4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2011, 2017 by Delphix. All rights reserved.
25 * Copyright (c) 2015, Nexenta Systems, Inc. All rights reserved.
26 * Copyright (c) 2013 Martin Matuska <mm@FreeBSD.org>. All rights reserved.
27 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
28 * Copyright 2013 Saso Kiselkov. All rights reserved.
29 * Copyright (c) 2014 Integros [integros.com]
30 * Copyright 2016 Toomas Soome <tsoome@me.com>
31 * Copyright (c) 2017 Datto Inc.
35 * SPA: Storage Pool Allocator
37 * This file contains all the routines used when modifying on-disk SPA state.
38 * This includes opening, importing, destroying, exporting a pool, and syncing a
42 #include <sys/zfs_context.h>
43 #include <sys/fm/fs/zfs.h>
44 #include <sys/spa_impl.h>
46 #include <sys/zio_checksum.h>
48 #include <sys/dmu_tx.h>
52 #include <sys/vdev_impl.h>
53 #include <sys/metaslab.h>
54 #include <sys/metaslab_impl.h>
55 #include <sys/uberblock_impl.h>
58 #include <sys/dmu_traverse.h>
59 #include <sys/dmu_objset.h>
60 #include <sys/unique.h>
61 #include <sys/dsl_pool.h>
62 #include <sys/dsl_dataset.h>
63 #include <sys/dsl_dir.h>
64 #include <sys/dsl_prop.h>
65 #include <sys/dsl_synctask.h>
66 #include <sys/fs/zfs.h>
68 #include <sys/callb.h>
69 #include <sys/spa_boot.h>
70 #include <sys/zfs_ioctl.h>
71 #include <sys/dsl_scan.h>
72 #include <sys/dmu_send.h>
73 #include <sys/dsl_destroy.h>
74 #include <sys/dsl_userhold.h>
75 #include <sys/zfeature.h>
77 #include <sys/trim_map.h>
81 #include <sys/callb.h>
82 #include <sys/cpupart.h>
87 #include "zfs_comutil.h"
89 /* Check hostid on import? */
90 static int check_hostid = 1;
93 * The interval, in seconds, at which failed configuration cache file writes
96 static int zfs_ccw_retry_interval = 300;
98 SYSCTL_DECL(_vfs_zfs);
99 SYSCTL_INT(_vfs_zfs, OID_AUTO, check_hostid, CTLFLAG_RWTUN, &check_hostid, 0,
100 "Check hostid on import?");
101 TUNABLE_INT("vfs.zfs.ccw_retry_interval", &zfs_ccw_retry_interval);
102 SYSCTL_INT(_vfs_zfs, OID_AUTO, ccw_retry_interval, CTLFLAG_RW,
103 &zfs_ccw_retry_interval, 0,
104 "Configuration cache file write, retry after failure, interval (seconds)");
106 typedef enum zti_modes {
107 ZTI_MODE_FIXED, /* value is # of threads (min 1) */
108 ZTI_MODE_BATCH, /* cpu-intensive; value is ignored */
109 ZTI_MODE_NULL, /* don't create a taskq */
113 #define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) }
114 #define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 }
115 #define ZTI_NULL { ZTI_MODE_NULL, 0, 0 }
117 #define ZTI_N(n) ZTI_P(n, 1)
118 #define ZTI_ONE ZTI_N(1)
120 typedef struct zio_taskq_info {
121 zti_modes_t zti_mode;
126 static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
127 "issue", "issue_high", "intr", "intr_high"
131 * This table defines the taskq settings for each ZFS I/O type. When
132 * initializing a pool, we use this table to create an appropriately sized
133 * taskq. Some operations are low volume and therefore have a small, static
134 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE
135 * macros. Other operations process a large amount of data; the ZTI_BATCH
136 * macro causes us to create a taskq oriented for throughput. Some operations
137 * are so high frequency and short-lived that the taskq itself can become a a
138 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an
139 * additional degree of parallelism specified by the number of threads per-
140 * taskq and the number of taskqs; when dispatching an event in this case, the
141 * particular taskq is chosen at random.
143 * The different taskq priorities are to handle the different contexts (issue
144 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that
145 * need to be handled with minimum delay.
147 const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
148 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
149 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* NULL */
150 { ZTI_N(8), ZTI_NULL, ZTI_P(12, 8), ZTI_NULL }, /* READ */
151 { ZTI_BATCH, ZTI_N(5), ZTI_N(8), ZTI_N(5) }, /* WRITE */
152 { ZTI_P(12, 8), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FREE */
153 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* CLAIM */
154 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* IOCTL */
157 static sysevent_t *spa_event_create(spa_t *spa, vdev_t *vd, const char *name);
158 static void spa_event_post(sysevent_t *ev);
159 static void spa_sync_version(void *arg, dmu_tx_t *tx);
160 static void spa_sync_props(void *arg, dmu_tx_t *tx);
161 static boolean_t spa_has_active_shared_spare(spa_t *spa);
162 static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
163 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
165 static void spa_vdev_resilver_done(spa_t *spa);
167 uint_t zio_taskq_batch_pct = 75; /* 1 thread per cpu in pset */
169 id_t zio_taskq_psrset_bind = PS_NONE;
172 boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */
173 uint_t zio_taskq_basedc = 80; /* base duty cycle */
176 boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */
177 extern int zfs_sync_pass_deferred_free;
180 * This (illegal) pool name is used when temporarily importing a spa_t in order
181 * to get the vdev stats associated with the imported devices.
183 #define TRYIMPORT_NAME "$import"
186 * ==========================================================================
187 * SPA properties routines
188 * ==========================================================================
192 * Add a (source=src, propname=propval) list to an nvlist.
195 spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
196 uint64_t intval, zprop_source_t src)
198 const char *propname = zpool_prop_to_name(prop);
201 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
202 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
205 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
207 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
209 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
210 nvlist_free(propval);
214 * Get property values from the spa configuration.
217 spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
219 vdev_t *rvd = spa->spa_root_vdev;
220 dsl_pool_t *pool = spa->spa_dsl_pool;
221 uint64_t size, alloc, cap, version;
222 zprop_source_t src = ZPROP_SRC_NONE;
223 spa_config_dirent_t *dp;
224 metaslab_class_t *mc = spa_normal_class(spa);
226 ASSERT(MUTEX_HELD(&spa->spa_props_lock));
229 alloc = metaslab_class_get_alloc(spa_normal_class(spa));
230 size = metaslab_class_get_space(spa_normal_class(spa));
231 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
232 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
233 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
234 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
237 spa_prop_add_list(*nvp, ZPOOL_PROP_FRAGMENTATION, NULL,
238 metaslab_class_fragmentation(mc), src);
239 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL,
240 metaslab_class_expandable_space(mc), src);
241 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
242 (spa_mode(spa) == FREAD), src);
244 cap = (size == 0) ? 0 : (alloc * 100 / size);
245 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
247 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
248 ddt_get_pool_dedup_ratio(spa), src);
250 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
251 rvd->vdev_state, src);
253 version = spa_version(spa);
254 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
255 src = ZPROP_SRC_DEFAULT;
257 src = ZPROP_SRC_LOCAL;
258 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
263 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
264 * when opening pools before this version freedir will be NULL.
266 if (pool->dp_free_dir != NULL) {
267 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
268 dsl_dir_phys(pool->dp_free_dir)->dd_used_bytes,
271 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
275 if (pool->dp_leak_dir != NULL) {
276 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, NULL,
277 dsl_dir_phys(pool->dp_leak_dir)->dd_used_bytes,
280 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED,
285 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
287 if (spa->spa_comment != NULL) {
288 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
292 if (spa->spa_root != NULL)
293 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
296 if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) {
297 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL,
298 MIN(zfs_max_recordsize, SPA_MAXBLOCKSIZE), ZPROP_SRC_NONE);
300 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL,
301 SPA_OLD_MAXBLOCKSIZE, ZPROP_SRC_NONE);
304 if ((dp = list_head(&spa->spa_config_list)) != NULL) {
305 if (dp->scd_path == NULL) {
306 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
307 "none", 0, ZPROP_SRC_LOCAL);
308 } else if (strcmp(dp->scd_path, spa_config_path) != 0) {
309 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
310 dp->scd_path, 0, ZPROP_SRC_LOCAL);
316 * Get zpool property values.
319 spa_prop_get(spa_t *spa, nvlist_t **nvp)
321 objset_t *mos = spa->spa_meta_objset;
326 VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
328 mutex_enter(&spa->spa_props_lock);
331 * Get properties from the spa config.
333 spa_prop_get_config(spa, nvp);
335 /* If no pool property object, no more prop to get. */
336 if (mos == NULL || spa->spa_pool_props_object == 0) {
337 mutex_exit(&spa->spa_props_lock);
342 * Get properties from the MOS pool property object.
344 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
345 (err = zap_cursor_retrieve(&zc, &za)) == 0;
346 zap_cursor_advance(&zc)) {
349 zprop_source_t src = ZPROP_SRC_DEFAULT;
352 if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
355 switch (za.za_integer_length) {
357 /* integer property */
358 if (za.za_first_integer !=
359 zpool_prop_default_numeric(prop))
360 src = ZPROP_SRC_LOCAL;
362 if (prop == ZPOOL_PROP_BOOTFS) {
364 dsl_dataset_t *ds = NULL;
366 dp = spa_get_dsl(spa);
367 dsl_pool_config_enter(dp, FTAG);
368 if (err = dsl_dataset_hold_obj(dp,
369 za.za_first_integer, FTAG, &ds)) {
370 dsl_pool_config_exit(dp, FTAG);
374 strval = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN,
376 dsl_dataset_name(ds, strval);
377 dsl_dataset_rele(ds, FTAG);
378 dsl_pool_config_exit(dp, FTAG);
381 intval = za.za_first_integer;
384 spa_prop_add_list(*nvp, prop, strval, intval, src);
387 kmem_free(strval, ZFS_MAX_DATASET_NAME_LEN);
392 /* string property */
393 strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
394 err = zap_lookup(mos, spa->spa_pool_props_object,
395 za.za_name, 1, za.za_num_integers, strval);
397 kmem_free(strval, za.za_num_integers);
400 spa_prop_add_list(*nvp, prop, strval, 0, src);
401 kmem_free(strval, za.za_num_integers);
408 zap_cursor_fini(&zc);
409 mutex_exit(&spa->spa_props_lock);
411 if (err && err != ENOENT) {
421 * Validate the given pool properties nvlist and modify the list
422 * for the property values to be set.
425 spa_prop_validate(spa_t *spa, nvlist_t *props)
428 int error = 0, reset_bootfs = 0;
430 boolean_t has_feature = B_FALSE;
433 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
435 char *strval, *slash, *check, *fname;
436 const char *propname = nvpair_name(elem);
437 zpool_prop_t prop = zpool_name_to_prop(propname);
441 if (!zpool_prop_feature(propname)) {
442 error = SET_ERROR(EINVAL);
447 * Sanitize the input.
449 if (nvpair_type(elem) != DATA_TYPE_UINT64) {
450 error = SET_ERROR(EINVAL);
454 if (nvpair_value_uint64(elem, &intval) != 0) {
455 error = SET_ERROR(EINVAL);
460 error = SET_ERROR(EINVAL);
464 fname = strchr(propname, '@') + 1;
465 if (zfeature_lookup_name(fname, NULL) != 0) {
466 error = SET_ERROR(EINVAL);
470 has_feature = B_TRUE;
473 case ZPOOL_PROP_VERSION:
474 error = nvpair_value_uint64(elem, &intval);
476 (intval < spa_version(spa) ||
477 intval > SPA_VERSION_BEFORE_FEATURES ||
479 error = SET_ERROR(EINVAL);
482 case ZPOOL_PROP_DELEGATION:
483 case ZPOOL_PROP_AUTOREPLACE:
484 case ZPOOL_PROP_LISTSNAPS:
485 case ZPOOL_PROP_AUTOEXPAND:
486 error = nvpair_value_uint64(elem, &intval);
487 if (!error && intval > 1)
488 error = SET_ERROR(EINVAL);
491 case ZPOOL_PROP_BOOTFS:
493 * If the pool version is less than SPA_VERSION_BOOTFS,
494 * or the pool is still being created (version == 0),
495 * the bootfs property cannot be set.
497 if (spa_version(spa) < SPA_VERSION_BOOTFS) {
498 error = SET_ERROR(ENOTSUP);
503 * Make sure the vdev config is bootable
505 if (!vdev_is_bootable(spa->spa_root_vdev)) {
506 error = SET_ERROR(ENOTSUP);
512 error = nvpair_value_string(elem, &strval);
518 if (strval == NULL || strval[0] == '\0') {
519 objnum = zpool_prop_default_numeric(
524 if (error = dmu_objset_hold(strval, FTAG, &os))
528 * Must be ZPL, and its property settings
529 * must be supported by GRUB (compression
530 * is not gzip, and large blocks are not used).
533 if (dmu_objset_type(os) != DMU_OST_ZFS) {
534 error = SET_ERROR(ENOTSUP);
536 dsl_prop_get_int_ds(dmu_objset_ds(os),
537 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
539 !BOOTFS_COMPRESS_VALID(propval)) {
540 error = SET_ERROR(ENOTSUP);
542 objnum = dmu_objset_id(os);
544 dmu_objset_rele(os, FTAG);
548 case ZPOOL_PROP_FAILUREMODE:
549 error = nvpair_value_uint64(elem, &intval);
550 if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
551 intval > ZIO_FAILURE_MODE_PANIC))
552 error = SET_ERROR(EINVAL);
555 * This is a special case which only occurs when
556 * the pool has completely failed. This allows
557 * the user to change the in-core failmode property
558 * without syncing it out to disk (I/Os might
559 * currently be blocked). We do this by returning
560 * EIO to the caller (spa_prop_set) to trick it
561 * into thinking we encountered a property validation
564 if (!error && spa_suspended(spa)) {
565 spa->spa_failmode = intval;
566 error = SET_ERROR(EIO);
570 case ZPOOL_PROP_CACHEFILE:
571 if ((error = nvpair_value_string(elem, &strval)) != 0)
574 if (strval[0] == '\0')
577 if (strcmp(strval, "none") == 0)
580 if (strval[0] != '/') {
581 error = SET_ERROR(EINVAL);
585 slash = strrchr(strval, '/');
586 ASSERT(slash != NULL);
588 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
589 strcmp(slash, "/..") == 0)
590 error = SET_ERROR(EINVAL);
593 case ZPOOL_PROP_COMMENT:
594 if ((error = nvpair_value_string(elem, &strval)) != 0)
596 for (check = strval; *check != '\0'; check++) {
598 * The kernel doesn't have an easy isprint()
599 * check. For this kernel check, we merely
600 * check ASCII apart from DEL. Fix this if
601 * there is an easy-to-use kernel isprint().
603 if (*check >= 0x7f) {
604 error = SET_ERROR(EINVAL);
608 if (strlen(strval) > ZPROP_MAX_COMMENT)
612 case ZPOOL_PROP_DEDUPDITTO:
613 if (spa_version(spa) < SPA_VERSION_DEDUP)
614 error = SET_ERROR(ENOTSUP);
616 error = nvpair_value_uint64(elem, &intval);
618 intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
619 error = SET_ERROR(EINVAL);
627 if (!error && reset_bootfs) {
628 error = nvlist_remove(props,
629 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
632 error = nvlist_add_uint64(props,
633 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
641 spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
644 spa_config_dirent_t *dp;
646 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
650 dp = kmem_alloc(sizeof (spa_config_dirent_t),
653 if (cachefile[0] == '\0')
654 dp->scd_path = spa_strdup(spa_config_path);
655 else if (strcmp(cachefile, "none") == 0)
658 dp->scd_path = spa_strdup(cachefile);
660 list_insert_head(&spa->spa_config_list, dp);
662 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
666 spa_prop_set(spa_t *spa, nvlist_t *nvp)
669 nvpair_t *elem = NULL;
670 boolean_t need_sync = B_FALSE;
672 if ((error = spa_prop_validate(spa, nvp)) != 0)
675 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
676 zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem));
678 if (prop == ZPOOL_PROP_CACHEFILE ||
679 prop == ZPOOL_PROP_ALTROOT ||
680 prop == ZPOOL_PROP_READONLY)
683 if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) {
686 if (prop == ZPOOL_PROP_VERSION) {
687 VERIFY(nvpair_value_uint64(elem, &ver) == 0);
689 ASSERT(zpool_prop_feature(nvpair_name(elem)));
690 ver = SPA_VERSION_FEATURES;
694 /* Save time if the version is already set. */
695 if (ver == spa_version(spa))
699 * In addition to the pool directory object, we might
700 * create the pool properties object, the features for
701 * read object, the features for write object, or the
702 * feature descriptions object.
704 error = dsl_sync_task(spa->spa_name, NULL,
705 spa_sync_version, &ver,
706 6, ZFS_SPACE_CHECK_RESERVED);
717 return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props,
718 nvp, 6, ZFS_SPACE_CHECK_RESERVED));
725 * If the bootfs property value is dsobj, clear it.
728 spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
730 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
731 VERIFY(zap_remove(spa->spa_meta_objset,
732 spa->spa_pool_props_object,
733 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
740 spa_change_guid_check(void *arg, dmu_tx_t *tx)
742 uint64_t *newguid = arg;
743 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
744 vdev_t *rvd = spa->spa_root_vdev;
747 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
748 vdev_state = rvd->vdev_state;
749 spa_config_exit(spa, SCL_STATE, FTAG);
751 if (vdev_state != VDEV_STATE_HEALTHY)
752 return (SET_ERROR(ENXIO));
754 ASSERT3U(spa_guid(spa), !=, *newguid);
760 spa_change_guid_sync(void *arg, dmu_tx_t *tx)
762 uint64_t *newguid = arg;
763 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
765 vdev_t *rvd = spa->spa_root_vdev;
767 oldguid = spa_guid(spa);
769 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
770 rvd->vdev_guid = *newguid;
771 rvd->vdev_guid_sum += (*newguid - oldguid);
772 vdev_config_dirty(rvd);
773 spa_config_exit(spa, SCL_STATE, FTAG);
775 spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu",
780 * Change the GUID for the pool. This is done so that we can later
781 * re-import a pool built from a clone of our own vdevs. We will modify
782 * the root vdev's guid, our own pool guid, and then mark all of our
783 * vdevs dirty. Note that we must make sure that all our vdevs are
784 * online when we do this, or else any vdevs that weren't present
785 * would be orphaned from our pool. We are also going to issue a
786 * sysevent to update any watchers.
789 spa_change_guid(spa_t *spa)
794 mutex_enter(&spa->spa_vdev_top_lock);
795 mutex_enter(&spa_namespace_lock);
796 guid = spa_generate_guid(NULL);
798 error = dsl_sync_task(spa->spa_name, spa_change_guid_check,
799 spa_change_guid_sync, &guid, 5, ZFS_SPACE_CHECK_RESERVED);
802 spa_config_sync(spa, B_FALSE, B_TRUE);
803 spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID);
806 mutex_exit(&spa_namespace_lock);
807 mutex_exit(&spa->spa_vdev_top_lock);
813 * ==========================================================================
814 * SPA state manipulation (open/create/destroy/import/export)
815 * ==========================================================================
819 spa_error_entry_compare(const void *a, const void *b)
821 spa_error_entry_t *sa = (spa_error_entry_t *)a;
822 spa_error_entry_t *sb = (spa_error_entry_t *)b;
825 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
826 sizeof (zbookmark_phys_t));
837 * Utility function which retrieves copies of the current logs and
838 * re-initializes them in the process.
841 spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
843 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
845 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
846 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
848 avl_create(&spa->spa_errlist_scrub,
849 spa_error_entry_compare, sizeof (spa_error_entry_t),
850 offsetof(spa_error_entry_t, se_avl));
851 avl_create(&spa->spa_errlist_last,
852 spa_error_entry_compare, sizeof (spa_error_entry_t),
853 offsetof(spa_error_entry_t, se_avl));
857 spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
859 const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
860 enum zti_modes mode = ztip->zti_mode;
861 uint_t value = ztip->zti_value;
862 uint_t count = ztip->zti_count;
863 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
866 boolean_t batch = B_FALSE;
868 if (mode == ZTI_MODE_NULL) {
870 tqs->stqs_taskq = NULL;
874 ASSERT3U(count, >, 0);
876 tqs->stqs_count = count;
877 tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP);
881 ASSERT3U(value, >=, 1);
882 value = MAX(value, 1);
887 flags |= TASKQ_THREADS_CPU_PCT;
888 value = zio_taskq_batch_pct;
892 panic("unrecognized mode for %s_%s taskq (%u:%u) in "
894 zio_type_name[t], zio_taskq_types[q], mode, value);
898 for (uint_t i = 0; i < count; i++) {
902 (void) snprintf(name, sizeof (name), "%s_%s_%u",
903 zio_type_name[t], zio_taskq_types[q], i);
905 (void) snprintf(name, sizeof (name), "%s_%s",
906 zio_type_name[t], zio_taskq_types[q]);
910 if (zio_taskq_sysdc && spa->spa_proc != &p0) {
912 flags |= TASKQ_DC_BATCH;
914 tq = taskq_create_sysdc(name, value, 50, INT_MAX,
915 spa->spa_proc, zio_taskq_basedc, flags);
918 pri_t pri = maxclsyspri;
920 * The write issue taskq can be extremely CPU
921 * intensive. Run it at slightly lower priority
922 * than the other taskqs.
924 * - numerically higher priorities are lower priorities;
925 * - if priorities divided by four (RQ_PPQ) are equal
926 * then a difference between them is insignificant.
928 if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE)
935 tq = taskq_create_proc(name, value, pri, 50,
936 INT_MAX, spa->spa_proc, flags);
941 tqs->stqs_taskq[i] = tq;
946 spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
948 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
950 if (tqs->stqs_taskq == NULL) {
951 ASSERT0(tqs->stqs_count);
955 for (uint_t i = 0; i < tqs->stqs_count; i++) {
956 ASSERT3P(tqs->stqs_taskq[i], !=, NULL);
957 taskq_destroy(tqs->stqs_taskq[i]);
960 kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *));
961 tqs->stqs_taskq = NULL;
965 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
966 * Note that a type may have multiple discrete taskqs to avoid lock contention
967 * on the taskq itself. In that case we choose which taskq at random by using
968 * the low bits of gethrtime().
