);
892 spa_thread(void *arg)
897 user_t *pu = PTOU(curproc);
899 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
902 ASSERT(curproc != &p0);
903 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
904 "zpool-%s", spa->spa_name);
905 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
908 /* bind this thread to the requested psrset */
909 if (zio_taskq_psrset_bind != PS_NONE) {
911 mutex_enter(&cpu_lock);
912 mutex_enter(&pidlock);
913 mutex_enter(&curproc->p_lock);
915 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
916 0, NULL, NULL) == 0) {
917 curthread->t_bind_pset = zio_taskq_psrset_bind;
920 "Couldn't bind process for zfs pool \"%s\" to "
921 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
924 mutex_exit(&curproc->p_lock);
925 mutex_exit(&pidlock);
926 mutex_exit(&cpu_lock);
932 if (zio_taskq_sysdc) {
933 sysdc_thread_enter(curthread, 100, 0);
937 spa->spa_proc = curproc;
938 spa->spa_did = curthread->t_did;
940 spa_create_zio_taskqs(spa);
942 mutex_enter(&spa->spa_proc_lock);
943 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
945 spa->spa_proc_state = SPA_PROC_ACTIVE;
946 cv_broadcast(&spa->spa_proc_cv);
948 CALLB_CPR_SAFE_BEGIN(&cprinfo);
949 while (spa->spa_proc_state == SPA_PROC_ACTIVE)
950 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
951 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
953 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
954 spa->spa_proc_state = SPA_PROC_GONE;
956 cv_broadcast(&spa->spa_proc_cv);
957 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */
959 mutex_enter(&curproc->p_lock);
962 #endif /* SPA_PROCESS */
966 * Activate an uninitialized pool.
969 spa_activate(spa_t *spa, int mode)
971 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
973 spa->spa_state = POOL_STATE_ACTIVE;
974 spa->spa_mode = mode;
976 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
977 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
979 /* Try to create a covering process */
980 mutex_enter(&spa->spa_proc_lock);
981 ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
982 ASSERT(spa->spa_proc == &p0);
986 /* Only create a process if we're going to be around a while. */
987 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
988 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
990 spa->spa_proc_state = SPA_PROC_CREATED;
991 while (spa->spa_proc_state == SPA_PROC_CREATED) {
992 cv_wait(&spa->spa_proc_cv,
993 &spa->spa_proc_lock);
995 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
996 ASSERT(spa->spa_proc != &p0);
997 ASSERT(spa->spa_did != 0);
1001 "Couldn't create process for zfs pool \"%s\"\n",
1006 #endif /* SPA_PROCESS */
1007 mutex_exit(&spa->spa_proc_lock);
1009 /* If we didn't create a process, we need to create our taskqs. */
1010 ASSERT(spa->spa_proc == &p0);
1011 if (spa->spa_proc == &p0) {
1012 spa_create_zio_taskqs(spa);
1016 * Start TRIM thread.
1018 trim_thread_create(spa);
1020 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
1021 offsetof(vdev_t, vdev_config_dirty_node));
1022 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
1023 offsetof(vdev_t, vdev_state_dirty_node));
1025 txg_list_create(&spa->spa_vdev_txg_list,
1026 offsetof(struct vdev, vdev_txg_node));
1028 avl_create(&spa->spa_errlist_scrub,
1029 spa_error_entry_compare, sizeof (spa_error_entry_t),
1030 offsetof(spa_error_entry_t, se_avl));
1031 avl_create(&spa->spa_errlist_last,
1032 spa_error_entry_compare, sizeof (spa_error_entry_t),
1033 offsetof(spa_error_entry_t, se_avl));
1037 * Opposite of spa_activate().
1040 spa_deactivate(spa_t *spa)
1042 ASSERT(spa->spa_sync_on == B_FALSE);
1043 ASSERT(spa->spa_dsl_pool == NULL);
1044 ASSERT(spa->spa_root_vdev == NULL);
1045 ASSERT(spa->spa_async_zio_root == NULL);
1046 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
1049 * Stop TRIM thread in case spa_unload() wasn't called directly
1050 * before spa_deactivate().
1052 trim_thread_destroy(spa);
1054 txg_list_destroy(&spa->spa_vdev_txg_list);
1056 list_destroy(&spa->spa_config_dirty_list);
1057 list_destroy(&spa->spa_state_dirty_list);
1059 for (int t = 0; t < ZIO_TYPES; t++) {
1060 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1061 if (spa->spa_zio_taskq[t][q] != NULL)
1062 taskq_destroy(spa->spa_zio_taskq[t][q]);
1063 spa->spa_zio_taskq[t][q] = NULL;
1067 metaslab_class_destroy(spa->spa_normal_class);
1068 spa->spa_normal_class = NULL;
1070 metaslab_class_destroy(spa->spa_log_class);
1071 spa->spa_log_class = NULL;
1074 * If this was part of an import or the open otherwise failed, we may
1075 * still have errors left in the queues. Empty them just in case.
1077 spa_errlog_drain(spa);
1079 avl_destroy(&spa->spa_errlist_scrub);
1080 avl_destroy(&spa->spa_errlist_last);
1082 spa->spa_state = POOL_STATE_UNINITIALIZED;
1084 mutex_enter(&spa->spa_proc_lock);
1085 if (spa->spa_proc_state != SPA_PROC_NONE) {
1086 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1087 spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1088 cv_broadcast(&spa->spa_proc_cv);
1089 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1090 ASSERT(spa->spa_proc != &p0);
1091 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1093 ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1094 spa->spa_proc_state = SPA_PROC_NONE;
1096 ASSERT(spa->spa_proc == &p0);
1097 mutex_exit(&spa->spa_proc_lock);
1101 * We want to make sure spa_thread() has actually exited the ZFS
1102 * module, so that the module can't be unloaded out from underneath
1105 if (spa->spa_did != 0) {
1106 thread_join(spa->spa_did);
1109 #endif /* SPA_PROCESS */
1113 * Verify a pool configuration, and construct the vdev tree appropriately. This
1114 * will create all the necessary vdevs in the appropriate layout, with each vdev
1115 * in the CLOSED state. This will prep the pool before open/creation/import.
1116 * All vdev validation is done by the vdev_alloc() routine.
1119 spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1120 uint_t id, int atype)
1126 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1129 if ((*vdp)->vdev_ops->vdev_op_leaf)
1132 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1135 if (error == ENOENT)
1141 return (SET_ERROR(EINVAL));
1144 for (int c = 0; c < children; c++) {
1146 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1154 ASSERT(*vdp != NULL);
1160 * Opposite of spa_load().
1163 spa_unload(spa_t *spa)
1167 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1172 trim_thread_destroy(spa);
1177 spa_async_suspend(spa);
1182 if (spa->spa_sync_on) {
1183 txg_sync_stop(spa->spa_dsl_pool);
1184 spa->spa_sync_on = B_FALSE;
1188 * Wait for any outstanding async I/O to complete.
1190 if (spa->spa_async_zio_root != NULL) {
1191 (void) zio_wait(spa->spa_async_zio_root);
1192 spa->spa_async_zio_root = NULL;
1195 bpobj_close(&spa->spa_deferred_bpobj);
1198 * Close the dsl pool.
1200 if (spa->spa_dsl_pool) {
1201 dsl_pool_close(spa->spa_dsl_pool);
1202 spa->spa_dsl_pool = NULL;
1203 spa->spa_meta_objset = NULL;
1208 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1211 * Drop and purge level 2 cache
1213 spa_l2cache_drop(spa);
1218 if (spa->spa_root_vdev)
1219 vdev_free(spa->spa_root_vdev);
1220 ASSERT(spa->spa_root_vdev == NULL);
1222 for (i = 0; i < spa->spa_spares.sav_count; i++)
1223 vdev_free(spa->spa_spares.sav_vdevs[i]);
1224 if (spa->spa_spares.sav_vdevs) {
1225 kmem_free(spa->spa_spares.sav_vdevs,
1226 spa->spa_spares.sav_count * sizeof (void *));
1227 spa->spa_spares.sav_vdevs = NULL;
1229 if (spa->spa_spares.sav_config) {
1230 nvlist_free(spa->spa_spares.sav_config);
1231 spa->spa_spares.sav_config = NULL;
1233 spa->spa_spares.sav_count = 0;
1235 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1236 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1237 vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1239 if (spa->spa_l2cache.sav_vdevs) {
1240 kmem_free(spa->spa_l2cache.sav_vdevs,
1241 spa->spa_l2cache.sav_count * sizeof (void *));
1242 spa->spa_l2cache.sav_vdevs = NULL;
1244 if (spa->spa_l2cache.sav_config) {
1245 nvlist_free(spa->spa_l2cache.sav_config);
1246 spa->spa_l2cache.sav_config = NULL;
1248 spa->spa_l2cache.sav_count = 0;
1250 spa->spa_async_suspended = 0;
1252 if (spa->spa_comment != NULL) {
1253 spa_strfree(spa->spa_comment);
1254 spa->spa_comment = NULL;
1257 spa_config_exit(spa, SCL_ALL, FTAG);
1261 * Load (or re-load) the current list of vdevs describing the active spares for
1262 * this pool. When this is called, we have some form of basic information in
1263 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1264 * then re-generate a more complete list including status information.
1267 spa_load_spares(spa_t *spa)
1274 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1277 * First, close and free any existing spare vdevs.
1279 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1280 vd = spa->spa_spares.sav_vdevs[i];
1282 /* Undo the call to spa_activate() below */
1283 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1284 B_FALSE)) != NULL && tvd->vdev_isspare)
1285 spa_spare_remove(tvd);
1290 if (spa->spa_spares.sav_vdevs)
1291 kmem_free(spa->spa_spares.sav_vdevs,
1292 spa->spa_spares.sav_count * sizeof (void *));
1294 if (spa->spa_spares.sav_config == NULL)
1297 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1298 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1300 spa->spa_spares.sav_count = (int)nspares;
1301 spa->spa_spares.sav_vdevs = NULL;
1307 * Construct the array of vdevs, opening them to get status in the
1308 * process. For each spare, there is potentially two different vdev_t
1309 * structures associated with it: one in the list of spares (used only
1310 * for basic validation purposes) and one in the active vdev
1311 * configuration (if it's spared in). During this phase we open and
1312 * validate each vdev on the spare list. If the vdev also exists in the
1313 * active configuration, then we also mark this vdev as an active spare.
1315 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1317 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1318 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1319 VDEV_ALLOC_SPARE) == 0);
1322 spa->spa_spares.sav_vdevs[i] = vd;
1324 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1325 B_FALSE)) != NULL) {
1326 if (!tvd->vdev_isspare)
1330 * We only mark the spare active if we were successfully
1331 * able to load the vdev. Otherwise, importing a pool
1332 * with a bad active spare would result in strange
1333 * behavior, because multiple pool would think the spare
1334 * is actively in use.
1336 * There is a vulnerability here to an equally bizarre
1337 * circumstance, where a dead active spare is later
1338 * brought back to life (onlined or otherwise). Given
1339 * the rarity of this scenario, and the extra complexity
1340 * it adds, we ignore the possibility.
1342 if (!vdev_is_dead(tvd))
1343 spa_spare_activate(tvd);
1347 vd->vdev_aux = &spa->spa_spares;
1349 if (vdev_open(vd) != 0)
1352 if (vdev_validate_aux(vd) == 0)
1357 * Recompute the stashed list of spares, with status information
1360 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1361 DATA_TYPE_NVLIST_ARRAY) == 0);
1363 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1365 for (i = 0; i < spa->spa_spares.sav_count; i++)
1366 spares[i] = vdev_config_generate(spa,
1367 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1368 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1369 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1370 for (i = 0; i < spa->spa_spares.sav_count; i++)
1371 nvlist_free(spares[i]);
1372 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1376 * Load (or re-load) the current list of vdevs describing the active l2cache for
1377 * this pool. When this is called, we have some form of basic information in
1378 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1379 * then re-generate a more complete list including status information.
1380 * Devices which are already active have their details maintained, and are
1384 spa_load_l2cache(spa_t *spa)
1388 int i, j, oldnvdevs;
1390 vdev_t *vd, **oldvdevs, **newvdevs;
1391 spa_aux_vdev_t *sav = &spa->spa_l2cache;
1393 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1395 if (sav->sav_config != NULL) {
1396 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1397 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1398 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1404 oldvdevs = sav->sav_vdevs;
1405 oldnvdevs = sav->sav_count;
1406 sav->sav_vdevs = NULL;
1410 * Process new nvlist of vdevs.
1412 for (i = 0; i < nl2cache; i++) {
1413 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1417 for (j = 0; j < oldnvdevs; j++) {
1419 if (vd != NULL && guid == vd->vdev_guid) {
1421 * Retain previous vdev for add/remove ops.
1429 if (newvdevs[i] == NULL) {
1433 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1434 VDEV_ALLOC_L2CACHE) == 0);
1439 * Commit this vdev as an l2cache device,
1440 * even if it fails to open.
1442 spa_l2cache_add(vd);
1447 spa_l2cache_activate(vd);
1449 if (vdev_open(vd) != 0)
1452 (void) vdev_validate_aux(vd);
1454 if (!vdev_is_dead(vd))
1455 l2arc_add_vdev(spa, vd);
1460 * Purge vdevs that were dropped
1462 for (i = 0; i < oldnvdevs; i++) {
1467 ASSERT(vd->vdev_isl2cache);
1469 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1470 pool != 0ULL && l2arc_vdev_present(vd))
1471 l2arc_remove_vdev(vd);
1472 vdev_clear_stats(vd);
1478 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1480 if (sav->sav_config == NULL)
1483 sav->sav_vdevs = newvdevs;
1484 sav->sav_count = (int)nl2cache;
1487 * Recompute the stashed list of l2cache devices, with status
1488 * information this time.
1490 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1491 DATA_TYPE_NVLIST_ARRAY) == 0);
1493 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1494 for (i = 0; i < sav->sav_count; i++)
1495 l2cache[i] = vdev_config_generate(spa,
1496 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1497 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1498 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1500 for (i = 0; i < sav->sav_count; i++)
1501 nvlist_free(l2cache[i]);
1503 kmem_free(l2cache, sav->sav_count * sizeof (void *));
1507 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1510 char *packed = NULL;
1515 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
1516 nvsize = *(uint64_t *)db->db_data;
1517 dmu_buf_rele(db, FTAG);
1519 packed = kmem_alloc(nvsize, KM_SLEEP);
1520 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1523 error = nvlist_unpack(packed, nvsize, value, 0);
1524 kmem_free(packed, nvsize);
1530 * Checks to see if the given vdev could not be opened, in which case we post a
1531 * sysevent to notify the autoreplace code that the device has been removed.
1534 spa_check_removed(vdev_t *vd)
1536 for (int c = 0; c < vd->vdev_children; c++)
1537 spa_check_removed(vd->vdev_child[c]);
1539 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) &&
1541 zfs_post_autoreplace(vd->vdev_spa, vd);
1542 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1547 * Validate the current config against the MOS config
1550 spa_config_valid(spa_t *spa, nvlist_t *config)
1552 vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1555 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1557 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1558 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1560 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1563 * If we're doing a normal import, then build up any additional
1564 * diagnostic information about missing devices in this config.
1565 * We'll pass this up to the user for further processing.
1567 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1568 nvlist_t **child, *nv;
1571 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1573 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1575 for (int c = 0; c < rvd->vdev_children; c++) {
1576 vdev_t *tvd = rvd->vdev_child[c];
1577 vdev_t *mtvd = mrvd->vdev_child[c];
1579 if (tvd->vdev_ops == &vdev_missing_ops &&
1580 mtvd->vdev_ops != &vdev_missing_ops &&
1582 child[idx++] = vdev_config_generate(spa, mtvd,
1587 VERIFY(nvlist_add_nvlist_array(nv,
1588 ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1589 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1590 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1592 for (int i = 0; i < idx; i++)
1593 nvlist_free(child[i]);
1596 kmem_free(child, rvd->vdev_children * sizeof (char **));
1600 * Compare the root vdev tree with the information we have
1601 * from the MOS config (mrvd). Check each top-level vdev
1602 * with the corresponding MOS config top-level (mtvd).
1604 for (int c = 0; c < rvd->vdev_children; c++) {
1605 vdev_t *tvd = rvd->vdev_child[c];
1606 vdev_t *mtvd = mrvd->vdev_child[c];
1609 * Resolve any "missing" vdevs in the current configuration.
1610 * If we find that the MOS config has more accurate information
1611 * about the top-level vdev then use that vdev instead.
1613 if (tvd->vdev_ops == &vdev_missing_ops &&
1614 mtvd->vdev_ops != &vdev_missing_ops) {
1616 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1620 * Device specific actions.
1622 if (mtvd->vdev_islog) {
1623 spa_set_log_state(spa, SPA_LOG_CLEAR);
1626 * XXX - once we have 'readonly' pool
1627 * support we should be able to handle
1628 * missing data devices by transitioning
1629 * the pool to readonly.
