4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2011, 2014 by Delphix. All rights reserved.
25 * Copyright (c) 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]
33 * SPA: Storage Pool Allocator
35 * This file contains all the routines used when modifying on-disk SPA state.
36 * This includes opening, importing, destroying, exporting a pool, and syncing a
40 #include <sys/zfs_context.h>
41 #include <sys/fm/fs/zfs.h>
42 #include <sys/spa_impl.h>
44 #include <sys/zio_checksum.h>
46 #include <sys/dmu_tx.h>
50 #include <sys/vdev_impl.h>
51 #include <sys/metaslab.h>
52 #include <sys/metaslab_impl.h>
53 #include <sys/uberblock_impl.h>
56 #include <sys/dmu_traverse.h>
57 #include <sys/dmu_objset.h>
58 #include <sys/unique.h>
59 #include <sys/dsl_pool.h>
60 #include <sys/dsl_dataset.h>
61 #include <sys/dsl_dir.h>
62 #include <sys/dsl_prop.h>
63 #include <sys/dsl_synctask.h>
64 #include <sys/fs/zfs.h>
66 #include <sys/callb.h>
67 #include <sys/spa_boot.h>
68 #include <sys/zfs_ioctl.h>
69 #include <sys/dsl_scan.h>
70 #include <sys/dmu_send.h>
71 #include <sys/dsl_destroy.h>
72 #include <sys/dsl_userhold.h>
73 #include <sys/zfeature.h>
75 #include <sys/trim_map.h>
78 #include <sys/callb.h>
79 #include <sys/cpupart.h>
84 #include "zfs_comutil.h"
86 /* Check hostid on import? */
87 static int check_hostid = 1;
90 * The interval, in seconds, at which failed configuration cache file writes
93 static int zfs_ccw_retry_interval = 300;
95 SYSCTL_DECL(_vfs_zfs);
96 TUNABLE_INT("vfs.zfs.check_hostid", &check_hostid);
97 SYSCTL_INT(_vfs_zfs, OID_AUTO, check_hostid, CTLFLAG_RWTUN, &check_hostid, 0,
98 "Check hostid on import?");
99 TUNABLE_INT("vfs.zfs.ccw_retry_interval", &zfs_ccw_retry_interval);
100 SYSCTL_INT(_vfs_zfs, OID_AUTO, ccw_retry_interval, CTLFLAG_RW,
101 &zfs_ccw_retry_interval, 0,
102 "Configuration cache file write, retry after failure, interval (seconds)");
104 typedef enum zti_modes {
105 ZTI_MODE_FIXED, /* value is # of threads (min 1) */
106 ZTI_MODE_BATCH, /* cpu-intensive; value is ignored */
107 ZTI_MODE_NULL, /* don't create a taskq */
111 #define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) }
112 #define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 }
113 #define ZTI_NULL { ZTI_MODE_NULL, 0, 0 }
115 #define ZTI_N(n) ZTI_P(n, 1)
116 #define ZTI_ONE ZTI_N(1)
118 typedef struct zio_taskq_info {
119 zti_modes_t zti_mode;
124 static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
125 "issue", "issue_high", "intr", "intr_high"
129 * This table defines the taskq settings for each ZFS I/O type. When
130 * initializing a pool, we use this table to create an appropriately sized
131 * taskq. Some operations are low volume and therefore have a small, static
132 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE
133 * macros. Other operations process a large amount of data; the ZTI_BATCH
134 * macro causes us to create a taskq oriented for throughput. Some operations
135 * are so high frequency and short-lived that the taskq itself can become a a
136 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an
137 * additional degree of parallelism specified by the number of threads per-
138 * taskq and the number of taskqs; when dispatching an event in this case, the
139 * particular taskq is chosen at random.
141 * The different taskq priorities are to handle the different contexts (issue
142 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that
143 * need to be handled with minimum delay.
145 const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
146 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
147 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* NULL */
148 { ZTI_N(8), ZTI_NULL, ZTI_P(12, 8), ZTI_NULL }, /* READ */
149 { ZTI_BATCH, ZTI_N(5), ZTI_N(8), ZTI_N(5) }, /* WRITE */
150 { ZTI_P(12, 8), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FREE */
151 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* CLAIM */
152 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* IOCTL */
155 static sysevent_t *spa_event_create(spa_t *spa, vdev_t *vd, const char *name);
156 static void spa_event_post(sysevent_t *ev);
157 static void spa_sync_version(void *arg, dmu_tx_t *tx);
158 static void spa_sync_props(void *arg, dmu_tx_t *tx);
159 static boolean_t spa_has_active_shared_spare(spa_t *spa);
160 static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
161 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
163 static void spa_vdev_resilver_done(spa_t *spa);
165 uint_t zio_taskq_batch_pct = 75; /* 1 thread per cpu in pset */
167 id_t zio_taskq_psrset_bind = PS_NONE;
170 boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */
171 uint_t zio_taskq_basedc = 80; /* base duty cycle */
174 boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */
175 extern int zfs_sync_pass_deferred_free;
178 extern void spa_deadman(void *arg);
182 * This (illegal) pool name is used when temporarily importing a spa_t in order
183 * to get the vdev stats associated with the imported devices.
185 #define TRYIMPORT_NAME "$import"
188 * ==========================================================================
189 * SPA properties routines
190 * ==========================================================================
194 * Add a (source=src, propname=propval) list to an nvlist.
197 spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
198 uint64_t intval, zprop_source_t src)
200 const char *propname = zpool_prop_to_name(prop);
203 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
204 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
207 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
209 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
211 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
212 nvlist_free(propval);
216 * Get property values from the spa configuration.
219 spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
221 vdev_t *rvd = spa->spa_root_vdev;
222 dsl_pool_t *pool = spa->spa_dsl_pool;
223 uint64_t size, alloc, cap, version;
224 zprop_source_t src = ZPROP_SRC_NONE;
225 spa_config_dirent_t *dp;
226 metaslab_class_t *mc = spa_normal_class(spa);
228 ASSERT(MUTEX_HELD(&spa->spa_props_lock));
231 alloc = metaslab_class_get_alloc(spa_normal_class(spa));
232 size = metaslab_class_get_space(spa_normal_class(spa));
233 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
234 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
235 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
236 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
239 spa_prop_add_list(*nvp, ZPOOL_PROP_FRAGMENTATION, NULL,
240 metaslab_class_fragmentation(mc), src);
241 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL,
242 metaslab_class_expandable_space(mc), src);
243 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
244 (spa_mode(spa) == FREAD), src);
246 cap = (size == 0) ? 0 : (alloc * 100 / size);
247 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
249 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
250 ddt_get_pool_dedup_ratio(spa), src);
252 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
253 rvd->vdev_state, src);
255 version = spa_version(spa);
256 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
257 src = ZPROP_SRC_DEFAULT;
259 src = ZPROP_SRC_LOCAL;
260 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
265 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
266 * when opening pools before this version freedir will be NULL.
268 if (pool->dp_free_dir != NULL) {
269 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
270 dsl_dir_phys(pool->dp_free_dir)->dd_used_bytes,
273 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
277 if (pool->dp_leak_dir != NULL) {
278 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, NULL,
279 dsl_dir_phys(pool->dp_leak_dir)->dd_used_bytes,
282 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED,
287 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
289 if (spa->spa_comment != NULL) {
290 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
294 if (spa->spa_root != NULL)
295 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
298 if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) {
299 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL,
300 MIN(zfs_max_recordsize, SPA_MAXBLOCKSIZE), ZPROP_SRC_NONE);
302 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL,
303 SPA_OLD_MAXBLOCKSIZE, ZPROP_SRC_NONE);
306 if ((dp = list_head(&spa->spa_config_list)) != NULL) {
307 if (dp->scd_path == NULL) {
308 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
309 "none", 0, ZPROP_SRC_LOCAL);
310 } else if (strcmp(dp->scd_path, spa_config_path) != 0) {
311 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
312 dp->scd_path, 0, ZPROP_SRC_LOCAL);
318 * Get zpool property values.
321 spa_prop_get(spa_t *spa, nvlist_t **nvp)
323 objset_t *mos = spa->spa_meta_objset;
328 VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
330 mutex_enter(&spa->spa_props_lock);
333 * Get properties from the spa config.
335 spa_prop_get_config(spa, nvp);
337 /* If no pool property object, no more prop to get. */
338 if (mos == NULL || spa->spa_pool_props_object == 0) {
339 mutex_exit(&spa->spa_props_lock);
344 * Get properties from the MOS pool property object.
346 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
347 (err = zap_cursor_retrieve(&zc, &za)) == 0;
348 zap_cursor_advance(&zc)) {
351 zprop_source_t src = ZPROP_SRC_DEFAULT;
354 if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
357 switch (za.za_integer_length) {
359 /* integer property */
360 if (za.za_first_integer !=
361 zpool_prop_default_numeric(prop))
362 src = ZPROP_SRC_LOCAL;
364 if (prop == ZPOOL_PROP_BOOTFS) {
366 dsl_dataset_t *ds = NULL;
368 dp = spa_get_dsl(spa);
369 dsl_pool_config_enter(dp, FTAG);
370 if (err = dsl_dataset_hold_obj(dp,
371 za.za_first_integer, FTAG, &ds)) {
372 dsl_pool_config_exit(dp, FTAG);
376 strval = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN,
378 dsl_dataset_name(ds, strval);
379 dsl_dataset_rele(ds, FTAG);
380 dsl_pool_config_exit(dp, FTAG);
383 intval = za.za_first_integer;
386 spa_prop_add_list(*nvp, prop, strval, intval, src);
389 kmem_free(strval, ZFS_MAX_DATASET_NAME_LEN);
394 /* string property */
395 strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
396 err = zap_lookup(mos, spa->spa_pool_props_object,
397 za.za_name, 1, za.za_num_integers, strval);
399 kmem_free(strval, za.za_num_integers);
402 spa_prop_add_list(*nvp, prop, strval, 0, src);
403 kmem_free(strval, za.za_num_integers);
410 zap_cursor_fini(&zc);
411 mutex_exit(&spa->spa_props_lock);
413 if (err && err != ENOENT) {
423 * Validate the given pool properties nvlist and modify the list
424 * for the property values to be set.
427 spa_prop_validate(spa_t *spa, nvlist_t *props)
430 int error = 0, reset_bootfs = 0;
432 boolean_t has_feature = B_FALSE;
435 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
437 char *strval, *slash, *check, *fname;
438 const char *propname = nvpair_name(elem);
439 zpool_prop_t prop = zpool_name_to_prop(propname);
443 if (!zpool_prop_feature(propname)) {
444 error = SET_ERROR(EINVAL);
449 * Sanitize the input.
451 if (nvpair_type(elem) != DATA_TYPE_UINT64) {
452 error = SET_ERROR(EINVAL);
456 if (nvpair_value_uint64(elem, &intval) != 0) {
457 error = SET_ERROR(EINVAL);
462 error = SET_ERROR(EINVAL);
466 fname = strchr(propname, '@') + 1;
467 if (zfeature_lookup_name(fname, NULL) != 0) {
468 error = SET_ERROR(EINVAL);
472 has_feature = B_TRUE;
475 case ZPOOL_PROP_VERSION:
476 error = nvpair_value_uint64(elem, &intval);
478 (intval < spa_version(spa) ||
479 intval > SPA_VERSION_BEFORE_FEATURES ||
481 error = SET_ERROR(EINVAL);
484 case ZPOOL_PROP_DELEGATION:
485 case ZPOOL_PROP_AUTOREPLACE:
486 case ZPOOL_PROP_LISTSNAPS:
487 case ZPOOL_PROP_AUTOEXPAND:
488 error = nvpair_value_uint64(elem, &intval);
489 if (!error && intval > 1)
490 error = SET_ERROR(EINVAL);
493 case ZPOOL_PROP_BOOTFS:
495 * If the pool version is less than SPA_VERSION_BOOTFS,
496 * or the pool is still being created (version == 0),
497 * the bootfs property cannot be set.
499 if (spa_version(spa) < SPA_VERSION_BOOTFS) {
500 error = SET_ERROR(ENOTSUP);
505 * Make sure the vdev config is bootable
507 if (!vdev_is_bootable(spa->spa_root_vdev)) {
508 error = SET_ERROR(ENOTSUP);
514 error = nvpair_value_string(elem, &strval);
520 if (strval == NULL || strval[0] == '\0') {
521 objnum = zpool_prop_default_numeric(
526 if (error = dmu_objset_hold(strval, FTAG, &os))
530 * Must be ZPL, and its property settings
531 * must be supported by GRUB (compression
532 * is not gzip, and large blocks are not used).