971 spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q,
972 task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent)
974 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
977 ASSERT3P(tqs->stqs_taskq, !=, NULL);
978 ASSERT3U(tqs->stqs_count, !=, 0);
980 if (tqs->stqs_count == 1) {
981 tq = tqs->stqs_taskq[0];
984 tq = tqs->stqs_taskq[cpu_ticks() % tqs->stqs_count];
986 tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count];
990 taskq_dispatch_ent(tq, func, arg, flags, ent);
994 spa_create_zio_taskqs(spa_t *spa)
996 for (int t = 0; t < ZIO_TYPES; t++) {
997 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
998 spa_taskqs_init(spa, t, q);
1006 spa_thread(void *arg)
1008 callb_cpr_t cprinfo;
1011 user_t *pu = PTOU(curproc);
1013 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
1016 ASSERT(curproc != &p0);
1017 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
1018 "zpool-%s", spa->spa_name);
1019 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
1022 /* bind this thread to the requested psrset */
1023 if (zio_taskq_psrset_bind != PS_NONE) {
1025 mutex_enter(&cpu_lock);
1026 mutex_enter(&pidlock);
1027 mutex_enter(&curproc->p_lock);
1029 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
1030 0, NULL, NULL) == 0) {
1031 curthread->t_bind_pset = zio_taskq_psrset_bind;
1034 "Couldn't bind process for zfs pool \"%s\" to "
1035 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
1038 mutex_exit(&curproc->p_lock);
1039 mutex_exit(&pidlock);
1040 mutex_exit(&cpu_lock);
1046 if (zio_taskq_sysdc) {
1047 sysdc_thread_enter(curthread, 100, 0);
1051 spa->spa_proc = curproc;
1052 spa->spa_did = curthread->t_did;
1054 spa_create_zio_taskqs(spa);
1056 mutex_enter(&spa->spa_proc_lock);
1057 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
1059 spa->spa_proc_state = SPA_PROC_ACTIVE;
1060 cv_broadcast(&spa->spa_proc_cv);
1062 CALLB_CPR_SAFE_BEGIN(&cprinfo);
1063 while (spa->spa_proc_state == SPA_PROC_ACTIVE)
1064 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1065 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
1067 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
1068 spa->spa_proc_state = SPA_PROC_GONE;
1069 spa->spa_proc = &p0;
1070 cv_broadcast(&spa->spa_proc_cv);
1071 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */
1073 mutex_enter(&curproc->p_lock);
1076 #endif /* SPA_PROCESS */
1080 * Activate an uninitialized pool.
1083 spa_activate(spa_t *spa, int mode)
1085 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
1087 spa->spa_state = POOL_STATE_ACTIVE;
1088 spa->spa_mode = mode;
1090 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
1091 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
1093 /* Try to create a covering process */
1094 mutex_enter(&spa->spa_proc_lock);
1095 ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
1096 ASSERT(spa->spa_proc == &p0);
1100 /* Only create a process if we're going to be around a while. */
1101 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
1102 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
1104 spa->spa_proc_state = SPA_PROC_CREATED;
1105 while (spa->spa_proc_state == SPA_PROC_CREATED) {
1106 cv_wait(&spa->spa_proc_cv,
1107 &spa->spa_proc_lock);
1109 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1110 ASSERT(spa->spa_proc != &p0);
1111 ASSERT(spa->spa_did != 0);
1115 "Couldn't create process for zfs pool \"%s\"\n",
1120 #endif /* SPA_PROCESS */
1121 mutex_exit(&spa->spa_proc_lock);
1123 /* If we didn't create a process, we need to create our taskqs. */
1124 ASSERT(spa->spa_proc == &p0);
1125 if (spa->spa_proc == &p0) {
1126 spa_create_zio_taskqs(spa);
1130 * Start TRIM thread.
1132 trim_thread_create(spa);
1134 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
1135 offsetof(vdev_t, vdev_config_dirty_node));
1136 list_create(&spa->spa_evicting_os_list, sizeof (objset_t),
1137 offsetof(objset_t, os_evicting_node));
1138 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
1139 offsetof(vdev_t, vdev_state_dirty_node));
1141 txg_list_create(&spa->spa_vdev_txg_list, spa,
1142 offsetof(struct vdev, vdev_txg_node));
1144 avl_create(&spa->spa_errlist_scrub,
1145 spa_error_entry_compare, sizeof (spa_error_entry_t),
1146 offsetof(spa_error_entry_t, se_avl));
1147 avl_create(&spa->spa_errlist_last,
1148 spa_error_entry_compare, sizeof (spa_error_entry_t),
1149 offsetof(spa_error_entry_t, se_avl));
1153 * Opposite of spa_activate().
1156 spa_deactivate(spa_t *spa)
1158 ASSERT(spa->spa_sync_on == B_FALSE);
1159 ASSERT(spa->spa_dsl_pool == NULL);
1160 ASSERT(spa->spa_root_vdev == NULL);
1161 ASSERT(spa->spa_async_zio_root == NULL);
1162 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
1165 * Stop TRIM thread in case spa_unload() wasn't called directly
1166 * before spa_deactivate().
1168 trim_thread_destroy(spa);
1170 spa_evicting_os_wait(spa);
1172 txg_list_destroy(&spa->spa_vdev_txg_list);
1174 list_destroy(&spa->spa_config_dirty_list);
1175 list_destroy(&spa->spa_evicting_os_list);
1176 list_destroy(&spa->spa_state_dirty_list);
1178 for (int t = 0; t < ZIO_TYPES; t++) {
1179 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1180 spa_taskqs_fini(spa, t, q);
1184 metaslab_class_destroy(spa->spa_normal_class);
1185 spa->spa_normal_class = NULL;
1187 metaslab_class_destroy(spa->spa_log_class);
1188 spa->spa_log_class = NULL;
1191 * If this was part of an import or the open otherwise failed, we may
1192 * still have errors left in the queues. Empty them just in case.
1194 spa_errlog_drain(spa);
1196 avl_destroy(&spa->spa_errlist_scrub);
1197 avl_destroy(&spa->spa_errlist_last);
1199 spa->spa_state = POOL_STATE_UNINITIALIZED;
1201 mutex_enter(&spa->spa_proc_lock);
1202 if (spa->spa_proc_state != SPA_PROC_NONE) {
1203 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1204 spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1205 cv_broadcast(&spa->spa_proc_cv);
1206 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1207 ASSERT(spa->spa_proc != &p0);
1208 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1210 ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1211 spa->spa_proc_state = SPA_PROC_NONE;
1213 ASSERT(spa->spa_proc == &p0);
1214 mutex_exit(&spa->spa_proc_lock);
1218 * We want to make sure spa_thread() has actually exited the ZFS
1219 * module, so that the module can't be unloaded out from underneath
1222 if (spa->spa_did != 0) {
1223 thread_join(spa->spa_did);
1226 #endif /* SPA_PROCESS */
1230 * Verify a pool configuration, and construct the vdev tree appropriately. This
1231 * will create all the necessary vdevs in the appropriate layout, with each vdev
1232 * in the CLOSED state. This will prep the pool before open/creation/import.
1233 * All vdev validation is done by the vdev_alloc() routine.
1236 spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1237 uint_t id, int atype)
1243 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1246 if ((*vdp)->vdev_ops->vdev_op_leaf)
1249 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1252 if (error == ENOENT)
1258 return (SET_ERROR(EINVAL));
1261 for (int c = 0; c < children; c++) {
1263 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1271 ASSERT(*vdp != NULL);
1277 * Opposite of spa_load().
1280 spa_unload(spa_t *spa)
1284 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1289 trim_thread_destroy(spa);
1294 spa_async_suspend(spa);
1299 if (spa->spa_sync_on) {
1300 txg_sync_stop(spa->spa_dsl_pool);
1301 spa->spa_sync_on = B_FALSE;
1305 * Even though vdev_free() also calls vdev_metaslab_fini, we need
1306 * to call it earlier, before we wait for async i/o to complete.
1307 * This ensures that there is no async metaslab prefetching, by
1308 * calling taskq_wait(mg_taskq).
1310 if (spa->spa_root_vdev != NULL) {
1311 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1312 for (int c = 0; c < spa->spa_root_vdev->vdev_children; c++)
1313 vdev_metaslab_fini(spa->spa_root_vdev->vdev_child[c]);
1314 spa_config_exit(spa, SCL_ALL, FTAG);
1318 * Wait for any outstanding async I/O to complete.
1320 if (spa->spa_async_zio_root != NULL) {
1321 for (int i = 0; i < max_ncpus; i++)
1322 (void) zio_wait(spa->spa_async_zio_root[i]);
1323 kmem_free(spa->spa_async_zio_root, max_ncpus * sizeof (void *));
1324 spa->spa_async_zio_root = NULL;
1327 bpobj_close(&spa->spa_deferred_bpobj);
1329 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1334 if (spa->spa_root_vdev)
1335 vdev_free(spa->spa_root_vdev);
1336 ASSERT(spa->spa_root_vdev == NULL);
1339 * Close the dsl pool.
1341 if (spa->spa_dsl_pool) {
1342 dsl_pool_close(spa->spa_dsl_pool);
1343 spa->spa_dsl_pool = NULL;
1344 spa->spa_meta_objset = NULL;
1350 * Drop and purge level 2 cache
1352 spa_l2cache_drop(spa);
1354 for (i = 0; i < spa->spa_spares.sav_count; i++)
1355 vdev_free(spa->spa_spares.sav_vdevs[i]);
1356 if (spa->spa_spares.sav_vdevs) {
1357 kmem_free(spa->spa_spares.sav_vdevs,
1358 spa->spa_spares.sav_count * sizeof (void *));
1359 spa->spa_spares.sav_vdevs = NULL;
1361 if (spa->spa_spares.sav_config) {
1362 nvlist_free(spa->spa_spares.sav_config);
1363 spa->spa_spares.sav_config = NULL;
1365 spa->spa_spares.sav_count = 0;
1367 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1368 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1369 vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1371 if (spa->spa_l2cache.sav_vdevs) {
1372 kmem_free(spa->spa_l2cache.sav_vdevs,
1373 spa->spa_l2cache.sav_count * sizeof (void *));
1374 spa->spa_l2cache.sav_vdevs = NULL;
1376 if (spa->spa_l2cache.sav_config) {
1377 nvlist_free(spa->spa_l2cache.sav_config);
1378 spa->spa_l2cache.sav_config = NULL;
1380 spa->spa_l2cache.sav_count = 0;
1382 spa->spa_async_suspended = 0;
1384 if (spa->spa_comment != NULL) {
1385 spa_strfree(spa->spa_comment);
1386 spa->spa_comment = NULL;
1389 spa_config_exit(spa, SCL_ALL, FTAG);
1393 * Load (or re-load) the current list of vdevs describing the active spares for
1394 * this pool. When this is called, we have some form of basic information in
1395 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1396 * then re-generate a more complete list including status information.
1399 spa_load_spares(spa_t *spa)
1406 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1409 * First, close and free any existing spare vdevs.
1411 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1412 vd = spa->spa_spares.sav_vdevs[i];
1414 /* Undo the call to spa_activate() below */
1415 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1416 B_FALSE)) != NULL && tvd->vdev_isspare)
1417 spa_spare_remove(tvd);
1422 if (spa->spa_spares.sav_vdevs)
1423 kmem_free(spa->spa_spares.sav_vdevs,
1424 spa->spa_spares.sav_count * sizeof (void *));
1426 if (spa->spa_spares.sav_config == NULL)
1429 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1430 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1432 spa->spa_spares.sav_count = (int)nspares;
1433 spa->spa_spares.sav_vdevs = NULL;
1439 * Construct the array of vdevs, opening them to get status in the
1440 * process. For each spare, there is potentially two different vdev_t
1441 * structures associated with it: one in the list of spares (used only
1442 * for basic validation purposes) and one in the active vdev
1443 * configuration (if it's spared in). During this phase we open and
1444 * validate each vdev on the spare list. If the vdev also exists in the
1445 * active configuration, then we also mark this vdev as an active spare.
1447 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1449 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1450 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1451 VDEV_ALLOC_SPARE) == 0);
1454 spa->spa_spares.sav_vdevs[i] = vd;
1456 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1457 B_FALSE)) != NULL) {
1458 if (!tvd->vdev_isspare)
1462 * We only mark the spare active if we were successfully
1463 * able to load the vdev. Otherwise, importing a pool
1464 * with a bad active spare would result in strange
1465 * behavior, because multiple pool would think the spare
1466 * is actively in use.
1468 * There is a vulnerability here to an equally bizarre
1469 * circumstance, where a dead active spare is later
1470 * brought back to life (onlined or otherwise). Given
1471 * the rarity of this scenario, and the extra complexity
1472 * it adds, we ignore the possibility.
1474 if (!vdev_is_dead(tvd))
1475 spa_spare_activate(tvd);
1479 vd->vdev_aux = &spa->spa_spares;
1481 if (vdev_open(vd) != 0)
1484 if (vdev_validate_aux(vd) == 0)
1489 * Recompute the stashed list of spares, with status information
1492 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1493 DATA_TYPE_NVLIST_ARRAY) == 0);
1495 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1497 for (i = 0; i < spa->spa_spares.sav_count; i++)
1498 spares[i] = vdev_config_generate(spa,
1499 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1500 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1501 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1502 for (i = 0; i < spa->spa_spares.sav_count; i++)
1503 nvlist_free(spares[i]);
1504 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1508 * Load (or re-load) the current list of vdevs describing the active l2cache for
1509 * this pool. When this is called, we have some form of basic information in
1510 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1511 * then re-generate a more complete list including status information.
1512 * Devices which are already active have their details maintained, and are
1516 spa_load_l2cache(spa_t *spa)
1520 int i, j, oldnvdevs;
1522 vdev_t *vd, **oldvdevs, **newvdevs;
1523 spa_aux_vdev_t *sav = &spa->spa_l2cache;
1525 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1527 if (sav->sav_config != NULL) {
1528 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1529 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1530 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1536 oldvdevs = sav->sav_vdevs;
1537 oldnvdevs = sav->sav_count;
1538 sav->sav_vdevs = NULL;
1542 * Process new nvlist of vdevs.
1544 for (i = 0; i < nl2cache; i++) {
1545 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1549 for (j = 0; j < oldnvdevs; j++) {
1551 if (vd != NULL && guid == vd->vdev_guid) {
1553 * Retain previous vdev for add/remove ops.
1561 if (newvdevs[i] == NULL) {
1565 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1566 VDEV_ALLOC_L2CACHE) == 0);
1571 * Commit this vdev as an l2cache device,
1572 * even if it fails to open.
1574 spa_l2cache_add(vd);
1579 spa_l2cache_activate(vd);
1581 if (vdev_open(vd) != 0)
1584 (void) vdev_validate_aux(vd);
1586 if (!vdev_is_dead(vd))
1587 l2arc_add_vdev(spa, vd);
1592 * Purge vdevs that were dropped
1594 for (i = 0; i < oldnvdevs; i++) {
1599 ASSERT(vd->vdev_isl2cache);
1601 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1602 pool != 0ULL && l2arc_vdev_present(vd))
1603 l2arc_remove_vdev(vd);
1604 vdev_clear_stats(vd);
1610 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1612 if (sav->sav_config == NULL)
1615 sav->sav_vdevs = newvdevs;
1616 sav->sav_count = (int)nl2cache;
1619 * Recompute the stashed list of l2cache devices, with status
1620 * information this time.
1622 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1623 DATA_TYPE_NVLIST_ARRAY) == 0);
1625 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1626 for (i = 0; i < sav->sav_count; i++)
1627 l2cache[i] = vdev_config_generate(spa,
1628 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1629 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1630 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1632 for (i = 0; i < sav->sav_count; i++)
1633 nvlist_free(l2cache[i]);
1635 kmem_free(l2cache, sav->sav_count * sizeof (void *));
1639 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1642 char *packed = NULL;
1647 error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db);
1651 nvsize = *(uint64_t *)db->db_data;
1652 dmu_buf_rele(db, FTAG);
1654 packed = kmem_alloc(nvsize, KM_SLEEP);
1655 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1658 error = nvlist_unpack(packed, nvsize, value, 0);
1659 kmem_free(packed, nvsize);
1665 * Checks to see if the given vdev could not be opened, in which case we post a
1666 * sysevent to notify the autoreplace code that the device has been removed.
1669 spa_check_removed(vdev_t *vd)
1671 for (int c = 0; c < vd->vdev_children; c++)
1672 spa_check_removed(vd->vdev_child[c]);
1674 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) &&
1676 zfs_post_autoreplace(vd->vdev_spa, vd);
1677 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1682 spa_config_valid_zaps(vdev_t *vd, vdev_t *mvd)
1684 ASSERT3U(vd->vdev_children, ==, mvd->vdev_children);
1686 vd->vdev_top_zap = mvd->vdev_top_zap;
1687 vd->vdev_leaf_zap = mvd->vdev_leaf_zap;
1689 for (uint64_t i = 0; i < vd->vdev_children; i++) {
1690 spa_config_valid_zaps(vd->vdev_child[i], mvd->vdev_child[i]);
1695 * Validate the current config against the MOS config
1698 spa_config_valid(spa_t *spa, nvlist_t *config)
1700 vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1703 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1705 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1706 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1708 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1711 * If we're doing a normal import, then build up any additional
1712 * diagnostic information about missing devices in this config.
1713 * We'll pass this up to the user for further processing.
1715 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1716 nvlist_t **child, *nv;
1719 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1721 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1723 for (int c = 0; c < rvd->vdev_children; c++) {
1724 vdev_t *tvd = rvd->vdev_child[c];
1725 vdev_t *mtvd = mrvd->vdev_child[c];
1727 if (tvd->vdev_ops == &vdev_missing_ops &&
1728 mtvd->vdev_ops != &vdev_missing_ops &&
1730 child[idx++] = vdev_config_generate(spa, mtvd,
1735 VERIFY(nvlist_add_nvlist_array(nv,
1736 ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1737 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1738 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1740 for (int i = 0; i < idx; i++)
1741 nvlist_free(child[i]);
1744 kmem_free(child, rvd->vdev_children * sizeof (char **));
1748 * Compare the root vdev tree with the information we have
1749 * from the MOS config (mrvd). Check each top-level vdev
1750 * with the corresponding MOS config top-level (mtvd).
1752 for (int c = 0; c < rvd->vdev_children; c++) {
1753 vdev_t *tvd = rvd->vdev_child[c];
1754 vdev_t *mtvd = mrvd->vdev_child[c];
1757 * Resolve any "missing" vdevs in the current configuration.
1758 * If we find that the MOS config has more accurate information
1759 * about the top-level vdev then use that vdev instead.
1761 if (tvd->vdev_ops == &vdev_missing_ops &&
1762 mtvd->vdev_ops != &vdev_missing_ops) {
1764 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1768 * Device specific actions.
1770 if (mtvd->vdev_islog) {
1771 spa_set_log_state(spa, SPA_LOG_CLEAR);
1774 * XXX - once we have 'readonly' pool
1775 * support we should be able to handle
1776 * missing data devices by transitioning
1777 * the pool to readonly.
1783 * Swap the missing vdev with the data we were
1784 * able to obtain from the MOS config.
1786 vdev_remove_child(rvd, tvd);
1787 vdev_remove_child(mrvd, mtvd);
1789 vdev_add_child(rvd, mtvd);
1790 vdev_add_child(mrvd, tvd);
1792 spa_config_exit(spa, SCL_ALL, FTAG);
1794 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1798 if (mtvd->vdev_islog) {
1800 * Load the slog device's state from the MOS
1801 * config since it's possible that the label
1802 * does not contain the most up-to-date
1805 vdev_load_log_state(tvd, mtvd);
1810 * Per-vdev ZAP info is stored exclusively in the MOS.
1812 spa_config_valid_zaps(tvd, mtvd);
1817 spa_config_exit(spa, SCL_ALL, FTAG);
1820 * Ensure we were able to validate the config.
1822 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1826 * Check for missing log devices
1829 spa_check_logs(spa_t *spa)
1831 boolean_t rv = B_FALSE;
1832 dsl_pool_t *dp = spa_get_dsl(spa);
1834 switch (spa->spa_log_state) {
1835 case SPA_LOG_MISSING:
1836 /* need to recheck in case slog has been restored */
1837 case SPA_LOG_UNKNOWN:
1838 rv = (dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
1839 zil_check_log_chain, NULL, DS_FIND_CHILDREN) != 0);
1841 spa_set_log_state(spa, SPA_LOG_MISSING);
1848 spa_passivate_log(spa_t *spa)
1850 vdev_t *rvd = spa->spa_root_vdev;
1851 boolean_t slog_found = B_FALSE;
1853 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1855 if (!spa_has_slogs(spa))
1858 for (int c = 0; c < rvd->vdev_children; c++) {
1859 vdev_t *tvd = rvd->vdev_child[c];
1860 metaslab_group_t *mg = tvd->vdev_mg;
1862 if (tvd->vdev_islog) {
1863 metaslab_group_passivate(mg);
1864 slog_found = B_TRUE;
1868 return (slog_found);
1872 spa_activate_log(spa_t *spa)
1874 vdev_t *rvd = spa->spa_root_vdev;
1876 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1878 for (int c = 0; c < rvd->vdev_children; c++) {
1879 vdev_t *tvd = rvd->vdev_child[c];
1880 metaslab_group_t *mg = tvd->vdev_mg;
1882 if (tvd->vdev_islog)
1883 metaslab_group_activate(mg);
1888 spa_offline_log(spa_t *spa)
1892 error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1893 NULL, DS_FIND_CHILDREN);
1896 * We successfully offlined the log device, sync out the
1897 * current txg so that the "stubby" block can be removed
1900 txg_wait_synced(spa->spa_dsl_pool, 0);
1906 spa_aux_check_removed(spa_aux_vdev_t *sav)
1910 for (i = 0; i < sav->sav_count; i++)
1911 spa_check_removed(sav->sav_vdevs[i]);
1915 spa_claim_notify(zio_t *zio)
1917 spa_t *spa = zio->io_spa;
1922 mutex_enter(&spa->spa_props_lock); /* any mutex will do */
1923 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1924 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1925 mutex_exit(&spa->spa_props_lock);
1928 typedef struct spa_load_error {
1929 uint64_t sle_meta_count;
1930 uint64_t sle_data_count;
1934 spa_load_verify_done(zio_t *zio)
1936 blkptr_t *bp = zio->io_bp;
1937 spa_load_error_t *sle = zio->io_private;
1938 dmu_object_type_t type = BP_GET_TYPE(bp);
1939 int error = zio->io_error;
1940 spa_t *spa = zio->io_spa;
1942 abd_free(zio->io_abd);
1944 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1945 type != DMU_OT_INTENT_LOG)
1946 atomic_inc_64(&sle->sle_meta_count);
1948 atomic_inc_64(&sle->sle_data_count);
1951 mutex_enter(&spa->spa_scrub_lock);
1952 spa->spa_scrub_inflight--;
1953 cv_broadcast(&spa->spa_scrub_io_cv);
1954 mutex_exit(&spa->spa_scrub_lock);
1958 * Maximum number of concurrent scrub i/os to create while verifying
1959 * a pool while importing it.