1635 * Swap the missing vdev with the data we were
1636 * able to obtain from the MOS config.
1638 vdev_remove_child(rvd, tvd);
1639 vdev_remove_child(mrvd, mtvd);
1641 vdev_add_child(rvd, mtvd);
1642 vdev_add_child(mrvd, tvd);
1644 spa_config_exit(spa, SCL_ALL, FTAG);
1646 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1649 } else if (mtvd->vdev_islog) {
1651 * Load the slog device's state from the MOS config
1652 * since it's possible that the label does not
1653 * contain the most up-to-date information.
1655 vdev_load_log_state(tvd, mtvd);
1660 spa_config_exit(spa, SCL_ALL, FTAG);
1663 * Ensure we were able to validate the config.
1665 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1669 * Check for missing log devices
1672 spa_check_logs(spa_t *spa)
1674 boolean_t rv = B_FALSE;
1676 switch (spa->spa_log_state) {
1677 case SPA_LOG_MISSING:
1678 /* need to recheck in case slog has been restored */
1679 case SPA_LOG_UNKNOWN:
1680 rv = (dmu_objset_find(spa->spa_name, zil_check_log_chain,
1681 NULL, DS_FIND_CHILDREN) != 0);
1683 spa_set_log_state(spa, SPA_LOG_MISSING);
1690 spa_passivate_log(spa_t *spa)
1692 vdev_t *rvd = spa->spa_root_vdev;
1693 boolean_t slog_found = B_FALSE;
1695 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1697 if (!spa_has_slogs(spa))
1700 for (int c = 0; c < rvd->vdev_children; c++) {
1701 vdev_t *tvd = rvd->vdev_child[c];
1702 metaslab_group_t *mg = tvd->vdev_mg;
1704 if (tvd->vdev_islog) {
1705 metaslab_group_passivate(mg);
1706 slog_found = B_TRUE;
1710 return (slog_found);
1714 spa_activate_log(spa_t *spa)
1716 vdev_t *rvd = spa->spa_root_vdev;
1718 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1720 for (int c = 0; c < rvd->vdev_children; c++) {
1721 vdev_t *tvd = rvd->vdev_child[c];
1722 metaslab_group_t *mg = tvd->vdev_mg;
1724 if (tvd->vdev_islog)
1725 metaslab_group_activate(mg);
1730 spa_offline_log(spa_t *spa)
1734 error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1735 NULL, DS_FIND_CHILDREN);
1738 * We successfully offlined the log device, sync out the
1739 * current txg so that the "stubby" block can be removed
1742 txg_wait_synced(spa->spa_dsl_pool, 0);
1748 spa_aux_check_removed(spa_aux_vdev_t *sav)
1752 for (i = 0; i < sav->sav_count; i++)
1753 spa_check_removed(sav->sav_vdevs[i]);
1757 spa_claim_notify(zio_t *zio)
1759 spa_t *spa = zio->io_spa;
1764 mutex_enter(&spa->spa_props_lock); /* any mutex will do */
1765 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1766 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1767 mutex_exit(&spa->spa_props_lock);
1770 typedef struct spa_load_error {
1771 uint64_t sle_meta_count;
1772 uint64_t sle_data_count;
1776 spa_load_verify_done(zio_t *zio)
1778 blkptr_t *bp = zio->io_bp;
1779 spa_load_error_t *sle = zio->io_private;
1780 dmu_object_type_t type = BP_GET_TYPE(bp);
1781 int error = zio->io_error;
1784 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1785 type != DMU_OT_INTENT_LOG)
1786 atomic_add_64(&sle->sle_meta_count, 1);
1788 atomic_add_64(&sle->sle_data_count, 1);
1790 zio_data_buf_free(zio->io_data, zio->io_size);
1795 spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1796 const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1800 size_t size = BP_GET_PSIZE(bp);
1801 void *data = zio_data_buf_alloc(size);
1803 zio_nowait(zio_read(rio, spa, bp, data, size,
1804 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1805 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1806 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1812 spa_load_verify(spa_t *spa)
1815 spa_load_error_t sle = { 0 };
1816 zpool_rewind_policy_t policy;
1817 boolean_t verify_ok = B_FALSE;
1820 zpool_get_rewind_policy(spa->spa_config, &policy);
1822 if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1825 rio = zio_root(spa, NULL, &sle,
1826 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1828 error = traverse_pool(spa, spa->spa_verify_min_txg,
1829 TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio);
1831 (void) zio_wait(rio);
1833 spa->spa_load_meta_errors = sle.sle_meta_count;
1834 spa->spa_load_data_errors = sle.sle_data_count;
1836 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1837 sle.sle_data_count <= policy.zrp_maxdata) {
1841 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1842 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1844 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1845 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1846 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1847 VERIFY(nvlist_add_int64(spa->spa_load_info,
1848 ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1849 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1850 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1852 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1856 if (error != ENXIO && error != EIO)
1857 error = SET_ERROR(EIO);
1861 return (verify_ok ? 0 : EIO);
1865 * Find a value in the pool props object.
1868 spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1870 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1871 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1875 * Find a value in the pool directory object.
1878 spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1880 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1881 name, sizeof (uint64_t), 1, val));
1885 spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1887 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1892 * Fix up config after a partly-completed split. This is done with the
1893 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
1894 * pool have that entry in their config, but only the splitting one contains
1895 * a list of all the guids of the vdevs that are being split off.
1897 * This function determines what to do with that list: either rejoin
1898 * all the disks to the pool, or complete the splitting process. To attempt
1899 * the rejoin, each disk that is offlined is marked online again, and
1900 * we do a reopen() call. If the vdev label for every disk that was
1901 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1902 * then we call vdev_split() on each disk, and complete the split.
1904 * Otherwise we leave the config alone, with all the vdevs in place in
1905 * the original pool.
1908 spa_try_repair(spa_t *spa, nvlist_t *config)
1915 boolean_t attempt_reopen;
1917 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
1920 /* check that the config is complete */
1921 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
1922 &glist, &gcount) != 0)
1925 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
1927 /* attempt to online all the vdevs & validate */
1928 attempt_reopen = B_TRUE;
1929 for (i = 0; i < gcount; i++) {
1930 if (glist[i] == 0) /* vdev is hole */
1933 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
1934 if (vd[i] == NULL) {
1936 * Don't bother attempting to reopen the disks;
1937 * just do the split.
1939 attempt_reopen = B_FALSE;
1941 /* attempt to re-online it */
1942 vd[i]->vdev_offline = B_FALSE;
1946 if (attempt_reopen) {
1947 vdev_reopen(spa->spa_root_vdev);
1949 /* check each device to see what state it's in */
1950 for (extracted = 0, i = 0; i < gcount; i++) {
1951 if (vd[i] != NULL &&
1952 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
1959 * If every disk has been moved to the new pool, or if we never
1960 * even attempted to look at them, then we split them off for
1963 if (!attempt_reopen || gcount == extracted) {
1964 for (i = 0; i < gcount; i++)
1967 vdev_reopen(spa->spa_root_vdev);
1970 kmem_free(vd, gcount * sizeof (vdev_t *));
1974 spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
1975 boolean_t mosconfig)
1977 nvlist_t *config = spa->spa_config;
1978 char *ereport = FM_EREPORT_ZFS_POOL;
1984 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
1985 return (SET_ERROR(EINVAL));
1987 ASSERT(spa->spa_comment == NULL);
1988 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
1989 spa->spa_comment = spa_strdup(comment);
1992 * Versioning wasn't explicitly added to the label until later, so if
1993 * it's not present treat it as the initial version.
1995 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
1996 &spa->spa_ubsync.ub_version) != 0)
1997 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
1999 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
2000 &spa->spa_config_txg);
2002 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
2003 spa_guid_exists(pool_guid, 0)) {
2004 error = SET_ERROR(EEXIST);
2006 spa->spa_config_guid = pool_guid;
2008 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
2010 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
2014 nvlist_free(spa->spa_load_info);
2015 spa->spa_load_info = fnvlist_alloc();
2017 gethrestime(&spa->spa_loaded_ts);
2018 error = spa_load_impl(spa, pool_guid, config, state, type,
2019 mosconfig, &ereport);
2022 spa->spa_minref = refcount_count(&spa->spa_refcount);
2024 if (error != EEXIST) {
2025 spa->spa_loaded_ts.tv_sec = 0;
2026 spa->spa_loaded_ts.tv_nsec = 0;
2028 if (error != EBADF) {
2029 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
2032 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2039 * Load an existing storage pool, using the pool's builtin spa_config as a
2040 * source of configuration information.
2043 spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
2044 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
2048 nvlist_t *nvroot = NULL;
2051 uberblock_t *ub = &spa->spa_uberblock;
2052 uint64_t children, config_cache_txg = spa->spa_config_txg;
2053 int orig_mode = spa->spa_mode;
2056 boolean_t missing_feat_write = B_FALSE;
2059 * If this is an untrusted config, access the pool in read-only mode.
2060 * This prevents things like resilvering recently removed devices.
2063 spa->spa_mode = FREAD;
2065 ASSERT(MUTEX_HELD(&spa_namespace_lock));
2067 spa->spa_load_state = state;
2069 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
2070 return (SET_ERROR(EINVAL));
2072 parse = (type == SPA_IMPORT_EXISTING ?
2073 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2076 * Create "The Godfather" zio to hold all async IOs
2078 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2079 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2082 * Parse the configuration into a vdev tree. We explicitly set the
2083 * value that will be returned by spa_version() since parsing the
2084 * configuration requires knowing the version number.
2086 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2087 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
2088 spa_config_exit(spa, SCL_ALL, FTAG);
2093 ASSERT(spa->spa_root_vdev == rvd);
2095 if (type != SPA_IMPORT_ASSEMBLE) {
2096 ASSERT(spa_guid(spa) == pool_guid);
2100 * Try to open all vdevs, loading each label in the process.
2102 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2103 error = vdev_open(rvd);
2104 spa_config_exit(spa, SCL_ALL, FTAG);
2109 * We need to validate the vdev labels against the configuration that
2110 * we have in hand, which is dependent on the setting of mosconfig. If
2111 * mosconfig is true then we're validating the vdev labels based on
2112 * that config. Otherwise, we're validating against the cached config
2113 * (zpool.cache) that was read when we loaded the zfs module, and then
2114 * later we will recursively call spa_load() and validate against
2117 * If we're assembling a new pool that's been split off from an
2118 * existing pool, the labels haven't yet been updated so we skip
2119 * validation for now.
2121 if (type != SPA_IMPORT_ASSEMBLE) {
2122 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2123 error = vdev_validate(rvd, mosconfig);
2124 spa_config_exit(spa, SCL_ALL, FTAG);
2129 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2130 return (SET_ERROR(ENXIO));
2134 * Find the best uberblock.
2136 vdev_uberblock_load(rvd, ub, &label);
2139 * If we weren't able to find a single valid uberblock, return failure.
2141 if (ub->ub_txg == 0) {
2143 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2147 * If the pool has an unsupported version we can't open it.
2149 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2151 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2154 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2158 * If we weren't able to find what's necessary for reading the
2159 * MOS in the label, return failure.
2161 if (label == NULL || nvlist_lookup_nvlist(label,
2162 ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) {
2164 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2169 * Update our in-core representation with the definitive values
2172 nvlist_free(spa->spa_label_features);
2173 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2179 * Look through entries in the label nvlist's features_for_read. If
2180 * there is a feature listed there which we don't understand then we
2181 * cannot open a pool.
2183 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2184 nvlist_t *unsup_feat;
2186 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2189 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2191 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2192 if (!zfeature_is_supported(nvpair_name(nvp))) {
2193 VERIFY(nvlist_add_string(unsup_feat,
2194 nvpair_name(nvp), "") == 0);
2198 if (!nvlist_empty(unsup_feat)) {
2199 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2200 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2201 nvlist_free(unsup_feat);
2202 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2206 nvlist_free(unsup_feat);
2210 * If the vdev guid sum doesn't match the uberblock, we have an
2211 * incomplete configuration. We first check to see if the pool
2212 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2213 * If it is, defer the vdev_guid_sum check till later so we
2214 * can handle missing vdevs.
2216 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
2217 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
2218 rvd->vdev_guid_sum != ub->ub_guid_sum)
2219 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2221 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2222 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2223 spa_try_repair(spa, config);
2224 spa_config_exit(spa, SCL_ALL, FTAG);
2225 nvlist_free(spa->spa_config_splitting);
2226 spa->spa_config_splitting = NULL;
2230 * Initialize internal SPA structures.
2232 spa->spa_state = POOL_STATE_ACTIVE;
2233 spa->spa_ubsync = spa->spa_uberblock;
2234 spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2235 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2236 spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2237 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2238 spa->spa_claim_max_txg = spa->spa_first_txg;
2239 spa->spa_prev_software_version = ub->ub_software_version;
2241 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2243 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2244 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2246 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2247 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2249 if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2250 boolean_t missing_feat_read = B_FALSE;
2251 nvlist_t *unsup_feat, *enabled_feat;
2253 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2254 &spa->spa_feat_for_read_obj) != 0) {
2255 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2258 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2259 &spa->spa_feat_for_write_obj) != 0) {
2260 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2263 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2264 &spa->spa_feat_desc_obj) != 0) {
2265 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2268 enabled_feat = fnvlist_alloc();
2269 unsup_feat = fnvlist_alloc();
2271 if (!feature_is_supported(spa->spa_meta_objset,
2272 spa->spa_feat_for_read_obj, spa->spa_feat_desc_obj,
2273 unsup_feat, enabled_feat))
2274 missing_feat_read = B_TRUE;
2276 if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) {
2277 if (!feature_is_supported(spa->spa_meta_objset,
2278 spa->spa_feat_for_write_obj, spa->spa_feat_desc_obj,
2279 unsup_feat, enabled_feat)) {
2280 missing_feat_write = B_TRUE;
2284 fnvlist_add_nvlist(spa->spa_load_info,
2285 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
2287 if (!nvlist_empty(unsup_feat)) {
2288 fnvlist_add_nvlist(spa->spa_load_info,
2289 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
2292 fnvlist_free(enabled_feat);
2293 fnvlist_free(unsup_feat);
2295 if (!missing_feat_read) {
2296 fnvlist_add_boolean(spa->spa_load_info,
2297 ZPOOL_CONFIG_CAN_RDONLY);
2301 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2302 * twofold: to determine whether the pool is available for
2303 * import in read-write mode and (if it is not) whether the
2304 * pool is available for import in read-only mode. If the pool
2305 * is available for import in read-write mode, it is displayed
2306 * as available in userland; if it is not available for import
2307 * in read-only mode, it is displayed as unavailable in
2308 * userland. If the pool is available for import in read-only
2309 * mode but not read-write mode, it is displayed as unavailable
2310 * in userland with a special note that the pool is actually
2311 * available for open in read-only mode.
2313 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2314 * missing a feature for write, we must first determine whether
2315 * the pool can be opened read-only before returning to
2316 * userland in order to know whether to display the
2317 * abovementioned note.
2319 if (missing_feat_read || (missing_feat_write &&
2320 spa_writeable(spa))) {
2321 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2326 spa->spa_is_initializing = B_TRUE;
2327 error = dsl_pool_open(spa->spa_dsl_pool);
2328 spa->spa_is_initializing = B_FALSE;
2330 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2334 nvlist_t *policy = NULL, *nvconfig;
2336 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2337 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2339 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2340 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2342 unsigned long myhostid = 0;
2344 VERIFY(nvlist_lookup_string(nvconfig,
2345 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2348 myhostid = zone_get_hostid(NULL);
2351 * We're emulating the system's hostid in userland, so
2352 * we can't use zone_get_hostid().
2354 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2355 #endif /* _KERNEL */
2356 if (check_hostid && hostid != 0 && myhostid != 0 &&
2357 hostid != myhostid) {
2358 nvlist_free(nvconfig);
2359 cmn_err(CE_WARN, "pool '%s' could not be "
2360 "loaded as it was last accessed by "
2361 "another system (host: %s hostid: 0x%lx). "
2362 "See: http://illumos.org/msg/ZFS-8000-EY",
2363 spa_name(spa), hostname,
2364 (unsigned long)hostid);
2365 return (SET_ERROR(EBADF));
2368 if (nvlist_lookup_nvlist(spa->spa_config,
2369 ZPOOL_REWIND_POLICY, &policy) == 0)
2370 VERIFY(nvlist_add_nvlist(nvconfig,
2371 ZPOOL_REWIND_POLICY, policy) == 0);
2373 spa_config_set(spa, nvconfig);
2375 spa_deactivate(spa);
2376 spa_activate(spa, orig_mode);
2378 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2381 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2382 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2383 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2385 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2388 * Load the bit that tells us to use the new accounting function
2389 * (raid-z deflation). If we have an older pool, this will not
2392 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2393 if (error != 0 && error != ENOENT)
2394 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2396 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2397 &spa->spa_creation_version);
2398 if (error != 0 && error != ENOENT)
2399 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2402 * Load the persistent error log. If we have an older pool, this will
2405 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2406 if (error != 0 && error != ENOENT)
2407 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2409 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2410 &spa->spa_errlog_scrub);
2411 if (error != 0 && error != ENOENT)
2412 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2415 * Load the history object. If we have an older pool, this
2416 * will not be present.