535 if (dmu_objset_type(os) != DMU_OST_ZFS) {
536 error = SET_ERROR(ENOTSUP);
538 dsl_prop_get_int_ds(dmu_objset_ds(os),
539 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
541 !BOOTFS_COMPRESS_VALID(propval)) {
542 error = SET_ERROR(ENOTSUP);
544 dsl_prop_get_int_ds(dmu_objset_ds(os),
545 zfs_prop_to_name(ZFS_PROP_RECORDSIZE),
547 propval > SPA_OLD_MAXBLOCKSIZE) {
548 error = SET_ERROR(ENOTSUP);
550 objnum = dmu_objset_id(os);
552 dmu_objset_rele(os, FTAG);
556 case ZPOOL_PROP_FAILUREMODE:
557 error = nvpair_value_uint64(elem, &intval);
558 if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
559 intval > ZIO_FAILURE_MODE_PANIC))
560 error = SET_ERROR(EINVAL);
563 * This is a special case which only occurs when
564 * the pool has completely failed. This allows
565 * the user to change the in-core failmode property
566 * without syncing it out to disk (I/Os might
567 * currently be blocked). We do this by returning
568 * EIO to the caller (spa_prop_set) to trick it
569 * into thinking we encountered a property validation
572 if (!error && spa_suspended(spa)) {
573 spa->spa_failmode = intval;
574 error = SET_ERROR(EIO);
578 case ZPOOL_PROP_CACHEFILE:
579 if ((error = nvpair_value_string(elem, &strval)) != 0)
582 if (strval[0] == '\0')
585 if (strcmp(strval, "none") == 0)
588 if (strval[0] != '/') {
589 error = SET_ERROR(EINVAL);
593 slash = strrchr(strval, '/');
594 ASSERT(slash != NULL);
596 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
597 strcmp(slash, "/..") == 0)
598 error = SET_ERROR(EINVAL);
601 case ZPOOL_PROP_COMMENT:
602 if ((error = nvpair_value_string(elem, &strval)) != 0)
604 for (check = strval; *check != '\0'; check++) {
606 * The kernel doesn't have an easy isprint()
607 * check. For this kernel check, we merely
608 * check ASCII apart from DEL. Fix this if
609 * there is an easy-to-use kernel isprint().
611 if (*check >= 0x7f) {
612 error = SET_ERROR(EINVAL);
616 if (strlen(strval) > ZPROP_MAX_COMMENT)
620 case ZPOOL_PROP_DEDUPDITTO:
621 if (spa_version(spa) < SPA_VERSION_DEDUP)
622 error = SET_ERROR(ENOTSUP);
624 error = nvpair_value_uint64(elem, &intval);
626 intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
627 error = SET_ERROR(EINVAL);
635 if (!error && reset_bootfs) {
636 error = nvlist_remove(props,
637 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
640 error = nvlist_add_uint64(props,
641 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
649 spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
652 spa_config_dirent_t *dp;
654 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
658 dp = kmem_alloc(sizeof (spa_config_dirent_t),
661 if (cachefile[0] == '\0')
662 dp->scd_path = spa_strdup(spa_config_path);
663 else if (strcmp(cachefile, "none") == 0)
666 dp->scd_path = spa_strdup(cachefile);
668 list_insert_head(&spa->spa_config_list, dp);
670 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
674 spa_prop_set(spa_t *spa, nvlist_t *nvp)
677 nvpair_t *elem = NULL;
678 boolean_t need_sync = B_FALSE;
680 if ((error = spa_prop_validate(spa, nvp)) != 0)
683 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
684 zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem));
686 if (prop == ZPOOL_PROP_CACHEFILE ||
687 prop == ZPOOL_PROP_ALTROOT ||
688 prop == ZPOOL_PROP_READONLY)
691 if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) {
694 if (prop == ZPOOL_PROP_VERSION) {
695 VERIFY(nvpair_value_uint64(elem, &ver) == 0);
697 ASSERT(zpool_prop_feature(nvpair_name(elem)));
698 ver = SPA_VERSION_FEATURES;
702 /* Save time if the version is already set. */
703 if (ver == spa_version(spa))
707 * In addition to the pool directory object, we might
708 * create the pool properties object, the features for
709 * read object, the features for write object, or the
710 * feature descriptions object.
712 error = dsl_sync_task(spa->spa_name, NULL,
713 spa_sync_version, &ver,
714 6, ZFS_SPACE_CHECK_RESERVED);
725 return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props,
726 nvp, 6, ZFS_SPACE_CHECK_RESERVED));
733 * If the bootfs property value is dsobj, clear it.
736 spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
738 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
739 VERIFY(zap_remove(spa->spa_meta_objset,
740 spa->spa_pool_props_object,
741 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
748 spa_change_guid_check(void *arg, dmu_tx_t *tx)
750 uint64_t *newguid = arg;
751 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
752 vdev_t *rvd = spa->spa_root_vdev;
755 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
756 vdev_state = rvd->vdev_state;
757 spa_config_exit(spa, SCL_STATE, FTAG);
759 if (vdev_state != VDEV_STATE_HEALTHY)
760 return (SET_ERROR(ENXIO));
762 ASSERT3U(spa_guid(spa), !=, *newguid);
768 spa_change_guid_sync(void *arg, dmu_tx_t *tx)
770 uint64_t *newguid = arg;
771 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
773 vdev_t *rvd = spa->spa_root_vdev;
775 oldguid = spa_guid(spa);
777 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
778 rvd->vdev_guid = *newguid;
779 rvd->vdev_guid_sum += (*newguid - oldguid);
780 vdev_config_dirty(rvd);
781 spa_config_exit(spa, SCL_STATE, FTAG);
783 spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu",
788 * Change the GUID for the pool. This is done so that we can later
789 * re-import a pool built from a clone of our own vdevs. We will modify
790 * the root vdev's guid, our own pool guid, and then mark all of our
791 * vdevs dirty. Note that we must make sure that all our vdevs are
792 * online when we do this, or else any vdevs that weren't present
793 * would be orphaned from our pool. We are also going to issue a
794 * sysevent to update any watchers.
797 spa_change_guid(spa_t *spa)
802 mutex_enter(&spa->spa_vdev_top_lock);
803 mutex_enter(&spa_namespace_lock);
804 guid = spa_generate_guid(NULL);
806 error = dsl_sync_task(spa->spa_name, spa_change_guid_check,
807 spa_change_guid_sync, &guid, 5, ZFS_SPACE_CHECK_RESERVED);
810 spa_config_sync(spa, B_FALSE, B_TRUE);
811 spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID);
814 mutex_exit(&spa_namespace_lock);
815 mutex_exit(&spa->spa_vdev_top_lock);
821 * ==========================================================================
822 * SPA state manipulation (open/create/destroy/import/export)
823 * ==========================================================================
827 spa_error_entry_compare(const void *a, const void *b)
829 spa_error_entry_t *sa = (spa_error_entry_t *)a;
830 spa_error_entry_t *sb = (spa_error_entry_t *)b;
833 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
834 sizeof (zbookmark_phys_t));
845 * Utility function which retrieves copies of the current logs and
846 * re-initializes them in the process.
849 spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
851 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
853 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
854 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
856 avl_create(&spa->spa_errlist_scrub,
857 spa_error_entry_compare, sizeof (spa_error_entry_t),
858 offsetof(spa_error_entry_t, se_avl));
859 avl_create(&spa->spa_errlist_last,
860 spa_error_entry_compare, sizeof (spa_error_entry_t),
861 offsetof(spa_error_entry_t, se_avl));
865 spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
867 const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
868 enum zti_modes mode = ztip->zti_mode;
869 uint_t value = ztip->zti_value;
870 uint_t count = ztip->zti_count;
871 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
874 boolean_t batch = B_FALSE;
876 if (mode == ZTI_MODE_NULL) {
878 tqs->stqs_taskq = NULL;
882 ASSERT3U(count, >, 0);
884 tqs->stqs_count = count;
885 tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP);
889 ASSERT3U(value, >=, 1);
890 value = MAX(value, 1);
895 flags |= TASKQ_THREADS_CPU_PCT;
896 value = zio_taskq_batch_pct;
900 panic("unrecognized mode for %s_%s taskq (%u:%u) in "
902 zio_type_name[t], zio_taskq_types[q], mode, value);
906 for (uint_t i = 0; i < count; i++) {
910 (void) snprintf(name, sizeof (name), "%s_%s_%u",
911 zio_type_name[t], zio_taskq_types[q], i);
913 (void) snprintf(name, sizeof (name), "%s_%s",
914 zio_type_name[t], zio_taskq_types[q]);
918 if (zio_taskq_sysdc && spa->spa_proc != &p0) {
920 flags |= TASKQ_DC_BATCH;
922 tq = taskq_create_sysdc(name, value, 50, INT_MAX,
923 spa->spa_proc, zio_taskq_basedc, flags);
926 pri_t pri = maxclsyspri;
928 * The write issue taskq can be extremely CPU
929 * intensive. Run it at slightly lower priority
930 * than the other taskqs.
932 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,
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 * Wait for any outstanding async I/O to complete.
1307 if (spa->spa_async_zio_root != NULL) {
1308 for (int i = 0; i < max_ncpus; i++)
1309 (void) zio_wait(spa->spa_async_zio_root[i]);
1310 kmem_free(spa->spa_async_zio_root, max_ncpus * sizeof (void *));
1311 spa->spa_async_zio_root = NULL;
1314 bpobj_close(&spa->spa_deferred_bpobj);
1316 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1321 if (spa->spa_root_vdev)
1322 vdev_free(spa->spa_root_vdev);
1323 ASSERT(spa->spa_root_vdev == NULL);
1326 * Close the dsl pool.
1328 if (spa->spa_dsl_pool) {
1329 dsl_pool_close(spa->spa_dsl_pool);
1330 spa->spa_dsl_pool = NULL;
1331 spa->spa_meta_objset = NULL;
1337 * Drop and purge level 2 cache
1339 spa_l2cache_drop(spa);
1341 for (i = 0; i < spa->spa_spares.sav_count; i++)
1342 vdev_free(spa->spa_spares.sav_vdevs[i]);
1343 if (spa->spa_spares.sav_vdevs) {
1344 kmem_free(spa->spa_spares.sav_vdevs,
1345 spa->spa_spares.sav_count * sizeof (void *));
1346 spa->spa_spares.sav_vdevs = NULL;
1348 if (spa->spa_spares.sav_config) {
1349 nvlist_free(spa->spa_spares.sav_config);
1350 spa->spa_spares.sav_config = NULL;
1352 spa->spa_spares.sav_count = 0;
1354 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1355 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1356 vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1358 if (spa->spa_l2cache.sav_vdevs) {
1359 kmem_free(spa->spa_l2cache.sav_vdevs,
1360 spa->spa_l2cache.sav_count * sizeof (void *));
1361 spa->spa_l2cache.sav_vdevs = NULL;
1363 if (spa->spa_l2cache.sav_config) {
1364 nvlist_free(spa->spa_l2cache.sav_config);
1365 spa->spa_l2cache.sav_config = NULL;
1367 spa->spa_l2cache.sav_count = 0;
1369 spa->spa_async_suspended = 0;
1371 if (spa->spa_comment != NULL) {
1372 spa_strfree(spa->spa_comment);
1373 spa->spa_comment = NULL;
1376 spa_config_exit(spa, SCL_ALL, FTAG);
1380 * Load (or re-load) the current list of vdevs describing the active spares for
1381 * this pool. When this is called, we have some form of basic information in
1382 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1383 * then re-generate a more complete list including status information.
1386 spa_load_spares(spa_t *spa)
1393 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1396 * First, close and free any existing spare vdevs.
1398 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1399 vd = spa->spa_spares.sav_vdevs[i];
1401 /* Undo the call to spa_activate() below */
1402 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1403 B_FALSE)) != NULL && tvd->vdev_isspare)
1404 spa_spare_remove(tvd);
1409 if (spa->spa_spares.sav_vdevs)
1410 kmem_free(spa->spa_spares.sav_vdevs,
1411 spa->spa_spares.sav_count * sizeof (void *));
1413 if (spa->spa_spares.sav_config == NULL)
1416 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1417 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1419 spa->spa_spares.sav_count = (int)nspares;
1420 spa->spa_spares.sav_vdevs = NULL;
1426 * Construct the array of vdevs, opening them to get status in the
1427 * process. For each spare, there is potentially two different vdev_t
1428 * structures associated with it: one in the list of spares (used only
1429 * for basic validation purposes) and one in the active vdev
1430 * configuration (if it's spared in). During this phase we open and
1431 * validate each vdev on the spare list. If the vdev also exists in the
1432 * active configuration, then we also mark this vdev as an active spare.
1434 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1436 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1437 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1438 VDEV_ALLOC_SPARE) == 0);
1441 spa->spa_spares.sav_vdevs[i] = vd;
1443 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1444 B_FALSE)) != NULL) {
1445 if (!tvd->vdev_isspare)
1449 * We only mark the spare active if we were successfully
1450 * able to load the vdev. Otherwise, importing a pool
1451 * with a bad active spare would result in strange
1452 * behavior, because multiple pool would think the spare
1453 * is actively in use.
1455 * There is a vulnerability here to an equally bizarre
1456 * circumstance, where a dead active spare is later
1457 * brought back to life (onlined or otherwise). Given
1458 * the rarity of this scenario, and the extra complexity
1459 * it adds, we ignore the possibility.
1461 if (!vdev_is_dead(tvd))
1462 spa_spare_activate(tvd);
1466 vd->vdev_aux = &spa->spa_spares;
1468 if (vdev_open(vd) != 0)
1471 if (vdev_validate_aux(vd) == 0)
1476 * Recompute the stashed list of spares, with status information
1479 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1480 DATA_TYPE_NVLIST_ARRAY) == 0);
1482 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1484 for (i = 0; i < spa->spa_spares.sav_count; i++)
1485 spares[i] = vdev_config_generate(spa,
1486 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1487 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1488 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1489 for (i = 0; i < spa->spa_spares.sav_count; i++)
1490 nvlist_free(spares[i]);
1491 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1495 * Load (or re-load) the current list of vdevs describing the active l2cache for
1496 * this pool. When this is called, we have some form of basic information in
1497 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1498 * then re-generate a more complete list including status information.