1961 int spa_load_verify_maxinflight = 10000;
1962 boolean_t spa_load_verify_metadata = B_TRUE;
1963 boolean_t spa_load_verify_data = B_TRUE;
1965 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_maxinflight, CTLFLAG_RWTUN,
1966 &spa_load_verify_maxinflight, 0,
1967 "Maximum number of concurrent scrub I/Os to create while verifying a "
1968 "pool while importing it");
1970 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_metadata, CTLFLAG_RWTUN,
1971 &spa_load_verify_metadata, 0,
1972 "Check metadata on import?");
1974 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_data, CTLFLAG_RWTUN,
1975 &spa_load_verify_data, 0,
1976 "Check user data on import?");
1980 spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1981 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
1983 if (bp == NULL || BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp))
1986 * Note: normally this routine will not be called if
1987 * spa_load_verify_metadata is not set. However, it may be useful
1988 * to manually set the flag after the traversal has begun.
1990 if (!spa_load_verify_metadata)
1992 if (!BP_IS_METADATA(bp) && !spa_load_verify_data)
1996 size_t size = BP_GET_PSIZE(bp);
1998 mutex_enter(&spa->spa_scrub_lock);
1999 while (spa->spa_scrub_inflight >= spa_load_verify_maxinflight)
2000 cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
2001 spa->spa_scrub_inflight++;
2002 mutex_exit(&spa->spa_scrub_lock);
2004 zio_nowait(zio_read(rio, spa, bp, abd_alloc_for_io(size, B_FALSE), size,
2005 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
2006 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
2007 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
2013 verify_dataset_name_len(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg)
2015 if (dsl_dataset_namelen(ds) >= ZFS_MAX_DATASET_NAME_LEN)
2016 return (SET_ERROR(ENAMETOOLONG));
2022 spa_load_verify(spa_t *spa)
2025 spa_load_error_t sle = { 0 };
2026 zpool_rewind_policy_t policy;
2027 boolean_t verify_ok = B_FALSE;
2030 zpool_get_rewind_policy(spa->spa_config, &policy);
2032 if (policy.zrp_request & ZPOOL_NEVER_REWIND)
2035 dsl_pool_config_enter(spa->spa_dsl_pool, FTAG);
2036 error = dmu_objset_find_dp(spa->spa_dsl_pool,
2037 spa->spa_dsl_pool->dp_root_dir_obj, verify_dataset_name_len, NULL,
2039 dsl_pool_config_exit(spa->spa_dsl_pool, FTAG);
2043 rio = zio_root(spa, NULL, &sle,
2044 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
2046 if (spa_load_verify_metadata) {
2047 error = traverse_pool(spa, spa->spa_verify_min_txg,
2048 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA,
2049 spa_load_verify_cb, rio);
2052 (void) zio_wait(rio);
2054 spa->spa_load_meta_errors = sle.sle_meta_count;
2055 spa->spa_load_data_errors = sle.sle_data_count;
2057 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
2058 sle.sle_data_count <= policy.zrp_maxdata) {
2062 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
2063 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
2065 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
2066 VERIFY(nvlist_add_uint64(spa->spa_load_info,
2067 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
2068 VERIFY(nvlist_add_int64(spa->spa_load_info,
2069 ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
2070 VERIFY(nvlist_add_uint64(spa->spa_load_info,
2071 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
2073 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
2077 if (error != ENXIO && error != EIO)
2078 error = SET_ERROR(EIO);
2082 return (verify_ok ? 0 : EIO);
2086 * Find a value in the pool props object.
2089 spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
2091 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
2092 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
2096 * Find a value in the pool directory object.
2099 spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
2101 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
2102 name, sizeof (uint64_t), 1, val));
2106 spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
2108 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
2113 * Fix up config after a partly-completed split. This is done with the
2114 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
2115 * pool have that entry in their config, but only the splitting one contains
2116 * a list of all the guids of the vdevs that are being split off.
2118 * This function determines what to do with that list: either rejoin
2119 * all the disks to the pool, or complete the splitting process. To attempt
2120 * the rejoin, each disk that is offlined is marked online again, and
2121 * we do a reopen() call. If the vdev label for every disk that was
2122 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
2123 * then we call vdev_split() on each disk, and complete the split.
2125 * Otherwise we leave the config alone, with all the vdevs in place in
2126 * the original pool.
2129 spa_try_repair(spa_t *spa, nvlist_t *config)
2136 boolean_t attempt_reopen;
2138 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
2141 /* check that the config is complete */
2142 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
2143 &glist, &gcount) != 0)
2146 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
2148 /* attempt to online all the vdevs & validate */
2149 attempt_reopen = B_TRUE;
2150 for (i = 0; i < gcount; i++) {
2151 if (glist[i] == 0) /* vdev is hole */
2154 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
2155 if (vd[i] == NULL) {
2157 * Don't bother attempting to reopen the disks;
2158 * just do the split.
2160 attempt_reopen = B_FALSE;
2162 /* attempt to re-online it */
2163 vd[i]->vdev_offline = B_FALSE;
2167 if (attempt_reopen) {
2168 vdev_reopen(spa->spa_root_vdev);
2170 /* check each device to see what state it's in */
2171 for (extracted = 0, i = 0; i < gcount; i++) {
2172 if (vd[i] != NULL &&
2173 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
2180 * If every disk has been moved to the new pool, or if we never
2181 * even attempted to look at them, then we split them off for
2184 if (!attempt_reopen || gcount == extracted) {
2185 for (i = 0; i < gcount; i++)
2188 vdev_reopen(spa->spa_root_vdev);
2191 kmem_free(vd, gcount * sizeof (vdev_t *));
2195 spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
2196 boolean_t mosconfig)
2198 nvlist_t *config = spa->spa_config;
2199 char *ereport = FM_EREPORT_ZFS_POOL;
2205 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
2206 return (SET_ERROR(EINVAL));
2208 ASSERT(spa->spa_comment == NULL);
2209 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
2210 spa->spa_comment = spa_strdup(comment);
2213 * Versioning wasn't explicitly added to the label until later, so if
2214 * it's not present treat it as the initial version.
2216 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
2217 &spa->spa_ubsync.ub_version) != 0)
2218 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
2220 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
2221 &spa->spa_config_txg);
2223 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
2224 spa_guid_exists(pool_guid, 0)) {
2225 error = SET_ERROR(EEXIST);
2227 spa->spa_config_guid = pool_guid;
2229 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
2231 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
2235 nvlist_free(spa->spa_load_info);
2236 spa->spa_load_info = fnvlist_alloc();
2238 gethrestime(&spa->spa_loaded_ts);
2239 error = spa_load_impl(spa, pool_guid, config, state, type,
2240 mosconfig, &ereport);
2244 * Don't count references from objsets that are already closed
2245 * and are making their way through the eviction process.
2247 spa_evicting_os_wait(spa);
2248 spa->spa_minref = refcount_count(&spa->spa_refcount);
2250 if (error != EEXIST) {
2251 spa->spa_loaded_ts.tv_sec = 0;
2252 spa->spa_loaded_ts.tv_nsec = 0;
2254 if (error != EBADF) {
2255 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
2258 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2265 * Count the number of per-vdev ZAPs associated with all of the vdevs in the
2266 * vdev tree rooted in the given vd, and ensure that each ZAP is present in the
2267 * spa's per-vdev ZAP list.
2270 vdev_count_verify_zaps(vdev_t *vd)
2272 spa_t *spa = vd->vdev_spa;
2274 if (vd->vdev_top_zap != 0) {
2276 ASSERT0(zap_lookup_int(spa->spa_meta_objset,
2277 spa->spa_all_vdev_zaps, vd->vdev_top_zap));
2279 if (vd->vdev_leaf_zap != 0) {
2281 ASSERT0(zap_lookup_int(spa->spa_meta_objset,
2282 spa->spa_all_vdev_zaps, vd->vdev_leaf_zap));
2285 for (uint64_t i = 0; i < vd->vdev_children; i++) {
2286 total += vdev_count_verify_zaps(vd->vdev_child[i]);
2293 * Load an existing storage pool, using the pool's builtin spa_config as a
2294 * source of configuration information.
2297 spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
2298 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
2302 nvlist_t *nvroot = NULL;
2305 uberblock_t *ub = &spa->spa_uberblock;
2306 uint64_t children, config_cache_txg = spa->spa_config_txg;
2307 int orig_mode = spa->spa_mode;
2310 boolean_t missing_feat_write = B_FALSE;
2313 * If this is an untrusted config, access the pool in read-only mode.
2314 * This prevents things like resilvering recently removed devices.
2317 spa->spa_mode = FREAD;
2319 ASSERT(MUTEX_HELD(&spa_namespace_lock));
2321 spa->spa_load_state = state;
2323 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
2324 return (SET_ERROR(EINVAL));
2326 parse = (type == SPA_IMPORT_EXISTING ?
2327 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2330 * Create "The Godfather" zio to hold all async IOs
2332 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
2334 for (int i = 0; i < max_ncpus; i++) {
2335 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
2336 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
2337 ZIO_FLAG_GODFATHER);
2341 * Parse the configuration into a vdev tree. We explicitly set the
2342 * value that will be returned by spa_version() since parsing the
2343 * configuration requires knowing the version number.
2345 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2346 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
2347 spa_config_exit(spa, SCL_ALL, FTAG);
2352 ASSERT(spa->spa_root_vdev == rvd);
2353 ASSERT3U(spa->spa_min_ashift, >=, SPA_MINBLOCKSHIFT);
2354 ASSERT3U(spa->spa_max_ashift, <=, SPA_MAXBLOCKSHIFT);
2356 if (type != SPA_IMPORT_ASSEMBLE) {
2357 ASSERT(spa_guid(spa) == pool_guid);
2361 * Try to open all vdevs, loading each label in the process.
2363 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2364 error = vdev_open(rvd);
2365 spa_config_exit(spa, SCL_ALL, FTAG);
2370 * We need to validate the vdev labels against the configuration that
2371 * we have in hand, which is dependent on the setting of mosconfig. If
2372 * mosconfig is true then we're validating the vdev labels based on
2373 * that config. Otherwise, we're validating against the cached config
2374 * (zpool.cache) that was read when we loaded the zfs module, and then
2375 * later we will recursively call spa_load() and validate against
2378 * If we're assembling a new pool that's been split off from an
2379 * existing pool, the labels haven't yet been updated so we skip
2380 * validation for now.
2382 if (type != SPA_IMPORT_ASSEMBLE) {
2383 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2384 error = vdev_validate(rvd, mosconfig);
2385 spa_config_exit(spa, SCL_ALL, FTAG);
2390 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2391 return (SET_ERROR(ENXIO));
2395 * Find the best uberblock.
2397 vdev_uberblock_load(rvd, ub, &label);
2400 * If we weren't able to find a single valid uberblock, return failure.
2402 if (ub->ub_txg == 0) {
2404 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2408 * If the pool has an unsupported version we can't open it.
2410 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2412 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2415 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2419 * If we weren't able to find what's necessary for reading the
2420 * MOS in the label, return failure.
2422 if (label == NULL || nvlist_lookup_nvlist(label,
2423 ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) {
2425 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2430 * Update our in-core representation with the definitive values
2433 nvlist_free(spa->spa_label_features);
2434 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2440 * Look through entries in the label nvlist's features_for_read. If
2441 * there is a feature listed there which we don't understand then we
2442 * cannot open a pool.
2444 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2445 nvlist_t *unsup_feat;
2447 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2450 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2452 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2453 if (!zfeature_is_supported(nvpair_name(nvp))) {
2454 VERIFY(nvlist_add_string(unsup_feat,
2455 nvpair_name(nvp), "") == 0);
2459 if (!nvlist_empty(unsup_feat)) {
2460 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2461 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2462 nvlist_free(unsup_feat);
2463 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2467 nvlist_free(unsup_feat);
2471 * If the vdev guid sum doesn't match the uberblock, we have an
2472 * incomplete configuration. We first check to see if the pool
2473 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2474 * If it is, defer the vdev_guid_sum check till later so we
2475 * can handle missing vdevs.
2477 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
2478 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
2479 rvd->vdev_guid_sum != ub->ub_guid_sum)
2480 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2482 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2483 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2484 spa_try_repair(spa, config);
2485 spa_config_exit(spa, SCL_ALL, FTAG);
2486 nvlist_free(spa->spa_config_splitting);
2487 spa->spa_config_splitting = NULL;
2491 * Initialize internal SPA structures.
2493 spa->spa_state = POOL_STATE_ACTIVE;
2494 spa->spa_ubsync = spa->spa_uberblock;
2495 spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2496 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2497 spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2498 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2499 spa->spa_claim_max_txg = spa->spa_first_txg;
2500 spa->spa_prev_software_version = ub->ub_software_version;
2502 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2504 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2505 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2507 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2508 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2510 if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2511 boolean_t missing_feat_read = B_FALSE;
2512 nvlist_t *unsup_feat, *enabled_feat;
2514 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2515 &spa->spa_feat_for_read_obj) != 0) {
2516 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2519 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2520 &spa->spa_feat_for_write_obj) != 0) {
2521 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2524 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2525 &spa->spa_feat_desc_obj) != 0) {
2526 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2529 enabled_feat = fnvlist_alloc();
2530 unsup_feat = fnvlist_alloc();
2532 if (!spa_features_check(spa, B_FALSE,
2533 unsup_feat, enabled_feat))
2534 missing_feat_read = B_TRUE;
2536 if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) {
2537 if (!spa_features_check(spa, B_TRUE,
2538 unsup_feat, enabled_feat)) {
2539 missing_feat_write = B_TRUE;
2543 fnvlist_add_nvlist(spa->spa_load_info,
2544 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
2546 if (!nvlist_empty(unsup_feat)) {
2547 fnvlist_add_nvlist(spa->spa_load_info,
2548 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
2551 fnvlist_free(enabled_feat);
2552 fnvlist_free(unsup_feat);
2554 if (!missing_feat_read) {
2555 fnvlist_add_boolean(spa->spa_load_info,
2556 ZPOOL_CONFIG_CAN_RDONLY);
2560 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2561 * twofold: to determine whether the pool is available for
2562 * import in read-write mode and (if it is not) whether the
2563 * pool is available for import in read-only mode. If the pool
2564 * is available for import in read-write mode, it is displayed
2565 * as available in userland; if it is not available for import
2566 * in read-only mode, it is displayed as unavailable in
2567 * userland. If the pool is available for import in read-only
2568 * mode but not read-write mode, it is displayed as unavailable
2569 * in userland with a special note that the pool is actually
2570 * available for open in read-only mode.
2572 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2573 * missing a feature for write, we must first determine whether
2574 * the pool can be opened read-only before returning to
2575 * userland in order to know whether to display the
2576 * abovementioned note.
2578 if (missing_feat_read || (missing_feat_write &&
2579 spa_writeable(spa))) {
2580 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2585 * Load refcounts for ZFS features from disk into an in-memory
2586 * cache during SPA initialization.
2588 for (spa_feature_t i = 0; i < SPA_FEATURES; i++) {
2591 error = feature_get_refcount_from_disk(spa,
2592 &spa_feature_table[i], &refcount);
2594 spa->spa_feat_refcount_cache[i] = refcount;
2595 } else if (error == ENOTSUP) {
2596 spa->spa_feat_refcount_cache[i] =
2597 SPA_FEATURE_DISABLED;
2599 return (spa_vdev_err(rvd,
2600 VDEV_AUX_CORRUPT_DATA, EIO));
2605 if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) {
2606 if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG,
2607 &spa->spa_feat_enabled_txg_obj) != 0)
2608 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2611 spa->spa_is_initializing = B_TRUE;
2612 error = dsl_pool_open(spa->spa_dsl_pool);
2613 spa->spa_is_initializing = B_FALSE;
2615 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2619 nvlist_t *policy = NULL, *nvconfig;
2621 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2622 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2624 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2625 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2627 unsigned long myhostid = 0;
2629 VERIFY(nvlist_lookup_string(nvconfig,
2630 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2633 myhostid = zone_get_hostid(NULL);
2636 * We're emulating the system's hostid in userland, so
2637 * we can't use zone_get_hostid().
2639 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2640 #endif /* _KERNEL */
2641 if (check_hostid && hostid != 0 && myhostid != 0 &&
2642 hostid != myhostid) {
2643 nvlist_free(nvconfig);
2644 cmn_err(CE_WARN, "pool '%s' could not be "
2645 "loaded as it was last accessed by "
2646 "another system (host: %s hostid: 0x%lx). "
2647 "See: http://illumos.org/msg/ZFS-8000-EY",
2648 spa_name(spa), hostname,
2649 (unsigned long)hostid);
2650 return (SET_ERROR(EBADF));
2653 if (nvlist_lookup_nvlist(spa->spa_config,
2654 ZPOOL_REWIND_POLICY, &policy) == 0)
2655 VERIFY(nvlist_add_nvlist(nvconfig,
2656 ZPOOL_REWIND_POLICY, policy) == 0);
2658 spa_config_set(spa, nvconfig);
2660 spa_deactivate(spa);
2661 spa_activate(spa, orig_mode);
2663 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2666 /* Grab the secret checksum salt from the MOS. */
2667 error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
2668 DMU_POOL_CHECKSUM_SALT, 1,
2669 sizeof (spa->spa_cksum_salt.zcs_bytes),
2670 spa->spa_cksum_salt.zcs_bytes);
2671 if (error == ENOENT) {
2672 /* Generate a new salt for subsequent use */
2673 (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes,
2674 sizeof (spa->spa_cksum_salt.zcs_bytes));
2675 } else if (error != 0) {
2676 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2679 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2680 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2681 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2683 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2686 * Load the bit that tells us to use the new accounting function
2687 * (raid-z deflation). If we have an older pool, this will not
2690 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2691 if (error != 0 && error != ENOENT)
2692 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2694 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2695 &spa->spa_creation_version);
2696 if (error != 0 && error != ENOENT)
2697 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2700 * Load the persistent error log. If we have an older pool, this will
2703 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2704 if (error != 0 && error != ENOENT)
2705 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2707 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2708 &spa->spa_errlog_scrub);
2709 if (error != 0 && error != ENOENT)
2710 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2713 * Load the history object. If we have an older pool, this
2714 * will not be present.
2716 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2717 if (error != 0 && error != ENOENT)
2718 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2721 * Load the per-vdev ZAP map. If we have an older pool, this will not
2722 * be present; in this case, defer its creation to a later time to
2723 * avoid dirtying the MOS this early / out of sync context. See
2724 * spa_sync_config_object.
2727 /* The sentinel is only available in the MOS config. */
2728 nvlist_t *mos_config;
2729 if (load_nvlist(spa, spa->spa_config_object, &mos_config) != 0)
2730 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2732 error = spa_dir_prop(spa, DMU_POOL_VDEV_ZAP_MAP,
2733 &spa->spa_all_vdev_zaps);
2735 if (error == ENOENT) {
2736 VERIFY(!nvlist_exists(mos_config,
2737 ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS));
2738 spa->spa_avz_action = AVZ_ACTION_INITIALIZE;
2739 ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev));
2740 } else if (error != 0) {
2741 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2742 } else if (!nvlist_exists(mos_config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)) {
2744 * An older version of ZFS overwrote the sentinel value, so
2745 * we have orphaned per-vdev ZAPs in the MOS. Defer their
2746 * destruction to later; see spa_sync_config_object.
2748 spa->spa_avz_action = AVZ_ACTION_DESTROY;
2750 * We're assuming that no vdevs have had their ZAPs created
2751 * before this. Better be sure of it.
2753 ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev));
2755 nvlist_free(mos_config);
2758 * If we're assembling the pool from the split-off vdevs of
2759 * an existing pool, we don't want to attach the spares & cache
2764 * Load any hot spares for this pool.
2766 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2767 if (error != 0 && error != ENOENT)
2768 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2769 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2770 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2771 if (load_nvlist(spa, spa->spa_spares.sav_object,
2772 &spa->spa_spares.sav_config) != 0)
2773 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2775 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2776 spa_load_spares(spa);
2777 spa_config_exit(spa, SCL_ALL, FTAG);
2778 } else if (error == 0) {
2779 spa->spa_spares.sav_sync = B_TRUE;
2783 * Load any level 2 ARC devices for this pool.
2785 error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2786 &spa->spa_l2cache.sav_object);
2787 if (error != 0 && error != ENOENT)
2788 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2789 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2790 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2791 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2792 &spa->spa_l2cache.sav_config) != 0)
2793 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2795 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2796 spa_load_l2cache(spa);
2797 spa_config_exit(spa, SCL_ALL, FTAG);
2798 } else if (error == 0) {
2799 spa->spa_l2cache.sav_sync = B_TRUE;
2802 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2804 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2805 if (error && error != ENOENT)
2806 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2809 uint64_t autoreplace;
2811 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2812 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2813 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2814 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2815 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2816 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2817 &spa->spa_dedup_ditto);
2819 spa->spa_autoreplace = (autoreplace != 0);
2823 * If the 'autoreplace' property is set, then post a resource notifying
2824 * the ZFS DE that it should not issue any faults for unopenable
2825 * devices. We also iterate over the vdevs, and post a sysevent for any
2826 * unopenable vdevs so that the normal autoreplace handler can take
2829 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2830 spa_check_removed(spa->spa_root_vdev);
2832 * For the import case, this is done in spa_import(), because
2833 * at this point we're using the spare definitions from
2834 * the MOS config, not necessarily from the userland config.
2836 if (state != SPA_LOAD_IMPORT) {
2837 spa_aux_check_removed(&spa->spa_spares);
2838 spa_aux_check_removed(&spa->spa_l2cache);
2843 * Load the vdev state for all toplevel vdevs.
2848 * Propagate the leaf DTLs we just loaded all the way up the tree.