2418 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2419 if (error != 0 && error != ENOENT)
2420 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2423 * If we're assembling the pool from the split-off vdevs of
2424 * an existing pool, we don't want to attach the spares & cache
2429 * Load any hot spares for this pool.
2431 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2432 if (error != 0 && error != ENOENT)
2433 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2434 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2435 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2436 if (load_nvlist(spa, spa->spa_spares.sav_object,
2437 &spa->spa_spares.sav_config) != 0)
2438 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2440 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2441 spa_load_spares(spa);
2442 spa_config_exit(spa, SCL_ALL, FTAG);
2443 } else if (error == 0) {
2444 spa->spa_spares.sav_sync = B_TRUE;
2448 * Load any level 2 ARC devices for this pool.
2450 error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2451 &spa->spa_l2cache.sav_object);
2452 if (error != 0 && error != ENOENT)
2453 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2454 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2455 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2456 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2457 &spa->spa_l2cache.sav_config) != 0)
2458 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2460 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2461 spa_load_l2cache(spa);
2462 spa_config_exit(spa, SCL_ALL, FTAG);
2463 } else if (error == 0) {
2464 spa->spa_l2cache.sav_sync = B_TRUE;
2467 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2469 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2470 if (error && error != ENOENT)
2471 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2474 uint64_t autoreplace;
2476 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2477 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2478 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2479 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2480 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2481 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2482 &spa->spa_dedup_ditto);
2484 spa->spa_autoreplace = (autoreplace != 0);
2488 * If the 'autoreplace' property is set, then post a resource notifying
2489 * the ZFS DE that it should not issue any faults for unopenable
2490 * devices. We also iterate over the vdevs, and post a sysevent for any
2491 * unopenable vdevs so that the normal autoreplace handler can take
2494 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2495 spa_check_removed(spa->spa_root_vdev);
2497 * For the import case, this is done in spa_import(), because
2498 * at this point we're using the spare definitions from
2499 * the MOS config, not necessarily from the userland config.
2501 if (state != SPA_LOAD_IMPORT) {
2502 spa_aux_check_removed(&spa->spa_spares);
2503 spa_aux_check_removed(&spa->spa_l2cache);
2508 * Load the vdev state for all toplevel vdevs.
2513 * Propagate the leaf DTLs we just loaded all the way up the tree.
2515 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2516 vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2517 spa_config_exit(spa, SCL_ALL, FTAG);
2520 * Load the DDTs (dedup tables).
2522 error = ddt_load(spa);
2524 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2526 spa_update_dspace(spa);
2529 * Validate the config, using the MOS config to fill in any
2530 * information which might be missing. If we fail to validate
2531 * the config then declare the pool unfit for use. If we're
2532 * assembling a pool from a split, the log is not transferred
2535 if (type != SPA_IMPORT_ASSEMBLE) {
2538 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2539 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2541 if (!spa_config_valid(spa, nvconfig)) {
2542 nvlist_free(nvconfig);
2543 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2546 nvlist_free(nvconfig);
2549 * Now that we've validated the config, check the state of the
2550 * root vdev. If it can't be opened, it indicates one or
2551 * more toplevel vdevs are faulted.
2553 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2554 return (SET_ERROR(ENXIO));
2556 if (spa_check_logs(spa)) {
2557 *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2558 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2562 if (missing_feat_write) {
2563 ASSERT(state == SPA_LOAD_TRYIMPORT);
2566 * At this point, we know that we can open the pool in
2567 * read-only mode but not read-write mode. We now have enough
2568 * information and can return to userland.
2570 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
2574 * We've successfully opened the pool, verify that we're ready
2575 * to start pushing transactions.
2577 if (state != SPA_LOAD_TRYIMPORT) {
2578 if (error = spa_load_verify(spa))
2579 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2583 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2584 spa->spa_load_max_txg == UINT64_MAX)) {
2586 int need_update = B_FALSE;
2588 ASSERT(state != SPA_LOAD_TRYIMPORT);
2591 * Claim log blocks that haven't been committed yet.
2592 * This must all happen in a single txg.
2593 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2594 * invoked from zil_claim_log_block()'s i/o done callback.
2595 * Price of rollback is that we abandon the log.
2597 spa->spa_claiming = B_TRUE;
2599 tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2600 spa_first_txg(spa));
2601 (void) dmu_objset_find(spa_name(spa),
2602 zil_claim, tx, DS_FIND_CHILDREN);
2605 spa->spa_claiming = B_FALSE;
2607 spa_set_log_state(spa, SPA_LOG_GOOD);
2608 spa->spa_sync_on = B_TRUE;
2609 txg_sync_start(spa->spa_dsl_pool);
2612 * Wait for all claims to sync. We sync up to the highest
2613 * claimed log block birth time so that claimed log blocks
2614 * don't appear to be from the future. spa_claim_max_txg
2615 * will have been set for us by either zil_check_log_chain()
2616 * (invoked from spa_check_logs()) or zil_claim() above.
2618 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2621 * If the config cache is stale, or we have uninitialized
2622 * metaslabs (see spa_vdev_add()), then update the config.
2624 * If this is a verbatim import, trust the current
2625 * in-core spa_config and update the disk labels.
2627 if (config_cache_txg != spa->spa_config_txg ||
2628 state == SPA_LOAD_IMPORT ||
2629 state == SPA_LOAD_RECOVER ||
2630 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2631 need_update = B_TRUE;
2633 for (int c = 0; c < rvd->vdev_children; c++)
2634 if (rvd->vdev_child[c]->vdev_ms_array == 0)
2635 need_update = B_TRUE;
2638 * Update the config cache asychronously in case we're the
2639 * root pool, in which case the config cache isn't writable yet.
2642 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2645 * Check all DTLs to see if anything needs resilvering.
2647 if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2648 vdev_resilver_needed(rvd, NULL, NULL))
2649 spa_async_request(spa, SPA_ASYNC_RESILVER);
2652 * Log the fact that we booted up (so that we can detect if
2653 * we rebooted in the middle of an operation).
2655 spa_history_log_version(spa, "open");
2658 * Delete any inconsistent datasets.
2660 (void) dmu_objset_find(spa_name(spa),
2661 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2664 * Clean up any stale temporary dataset userrefs.
2666 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2673 spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2675 int mode = spa->spa_mode;
2678 spa_deactivate(spa);
2680 spa->spa_load_max_txg--;
2682 spa_activate(spa, mode);
2683 spa_async_suspend(spa);
2685 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2689 * If spa_load() fails this function will try loading prior txg's. If
2690 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2691 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2692 * function will not rewind the pool and will return the same error as
2696 spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2697 uint64_t max_request, int rewind_flags)
2699 nvlist_t *loadinfo = NULL;
2700 nvlist_t *config = NULL;
2701 int load_error, rewind_error;
2702 uint64_t safe_rewind_txg;
2705 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2706 spa->spa_load_max_txg = spa->spa_load_txg;
2707 spa_set_log_state(spa, SPA_LOG_CLEAR);
2709 spa->spa_load_max_txg = max_request;
2712 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2714 if (load_error == 0)
2717 if (spa->spa_root_vdev != NULL)
2718 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2720 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2721 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2723 if (rewind_flags & ZPOOL_NEVER_REWIND) {
2724 nvlist_free(config);
2725 return (load_error);
2728 if (state == SPA_LOAD_RECOVER) {
2729 /* Price of rolling back is discarding txgs, including log */
2730 spa_set_log_state(spa, SPA_LOG_CLEAR);
2733 * If we aren't rolling back save the load info from our first
2734 * import attempt so that we can restore it after attempting
2737 loadinfo = spa->spa_load_info;
2738 spa->spa_load_info = fnvlist_alloc();
2741 spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2742 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2743 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2744 TXG_INITIAL : safe_rewind_txg;
2747 * Continue as long as we're finding errors, we're still within
2748 * the acceptable rewind range, and we're still finding uberblocks
2750 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2751 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2752 if (spa->spa_load_max_txg < safe_rewind_txg)
2753 spa->spa_extreme_rewind = B_TRUE;
2754 rewind_error = spa_load_retry(spa, state, mosconfig);
2757 spa->spa_extreme_rewind = B_FALSE;
2758 spa->spa_load_max_txg = UINT64_MAX;
2760 if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2761 spa_config_set(spa, config);
2763 if (state == SPA_LOAD_RECOVER) {
2764 ASSERT3P(loadinfo, ==, NULL);
2765 return (rewind_error);
2767 /* Store the rewind info as part of the initial load info */
2768 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
2769 spa->spa_load_info);
2771 /* Restore the initial load info */
2772 fnvlist_free(spa->spa_load_info);
2773 spa->spa_load_info = loadinfo;
2775 return (load_error);
2782 * The import case is identical to an open except that the configuration is sent
2783 * down from userland, instead of grabbed from the configuration cache. For the
2784 * case of an open, the pool configuration will exist in the
2785 * POOL_STATE_UNINITIALIZED state.
2787 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2788 * the same time open the pool, without having to keep around the spa_t in some
2792 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2796 spa_load_state_t state = SPA_LOAD_OPEN;
2798 int locked = B_FALSE;
2799 int firstopen = B_FALSE;
2804 * As disgusting as this is, we need to support recursive calls to this
2805 * function because dsl_dir_open() is called during spa_load(), and ends
2806 * up calling spa_open() again. The real fix is to figure out how to
2807 * avoid dsl_dir_open() calling this in the first place.
2809 if (mutex_owner(&spa_namespace_lock) != curthread) {
2810 mutex_enter(&spa_namespace_lock);
2814 if ((spa = spa_lookup(pool)) == NULL) {
2816 mutex_exit(&spa_namespace_lock);
2817 return (SET_ERROR(ENOENT));
2820 if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2821 zpool_rewind_policy_t policy;
2825 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2827 if (policy.zrp_request & ZPOOL_DO_REWIND)
2828 state = SPA_LOAD_RECOVER;
2830 spa_activate(spa, spa_mode_global);
2832 if (state != SPA_LOAD_RECOVER)
2833 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2835 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2836 policy.zrp_request);
2838 if (error == EBADF) {
2840 * If vdev_validate() returns failure (indicated by
2841 * EBADF), it indicates that one of the vdevs indicates
2842 * that the pool has been exported or destroyed. If
2843 * this is the case, the config cache is out of sync and
2844 * we should remove the pool from the namespace.
2847 spa_deactivate(spa);
2848 spa_config_sync(spa, B_TRUE, B_TRUE);
2851 mutex_exit(&spa_namespace_lock);
2852 return (SET_ERROR(ENOENT));
2857 * We can't open the pool, but we still have useful
2858 * information: the state of each vdev after the
2859 * attempted vdev_open(). Return this to the user.
2861 if (config != NULL && spa->spa_config) {
2862 VERIFY(nvlist_dup(spa->spa_config, config,
2864 VERIFY(nvlist_add_nvlist(*config,
2865 ZPOOL_CONFIG_LOAD_INFO,
2866 spa->spa_load_info) == 0);
2869 spa_deactivate(spa);
2870 spa->spa_last_open_failed = error;
2872 mutex_exit(&spa_namespace_lock);
2878 spa_open_ref(spa, tag);
2881 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2884 * If we've recovered the pool, pass back any information we
2885 * gathered while doing the load.
2887 if (state == SPA_LOAD_RECOVER) {
2888 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
2889 spa->spa_load_info) == 0);
2893 spa->spa_last_open_failed = 0;
2894 spa->spa_last_ubsync_txg = 0;
2895 spa->spa_load_txg = 0;
2896 mutex_exit(&spa_namespace_lock);
2900 zvol_create_minors(spa->spa_name);
2911 spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
2914 return (spa_open_common(name, spapp, tag, policy, config));
2918 spa_open(const char *name, spa_t **spapp, void *tag)
2920 return (spa_open_common(name, spapp, tag, NULL, NULL));
2924 * Lookup the given spa_t, incrementing the inject count in the process,
2925 * preventing it from being exported or destroyed.
2928 spa_inject_addref(char *name)
2932 mutex_enter(&spa_namespace_lock);
2933 if ((spa = spa_lookup(name)) == NULL) {
2934 mutex_exit(&spa_namespace_lock);
2937 spa->spa_inject_ref++;
2938 mutex_exit(&spa_namespace_lock);
2944 spa_inject_delref(spa_t *spa)
2946 mutex_enter(&spa_namespace_lock);
2947 spa->spa_inject_ref--;
2948 mutex_exit(&spa_namespace_lock);
2952 * Add spares device information to the nvlist.
2955 spa_add_spares(spa_t *spa, nvlist_t *config)
2965 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2967 if (spa->spa_spares.sav_count == 0)
2970 VERIFY(nvlist_lookup_nvlist(config,
2971 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2972 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
2973 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2975 VERIFY(nvlist_add_nvlist_array(nvroot,
2976 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2977 VERIFY(nvlist_lookup_nvlist_array(nvroot,
2978 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2981 * Go through and find any spares which have since been
2982 * repurposed as an active spare. If this is the case, update
2983 * their status appropriately.
2985 for (i = 0; i < nspares; i++) {
2986 VERIFY(nvlist_lookup_uint64(spares[i],
2987 ZPOOL_CONFIG_GUID, &guid) == 0);
2988 if (spa_spare_exists(guid, &pool, NULL) &&
2990 VERIFY(nvlist_lookup_uint64_array(
2991 spares[i], ZPOOL_CONFIG_VDEV_STATS,
2992 (uint64_t **)&vs, &vsc) == 0);
2993 vs->vs_state = VDEV_STATE_CANT_OPEN;
2994 vs->vs_aux = VDEV_AUX_SPARED;
3001 * Add l2cache device information to the nvlist, including vdev stats.
3004 spa_add_l2cache(spa_t *spa, nvlist_t *config)
3007 uint_t i, j, nl2cache;
3014 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3016 if (spa->spa_l2cache.sav_count == 0)
3019 VERIFY(nvlist_lookup_nvlist(config,
3020 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3021 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
3022 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3023 if (nl2cache != 0) {
3024 VERIFY(nvlist_add_nvlist_array(nvroot,
3025 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3026 VERIFY(nvlist_lookup_nvlist_array(nvroot,
3027 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3030 * Update level 2 cache device stats.
3033 for (i = 0; i < nl2cache; i++) {
3034 VERIFY(nvlist_lookup_uint64(l2cache[i],
3035 ZPOOL_CONFIG_GUID, &guid) == 0);
3038 for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
3040 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
3041 vd = spa->spa_l2cache.sav_vdevs[j];
3047 VERIFY(nvlist_lookup_uint64_array(l2cache[i],
3048 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
3050 vdev_get_stats(vd, vs);
3056 spa_add_feature_stats(spa_t *spa, nvlist_t *config)
3062 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3063 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3065 /* We may be unable to read features if pool is suspended. */
3066 if (spa_suspended(spa))
3069 if (spa->spa_feat_for_read_obj != 0) {
3070 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3071 spa->spa_feat_for_read_obj);
3072 zap_cursor_retrieve(&zc, &za) == 0;
3073 zap_cursor_advance(&zc)) {
3074 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3075 za.za_num_integers == 1);
3076 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3077 za.za_first_integer));
3079 zap_cursor_fini(&zc);
3082 if (spa->spa_feat_for_write_obj != 0) {
3083 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3084 spa->spa_feat_for_write_obj);
3085 zap_cursor_retrieve(&zc, &za) == 0;
3086 zap_cursor_advance(&zc)) {
3087 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3088 za.za_num_integers == 1);
3089 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3090 za.za_first_integer));
3092 zap_cursor_fini(&zc);
3096 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
3098 nvlist_free(features);
3102 spa_get_stats(const char *name, nvlist_t **config,
3103 char *altroot, size_t buflen)
3109 error = spa_open_common(name, &spa, FTAG, NULL, config);
3113 * This still leaves a window of inconsistency where the spares
3114 * or l2cache devices could change and the config would be
3115 * self-inconsistent.
3117 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3119 if (*config != NULL) {
3120 uint64_t loadtimes[2];
3122 loadtimes[0] = spa->spa_loaded_ts.tv_sec;
3123 loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
3124 VERIFY(nvlist_add_uint64_array(*config,
3125 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
3127 VERIFY(nvlist_add_uint64(*config,
3128 ZPOOL_CONFIG_ERRCOUNT,
3129 spa_get_errlog_size(spa)) == 0);
3131 if (spa_suspended(spa))
3132 VERIFY(nvlist_add_uint64(*config,
3133 ZPOOL_CONFIG_SUSPENDED,
3134 spa->spa_failmode) == 0);
3136 spa_add_spares(spa, *config);
3137 spa_add_l2cache(spa, *config);
3138 spa_add_feature_stats(spa, *config);
3143 * We want to get the alternate root even for faulted pools, so we cheat
3144 * and call spa_lookup() directly.