1499 * Devices which are already active have their details maintained, and are
1503 spa_load_l2cache(spa_t *spa)
1507 int i, j, oldnvdevs;
1509 vdev_t *vd, **oldvdevs, **newvdevs;
1510 spa_aux_vdev_t *sav = &spa->spa_l2cache;
1512 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1514 if (sav->sav_config != NULL) {
1515 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1516 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1517 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1523 oldvdevs = sav->sav_vdevs;
1524 oldnvdevs = sav->sav_count;
1525 sav->sav_vdevs = NULL;
1529 * Process new nvlist of vdevs.
1531 for (i = 0; i < nl2cache; i++) {
1532 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1536 for (j = 0; j < oldnvdevs; j++) {
1538 if (vd != NULL && guid == vd->vdev_guid) {
1540 * Retain previous vdev for add/remove ops.
1548 if (newvdevs[i] == NULL) {
1552 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1553 VDEV_ALLOC_L2CACHE) == 0);
1558 * Commit this vdev as an l2cache device,
1559 * even if it fails to open.
1561 spa_l2cache_add(vd);
1566 spa_l2cache_activate(vd);
1568 if (vdev_open(vd) != 0)
1571 (void) vdev_validate_aux(vd);
1573 if (!vdev_is_dead(vd))
1574 l2arc_add_vdev(spa, vd);
1579 * Purge vdevs that were dropped
1581 for (i = 0; i < oldnvdevs; i++) {
1586 ASSERT(vd->vdev_isl2cache);
1588 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1589 pool != 0ULL && l2arc_vdev_present(vd))
1590 l2arc_remove_vdev(vd);
1591 vdev_clear_stats(vd);
1597 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1599 if (sav->sav_config == NULL)
1602 sav->sav_vdevs = newvdevs;
1603 sav->sav_count = (int)nl2cache;
1606 * Recompute the stashed list of l2cache devices, with status
1607 * information this time.
1609 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1610 DATA_TYPE_NVLIST_ARRAY) == 0);
1612 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1613 for (i = 0; i < sav->sav_count; i++)
1614 l2cache[i] = vdev_config_generate(spa,
1615 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1616 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1617 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1619 for (i = 0; i < sav->sav_count; i++)
1620 nvlist_free(l2cache[i]);
1622 kmem_free(l2cache, sav->sav_count * sizeof (void *));
1626 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1629 char *packed = NULL;
1634 error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db);
1638 nvsize = *(uint64_t *)db->db_data;
1639 dmu_buf_rele(db, FTAG);
1641 packed = kmem_alloc(nvsize, KM_SLEEP);
1642 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1645 error = nvlist_unpack(packed, nvsize, value, 0);
1646 kmem_free(packed, nvsize);
1652 * Checks to see if the given vdev could not be opened, in which case we post a
1653 * sysevent to notify the autoreplace code that the device has been removed.
1656 spa_check_removed(vdev_t *vd)
1658 for (int c = 0; c < vd->vdev_children; c++)
1659 spa_check_removed(vd->vdev_child[c]);
1661 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) &&
1663 zfs_post_autoreplace(vd->vdev_spa, vd);
1664 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1669 * Validate the current config against the MOS config
1672 spa_config_valid(spa_t *spa, nvlist_t *config)
1674 vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1677 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1679 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1680 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1682 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1685 * If we're doing a normal import, then build up any additional
1686 * diagnostic information about missing devices in this config.
1687 * We'll pass this up to the user for further processing.
1689 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1690 nvlist_t **child, *nv;
1693 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1695 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1697 for (int c = 0; c < rvd->vdev_children; c++) {
1698 vdev_t *tvd = rvd->vdev_child[c];
1699 vdev_t *mtvd = mrvd->vdev_child[c];
1701 if (tvd->vdev_ops == &vdev_missing_ops &&
1702 mtvd->vdev_ops != &vdev_missing_ops &&
1704 child[idx++] = vdev_config_generate(spa, mtvd,
1709 VERIFY(nvlist_add_nvlist_array(nv,
1710 ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1711 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1712 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1714 for (int i = 0; i < idx; i++)
1715 nvlist_free(child[i]);
1718 kmem_free(child, rvd->vdev_children * sizeof (char **));
1722 * Compare the root vdev tree with the information we have
1723 * from the MOS config (mrvd). Check each top-level vdev
1724 * with the corresponding MOS config top-level (mtvd).
1726 for (int c = 0; c < rvd->vdev_children; c++) {
1727 vdev_t *tvd = rvd->vdev_child[c];
1728 vdev_t *mtvd = mrvd->vdev_child[c];
1731 * Resolve any "missing" vdevs in the current configuration.
1732 * If we find that the MOS config has more accurate information
1733 * about the top-level vdev then use that vdev instead.
1735 if (tvd->vdev_ops == &vdev_missing_ops &&
1736 mtvd->vdev_ops != &vdev_missing_ops) {
1738 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1742 * Device specific actions.
1744 if (mtvd->vdev_islog) {
1745 spa_set_log_state(spa, SPA_LOG_CLEAR);
1748 * XXX - once we have 'readonly' pool
1749 * support we should be able to handle
1750 * missing data devices by transitioning
1751 * the pool to readonly.
1757 * Swap the missing vdev with the data we were
1758 * able to obtain from the MOS config.
1760 vdev_remove_child(rvd, tvd);
1761 vdev_remove_child(mrvd, mtvd);
1763 vdev_add_child(rvd, mtvd);
1764 vdev_add_child(mrvd, tvd);
1766 spa_config_exit(spa, SCL_ALL, FTAG);
1768 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1771 } else if (mtvd->vdev_islog) {
1773 * Load the slog device's state from the MOS config
1774 * since it's possible that the label does not
1775 * contain the most up-to-date information.
1777 vdev_load_log_state(tvd, mtvd);
1782 spa_config_exit(spa, SCL_ALL, FTAG);
1785 * Ensure we were able to validate the config.
1787 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1791 * Check for missing log devices
1794 spa_check_logs(spa_t *spa)
1796 boolean_t rv = B_FALSE;
1797 dsl_pool_t *dp = spa_get_dsl(spa);
1799 switch (spa->spa_log_state) {
1800 case SPA_LOG_MISSING:
1801 /* need to recheck in case slog has been restored */
1802 case SPA_LOG_UNKNOWN:
1803 rv = (dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
1804 zil_check_log_chain, NULL, DS_FIND_CHILDREN) != 0);
1806 spa_set_log_state(spa, SPA_LOG_MISSING);
1813 spa_passivate_log(spa_t *spa)
1815 vdev_t *rvd = spa->spa_root_vdev;
1816 boolean_t slog_found = B_FALSE;
1818 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1820 if (!spa_has_slogs(spa))
1823 for (int c = 0; c < rvd->vdev_children; c++) {
1824 vdev_t *tvd = rvd->vdev_child[c];
1825 metaslab_group_t *mg = tvd->vdev_mg;
1827 if (tvd->vdev_islog) {
1828 metaslab_group_passivate(mg);
1829 slog_found = B_TRUE;
1833 return (slog_found);
1837 spa_activate_log(spa_t *spa)
1839 vdev_t *rvd = spa->spa_root_vdev;
1841 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1843 for (int c = 0; c < rvd->vdev_children; c++) {
1844 vdev_t *tvd = rvd->vdev_child[c];
1845 metaslab_group_t *mg = tvd->vdev_mg;
1847 if (tvd->vdev_islog)
1848 metaslab_group_activate(mg);
1853 spa_offline_log(spa_t *spa)
1857 error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1858 NULL, DS_FIND_CHILDREN);
1861 * We successfully offlined the log device, sync out the
1862 * current txg so that the "stubby" block can be removed
1865 txg_wait_synced(spa->spa_dsl_pool, 0);
1871 spa_aux_check_removed(spa_aux_vdev_t *sav)
1875 for (i = 0; i < sav->sav_count; i++)
1876 spa_check_removed(sav->sav_vdevs[i]);
1880 spa_claim_notify(zio_t *zio)
1882 spa_t *spa = zio->io_spa;
1887 mutex_enter(&spa->spa_props_lock); /* any mutex will do */
1888 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1889 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1890 mutex_exit(&spa->spa_props_lock);
1893 typedef struct spa_load_error {
1894 uint64_t sle_meta_count;
1895 uint64_t sle_data_count;
1899 spa_load_verify_done(zio_t *zio)
1901 blkptr_t *bp = zio->io_bp;
1902 spa_load_error_t *sle = zio->io_private;
1903 dmu_object_type_t type = BP_GET_TYPE(bp);
1904 int error = zio->io_error;
1905 spa_t *spa = zio->io_spa;
1908 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1909 type != DMU_OT_INTENT_LOG)
1910 atomic_inc_64(&sle->sle_meta_count);
1912 atomic_inc_64(&sle->sle_data_count);
1914 zio_data_buf_free(zio->io_data, zio->io_size);
1916 mutex_enter(&spa->spa_scrub_lock);
1917 spa->spa_scrub_inflight--;
1918 cv_broadcast(&spa->spa_scrub_io_cv);
1919 mutex_exit(&spa->spa_scrub_lock);
1923 * Maximum number of concurrent scrub i/os to create while verifying
1924 * a pool while importing it.
1926 int spa_load_verify_maxinflight = 10000;
1927 boolean_t spa_load_verify_metadata = B_TRUE;
1928 boolean_t spa_load_verify_data = B_TRUE;
1930 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_maxinflight, CTLFLAG_RWTUN,
1931 &spa_load_verify_maxinflight, 0,
1932 "Maximum number of concurrent scrub I/Os to create while verifying a "
1933 "pool while importing it");
1935 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_metadata, CTLFLAG_RWTUN,
1936 &spa_load_verify_metadata, 0,
1937 "Check metadata on import?");
1939 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_data, CTLFLAG_RWTUN,
1940 &spa_load_verify_data, 0,
1941 "Check user data on import?");
1945 spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1946 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
1948 if (bp == NULL || BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp))
1951 * Note: normally this routine will not be called if
1952 * spa_load_verify_metadata is not set. However, it may be useful
1953 * to manually set the flag after the traversal has begun.
1955 if (!spa_load_verify_metadata)
1957 if (BP_GET_BUFC_TYPE(bp) == ARC_BUFC_DATA && !spa_load_verify_data)
1961 size_t size = BP_GET_PSIZE(bp);
1962 void *data = zio_data_buf_alloc(size);
1964 mutex_enter(&spa->spa_scrub_lock);
1965 while (spa->spa_scrub_inflight >= spa_load_verify_maxinflight)
1966 cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
1967 spa->spa_scrub_inflight++;
1968 mutex_exit(&spa->spa_scrub_lock);
1970 zio_nowait(zio_read(rio, spa, bp, data, size,
1971 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1972 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1973 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1979 verify_dataset_name_len(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg)
1981 if (dsl_dataset_namelen(ds) >= ZFS_MAX_DATASET_NAME_LEN)
1982 return (SET_ERROR(ENAMETOOLONG));
1988 spa_load_verify(spa_t *spa)
1991 spa_load_error_t sle = { 0 };
1992 zpool_rewind_policy_t policy;
1993 boolean_t verify_ok = B_FALSE;
1996 zpool_get_rewind_policy(spa->spa_config, &policy);
1998 if (policy.zrp_request & ZPOOL_NEVER_REWIND)
2001 dsl_pool_config_enter(spa->spa_dsl_pool, FTAG);
2002 error = dmu_objset_find_dp(spa->spa_dsl_pool,
2003 spa->spa_dsl_pool->dp_root_dir_obj, verify_dataset_name_len, NULL,
2005 dsl_pool_config_exit(spa->spa_dsl_pool, FTAG);
2009 rio = zio_root(spa, NULL, &sle,
2010 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
2012 if (spa_load_verify_metadata) {
2013 error = traverse_pool(spa, spa->spa_verify_min_txg,
2014 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA,
2015 spa_load_verify_cb, rio);
2018 (void) zio_wait(rio);
2020 spa->spa_load_meta_errors = sle.sle_meta_count;
2021 spa->spa_load_data_errors = sle.sle_data_count;
2023 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
2024 sle.sle_data_count <= policy.zrp_maxdata) {
2028 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
2029 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
2031 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
2032 VERIFY(nvlist_add_uint64(spa->spa_load_info,
2033 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
2034 VERIFY(nvlist_add_int64(spa->spa_load_info,
2035 ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
2036 VERIFY(nvlist_add_uint64(spa->spa_load_info,
2037 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
2039 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
2043 if (error != ENXIO && error != EIO)
2044 error = SET_ERROR(EIO);
2048 return (verify_ok ? 0 : EIO);
2052 * Find a value in the pool props object.
2055 spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
2057 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
2058 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
2062 * Find a value in the pool directory object.