2850 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2851 vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2852 spa_config_exit(spa, SCL_ALL, FTAG);
2855 * Load the DDTs (dedup tables).
2857 error = ddt_load(spa);
2859 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2861 spa_update_dspace(spa);
2864 * Validate the config, using the MOS config to fill in any
2865 * information which might be missing. If we fail to validate
2866 * the config then declare the pool unfit for use. If we're
2867 * assembling a pool from a split, the log is not transferred
2870 if (type != SPA_IMPORT_ASSEMBLE) {
2873 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2874 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2876 if (!spa_config_valid(spa, nvconfig)) {
2877 nvlist_free(nvconfig);
2878 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2881 nvlist_free(nvconfig);
2884 * Now that we've validated the config, check the state of the
2885 * root vdev. If it can't be opened, it indicates one or
2886 * more toplevel vdevs are faulted.
2888 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2889 return (SET_ERROR(ENXIO));
2891 if (spa_writeable(spa) && spa_check_logs(spa)) {
2892 *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2893 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2897 if (missing_feat_write) {
2898 ASSERT(state == SPA_LOAD_TRYIMPORT);
2901 * At this point, we know that we can open the pool in
2902 * read-only mode but not read-write mode. We now have enough
2903 * information and can return to userland.
2905 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
2909 * We've successfully opened the pool, verify that we're ready
2910 * to start pushing transactions.
2912 if (state != SPA_LOAD_TRYIMPORT) {
2913 if (error = spa_load_verify(spa))
2914 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2918 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2919 spa->spa_load_max_txg == UINT64_MAX)) {
2921 int need_update = B_FALSE;
2922 dsl_pool_t *dp = spa_get_dsl(spa);
2924 ASSERT(state != SPA_LOAD_TRYIMPORT);
2927 * Claim log blocks that haven't been committed yet.
2928 * This must all happen in a single txg.
2929 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2930 * invoked from zil_claim_log_block()'s i/o done callback.
2931 * Price of rollback is that we abandon the log.
2933 spa->spa_claiming = B_TRUE;
2935 tx = dmu_tx_create_assigned(dp, spa_first_txg(spa));
2936 (void) dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
2937 zil_claim, tx, DS_FIND_CHILDREN);
2940 spa->spa_claiming = B_FALSE;
2942 spa_set_log_state(spa, SPA_LOG_GOOD);
2943 spa->spa_sync_on = B_TRUE;
2944 txg_sync_start(spa->spa_dsl_pool);
2947 * Wait for all claims to sync. We sync up to the highest
2948 * claimed log block birth time so that claimed log blocks
2949 * don't appear to be from the future. spa_claim_max_txg
2950 * will have been set for us by either zil_check_log_chain()
2951 * (invoked from spa_check_logs()) or zil_claim() above.
2953 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2956 * If the config cache is stale, or we have uninitialized
2957 * metaslabs (see spa_vdev_add()), then update the config.
2959 * If this is a verbatim import, trust the current
2960 * in-core spa_config and update the disk labels.
2962 if (config_cache_txg != spa->spa_config_txg ||
2963 state == SPA_LOAD_IMPORT ||
2964 state == SPA_LOAD_RECOVER ||
2965 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2966 need_update = B_TRUE;
2968 for (int c = 0; c < rvd->vdev_children; c++)
2969 if (rvd->vdev_child[c]->vdev_ms_array == 0)
2970 need_update = B_TRUE;
2973 * Update the config cache asychronously in case we're the
2974 * root pool, in which case the config cache isn't writable yet.
2977 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2980 * Check all DTLs to see if anything needs resilvering.
2982 if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2983 vdev_resilver_needed(rvd, NULL, NULL))
2984 spa_async_request(spa, SPA_ASYNC_RESILVER);
2987 * Log the fact that we booted up (so that we can detect if
2988 * we rebooted in the middle of an operation).
2990 spa_history_log_version(spa, "open");
2993 * Delete any inconsistent datasets.
2995 (void) dmu_objset_find(spa_name(spa),
2996 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2999 * Clean up any stale temporary dataset userrefs.
3001 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
3008 spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
3010 int mode = spa->spa_mode;
3013 spa_deactivate(spa);
3015 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg - 1;
3017 spa_activate(spa, mode);
3018 spa_async_suspend(spa);
3020 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
3024 * If spa_load() fails this function will try loading prior txg's. If
3025 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
3026 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
3027 * function will not rewind the pool and will return the same error as
3031 spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
3032 uint64_t max_request, int rewind_flags)
3034 nvlist_t *loadinfo = NULL;
3035 nvlist_t *config = NULL;
3036 int load_error, rewind_error;
3037 uint64_t safe_rewind_txg;
3040 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
3041 spa->spa_load_max_txg = spa->spa_load_txg;
3042 spa_set_log_state(spa, SPA_LOG_CLEAR);
3044 spa->spa_load_max_txg = max_request;
3045 if (max_request != UINT64_MAX)
3046 spa->spa_extreme_rewind = B_TRUE;
3049 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
3051 if (load_error == 0)
3054 if (spa->spa_root_vdev != NULL)
3055 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
3057 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
3058 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
3060 if (rewind_flags & ZPOOL_NEVER_REWIND) {
3061 nvlist_free(config);
3062 return (load_error);
3065 if (state == SPA_LOAD_RECOVER) {
3066 /* Price of rolling back is discarding txgs, including log */
3067 spa_set_log_state(spa, SPA_LOG_CLEAR);
3070 * If we aren't rolling back save the load info from our first
3071 * import attempt so that we can restore it after attempting
3074 loadinfo = spa->spa_load_info;
3075 spa->spa_load_info = fnvlist_alloc();
3078 spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
3079 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
3080 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
3081 TXG_INITIAL : safe_rewind_txg;
3084 * Continue as long as we're finding errors, we're still within
3085 * the acceptable rewind range, and we're still finding uberblocks
3087 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
3088 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
3089 if (spa->spa_load_max_txg < safe_rewind_txg)
3090 spa->spa_extreme_rewind = B_TRUE;
3091 rewind_error = spa_load_retry(spa, state, mosconfig);
3094 spa->spa_extreme_rewind = B_FALSE;
3095 spa->spa_load_max_txg = UINT64_MAX;
3097 if (config && (rewind_error || state != SPA_LOAD_RECOVER))
3098 spa_config_set(spa, config);
3100 nvlist_free(config);
3102 if (state == SPA_LOAD_RECOVER) {
3103 ASSERT3P(loadinfo, ==, NULL);
3104 return (rewind_error);
3106 /* Store the rewind info as part of the initial load info */
3107 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
3108 spa->spa_load_info);
3110 /* Restore the initial load info */
3111 fnvlist_free(spa->spa_load_info);
3112 spa->spa_load_info = loadinfo;
3114 return (load_error);
3121 * The import case is identical to an open except that the configuration is sent
3122 * down from userland, instead of grabbed from the configuration cache. For the
3123 * case of an open, the pool configuration will exist in the
3124 * POOL_STATE_UNINITIALIZED state.
3126 * The stats information (gen/count/ustats) is used to gather vdev statistics at
3127 * the same time open the pool, without having to keep around the spa_t in some
3131 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
3135 spa_load_state_t state = SPA_LOAD_OPEN;
3137 int locked = B_FALSE;
3138 int firstopen = B_FALSE;
3143 * As disgusting as this is, we need to support recursive calls to this
3144 * function because dsl_dir_open() is called during spa_load(), and ends
3145 * up calling spa_open() again. The real fix is to figure out how to
3146 * avoid dsl_dir_open() calling this in the first place.
3148 if (mutex_owner(&spa_namespace_lock) != curthread) {
3149 mutex_enter(&spa_namespace_lock);
3153 if ((spa = spa_lookup(pool)) == NULL) {
3155 mutex_exit(&spa_namespace_lock);
3156 return (SET_ERROR(ENOENT));
3159 if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
3160 zpool_rewind_policy_t policy;
3164 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
3166 if (policy.zrp_request & ZPOOL_DO_REWIND)
3167 state = SPA_LOAD_RECOVER;
3169 spa_activate(spa, spa_mode_global);
3171 if (state != SPA_LOAD_RECOVER)
3172 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3174 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
3175 policy.zrp_request);
3177 if (error == EBADF) {
3179 * If vdev_validate() returns failure (indicated by
3180 * EBADF), it indicates that one of the vdevs indicates
3181 * that the pool has been exported or destroyed. If
3182 * this is the case, the config cache is out of sync and
3183 * we should remove the pool from the namespace.
3186 spa_deactivate(spa);
3187 spa_config_sync(spa, B_TRUE, B_TRUE);
3190 mutex_exit(&spa_namespace_lock);
3191 return (SET_ERROR(ENOENT));
3196 * We can't open the pool, but we still have useful
3197 * information: the state of each vdev after the
3198 * attempted vdev_open(). Return this to the user.
3200 if (config != NULL && spa->spa_config) {
3201 VERIFY(nvlist_dup(spa->spa_config, config,
3203 VERIFY(nvlist_add_nvlist(*config,
3204 ZPOOL_CONFIG_LOAD_INFO,
3205 spa->spa_load_info) == 0);
3208 spa_deactivate(spa);
3209 spa->spa_last_open_failed = error;
3211 mutex_exit(&spa_namespace_lock);
3217 spa_open_ref(spa, tag);
3220 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
3223 * If we've recovered the pool, pass back any information we
3224 * gathered while doing the load.
3226 if (state == SPA_LOAD_RECOVER) {
3227 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
3228 spa->spa_load_info) == 0);
3232 spa->spa_last_open_failed = 0;
3233 spa->spa_last_ubsync_txg = 0;
3234 spa->spa_load_txg = 0;
3235 mutex_exit(&spa_namespace_lock);
3239 zvol_create_minors(spa->spa_name);
3250 spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
3253 return (spa_open_common(name, spapp, tag, policy, config));
3257 spa_open(const char *name, spa_t **spapp, void *tag)
3259 return (spa_open_common(name, spapp, tag, NULL, NULL));
3263 * Lookup the given spa_t, incrementing the inject count in the process,
3264 * preventing it from being exported or destroyed.
3267 spa_inject_addref(char *name)
3271 mutex_enter(&spa_namespace_lock);
3272 if ((spa = spa_lookup(name)) == NULL) {
3273 mutex_exit(&spa_namespace_lock);
3276 spa->spa_inject_ref++;
3277 mutex_exit(&spa_namespace_lock);
3283 spa_inject_delref(spa_t *spa)
3285 mutex_enter(&spa_namespace_lock);
3286 spa->spa_inject_ref--;
3287 mutex_exit(&spa_namespace_lock);
3291 * Add spares device information to the nvlist.
3294 spa_add_spares(spa_t *spa, nvlist_t *config)
3304 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3306 if (spa->spa_spares.sav_count == 0)
3309 VERIFY(nvlist_lookup_nvlist(config,
3310 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3311 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
3312 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3314 VERIFY(nvlist_add_nvlist_array(nvroot,
3315 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3316 VERIFY(nvlist_lookup_nvlist_array(nvroot,
3317 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3320 * Go through and find any spares which have since been
3321 * repurposed as an active spare. If this is the case, update
3322 * their status appropriately.
3324 for (i = 0; i < nspares; i++) {
3325 VERIFY(nvlist_lookup_uint64(spares[i],
3326 ZPOOL_CONFIG_GUID, &guid) == 0);
3327 if (spa_spare_exists(guid, &pool, NULL) &&
3329 VERIFY(nvlist_lookup_uint64_array(
3330 spares[i], ZPOOL_CONFIG_VDEV_STATS,
3331 (uint64_t **)&vs, &vsc) == 0);
3332 vs->vs_state = VDEV_STATE_CANT_OPEN;
3333 vs->vs_aux = VDEV_AUX_SPARED;
3340 * Add l2cache device information to the nvlist, including vdev stats.
3343 spa_add_l2cache(spa_t *spa, nvlist_t *config)
3346 uint_t i, j, nl2cache;
3353 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3355 if (spa->spa_l2cache.sav_count == 0)
3358 VERIFY(nvlist_lookup_nvlist(config,
3359 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3360 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
3361 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3362 if (nl2cache != 0) {
3363 VERIFY(nvlist_add_nvlist_array(nvroot,
3364 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3365 VERIFY(nvlist_lookup_nvlist_array(nvroot,
3366 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3369 * Update level 2 cache device stats.
3372 for (i = 0; i < nl2cache; i++) {
3373 VERIFY(nvlist_lookup_uint64(l2cache[i],
3374 ZPOOL_CONFIG_GUID, &guid) == 0);
3377 for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
3379 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
3380 vd = spa->spa_l2cache.sav_vdevs[j];
3386 VERIFY(nvlist_lookup_uint64_array(l2cache[i],
3387 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
3389 vdev_get_stats(vd, vs);
3395 spa_add_feature_stats(spa_t *spa, nvlist_t *config)
3401 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3402 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3404 /* We may be unable to read features if pool is suspended. */
3405 if (spa_suspended(spa))
3408 if (spa->spa_feat_for_read_obj != 0) {
3409 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3410 spa->spa_feat_for_read_obj);
3411 zap_cursor_retrieve(&zc, &za) == 0;
3412 zap_cursor_advance(&zc)) {
3413 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3414 za.za_num_integers == 1);
3415 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3416 za.za_first_integer));
3418 zap_cursor_fini(&zc);
3421 if (spa->spa_feat_for_write_obj != 0) {
3422 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3423 spa->spa_feat_for_write_obj);
3424 zap_cursor_retrieve(&zc, &za) == 0;
3425 zap_cursor_advance(&zc)) {
3426 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3427 za.za_num_integers == 1);
3428 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3429 za.za_first_integer));
3431 zap_cursor_fini(&zc);
3435 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
3437 nvlist_free(features);
3441 spa_get_stats(const char *name, nvlist_t **config,
3442 char *altroot, size_t buflen)
3448 error = spa_open_common(name, &spa, FTAG, NULL, config);
3452 * This still leaves a window of inconsistency where the spares
3453 * or l2cache devices could change and the config would be
3454 * self-inconsistent.
3456 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3458 if (*config != NULL) {
3459 uint64_t loadtimes[2];
3461 loadtimes[0] = spa->spa_loaded_ts.tv_sec;
3462 loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
3463 VERIFY(nvlist_add_uint64_array(*config,
3464 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
3466 VERIFY(nvlist_add_uint64(*config,
3467 ZPOOL_CONFIG_ERRCOUNT,
3468 spa_get_errlog_size(spa)) == 0);
3470 if (spa_suspended(spa))
3471 VERIFY(nvlist_add_uint64(*config,
3472 ZPOOL_CONFIG_SUSPENDED,
3473 spa->spa_failmode) == 0);
3475 spa_add_spares(spa, *config);
3476 spa_add_l2cache(spa, *config);
3477 spa_add_feature_stats(spa, *config);
3482 * We want to get the alternate root even for faulted pools, so we cheat
3483 * and call spa_lookup() directly.
3487 mutex_enter(&spa_namespace_lock);
3488 spa = spa_lookup(name);
3490 spa_altroot(spa, altroot, buflen);
3494 mutex_exit(&spa_namespace_lock);
3496 spa_altroot(spa, altroot, buflen);
3501 spa_config_exit(spa, SCL_CONFIG, FTAG);
3502 spa_close(spa, FTAG);
3509 * Validate that the auxiliary device array is well formed. We must have an
3510 * array of nvlists, each which describes a valid leaf vdev. If this is an
3511 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3512 * specified, as long as they are well-formed.
3515 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
3516 spa_aux_vdev_t *sav, const char *config, uint64_t version,
3517 vdev_labeltype_t label)
3524 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3527 * It's acceptable to have no devs specified.
3529 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
3533 return (SET_ERROR(EINVAL));
3536 * Make sure the pool is formatted with a version that supports this
3539 if (spa_version(spa) < version)
3540 return (SET_ERROR(ENOTSUP));
3543 * Set the pending device list so we correctly handle device in-use
3546 sav->sav_pending = dev;
3547 sav->sav_npending = ndev;
3549 for (i = 0; i < ndev; i++) {
3550 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
3554 if (!vd->vdev_ops->vdev_op_leaf) {
3556 error = SET_ERROR(EINVAL);
3561 * The L2ARC currently only supports disk devices in
3562 * kernel context. For user-level testing, we allow it.
3565 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
3566 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
3567 error = SET_ERROR(ENOTBLK);
3574 if ((error = vdev_open(vd)) == 0 &&
3575 (error = vdev_label_init(vd, crtxg, label)) == 0) {
3576 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
3577 vd->vdev_guid) == 0);
3583 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
3590 sav->sav_pending = NULL;
3591 sav->sav_npending = 0;
3596 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
3600 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3602 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3603 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
3604 VDEV_LABEL_SPARE)) != 0) {
3608 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3609 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
3610 VDEV_LABEL_L2CACHE));
3614 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
3619 if (sav->sav_config != NULL) {
3625 * Generate new dev list by concatentating with the
3628 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
3629 &olddevs, &oldndevs) == 0);
3631 newdevs = kmem_alloc(sizeof (void *) *
3632 (ndevs + oldndevs), KM_SLEEP);
3633 for (i = 0; i < oldndevs; i++)
3634 VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
3636 for (i = 0; i < ndevs; i++)
3637 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
3640 VERIFY(nvlist_remove(sav->sav_config, config,
3641 DATA_TYPE_NVLIST_ARRAY) == 0);
3643 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3644 config, newdevs, ndevs + oldndevs) == 0);
3645 for (i = 0; i < oldndevs + ndevs; i++)
3646 nvlist_free(newdevs[i]);
3647 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
3650 * Generate a new dev list.
3652 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
3654 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
3660 * Stop and drop level 2 ARC devices
3663 spa_l2cache_drop(spa_t *spa)
3667 spa_aux_vdev_t *sav = &spa->spa_l2cache;
3669 for (i = 0; i < sav->sav_count; i++) {
3672 vd = sav->sav_vdevs[i];
3675 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
3676 pool != 0ULL && l2arc_vdev_present(vd))
3677 l2arc_remove_vdev(vd);
3685 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
3689 char *altroot = NULL;
3694 uint64_t txg = TXG_INITIAL;
3695 nvlist_t **spares, **l2cache;
3696 uint_t nspares, nl2cache;
3697 uint64_t version, obj;
3698 boolean_t has_features;
3701 * If this pool already exists, return failure.
3703 mutex_enter(&spa_namespace_lock);
3704 if (spa_lookup(pool) != NULL) {
3705 mutex_exit(&spa_namespace_lock);
3706 return (SET_ERROR(EEXIST));
3710 * Allocate a new spa_t structure.
3712 (void) nvlist_lookup_string(props,
3713 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3714 spa = spa_add(pool, NULL, altroot);
3715 spa_activate(spa, spa_mode_global);
3717 if (props && (error = spa_prop_validate(spa, props))) {
3718 spa_deactivate(spa);
3720 mutex_exit(&spa_namespace_lock);
3724 has_features = B_FALSE;
3725 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
3726 elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
3727 if (zpool_prop_feature(nvpair_name(elem)))
3728 has_features = B_TRUE;
3731 if (has_features || nvlist_lookup_uint64(props,
3732 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
3733 version = SPA_VERSION;
3735 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
3737 spa->spa_first_txg = txg;
3738 spa->spa_uberblock.ub_txg = txg - 1;
3739 spa->spa_uberblock.ub_version = version;
3740 spa->spa_ubsync = spa->spa_uberblock;
3741 spa->spa_load_state = SPA_LOAD_CREATE;
3744 * Create "The Godfather" zio to hold all async IOs
3746 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
3748 for (int i = 0; i < max_ncpus; i++) {
3749 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
3750 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
3751 ZIO_FLAG_GODFATHER);
3755 * Create the root vdev.
3757 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3759 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3761 ASSERT(error != 0 || rvd != NULL);
3762 ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3764 if (error == 0 && !zfs_allocatable_devs(nvroot))
3765 error = SET_ERROR(EINVAL);
3768 (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3769 (error = spa_validate_aux(spa, nvroot, txg,
3770 VDEV_ALLOC_ADD)) == 0) {
3771 for (int c = 0; c < rvd->vdev_children; c++) {
3772 vdev_ashift_optimize(rvd->vdev_child[c]);
3773 vdev_metaslab_set_size(rvd->vdev_child[c]);
3774 vdev_expand(rvd->vdev_child[c], txg);
3778 spa_config_exit(spa, SCL_ALL, FTAG);
3782 spa_deactivate(spa);
3784 mutex_exit(&spa_namespace_lock);
3789 * Get the list of spares, if specified.
3791 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3792 &spares, &nspares) == 0) {
3793 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3795 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3796 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3797 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3798 spa_load_spares(spa);
3799 spa_config_exit(spa, SCL_ALL, FTAG);
3800 spa->spa_spares.sav_sync = B_TRUE;
3804 * Get the list of level 2 cache devices, if specified.
3806 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3807 &l2cache, &nl2cache) == 0) {
3808 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3809 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3810 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3811 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3812 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3813 spa_load_l2cache(spa);
3814 spa_config_exit(spa, SCL_ALL, FTAG);
3815 spa->spa_l2cache.sav_sync = B_TRUE;
3818 spa->spa_is_initializing = B_TRUE;
3819 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3820 spa->spa_meta_objset = dp->dp_meta_objset;
3821 spa->spa_is_initializing = B_FALSE;
3824 * Create DDTs (dedup tables).
3828 spa_update_dspace(spa);
3830 tx = dmu_tx_create_assigned(dp, txg);
3833 * Create the pool config object.
3835 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3836 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3837 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3839 if (zap_add(spa->spa_meta_objset,
3840 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3841 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3842 cmn_err(CE_PANIC, "failed to add pool config");
3845 if (spa_version(spa) >= SPA_VERSION_FEATURES)
3846 spa_feature_create_zap_objects(spa, tx);
3848 if (zap_add(spa->spa_meta_objset,
3849 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3850 sizeof (uint64_t), 1, &version, tx) != 0) {
3851 cmn_err(CE_PANIC, "failed to add pool version");
3854 /* Newly created pools with the right version are always deflated. */
3855 if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3856 spa->spa_deflate = TRUE;
3857 if (zap_add(spa->spa_meta_objset,
3858 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3859 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3860 cmn_err(CE_PANIC, "failed to add deflate");
3865 * Create the deferred-free bpobj. Turn off compression
3866 * because sync-to-convergence takes longer if the blocksize
3869 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3870 dmu_object_set_compress(spa->spa_meta_objset, obj,
3871 ZIO_COMPRESS_OFF, tx);
3872 if (zap_add(spa->spa_meta_objset,
3873 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3874 sizeof (uint64_t), 1, &obj, tx) != 0) {
3875 cmn_err(CE_PANIC, "failed to add bpobj");
3877 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3878 spa->spa_meta_objset, obj));
3881 * Create the pool's history object.