3148 mutex_enter(&spa_namespace_lock);
3149 spa = spa_lookup(name);
3151 spa_altroot(spa, altroot, buflen);
3155 mutex_exit(&spa_namespace_lock);
3157 spa_altroot(spa, altroot, buflen);
3162 spa_config_exit(spa, SCL_CONFIG, FTAG);
3163 spa_close(spa, FTAG);
3170 * Validate that the auxiliary device array is well formed. We must have an
3171 * array of nvlists, each which describes a valid leaf vdev. If this is an
3172 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3173 * specified, as long as they are well-formed.
3176 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
3177 spa_aux_vdev_t *sav, const char *config, uint64_t version,
3178 vdev_labeltype_t label)
3185 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3188 * It's acceptable to have no devs specified.
3190 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
3194 return (SET_ERROR(EINVAL));
3197 * Make sure the pool is formatted with a version that supports this
3200 if (spa_version(spa) < version)
3201 return (SET_ERROR(ENOTSUP));
3204 * Set the pending device list so we correctly handle device in-use
3207 sav->sav_pending = dev;
3208 sav->sav_npending = ndev;
3210 for (i = 0; i < ndev; i++) {
3211 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
3215 if (!vd->vdev_ops->vdev_op_leaf) {
3217 error = SET_ERROR(EINVAL);
3222 * The L2ARC currently only supports disk devices in
3223 * kernel context. For user-level testing, we allow it.
3226 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
3227 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
3228 error = SET_ERROR(ENOTBLK);
3235 if ((error = vdev_open(vd)) == 0 &&
3236 (error = vdev_label_init(vd, crtxg, label)) == 0) {
3237 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
3238 vd->vdev_guid) == 0);
3244 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
3251 sav->sav_pending = NULL;
3252 sav->sav_npending = 0;
3257 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
3261 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3263 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3264 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
3265 VDEV_LABEL_SPARE)) != 0) {
3269 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3270 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
3271 VDEV_LABEL_L2CACHE));
3275 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
3280 if (sav->sav_config != NULL) {
3286 * Generate new dev list by concatentating with the
3289 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
3290 &olddevs, &oldndevs) == 0);
3292 newdevs = kmem_alloc(sizeof (void *) *
3293 (ndevs + oldndevs), KM_SLEEP);
3294 for (i = 0; i < oldndevs; i++)
3295 VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
3297 for (i = 0; i < ndevs; i++)
3298 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
3301 VERIFY(nvlist_remove(sav->sav_config, config,
3302 DATA_TYPE_NVLIST_ARRAY) == 0);
3304 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3305 config, newdevs, ndevs + oldndevs) == 0);
3306 for (i = 0; i < oldndevs + ndevs; i++)
3307 nvlist_free(newdevs[i]);
3308 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
3311 * Generate a new dev list.
3313 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
3315 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
3321 * Stop and drop level 2 ARC devices
3324 spa_l2cache_drop(spa_t *spa)
3328 spa_aux_vdev_t *sav = &spa->spa_l2cache;
3330 for (i = 0; i < sav->sav_count; i++) {
3333 vd = sav->sav_vdevs[i];
3336 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
3337 pool != 0ULL && l2arc_vdev_present(vd))
3338 l2arc_remove_vdev(vd);
3346 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
3350 char *altroot = NULL;
3355 uint64_t txg = TXG_INITIAL;
3356 nvlist_t **spares, **l2cache;
3357 uint_t nspares, nl2cache;
3358 uint64_t version, obj;
3359 boolean_t has_features;
3362 * If this pool already exists, return failure.
3364 mutex_enter(&spa_namespace_lock);
3365 if (spa_lookup(pool) != NULL) {
3366 mutex_exit(&spa_namespace_lock);
3367 return (SET_ERROR(EEXIST));
3371 * Allocate a new spa_t structure.
3373 (void) nvlist_lookup_string(props,
3374 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3375 spa = spa_add(pool, NULL, altroot);
3376 spa_activate(spa, spa_mode_global);
3378 if (props && (error = spa_prop_validate(spa, props))) {
3379 spa_deactivate(spa);
3381 mutex_exit(&spa_namespace_lock);
3385 has_features = B_FALSE;
3386 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
3387 elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
3388 if (zpool_prop_feature(nvpair_name(elem)))
3389 has_features = B_TRUE;
3392 if (has_features || nvlist_lookup_uint64(props,
3393 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
3394 version = SPA_VERSION;
3396 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
3398 spa->spa_first_txg = txg;
3399 spa->spa_uberblock.ub_txg = txg - 1;
3400 spa->spa_uberblock.ub_version = version;
3401 spa->spa_ubsync = spa->spa_uberblock;
3404 * Create "The Godfather" zio to hold all async IOs
3406 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
3407 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
3410 * Create the root vdev.
3412 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3414 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3416 ASSERT(error != 0 || rvd != NULL);
3417 ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3419 if (error == 0 && !zfs_allocatable_devs(nvroot))
3420 error = SET_ERROR(EINVAL);
3423 (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3424 (error = spa_validate_aux(spa, nvroot, txg,
3425 VDEV_ALLOC_ADD)) == 0) {
3426 for (int c = 0; c < rvd->vdev_children; c++) {
3427 vdev_ashift_optimize(rvd->vdev_child[c]);
3428 vdev_metaslab_set_size(rvd->vdev_child[c]);
3429 vdev_expand(rvd->vdev_child[c], txg);
3433 spa_config_exit(spa, SCL_ALL, FTAG);
3437 spa_deactivate(spa);
3439 mutex_exit(&spa_namespace_lock);
3444 * Get the list of spares, if specified.
3446 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3447 &spares, &nspares) == 0) {
3448 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3450 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3451 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3452 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3453 spa_load_spares(spa);
3454 spa_config_exit(spa, SCL_ALL, FTAG);
3455 spa->spa_spares.sav_sync = B_TRUE;
3459 * Get the list of level 2 cache devices, if specified.
3461 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3462 &l2cache, &nl2cache) == 0) {
3463 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3464 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3465 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3466 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3467 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3468 spa_load_l2cache(spa);
3469 spa_config_exit(spa, SCL_ALL, FTAG);
3470 spa->spa_l2cache.sav_sync = B_TRUE;
3473 spa->spa_is_initializing = B_TRUE;
3474 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3475 spa->spa_meta_objset = dp->dp_meta_objset;
3476 spa->spa_is_initializing = B_FALSE;
3479 * Create DDTs (dedup tables).
3483 spa_update_dspace(spa);
3485 tx = dmu_tx_create_assigned(dp, txg);
3488 * Create the pool config object.
3490 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3491 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3492 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3494 if (zap_add(spa->spa_meta_objset,
3495 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3496 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3497 cmn_err(CE_PANIC, "failed to add pool config");
3500 if (spa_version(spa) >= SPA_VERSION_FEATURES)
3501 spa_feature_create_zap_objects(spa, tx);
3503 if (zap_add(spa->spa_meta_objset,
3504 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3505 sizeof (uint64_t), 1, &version, tx) != 0) {
3506 cmn_err(CE_PANIC, "failed to add pool version");
3509 /* Newly created pools with the right version are always deflated. */
3510 if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3511 spa->spa_deflate = TRUE;
3512 if (zap_add(spa->spa_meta_objset,
3513 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3514 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3515 cmn_err(CE_PANIC, "failed to add deflate");
3520 * Create the deferred-free bpobj. Turn off compression
3521 * because sync-to-convergence takes longer if the blocksize
3524 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3525 dmu_object_set_compress(spa->spa_meta_objset, obj,
3526 ZIO_COMPRESS_OFF, tx);
3527 if (zap_add(spa->spa_meta_objset,
3528 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3529 sizeof (uint64_t), 1, &obj, tx) != 0) {
3530 cmn_err(CE_PANIC, "failed to add bpobj");
3532 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3533 spa->spa_meta_objset, obj));
3536 * Create the pool's history object.
3538 if (version >= SPA_VERSION_ZPOOL_HISTORY)
3539 spa_history_create_obj(spa, tx);
3542 * Set pool properties.
3544 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3545 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3546 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3547 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3549 if (props != NULL) {
3550 spa_configfile_set(spa, props, B_FALSE);
3551 spa_sync_props(props, tx);
3556 spa->spa_sync_on = B_TRUE;
3557 txg_sync_start(spa->spa_dsl_pool);
3560 * We explicitly wait for the first transaction to complete so that our
3561 * bean counters are appropriately updated.
3563 txg_wait_synced(spa->spa_dsl_pool, txg);
3565 spa_config_sync(spa, B_FALSE, B_TRUE);
3567 spa_history_log_version(spa, "create");
3569 spa->spa_minref = refcount_count(&spa->spa_refcount);
3571 mutex_exit(&spa_namespace_lock);
3579 * Get the root pool information from the root disk, then import the root pool
3580 * during the system boot up time.
3582 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3585 spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3588 nvlist_t *nvtop, *nvroot;
3591 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3595 * Add this top-level vdev to the child array.
3597 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3599 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3601 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3604 * Put this pool's top-level vdevs into a root vdev.
3606 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3607 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3608 VDEV_TYPE_ROOT) == 0);
3609 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3610 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3611 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3615 * Replace the existing vdev_tree with the new root vdev in
3616 * this pool's configuration (remove the old, add the new).
3618 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3619 nvlist_free(nvroot);
3624 * Walk the vdev tree and see if we can find a device with "better"
3625 * configuration. A configuration is "better" if the label on that
3626 * device has a more recent txg.
3629 spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3631 for (int c = 0; c < vd->vdev_children; c++)
3632 spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3634 if (vd->vdev_ops->vdev_op_leaf) {
3638 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3642 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3646 * Do we have a better boot device?
3648 if (label_txg > *txg) {
3657 * Import a root pool.
3659 * For x86. devpath_list will consist of devid and/or physpath name of
3660 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3661 * The GRUB "findroot" command will return the vdev we should boot.
3663 * For Sparc, devpath_list consists the physpath name of the booting device
3664 * no matter the rootpool is a single device pool or a mirrored pool.
3666 * "/pci@1f,0/ide@d/disk@0,0:a"
3669 spa_import_rootpool(char *devpath, char *devid)
3672 vdev_t *rvd, *bvd, *avd = NULL;
3673 nvlist_t *config, *nvtop;
3679 * Read the label from the boot device and generate a configuration.
3681 config = spa_generate_rootconf(devpath, devid, &guid);
3682 #if defined(_OBP) && defined(_KERNEL)
3683 if (config == NULL) {
3684 if (strstr(devpath, "/iscsi/ssd") != NULL) {
3686 get_iscsi_bootpath_phy(devpath);
3687 config = spa_generate_rootconf(devpath, devid, &guid);
3691 if (config == NULL) {
3692 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
3694 return (SET_ERROR(EIO));
3697 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3699 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3701 mutex_enter(&spa_namespace_lock);
3702 if ((spa = spa_lookup(pname)) != NULL) {
3704 * Remove the existing root pool from the namespace so that we
3705 * can replace it with the correct config we just read in.
3710 spa = spa_add(pname, config, NULL);
3711 spa->spa_is_root = B_TRUE;
3712 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3715 * Build up a vdev tree based on the boot device's label config.
3717 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3719 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3720 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3721 VDEV_ALLOC_ROOTPOOL);
3722 spa_config_exit(spa, SCL_ALL, FTAG);
3724 mutex_exit(&spa_namespace_lock);
3725 nvlist_free(config);
3726 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3732 * Get the boot vdev.
3734 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3735 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3736 (u_longlong_t)guid);
3737 error = SET_ERROR(ENOENT);
3742 * Determine if there is a better boot device.
3745 spa_alt_rootvdev(rvd, &avd, &txg);
3747 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3748 "try booting from '%s'", avd->vdev_path);
3749 error = SET_ERROR(EINVAL);
3754 * If the boot device is part of a spare vdev then ensure that
3755 * we're booting off the active spare.
3757 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3758 !bvd->vdev_isspare) {
3759 cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3760 "try booting from '%s'",
3762 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3763 error = SET_ERROR(EINVAL);
3769 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3771 spa_config_exit(spa, SCL_ALL, FTAG);
3772 mutex_exit(&spa_namespace_lock);
3774 nvlist_free(config);
3780 extern int vdev_geom_read_pool_label(const char *name, nvlist_t ***configs,
3784 spa_generate_rootconf(const char *name)
3786 nvlist_t **configs, **tops;
3788 nvlist_t *best_cfg, *nvtop, *nvroot;
3797 if (vdev_geom_read_pool_label(name, &configs, &count) != 0)
3800 ASSERT3U(count, !=, 0);
3802 for (i = 0; i < count; i++) {
3805 VERIFY(nvlist_lookup_uint64(configs[i], ZPOOL_CONFIG_POOL_TXG,
3807 if (txg > best_txg) {
3809 best_cfg = configs[i];
3814 * Multi-vdev root pool configuration discovery is not supported yet.
3817 nvlist_lookup_uint64(best_cfg, ZPOOL_CONFIG_VDEV_CHILDREN, &nchildren);
3819 nvlist_lookup_uint64_array(best_cfg, ZPOOL_CONFIG_HOLE_ARRAY,
3822 tops = kmem_zalloc(nchildren * sizeof(void *), KM_SLEEP);
3823 for (i = 0; i < nchildren; i++) {
3826 if (configs[i] == NULL)
3828 VERIFY(nvlist_lookup_nvlist(configs[i], ZPOOL_CONFIG_VDEV_TREE,
3830 nvlist_dup(nvtop, &tops[i], KM_SLEEP);
3832 for (i = 0; holes != NULL && i < nholes; i++) {
3835 if (tops[holes[i]] != NULL)
3837 nvlist_alloc(&tops[holes[i]], NV_UNIQUE_NAME, KM_SLEEP);
3838 VERIFY(nvlist_add_string(tops[holes[i]], ZPOOL_CONFIG_TYPE,
3839 VDEV_TYPE_HOLE) == 0);
3840 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_ID,
3842 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_GUID,
3845 for (i = 0; i < nchildren; i++) {
3846 if (tops[i] != NULL)
3848 nvlist_alloc(&tops[i], NV_UNIQUE_NAME, KM_SLEEP);
3849 VERIFY(nvlist_add_string(tops[i], ZPOOL_CONFIG_TYPE,
3850 VDEV_TYPE_MISSING) == 0);
3851 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_ID,
3853 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_GUID,
3858 * Create pool config based on the best vdev config.
3860 nvlist_dup(best_cfg, &config, KM_SLEEP);
3863 * Put this pool's top-level vdevs into a root vdev.
3865 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3867 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3868 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3869 VDEV_TYPE_ROOT) == 0);
3870 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3871 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3872 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3873 tops, nchildren) == 0);
3876 * Replace the existing vdev_tree with the new root vdev in
3877 * this pool's configuration (remove the old, add the new).
3879 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3882 * Drop vdev config elements that should not be present at pool level.
3884 nvlist_remove(config, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64);
3885 nvlist_remove(config, ZPOOL_CONFIG_TOP_GUID, DATA_TYPE_UINT64);
3887 for (i = 0; i < count; i++)
3888 nvlist_free(configs[i]);
3889 kmem_free(configs, count * sizeof(void *));
3890 for (i = 0; i < nchildren; i++)
3891 nvlist_free(tops[i]);
3892 kmem_free(tops, nchildren * sizeof(void *));
3893 nvlist_free(nvroot);
3898 spa_import_rootpool(const char *name)
3901 vdev_t *rvd, *bvd, *avd = NULL;
3902 nvlist_t *config, *nvtop;
3908 * Read the label from the boot device and generate a configuration.
3910 config = spa_generate_rootconf(name);
3912 mutex_enter(&spa_namespace_lock);
3913 if (config != NULL) {
3914 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3915 &pname) == 0 && strcmp(name, pname) == 0);
3916 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg)
3919 if ((spa = spa_lookup(pname)) != NULL) {
3921 * Remove the existing root pool from the namespace so
3922 * that we can replace it with the correct config
3927 spa = spa_add(pname, config, NULL);
3930 * Set spa_ubsync.ub_version as it can be used in vdev_alloc()
3931 * via spa_version().
3933 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
3934 &spa->spa_ubsync.ub_version) != 0)
3935 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
3936 } else if ((spa = spa_lookup(name)) == NULL) {
3937 cmn_err(CE_NOTE, "Cannot find the pool label for '%s'",
3941 VERIFY(nvlist_dup(spa->spa_config, &config, KM_SLEEP) == 0);
3943 spa->spa_is_root = B_TRUE;
3944 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3947 * Build up a vdev tree based on the boot device's label config.
3949 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3951 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3952 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3953 VDEV_ALLOC_ROOTPOOL);
3954 spa_config_exit(spa, SCL_ALL, FTAG);
3956 mutex_exit(&spa_namespace_lock);
3957 nvlist_free(config);
3958 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3963 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3965 spa_config_exit(spa, SCL_ALL, FTAG);
3966 mutex_exit(&spa_namespace_lock);
3968 nvlist_free(config);
3976 * Import a non-root pool into the system.