2065 spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
2067 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
2068 name, sizeof (uint64_t), 1, val));
2072 spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
2074 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
2079 * Fix up config after a partly-completed split. This is done with the
2080 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
2081 * pool have that entry in their config, but only the splitting one contains
2082 * a list of all the guids of the vdevs that are being split off.
2084 * This function determines what to do with that list: either rejoin
2085 * all the disks to the pool, or complete the splitting process. To attempt
2086 * the rejoin, each disk that is offlined is marked online again, and
2087 * we do a reopen() call. If the vdev label for every disk that was
2088 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
2089 * then we call vdev_split() on each disk, and complete the split.
2091 * Otherwise we leave the config alone, with all the vdevs in place in
2092 * the original pool.
2095 spa_try_repair(spa_t *spa, nvlist_t *config)
2102 boolean_t attempt_reopen;
2104 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
2107 /* check that the config is complete */
2108 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
2109 &glist, &gcount) != 0)
2112 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
2114 /* attempt to online all the vdevs & validate */
2115 attempt_reopen = B_TRUE;
2116 for (i = 0; i < gcount; i++) {
2117 if (glist[i] == 0) /* vdev is hole */
2120 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
2121 if (vd[i] == NULL) {
2123 * Don't bother attempting to reopen the disks;
2124 * just do the split.
2126 attempt_reopen = B_FALSE;
2128 /* attempt to re-online it */
2129 vd[i]->vdev_offline = B_FALSE;
2133 if (attempt_reopen) {
2134 vdev_reopen(spa->spa_root_vdev);
2136 /* check each device to see what state it's in */
2137 for (extracted = 0, i = 0; i < gcount; i++) {
2138 if (vd[i] != NULL &&
2139 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
2146 * If every disk has been moved to the new pool, or if we never
2147 * even attempted to look at them, then we split them off for
2150 if (!attempt_reopen || gcount == extracted) {
2151 for (i = 0; i < gcount; i++)
2154 vdev_reopen(spa->spa_root_vdev);
2157 kmem_free(vd, gcount * sizeof (vdev_t *));
2161 spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
2162 boolean_t mosconfig)
2164 nvlist_t *config = spa->spa_config;
2165 char *ereport = FM_EREPORT_ZFS_POOL;
2171 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
2172 return (SET_ERROR(EINVAL));
2174 ASSERT(spa->spa_comment == NULL);
2175 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
2176 spa->spa_comment = spa_strdup(comment);
2179 * Versioning wasn't explicitly added to the label until later, so if
2180 * it's not present treat it as the initial version.
2182 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
2183 &spa->spa_ubsync.ub_version) != 0)
2184 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
2186 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
2187 &spa->spa_config_txg);
2189 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
2190 spa_guid_exists(pool_guid, 0)) {
2191 error = SET_ERROR(EEXIST);
2193 spa->spa_config_guid = pool_guid;
2195 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
2197 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
2201 nvlist_free(spa->spa_load_info);
2202 spa->spa_load_info = fnvlist_alloc();
2204 gethrestime(&spa->spa_loaded_ts);
2205 error = spa_load_impl(spa, pool_guid, config, state, type,
2206 mosconfig, &ereport);
2210 * Don't count references from objsets that are already closed
2211 * and are making their way through the eviction process.
2213 spa_evicting_os_wait(spa);
2214 spa->spa_minref = refcount_count(&spa->spa_refcount);
2216 if (error != EEXIST) {
2217 spa->spa_loaded_ts.tv_sec = 0;
2218 spa->spa_loaded_ts.tv_nsec = 0;
2220 if (error != EBADF) {
2221 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
2224 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2231 * Load an existing storage pool, using the pool's builtin spa_config as a
2232 * source of configuration information.
2235 spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
2236 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
2240 nvlist_t *nvroot = NULL;
2243 uberblock_t *ub = &spa->spa_uberblock;
2244 uint64_t children, config_cache_txg = spa->spa_config_txg;
2245 int orig_mode = spa->spa_mode;
2248 boolean_t missing_feat_write = B_FALSE;
2251 * If this is an untrusted config, access the pool in read-only mode.
2252 * This prevents things like resilvering recently removed devices.
2255 spa->spa_mode = FREAD;
2257 ASSERT(MUTEX_HELD(&spa_namespace_lock));
2259 spa->spa_load_state = state;
2261 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
2262 return (SET_ERROR(EINVAL));
2264 parse = (type == SPA_IMPORT_EXISTING ?
2265 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2268 * Create "The Godfather" zio to hold all async IOs
2270 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
2272 for (int i = 0; i < max_ncpus; i++) {
2273 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
2274 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
2275 ZIO_FLAG_GODFATHER);
2279 * Parse the configuration into a vdev tree. We explicitly set the
2280 * value that will be returned by spa_version() since parsing the
2281 * configuration requires knowing the version number.
2283 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2284 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
2285 spa_config_exit(spa, SCL_ALL, FTAG);
2290 ASSERT(spa->spa_root_vdev == rvd);
2291 ASSERT3U(spa->spa_min_ashift, >=, SPA_MINBLOCKSHIFT);
2292 ASSERT3U(spa->spa_max_ashift, <=, SPA_MAXBLOCKSHIFT);
2294 if (type != SPA_IMPORT_ASSEMBLE) {
2295 ASSERT(spa_guid(spa) == pool_guid);
2299 * Try to open all vdevs, loading each label in the process.
2301 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2302 error = vdev_open(rvd);
2303 spa_config_exit(spa, SCL_ALL, FTAG);
2308 * We need to validate the vdev labels against the configuration that
2309 * we have in hand, which is dependent on the setting of mosconfig. If
2310 * mosconfig is true then we're validating the vdev labels based on
2311 * that config. Otherwise, we're validating against the cached config
2312 * (zpool.cache) that was read when we loaded the zfs module, and then
2313 * later we will recursively call spa_load() and validate against
2316 * If we're assembling a new pool that's been split off from an
2317 * existing pool, the labels haven't yet been updated so we skip
2318 * validation for now.
2320 if (type != SPA_IMPORT_ASSEMBLE) {
2321 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2322 error = vdev_validate(rvd, mosconfig);
2323 spa_config_exit(spa, SCL_ALL, FTAG);
2328 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2329 return (SET_ERROR(ENXIO));
2333 * Find the best uberblock.
2335 vdev_uberblock_load(rvd, ub, &label);
2338 * If we weren't able to find a single valid uberblock, return failure.
2340 if (ub->ub_txg == 0) {
2342 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2346 * If the pool has an unsupported version we can't open it.
2348 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2350 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2353 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2357 * If we weren't able to find what's necessary for reading the
2358 * MOS in the label, return failure.
2360 if (label == NULL || nvlist_lookup_nvlist(label,
2361 ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) {
2363 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2368 * Update our in-core representation with the definitive values
2371 nvlist_free(spa->spa_label_features);
2372 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2378 * Look through entries in the label nvlist's features_for_read. If
2379 * there is a feature listed there which we don't understand then we
2380 * cannot open a pool.
2382 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2383 nvlist_t *unsup_feat;
2385 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2388 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2390 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2391 if (!zfeature_is_supported(nvpair_name(nvp))) {
2392 VERIFY(nvlist_add_string(unsup_feat,
2393 nvpair_name(nvp), "") == 0);
2397 if (!nvlist_empty(unsup_feat)) {
2398 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2399 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2400 nvlist_free(unsup_feat);
2401 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2405 nvlist_free(unsup_feat);
2409 * If the vdev guid sum doesn't match the uberblock, we have an
2410 * incomplete configuration. We first check to see if the pool
2411 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2412 * If it is, defer the vdev_guid_sum check till later so we
2413 * can handle missing vdevs.
2415 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
2416 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
2417 rvd->vdev_guid_sum != ub->ub_guid_sum)
2418 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2420 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2421 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2422 spa_try_repair(spa, config);
2423 spa_config_exit(spa, SCL_ALL, FTAG);
2424 nvlist_free(spa->spa_config_splitting);
2425 spa->spa_config_splitting = NULL;
2429 * Initialize internal SPA structures.
2431 spa->spa_state = POOL_STATE_ACTIVE;
2432 spa->spa_ubsync = spa->spa_uberblock;
2433 spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2434 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2435 spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2436 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2437 spa->spa_claim_max_txg = spa->spa_first_txg;
2438 spa->spa_prev_software_version = ub->ub_software_version;
2440 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2442 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2443 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2445 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2446 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2448 if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2449 boolean_t missing_feat_read = B_FALSE;
2450 nvlist_t *unsup_feat, *enabled_feat;
2452 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2453 &spa->spa_feat_for_read_obj) != 0) {
2454 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2457 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2458 &spa->spa_feat_for_write_obj) != 0) {
2459 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2462 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2463 &spa->spa_feat_desc_obj) != 0) {
2464 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2467 enabled_feat = fnvlist_alloc();
2468 unsup_feat = fnvlist_alloc();
2470 if (!spa_features_check(spa, B_FALSE,
2471 unsup_feat, enabled_feat))
2472 missing_feat_read = B_TRUE;
2474 if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) {
2475 if (!spa_features_check(spa, B_TRUE,
2476 unsup_feat, enabled_feat)) {
2477 missing_feat_write = B_TRUE;
2481 fnvlist_add_nvlist(spa->spa_load_info,
2482 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
2484 if (!nvlist_empty(unsup_feat)) {
2485 fnvlist_add_nvlist(spa->spa_load_info,
2486 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
2489 fnvlist_free(enabled_feat);
2490 fnvlist_free(unsup_feat);
2492 if (!missing_feat_read) {
2493 fnvlist_add_boolean(spa->spa_load_info,
2494 ZPOOL_CONFIG_CAN_RDONLY);
2498 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2499 * twofold: to determine whether the pool is available for
2500 * import in read-write mode and (if it is not) whether the
2501 * pool is available for import in read-only mode. If the pool
2502 * is available for import in read-write mode, it is displayed
2503 * as available in userland; if it is not available for import
2504 * in read-only mode, it is displayed as unavailable in
2505 * userland. If the pool is available for import in read-only
2506 * mode but not read-write mode, it is displayed as unavailable
2507 * in userland with a special note that the pool is actually
2508 * available for open in read-only mode.
2510 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2511 * missing a feature for write, we must first determine whether
2512 * the pool can be opened read-only before returning to
2513 * userland in order to know whether to display the
2514 * abovementioned note.
2516 if (missing_feat_read || (missing_feat_write &&
2517 spa_writeable(spa))) {
2518 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2523 * Load refcounts for ZFS features from disk into an in-memory
2524 * cache during SPA initialization.
2526 for (spa_feature_t i = 0; i < SPA_FEATURES; i++) {
2529 error = feature_get_refcount_from_disk(spa,
2530 &spa_feature_table[i], &refcount);
2532 spa->spa_feat_refcount_cache[i] = refcount;
2533 } else if (error == ENOTSUP) {
2534 spa->spa_feat_refcount_cache[i] =
2535 SPA_FEATURE_DISABLED;
2537 return (spa_vdev_err(rvd,
2538 VDEV_AUX_CORRUPT_DATA, EIO));
2543 if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) {
2544 if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG,
2545 &spa->spa_feat_enabled_txg_obj) != 0)
2546 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2549 spa->spa_is_initializing = B_TRUE;
2550 error = dsl_pool_open(spa->spa_dsl_pool);
2551 spa->spa_is_initializing = B_FALSE;
2553 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2557 nvlist_t *policy = NULL, *nvconfig;
2559 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2560 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2562 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2563 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2565 unsigned long myhostid = 0;
2567 VERIFY(nvlist_lookup_string(nvconfig,
2568 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2571 myhostid = zone_get_hostid(NULL);
2574 * We're emulating the system's hostid in userland, so
2575 * we can't use zone_get_hostid().
2577 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2578 #endif /* _KERNEL */
2579 if (check_hostid && hostid != 0 && myhostid != 0 &&
2580 hostid != myhostid) {
2581 nvlist_free(nvconfig);
2582 cmn_err(CE_WARN, "pool '%s' could not be "
2583 "loaded as it was last accessed by "
2584 "another system (host: %s hostid: 0x%lx). "
2585 "See: http://illumos.org/msg/ZFS-8000-EY",
2586 spa_name(spa), hostname,
2587 (unsigned long)hostid);
2588 return (SET_ERROR(EBADF));
2591 if (nvlist_lookup_nvlist(spa->spa_config,
2592 ZPOOL_REWIND_POLICY, &policy) == 0)
2593 VERIFY(nvlist_add_nvlist(nvconfig,
2594 ZPOOL_REWIND_POLICY, policy) == 0);
2596 spa_config_set(spa, nvconfig);
2598 spa_deactivate(spa);
2599 spa_activate(spa, orig_mode);
2601 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2604 /* Grab the secret checksum salt from the MOS. */
2605 error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
2606 DMU_POOL_CHECKSUM_SALT, 1,
2607 sizeof (spa->spa_cksum_salt.zcs_bytes),
2608 spa->spa_cksum_salt.zcs_bytes);
2609 if (error == ENOENT) {
2610 /* Generate a new salt for subsequent use */
2611 (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes,
2612 sizeof (spa->spa_cksum_salt.zcs_bytes));
2613 } else if (error != 0) {
2614 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2617 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2618 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2619 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2621 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2624 * Load the bit that tells us to use the new accounting function
2625 * (raid-z deflation). If we have an older pool, this will not
2628 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2629 if (error != 0 && error != ENOENT)
2630 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2632 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2633 &spa->spa_creation_version);
2634 if (error != 0 && error != ENOENT)
2635 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2638 * Load the persistent error log. If we have an older pool, this will
2641 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2642 if (error != 0 && error != ENOENT)
2643 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2645 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2646 &spa->spa_errlog_scrub);
2647 if (error != 0 && error != ENOENT)
2648 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2651 * Load the history object. If we have an older pool, this
2652 * will not be present.