3883 if (version >= SPA_VERSION_ZPOOL_HISTORY)
3884 spa_history_create_obj(spa, tx);
3887 * Generate some random noise for salted checksums to operate on.
3889 (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes,
3890 sizeof (spa->spa_cksum_salt.zcs_bytes));
3893 * Set pool properties.
3895 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3896 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3897 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3898 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3900 if (props != NULL) {
3901 spa_configfile_set(spa, props, B_FALSE);
3902 spa_sync_props(props, tx);
3907 spa->spa_sync_on = B_TRUE;
3908 txg_sync_start(spa->spa_dsl_pool);
3911 * We explicitly wait for the first transaction to complete so that our
3912 * bean counters are appropriately updated.
3914 txg_wait_synced(spa->spa_dsl_pool, txg);
3916 spa_config_sync(spa, B_FALSE, B_TRUE);
3917 spa_event_notify(spa, NULL, ESC_ZFS_POOL_CREATE);
3919 spa_history_log_version(spa, "create");
3922 * Don't count references from objsets that are already closed
3923 * and are making their way through the eviction process.
3925 spa_evicting_os_wait(spa);
3926 spa->spa_minref = refcount_count(&spa->spa_refcount);
3927 spa->spa_load_state = SPA_LOAD_NONE;
3929 mutex_exit(&spa_namespace_lock);
3937 * Get the root pool information from the root disk, then import the root pool
3938 * during the system boot up time.
3940 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3943 spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3946 nvlist_t *nvtop, *nvroot;
3949 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3953 * Add this top-level vdev to the child array.
3955 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3957 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3959 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3962 * Put this pool's top-level vdevs into a root vdev.
3964 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3965 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3966 VDEV_TYPE_ROOT) == 0);
3967 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3968 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3969 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3973 * Replace the existing vdev_tree with the new root vdev in
3974 * this pool's configuration (remove the old, add the new).
3976 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3977 nvlist_free(nvroot);
3982 * Walk the vdev tree and see if we can find a device with "better"
3983 * configuration. A configuration is "better" if the label on that
3984 * device has a more recent txg.
3987 spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3989 for (int c = 0; c < vd->vdev_children; c++)
3990 spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3992 if (vd->vdev_ops->vdev_op_leaf) {
3996 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
4000 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
4004 * Do we have a better boot device?
4006 if (label_txg > *txg) {
4015 * Import a root pool.
4017 * For x86. devpath_list will consist of devid and/or physpath name of
4018 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
4019 * The GRUB "findroot" command will return the vdev we should boot.
4021 * For Sparc, devpath_list consists the physpath name of the booting device
4022 * no matter the rootpool is a single device pool or a mirrored pool.
4024 * "/pci@1f,0/ide@d/disk@0,0:a"
4027 spa_import_rootpool(char *devpath, char *devid)
4030 vdev_t *rvd, *bvd, *avd = NULL;
4031 nvlist_t *config, *nvtop;
4037 * Read the label from the boot device and generate a configuration.
4039 config = spa_generate_rootconf(devpath, devid, &guid);
4040 #if defined(_OBP) && defined(_KERNEL)
4041 if (config == NULL) {
4042 if (strstr(devpath, "/iscsi/ssd") != NULL) {
4044 get_iscsi_bootpath_phy(devpath);
4045 config = spa_generate_rootconf(devpath, devid, &guid);
4049 if (config == NULL) {
4050 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
4052 return (SET_ERROR(EIO));
4055 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
4057 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
4059 mutex_enter(&spa_namespace_lock);
4060 if ((spa = spa_lookup(pname)) != NULL) {
4062 * Remove the existing root pool from the namespace so that we
4063 * can replace it with the correct config we just read in.
4068 spa = spa_add(pname, config, NULL);
4069 spa->spa_is_root = B_TRUE;
4070 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
4073 * Build up a vdev tree based on the boot device's label config.
4075 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4077 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4078 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
4079 VDEV_ALLOC_ROOTPOOL);
4080 spa_config_exit(spa, SCL_ALL, FTAG);
4082 mutex_exit(&spa_namespace_lock);
4083 nvlist_free(config);
4084 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
4090 * Get the boot vdev.
4092 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
4093 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
4094 (u_longlong_t)guid);
4095 error = SET_ERROR(ENOENT);
4100 * Determine if there is a better boot device.
4103 spa_alt_rootvdev(rvd, &avd, &txg);
4105 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
4106 "try booting from '%s'", avd->vdev_path);
4107 error = SET_ERROR(EINVAL);
4112 * If the boot device is part of a spare vdev then ensure that
4113 * we're booting off the active spare.
4115 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
4116 !bvd->vdev_isspare) {
4117 cmn_err(CE_NOTE, "The boot device is currently spared. Please "
4118 "try booting from '%s'",
4120 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
4121 error = SET_ERROR(EINVAL);
4127 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4129 spa_config_exit(spa, SCL_ALL, FTAG);
4130 mutex_exit(&spa_namespace_lock);
4132 nvlist_free(config);
4136 #else /* !illumos */
4138 extern int vdev_geom_read_pool_label(const char *name, nvlist_t ***configs,
4142 spa_generate_rootconf(const char *name)
4144 nvlist_t **configs, **tops;
4146 nvlist_t *best_cfg, *nvtop, *nvroot;
4155 if (vdev_geom_read_pool_label(name, &configs, &count) != 0)
4158 ASSERT3U(count, !=, 0);
4160 for (i = 0; i < count; i++) {
4163 VERIFY(nvlist_lookup_uint64(configs[i], ZPOOL_CONFIG_POOL_TXG,
4165 if (txg > best_txg) {
4167 best_cfg = configs[i];
4172 nvlist_lookup_uint64(best_cfg, ZPOOL_CONFIG_VDEV_CHILDREN, &nchildren);
4174 nvlist_lookup_uint64_array(best_cfg, ZPOOL_CONFIG_HOLE_ARRAY,
4177 tops = kmem_zalloc(nchildren * sizeof(void *), KM_SLEEP);
4178 for (i = 0; i < nchildren; i++) {
4181 if (configs[i] == NULL)
4183 VERIFY(nvlist_lookup_nvlist(configs[i], ZPOOL_CONFIG_VDEV_TREE,
4185 nvlist_dup(nvtop, &tops[i], KM_SLEEP);
4187 for (i = 0; holes != NULL && i < nholes; i++) {
4190 if (tops[holes[i]] != NULL)
4192 nvlist_alloc(&tops[holes[i]], NV_UNIQUE_NAME, KM_SLEEP);
4193 VERIFY(nvlist_add_string(tops[holes[i]], ZPOOL_CONFIG_TYPE,
4194 VDEV_TYPE_HOLE) == 0);
4195 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_ID,
4197 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_GUID,
4200 for (i = 0; i < nchildren; i++) {
4201 if (tops[i] != NULL)
4203 nvlist_alloc(&tops[i], NV_UNIQUE_NAME, KM_SLEEP);
4204 VERIFY(nvlist_add_string(tops[i], ZPOOL_CONFIG_TYPE,
4205 VDEV_TYPE_MISSING) == 0);
4206 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_ID,
4208 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_GUID,
4213 * Create pool config based on the best vdev config.
4215 nvlist_dup(best_cfg, &config, KM_SLEEP);
4218 * Put this pool's top-level vdevs into a root vdev.
4220 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4222 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4223 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
4224 VDEV_TYPE_ROOT) == 0);
4225 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
4226 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
4227 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
4228 tops, nchildren) == 0);
4231 * Replace the existing vdev_tree with the new root vdev in
4232 * this pool's configuration (remove the old, add the new).
4234 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
4237 * Drop vdev config elements that should not be present at pool level.
4239 nvlist_remove(config, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64);
4240 nvlist_remove(config, ZPOOL_CONFIG_TOP_GUID, DATA_TYPE_UINT64);
4242 for (i = 0; i < count; i++)
4243 nvlist_free(configs[i]);
4244 kmem_free(configs, count * sizeof(void *));
4245 for (i = 0; i < nchildren; i++)
4246 nvlist_free(tops[i]);
4247 kmem_free(tops, nchildren * sizeof(void *));
4248 nvlist_free(nvroot);
4253 spa_import_rootpool(const char *name)
4256 vdev_t *rvd, *bvd, *avd = NULL;
4257 nvlist_t *config, *nvtop;
4263 * Read the label from the boot device and generate a configuration.
4265 config = spa_generate_rootconf(name);
4267 mutex_enter(&spa_namespace_lock);
4268 if (config != NULL) {
4269 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
4270 &pname) == 0 && strcmp(name, pname) == 0);
4271 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg)
4274 if ((spa = spa_lookup(pname)) != NULL) {
4276 * The pool could already be imported,
4277 * e.g., after reboot -r.
4279 if (spa->spa_state == POOL_STATE_ACTIVE) {
4280 mutex_exit(&spa_namespace_lock);
4281 nvlist_free(config);
4286 * Remove the existing root pool from the namespace so
4287 * that we can replace it with the correct config
4292 spa = spa_add(pname, config, NULL);
4295 * Set spa_ubsync.ub_version as it can be used in vdev_alloc()
4296 * via spa_version().
4298 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
4299 &spa->spa_ubsync.ub_version) != 0)
4300 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
4301 } else if ((spa = spa_lookup(name)) == NULL) {
4302 mutex_exit(&spa_namespace_lock);
4303 nvlist_free(config);
4304 cmn_err(CE_NOTE, "Cannot find the pool label for '%s'",
4308 VERIFY(nvlist_dup(spa->spa_config, &config, KM_SLEEP) == 0);
4310 spa->spa_is_root = B_TRUE;
4311 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
4314 * Build up a vdev tree based on the boot device's label config.
4316 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4318 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4319 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
4320 VDEV_ALLOC_ROOTPOOL);
4321 spa_config_exit(spa, SCL_ALL, FTAG);
4323 mutex_exit(&spa_namespace_lock);
4324 nvlist_free(config);
4325 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
4330 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4332 spa_config_exit(spa, SCL_ALL, FTAG);
4333 mutex_exit(&spa_namespace_lock);
4335 nvlist_free(config);
4339 #endif /* illumos */
4340 #endif /* _KERNEL */
4343 * Import a non-root pool into the system.
4346 spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
4349 char *altroot = NULL;
4350 spa_load_state_t state = SPA_LOAD_IMPORT;
4351 zpool_rewind_policy_t policy;
4352 uint64_t mode = spa_mode_global;
4353 uint64_t readonly = B_FALSE;
4356 nvlist_t **spares, **l2cache;
4357 uint_t nspares, nl2cache;
4360 * If a pool with this name exists, return failure.
4362 mutex_enter(&spa_namespace_lock);
4363 if (spa_lookup(pool) != NULL) {
4364 mutex_exit(&spa_namespace_lock);
4365 return (SET_ERROR(EEXIST));
4369 * Create and initialize the spa structure.
4371 (void) nvlist_lookup_string(props,
4372 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4373 (void) nvlist_lookup_uint64(props,
4374 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
4377 spa = spa_add(pool, config, altroot);
4378 spa->spa_import_flags = flags;
4381 * Verbatim import - Take a pool and insert it into the namespace
4382 * as if it had been loaded at boot.
4384 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
4386 spa_configfile_set(spa, props, B_FALSE);
4388 spa_config_sync(spa, B_FALSE, B_TRUE);
4389 spa_event_notify(spa, NULL, ESC_ZFS_POOL_IMPORT);
4391 mutex_exit(&spa_namespace_lock);
4395 spa_activate(spa, mode);
4398 * Don't start async tasks until we know everything is healthy.
4400 spa_async_suspend(spa);
4402 zpool_get_rewind_policy(config, &policy);
4403 if (policy.zrp_request & ZPOOL_DO_REWIND)
4404 state = SPA_LOAD_RECOVER;
4407 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
4408 * because the user-supplied config is actually the one to trust when
4411 if (state != SPA_LOAD_RECOVER)
4412 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
4414 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
4415 policy.zrp_request);
4418 * Propagate anything learned while loading the pool and pass it
4419 * back to caller (i.e. rewind info, missing devices, etc).
4421 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4422 spa->spa_load_info) == 0);
4424 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4426 * Toss any existing sparelist, as it doesn't have any validity
4427 * anymore, and conflicts with spa_has_spare().
4429 if (spa->spa_spares.sav_config) {
4430 nvlist_free(spa->spa_spares.sav_config);
4431 spa->spa_spares.sav_config = NULL;
4432 spa_load_spares(spa);
4434 if (spa->spa_l2cache.sav_config) {
4435 nvlist_free(spa->spa_l2cache.sav_config);
4436 spa->spa_l2cache.sav_config = NULL;
4437 spa_load_l2cache(spa);
4440 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4443 error = spa_validate_aux(spa, nvroot, -1ULL,
4446 error = spa_validate_aux(spa, nvroot, -1ULL,
4447 VDEV_ALLOC_L2CACHE);
4448 spa_config_exit(spa, SCL_ALL, FTAG);
4451 spa_configfile_set(spa, props, B_FALSE);
4453 if (error != 0 || (props && spa_writeable(spa) &&
4454 (error = spa_prop_set(spa, props)))) {
4456 spa_deactivate(spa);
4458 mutex_exit(&spa_namespace_lock);
4462 spa_async_resume(spa);
4465 * Override any spares and level 2 cache devices as specified by
4466 * the user, as these may have correct device names/devids, etc.
4468 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
4469 &spares, &nspares) == 0) {
4470 if (spa->spa_spares.sav_config)
4471 VERIFY(nvlist_remove(spa->spa_spares.sav_config,
4472 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
4474 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
4475 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4476 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
4477 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
4478 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4479 spa_load_spares(spa);
4480 spa_config_exit(spa, SCL_ALL, FTAG);
4481 spa->spa_spares.sav_sync = B_TRUE;
4483 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
4484 &l2cache, &nl2cache) == 0) {
4485 if (spa->spa_l2cache.sav_config)
4486 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
4487 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
4489 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
4490 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4491 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
4492 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
4493 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4494 spa_load_l2cache(spa);
4495 spa_config_exit(spa, SCL_ALL, FTAG);
4496 spa->spa_l2cache.sav_sync = B_TRUE;
4500 * Check for any removed devices.
4502 if (spa->spa_autoreplace) {
4503 spa_aux_check_removed(&spa->spa_spares);
4504 spa_aux_check_removed(&spa->spa_l2cache);
4507 if (spa_writeable(spa)) {
4509 * Update the config cache to include the newly-imported pool.
4511 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4515 * It's possible that the pool was expanded while it was exported.
4516 * We kick off an async task to handle this for us.
4518 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
4520 spa_history_log_version(spa, "import");
4522 spa_event_notify(spa, NULL, ESC_ZFS_POOL_IMPORT);
4524 mutex_exit(&spa_namespace_lock);
4528 zvol_create_minors(pool);
4535 spa_tryimport(nvlist_t *tryconfig)
4537 nvlist_t *config = NULL;
4543 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
4546 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
4550 * Create and initialize the spa structure.
4552 mutex_enter(&spa_namespace_lock);
4553 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
4554 spa_activate(spa, FREAD);
4557 * Pass off the heavy lifting to spa_load().
4558 * Pass TRUE for mosconfig because the user-supplied config
4559 * is actually the one to trust when doing an import.
4561 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
4564 * If 'tryconfig' was at least parsable, return the current config.
4566 if (spa->spa_root_vdev != NULL) {
4567 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4568 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
4570 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4572 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
4573 spa->spa_uberblock.ub_timestamp) == 0);
4574 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4575 spa->spa_load_info) == 0);
4578 * If the bootfs property exists on this pool then we
4579 * copy it out so that external consumers can tell which
4580 * pools are bootable.
4582 if ((!error || error == EEXIST) && spa->spa_bootfs) {
4583 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4586 * We have to play games with the name since the
4587 * pool was opened as TRYIMPORT_NAME.
4589 if (dsl_dsobj_to_dsname(spa_name(spa),
4590 spa->spa_bootfs, tmpname) == 0) {
4592 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4594 cp = strchr(tmpname, '/');
4596 (void) strlcpy(dsname, tmpname,
4599 (void) snprintf(dsname, MAXPATHLEN,
4600 "%s/%s", poolname, ++cp);
4602 VERIFY(nvlist_add_string(config,
4603 ZPOOL_CONFIG_BOOTFS, dsname) == 0);
4604 kmem_free(dsname, MAXPATHLEN);
4606 kmem_free(tmpname, MAXPATHLEN);
4610 * Add the list of hot spares and level 2 cache devices.
4612 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4613 spa_add_spares(spa, config);
4614 spa_add_l2cache(spa, config);
4615 spa_config_exit(spa, SCL_CONFIG, FTAG);
4619 spa_deactivate(spa);
4621 mutex_exit(&spa_namespace_lock);
4627 * Pool export/destroy
4629 * The act of destroying or exporting a pool is very simple. We make sure there
4630 * is no more pending I/O and any references to the pool are gone. Then, we
4631 * update the pool state and sync all the labels to disk, removing the
4632 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4633 * we don't sync the labels or remove the configuration cache.
4636 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
4637 boolean_t force, boolean_t hardforce)
4644 if (!(spa_mode_global & FWRITE))
4645 return (SET_ERROR(EROFS));
4647 mutex_enter(&spa_namespace_lock);
4648 if ((spa = spa_lookup(pool)) == NULL) {
4649 mutex_exit(&spa_namespace_lock);
4650 return (SET_ERROR(ENOENT));
4654 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4655 * reacquire the namespace lock, and see if we can export.
4657 spa_open_ref(spa, FTAG);
4658 mutex_exit(&spa_namespace_lock);
4659 spa_async_suspend(spa);
4660 mutex_enter(&spa_namespace_lock);
4661 spa_close(spa, FTAG);
4664 * The pool will be in core if it's openable,
4665 * in which case we can modify its state.
4667 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
4669 * Objsets may be open only because they're dirty, so we
4670 * have to force it to sync before checking spa_refcnt.
4672 txg_wait_synced(spa->spa_dsl_pool, 0);
4673 spa_evicting_os_wait(spa);
4676 * A pool cannot be exported or destroyed if there are active
4677 * references. If we are resetting a pool, allow references by
4678 * fault injection handlers.
4680 if (!spa_refcount_zero(spa) ||
4681 (spa->spa_inject_ref != 0 &&
4682 new_state != POOL_STATE_UNINITIALIZED)) {
4683 spa_async_resume(spa);
4684 mutex_exit(&spa_namespace_lock);
4685 return (SET_ERROR(EBUSY));
4689 * A pool cannot be exported if it has an active shared spare.
4690 * This is to prevent other pools stealing the active spare
4691 * from an exported pool. At user's own will, such pool can
4692 * be forcedly exported.
4694 if (!force && new_state == POOL_STATE_EXPORTED &&
4695 spa_has_active_shared_spare(spa)) {
4696 spa_async_resume(spa);
4697 mutex_exit(&spa_namespace_lock);
4698 return (SET_ERROR(EXDEV));
4702 * We want this to be reflected on every label,
4703 * so mark them all dirty. spa_unload() will do the
4704 * final sync that pushes these changes out.
4706 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
4707 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4708 spa->spa_state = new_state;
4709 spa->spa_final_txg = spa_last_synced_txg(spa) +
4711 vdev_config_dirty(spa->spa_root_vdev);
4712 spa_config_exit(spa, SCL_ALL, FTAG);
4716 spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
4718 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4720 spa_deactivate(spa);
4723 if (oldconfig && spa->spa_config)
4724 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
4726 if (new_state != POOL_STATE_UNINITIALIZED) {
4728 spa_config_sync(spa, B_TRUE, B_TRUE);
4731 mutex_exit(&spa_namespace_lock);
4737 * Destroy a storage pool.
4740 spa_destroy(char *pool)
4742 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
4747 * Export a storage pool.
4750 spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
4751 boolean_t hardforce)
4753 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
4758 * Similar to spa_export(), this unloads the spa_t without actually removing it
4759 * from the namespace in any way.
4762 spa_reset(char *pool)
4764 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
4769 * ==========================================================================
4770 * Device manipulation
4771 * ==========================================================================
4775 * Add a device to a storage pool.
4778 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
4782 vdev_t *rvd = spa->spa_root_vdev;
4784 nvlist_t **spares, **l2cache;
4785 uint_t nspares, nl2cache;
4787 ASSERT(spa_writeable(spa));
4789 txg = spa_vdev_enter(spa);
4791 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
4792 VDEV_ALLOC_ADD)) != 0)
4793 return (spa_vdev_exit(spa, NULL, txg, error));
4795 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */
4797 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
4801 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
4805 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
4806 return (spa_vdev_exit(spa, vd, txg, EINVAL));
4808 if (vd->vdev_children != 0 &&
4809 (error = vdev_create(vd, txg, B_FALSE)) != 0)
4810 return (spa_vdev_exit(spa, vd, txg, error));
4813 * We must validate the spares and l2cache devices after checking the
4814 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
4816 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
4817 return (spa_vdev_exit(spa, vd, txg, error));
4820 * Transfer each new top-level vdev from vd to rvd.
4822 for (int c = 0; c < vd->vdev_children; c++) {
4825 * Set the vdev id to the first hole, if one exists.
4827 for (id = 0; id < rvd->vdev_children; id++) {
4828 if (rvd->vdev_child[id]->vdev_ishole) {
4829 vdev_free(rvd->vdev_child[id]);
4833 tvd = vd->vdev_child[c];
4834 vdev_remove_child(vd, tvd);
4836 vdev_add_child(rvd, tvd);
4837 vdev_config_dirty(tvd);
4841 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
4842 ZPOOL_CONFIG_SPARES);
4843 spa_load_spares(spa);
4844 spa->spa_spares.sav_sync = B_TRUE;
4847 if (nl2cache != 0) {
4848 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
4849 ZPOOL_CONFIG_L2CACHE);
4850 spa_load_l2cache(spa);
4851 spa->spa_l2cache.sav_sync = B_TRUE;
4855 * We have to be careful when adding new vdevs to an existing pool.