3979 spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
3982 char *altroot = NULL;
3983 spa_load_state_t state = SPA_LOAD_IMPORT;
3984 zpool_rewind_policy_t policy;
3985 uint64_t mode = spa_mode_global;
3986 uint64_t readonly = B_FALSE;
3989 nvlist_t **spares, **l2cache;
3990 uint_t nspares, nl2cache;
3993 * If a pool with this name exists, return failure.
3995 mutex_enter(&spa_namespace_lock);
3996 if (spa_lookup(pool) != NULL) {
3997 mutex_exit(&spa_namespace_lock);
3998 return (SET_ERROR(EEXIST));
4002 * Create and initialize the spa structure.
4004 (void) nvlist_lookup_string(props,
4005 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4006 (void) nvlist_lookup_uint64(props,
4007 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
4010 spa = spa_add(pool, config, altroot);
4011 spa->spa_import_flags = flags;
4014 * Verbatim import - Take a pool and insert it into the namespace
4015 * as if it had been loaded at boot.
4017 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
4019 spa_configfile_set(spa, props, B_FALSE);
4021 spa_config_sync(spa, B_FALSE, B_TRUE);
4023 mutex_exit(&spa_namespace_lock);
4027 spa_activate(spa, mode);
4030 * Don't start async tasks until we know everything is healthy.
4032 spa_async_suspend(spa);
4034 zpool_get_rewind_policy(config, &policy);
4035 if (policy.zrp_request & ZPOOL_DO_REWIND)
4036 state = SPA_LOAD_RECOVER;
4039 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
4040 * because the user-supplied config is actually the one to trust when
4043 if (state != SPA_LOAD_RECOVER)
4044 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
4046 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
4047 policy.zrp_request);
4050 * Propagate anything learned while loading the pool and pass it
4051 * back to caller (i.e. rewind info, missing devices, etc).
4053 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4054 spa->spa_load_info) == 0);
4056 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4058 * Toss any existing sparelist, as it doesn't have any validity
4059 * anymore, and conflicts with spa_has_spare().
4061 if (spa->spa_spares.sav_config) {
4062 nvlist_free(spa->spa_spares.sav_config);
4063 spa->spa_spares.sav_config = NULL;
4064 spa_load_spares(spa);
4066 if (spa->spa_l2cache.sav_config) {
4067 nvlist_free(spa->spa_l2cache.sav_config);
4068 spa->spa_l2cache.sav_config = NULL;
4069 spa_load_l2cache(spa);
4072 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4075 error = spa_validate_aux(spa, nvroot, -1ULL,
4078 error = spa_validate_aux(spa, nvroot, -1ULL,
4079 VDEV_ALLOC_L2CACHE);
4080 spa_config_exit(spa, SCL_ALL, FTAG);
4083 spa_configfile_set(spa, props, B_FALSE);
4085 if (error != 0 || (props && spa_writeable(spa) &&
4086 (error = spa_prop_set(spa, props)))) {
4088 spa_deactivate(spa);
4090 mutex_exit(&spa_namespace_lock);
4094 spa_async_resume(spa);
4097 * Override any spares and level 2 cache devices as specified by
4098 * the user, as these may have correct device names/devids, etc.
4100 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
4101 &spares, &nspares) == 0) {
4102 if (spa->spa_spares.sav_config)
4103 VERIFY(nvlist_remove(spa->spa_spares.sav_config,
4104 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
4106 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
4107 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4108 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
4109 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
4110 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4111 spa_load_spares(spa);
4112 spa_config_exit(spa, SCL_ALL, FTAG);
4113 spa->spa_spares.sav_sync = B_TRUE;
4115 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
4116 &l2cache, &nl2cache) == 0) {
4117 if (spa->spa_l2cache.sav_config)
4118 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
4119 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
4121 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
4122 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4123 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
4124 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
4125 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4126 spa_load_l2cache(spa);
4127 spa_config_exit(spa, SCL_ALL, FTAG);
4128 spa->spa_l2cache.sav_sync = B_TRUE;
4132 * Check for any removed devices.
4134 if (spa->spa_autoreplace) {
4135 spa_aux_check_removed(&spa->spa_spares);
4136 spa_aux_check_removed(&spa->spa_l2cache);
4139 if (spa_writeable(spa)) {
4141 * Update the config cache to include the newly-imported pool.
4143 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4147 * It's possible that the pool was expanded while it was exported.
4148 * We kick off an async task to handle this for us.
4150 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
4152 mutex_exit(&spa_namespace_lock);
4153 spa_history_log_version(spa, "import");
4157 zvol_create_minors(pool);
4164 spa_tryimport(nvlist_t *tryconfig)
4166 nvlist_t *config = NULL;
4172 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
4175 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
4179 * Create and initialize the spa structure.
4181 mutex_enter(&spa_namespace_lock);
4182 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
4183 spa_activate(spa, FREAD);
4186 * Pass off the heavy lifting to spa_load().
4187 * Pass TRUE for mosconfig because the user-supplied config
4188 * is actually the one to trust when doing an import.
4190 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
4193 * If 'tryconfig' was at least parsable, return the current config.
4195 if (spa->spa_root_vdev != NULL) {
4196 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4197 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
4199 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4201 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
4202 spa->spa_uberblock.ub_timestamp) == 0);
4203 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4204 spa->spa_load_info) == 0);
4207 * If the bootfs property exists on this pool then we
4208 * copy it out so that external consumers can tell which
4209 * pools are bootable.
4211 if ((!error || error == EEXIST) && spa->spa_bootfs) {
4212 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4215 * We have to play games with the name since the
4216 * pool was opened as TRYIMPORT_NAME.
4218 if (dsl_dsobj_to_dsname(spa_name(spa),
4219 spa->spa_bootfs, tmpname) == 0) {
4221 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4223 cp = strchr(tmpname, '/');
4225 (void) strlcpy(dsname, tmpname,
4228 (void) snprintf(dsname, MAXPATHLEN,
4229 "%s/%s", poolname, ++cp);
4231 VERIFY(nvlist_add_string(config,
4232 ZPOOL_CONFIG_BOOTFS, dsname) == 0);
4233 kmem_free(dsname, MAXPATHLEN);
4235 kmem_free(tmpname, MAXPATHLEN);
4239 * Add the list of hot spares and level 2 cache devices.
4241 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4242 spa_add_spares(spa, config);
4243 spa_add_l2cache(spa, config);
4244 spa_config_exit(spa, SCL_CONFIG, FTAG);
4248 spa_deactivate(spa);
4250 mutex_exit(&spa_namespace_lock);
4256 * Pool export/destroy
4258 * The act of destroying or exporting a pool is very simple. We make sure there
4259 * is no more pending I/O and any references to the pool are gone. Then, we
4260 * update the pool state and sync all the labels to disk, removing the
4261 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4262 * we don't sync the labels or remove the configuration cache.
4265 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
4266 boolean_t force, boolean_t hardforce)
4273 if (!(spa_mode_global & FWRITE))
4274 return (SET_ERROR(EROFS));
4276 mutex_enter(&spa_namespace_lock);
4277 if ((spa = spa_lookup(pool)) == NULL) {
4278 mutex_exit(&spa_namespace_lock);
4279 return (SET_ERROR(ENOENT));
4283 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4284 * reacquire the namespace lock, and see if we can export.
4286 spa_open_ref(spa, FTAG);
4287 mutex_exit(&spa_namespace_lock);
4288 spa_async_suspend(spa);
4289 mutex_enter(&spa_namespace_lock);
4290 spa_close(spa, FTAG);
4293 * The pool will be in core if it's openable,
4294 * in which case we can modify its state.
4296 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
4298 * Objsets may be open only because they're dirty, so we
4299 * have to force it to sync before checking spa_refcnt.
4301 txg_wait_synced(spa->spa_dsl_pool, 0);
4304 * A pool cannot be exported or destroyed if there are active
4305 * references. If we are resetting a pool, allow references by
4306 * fault injection handlers.
4308 if (!spa_refcount_zero(spa) ||
4309 (spa->spa_inject_ref != 0 &&
4310 new_state != POOL_STATE_UNINITIALIZED)) {
4311 spa_async_resume(spa);
4312 mutex_exit(&spa_namespace_lock);
4313 return (SET_ERROR(EBUSY));
4317 * A pool cannot be exported if it has an active shared spare.
4318 * This is to prevent other pools stealing the active spare
4319 * from an exported pool. At user's own will, such pool can
4320 * be forcedly exported.
4322 if (!force && new_state == POOL_STATE_EXPORTED &&
4323 spa_has_active_shared_spare(spa)) {
4324 spa_async_resume(spa);
4325 mutex_exit(&spa_namespace_lock);
4326 return (SET_ERROR(EXDEV));
4330 * We want this to be reflected on every label,
4331 * so mark them all dirty. spa_unload() will do the
4332 * final sync that pushes these changes out.
4334 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
4335 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4336 spa->spa_state = new_state;
4337 spa->spa_final_txg = spa_last_synced_txg(spa) +
4339 vdev_config_dirty(spa->spa_root_vdev);
4340 spa_config_exit(spa, SCL_ALL, FTAG);
4344 spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
4346 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4348 spa_deactivate(spa);
4351 if (oldconfig && spa->spa_config)
4352 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
4354 if (new_state != POOL_STATE_UNINITIALIZED) {
4356 spa_config_sync(spa, B_TRUE, B_TRUE);
4359 mutex_exit(&spa_namespace_lock);
4365 * Destroy a storage pool.
4368 spa_destroy(char *pool)
4370 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
4375 * Export a storage pool.
4378 spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
4379 boolean_t hardforce)
4381 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
4386 * Similar to spa_export(), this unloads the spa_t without actually removing it
4387 * from the namespace in any way.
4390 spa_reset(char *pool)
4392 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
4397 * ==========================================================================
4398 * Device manipulation
4399 * ==========================================================================
4403 * Add a device to a storage pool.
4406 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
4410 vdev_t *rvd = spa->spa_root_vdev;
4412 nvlist_t **spares, **l2cache;
4413 uint_t nspares, nl2cache;
4415 ASSERT(spa_writeable(spa));
4417 txg = spa_vdev_enter(spa);
4419 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
4420 VDEV_ALLOC_ADD)) != 0)
4421 return (spa_vdev_exit(spa, NULL, txg, error));
4423 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */
4425 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
4429 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
4433 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
4434 return (spa_vdev_exit(spa, vd, txg, EINVAL));
4436 if (vd->vdev_children != 0 &&
4437 (error = vdev_create(vd, txg, B_FALSE)) != 0)
4438 return (spa_vdev_exit(spa, vd, txg, error));
4441 * We must validate the spares and l2cache devices after checking the
4442 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
4444 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
4445 return (spa_vdev_exit(spa, vd, txg, error));
4448 * Transfer each new top-level vdev from vd to rvd.
4450 for (int c = 0; c < vd->vdev_children; c++) {
4453 * Set the vdev id to the first hole, if one exists.
4455 for (id = 0; id < rvd->vdev_children; id++) {
4456 if (rvd->vdev_child[id]->vdev_ishole) {
4457 vdev_free(rvd->vdev_child[id]);
4461 tvd = vd->vdev_child[c];
4462 vdev_remove_child(vd, tvd);
4464 vdev_add_child(rvd, tvd);
4465 vdev_config_dirty(tvd);
4469 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
4470 ZPOOL_CONFIG_SPARES);
4471 spa_load_spares(spa);
4472 spa->spa_spares.sav_sync = B_TRUE;
4475 if (nl2cache != 0) {
4476 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
4477 ZPOOL_CONFIG_L2CACHE);
4478 spa_load_l2cache(spa);
4479 spa->spa_l2cache.sav_sync = B_TRUE;
4483 * We have to be careful when adding new vdevs to an existing pool.
4484 * If other threads start allocating from these vdevs before we
4485 * sync the config cache, and we lose power, then upon reboot we may
4486 * fail to open the pool because there are DVAs that the config cache
4487 * can't translate. Therefore, we first add the vdevs without
4488 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4489 * and then let spa_config_update() initialize the new metaslabs.
4491 * spa_load() checks for added-but-not-initialized vdevs, so that
4492 * if we lose power at any point in this sequence, the remaining
4493 * steps will be completed the next time we load the pool.
4495 (void) spa_vdev_exit(spa, vd, txg, 0);
4497 mutex_enter(&spa_namespace_lock);
4498 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4499 mutex_exit(&spa_namespace_lock);
4505 * Attach a device to a mirror. The arguments are the path to any device
4506 * in the mirror, and the nvroot for the new device. If the path specifies
4507 * a device that is not mirrored, we automatically insert the mirror vdev.
4509 * If 'replacing' is specified, the new device is intended to replace the
4510 * existing device; in this case the two devices are made into their own
4511 * mirror using the 'replacing' vdev, which is functionally identical to
4512 * the mirror vdev (it actually reuses all the same ops) but has a few
4513 * extra rules: you can't attach to it after it's been created, and upon
4514 * completion of resilvering, the first disk (the one being replaced)
4515 * is automatically detached.
4518 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
4520 uint64_t txg, dtl_max_txg;
4521 vdev_t *rvd = spa->spa_root_vdev;
4522 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
4524 char *oldvdpath, *newvdpath;
4528 ASSERT(spa_writeable(spa));
4530 txg = spa_vdev_enter(spa);
4532 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
4535 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4537 if (!oldvd->vdev_ops->vdev_op_leaf)
4538 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4540 pvd = oldvd->vdev_parent;
4542 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
4543 VDEV_ALLOC_ATTACH)) != 0)
4544 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4546 if (newrootvd->vdev_children != 1)
4547 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4549 newvd = newrootvd->vdev_child[0];
4551 if (!newvd->vdev_ops->vdev_op_leaf)
4552 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4554 if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
4555 return (spa_vdev_exit(spa, newrootvd, txg, error));
4558 * Spares can't replace logs
4560 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
4561 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4565 * For attach, the only allowable parent is a mirror or the root
4568 if (pvd->vdev_ops != &vdev_mirror_ops &&
4569 pvd->vdev_ops != &vdev_root_ops)
4570 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4572 pvops = &vdev_mirror_ops;
4575 * Active hot spares can only be replaced by inactive hot
4578 if (pvd->vdev_ops == &vdev_spare_ops &&
4579 oldvd->vdev_isspare &&
4580 !spa_has_spare(spa, newvd->vdev_guid))
4581 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4584 * If the source is a hot spare, and the parent isn't already a
4585 * spare, then we want to create a new hot spare. Otherwise, we
4586 * want to create a replacing vdev. The user is not allowed to
4587 * attach to a spared vdev child unless the 'isspare' state is
4588 * the same (spare replaces spare, non-spare replaces
4591 if (pvd->vdev_ops == &vdev_replacing_ops &&
4592 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
4593 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4594 } else if (pvd->vdev_ops == &vdev_spare_ops &&
4595 newvd->vdev_isspare != oldvd->vdev_isspare) {
4596 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4599 if (newvd->vdev_isspare)
4600 pvops = &vdev_spare_ops;
4602 pvops = &vdev_replacing_ops;
4606 * Make sure the new device is big enough.
4608 if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
4609 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
4612 * The new device cannot have a higher alignment requirement
4613 * than the top-level vdev.
4615 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4616 return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4619 * If this is an in-place replacement, update oldvd's path and devid
4620 * to make it distinguishable from newvd, and unopenable from now on.
4622 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4623 spa_strfree(oldvd->vdev_path);
4624 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4626 (void) sprintf(oldvd->vdev_path, "%s/%s",
4627 newvd->vdev_path, "old");
4628 if (oldvd->vdev_devid != NULL) {
4629 spa_strfree(oldvd->vdev_devid);
4630 oldvd->vdev_devid = NULL;
4634 /* mark the device being resilvered */
4635 newvd->vdev_resilver_txg = txg;
4638 * If the parent is not a mirror, or if we're replacing, insert the new
4639 * mirror/replacing/spare vdev above oldvd.
4641 if (pvd->vdev_ops != pvops)
4642 pvd = vdev_add_parent(oldvd, pvops);
4644 ASSERT(pvd->vdev_top->vdev_parent == rvd);
4645 ASSERT(pvd->vdev_ops == pvops);
4646 ASSERT(oldvd->vdev_parent == pvd);
4649 * Extract the new device from its root and add it to pvd.
4651 vdev_remove_child(newrootvd, newvd);
4652 newvd->vdev_id = pvd->vdev_children;
4653 newvd->vdev_crtxg = oldvd->vdev_crtxg;
4654 vdev_add_child(pvd, newvd);
4656 tvd = newvd->vdev_top;
4657 ASSERT(pvd->vdev_top == tvd);
4658 ASSERT(tvd->vdev_parent == rvd);
4660 vdev_config_dirty(tvd);
4663 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4664 * for any dmu_sync-ed blocks. It will propagate upward when
4665 * spa_vdev_exit() calls vdev_dtl_reassess().