2654 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2655 if (error != 0 && error != ENOENT)
2656 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2659 * If we're assembling the pool from the split-off vdevs of
2660 * an existing pool, we don't want to attach the spares & cache
2665 * Load any hot spares for this pool.
2667 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2668 if (error != 0 && error != ENOENT)
2669 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2670 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2671 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2672 if (load_nvlist(spa, spa->spa_spares.sav_object,
2673 &spa->spa_spares.sav_config) != 0)
2674 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2676 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2677 spa_load_spares(spa);
2678 spa_config_exit(spa, SCL_ALL, FTAG);
2679 } else if (error == 0) {
2680 spa->spa_spares.sav_sync = B_TRUE;
2684 * Load any level 2 ARC devices for this pool.
2686 error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2687 &spa->spa_l2cache.sav_object);
2688 if (error != 0 && error != ENOENT)
2689 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2690 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2691 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2692 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2693 &spa->spa_l2cache.sav_config) != 0)
2694 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2696 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2697 spa_load_l2cache(spa);
2698 spa_config_exit(spa, SCL_ALL, FTAG);
2699 } else if (error == 0) {
2700 spa->spa_l2cache.sav_sync = B_TRUE;
2703 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2705 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2706 if (error && error != ENOENT)
2707 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2710 uint64_t autoreplace;
2712 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2713 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2714 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2715 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2716 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2717 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2718 &spa->spa_dedup_ditto);
2720 spa->spa_autoreplace = (autoreplace != 0);
2724 * If the 'autoreplace' property is set, then post a resource notifying
2725 * the ZFS DE that it should not issue any faults for unopenable
2726 * devices. We also iterate over the vdevs, and post a sysevent for any
2727 * unopenable vdevs so that the normal autoreplace handler can take
2730 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2731 spa_check_removed(spa->spa_root_vdev);
2733 * For the import case, this is done in spa_import(), because
2734 * at this point we're using the spare definitions from
2735 * the MOS config, not necessarily from the userland config.
2737 if (state != SPA_LOAD_IMPORT) {
2738 spa_aux_check_removed(&spa->spa_spares);
2739 spa_aux_check_removed(&spa->spa_l2cache);
2744 * Load the vdev state for all toplevel vdevs.
2749 * Propagate the leaf DTLs we just loaded all the way up the tree.
2751 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2752 vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2753 spa_config_exit(spa, SCL_ALL, FTAG);
2756 * Load the DDTs (dedup tables).
2758 error = ddt_load(spa);
2760 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2762 spa_update_dspace(spa);
2765 * Validate the config, using the MOS config to fill in any
2766 * information which might be missing. If we fail to validate
2767 * the config then declare the pool unfit for use. If we're
2768 * assembling a pool from a split, the log is not transferred
2771 if (type != SPA_IMPORT_ASSEMBLE) {
2774 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2775 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2777 if (!spa_config_valid(spa, nvconfig)) {
2778 nvlist_free(nvconfig);
2779 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2782 nvlist_free(nvconfig);
2785 * Now that we've validated the config, check the state of the
2786 * root vdev. If it can't be opened, it indicates one or
2787 * more toplevel vdevs are faulted.
2789 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2790 return (SET_ERROR(ENXIO));
2792 if (spa_writeable(spa) && spa_check_logs(spa)) {
2793 *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2794 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2798 if (missing_feat_write) {
2799 ASSERT(state == SPA_LOAD_TRYIMPORT);
2802 * At this point, we know that we can open the pool in
2803 * read-only mode but not read-write mode. We now have enough
2804 * information and can return to userland.
2806 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
2810 * We've successfully opened the pool, verify that we're ready
2811 * to start pushing transactions.
2813 if (state != SPA_LOAD_TRYIMPORT) {
2814 if (error = spa_load_verify(spa))
2815 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2819 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2820 spa->spa_load_max_txg == UINT64_MAX)) {
2822 int need_update = B_FALSE;
2823 dsl_pool_t *dp = spa_get_dsl(spa);
2825 ASSERT(state != SPA_LOAD_TRYIMPORT);
2828 * Claim log blocks that haven't been committed yet.
2829 * This must all happen in a single txg.
2830 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2831 * invoked from zil_claim_log_block()'s i/o done callback.
2832 * Price of rollback is that we abandon the log.
2834 spa->spa_claiming = B_TRUE;
2836 tx = dmu_tx_create_assigned(dp, spa_first_txg(spa));
2837 (void) dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
2838 zil_claim, tx, DS_FIND_CHILDREN);
2841 spa->spa_claiming = B_FALSE;
2843 spa_set_log_state(spa, SPA_LOG_GOOD);
2844 spa->spa_sync_on = B_TRUE;
2845 txg_sync_start(spa->spa_dsl_pool);
2848 * Wait for all claims to sync. We sync up to the highest
2849 * claimed log block birth time so that claimed log blocks
2850 * don't appear to be from the future. spa_claim_max_txg
2851 * will have been set for us by either zil_check_log_chain()
2852 * (invoked from spa_check_logs()) or zil_claim() above.
2854 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2857 * If the config cache is stale, or we have uninitialized
2858 * metaslabs (see spa_vdev_add()), then update the config.
2860 * If this is a verbatim import, trust the current
2861 * in-core spa_config and update the disk labels.
2863 if (config_cache_txg != spa->spa_config_txg ||
2864 state == SPA_LOAD_IMPORT ||
2865 state == SPA_LOAD_RECOVER ||
2866 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2867 need_update = B_TRUE;
2869 for (int c = 0; c < rvd->vdev_children; c++)
2870 if (rvd->vdev_child[c]->vdev_ms_array == 0)
2871 need_update = B_TRUE;
2874 * Update the config cache asychronously in case we're the
2875 * root pool, in which case the config cache isn't writable yet.
2878 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2881 * Check all DTLs to see if anything needs resilvering.
2883 if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2884 vdev_resilver_needed(rvd, NULL, NULL))
2885 spa_async_request(spa, SPA_ASYNC_RESILVER);
2888 * Log the fact that we booted up (so that we can detect if
2889 * we rebooted in the middle of an operation).
2891 spa_history_log_version(spa, "open");
2894 * Delete any inconsistent datasets.
2896 (void) dmu_objset_find(spa_name(spa),
2897 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2900 * Clean up any stale temporary dataset userrefs.
2902 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2909 spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2911 int mode = spa->spa_mode;
2914 spa_deactivate(spa);
2916 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg - 1;
2918 spa_activate(spa, mode);
2919 spa_async_suspend(spa);
2921 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2925 * If spa_load() fails this function will try loading prior txg's. If
2926 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2927 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2928 * function will not rewind the pool and will return the same error as
2932 spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2933 uint64_t max_request, int rewind_flags)
2935 nvlist_t *loadinfo = NULL;
2936 nvlist_t *config = NULL;
2937 int load_error, rewind_error;
2938 uint64_t safe_rewind_txg;
2941 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2942 spa->spa_load_max_txg = spa->spa_load_txg;
2943 spa_set_log_state(spa, SPA_LOG_CLEAR);
2945 spa->spa_load_max_txg = max_request;
2946 if (max_request != UINT64_MAX)
2947 spa->spa_extreme_rewind = B_TRUE;
2950 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2952 if (load_error == 0)
2955 if (spa->spa_root_vdev != NULL)
2956 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2958 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2959 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2961 if (rewind_flags & ZPOOL_NEVER_REWIND) {
2962 nvlist_free(config);
2963 return (load_error);
2966 if (state == SPA_LOAD_RECOVER) {
2967 /* Price of rolling back is discarding txgs, including log */
2968 spa_set_log_state(spa, SPA_LOG_CLEAR);
2971 * If we aren't rolling back save the load info from our first
2972 * import attempt so that we can restore it after attempting
2975 loadinfo = spa->spa_load_info;
2976 spa->spa_load_info = fnvlist_alloc();
2979 spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2980 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2981 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2982 TXG_INITIAL : safe_rewind_txg;
2985 * Continue as long as we're finding errors, we're still within
2986 * the acceptable rewind range, and we're still finding uberblocks
2988 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2989 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2990 if (spa->spa_load_max_txg < safe_rewind_txg)
2991 spa->spa_extreme_rewind = B_TRUE;
2992 rewind_error = spa_load_retry(spa, state, mosconfig);
2995 spa->spa_extreme_rewind = B_FALSE;
2996 spa->spa_load_max_txg = UINT64_MAX;
2998 if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2999 spa_config_set(spa, config);
3001 if (state == SPA_LOAD_RECOVER) {
3002 ASSERT3P(loadinfo, ==, NULL);
3003 return (rewind_error);
3005 /* Store the rewind info as part of the initial load info */
3006 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
3007 spa->spa_load_info);
3009 /* Restore the initial load info */
3010 fnvlist_free(spa->spa_load_info);
3011 spa->spa_load_info = loadinfo;
3013 return (load_error);
3020 * The import case is identical to an open except that the configuration is sent
3021 * down from userland, instead of grabbed from the configuration cache. For the
3022 * case of an open, the pool configuration will exist in the
3023 * POOL_STATE_UNINITIALIZED state.
3025 * The stats information (gen/count/ustats) is used to gather vdev statistics at
3026 * the same time open the pool, without having to keep around the spa_t in some
3030 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
3034 spa_load_state_t state = SPA_LOAD_OPEN;
3036 int locked = B_FALSE;
3037 int firstopen = B_FALSE;
3042 * As disgusting as this is, we need to support recursive calls to this
3043 * function because dsl_dir_open() is called during spa_load(), and ends
3044 * up calling spa_open() again. The real fix is to figure out how to
3045 * avoid dsl_dir_open() calling this in the first place.
3047 if (mutex_owner(&spa_namespace_lock) != curthread) {
3048 mutex_enter(&spa_namespace_lock);
3052 if ((spa = spa_lookup(pool)) == NULL) {
3054 mutex_exit(&spa_namespace_lock);
3055 return (SET_ERROR(ENOENT));
3058 if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
3059 zpool_rewind_policy_t policy;
3063 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
3065 if (policy.zrp_request & ZPOOL_DO_REWIND)
3066 state = SPA_LOAD_RECOVER;
3068 spa_activate(spa, spa_mode_global);
3070 if (state != SPA_LOAD_RECOVER)
3071 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3073 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
3074 policy.zrp_request);
3076 if (error == EBADF) {
3078 * If vdev_validate() returns failure (indicated by
3079 * EBADF), it indicates that one of the vdevs indicates
3080 * that the pool has been exported or destroyed. If
3081 * this is the case, the config cache is out of sync and
3082 * we should remove the pool from the namespace.
3085 spa_deactivate(spa);
3086 spa_config_sync(spa, B_TRUE, B_TRUE);
3089 mutex_exit(&spa_namespace_lock);
3090 return (SET_ERROR(ENOENT));
3095 * We can't open the pool, but we still have useful
3096 * information: the state of each vdev after the
3097 * attempted vdev_open(). Return this to the user.
3099 if (config != NULL && spa->spa_config) {
3100 VERIFY(nvlist_dup(spa->spa_config, config,
3102 VERIFY(nvlist_add_nvlist(*config,
3103 ZPOOL_CONFIG_LOAD_INFO,
3104 spa->spa_load_info) == 0);
3107 spa_deactivate(spa);
3108 spa->spa_last_open_failed = error;
3110 mutex_exit(&spa_namespace_lock);
3116 spa_open_ref(spa, tag);
3119 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
3122 * If we've recovered the pool, pass back any information we
3123 * gathered while doing the load.
3125 if (state == SPA_LOAD_RECOVER) {
3126 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
3127 spa->spa_load_info) == 0);
3131 spa->spa_last_open_failed = 0;
3132 spa->spa_last_ubsync_txg = 0;
3133 spa->spa_load_txg = 0;
3134 mutex_exit(&spa_namespace_lock);
3138 zvol_create_minors(spa->spa_name);
3149 spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
3152 return (spa_open_common(name, spapp, tag, policy, config));
3156 spa_open(const char *name, spa_t **spapp, void *tag)
3158 return (spa_open_common(name, spapp, tag, NULL, NULL));
3162 * Lookup the given spa_t, incrementing the inject count in the process,
3163 * preventing it from being exported or destroyed.
3166 spa_inject_addref(char *name)
3170 mutex_enter(&spa_namespace_lock);
3171 if ((spa = spa_lookup(name)) == NULL) {
3172 mutex_exit(&spa_namespace_lock);
3175 spa->spa_inject_ref++;
3176 mutex_exit(&spa_namespace_lock);
3182 spa_inject_delref(spa_t *spa)
3184 mutex_enter(&spa_namespace_lock);
3185 spa->spa_inject_ref--;
3186 mutex_exit(&spa_namespace_lock);
3190 * Add spares device information to the nvlist.