4856 * If other threads start allocating from these vdevs before we
4857 * sync the config cache, and we lose power, then upon reboot we may
4858 * fail to open the pool because there are DVAs that the config cache
4859 * can't translate. Therefore, we first add the vdevs without
4860 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4861 * and then let spa_config_update() initialize the new metaslabs.
4863 * spa_load() checks for added-but-not-initialized vdevs, so that
4864 * if we lose power at any point in this sequence, the remaining
4865 * steps will be completed the next time we load the pool.
4867 (void) spa_vdev_exit(spa, vd, txg, 0);
4869 mutex_enter(&spa_namespace_lock);
4870 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4871 spa_event_notify(spa, NULL, ESC_ZFS_VDEV_ADD);
4872 mutex_exit(&spa_namespace_lock);
4878 * Attach a device to a mirror. The arguments are the path to any device
4879 * in the mirror, and the nvroot for the new device. If the path specifies
4880 * a device that is not mirrored, we automatically insert the mirror vdev.
4882 * If 'replacing' is specified, the new device is intended to replace the
4883 * existing device; in this case the two devices are made into their own
4884 * mirror using the 'replacing' vdev, which is functionally identical to
4885 * the mirror vdev (it actually reuses all the same ops) but has a few
4886 * extra rules: you can't attach to it after it's been created, and upon
4887 * completion of resilvering, the first disk (the one being replaced)
4888 * is automatically detached.
4891 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
4893 uint64_t txg, dtl_max_txg;
4894 vdev_t *rvd = spa->spa_root_vdev;
4895 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
4897 char *oldvdpath, *newvdpath;
4901 ASSERT(spa_writeable(spa));
4903 txg = spa_vdev_enter(spa);
4905 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
4908 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4910 if (!oldvd->vdev_ops->vdev_op_leaf)
4911 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4913 pvd = oldvd->vdev_parent;
4915 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
4916 VDEV_ALLOC_ATTACH)) != 0)
4917 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4919 if (newrootvd->vdev_children != 1)
4920 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4922 newvd = newrootvd->vdev_child[0];
4924 if (!newvd->vdev_ops->vdev_op_leaf)
4925 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4927 if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
4928 return (spa_vdev_exit(spa, newrootvd, txg, error));
4931 * Spares can't replace logs
4933 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
4934 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4938 * For attach, the only allowable parent is a mirror or the root
4941 if (pvd->vdev_ops != &vdev_mirror_ops &&
4942 pvd->vdev_ops != &vdev_root_ops)
4943 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4945 pvops = &vdev_mirror_ops;
4948 * Active hot spares can only be replaced by inactive hot
4951 if (pvd->vdev_ops == &vdev_spare_ops &&
4952 oldvd->vdev_isspare &&
4953 !spa_has_spare(spa, newvd->vdev_guid))
4954 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4957 * If the source is a hot spare, and the parent isn't already a
4958 * spare, then we want to create a new hot spare. Otherwise, we
4959 * want to create a replacing vdev. The user is not allowed to
4960 * attach to a spared vdev child unless the 'isspare' state is
4961 * the same (spare replaces spare, non-spare replaces
4964 if (pvd->vdev_ops == &vdev_replacing_ops &&
4965 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
4966 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4967 } else if (pvd->vdev_ops == &vdev_spare_ops &&
4968 newvd->vdev_isspare != oldvd->vdev_isspare) {
4969 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4972 if (newvd->vdev_isspare)
4973 pvops = &vdev_spare_ops;
4975 pvops = &vdev_replacing_ops;
4979 * Make sure the new device is big enough.
4981 if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
4982 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
4985 * The new device cannot have a higher alignment requirement
4986 * than the top-level vdev.
4988 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4989 return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4992 * If this is an in-place replacement, update oldvd's path and devid
4993 * to make it distinguishable from newvd, and unopenable from now on.
4995 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4996 spa_strfree(oldvd->vdev_path);
4997 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4999 (void) sprintf(oldvd->vdev_path, "%s/%s",
5000 newvd->vdev_path, "old");
5001 if (oldvd->vdev_devid != NULL) {
5002 spa_strfree(oldvd->vdev_devid);
5003 oldvd->vdev_devid = NULL;
5007 /* mark the device being resilvered */
5008 newvd->vdev_resilver_txg = txg;
5011 * If the parent is not a mirror, or if we're replacing, insert the new
5012 * mirror/replacing/spare vdev above oldvd.
5014 if (pvd->vdev_ops != pvops)
5015 pvd = vdev_add_parent(oldvd, pvops);
5017 ASSERT(pvd->vdev_top->vdev_parent == rvd);
5018 ASSERT(pvd->vdev_ops == pvops);
5019 ASSERT(oldvd->vdev_parent == pvd);
5022 * Extract the new device from its root and add it to pvd.
5024 vdev_remove_child(newrootvd, newvd);
5025 newvd->vdev_id = pvd->vdev_children;
5026 newvd->vdev_crtxg = oldvd->vdev_crtxg;
5027 vdev_add_child(pvd, newvd);
5029 tvd = newvd->vdev_top;
5030 ASSERT(pvd->vdev_top == tvd);
5031 ASSERT(tvd->vdev_parent == rvd);
5033 vdev_config_dirty(tvd);
5036 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
5037 * for any dmu_sync-ed blocks. It will propagate upward when
5038 * spa_vdev_exit() calls vdev_dtl_reassess().
5040 dtl_max_txg = txg + TXG_CONCURRENT_STATES;
5042 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
5043 dtl_max_txg - TXG_INITIAL);
5045 if (newvd->vdev_isspare) {
5046 spa_spare_activate(newvd);
5047 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
5050 oldvdpath = spa_strdup(oldvd->vdev_path);
5051 newvdpath = spa_strdup(newvd->vdev_path);
5052 newvd_isspare = newvd->vdev_isspare;
5055 * Mark newvd's DTL dirty in this txg.
5057 vdev_dirty(tvd, VDD_DTL, newvd, txg);
5060 * Schedule the resilver to restart in the future. We do this to
5061 * ensure that dmu_sync-ed blocks have been stitched into the
5062 * respective datasets.
5064 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
5066 if (spa->spa_bootfs)
5067 spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
5069 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_ATTACH);
5074 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
5076 spa_history_log_internal(spa, "vdev attach", NULL,
5077 "%s vdev=%s %s vdev=%s",
5078 replacing && newvd_isspare ? "spare in" :
5079 replacing ? "replace" : "attach", newvdpath,
5080 replacing ? "for" : "to", oldvdpath);
5082 spa_strfree(oldvdpath);
5083 spa_strfree(newvdpath);
5089 * Detach a device from a mirror or replacing vdev.
5091 * If 'replace_done' is specified, only detach if the parent
5092 * is a replacing vdev.
5095 spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
5099 vdev_t *rvd = spa->spa_root_vdev;
5100 vdev_t *vd, *pvd, *cvd, *tvd;
5101 boolean_t unspare = B_FALSE;
5102 uint64_t unspare_guid = 0;
5105 ASSERT(spa_writeable(spa));
5107 txg = spa_vdev_enter(spa);
5109 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5112 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
5114 if (!vd->vdev_ops->vdev_op_leaf)
5115 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
5117 pvd = vd->vdev_parent;
5120 * If the parent/child relationship is not as expected, don't do it.
5121 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
5122 * vdev that's replacing B with C. The user's intent in replacing
5123 * is to go from M(A,B) to M(A,C). If the user decides to cancel
5124 * the replace by detaching C, the expected behavior is to end up
5125 * M(A,B). But suppose that right after deciding to detach C,
5126 * the replacement of B completes. We would have M(A,C), and then
5127 * ask to detach C, which would leave us with just A -- not what
5128 * the user wanted. To prevent this, we make sure that the
5129 * parent/child relationship hasn't changed -- in this example,
5130 * that C's parent is still the replacing vdev R.
5132 if (pvd->vdev_guid != pguid && pguid != 0)
5133 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
5136 * Only 'replacing' or 'spare' vdevs can be replaced.
5138 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
5139 pvd->vdev_ops != &vdev_spare_ops)
5140 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
5142 ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
5143 spa_version(spa) >= SPA_VERSION_SPARES);
5146 * Only mirror, replacing, and spare vdevs support detach.
5148 if (pvd->vdev_ops != &vdev_replacing_ops &&
5149 pvd->vdev_ops != &vdev_mirror_ops &&
5150 pvd->vdev_ops != &vdev_spare_ops)
5151 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
5154 * If this device has the only valid copy of some data,
5155 * we cannot safely detach it.
5157 if (vdev_dtl_required(vd))
5158 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
5160 ASSERT(pvd->vdev_children >= 2);
5163 * If we are detaching the second disk from a replacing vdev, then
5164 * check to see if we changed the original vdev's path to have "/old"
5165 * at the end in spa_vdev_attach(). If so, undo that change now.
5167 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
5168 vd->vdev_path != NULL) {
5169 size_t len = strlen(vd->vdev_path);
5171 for (int c = 0; c < pvd->vdev_children; c++) {
5172 cvd = pvd->vdev_child[c];
5174 if (cvd == vd || cvd->vdev_path == NULL)
5177 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
5178 strcmp(cvd->vdev_path + len, "/old") == 0) {
5179 spa_strfree(cvd->vdev_path);
5180 cvd->vdev_path = spa_strdup(vd->vdev_path);
5187 * If we are detaching the original disk from a spare, then it implies
5188 * that the spare should become a real disk, and be removed from the
5189 * active spare list for the pool.
5191 if (pvd->vdev_ops == &vdev_spare_ops &&
5193 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
5197 * Erase the disk labels so the disk can be used for other things.
5198 * This must be done after all other error cases are handled,
5199 * but before we disembowel vd (so we can still do I/O to it).
5200 * But if we can't do it, don't treat the error as fatal --
5201 * it may be that the unwritability of the disk is the reason
5202 * it's being detached!
5204 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5207 * Remove vd from its parent and compact the parent's children.
5209 vdev_remove_child(pvd, vd);
5210 vdev_compact_children(pvd);
5213 * Remember one of the remaining children so we can get tvd below.
5215 cvd = pvd->vdev_child[pvd->vdev_children - 1];
5218 * If we need to remove the remaining child from the list of hot spares,
5219 * do it now, marking the vdev as no longer a spare in the process.
5220 * We must do this before vdev_remove_parent(), because that can
5221 * change the GUID if it creates a new toplevel GUID. For a similar
5222 * reason, we must remove the spare now, in the same txg as the detach;
5223 * otherwise someone could attach a new sibling, change the GUID, and
5224 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
5227 ASSERT(cvd->vdev_isspare);
5228 spa_spare_remove(cvd);
5229 unspare_guid = cvd->vdev_guid;
5230 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
5231 cvd->vdev_unspare = B_TRUE;
5235 * If the parent mirror/replacing vdev only has one child,
5236 * the parent is no longer needed. Remove it from the tree.
5238 if (pvd->vdev_children == 1) {
5239 if (pvd->vdev_ops == &vdev_spare_ops)
5240 cvd->vdev_unspare = B_FALSE;
5241 vdev_remove_parent(cvd);
5246 * We don't set tvd until now because the parent we just removed
5247 * may have been the previous top-level vdev.
5249 tvd = cvd->vdev_top;
5250 ASSERT(tvd->vdev_parent == rvd);
5253 * Reevaluate the parent vdev state.
5255 vdev_propagate_state(cvd);
5258 * If the 'autoexpand' property is set on the pool then automatically
5259 * try to expand the size of the pool. For example if the device we
5260 * just detached was smaller than the others, it may be possible to
5261 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
5262 * first so that we can obtain the updated sizes of the leaf vdevs.
5264 if (spa->spa_autoexpand) {
5266 vdev_expand(tvd, txg);
5269 vdev_config_dirty(tvd);
5272 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
5273 * vd->vdev_detached is set and free vd's DTL object in syncing context.
5274 * But first make sure we're not on any *other* txg's DTL list, to
5275 * prevent vd from being accessed after it's freed.
5277 vdpath = spa_strdup(vd->vdev_path);
5278 for (int t = 0; t < TXG_SIZE; t++)
5279 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
5280 vd->vdev_detached = B_TRUE;
5281 vdev_dirty(tvd, VDD_DTL, vd, txg);
5283 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
5285 /* hang on to the spa before we release the lock */
5286 spa_open_ref(spa, FTAG);
5288 error = spa_vdev_exit(spa, vd, txg, 0);
5290 spa_history_log_internal(spa, "detach", NULL,
5292 spa_strfree(vdpath);
5295 * If this was the removal of the original device in a hot spare vdev,
5296 * then we want to go through and remove the device from the hot spare
5297 * list of every other pool.
5300 spa_t *altspa = NULL;
5302 mutex_enter(&spa_namespace_lock);
5303 while ((altspa = spa_next(altspa)) != NULL) {
5304 if (altspa->spa_state != POOL_STATE_ACTIVE ||
5308 spa_open_ref(altspa, FTAG);
5309 mutex_exit(&spa_namespace_lock);
5310 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
5311 mutex_enter(&spa_namespace_lock);
5312 spa_close(altspa, FTAG);
5314 mutex_exit(&spa_namespace_lock);
5316 /* search the rest of the vdevs for spares to remove */
5317 spa_vdev_resilver_done(spa);
5320 /* all done with the spa; OK to release */
5321 mutex_enter(&spa_namespace_lock);
5322 spa_close(spa, FTAG);
5323 mutex_exit(&spa_namespace_lock);
5329 * Split a set of devices from their mirrors, and create a new pool from them.
5332 spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
5333 nvlist_t *props, boolean_t exp)
5336 uint64_t txg, *glist;
5338 uint_t c, children, lastlog;
5339 nvlist_t **child, *nvl, *tmp;
5341 char *altroot = NULL;
5342 vdev_t *rvd, **vml = NULL; /* vdev modify list */
5343 boolean_t activate_slog;
5345 ASSERT(spa_writeable(spa));
5347 txg = spa_vdev_enter(spa);
5349 /* clear the log and flush everything up to now */
5350 activate_slog = spa_passivate_log(spa);
5351 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5352 error = spa_offline_log(spa);
5353 txg = spa_vdev_config_enter(spa);
5356 spa_activate_log(spa);
5359 return (spa_vdev_exit(spa, NULL, txg, error));
5361 /* check new spa name before going any further */
5362 if (spa_lookup(newname) != NULL)
5363 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
5366 * scan through all the children to ensure they're all mirrors
5368 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
5369 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
5371 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5373 /* first, check to ensure we've got the right child count */
5374 rvd = spa->spa_root_vdev;
5376 for (c = 0; c < rvd->vdev_children; c++) {
5377 vdev_t *vd = rvd->vdev_child[c];
5379 /* don't count the holes & logs as children */
5380 if (vd->vdev_islog || vd->vdev_ishole) {
5388 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
5389 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5391 /* next, ensure no spare or cache devices are part of the split */
5392 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
5393 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
5394 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5396 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
5397 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
5399 /* then, loop over each vdev and validate it */
5400 for (c = 0; c < children; c++) {
5401 uint64_t is_hole = 0;
5403 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
5407 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
5408 spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
5411 error = SET_ERROR(EINVAL);
5416 /* which disk is going to be split? */
5417 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
5419 error = SET_ERROR(EINVAL);
5423 /* look it up in the spa */
5424 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
5425 if (vml[c] == NULL) {
5426 error = SET_ERROR(ENODEV);
5430 /* make sure there's nothing stopping the split */
5431 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
5432 vml[c]->vdev_islog ||
5433 vml[c]->vdev_ishole ||
5434 vml[c]->vdev_isspare ||
5435 vml[c]->vdev_isl2cache ||
5436 !vdev_writeable(vml[c]) ||
5437 vml[c]->vdev_children != 0 ||
5438 vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
5439 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
5440 error = SET_ERROR(EINVAL);
5444 if (vdev_dtl_required(vml[c])) {
5445 error = SET_ERROR(EBUSY);
5449 /* we need certain info from the top level */
5450 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
5451 vml[c]->vdev_top->vdev_ms_array) == 0);
5452 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
5453 vml[c]->vdev_top->vdev_ms_shift) == 0);
5454 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
5455 vml[c]->vdev_top->vdev_asize) == 0);
5456 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
5457 vml[c]->vdev_top->vdev_ashift) == 0);
5459 /* transfer per-vdev ZAPs */
5460 ASSERT3U(vml[c]->vdev_leaf_zap, !=, 0);
5461 VERIFY0(nvlist_add_uint64(child[c],
5462 ZPOOL_CONFIG_VDEV_LEAF_ZAP, vml[c]->vdev_leaf_zap));
5464 ASSERT3U(vml[c]->vdev_top->vdev_top_zap, !=, 0);
5465 VERIFY0(nvlist_add_uint64(child[c],
5466 ZPOOL_CONFIG_VDEV_TOP_ZAP,
5467 vml[c]->vdev_parent->vdev_top_zap));
5471 kmem_free(vml, children * sizeof (vdev_t *));
5472 kmem_free(glist, children * sizeof (uint64_t));
5473 return (spa_vdev_exit(spa, NULL, txg, error));
5476 /* stop writers from using the disks */
5477 for (c = 0; c < children; c++) {
5479 vml[c]->vdev_offline = B_TRUE;
5481 vdev_reopen(spa->spa_root_vdev);
5484 * Temporarily record the splitting vdevs in the spa config. This
5485 * will disappear once the config is regenerated.
5487 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5488 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
5489 glist, children) == 0);
5490 kmem_free(glist, children * sizeof (uint64_t));
5492 mutex_enter(&spa->spa_props_lock);
5493 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
5495 mutex_exit(&spa->spa_props_lock);
5496 spa->spa_config_splitting = nvl;
5497 vdev_config_dirty(spa->spa_root_vdev);
5499 /* configure and create the new pool */
5500 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
5501 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
5502 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
5503 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5504 spa_version(spa)) == 0);
5505 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
5506 spa->spa_config_txg) == 0);
5507 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
5508 spa_generate_guid(NULL)) == 0);
5509 VERIFY0(nvlist_add_boolean(config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS));
5510 (void) nvlist_lookup_string(props,
5511 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
5513 /* add the new pool to the namespace */
5514 newspa = spa_add(newname, config, altroot);
5515 newspa->spa_avz_action = AVZ_ACTION_REBUILD;
5516 newspa->spa_config_txg = spa->spa_config_txg;
5517 spa_set_log_state(newspa, SPA_LOG_CLEAR);
5519 /* release the spa config lock, retaining the namespace lock */
5520 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5522 if (zio_injection_enabled)
5523 zio_handle_panic_injection(spa, FTAG, 1);
5525 spa_activate(newspa, spa_mode_global);
5526 spa_async_suspend(newspa);
5529 /* mark that we are creating new spa by splitting */
5530 newspa->spa_splitting_newspa = B_TRUE;
5532 /* create the new pool from the disks of the original pool */
5533 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
5535 newspa->spa_splitting_newspa = B_FALSE;
5540 /* if that worked, generate a real config for the new pool */
5541 if (newspa->spa_root_vdev != NULL) {
5542 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
5543 NV_UNIQUE_NAME, KM_SLEEP) == 0);
5544 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
5545 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
5546 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
5551 if (props != NULL) {
5552 spa_configfile_set(newspa, props, B_FALSE);
5553 error = spa_prop_set(newspa, props);
5558 /* flush everything */
5559 txg = spa_vdev_config_enter(newspa);
5560 vdev_config_dirty(newspa->spa_root_vdev);
5561 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
5563 if (zio_injection_enabled)
5564 zio_handle_panic_injection(spa, FTAG, 2);
5566 spa_async_resume(newspa);
5568 /* finally, update the original pool's config */
5569 txg = spa_vdev_config_enter(spa);
5570 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
5571 error = dmu_tx_assign(tx, TXG_WAIT);
5574 for (c = 0; c < children; c++) {
5575 if (vml[c] != NULL) {
5578 spa_history_log_internal(spa, "detach", tx,
5579 "vdev=%s", vml[c]->vdev_path);
5584 spa->spa_avz_action = AVZ_ACTION_REBUILD;
5585 vdev_config_dirty(spa->spa_root_vdev);
5586 spa->spa_config_splitting = NULL;
5590 (void) spa_vdev_exit(spa, NULL, txg, 0);
5592 if (zio_injection_enabled)
5593 zio_handle_panic_injection(spa, FTAG, 3);
5595 /* split is complete; log a history record */
5596 spa_history_log_internal(newspa, "split", NULL,
5597 "from pool %s", spa_name(spa));
5599 kmem_free(vml, children * sizeof (vdev_t *));
5601 /* if we're not going to mount the filesystems in userland, export */
5603 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
5610 spa_deactivate(newspa);
5613 txg = spa_vdev_config_enter(spa);
5615 /* re-online all offlined disks */
5616 for (c = 0; c < children; c++) {
5618 vml[c]->vdev_offline = B_FALSE;
5620 vdev_reopen(spa->spa_root_vdev);
5622 nvlist_free(spa->spa_config_splitting);
5623 spa->spa_config_splitting = NULL;
5624 (void) spa_vdev_exit(spa, NULL, txg, error);
5626 kmem_free(vml, children * sizeof (vdev_t *));
5631 spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
5633 for (int i = 0; i < count; i++) {
5636 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
5639 if (guid == target_guid)
5647 spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
5648 nvlist_t *dev_to_remove)
5650 nvlist_t **newdev = NULL;
5653 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
5655 for (int i = 0, j = 0; i < count; i++) {
5656 if (dev[i] == dev_to_remove)
5658 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
5661 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
5662 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
5664 for (int i = 0; i < count - 1; i++)
5665 nvlist_free(newdev[i]);
5668 kmem_free(newdev, (count - 1) * sizeof (void *));
5672 * Evacuate the device.
5675 spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
5680 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5681 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5682 ASSERT(vd == vd->vdev_top);
5685 * Evacuate the device. We don't hold the config lock as writer
5686 * since we need to do I/O but we do keep the
5687 * spa_namespace_lock held. Once this completes the device
5688 * should no longer have any blocks allocated on it.
5690 if (vd->vdev_islog) {
5691 if (vd->vdev_stat.vs_alloc != 0)
5692 error = spa_offline_log(spa);
5694 error = SET_ERROR(ENOTSUP);
5701 * The evacuation succeeded. Remove any remaining MOS metadata
5702 * associated with this vdev, and wait for these changes to sync.
5704 ASSERT0(vd->vdev_stat.vs_alloc);
5705 txg = spa_vdev_config_enter(spa);
5706 vd->vdev_removing = B_TRUE;
5707 vdev_dirty_leaves(vd, VDD_DTL, txg);
5708 vdev_config_dirty(vd);
5709 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5715 * Complete the removal by cleaning up the namespace.