4667 dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4669 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4670 dtl_max_txg - TXG_INITIAL);
4672 if (newvd->vdev_isspare) {
4673 spa_spare_activate(newvd);
4674 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
4677 oldvdpath = spa_strdup(oldvd->vdev_path);
4678 newvdpath = spa_strdup(newvd->vdev_path);
4679 newvd_isspare = newvd->vdev_isspare;
4682 * Mark newvd's DTL dirty in this txg.
4684 vdev_dirty(tvd, VDD_DTL, newvd, txg);
4687 * Restart the resilver
4689 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4694 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4696 spa_history_log_internal(spa, "vdev attach", NULL,
4697 "%s vdev=%s %s vdev=%s",
4698 replacing && newvd_isspare ? "spare in" :
4699 replacing ? "replace" : "attach", newvdpath,
4700 replacing ? "for" : "to", oldvdpath);
4702 spa_strfree(oldvdpath);
4703 spa_strfree(newvdpath);
4705 if (spa->spa_bootfs)
4706 spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
4712 * Detach a device from a mirror or replacing vdev.
4714 * If 'replace_done' is specified, only detach if the parent
4715 * is a replacing vdev.
4718 spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4722 vdev_t *rvd = spa->spa_root_vdev;
4723 vdev_t *vd, *pvd, *cvd, *tvd;
4724 boolean_t unspare = B_FALSE;
4725 uint64_t unspare_guid = 0;
4728 ASSERT(spa_writeable(spa));
4730 txg = spa_vdev_enter(spa);
4732 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4735 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4737 if (!vd->vdev_ops->vdev_op_leaf)
4738 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4740 pvd = vd->vdev_parent;
4743 * If the parent/child relationship is not as expected, don't do it.
4744 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4745 * vdev that's replacing B with C. The user's intent in replacing
4746 * is to go from M(A,B) to M(A,C). If the user decides to cancel
4747 * the replace by detaching C, the expected behavior is to end up
4748 * M(A,B). But suppose that right after deciding to detach C,
4749 * the replacement of B completes. We would have M(A,C), and then
4750 * ask to detach C, which would leave us with just A -- not what
4751 * the user wanted. To prevent this, we make sure that the
4752 * parent/child relationship hasn't changed -- in this example,
4753 * that C's parent is still the replacing vdev R.
4755 if (pvd->vdev_guid != pguid && pguid != 0)
4756 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4759 * Only 'replacing' or 'spare' vdevs can be replaced.
4761 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4762 pvd->vdev_ops != &vdev_spare_ops)
4763 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4765 ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4766 spa_version(spa) >= SPA_VERSION_SPARES);
4769 * Only mirror, replacing, and spare vdevs support detach.
4771 if (pvd->vdev_ops != &vdev_replacing_ops &&
4772 pvd->vdev_ops != &vdev_mirror_ops &&
4773 pvd->vdev_ops != &vdev_spare_ops)
4774 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4777 * If this device has the only valid copy of some data,
4778 * we cannot safely detach it.
4780 if (vdev_dtl_required(vd))
4781 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4783 ASSERT(pvd->vdev_children >= 2);
4786 * If we are detaching the second disk from a replacing vdev, then
4787 * check to see if we changed the original vdev's path to have "/old"
4788 * at the end in spa_vdev_attach(). If so, undo that change now.
4790 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4791 vd->vdev_path != NULL) {
4792 size_t len = strlen(vd->vdev_path);
4794 for (int c = 0; c < pvd->vdev_children; c++) {
4795 cvd = pvd->vdev_child[c];
4797 if (cvd == vd || cvd->vdev_path == NULL)
4800 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4801 strcmp(cvd->vdev_path + len, "/old") == 0) {
4802 spa_strfree(cvd->vdev_path);
4803 cvd->vdev_path = spa_strdup(vd->vdev_path);
4810 * If we are detaching the original disk from a spare, then it implies
4811 * that the spare should become a real disk, and be removed from the
4812 * active spare list for the pool.
4814 if (pvd->vdev_ops == &vdev_spare_ops &&
4816 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4820 * Erase the disk labels so the disk can be used for other things.
4821 * This must be done after all other error cases are handled,
4822 * but before we disembowel vd (so we can still do I/O to it).
4823 * But if we can't do it, don't treat the error as fatal --
4824 * it may be that the unwritability of the disk is the reason
4825 * it's being detached!
4827 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4830 * Remove vd from its parent and compact the parent's children.
4832 vdev_remove_child(pvd, vd);
4833 vdev_compact_children(pvd);
4836 * Remember one of the remaining children so we can get tvd below.
4838 cvd = pvd->vdev_child[pvd->vdev_children - 1];
4841 * If we need to remove the remaining child from the list of hot spares,
4842 * do it now, marking the vdev as no longer a spare in the process.
4843 * We must do this before vdev_remove_parent(), because that can
4844 * change the GUID if it creates a new toplevel GUID. For a similar
4845 * reason, we must remove the spare now, in the same txg as the detach;
4846 * otherwise someone could attach a new sibling, change the GUID, and
4847 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4850 ASSERT(cvd->vdev_isspare);
4851 spa_spare_remove(cvd);
4852 unspare_guid = cvd->vdev_guid;
4853 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4854 cvd->vdev_unspare = B_TRUE;
4858 * If the parent mirror/replacing vdev only has one child,
4859 * the parent is no longer needed. Remove it from the tree.
4861 if (pvd->vdev_children == 1) {
4862 if (pvd->vdev_ops == &vdev_spare_ops)
4863 cvd->vdev_unspare = B_FALSE;
4864 vdev_remove_parent(cvd);
4869 * We don't set tvd until now because the parent we just removed
4870 * may have been the previous top-level vdev.
4872 tvd = cvd->vdev_top;
4873 ASSERT(tvd->vdev_parent == rvd);
4876 * Reevaluate the parent vdev state.
4878 vdev_propagate_state(cvd);
4881 * If the 'autoexpand' property is set on the pool then automatically
4882 * try to expand the size of the pool. For example if the device we
4883 * just detached was smaller than the others, it may be possible to
4884 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4885 * first so that we can obtain the updated sizes of the leaf vdevs.
4887 if (spa->spa_autoexpand) {
4889 vdev_expand(tvd, txg);
4892 vdev_config_dirty(tvd);
4895 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
4896 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4897 * But first make sure we're not on any *other* txg's DTL list, to
4898 * prevent vd from being accessed after it's freed.
4900 vdpath = spa_strdup(vd->vdev_path);
4901 for (int t = 0; t < TXG_SIZE; t++)
4902 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4903 vd->vdev_detached = B_TRUE;
4904 vdev_dirty(tvd, VDD_DTL, vd, txg);
4906 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
4908 /* hang on to the spa before we release the lock */
4909 spa_open_ref(spa, FTAG);
4911 error = spa_vdev_exit(spa, vd, txg, 0);
4913 spa_history_log_internal(spa, "detach", NULL,
4915 spa_strfree(vdpath);
4918 * If this was the removal of the original device in a hot spare vdev,
4919 * then we want to go through and remove the device from the hot spare
4920 * list of every other pool.
4923 spa_t *altspa = NULL;
4925 mutex_enter(&spa_namespace_lock);
4926 while ((altspa = spa_next(altspa)) != NULL) {
4927 if (altspa->spa_state != POOL_STATE_ACTIVE ||
4931 spa_open_ref(altspa, FTAG);
4932 mutex_exit(&spa_namespace_lock);
4933 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
4934 mutex_enter(&spa_namespace_lock);
4935 spa_close(altspa, FTAG);
4937 mutex_exit(&spa_namespace_lock);
4939 /* search the rest of the vdevs for spares to remove */
4940 spa_vdev_resilver_done(spa);
4943 /* all done with the spa; OK to release */
4944 mutex_enter(&spa_namespace_lock);
4945 spa_close(spa, FTAG);
4946 mutex_exit(&spa_namespace_lock);
4952 * Split a set of devices from their mirrors, and create a new pool from them.
4955 spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
4956 nvlist_t *props, boolean_t exp)
4959 uint64_t txg, *glist;
4961 uint_t c, children, lastlog;
4962 nvlist_t **child, *nvl, *tmp;
4964 char *altroot = NULL;
4965 vdev_t *rvd, **vml = NULL; /* vdev modify list */
4966 boolean_t activate_slog;
4968 ASSERT(spa_writeable(spa));
4970 txg = spa_vdev_enter(spa);
4972 /* clear the log and flush everything up to now */
4973 activate_slog = spa_passivate_log(spa);
4974 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4975 error = spa_offline_log(spa);
4976 txg = spa_vdev_config_enter(spa);
4979 spa_activate_log(spa);
4982 return (spa_vdev_exit(spa, NULL, txg, error));
4984 /* check new spa name before going any further */
4985 if (spa_lookup(newname) != NULL)
4986 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
4989 * scan through all the children to ensure they're all mirrors
4991 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
4992 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
4994 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4996 /* first, check to ensure we've got the right child count */
4997 rvd = spa->spa_root_vdev;
4999 for (c = 0; c < rvd->vdev_children; c++) {
5000 vdev_t *vd = rvd->vdev_child[c];
5002 /* don't count the holes & logs as children */
5003 if (vd->vdev_islog || vd->vdev_ishole) {
5011 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
5012 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5014 /* next, ensure no spare or cache devices are part of the split */
5015 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
5016 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
5017 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5019 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
5020 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
5022 /* then, loop over each vdev and validate it */
5023 for (c = 0; c < children; c++) {
5024 uint64_t is_hole = 0;
5026 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
5030 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
5031 spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
5034 error = SET_ERROR(EINVAL);
5039 /* which disk is going to be split? */
5040 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
5042 error = SET_ERROR(EINVAL);
5046 /* look it up in the spa */
5047 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
5048 if (vml[c] == NULL) {
5049 error = SET_ERROR(ENODEV);
5053 /* make sure there's nothing stopping the split */
5054 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
5055 vml[c]->vdev_islog ||
5056 vml[c]->vdev_ishole ||
5057 vml[c]->vdev_isspare ||
5058 vml[c]->vdev_isl2cache ||
5059 !vdev_writeable(vml[c]) ||
5060 vml[c]->vdev_children != 0 ||
5061 vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
5062 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
5063 error = SET_ERROR(EINVAL);
5067 if (vdev_dtl_required(vml[c])) {
5068 error = SET_ERROR(EBUSY);
5072 /* we need certain info from the top level */
5073 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
5074 vml[c]->vdev_top->vdev_ms_array) == 0);
5075 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
5076 vml[c]->vdev_top->vdev_ms_shift) == 0);
5077 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
5078 vml[c]->vdev_top->vdev_asize) == 0);
5079 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
5080 vml[c]->vdev_top->vdev_ashift) == 0);
5084 kmem_free(vml, children * sizeof (vdev_t *));
5085 kmem_free(glist, children * sizeof (uint64_t));
5086 return (spa_vdev_exit(spa, NULL, txg, error));
5089 /* stop writers from using the disks */
5090 for (c = 0; c < children; c++) {
5092 vml[c]->vdev_offline = B_TRUE;
5094 vdev_reopen(spa->spa_root_vdev);
5097 * Temporarily record the splitting vdevs in the spa config. This
5098 * will disappear once the config is regenerated.
5100 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5101 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
5102 glist, children) == 0);
5103 kmem_free(glist, children * sizeof (uint64_t));
5105 mutex_enter(&spa->spa_props_lock);
5106 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
5108 mutex_exit(&spa->spa_props_lock);
5109 spa->spa_config_splitting = nvl;
5110 vdev_config_dirty(spa->spa_root_vdev);
5112 /* configure and create the new pool */
5113 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
5114 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
5115 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
5116 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5117 spa_version(spa)) == 0);
5118 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
5119 spa->spa_config_txg) == 0);
5120 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
5121 spa_generate_guid(NULL)) == 0);
5122 (void) nvlist_lookup_string(props,
5123 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
5125 /* add the new pool to the namespace */
5126 newspa = spa_add(newname, config, altroot);
5127 newspa->spa_config_txg = spa->spa_config_txg;
5128 spa_set_log_state(newspa, SPA_LOG_CLEAR);
5130 /* release the spa config lock, retaining the namespace lock */
5131 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5133 if (zio_injection_enabled)
5134 zio_handle_panic_injection(spa, FTAG, 1);
5136 spa_activate(newspa, spa_mode_global);
5137 spa_async_suspend(newspa);
5140 /* mark that we are creating new spa by splitting */
5141 newspa->spa_splitting_newspa = B_TRUE;
5143 /* create the new pool from the disks of the original pool */
5144 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
5146 newspa->spa_splitting_newspa = B_FALSE;
5151 /* if that worked, generate a real config for the new pool */
5152 if (newspa->spa_root_vdev != NULL) {
5153 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
5154 NV_UNIQUE_NAME, KM_SLEEP) == 0);
5155 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
5156 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
5157 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
5162 if (props != NULL) {
5163 spa_configfile_set(newspa, props, B_FALSE);
5164 error = spa_prop_set(newspa, props);
5169 /* flush everything */
5170 txg = spa_vdev_config_enter(newspa);
5171 vdev_config_dirty(newspa->spa_root_vdev);
5172 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
5174 if (zio_injection_enabled)
5175 zio_handle_panic_injection(spa, FTAG, 2);
5177 spa_async_resume(newspa);
5179 /* finally, update the original pool's config */
5180 txg = spa_vdev_config_enter(spa);
5181 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
5182 error = dmu_tx_assign(tx, TXG_WAIT);
5185 for (c = 0; c < children; c++) {
5186 if (vml[c] != NULL) {
5189 spa_history_log_internal(spa, "detach", tx,
5190 "vdev=%s", vml[c]->vdev_path);
5194 vdev_config_dirty(spa->spa_root_vdev);
5195 spa->spa_config_splitting = NULL;
5199 (void) spa_vdev_exit(spa, NULL, txg, 0);
5201 if (zio_injection_enabled)
5202 zio_handle_panic_injection(spa, FTAG, 3);
5204 /* split is complete; log a history record */
5205 spa_history_log_internal(newspa, "split", NULL,
5206 "from pool %s", spa_name(spa));
5208 kmem_free(vml, children * sizeof (vdev_t *));
5210 /* if we're not going to mount the filesystems in userland, export */
5212 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
5219 spa_deactivate(newspa);
5222 txg = spa_vdev_config_enter(spa);
5224 /* re-online all offlined disks */
5225 for (c = 0; c < children; c++) {
5227 vml[c]->vdev_offline = B_FALSE;
5229 vdev_reopen(spa->spa_root_vdev);
5231 nvlist_free(spa->spa_config_splitting);
5232 spa->spa_config_splitting = NULL;
5233 (void) spa_vdev_exit(spa, NULL, txg, error);
5235 kmem_free(vml, children * sizeof (vdev_t *));
5240 spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
5242 for (int i = 0; i < count; i++) {
5245 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
5248 if (guid == target_guid)
5256 spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
5257 nvlist_t *dev_to_remove)
5259 nvlist_t **newdev = NULL;
5262 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
5264 for (int i = 0, j = 0; i < count; i++) {
5265 if (dev[i] == dev_to_remove)
5267 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
5270 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
5271 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
5273 for (int i = 0; i < count - 1; i++)
5274 nvlist_free(newdev[i]);
5277 kmem_free(newdev, (count - 1) * sizeof (void *));
5281 * Evacuate the device.
5284 spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
5289 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5290 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5291 ASSERT(vd == vd->vdev_top);
5294 * Evacuate the device. We don't hold the config lock as writer
5295 * since we need to do I/O but we do keep the
5296 * spa_namespace_lock held. Once this completes the device
5297 * should no longer have any blocks allocated on it.
5299 if (vd->vdev_islog) {
5300 if (vd->vdev_stat.vs_alloc != 0)
5301 error = spa_offline_log(spa);
5303 error = SET_ERROR(ENOTSUP);
5310 * The evacuation succeeded. Remove any remaining MOS metadata
5311 * associated with this vdev, and wait for these changes to sync.
5313 ASSERT0(vd->vdev_stat.vs_alloc);
5314 txg = spa_vdev_config_enter(spa);
5315 vd->vdev_removing = B_TRUE;
5316 vdev_dirty(vd, 0, NULL, txg);
5317 vdev_config_dirty(vd);
5318 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5324 * Complete the removal by cleaning up the namespace.
5327 spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
5329 vdev_t *rvd = spa->spa_root_vdev;
5330 uint64_t id = vd->vdev_id;
5331 boolean_t last_vdev = (id == (rvd->vdev_children - 1));
5333 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5334 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5335 ASSERT(vd == vd->vdev_top);
5338 * Only remove any devices which are empty.
5340 if (vd->vdev_stat.vs_alloc != 0)
5343 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5345 if (list_link_active(&vd->vdev_state_dirty_node))
5346 vdev_state_clean(vd);
5347 if (list_link_active(&vd->vdev_config_dirty_node))
5348 vdev_config_clean(vd);
5353 vdev_compact_children(rvd);
5355 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
5356 vdev_add_child(rvd, vd);
5358 vdev_config_dirty(rvd);
5361 * Reassess the health of our root vdev.