3193 spa_add_spares(spa_t *spa, nvlist_t *config)
3203 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3205 if (spa->spa_spares.sav_count == 0)
3208 VERIFY(nvlist_lookup_nvlist(config,
3209 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3210 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
3211 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3213 VERIFY(nvlist_add_nvlist_array(nvroot,
3214 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3215 VERIFY(nvlist_lookup_nvlist_array(nvroot,
3216 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3219 * Go through and find any spares which have since been
3220 * repurposed as an active spare. If this is the case, update
3221 * their status appropriately.
3223 for (i = 0; i < nspares; i++) {
3224 VERIFY(nvlist_lookup_uint64(spares[i],
3225 ZPOOL_CONFIG_GUID, &guid) == 0);
3226 if (spa_spare_exists(guid, &pool, NULL) &&
3228 VERIFY(nvlist_lookup_uint64_array(
3229 spares[i], ZPOOL_CONFIG_VDEV_STATS,
3230 (uint64_t **)&vs, &vsc) == 0);
3231 vs->vs_state = VDEV_STATE_CANT_OPEN;
3232 vs->vs_aux = VDEV_AUX_SPARED;
3239 * Add l2cache device information to the nvlist, including vdev stats.
3242 spa_add_l2cache(spa_t *spa, nvlist_t *config)
3245 uint_t i, j, nl2cache;
3252 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3254 if (spa->spa_l2cache.sav_count == 0)
3257 VERIFY(nvlist_lookup_nvlist(config,
3258 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3259 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
3260 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3261 if (nl2cache != 0) {
3262 VERIFY(nvlist_add_nvlist_array(nvroot,
3263 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3264 VERIFY(nvlist_lookup_nvlist_array(nvroot,
3265 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3268 * Update level 2 cache device stats.
3271 for (i = 0; i < nl2cache; i++) {
3272 VERIFY(nvlist_lookup_uint64(l2cache[i],
3273 ZPOOL_CONFIG_GUID, &guid) == 0);
3276 for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
3278 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
3279 vd = spa->spa_l2cache.sav_vdevs[j];
3285 VERIFY(nvlist_lookup_uint64_array(l2cache[i],
3286 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
3288 vdev_get_stats(vd, vs);
3294 spa_add_feature_stats(spa_t *spa, nvlist_t *config)
3300 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3301 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3303 /* We may be unable to read features if pool is suspended. */
3304 if (spa_suspended(spa))
3307 if (spa->spa_feat_for_read_obj != 0) {
3308 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3309 spa->spa_feat_for_read_obj);
3310 zap_cursor_retrieve(&zc, &za) == 0;
3311 zap_cursor_advance(&zc)) {
3312 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3313 za.za_num_integers == 1);
3314 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3315 za.za_first_integer));
3317 zap_cursor_fini(&zc);
3320 if (spa->spa_feat_for_write_obj != 0) {
3321 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3322 spa->spa_feat_for_write_obj);
3323 zap_cursor_retrieve(&zc, &za) == 0;
3324 zap_cursor_advance(&zc)) {
3325 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3326 za.za_num_integers == 1);
3327 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3328 za.za_first_integer));
3330 zap_cursor_fini(&zc);
3334 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
3336 nvlist_free(features);
3340 spa_get_stats(const char *name, nvlist_t **config,
3341 char *altroot, size_t buflen)
3347 error = spa_open_common(name, &spa, FTAG, NULL, config);
3351 * This still leaves a window of inconsistency where the spares
3352 * or l2cache devices could change and the config would be
3353 * self-inconsistent.
3355 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3357 if (*config != NULL) {
3358 uint64_t loadtimes[2];
3360 loadtimes[0] = spa->spa_loaded_ts.tv_sec;
3361 loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
3362 VERIFY(nvlist_add_uint64_array(*config,
3363 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
3365 VERIFY(nvlist_add_uint64(*config,
3366 ZPOOL_CONFIG_ERRCOUNT,
3367 spa_get_errlog_size(spa)) == 0);
3369 if (spa_suspended(spa))
3370 VERIFY(nvlist_add_uint64(*config,
3371 ZPOOL_CONFIG_SUSPENDED,
3372 spa->spa_failmode) == 0);
3374 spa_add_spares(spa, *config);
3375 spa_add_l2cache(spa, *config);
3376 spa_add_feature_stats(spa, *config);
3381 * We want to get the alternate root even for faulted pools, so we cheat
3382 * and call spa_lookup() directly.
3386 mutex_enter(&spa_namespace_lock);
3387 spa = spa_lookup(name);
3389 spa_altroot(spa, altroot, buflen);
3393 mutex_exit(&spa_namespace_lock);
3395 spa_altroot(spa, altroot, buflen);
3400 spa_config_exit(spa, SCL_CONFIG, FTAG);
3401 spa_close(spa, FTAG);
3408 * Validate that the auxiliary device array is well formed. We must have an
3409 * array of nvlists, each which describes a valid leaf vdev. If this is an
3410 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3411 * specified, as long as they are well-formed.
3414 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
3415 spa_aux_vdev_t *sav, const char *config, uint64_t version,
3416 vdev_labeltype_t label)
3423 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3426 * It's acceptable to have no devs specified.
3428 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
3432 return (SET_ERROR(EINVAL));
3435 * Make sure the pool is formatted with a version that supports this
3438 if (spa_version(spa) < version)
3439 return (SET_ERROR(ENOTSUP));
3442 * Set the pending device list so we correctly handle device in-use
3445 sav->sav_pending = dev;
3446 sav->sav_npending = ndev;
3448 for (i = 0; i < ndev; i++) {
3449 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
3453 if (!vd->vdev_ops->vdev_op_leaf) {
3455 error = SET_ERROR(EINVAL);
3460 * The L2ARC currently only supports disk devices in
3461 * kernel context. For user-level testing, we allow it.
3464 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
3465 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
3466 error = SET_ERROR(ENOTBLK);
3473 if ((error = vdev_open(vd)) == 0 &&
3474 (error = vdev_label_init(vd, crtxg, label)) == 0) {
3475 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
3476 vd->vdev_guid) == 0);
3482 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
3489 sav->sav_pending = NULL;
3490 sav->sav_npending = 0;
3495 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
3499 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3501 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3502 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
3503 VDEV_LABEL_SPARE)) != 0) {
3507 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3508 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
3509 VDEV_LABEL_L2CACHE));
3513 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
3518 if (sav->sav_config != NULL) {
3524 * Generate new dev list by concatentating with the
3527 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
3528 &olddevs, &oldndevs) == 0);
3530 newdevs = kmem_alloc(sizeof (void *) *
3531 (ndevs + oldndevs), KM_SLEEP);
3532 for (i = 0; i < oldndevs; i++)
3533 VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
3535 for (i = 0; i < ndevs; i++)
3536 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
3539 VERIFY(nvlist_remove(sav->sav_config, config,
3540 DATA_TYPE_NVLIST_ARRAY) == 0);
3542 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3543 config, newdevs, ndevs + oldndevs) == 0);
3544 for (i = 0; i < oldndevs + ndevs; i++)
3545 nvlist_free(newdevs[i]);
3546 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
3549 * Generate a new dev list.
3551 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
3553 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
3559 * Stop and drop level 2 ARC devices
3562 spa_l2cache_drop(spa_t *spa)
3566 spa_aux_vdev_t *sav = &spa->spa_l2cache;
3568 for (i = 0; i < sav->sav_count; i++) {
3571 vd = sav->sav_vdevs[i];
3574 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
3575 pool != 0ULL && l2arc_vdev_present(vd))
3576 l2arc_remove_vdev(vd);
3584 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
3588 char *altroot = NULL;
3593 uint64_t txg = TXG_INITIAL;
3594 nvlist_t **spares, **l2cache;
3595 uint_t nspares, nl2cache;
3596 uint64_t version, obj;
3597 boolean_t has_features;
3600 * If this pool already exists, return failure.
3602 mutex_enter(&spa_namespace_lock);
3603 if (spa_lookup(pool) != NULL) {
3604 mutex_exit(&spa_namespace_lock);
3605 return (SET_ERROR(EEXIST));
3609 * Allocate a new spa_t structure.
3611 (void) nvlist_lookup_string(props,
3612 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3613 spa = spa_add(pool, NULL, altroot);
3614 spa_activate(spa, spa_mode_global);
3616 if (props && (error = spa_prop_validate(spa, props))) {
3617 spa_deactivate(spa);
3619 mutex_exit(&spa_namespace_lock);
3623 has_features = B_FALSE;
3624 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
3625 elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
3626 if (zpool_prop_feature(nvpair_name(elem)))
3627 has_features = B_TRUE;
3630 if (has_features || nvlist_lookup_uint64(props,
3631 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
3632 version = SPA_VERSION;
3634 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
3636 spa->spa_first_txg = txg;
3637 spa->spa_uberblock.ub_txg = txg - 1;
3638 spa->spa_uberblock.ub_version = version;
3639 spa->spa_ubsync = spa->spa_uberblock;
3640 spa->spa_load_state = SPA_LOAD_CREATE;
3643 * Create "The Godfather" zio to hold all async IOs
3645 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
3647 for (int i = 0; i < max_ncpus; i++) {
3648 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
3649 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
3650 ZIO_FLAG_GODFATHER);
3654 * Create the root vdev.
3656 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3658 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3660 ASSERT(error != 0 || rvd != NULL);
3661 ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3663 if (error == 0 && !zfs_allocatable_devs(nvroot))
3664 error = SET_ERROR(EINVAL);
3667 (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3668 (error = spa_validate_aux(spa, nvroot, txg,
3669 VDEV_ALLOC_ADD)) == 0) {
3670 for (int c = 0; c < rvd->vdev_children; c++) {
3671 vdev_ashift_optimize(rvd->vdev_child[c]);
3672 vdev_metaslab_set_size(rvd->vdev_child[c]);
3673 vdev_expand(rvd->vdev_child[c], txg);
3677 spa_config_exit(spa, SCL_ALL, FTAG);
3681 spa_deactivate(spa);
3683 mutex_exit(&spa_namespace_lock);
3688 * Get the list of spares, if specified.
3690 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3691 &spares, &nspares) == 0) {
3692 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3694 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3695 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3696 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3697 spa_load_spares(spa);
3698 spa_config_exit(spa, SCL_ALL, FTAG);
3699 spa->spa_spares.sav_sync = B_TRUE;
3703 * Get the list of level 2 cache devices, if specified.
3705 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3706 &l2cache, &nl2cache) == 0) {
3707 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3708 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3709 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3710 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3711 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3712 spa_load_l2cache(spa);
3713 spa_config_exit(spa, SCL_ALL, FTAG);
3714 spa->spa_l2cache.sav_sync = B_TRUE;
3717 spa->spa_is_initializing = B_TRUE;
3718 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3719 spa->spa_meta_objset = dp->dp_meta_objset;
3720 spa->spa_is_initializing = B_FALSE;
3723 * Create DDTs (dedup tables).
3727 spa_update_dspace(spa);
3729 tx = dmu_tx_create_assigned(dp, txg);
3732 * Create the pool config object.
3734 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3735 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3736 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3738 if (zap_add(spa->spa_meta_objset,
3739 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3740 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3741 cmn_err(CE_PANIC, "failed to add pool config");
3744 if (spa_version(spa) >= SPA_VERSION_FEATURES)
3745 spa_feature_create_zap_objects(spa, tx);
3747 if (zap_add(spa->spa_meta_objset,
3748 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3749 sizeof (uint64_t), 1, &version, tx) != 0) {
3750 cmn_err(CE_PANIC, "failed to add pool version");
3753 /* Newly created pools with the right version are always deflated. */
3754 if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3755 spa->spa_deflate = TRUE;
3756 if (zap_add(spa->spa_meta_objset,
3757 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3758 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3759 cmn_err(CE_PANIC, "failed to add deflate");
3764 * Create the deferred-free bpobj. Turn off compression
3765 * because sync-to-convergence takes longer if the blocksize
3768 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3769 dmu_object_set_compress(spa->spa_meta_objset, obj,
3770 ZIO_COMPRESS_OFF, tx);
3771 if (zap_add(spa->spa_meta_objset,
3772 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3773 sizeof (uint64_t), 1, &obj, tx) != 0) {
3774 cmn_err(CE_PANIC, "failed to add bpobj");
3776 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3777 spa->spa_meta_objset, obj));
3780 * Create the pool's history object.
3782 if (version >= SPA_VERSION_ZPOOL_HISTORY)
3783 spa_history_create_obj(spa, tx);
3786 * Generate some random noise for salted checksums to operate on.
3788 (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes,
3789 sizeof (spa->spa_cksum_salt.zcs_bytes));
3792 * Set pool properties.
3794 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3795 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3796 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3797 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3799 if (props != NULL) {
3800 spa_configfile_set(spa, props, B_FALSE);
3801 spa_sync_props(props, tx);
3806 spa->spa_sync_on = B_TRUE;
3807 txg_sync_start(spa->spa_dsl_pool);
3810 * We explicitly wait for the first transaction to complete so that our
3811 * bean counters are appropriately updated.
3813 txg_wait_synced(spa->spa_dsl_pool, txg);
3815 spa_config_sync(spa, B_FALSE, B_TRUE);
3816 spa_event_notify(spa, NULL, ESC_ZFS_POOL_CREATE);
3818 spa_history_log_version(spa, "create");
3821 * Don't count references from objsets that are already closed
3822 * and are making their way through the eviction process.