5718 spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
5720 vdev_t *rvd = spa->spa_root_vdev;
5721 uint64_t id = vd->vdev_id;
5722 boolean_t last_vdev = (id == (rvd->vdev_children - 1));
5724 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5725 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5726 ASSERT(vd == vd->vdev_top);
5729 * Only remove any devices which are empty.
5731 if (vd->vdev_stat.vs_alloc != 0)
5734 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5736 if (list_link_active(&vd->vdev_state_dirty_node))
5737 vdev_state_clean(vd);
5738 if (list_link_active(&vd->vdev_config_dirty_node))
5739 vdev_config_clean(vd);
5744 vdev_compact_children(rvd);
5746 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
5747 vdev_add_child(rvd, vd);
5749 vdev_config_dirty(rvd);
5752 * Reassess the health of our root vdev.
5758 * Remove a device from the pool -
5760 * Removing a device from the vdev namespace requires several steps
5761 * and can take a significant amount of time. As a result we use
5762 * the spa_vdev_config_[enter/exit] functions which allow us to
5763 * grab and release the spa_config_lock while still holding the namespace
5764 * lock. During each step the configuration is synced out.
5766 * Currently, this supports removing only hot spares, slogs, and level 2 ARC
5770 spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
5773 sysevent_t *ev = NULL;
5774 metaslab_group_t *mg;
5775 nvlist_t **spares, **l2cache, *nv;
5777 uint_t nspares, nl2cache;
5779 boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
5781 ASSERT(spa_writeable(spa));
5784 txg = spa_vdev_enter(spa);
5786 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5788 if (spa->spa_spares.sav_vdevs != NULL &&
5789 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5790 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
5791 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
5793 * Only remove the hot spare if it's not currently in use
5796 if (vd == NULL || unspare) {
5798 vd = spa_lookup_by_guid(spa, guid, B_TRUE);
5799 ev = spa_event_create(spa, vd, ESC_ZFS_VDEV_REMOVE_AUX);
5800 spa_vdev_remove_aux(spa->spa_spares.sav_config,
5801 ZPOOL_CONFIG_SPARES, spares, nspares, nv);
5802 spa_load_spares(spa);
5803 spa->spa_spares.sav_sync = B_TRUE;
5805 error = SET_ERROR(EBUSY);
5807 } else if (spa->spa_l2cache.sav_vdevs != NULL &&
5808 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5809 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
5810 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
5812 * Cache devices can always be removed.
5814 vd = spa_lookup_by_guid(spa, guid, B_TRUE);
5815 ev = spa_event_create(spa, vd, ESC_ZFS_VDEV_REMOVE_AUX);
5816 spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
5817 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
5818 spa_load_l2cache(spa);
5819 spa->spa_l2cache.sav_sync = B_TRUE;
5820 } else if (vd != NULL && vd->vdev_islog) {
5822 ASSERT(vd == vd->vdev_top);
5827 * Stop allocating from this vdev.
5829 metaslab_group_passivate(mg);
5832 * Wait for the youngest allocations and frees to sync,
5833 * and then wait for the deferral of those frees to finish.
5835 spa_vdev_config_exit(spa, NULL,
5836 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
5839 * Attempt to evacuate the vdev.
5841 error = spa_vdev_remove_evacuate(spa, vd);
5843 txg = spa_vdev_config_enter(spa);
5846 * If we couldn't evacuate the vdev, unwind.
5849 metaslab_group_activate(mg);
5850 return (spa_vdev_exit(spa, NULL, txg, error));
5854 * Clean up the vdev namespace.
5856 ev = spa_event_create(spa, vd, ESC_ZFS_VDEV_REMOVE_DEV);
5857 spa_vdev_remove_from_namespace(spa, vd);
5859 } else if (vd != NULL) {
5861 * Normal vdevs cannot be removed (yet).
5863 error = SET_ERROR(ENOTSUP);
5866 * There is no vdev of any kind with the specified guid.
5868 error = SET_ERROR(ENOENT);
5872 error = spa_vdev_exit(spa, NULL, txg, error);
5881 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5882 * currently spared, so we can detach it.
5885 spa_vdev_resilver_done_hunt(vdev_t *vd)
5887 vdev_t *newvd, *oldvd;
5889 for (int c = 0; c < vd->vdev_children; c++) {
5890 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
5896 * Check for a completed replacement. We always consider the first
5897 * vdev in the list to be the oldest vdev, and the last one to be
5898 * the newest (see spa_vdev_attach() for how that works). In
5899 * the case where the newest vdev is faulted, we will not automatically
5900 * remove it after a resilver completes. This is OK as it will require
5901 * user intervention to determine which disk the admin wishes to keep.
5903 if (vd->vdev_ops == &vdev_replacing_ops) {
5904 ASSERT(vd->vdev_children > 1);
5906 newvd = vd->vdev_child[vd->vdev_children - 1];
5907 oldvd = vd->vdev_child[0];
5909 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
5910 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5911 !vdev_dtl_required(oldvd))
5916 * Check for a completed resilver with the 'unspare' flag set.
5918 if (vd->vdev_ops == &vdev_spare_ops) {
5919 vdev_t *first = vd->vdev_child[0];
5920 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
5922 if (last->vdev_unspare) {
5925 } else if (first->vdev_unspare) {
5932 if (oldvd != NULL &&
5933 vdev_dtl_empty(newvd, DTL_MISSING) &&
5934 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5935 !vdev_dtl_required(oldvd))
5939 * If there are more than two spares attached to a disk,
5940 * and those spares are not required, then we want to
5941 * attempt to free them up now so that they can be used
5942 * by other pools. Once we're back down to a single
5943 * disk+spare, we stop removing them.
5945 if (vd->vdev_children > 2) {
5946 newvd = vd->vdev_child[1];
5948 if (newvd->vdev_isspare && last->vdev_isspare &&
5949 vdev_dtl_empty(last, DTL_MISSING) &&
5950 vdev_dtl_empty(last, DTL_OUTAGE) &&
5951 !vdev_dtl_required(newvd))
5960 spa_vdev_resilver_done(spa_t *spa)
5962 vdev_t *vd, *pvd, *ppvd;
5963 uint64_t guid, sguid, pguid, ppguid;
5965 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5967 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
5968 pvd = vd->vdev_parent;
5969 ppvd = pvd->vdev_parent;
5970 guid = vd->vdev_guid;
5971 pguid = pvd->vdev_guid;
5972 ppguid = ppvd->vdev_guid;
5975 * If we have just finished replacing a hot spared device, then
5976 * we need to detach the parent's first child (the original hot
5979 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
5980 ppvd->vdev_children == 2) {
5981 ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
5982 sguid = ppvd->vdev_child[1]->vdev_guid;
5984 ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd));
5986 spa_config_exit(spa, SCL_ALL, FTAG);
5987 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
5989 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
5991 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5994 spa_config_exit(spa, SCL_ALL, FTAG);
5998 * Update the stored path or FRU for this vdev.
6001 spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
6005 boolean_t sync = B_FALSE;
6007 ASSERT(spa_writeable(spa));
6009 spa_vdev_state_enter(spa, SCL_ALL);
6011 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
6012 return (spa_vdev_state_exit(spa, NULL, ENOENT));
6014 if (!vd->vdev_ops->vdev_op_leaf)
6015 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
6018 if (strcmp(value, vd->vdev_path) != 0) {
6019 spa_strfree(vd->vdev_path);
6020 vd->vdev_path = spa_strdup(value);
6024 if (vd->vdev_fru == NULL) {
6025 vd->vdev_fru = spa_strdup(value);
6027 } else if (strcmp(value, vd->vdev_fru) != 0) {
6028 spa_strfree(vd->vdev_fru);
6029 vd->vdev_fru = spa_strdup(value);
6034 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
6038 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
6040 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
6044 spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
6046 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
6050 * ==========================================================================
6052 * ==========================================================================
6055 spa_scrub_pause_resume(spa_t *spa, pool_scrub_cmd_t cmd)
6057 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
6059 if (dsl_scan_resilvering(spa->spa_dsl_pool))
6060 return (SET_ERROR(EBUSY));
6062 return (dsl_scrub_set_pause_resume(spa->spa_dsl_pool, cmd));
6066 spa_scan_stop(spa_t *spa)
6068 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
6069 if (dsl_scan_resilvering(spa->spa_dsl_pool))
6070 return (SET_ERROR(EBUSY));
6071 return (dsl_scan_cancel(spa->spa_dsl_pool));
6075 spa_scan(spa_t *spa, pool_scan_func_t func)
6077 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
6079 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
6080 return (SET_ERROR(ENOTSUP));
6083 * If a resilver was requested, but there is no DTL on a
6084 * writeable leaf device, we have nothing to do.
6086 if (func == POOL_SCAN_RESILVER &&
6087 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
6088 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
6092 return (dsl_scan(spa->spa_dsl_pool, func));
6096 * ==========================================================================
6097 * SPA async task processing
6098 * ==========================================================================
6102 spa_async_remove(spa_t *spa, vdev_t *vd)
6104 if (vd->vdev_remove_wanted) {
6105 vd->vdev_remove_wanted = B_FALSE;
6106 vd->vdev_delayed_close = B_FALSE;
6107 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
6110 * We want to clear the stats, but we don't want to do a full
6111 * vdev_clear() as that will cause us to throw away
6112 * degraded/faulted state as well as attempt to reopen the
6113 * device, all of which is a waste.
6115 vd->vdev_stat.vs_read_errors = 0;
6116 vd->vdev_stat.vs_write_errors = 0;
6117 vd->vdev_stat.vs_checksum_errors = 0;
6119 vdev_state_dirty(vd->vdev_top);
6120 /* Tell userspace that the vdev is gone. */
6121 zfs_post_remove(spa, vd);
6124 for (int c = 0; c < vd->vdev_children; c++)
6125 spa_async_remove(spa, vd->vdev_child[c]);
6129 spa_async_probe(spa_t *spa, vdev_t *vd)
6131 if (vd->vdev_probe_wanted) {
6132 vd->vdev_probe_wanted = B_FALSE;
6133 vdev_reopen(vd); /* vdev_open() does the actual probe */
6136 for (int c = 0; c < vd->vdev_children; c++)
6137 spa_async_probe(spa, vd->vdev_child[c]);
6141 spa_async_autoexpand(spa_t *spa, vdev_t *vd)
6147 if (!spa->spa_autoexpand)
6150 for (int c = 0; c < vd->vdev_children; c++) {
6151 vdev_t *cvd = vd->vdev_child[c];
6152 spa_async_autoexpand(spa, cvd);
6155 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
6158 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
6159 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
6161 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
6162 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
6164 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
6165 ESC_ZFS_VDEV_AUTOEXPAND, attr, &eid, DDI_SLEEP);
6168 kmem_free(physpath, MAXPATHLEN);
6172 spa_async_thread(void *arg)
6177 ASSERT(spa->spa_sync_on);
6179 mutex_enter(&spa->spa_async_lock);
6180 tasks = spa->spa_async_tasks;
6181 spa->spa_async_tasks &= SPA_ASYNC_REMOVE;
6182 mutex_exit(&spa->spa_async_lock);
6185 * See if the config needs to be updated.
6187 if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
6188 uint64_t old_space, new_space;
6190 mutex_enter(&spa_namespace_lock);
6191 old_space = metaslab_class_get_space(spa_normal_class(spa));
6192 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
6193 new_space = metaslab_class_get_space(spa_normal_class(spa));
6194 mutex_exit(&spa_namespace_lock);
6197 * If the pool grew as a result of the config update,
6198 * then log an internal history event.
6200 if (new_space != old_space) {
6201 spa_history_log_internal(spa, "vdev online", NULL,
6202 "pool '%s' size: %llu(+%llu)",
6203 spa_name(spa), new_space, new_space - old_space);
6207 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
6208 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
6209 spa_async_autoexpand(spa, spa->spa_root_vdev);
6210 spa_config_exit(spa, SCL_CONFIG, FTAG);
6214 * See if any devices need to be probed.
6216 if (tasks & SPA_ASYNC_PROBE) {
6217 spa_vdev_state_enter(spa, SCL_NONE);
6218 spa_async_probe(spa, spa->spa_root_vdev);
6219 (void) spa_vdev_state_exit(spa, NULL, 0);
6223 * If any devices are done replacing, detach them.
6225 if (tasks & SPA_ASYNC_RESILVER_DONE)
6226 spa_vdev_resilver_done(spa);
6229 * Kick off a resilver.
6231 if (tasks & SPA_ASYNC_RESILVER)
6232 dsl_resilver_restart(spa->spa_dsl_pool, 0);
6235 * Let the world know that we're done.
6237 mutex_enter(&spa->spa_async_lock);
6238 spa->spa_async_thread = NULL;
6239 cv_broadcast(&spa->spa_async_cv);
6240 mutex_exit(&spa->spa_async_lock);
6245 spa_async_thread_vd(void *arg)
6250 mutex_enter(&spa->spa_async_lock);
6251 tasks = spa->spa_async_tasks;
6253 spa->spa_async_tasks &= ~SPA_ASYNC_REMOVE;
6254 mutex_exit(&spa->spa_async_lock);
6257 * See if any devices need to be marked REMOVED.
6259 if (tasks & SPA_ASYNC_REMOVE) {
6260 spa_vdev_state_enter(spa, SCL_NONE);
6261 spa_async_remove(spa, spa->spa_root_vdev);
6262 for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
6263 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
6264 for (int i = 0; i < spa->spa_spares.sav_count; i++)
6265 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
6266 (void) spa_vdev_state_exit(spa, NULL, 0);
6270 * Let the world know that we're done.
6272 mutex_enter(&spa->spa_async_lock);
6273 tasks = spa->spa_async_tasks;
6274 if ((tasks & SPA_ASYNC_REMOVE) != 0)
6276 spa->spa_async_thread_vd = NULL;
6277 cv_broadcast(&spa->spa_async_cv);
6278 mutex_exit(&spa->spa_async_lock);
6283 spa_async_suspend(spa_t *spa)
6285 mutex_enter(&spa->spa_async_lock);
6286 spa->spa_async_suspended++;
6287 while (spa->spa_async_thread != NULL &&
6288 spa->spa_async_thread_vd != NULL)
6289 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
6290 mutex_exit(&spa->spa_async_lock);
6294 spa_async_resume(spa_t *spa)
6296 mutex_enter(&spa->spa_async_lock);
6297 ASSERT(spa->spa_async_suspended != 0);
6298 spa->spa_async_suspended--;
6299 mutex_exit(&spa->spa_async_lock);
6303 spa_async_tasks_pending(spa_t *spa)
6305 uint_t non_config_tasks;
6307 boolean_t config_task_suspended;
6309 non_config_tasks = spa->spa_async_tasks & ~(SPA_ASYNC_CONFIG_UPDATE |
6311 config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE;
6312 if (spa->spa_ccw_fail_time == 0) {
6313 config_task_suspended = B_FALSE;
6315 config_task_suspended =
6316 (gethrtime() - spa->spa_ccw_fail_time) <
6317 (zfs_ccw_retry_interval * NANOSEC);
6320 return (non_config_tasks || (config_task && !config_task_suspended));
6324 spa_async_dispatch(spa_t *spa)
6326 mutex_enter(&spa->spa_async_lock);
6327 if (spa_async_tasks_pending(spa) &&
6328 !spa->spa_async_suspended &&
6329 spa->spa_async_thread == NULL &&
6331 spa->spa_async_thread = thread_create(NULL, 0,
6332 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
6333 mutex_exit(&spa->spa_async_lock);
6337 spa_async_dispatch_vd(spa_t *spa)
6339 mutex_enter(&spa->spa_async_lock);
6340 if ((spa->spa_async_tasks & SPA_ASYNC_REMOVE) != 0 &&
6341 !spa->spa_async_suspended &&
6342 spa->spa_async_thread_vd == NULL &&
6344 spa->spa_async_thread_vd = thread_create(NULL, 0,
6345 spa_async_thread_vd, spa, 0, &p0, TS_RUN, maxclsyspri);
6346 mutex_exit(&spa->spa_async_lock);
6350 spa_async_request(spa_t *spa, int task)
6352 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
6353 mutex_enter(&spa->spa_async_lock);
6354 spa->spa_async_tasks |= task;
6355 mutex_exit(&spa->spa_async_lock);
6356 spa_async_dispatch_vd(spa);
6360 * ==========================================================================
6361 * SPA syncing routines
6362 * ==========================================================================
6366 bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
6369 bpobj_enqueue(bpo, bp, tx);
6374 spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
6378 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
6379 BP_GET_PSIZE(bp), zio->io_flags));
6384 * Note: this simple function is not inlined to make it easier to dtrace the
6385 * amount of time spent syncing frees.
6388 spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx)
6390 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6391 bplist_iterate(bpl, spa_free_sync_cb, zio, tx);
6392 VERIFY(zio_wait(zio) == 0);
6396 * Note: this simple function is not inlined to make it easier to dtrace the
6397 * amount of time spent syncing deferred frees.
6400 spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx)
6402 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6403 VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj,
6404 spa_free_sync_cb, zio, tx), ==, 0);
6405 VERIFY0(zio_wait(zio));
6410 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
6412 char *packed = NULL;
6417 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
6420 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
6421 * information. This avoids the dmu_buf_will_dirty() path and
6422 * saves us a pre-read to get data we don't actually care about.
6424 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
6425 packed = kmem_alloc(bufsize, KM_SLEEP);
6427 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
6429 bzero(packed + nvsize, bufsize - nvsize);
6431 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
6433 kmem_free(packed, bufsize);
6435 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
6436 dmu_buf_will_dirty(db, tx);
6437 *(uint64_t *)db->db_data = nvsize;
6438 dmu_buf_rele(db, FTAG);
6442 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
6443 const char *config, const char *entry)
6453 * Update the MOS nvlist describing the list of available devices.
6454 * spa_validate_aux() will have already made sure this nvlist is
6455 * valid and the vdevs are labeled appropriately.
6457 if (sav->sav_object == 0) {
6458 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
6459 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
6460 sizeof (uint64_t), tx);
6461 VERIFY(zap_update(spa->spa_meta_objset,
6462 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
6463 &sav->sav_object, tx) == 0);
6466 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
6467 if (sav->sav_count == 0) {
6468 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
6470 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
6471 for (i = 0; i < sav->sav_count; i++)
6472 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
6473 B_FALSE, VDEV_CONFIG_L2CACHE);
6474 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
6475 sav->sav_count) == 0);
6476 for (i = 0; i < sav->sav_count; i++)
6477 nvlist_free(list[i]);
6478 kmem_free(list, sav->sav_count * sizeof (void *));
6481 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
6482 nvlist_free(nvroot);
6484 sav->sav_sync = B_FALSE;
6488 * Rebuild spa's all-vdev ZAP from the vdev ZAPs indicated in each vdev_t.
6489 * The all-vdev ZAP must be empty.
6492 spa_avz_build(vdev_t *vd, uint64_t avz, dmu_tx_t *tx)
6494 spa_t *spa = vd->vdev_spa;
6495 if (vd->vdev_top_zap != 0) {
6496 VERIFY0(zap_add_int(spa->spa_meta_objset, avz,
6497 vd->vdev_top_zap, tx));
6499 if (vd->vdev_leaf_zap != 0) {
6500 VERIFY0(zap_add_int(spa->spa_meta_objset, avz,
6501 vd->vdev_leaf_zap, tx));
6503 for (uint64_t i = 0; i < vd->vdev_children; i++) {
6504 spa_avz_build(vd->vdev_child[i], avz, tx);
6509 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
6514 * If the pool is being imported from a pre-per-vdev-ZAP version of ZFS,
6515 * its config may not be dirty but we still need to build per-vdev ZAPs.
6516 * Similarly, if the pool is being assembled (e.g. after a split), we
6517 * need to rebuild the AVZ although the config may not be dirty.
6519 if (list_is_empty(&spa->spa_config_dirty_list) &&
6520 spa->spa_avz_action == AVZ_ACTION_NONE)
6523 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6525 ASSERT(spa->spa_avz_action == AVZ_ACTION_NONE ||
6526 spa->spa_avz_action == AVZ_ACTION_INITIALIZE ||
6527 spa->spa_all_vdev_zaps != 0);
6529 if (spa->spa_avz_action == AVZ_ACTION_REBUILD) {
6530 /* Make and build the new AVZ */
6531 uint64_t new_avz = zap_create(spa->spa_meta_objset,
6532 DMU_OTN_ZAP_METADATA, DMU_OT_NONE, 0, tx);
6533 spa_avz_build(spa->spa_root_vdev, new_avz, tx);
6535 /* Diff old AVZ with new one */
6539 for (zap_cursor_init(&zc, spa->spa_meta_objset,
6540 spa->spa_all_vdev_zaps);
6541 zap_cursor_retrieve(&zc, &za) == 0;
6542 zap_cursor_advance(&zc)) {
6543 uint64_t vdzap = za.za_first_integer;
6544 if (zap_lookup_int(spa->spa_meta_objset, new_avz,
6547 * ZAP is listed in old AVZ but not in new one;
6550 VERIFY0(zap_destroy(spa->spa_meta_objset, vdzap,
6555 zap_cursor_fini(&zc);
6557 /* Destroy the old AVZ */
6558 VERIFY0(zap_destroy(spa->spa_meta_objset,
6559 spa->spa_all_vdev_zaps, tx));
6561 /* Replace the old AVZ in the dir obj with the new one */
6562 VERIFY0(zap_update(spa->spa_meta_objset,
6563 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP,
6564 sizeof (new_avz), 1, &new_avz, tx));
6566 spa->spa_all_vdev_zaps = new_avz;
6567 } else if (spa->spa_avz_action == AVZ_ACTION_DESTROY) {
6571 /* Walk through the AVZ and destroy all listed ZAPs */
6572 for (zap_cursor_init(&zc, spa->spa_meta_objset,
6573 spa->spa_all_vdev_zaps);
6574 zap_cursor_retrieve(&zc, &za) == 0;
6575 zap_cursor_advance(&zc)) {
6576 uint64_t zap = za.za_first_integer;
6577 VERIFY0(zap_destroy(spa->spa_meta_objset, zap, tx));
6580 zap_cursor_fini(&zc);
6582 /* Destroy and unlink the AVZ itself */
6583 VERIFY0(zap_destroy(spa->spa_meta_objset,
6584 spa->spa_all_vdev_zaps, tx));
6585 VERIFY0(zap_remove(spa->spa_meta_objset,
6586 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP, tx));
6587 spa->spa_all_vdev_zaps = 0;
6590 if (spa->spa_all_vdev_zaps == 0) {
6591 spa->spa_all_vdev_zaps = zap_create_link(spa->spa_meta_objset,
6592 DMU_OTN_ZAP_METADATA, DMU_POOL_DIRECTORY_OBJECT,
6593 DMU_POOL_VDEV_ZAP_MAP, tx);
6595 spa->spa_avz_action = AVZ_ACTION_NONE;
6597 /* Create ZAPs for vdevs that don't have them. */
6598 vdev_construct_zaps(spa->spa_root_vdev, tx);
6600 config = spa_config_generate(spa, spa->spa_root_vdev,
6601 dmu_tx_get_txg(tx), B_FALSE);
6604 * If we're upgrading the spa version then make sure that
6605 * the config object gets updated with the correct version.