5367 * Remove a device from the pool -
5369 * Removing a device from the vdev namespace requires several steps
5370 * and can take a significant amount of time. As a result we use
5371 * the spa_vdev_config_[enter/exit] functions which allow us to
5372 * grab and release the spa_config_lock while still holding the namespace
5373 * lock. During each step the configuration is synced out.
5375 * Currently, this supports removing only hot spares, slogs, and level 2 ARC
5379 spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
5382 metaslab_group_t *mg;
5383 nvlist_t **spares, **l2cache, *nv;
5385 uint_t nspares, nl2cache;
5387 boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
5389 ASSERT(spa_writeable(spa));
5392 txg = spa_vdev_enter(spa);
5394 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5396 if (spa->spa_spares.sav_vdevs != NULL &&
5397 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5398 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
5399 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
5401 * Only remove the hot spare if it's not currently in use
5404 if (vd == NULL || unspare) {
5405 spa_vdev_remove_aux(spa->spa_spares.sav_config,
5406 ZPOOL_CONFIG_SPARES, spares, nspares, nv);
5407 spa_load_spares(spa);
5408 spa->spa_spares.sav_sync = B_TRUE;
5410 error = SET_ERROR(EBUSY);
5412 } else if (spa->spa_l2cache.sav_vdevs != NULL &&
5413 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5414 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
5415 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
5417 * Cache devices can always be removed.
5419 spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
5420 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
5421 spa_load_l2cache(spa);
5422 spa->spa_l2cache.sav_sync = B_TRUE;
5423 } else if (vd != NULL && vd->vdev_islog) {
5425 ASSERT(vd == vd->vdev_top);
5428 * XXX - Once we have bp-rewrite this should
5429 * become the common case.
5435 * Stop allocating from this vdev.
5437 metaslab_group_passivate(mg);
5440 * Wait for the youngest allocations and frees to sync,
5441 * and then wait for the deferral of those frees to finish.
5443 spa_vdev_config_exit(spa, NULL,
5444 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
5447 * Attempt to evacuate the vdev.
5449 error = spa_vdev_remove_evacuate(spa, vd);
5451 txg = spa_vdev_config_enter(spa);
5454 * If we couldn't evacuate the vdev, unwind.
5457 metaslab_group_activate(mg);
5458 return (spa_vdev_exit(spa, NULL, txg, error));
5462 * Clean up the vdev namespace.
5464 spa_vdev_remove_from_namespace(spa, vd);
5466 } else if (vd != NULL) {
5468 * Normal vdevs cannot be removed (yet).
5470 error = SET_ERROR(ENOTSUP);
5473 * There is no vdev of any kind with the specified guid.
5475 error = SET_ERROR(ENOENT);
5479 return (spa_vdev_exit(spa, NULL, txg, error));
5485 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5486 * currently spared, so we can detach it.
5489 spa_vdev_resilver_done_hunt(vdev_t *vd)
5491 vdev_t *newvd, *oldvd;
5493 for (int c = 0; c < vd->vdev_children; c++) {
5494 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
5500 * Check for a completed replacement. We always consider the first
5501 * vdev in the list to be the oldest vdev, and the last one to be
5502 * the newest (see spa_vdev_attach() for how that works). In
5503 * the case where the newest vdev is faulted, we will not automatically
5504 * remove it after a resilver completes. This is OK as it will require
5505 * user intervention to determine which disk the admin wishes to keep.
5507 if (vd->vdev_ops == &vdev_replacing_ops) {
5508 ASSERT(vd->vdev_children > 1);
5510 newvd = vd->vdev_child[vd->vdev_children - 1];
5511 oldvd = vd->vdev_child[0];
5513 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
5514 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5515 !vdev_dtl_required(oldvd))
5520 * Check for a completed resilver with the 'unspare' flag set.
5522 if (vd->vdev_ops == &vdev_spare_ops) {
5523 vdev_t *first = vd->vdev_child[0];
5524 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
5526 if (last->vdev_unspare) {
5529 } else if (first->vdev_unspare) {
5536 if (oldvd != NULL &&
5537 vdev_dtl_empty(newvd, DTL_MISSING) &&
5538 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5539 !vdev_dtl_required(oldvd))
5543 * If there are more than two spares attached to a disk,
5544 * and those spares are not required, then we want to
5545 * attempt to free them up now so that they can be used
5546 * by other pools. Once we're back down to a single
5547 * disk+spare, we stop removing them.
5549 if (vd->vdev_children > 2) {
5550 newvd = vd->vdev_child[1];
5552 if (newvd->vdev_isspare && last->vdev_isspare &&
5553 vdev_dtl_empty(last, DTL_MISSING) &&
5554 vdev_dtl_empty(last, DTL_OUTAGE) &&
5555 !vdev_dtl_required(newvd))
5564 spa_vdev_resilver_done(spa_t *spa)
5566 vdev_t *vd, *pvd, *ppvd;
5567 uint64_t guid, sguid, pguid, ppguid;
5569 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5571 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
5572 pvd = vd->vdev_parent;
5573 ppvd = pvd->vdev_parent;
5574 guid = vd->vdev_guid;
5575 pguid = pvd->vdev_guid;
5576 ppguid = ppvd->vdev_guid;
5579 * If we have just finished replacing a hot spared device, then
5580 * we need to detach the parent's first child (the original hot
5583 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
5584 ppvd->vdev_children == 2) {
5585 ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
5586 sguid = ppvd->vdev_child[1]->vdev_guid;
5588 ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd));
5590 spa_config_exit(spa, SCL_ALL, FTAG);
5591 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
5593 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
5595 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5598 spa_config_exit(spa, SCL_ALL, FTAG);
5602 * Update the stored path or FRU for this vdev.
5605 spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
5609 boolean_t sync = B_FALSE;
5611 ASSERT(spa_writeable(spa));
5613 spa_vdev_state_enter(spa, SCL_ALL);
5615 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
5616 return (spa_vdev_state_exit(spa, NULL, ENOENT));
5618 if (!vd->vdev_ops->vdev_op_leaf)
5619 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
5622 if (strcmp(value, vd->vdev_path) != 0) {
5623 spa_strfree(vd->vdev_path);
5624 vd->vdev_path = spa_strdup(value);
5628 if (vd->vdev_fru == NULL) {
5629 vd->vdev_fru = spa_strdup(value);
5631 } else if (strcmp(value, vd->vdev_fru) != 0) {
5632 spa_strfree(vd->vdev_fru);
5633 vd->vdev_fru = spa_strdup(value);
5638 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
5642 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
5644 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5648 spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5650 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5654 * ==========================================================================
5656 * ==========================================================================
5660 spa_scan_stop(spa_t *spa)
5662 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5663 if (dsl_scan_resilvering(spa->spa_dsl_pool))
5664 return (SET_ERROR(EBUSY));
5665 return (dsl_scan_cancel(spa->spa_dsl_pool));
5669 spa_scan(spa_t *spa, pool_scan_func_t func)
5671 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5673 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5674 return (SET_ERROR(ENOTSUP));
5677 * If a resilver was requested, but there is no DTL on a
5678 * writeable leaf device, we have nothing to do.
5680 if (func == POOL_SCAN_RESILVER &&
5681 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5682 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5686 return (dsl_scan(spa->spa_dsl_pool, func));
5690 * ==========================================================================
5691 * SPA async task processing
5692 * ==========================================================================
5696 spa_async_remove(spa_t *spa, vdev_t *vd)
5698 if (vd->vdev_remove_wanted) {
5699 vd->vdev_remove_wanted = B_FALSE;
5700 vd->vdev_delayed_close = B_FALSE;
5701 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5704 * We want to clear the stats, but we don't want to do a full
5705 * vdev_clear() as that will cause us to throw away
5706 * degraded/faulted state as well as attempt to reopen the
5707 * device, all of which is a waste.
5709 vd->vdev_stat.vs_read_errors = 0;
5710 vd->vdev_stat.vs_write_errors = 0;
5711 vd->vdev_stat.vs_checksum_errors = 0;
5713 vdev_state_dirty(vd->vdev_top);
5716 for (int c = 0; c < vd->vdev_children; c++)
5717 spa_async_remove(spa, vd->vdev_child[c]);
5721 spa_async_probe(spa_t *spa, vdev_t *vd)
5723 if (vd->vdev_probe_wanted) {
5724 vd->vdev_probe_wanted = B_FALSE;
5725 vdev_reopen(vd); /* vdev_open() does the actual probe */
5728 for (int c = 0; c < vd->vdev_children; c++)
5729 spa_async_probe(spa, vd->vdev_child[c]);
5733 spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5739 if (!spa->spa_autoexpand)
5742 for (int c = 0; c < vd->vdev_children; c++) {
5743 vdev_t *cvd = vd->vdev_child[c];
5744 spa_async_autoexpand(spa, cvd);
5747 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5750 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
5751 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
5753 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5754 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
5756 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
5757 ESC_ZFS_VDEV_AUTOEXPAND, attr, &eid, DDI_SLEEP);
5760 kmem_free(physpath, MAXPATHLEN);
5764 spa_async_thread(void *arg)
5769 ASSERT(spa->spa_sync_on);
5771 mutex_enter(&spa->spa_async_lock);
5772 tasks = spa->spa_async_tasks;
5773 spa->spa_async_tasks &= SPA_ASYNC_REMOVE;
5774 mutex_exit(&spa->spa_async_lock);
5777 * See if the config needs to be updated.
5779 if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5780 uint64_t old_space, new_space;
5782 mutex_enter(&spa_namespace_lock);
5783 old_space = metaslab_class_get_space(spa_normal_class(spa));
5784 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5785 new_space = metaslab_class_get_space(spa_normal_class(spa));
5786 mutex_exit(&spa_namespace_lock);
5789 * If the pool grew as a result of the config update,
5790 * then log an internal history event.
5792 if (new_space != old_space) {
5793 spa_history_log_internal(spa, "vdev online", NULL,
5794 "pool '%s' size: %llu(+%llu)",
5795 spa_name(spa), new_space, new_space - old_space);
5799 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5800 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5801 spa_async_autoexpand(spa, spa->spa_root_vdev);
5802 spa_config_exit(spa, SCL_CONFIG, FTAG);
5806 * See if any devices need to be probed.
5808 if (tasks & SPA_ASYNC_PROBE) {
5809 spa_vdev_state_enter(spa, SCL_NONE);
5810 spa_async_probe(spa, spa->spa_root_vdev);
5811 (void) spa_vdev_state_exit(spa, NULL, 0);
5815 * If any devices are done replacing, detach them.
5817 if (tasks & SPA_ASYNC_RESILVER_DONE)
5818 spa_vdev_resilver_done(spa);
5821 * Kick off a resilver.
5823 if (tasks & SPA_ASYNC_RESILVER)
5824 dsl_resilver_restart(spa->spa_dsl_pool, 0);
5827 * Let the world know that we're done.
5829 mutex_enter(&spa->spa_async_lock);
5830 spa->spa_async_thread = NULL;
5831 cv_broadcast(&spa->spa_async_cv);
5832 mutex_exit(&spa->spa_async_lock);
5837 spa_async_thread_vd(void *arg)
5842 ASSERT(spa->spa_sync_on);
5844 mutex_enter(&spa->spa_async_lock);
5845 tasks = spa->spa_async_tasks;
5847 spa->spa_async_tasks &= ~SPA_ASYNC_REMOVE;
5848 mutex_exit(&spa->spa_async_lock);
5851 * See if any devices need to be marked REMOVED.
5853 if (tasks & SPA_ASYNC_REMOVE) {
5854 spa_vdev_state_enter(spa, SCL_NONE);
5855 spa_async_remove(spa, spa->spa_root_vdev);
5856 for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
5857 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5858 for (int i = 0; i < spa->spa_spares.sav_count; i++)
5859 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5860 (void) spa_vdev_state_exit(spa, NULL, 0);
5864 * Let the world know that we're done.
5866 mutex_enter(&spa->spa_async_lock);
5867 tasks = spa->spa_async_tasks;
5868 if ((tasks & SPA_ASYNC_REMOVE) != 0)
5870 spa->spa_async_thread_vd = NULL;
5871 cv_broadcast(&spa->spa_async_cv);
5872 mutex_exit(&spa->spa_async_lock);
5877 spa_async_suspend(spa_t *spa)
5879 mutex_enter(&spa->spa_async_lock);
5880 spa->spa_async_suspended++;
5881 while (spa->spa_async_thread != NULL &&
5882 spa->spa_async_thread_vd != NULL)
5883 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5884 mutex_exit(&spa->spa_async_lock);
5888 spa_async_resume(spa_t *spa)
5890 mutex_enter(&spa->spa_async_lock);
5891 ASSERT(spa->spa_async_suspended != 0);
5892 spa->spa_async_suspended--;
5893 mutex_exit(&spa->spa_async_lock);
5897 spa_async_tasks_pending(spa_t *spa)
5899 uint_t non_config_tasks;
5901 boolean_t config_task_suspended;
5903 non_config_tasks = spa->spa_async_tasks & ~(SPA_ASYNC_CONFIG_UPDATE |
5905 config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE;
5906 if (spa->spa_ccw_fail_time == 0) {
5907 config_task_suspended = B_FALSE;
5909 config_task_suspended =
5910 (gethrtime() - spa->spa_ccw_fail_time) <
5911 (zfs_ccw_retry_interval * NANOSEC);
5914 return (non_config_tasks || (config_task && !config_task_suspended));
5918 spa_async_dispatch(spa_t *spa)
5920 mutex_enter(&spa->spa_async_lock);
5921 if (spa_async_tasks_pending(spa) &&
5922 !spa->spa_async_suspended &&
5923 spa->spa_async_thread == NULL &&
5925 spa->spa_async_thread = thread_create(NULL, 0,
5926 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
5927 mutex_exit(&spa->spa_async_lock);
5931 spa_async_dispatch_vd(spa_t *spa)
5933 mutex_enter(&spa->spa_async_lock);
5934 if ((spa->spa_async_tasks & SPA_ASYNC_REMOVE) != 0 &&
5935 !spa->spa_async_suspended &&
5936 spa->spa_async_thread_vd == NULL &&
5938 spa->spa_async_thread_vd = thread_create(NULL, 0,
5939 spa_async_thread_vd, spa, 0, &p0, TS_RUN, maxclsyspri);
5940 mutex_exit(&spa->spa_async_lock);
5944 spa_async_request(spa_t *spa, int task)
5946 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
5947 mutex_enter(&spa->spa_async_lock);
5948 spa->spa_async_tasks |= task;
5949 mutex_exit(&spa->spa_async_lock);
5950 spa_async_dispatch_vd(spa);
5954 * ==========================================================================
5955 * SPA syncing routines
5956 * ==========================================================================
5960 bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5963 bpobj_enqueue(bpo, bp, tx);
5968 spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5972 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
5973 BP_GET_PSIZE(bp), zio->io_flags));
5978 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
5980 char *packed = NULL;
5985 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
5988 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5989 * information. This avoids the dbuf_will_dirty() path and
5990 * saves us a pre-read to get data we don't actually care about.
5992 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
5993 packed = kmem_alloc(bufsize, KM_SLEEP);
5995 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
5997 bzero(packed + nvsize, bufsize - nvsize);
5999 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
6001 kmem_free(packed, bufsize);
6003 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
6004 dmu_buf_will_dirty(db, tx);
6005 *(uint64_t *)db->db_data = nvsize;
6006 dmu_buf_rele(db, FTAG);
6010 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
6011 const char *config, const char *entry)
6021 * Update the MOS nvlist describing the list of available devices.
6022 * spa_validate_aux() will have already made sure this nvlist is
6023 * valid and the vdevs are labeled appropriately.
6025 if (sav->sav_object == 0) {
6026 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
6027 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
6028 sizeof (uint64_t), tx);
6029 VERIFY(zap_update(spa->spa_meta_objset,
6030 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
6031 &sav->sav_object, tx) == 0);
6034 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
6035 if (sav->sav_count == 0) {
6036 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
6038 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
6039 for (i = 0; i < sav->sav_count; i++)
6040 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
6041 B_FALSE, VDEV_CONFIG_L2CACHE);
6042 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
6043 sav->sav_count) == 0);
6044 for (i = 0; i < sav->sav_count; i++)
6045 nvlist_free(list[i]);
6046 kmem_free(list, sav->sav_count * sizeof (void *));
6049 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
6050 nvlist_free(nvroot);
6052 sav->sav_sync = B_FALSE;
6056 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
6060 if (list_is_empty(&spa->spa_config_dirty_list))
6063 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6065 config = spa_config_generate(spa, spa->spa_root_vdev,
6066 dmu_tx_get_txg(tx), B_FALSE);
6069 * If we're upgrading the spa version then make sure that
6070 * the config object gets updated with the correct version.