3824 spa_evicting_os_wait(spa);
3825 spa->spa_minref = refcount_count(&spa->spa_refcount);
3826 spa->spa_load_state = SPA_LOAD_NONE;
3828 mutex_exit(&spa_namespace_lock);
3836 * Get the root pool information from the root disk, then import the root pool
3837 * during the system boot up time.
3839 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3842 spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3845 nvlist_t *nvtop, *nvroot;
3848 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3852 * Add this top-level vdev to the child array.
3854 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3856 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3858 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3861 * Put this pool's top-level vdevs into a root vdev.
3863 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3864 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3865 VDEV_TYPE_ROOT) == 0);
3866 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3867 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3868 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3872 * Replace the existing vdev_tree with the new root vdev in
3873 * this pool's configuration (remove the old, add the new).
3875 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3876 nvlist_free(nvroot);
3881 * Walk the vdev tree and see if we can find a device with "better"
3882 * configuration. A configuration is "better" if the label on that
3883 * device has a more recent txg.
3886 spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3888 for (int c = 0; c < vd->vdev_children; c++)
3889 spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3891 if (vd->vdev_ops->vdev_op_leaf) {
3895 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3899 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3903 * Do we have a better boot device?
3905 if (label_txg > *txg) {
3914 * Import a root pool.
3916 * For x86. devpath_list will consist of devid and/or physpath name of
3917 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3918 * The GRUB "findroot" command will return the vdev we should boot.
3920 * For Sparc, devpath_list consists the physpath name of the booting device
3921 * no matter the rootpool is a single device pool or a mirrored pool.
3923 * "/pci@1f,0/ide@d/disk@0,0:a"
3926 spa_import_rootpool(char *devpath, char *devid)
3929 vdev_t *rvd, *bvd, *avd = NULL;
3930 nvlist_t *config, *nvtop;
3936 * Read the label from the boot device and generate a configuration.
3938 config = spa_generate_rootconf(devpath, devid, &guid);
3939 #if defined(_OBP) && defined(_KERNEL)
3940 if (config == NULL) {
3941 if (strstr(devpath, "/iscsi/ssd") != NULL) {
3943 get_iscsi_bootpath_phy(devpath);
3944 config = spa_generate_rootconf(devpath, devid, &guid);
3948 if (config == NULL) {
3949 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
3951 return (SET_ERROR(EIO));
3954 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3956 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3958 mutex_enter(&spa_namespace_lock);
3959 if ((spa = spa_lookup(pname)) != NULL) {
3961 * Remove the existing root pool from the namespace so that we
3962 * can replace it with the correct config we just read in.
3967 spa = spa_add(pname, config, NULL);
3968 spa->spa_is_root = B_TRUE;
3969 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3972 * Build up a vdev tree based on the boot device's label config.
3974 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3976 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3977 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3978 VDEV_ALLOC_ROOTPOOL);
3979 spa_config_exit(spa, SCL_ALL, FTAG);
3981 mutex_exit(&spa_namespace_lock);
3982 nvlist_free(config);
3983 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3989 * Get the boot vdev.
3991 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3992 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3993 (u_longlong_t)guid);
3994 error = SET_ERROR(ENOENT);
3999 * Determine if there is a better boot device.
4002 spa_alt_rootvdev(rvd, &avd, &txg);
4004 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
4005 "try booting from '%s'", avd->vdev_path);
4006 error = SET_ERROR(EINVAL);
4011 * If the boot device is part of a spare vdev then ensure that
4012 * we're booting off the active spare.
4014 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
4015 !bvd->vdev_isspare) {
4016 cmn_err(CE_NOTE, "The boot device is currently spared. Please "
4017 "try booting from '%s'",
4019 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
4020 error = SET_ERROR(EINVAL);
4026 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4028 spa_config_exit(spa, SCL_ALL, FTAG);
4029 mutex_exit(&spa_namespace_lock);
4031 nvlist_free(config);
4035 #else /* !illumos */
4037 extern int vdev_geom_read_pool_label(const char *name, nvlist_t ***configs,
4041 spa_generate_rootconf(const char *name)
4043 nvlist_t **configs, **tops;
4045 nvlist_t *best_cfg, *nvtop, *nvroot;
4054 if (vdev_geom_read_pool_label(name, &configs, &count) != 0)
4057 ASSERT3U(count, !=, 0);
4059 for (i = 0; i < count; i++) {
4062 VERIFY(nvlist_lookup_uint64(configs[i], ZPOOL_CONFIG_POOL_TXG,
4064 if (txg > best_txg) {
4066 best_cfg = configs[i];
4071 * Multi-vdev root pool configuration discovery is not supported yet.
4074 nvlist_lookup_uint64(best_cfg, ZPOOL_CONFIG_VDEV_CHILDREN, &nchildren);
4076 nvlist_lookup_uint64_array(best_cfg, ZPOOL_CONFIG_HOLE_ARRAY,
4079 tops = kmem_zalloc(nchildren * sizeof(void *), KM_SLEEP);
4080 for (i = 0; i < nchildren; i++) {
4083 if (configs[i] == NULL)
4085 VERIFY(nvlist_lookup_nvlist(configs[i], ZPOOL_CONFIG_VDEV_TREE,
4087 nvlist_dup(nvtop, &tops[i], KM_SLEEP);
4089 for (i = 0; holes != NULL && i < nholes; i++) {
4092 if (tops[holes[i]] != NULL)
4094 nvlist_alloc(&tops[holes[i]], NV_UNIQUE_NAME, KM_SLEEP);
4095 VERIFY(nvlist_add_string(tops[holes[i]], ZPOOL_CONFIG_TYPE,
4096 VDEV_TYPE_HOLE) == 0);
4097 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_ID,
4099 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_GUID,
4102 for (i = 0; i < nchildren; i++) {
4103 if (tops[i] != NULL)
4105 nvlist_alloc(&tops[i], NV_UNIQUE_NAME, KM_SLEEP);
4106 VERIFY(nvlist_add_string(tops[i], ZPOOL_CONFIG_TYPE,
4107 VDEV_TYPE_MISSING) == 0);
4108 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_ID,
4110 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_GUID,
4115 * Create pool config based on the best vdev config.
4117 nvlist_dup(best_cfg, &config, KM_SLEEP);
4120 * Put this pool's top-level vdevs into a root vdev.
4122 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4124 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4125 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
4126 VDEV_TYPE_ROOT) == 0);
4127 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
4128 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
4129 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
4130 tops, nchildren) == 0);
4133 * Replace the existing vdev_tree with the new root vdev in
4134 * this pool's configuration (remove the old, add the new).
4136 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
4139 * Drop vdev config elements that should not be present at pool level.
4141 nvlist_remove(config, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64);
4142 nvlist_remove(config, ZPOOL_CONFIG_TOP_GUID, DATA_TYPE_UINT64);
4144 for (i = 0; i < count; i++)
4145 nvlist_free(configs[i]);
4146 kmem_free(configs, count * sizeof(void *));
4147 for (i = 0; i < nchildren; i++)
4148 nvlist_free(tops[i]);
4149 kmem_free(tops, nchildren * sizeof(void *));
4150 nvlist_free(nvroot);
4155 spa_import_rootpool(const char *name)
4158 vdev_t *rvd, *bvd, *avd = NULL;
4159 nvlist_t *config, *nvtop;
4165 * Read the label from the boot device and generate a configuration.
4167 config = spa_generate_rootconf(name);
4169 mutex_enter(&spa_namespace_lock);
4170 if (config != NULL) {
4171 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
4172 &pname) == 0 && strcmp(name, pname) == 0);
4173 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg)
4176 if ((spa = spa_lookup(pname)) != NULL) {
4178 * Remove the existing root pool from the namespace so
4179 * that we can replace it with the correct config
4184 spa = spa_add(pname, config, NULL);
4187 * Set spa_ubsync.ub_version as it can be used in vdev_alloc()
4188 * via spa_version().
4190 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
4191 &spa->spa_ubsync.ub_version) != 0)
4192 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
4193 } else if ((spa = spa_lookup(name)) == NULL) {
4194 mutex_exit(&spa_namespace_lock);
4195 nvlist_free(config);
4196 cmn_err(CE_NOTE, "Cannot find the pool label for '%s'",
4200 VERIFY(nvlist_dup(spa->spa_config, &config, KM_SLEEP) == 0);
4202 spa->spa_is_root = B_TRUE;
4203 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
4206 * Build up a vdev tree based on the boot device's label config.
4208 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4210 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4211 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
4212 VDEV_ALLOC_ROOTPOOL);
4213 spa_config_exit(spa, SCL_ALL, FTAG);
4215 mutex_exit(&spa_namespace_lock);
4216 nvlist_free(config);
4217 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
4222 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4224 spa_config_exit(spa, SCL_ALL, FTAG);
4225 mutex_exit(&spa_namespace_lock);
4227 nvlist_free(config);
4231 #endif /* illumos */
4232 #endif /* _KERNEL */
4235 * Import a non-root pool into the system.
4238 spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
4241 char *altroot = NULL;
4242 spa_load_state_t state = SPA_LOAD_IMPORT;
4243 zpool_rewind_policy_t policy;
4244 uint64_t mode = spa_mode_global;
4245 uint64_t readonly = B_FALSE;
4248 nvlist_t **spares, **l2cache;
4249 uint_t nspares, nl2cache;
4252 * If a pool with this name exists, return failure.
4254 mutex_enter(&spa_namespace_lock);
4255 if (spa_lookup(pool) != NULL) {
4256 mutex_exit(&spa_namespace_lock);
4257 return (SET_ERROR(EEXIST));
4261 * Create and initialize the spa structure.
4263 (void) nvlist_lookup_string(props,
4264 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4265 (void) nvlist_lookup_uint64(props,
4266 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
4269 spa = spa_add(pool, config, altroot);
4270 spa->spa_import_flags = flags;
4273 * Verbatim import - Take a pool and insert it into the namespace
4274 * as if it had been loaded at boot.
4276 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
4278 spa_configfile_set(spa, props, B_FALSE);
4280 spa_config_sync(spa, B_FALSE, B_TRUE);
4281 spa_event_notify(spa, NULL, ESC_ZFS_POOL_IMPORT);
4283 mutex_exit(&spa_namespace_lock);
4287 spa_activate(spa, mode);
4290 * Don't start async tasks until we know everything is healthy.
4292 spa_async_suspend(spa);
4294 zpool_get_rewind_policy(config, &policy);
4295 if (policy.zrp_request & ZPOOL_DO_REWIND)
4296 state = SPA_LOAD_RECOVER;
4299 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
4300 * because the user-supplied config is actually the one to trust when
4303 if (state != SPA_LOAD_RECOVER)
4304 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
4306 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
4307 policy.zrp_request);
4310 * Propagate anything learned while loading the pool and pass it
4311 * back to caller (i.e. rewind info, missing devices, etc).
4313 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4314 spa->spa_load_info) == 0);
4316 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4318 * Toss any existing sparelist, as it doesn't have any validity
4319 * anymore, and conflicts with spa_has_spare().
4321 if (spa->spa_spares.sav_config) {
4322 nvlist_free(spa->spa_spares.sav_config);
4323 spa->spa_spares.sav_config = NULL;
4324 spa_load_spares(spa);
4326 if (spa->spa_l2cache.sav_config) {
4327 nvlist_free(spa->spa_l2cache.sav_config);
4328 spa->spa_l2cache.sav_config = NULL;
4329 spa_load_l2cache(spa);
4332 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4335 error = spa_validate_aux(spa, nvroot, -1ULL,
4338 error = spa_validate_aux(spa, nvroot, -1ULL,
4339 VDEV_ALLOC_L2CACHE);
4340 spa_config_exit(spa, SCL_ALL, FTAG);
4343 spa_configfile_set(spa, props, B_FALSE);
4345 if (error != 0 || (props && spa_writeable(spa) &&
4346 (error = spa_prop_set(spa, props)))) {
4348 spa_deactivate(spa);
4350 mutex_exit(&spa_namespace_lock);
4354 spa_async_resume(spa);
4357 * Override any spares and level 2 cache devices as specified by
4358 * the user, as these may have correct device names/devids, etc.
4360 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
4361 &spares, &nspares) == 0) {
4362 if (spa->spa_spares.sav_config)
4363 VERIFY(nvlist_remove(spa->spa_spares.sav_config,
4364 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
4366 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
4367 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4368 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
4369 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
4370 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4371 spa_load_spares(spa);
4372 spa_config_exit(spa, SCL_ALL, FTAG);
4373 spa->spa_spares.sav_sync = B_TRUE;
4375 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
4376 &l2cache, &nl2cache) == 0) {
4377 if (spa->spa_l2cache.sav_config)
4378 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
4379 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
4381 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
4382 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4383 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
4384 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
4385 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4386 spa_load_l2cache(spa);
4387 spa_config_exit(spa, SCL_ALL, FTAG);
4388 spa->spa_l2cache.sav_sync = B_TRUE;
4392 * Check for any removed devices.
4394 if (spa->spa_autoreplace) {
4395 spa_aux_check_removed(&spa->spa_spares);
4396 spa_aux_check_removed(&spa->spa_l2cache);
4399 if (spa_writeable(spa)) {
4401 * Update the config cache to include the newly-imported pool.