6607 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
6608 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
6609 spa->spa_uberblock.ub_version);
6611 spa_config_exit(spa, SCL_STATE, FTAG);
6613 nvlist_free(spa->spa_config_syncing);
6614 spa->spa_config_syncing = config;
6616 spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
6620 spa_sync_version(void *arg, dmu_tx_t *tx)
6622 uint64_t *versionp = arg;
6623 uint64_t version = *versionp;
6624 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
6627 * Setting the version is special cased when first creating the pool.
6629 ASSERT(tx->tx_txg != TXG_INITIAL);
6631 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
6632 ASSERT(version >= spa_version(spa));
6634 spa->spa_uberblock.ub_version = version;
6635 vdev_config_dirty(spa->spa_root_vdev);
6636 spa_history_log_internal(spa, "set", tx, "version=%lld", version);
6640 * Set zpool properties.
6643 spa_sync_props(void *arg, dmu_tx_t *tx)
6645 nvlist_t *nvp = arg;
6646 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
6647 objset_t *mos = spa->spa_meta_objset;
6648 nvpair_t *elem = NULL;
6650 mutex_enter(&spa->spa_props_lock);
6652 while ((elem = nvlist_next_nvpair(nvp, elem))) {
6654 char *strval, *fname;
6656 const char *propname;
6657 zprop_type_t proptype;
6660 switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
6663 * We checked this earlier in spa_prop_validate().
6665 ASSERT(zpool_prop_feature(nvpair_name(elem)));
6667 fname = strchr(nvpair_name(elem), '@') + 1;
6668 VERIFY0(zfeature_lookup_name(fname, &fid));
6670 spa_feature_enable(spa, fid, tx);
6671 spa_history_log_internal(spa, "set", tx,
6672 "%s=enabled", nvpair_name(elem));
6675 case ZPOOL_PROP_VERSION:
6676 intval = fnvpair_value_uint64(elem);
6678 * The version is synced seperatly before other
6679 * properties and should be correct by now.
6681 ASSERT3U(spa_version(spa), >=, intval);
6684 case ZPOOL_PROP_ALTROOT:
6686 * 'altroot' is a non-persistent property. It should
6687 * have been set temporarily at creation or import time.
6689 ASSERT(spa->spa_root != NULL);
6692 case ZPOOL_PROP_READONLY:
6693 case ZPOOL_PROP_CACHEFILE:
6695 * 'readonly' and 'cachefile' are also non-persisitent
6699 case ZPOOL_PROP_COMMENT:
6700 strval = fnvpair_value_string(elem);
6701 if (spa->spa_comment != NULL)
6702 spa_strfree(spa->spa_comment);
6703 spa->spa_comment = spa_strdup(strval);
6705 * We need to dirty the configuration on all the vdevs
6706 * so that their labels get updated. It's unnecessary
6707 * to do this for pool creation since the vdev's
6708 * configuratoin has already been dirtied.
6710 if (tx->tx_txg != TXG_INITIAL)
6711 vdev_config_dirty(spa->spa_root_vdev);
6712 spa_history_log_internal(spa, "set", tx,
6713 "%s=%s", nvpair_name(elem), strval);
6717 * Set pool property values in the poolprops mos object.
6719 if (spa->spa_pool_props_object == 0) {
6720 spa->spa_pool_props_object =
6721 zap_create_link(mos, DMU_OT_POOL_PROPS,
6722 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
6726 /* normalize the property name */
6727 propname = zpool_prop_to_name(prop);
6728 proptype = zpool_prop_get_type(prop);
6730 if (nvpair_type(elem) == DATA_TYPE_STRING) {
6731 ASSERT(proptype == PROP_TYPE_STRING);
6732 strval = fnvpair_value_string(elem);
6733 VERIFY0(zap_update(mos,
6734 spa->spa_pool_props_object, propname,
6735 1, strlen(strval) + 1, strval, tx));
6736 spa_history_log_internal(spa, "set", tx,
6737 "%s=%s", nvpair_name(elem), strval);
6738 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
6739 intval = fnvpair_value_uint64(elem);
6741 if (proptype == PROP_TYPE_INDEX) {
6743 VERIFY0(zpool_prop_index_to_string(
6744 prop, intval, &unused));
6746 VERIFY0(zap_update(mos,
6747 spa->spa_pool_props_object, propname,
6748 8, 1, &intval, tx));
6749 spa_history_log_internal(spa, "set", tx,
6750 "%s=%lld", nvpair_name(elem), intval);
6752 ASSERT(0); /* not allowed */
6756 case ZPOOL_PROP_DELEGATION:
6757 spa->spa_delegation = intval;
6759 case ZPOOL_PROP_BOOTFS:
6760 spa->spa_bootfs = intval;
6762 case ZPOOL_PROP_FAILUREMODE:
6763 spa->spa_failmode = intval;
6765 case ZPOOL_PROP_AUTOEXPAND:
6766 spa->spa_autoexpand = intval;
6767 if (tx->tx_txg != TXG_INITIAL)
6768 spa_async_request(spa,
6769 SPA_ASYNC_AUTOEXPAND);
6771 case ZPOOL_PROP_DEDUPDITTO:
6772 spa->spa_dedup_ditto = intval;
6781 mutex_exit(&spa->spa_props_lock);
6785 * Perform one-time upgrade on-disk changes. spa_version() does not
6786 * reflect the new version this txg, so there must be no changes this
6787 * txg to anything that the upgrade code depends on after it executes.
6788 * Therefore this must be called after dsl_pool_sync() does the sync
6792 spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
6794 dsl_pool_t *dp = spa->spa_dsl_pool;
6796 ASSERT(spa->spa_sync_pass == 1);
6798 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
6800 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
6801 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
6802 dsl_pool_create_origin(dp, tx);
6804 /* Keeping the origin open increases spa_minref */
6805 spa->spa_minref += 3;
6808 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
6809 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
6810 dsl_pool_upgrade_clones(dp, tx);
6813 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
6814 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
6815 dsl_pool_upgrade_dir_clones(dp, tx);
6817 /* Keeping the freedir open increases spa_minref */
6818 spa->spa_minref += 3;
6821 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES &&
6822 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
6823 spa_feature_create_zap_objects(spa, tx);
6827 * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable
6828 * when possibility to use lz4 compression for metadata was added
6829 * Old pools that have this feature enabled must be upgraded to have
6830 * this feature active
6832 if (spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
6833 boolean_t lz4_en = spa_feature_is_enabled(spa,
6834 SPA_FEATURE_LZ4_COMPRESS);
6835 boolean_t lz4_ac = spa_feature_is_active(spa,
6836 SPA_FEATURE_LZ4_COMPRESS);
6838 if (lz4_en && !lz4_ac)
6839 spa_feature_incr(spa, SPA_FEATURE_LZ4_COMPRESS, tx);
6843 * If we haven't written the salt, do so now. Note that the
6844 * feature may not be activated yet, but that's fine since
6845 * the presence of this ZAP entry is backwards compatible.
6847 if (zap_contains(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
6848 DMU_POOL_CHECKSUM_SALT) == ENOENT) {
6849 VERIFY0(zap_add(spa->spa_meta_objset,
6850 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CHECKSUM_SALT, 1,
6851 sizeof (spa->spa_cksum_salt.zcs_bytes),
6852 spa->spa_cksum_salt.zcs_bytes, tx));
6855 rrw_exit(&dp->dp_config_rwlock, FTAG);
6859 * Sync the specified transaction group. New blocks may be dirtied as
6860 * part of the process, so we iterate until it converges.
6863 spa_sync(spa_t *spa, uint64_t txg)
6865 dsl_pool_t *dp = spa->spa_dsl_pool;
6866 objset_t *mos = spa->spa_meta_objset;
6867 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
6868 vdev_t *rvd = spa->spa_root_vdev;
6872 uint32_t max_queue_depth = zfs_vdev_async_write_max_active *
6873 zfs_vdev_queue_depth_pct / 100;
6875 VERIFY(spa_writeable(spa));
6878 * Lock out configuration changes.
6880 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
6882 spa->spa_syncing_txg = txg;
6883 spa->spa_sync_pass = 0;
6885 mutex_enter(&spa->spa_alloc_lock);
6886 VERIFY0(avl_numnodes(&spa->spa_alloc_tree));
6887 mutex_exit(&spa->spa_alloc_lock);
6890 * If there are any pending vdev state changes, convert them
6891 * into config changes that go out with this transaction group.
6893 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6894 while (list_head(&spa->spa_state_dirty_list) != NULL) {
6896 * We need the write lock here because, for aux vdevs,
6897 * calling vdev_config_dirty() modifies sav_config.
6898 * This is ugly and will become unnecessary when we
6899 * eliminate the aux vdev wart by integrating all vdevs
6900 * into the root vdev tree.
6902 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6903 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
6904 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
6905 vdev_state_clean(vd);
6906 vdev_config_dirty(vd);
6908 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6909 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
6911 spa_config_exit(spa, SCL_STATE, FTAG);
6913 tx = dmu_tx_create_assigned(dp, txg);
6915 spa->spa_sync_starttime = gethrtime();
6917 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid,
6918 spa->spa_sync_starttime + spa->spa_deadman_synctime));
6919 #else /* !illumos */
6921 callout_schedule(&spa->spa_deadman_cycid,
6922 hz * spa->spa_deadman_synctime / NANOSEC);
6924 #endif /* illumos */
6927 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
6928 * set spa_deflate if we have no raid-z vdevs.
6930 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
6931 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
6934 for (i = 0; i < rvd->vdev_children; i++) {
6935 vd = rvd->vdev_child[i];
6936 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
6939 if (i == rvd->vdev_children) {
6940 spa->spa_deflate = TRUE;
6941 VERIFY(0 == zap_add(spa->spa_meta_objset,
6942 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
6943 sizeof (uint64_t), 1, &spa->spa_deflate, tx));
6948 * Set the top-level vdev's max queue depth. Evaluate each
6949 * top-level's async write queue depth in case it changed.
6950 * The max queue depth will not change in the middle of syncing
6953 uint64_t queue_depth_total = 0;
6954 for (int c = 0; c < rvd->vdev_children; c++) {
6955 vdev_t *tvd = rvd->vdev_child[c];
6956 metaslab_group_t *mg = tvd->vdev_mg;
6958 if (mg == NULL || mg->mg_class != spa_normal_class(spa) ||
6959 !metaslab_group_initialized(mg))
6963 * It is safe to do a lock-free check here because only async
6964 * allocations look at mg_max_alloc_queue_depth, and async
6965 * allocations all happen from spa_sync().
6967 ASSERT0(refcount_count(&mg->mg_alloc_queue_depth));
6968 mg->mg_max_alloc_queue_depth = max_queue_depth;
6969 queue_depth_total += mg->mg_max_alloc_queue_depth;
6971 metaslab_class_t *mc = spa_normal_class(spa);
6972 ASSERT0(refcount_count(&mc->mc_alloc_slots));
6973 mc->mc_alloc_max_slots = queue_depth_total;
6974 mc->mc_alloc_throttle_enabled = zio_dva_throttle_enabled;
6976 ASSERT3U(mc->mc_alloc_max_slots, <=,
6977 max_queue_depth * rvd->vdev_children);
6980 * Iterate to convergence.
6983 int pass = ++spa->spa_sync_pass;
6985 spa_sync_config_object(spa, tx);
6986 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
6987 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
6988 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
6989 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
6990 spa_errlog_sync(spa, txg);
6991 dsl_pool_sync(dp, txg);
6993 if (pass < zfs_sync_pass_deferred_free) {
6994 spa_sync_frees(spa, free_bpl, tx);
6997 * We can not defer frees in pass 1, because
6998 * we sync the deferred frees later in pass 1.
7000 ASSERT3U(pass, >, 1);
7001 bplist_iterate(free_bpl, bpobj_enqueue_cb,
7002 &spa->spa_deferred_bpobj, tx);
7006 dsl_scan_sync(dp, tx);
7008 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
7012 spa_sync_upgrades(spa, tx);
7014 spa->spa_uberblock.ub_rootbp.blk_birth);
7016 * Note: We need to check if the MOS is dirty
7017 * because we could have marked the MOS dirty
7018 * without updating the uberblock (e.g. if we
7019 * have sync tasks but no dirty user data). We
7020 * need to check the uberblock's rootbp because
7021 * it is updated if we have synced out dirty
7022 * data (though in this case the MOS will most
7023 * likely also be dirty due to second order
7024 * effects, we don't want to rely on that here).
7026 if (spa->spa_uberblock.ub_rootbp.blk_birth < txg &&
7027 !dmu_objset_is_dirty(mos, txg)) {
7029 * Nothing changed on the first pass,
7030 * therefore this TXG is a no-op. Avoid
7031 * syncing deferred frees, so that we
7032 * can keep this TXG as a no-op.
7034 ASSERT(txg_list_empty(&dp->dp_dirty_datasets,
7036 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
7037 ASSERT(txg_list_empty(&dp->dp_sync_tasks, txg));
7040 spa_sync_deferred_frees(spa, tx);
7043 } while (dmu_objset_is_dirty(mos, txg));
7045 if (!list_is_empty(&spa->spa_config_dirty_list)) {
7047 * Make sure that the number of ZAPs for all the vdevs matches
7048 * the number of ZAPs in the per-vdev ZAP list. This only gets
7049 * called if the config is dirty; otherwise there may be
7050 * outstanding AVZ operations that weren't completed in
7051 * spa_sync_config_object.
7053 uint64_t all_vdev_zap_entry_count;
7054 ASSERT0(zap_count(spa->spa_meta_objset,
7055 spa->spa_all_vdev_zaps, &all_vdev_zap_entry_count));
7056 ASSERT3U(vdev_count_verify_zaps(spa->spa_root_vdev), ==,
7057 all_vdev_zap_entry_count);
7061 * Rewrite the vdev configuration (which includes the uberblock)
7062 * to commit the transaction group.
7064 * If there are no dirty vdevs, we sync the uberblock to a few
7065 * random top-level vdevs that are known to be visible in the
7066 * config cache (see spa_vdev_add() for a complete description).
7067 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
7071 * We hold SCL_STATE to prevent vdev open/close/etc.
7072 * while we're attempting to write the vdev labels.
7074 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
7076 if (list_is_empty(&spa->spa_config_dirty_list)) {
7077 vdev_t *svd[SPA_DVAS_PER_BP];
7079 int children = rvd->vdev_children;
7080 int c0 = spa_get_random(children);
7082 for (int c = 0; c < children; c++) {
7083 vd = rvd->vdev_child[(c0 + c) % children];
7084 if (vd->vdev_ms_array == 0 || vd->vdev_islog)
7086 svd[svdcount++] = vd;
7087 if (svdcount == SPA_DVAS_PER_BP)
7090 error = vdev_config_sync(svd, svdcount, txg);
7092 error = vdev_config_sync(rvd->vdev_child,
7093 rvd->vdev_children, txg);
7097 spa->spa_last_synced_guid = rvd->vdev_guid;
7099 spa_config_exit(spa, SCL_STATE, FTAG);
7103 zio_suspend(spa, NULL);
7104 zio_resume_wait(spa);
7109 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY));
7110 #else /* !illumos */
7112 callout_drain(&spa->spa_deadman_cycid);
7114 #endif /* illumos */
7117 * Clear the dirty config list.
7119 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
7120 vdev_config_clean(vd);
7123 * Now that the new config has synced transactionally,
7124 * let it become visible to the config cache.
7126 if (spa->spa_config_syncing != NULL) {
7127 spa_config_set(spa, spa->spa_config_syncing);
7128 spa->spa_config_txg = txg;
7129 spa->spa_config_syncing = NULL;
7132 dsl_pool_sync_done(dp, txg);
7134 mutex_enter(&spa->spa_alloc_lock);
7135 VERIFY0(avl_numnodes(&spa->spa_alloc_tree));
7136 mutex_exit(&spa->spa_alloc_lock);
7139 * Update usable space statistics.
7141 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
7142 vdev_sync_done(vd, txg);
7144 spa_update_dspace(spa);
7147 * It had better be the case that we didn't dirty anything
7148 * since vdev_config_sync().
7150 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
7151 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
7152 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
7154 spa->spa_sync_pass = 0;
7157 * Update the last synced uberblock here. We want to do this at
7158 * the end of spa_sync() so that consumers of spa_last_synced_txg()
7159 * will be guaranteed that all the processing associated with
7160 * that txg has been completed.
7162 spa->spa_ubsync = spa->spa_uberblock;
7163 spa_config_exit(spa, SCL_CONFIG, FTAG);
7165 spa_handle_ignored_writes(spa);
7168 * If any async tasks have been requested, kick them off.
7170 spa_async_dispatch(spa);
7171 spa_async_dispatch_vd(spa);
7175 * Sync all pools. We don't want to hold the namespace lock across these
7176 * operations, so we take a reference on the spa_t and drop the lock during the
7180 spa_sync_allpools(void)
7183 mutex_enter(&spa_namespace_lock);
7184 while ((spa = spa_next(spa)) != NULL) {
7185 if (spa_state(spa) != POOL_STATE_ACTIVE ||
7186 !spa_writeable(spa) || spa_suspended(spa))
7188 spa_open_ref(spa, FTAG);
7189 mutex_exit(&spa_namespace_lock);
7190 txg_wait_synced(spa_get_dsl(spa), 0);
7191 mutex_enter(&spa_namespace_lock);
7192 spa_close(spa, FTAG);
7194 mutex_exit(&spa_namespace_lock);
7198 * ==========================================================================
7199 * Miscellaneous routines
7200 * ==========================================================================
7204 * Remove all pools in the system.
7212 * Remove all cached state. All pools should be closed now,
7213 * so every spa in the AVL tree should be unreferenced.
7215 mutex_enter(&spa_namespace_lock);
7216 while ((spa = spa_next(NULL)) != NULL) {
7218 * Stop async tasks. The async thread may need to detach
7219 * a device that's been replaced, which requires grabbing
7220 * spa_namespace_lock, so we must drop it here.
7222 spa_open_ref(spa, FTAG);
7223 mutex_exit(&spa_namespace_lock);
7224 spa_async_suspend(spa);
7225 mutex_enter(&spa_namespace_lock);
7226 spa_close(spa, FTAG);
7228 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
7230 spa_deactivate(spa);
7234 mutex_exit(&spa_namespace_lock);
7238 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
7243 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
7247 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
7248 vd = spa->spa_l2cache.sav_vdevs[i];
7249 if (vd->vdev_guid == guid)
7253 for (i = 0; i < spa->spa_spares.sav_count; i++) {
7254 vd = spa->spa_spares.sav_vdevs[i];
7255 if (vd->vdev_guid == guid)
7264 spa_upgrade(spa_t *spa, uint64_t version)
7266 ASSERT(spa_writeable(spa));
7268 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
7271 * This should only be called for a non-faulted pool, and since a
7272 * future version would result in an unopenable pool, this shouldn't be
7275 ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version));
7276 ASSERT3U(version, >=, spa->spa_uberblock.ub_version);
7278 spa->spa_uberblock.ub_version = version;
7279 vdev_config_dirty(spa->spa_root_vdev);
7281 spa_config_exit(spa, SCL_ALL, FTAG);
7283 txg_wait_synced(spa_get_dsl(spa), 0);
7287 spa_has_spare(spa_t *spa, uint64_t guid)
7291 spa_aux_vdev_t *sav = &spa->spa_spares;
7293 for (i = 0; i < sav->sav_count; i++)
7294 if (sav->sav_vdevs[i]->vdev_guid == guid)
7297 for (i = 0; i < sav->sav_npending; i++) {
7298 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
7299 &spareguid) == 0 && spareguid == guid)
7307 * Check if a pool has an active shared spare device.
7308 * Note: reference count of an active spare is 2, as a spare and as a replace
7311 spa_has_active_shared_spare(spa_t *spa)
7315 spa_aux_vdev_t *sav = &spa->spa_spares;
7317 for (i = 0; i < sav->sav_count; i++) {
7318 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
7319 &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
7328 spa_event_create(spa_t *spa, vdev_t *vd, const char *name)
7330 sysevent_t *ev = NULL;
7332 sysevent_attr_list_t *attr = NULL;
7333 sysevent_value_t value;
7335 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
7339 value.value_type = SE_DATA_TYPE_STRING;
7340 value.value.sv_string = spa_name(spa);
7341 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
7344 value.value_type = SE_DATA_TYPE_UINT64;
7345 value.value.sv_uint64 = spa_guid(spa);
7346 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
7350 value.value_type = SE_DATA_TYPE_UINT64;
7351 value.value.sv_uint64 = vd->vdev_guid;
7352 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
7356 if (vd->vdev_path) {
7357 value.value_type = SE_DATA_TYPE_STRING;
7358 value.value.sv_string = vd->vdev_path;
7359 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
7360 &value, SE_SLEEP) != 0)
7365 if (sysevent_attach_attributes(ev, attr) != 0)
7371 sysevent_free_attr(attr);
7378 spa_event_post(sysevent_t *ev)
7383 (void) log_sysevent(ev, SE_SLEEP, &eid);
7389 * Post a sysevent corresponding to the given event. The 'name' must be one of
7390 * the event definitions in sys/sysevent/eventdefs.h. The payload will be
7391 * filled in from the spa and (optionally) the vdev. This doesn't do anything
7392 * in the userland libzpool, as we don't want consumers to misinterpret ztest
7393 * or zdb as real changes.
7396 spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
7398 spa_event_post(spa_event_create(spa, vd, name));