6072 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
6073 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
6074 spa->spa_uberblock.ub_version);
6076 spa_config_exit(spa, SCL_STATE, FTAG);
6078 if (spa->spa_config_syncing)
6079 nvlist_free(spa->spa_config_syncing);
6080 spa->spa_config_syncing = config;
6082 spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
6086 spa_sync_version(void *arg, dmu_tx_t *tx)
6088 uint64_t *versionp = arg;
6089 uint64_t version = *versionp;
6090 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
6093 * Setting the version is special cased when first creating the pool.
6095 ASSERT(tx->tx_txg != TXG_INITIAL);
6097 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
6098 ASSERT(version >= spa_version(spa));
6100 spa->spa_uberblock.ub_version = version;
6101 vdev_config_dirty(spa->spa_root_vdev);
6102 spa_history_log_internal(spa, "set", tx, "version=%lld", version);
6106 * Set zpool properties.
6109 spa_sync_props(void *arg, dmu_tx_t *tx)
6111 nvlist_t *nvp = arg;
6112 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
6113 objset_t *mos = spa->spa_meta_objset;
6114 nvpair_t *elem = NULL;
6116 mutex_enter(&spa->spa_props_lock);
6118 while ((elem = nvlist_next_nvpair(nvp, elem))) {
6120 char *strval, *fname;
6122 const char *propname;
6123 zprop_type_t proptype;
6124 zfeature_info_t *feature;
6126 switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
6129 * We checked this earlier in spa_prop_validate().
6131 ASSERT(zpool_prop_feature(nvpair_name(elem)));
6133 fname = strchr(nvpair_name(elem), '@') + 1;
6134 VERIFY3U(0, ==, zfeature_lookup_name(fname, &feature));
6136 spa_feature_enable(spa, feature, tx);
6137 spa_history_log_internal(spa, "set", tx,
6138 "%s=enabled", nvpair_name(elem));
6141 case ZPOOL_PROP_VERSION:
6142 VERIFY(nvpair_value_uint64(elem, &intval) == 0);
6144 * The version is synced seperatly before other
6145 * properties and should be correct by now.
6147 ASSERT3U(spa_version(spa), >=, intval);
6150 case ZPOOL_PROP_ALTROOT:
6152 * 'altroot' is a non-persistent property. It should
6153 * have been set temporarily at creation or import time.
6155 ASSERT(spa->spa_root != NULL);
6158 case ZPOOL_PROP_READONLY:
6159 case ZPOOL_PROP_CACHEFILE:
6161 * 'readonly' and 'cachefile' are also non-persisitent
6165 case ZPOOL_PROP_COMMENT:
6166 VERIFY(nvpair_value_string(elem, &strval) == 0);
6167 if (spa->spa_comment != NULL)
6168 spa_strfree(spa->spa_comment);
6169 spa->spa_comment = spa_strdup(strval);
6171 * We need to dirty the configuration on all the vdevs
6172 * so that their labels get updated. It's unnecessary
6173 * to do this for pool creation since the vdev's
6174 * configuratoin has already been dirtied.
6176 if (tx->tx_txg != TXG_INITIAL)
6177 vdev_config_dirty(spa->spa_root_vdev);
6178 spa_history_log_internal(spa, "set", tx,
6179 "%s=%s", nvpair_name(elem), strval);
6183 * Set pool property values in the poolprops mos object.
6185 if (spa->spa_pool_props_object == 0) {
6186 spa->spa_pool_props_object =
6187 zap_create_link(mos, DMU_OT_POOL_PROPS,
6188 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
6192 /* normalize the property name */
6193 propname = zpool_prop_to_name(prop);
6194 proptype = zpool_prop_get_type(prop);
6196 if (nvpair_type(elem) == DATA_TYPE_STRING) {
6197 ASSERT(proptype == PROP_TYPE_STRING);
6198 VERIFY(nvpair_value_string(elem, &strval) == 0);
6199 VERIFY(zap_update(mos,
6200 spa->spa_pool_props_object, propname,
6201 1, strlen(strval) + 1, strval, tx) == 0);
6202 spa_history_log_internal(spa, "set", tx,
6203 "%s=%s", nvpair_name(elem), strval);
6204 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
6205 VERIFY(nvpair_value_uint64(elem, &intval) == 0);
6207 if (proptype == PROP_TYPE_INDEX) {
6209 VERIFY(zpool_prop_index_to_string(
6210 prop, intval, &unused) == 0);
6212 VERIFY(zap_update(mos,
6213 spa->spa_pool_props_object, propname,
6214 8, 1, &intval, tx) == 0);
6215 spa_history_log_internal(spa, "set", tx,
6216 "%s=%lld", nvpair_name(elem), intval);
6218 ASSERT(0); /* not allowed */
6222 case ZPOOL_PROP_DELEGATION:
6223 spa->spa_delegation = intval;
6225 case ZPOOL_PROP_BOOTFS:
6226 spa->spa_bootfs = intval;
6228 case ZPOOL_PROP_FAILUREMODE:
6229 spa->spa_failmode = intval;
6231 case ZPOOL_PROP_AUTOEXPAND:
6232 spa->spa_autoexpand = intval;
6233 if (tx->tx_txg != TXG_INITIAL)
6234 spa_async_request(spa,
6235 SPA_ASYNC_AUTOEXPAND);
6237 case ZPOOL_PROP_DEDUPDITTO:
6238 spa->spa_dedup_ditto = intval;
6247 mutex_exit(&spa->spa_props_lock);
6251 * Perform one-time upgrade on-disk changes. spa_version() does not
6252 * reflect the new version this txg, so there must be no changes this
6253 * txg to anything that the upgrade code depends on after it executes.
6254 * Therefore this must be called after dsl_pool_sync() does the sync
6258 spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
6260 dsl_pool_t *dp = spa->spa_dsl_pool;
6262 ASSERT(spa->spa_sync_pass == 1);
6264 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
6266 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
6267 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
6268 dsl_pool_create_origin(dp, tx);
6270 /* Keeping the origin open increases spa_minref */
6271 spa->spa_minref += 3;
6274 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
6275 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
6276 dsl_pool_upgrade_clones(dp, tx);
6279 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
6280 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
6281 dsl_pool_upgrade_dir_clones(dp, tx);
6283 /* Keeping the freedir open increases spa_minref */
6284 spa->spa_minref += 3;
6287 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES &&
6288 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
6289 spa_feature_create_zap_objects(spa, tx);
6291 rrw_exit(&dp->dp_config_rwlock, FTAG);
6295 * Sync the specified transaction group. New blocks may be dirtied as
6296 * part of the process, so we iterate until it converges.
6299 spa_sync(spa_t *spa, uint64_t txg)
6301 dsl_pool_t *dp = spa->spa_dsl_pool;
6302 objset_t *mos = spa->spa_meta_objset;
6303 bpobj_t *defer_bpo = &spa->spa_deferred_bpobj;
6304 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
6305 vdev_t *rvd = spa->spa_root_vdev;
6310 VERIFY(spa_writeable(spa));
6313 * Lock out configuration changes.
6315 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
6317 spa->spa_syncing_txg = txg;
6318 spa->spa_sync_pass = 0;
6321 * If there are any pending vdev state changes, convert them
6322 * into config changes that go out with this transaction group.
6324 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6325 while (list_head(&spa->spa_state_dirty_list) != NULL) {
6327 * We need the write lock here because, for aux vdevs,
6328 * calling vdev_config_dirty() modifies sav_config.
6329 * This is ugly and will become unnecessary when we
6330 * eliminate the aux vdev wart by integrating all vdevs
6331 * into the root vdev tree.
6333 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6334 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
6335 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
6336 vdev_state_clean(vd);
6337 vdev_config_dirty(vd);
6339 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6340 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
6342 spa_config_exit(spa, SCL_STATE, FTAG);
6344 tx = dmu_tx_create_assigned(dp, txg);
6346 spa->spa_sync_starttime = gethrtime();
6348 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid,
6349 spa->spa_sync_starttime + spa->spa_deadman_synctime));
6352 callout_reset(&spa->spa_deadman_cycid,
6353 hz * spa->spa_deadman_synctime / NANOSEC, spa_deadman, spa);
6358 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
6359 * set spa_deflate if we have no raid-z vdevs.
6361 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
6362 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
6365 for (i = 0; i < rvd->vdev_children; i++) {
6366 vd = rvd->vdev_child[i];
6367 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
6370 if (i == rvd->vdev_children) {
6371 spa->spa_deflate = TRUE;
6372 VERIFY(0 == zap_add(spa->spa_meta_objset,
6373 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
6374 sizeof (uint64_t), 1, &spa->spa_deflate, tx));
6379 * If anything has changed in this txg, or if someone is waiting
6380 * for this txg to sync (eg, spa_vdev_remove()), push the
6381 * deferred frees from the previous txg. If not, leave them
6382 * alone so that we don't generate work on an otherwise idle
6385 if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
6386 !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
6387 !txg_list_empty(&dp->dp_sync_tasks, txg) ||
6388 ((dsl_scan_active(dp->dp_scan) ||
6389 txg_sync_waiting(dp)) && !spa_shutting_down(spa))) {
6390 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6391 VERIFY3U(bpobj_iterate(defer_bpo,
6392 spa_free_sync_cb, zio, tx), ==, 0);
6393 VERIFY0(zio_wait(zio));
6397 * Iterate to convergence.
6400 int pass = ++spa->spa_sync_pass;
6402 spa_sync_config_object(spa, tx);
6403 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
6404 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
6405 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
6406 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
6407 spa_errlog_sync(spa, txg);
6408 dsl_pool_sync(dp, txg);
6410 if (pass < zfs_sync_pass_deferred_free) {
6411 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6412 bplist_iterate(free_bpl, spa_free_sync_cb,
6414 VERIFY(zio_wait(zio) == 0);
6416 bplist_iterate(free_bpl, bpobj_enqueue_cb,
6421 dsl_scan_sync(dp, tx);
6423 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
6427 spa_sync_upgrades(spa, tx);
6429 } while (dmu_objset_is_dirty(mos, txg));
6432 * Rewrite the vdev configuration (which includes the uberblock)
6433 * to commit the transaction group.
6435 * If there are no dirty vdevs, we sync the uberblock to a few
6436 * random top-level vdevs that are known to be visible in the
6437 * config cache (see spa_vdev_add() for a complete description).
6438 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
6442 * We hold SCL_STATE to prevent vdev open/close/etc.
6443 * while we're attempting to write the vdev labels.
6445 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6447 if (list_is_empty(&spa->spa_config_dirty_list)) {
6448 vdev_t *svd[SPA_DVAS_PER_BP];
6450 int children = rvd->vdev_children;
6451 int c0 = spa_get_random(children);
6453 for (int c = 0; c < children; c++) {
6454 vd = rvd->vdev_child[(c0 + c) % children];
6455 if (vd->vdev_ms_array == 0 || vd->vdev_islog)
6457 svd[svdcount++] = vd;
6458 if (svdcount == SPA_DVAS_PER_BP)
6461 error = vdev_config_sync(svd, svdcount, txg, B_FALSE);
6463 error = vdev_config_sync(svd, svdcount, txg,
6466 error = vdev_config_sync(rvd->vdev_child,
6467 rvd->vdev_children, txg, B_FALSE);
6469 error = vdev_config_sync(rvd->vdev_child,
6470 rvd->vdev_children, txg, B_TRUE);
6474 spa->spa_last_synced_guid = rvd->vdev_guid;
6476 spa_config_exit(spa, SCL_STATE, FTAG);
6480 zio_suspend(spa, NULL);
6481 zio_resume_wait(spa);
6486 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY));
6489 callout_drain(&spa->spa_deadman_cycid);
6494 * Clear the dirty config list.
6496 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
6497 vdev_config_clean(vd);
6500 * Now that the new config has synced transactionally,
6501 * let it become visible to the config cache.
6503 if (spa->spa_config_syncing != NULL) {
6504 spa_config_set(spa, spa->spa_config_syncing);
6505 spa->spa_config_txg = txg;
6506 spa->spa_config_syncing = NULL;
6509 spa->spa_ubsync = spa->spa_uberblock;
6511 dsl_pool_sync_done(dp, txg);
6514 * Update usable space statistics.
6516 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
6517 vdev_sync_done(vd, txg);
6519 spa_update_dspace(spa);
6522 * It had better be the case that we didn't dirty anything
6523 * since vdev_config_sync().
6525 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
6526 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
6527 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
6529 spa->spa_sync_pass = 0;
6531 spa_config_exit(spa, SCL_CONFIG, FTAG);
6533 spa_handle_ignored_writes(spa);
6536 * If any async tasks have been requested, kick them off.
6538 spa_async_dispatch(spa);
6539 spa_async_dispatch_vd(spa);
6543 * Sync all pools. We don't want to hold the namespace lock across these
6544 * operations, so we take a reference on the spa_t and drop the lock during the
6548 spa_sync_allpools(void)
6551 mutex_enter(&spa_namespace_lock);
6552 while ((spa = spa_next(spa)) != NULL) {
6553 if (spa_state(spa) != POOL_STATE_ACTIVE ||
6554 !spa_writeable(spa) || spa_suspended(spa))
6556 spa_open_ref(spa, FTAG);
6557 mutex_exit(&spa_namespace_lock);
6558 txg_wait_synced(spa_get_dsl(spa), 0);
6559 mutex_enter(&spa_namespace_lock);
6560 spa_close(spa, FTAG);
6562 mutex_exit(&spa_namespace_lock);
6566 * ==========================================================================
6567 * Miscellaneous routines
6568 * ==========================================================================
6572 * Remove all pools in the system.
6580 * Remove all cached state. All pools should be closed now,
6581 * so every spa in the AVL tree should be unreferenced.
6583 mutex_enter(&spa_namespace_lock);
6584 while ((spa = spa_next(NULL)) != NULL) {
6586 * Stop async tasks. The async thread may need to detach
6587 * a device that's been replaced, which requires grabbing
6588 * spa_namespace_lock, so we must drop it here.
6590 spa_open_ref(spa, FTAG);
6591 mutex_exit(&spa_namespace_lock);
6592 spa_async_suspend(spa);
6593 mutex_enter(&spa_namespace_lock);
6594 spa_close(spa, FTAG);
6596 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
6598 spa_deactivate(spa);
6602 mutex_exit(&spa_namespace_lock);
6606 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
6611 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
6615 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
6616 vd = spa->spa_l2cache.sav_vdevs[i];
6617 if (vd->vdev_guid == guid)
6621 for (i = 0; i < spa->spa_spares.sav_count; i++) {
6622 vd = spa->spa_spares.sav_vdevs[i];
6623 if (vd->vdev_guid == guid)
6632 spa_upgrade(spa_t *spa, uint64_t version)
6634 ASSERT(spa_writeable(spa));
6636 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6639 * This should only be called for a non-faulted pool, and since a
6640 * future version would result in an unopenable pool, this shouldn't be
6643 ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version));
6644 ASSERT(version >= spa->spa_uberblock.ub_version);
6646 spa->spa_uberblock.ub_version = version;
6647 vdev_config_dirty(spa->spa_root_vdev);
6649 spa_config_exit(spa, SCL_ALL, FTAG);
6651 txg_wait_synced(spa_get_dsl(spa), 0);
6655 spa_has_spare(spa_t *spa, uint64_t guid)
6659 spa_aux_vdev_t *sav = &spa->spa_spares;
6661 for (i = 0; i < sav->sav_count; i++)
6662 if (sav->sav_vdevs[i]->vdev_guid == guid)
6665 for (i = 0; i < sav->sav_npending; i++) {
6666 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
6667 &spareguid) == 0 && spareguid == guid)
6675 * Check if a pool has an active shared spare device.
6676 * Note: reference count of an active spare is 2, as a spare and as a replace
6679 spa_has_active_shared_spare(spa_t *spa)
6683 spa_aux_vdev_t *sav = &spa->spa_spares;
6685 for (i = 0; i < sav->sav_count; i++) {
6686 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
6687 &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
6696 * Post a sysevent corresponding to the given event. The 'name' must be one of
6697 * the event definitions in sys/sysevent/eventdefs.h. The payload will be
6698 * filled in from the spa and (optionally) the vdev. This doesn't do anything
6699 * in the userland libzpool, as we don't want consumers to misinterpret ztest
6700 * or zdb as real changes.
6703 spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
6707 sysevent_attr_list_t *attr = NULL;
6708 sysevent_value_t value;
6711 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
6714 value.value_type = SE_DATA_TYPE_STRING;
6715 value.value.sv_string = spa_name(spa);
6716 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
6719 value.value_type = SE_DATA_TYPE_UINT64;
6720 value.value.sv_uint64 = spa_guid(spa);
6721 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
6725 value.value_type = SE_DATA_TYPE_UINT64;
6726 value.value.sv_uint64 = vd->vdev_guid;
6727 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
6731 if (vd->vdev_path) {
6732 value.value_type = SE_DATA_TYPE_STRING;
6733 value.value.sv_string = vd->vdev_path;
6734 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
6735 &value, SE_SLEEP) != 0)
6740 if (sysevent_attach_attributes(ev, attr) != 0)
6744 (void) log_sysevent(ev, SE_SLEEP, &eid);
6748 sysevent_free_attr(attr);