4403 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4407 * It's possible that the pool was expanded while it was exported.
4408 * We kick off an async task to handle this for us.
4410 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
4412 spa_history_log_version(spa, "import");
4414 spa_event_notify(spa, NULL, ESC_ZFS_POOL_IMPORT);
4416 mutex_exit(&spa_namespace_lock);
4420 zvol_create_minors(pool);
4427 spa_tryimport(nvlist_t *tryconfig)
4429 nvlist_t *config = NULL;
4435 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
4438 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
4442 * Create and initialize the spa structure.
4444 mutex_enter(&spa_namespace_lock);
4445 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
4446 spa_activate(spa, FREAD);
4449 * Pass off the heavy lifting to spa_load().
4450 * Pass TRUE for mosconfig because the user-supplied config
4451 * is actually the one to trust when doing an import.
4453 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
4456 * If 'tryconfig' was at least parsable, return the current config.
4458 if (spa->spa_root_vdev != NULL) {
4459 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4460 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
4462 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4464 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
4465 spa->spa_uberblock.ub_timestamp) == 0);
4466 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4467 spa->spa_load_info) == 0);
4470 * If the bootfs property exists on this pool then we
4471 * copy it out so that external consumers can tell which
4472 * pools are bootable.
4474 if ((!error || error == EEXIST) && spa->spa_bootfs) {
4475 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4478 * We have to play games with the name since the
4479 * pool was opened as TRYIMPORT_NAME.
4481 if (dsl_dsobj_to_dsname(spa_name(spa),
4482 spa->spa_bootfs, tmpname) == 0) {
4484 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4486 cp = strchr(tmpname, '/');
4488 (void) strlcpy(dsname, tmpname,
4491 (void) snprintf(dsname, MAXPATHLEN,
4492 "%s/%s", poolname, ++cp);
4494 VERIFY(nvlist_add_string(config,
4495 ZPOOL_CONFIG_BOOTFS, dsname) == 0);
4496 kmem_free(dsname, MAXPATHLEN);
4498 kmem_free(tmpname, MAXPATHLEN);
4502 * Add the list of hot spares and level 2 cache devices.
4504 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4505 spa_add_spares(spa, config);
4506 spa_add_l2cache(spa, config);
4507 spa_config_exit(spa, SCL_CONFIG, FTAG);
4511 spa_deactivate(spa);
4513 mutex_exit(&spa_namespace_lock);
4519 * Pool export/destroy
4521 * The act of destroying or exporting a pool is very simple. We make sure there
4522 * is no more pending I/O and any references to the pool are gone. Then, we
4523 * update the pool state and sync all the labels to disk, removing the
4524 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4525 * we don't sync the labels or remove the configuration cache.
4528 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
4529 boolean_t force, boolean_t hardforce)
4536 if (!(spa_mode_global & FWRITE))
4537 return (SET_ERROR(EROFS));
4539 mutex_enter(&spa_namespace_lock);
4540 if ((spa = spa_lookup(pool)) == NULL) {
4541 mutex_exit(&spa_namespace_lock);
4542 return (SET_ERROR(ENOENT));
4546 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4547 * reacquire the namespace lock, and see if we can export.
4549 spa_open_ref(spa, FTAG);
4550 mutex_exit(&spa_namespace_lock);
4551 spa_async_suspend(spa);
4552 mutex_enter(&spa_namespace_lock);
4553 spa_close(spa, FTAG);
4556 * The pool will be in core if it's openable,
4557 * in which case we can modify its state.
4559 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
4561 * Objsets may be open only because they're dirty, so we
4562 * have to force it to sync before checking spa_refcnt.
4564 txg_wait_synced(spa->spa_dsl_pool, 0);
4565 spa_evicting_os_wait(spa);
4568 * A pool cannot be exported or destroyed if there are active
4569 * references. If we are resetting a pool, allow references by
4570 * fault injection handlers.
4572 if (!spa_refcount_zero(spa) ||
4573 (spa->spa_inject_ref != 0 &&
4574 new_state != POOL_STATE_UNINITIALIZED)) {
4575 spa_async_resume(spa);
4576 mutex_exit(&spa_namespace_lock);
4577 return (SET_ERROR(EBUSY));
4581 * A pool cannot be exported if it has an active shared spare.
4582 * This is to prevent other pools stealing the active spare
4583 * from an exported pool. At user's own will, such pool can
4584 * be forcedly exported.
4586 if (!force && new_state == POOL_STATE_EXPORTED &&
4587 spa_has_active_shared_spare(spa)) {
4588 spa_async_resume(spa);
4589 mutex_exit(&spa_namespace_lock);
4590 return (SET_ERROR(EXDEV));
4594 * We want this to be reflected on every label,
4595 * so mark them all dirty. spa_unload() will do the
4596 * final sync that pushes these changes out.
4598 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
4599 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4600 spa->spa_state = new_state;
4601 spa->spa_final_txg = spa_last_synced_txg(spa) +
4603 vdev_config_dirty(spa->spa_root_vdev);
4604 spa_config_exit(spa, SCL_ALL, FTAG);
4608 spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
4610 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4612 spa_deactivate(spa);
4615 if (oldconfig && spa->spa_config)
4616 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
4618 if (new_state != POOL_STATE_UNINITIALIZED) {
4620 spa_config_sync(spa, B_TRUE, B_TRUE);
4623 mutex_exit(&spa_namespace_lock);
4629 * Destroy a storage pool.
4632 spa_destroy(char *pool)
4634 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
4639 * Export a storage pool.
4642 spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
4643 boolean_t hardforce)
4645 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
4650 * Similar to spa_export(), this unloads the spa_t without actually removing it
4651 * from the namespace in any way.
4654 spa_reset(char *pool)
4656 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
4661 * ==========================================================================
4662 * Device manipulation
4663 * ==========================================================================
4667 * Add a device to a storage pool.
4670 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
4674 vdev_t *rvd = spa->spa_root_vdev;
4676 nvlist_t **spares, **l2cache;
4677 uint_t nspares, nl2cache;
4679 ASSERT(spa_writeable(spa));
4681 txg = spa_vdev_enter(spa);
4683 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
4684 VDEV_ALLOC_ADD)) != 0)
4685 return (spa_vdev_exit(spa, NULL, txg, error));
4687 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */
4689 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
4693 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
4697 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
4698 return (spa_vdev_exit(spa, vd, txg, EINVAL));
4700 if (vd->vdev_children != 0 &&
4701 (error = vdev_create(vd, txg, B_FALSE)) != 0)
4702 return (spa_vdev_exit(spa, vd, txg, error));
4705 * We must validate the spares and l2cache devices after checking the
4706 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
4708 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
4709 return (spa_vdev_exit(spa, vd, txg, error));
4712 * Transfer each new top-level vdev from vd to rvd.
4714 for (int c = 0; c < vd->vdev_children; c++) {
4717 * Set the vdev id to the first hole, if one exists.
4719 for (id = 0; id < rvd->vdev_children; id++) {
4720 if (rvd->vdev_child[id]->vdev_ishole) {
4721 vdev_free(rvd->vdev_child[id]);
4725 tvd = vd->vdev_child[c];
4726 vdev_remove_child(vd, tvd);
4728 vdev_add_child(rvd, tvd);
4729 vdev_config_dirty(tvd);
4733 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
4734 ZPOOL_CONFIG_SPARES);
4735 spa_load_spares(spa);
4736 spa->spa_spares.sav_sync = B_TRUE;
4739 if (nl2cache != 0) {
4740 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
4741 ZPOOL_CONFIG_L2CACHE);
4742 spa_load_l2cache(spa);
4743 spa->spa_l2cache.sav_sync = B_TRUE;
4747 * We have to be careful when adding new vdevs to an existing pool.
4748 * If other threads start allocating from these vdevs before we
4749 * sync the config cache, and we lose power, then upon reboot we may
4750 * fail to open the pool because there are DVAs that the config cache
4751 * can't translate. Therefore, we first add the vdevs without
4752 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4753 * and then let spa_config_update() initialize the new metaslabs.
4755 * spa_load() checks for added-but-not-initialized vdevs, so that
4756 * if we lose power at any point in this sequence, the remaining
4757 * steps will be completed the next time we load the pool.
4759 (void) spa_vdev_exit(spa, vd, txg, 0);
4761 mutex_enter(&spa_namespace_lock);
4762 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4763 spa_event_notify(spa, NULL, ESC_ZFS_VDEV_ADD);
4764 mutex_exit(&spa_namespace_lock);
4770 * Attach a device to a mirror. The arguments are the path to any device
4771 * in the mirror, and the nvroot for the new device. If the path specifies
4772 * a device that is not mirrored, we automatically insert the mirror vdev.
4774 * If 'replacing' is specified, the new device is intended to replace the
4775 * existing device; in this case the two devices are made into their own
4776 * mirror using the 'replacing' vdev, which is functionally identical to
4777 * the mirror vdev (it actually reuses all the same ops) but has a few
4778 * extra rules: you can't attach to it after it's been created, and upon
4779 * completion of resilvering, the first disk (the one being replaced)
4780 * is automatically detached.
4783 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
4785 uint64_t txg, dtl_max_txg;
4786 vdev_t *rvd = spa->spa_root_vdev;
4787 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
4789 char *oldvdpath, *newvdpath;
4793 ASSERT(spa_writeable(spa));
4795 txg = spa_vdev_enter(spa);
4797 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
4800 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4802 if (!oldvd->vdev_ops->vdev_op_leaf)
4803 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4805 pvd = oldvd->vdev_parent;
4807 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
4808 VDEV_ALLOC_ATTACH)) != 0)
4809 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4811 if (newrootvd->vdev_children != 1)
4812 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4814 newvd = newrootvd->vdev_child[0];
4816 if (!newvd->vdev_ops->vdev_op_leaf)
4817 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4819 if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
4820 return (spa_vdev_exit(spa, newrootvd, txg, error));
4823 * Spares can't replace logs
4825 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
4826 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4830 * For attach, the only allowable parent is a mirror or the root
4833 if (pvd->vdev_ops != &vdev_mirror_ops &&
4834 pvd->vdev_ops != &vdev_root_ops)
4835 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4837 pvops = &vdev_mirror_ops;
4840 * Active hot spares can only be replaced by inactive hot
4843 if (pvd->vdev_ops == &vdev_spare_ops &&
4844 oldvd->vdev_isspare &&
4845 !spa_has_spare(spa, newvd->vdev_guid))
4846 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4849 * If the source is a hot spare, and the parent isn't already a
4850 * spare, then we want to create a new hot spare. Otherwise, we
4851 * want to create a replacing vdev. The user is not allowed to
4852 * attach to a spared vdev child unless the 'isspare' state is
4853 * the same (spare replaces spare, non-spare replaces
4856 if (pvd->vdev_ops == &vdev_replacing_ops &&
4857 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
4858 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4859 } else if (pvd->vdev_ops == &vdev_spare_ops &&
4860 newvd->vdev_isspare != oldvd->vdev_isspare) {
4861 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4864 if (newvd->vdev_isspare)
4865 pvops = &vdev_spare_ops;
4867 pvops = &vdev_replacing_ops;
4871 * Make sure the new device is big enough.
4873 if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
4874 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
4877 * The new device cannot have a higher alignment requirement
4878 * than the top-level vdev.
4880 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4881 return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4884 * If this is an in-place replacement, update oldvd's path and devid
4885 * to make it distinguishable from newvd, and unopenable from now on.
4887 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4888 spa_strfree(oldvd->vdev_path);
4889 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4891 (void) sprintf(oldvd->vdev_path, "%s/%s",
4892 newvd->vdev_path, "old");
4893 if (oldvd->vdev_devid != NULL) {
4894 spa_strfree(oldvd->vdev_devid);
4895 oldvd->vdev_devid = NULL;
4899 /* mark the device being resilvered */
4900 newvd->vdev_resilver_txg = txg;
4903 * If the parent is not a mirror, or if we're replacing, insert the new
4904 * mirror/replacing/spare vdev above oldvd.
4906 if (pvd->vdev_ops != pvops)
4907 pvd = vdev_add_parent(oldvd, pvops);
4909 ASSERT(pvd->vdev_top->vdev_parent == rvd);
4910 ASSERT(pvd->vdev_ops == pvops);
4911 ASSERT(oldvd->vdev_parent == pvd);
4914 * Extract the new device from its root and add it to pvd.
4916 vdev_remove_child(newrootvd, newvd);
4917 newvd->vdev_id = pvd->vdev_children;
4918 newvd->vdev_crtxg = oldvd->vdev_crtxg;
4919 vdev_add_child(pvd, newvd);
4921 tvd = newvd->vdev_top;
4922 ASSERT(pvd->vdev_top == tvd);
4923 ASSERT(tvd->vdev_parent == rvd);
4925 vdev_config_dirty(tvd);
4928 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4929 * for any dmu_sync-ed blocks. It will propagate upward when
4930 * spa_vdev_exit() calls vdev_dtl_reassess().
4932 dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4934 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4935 dtl_max_txg - TXG_INITIAL);
4937 if (newvd->vdev_isspare) {
4938 spa_spare_activate(newvd);
4939 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
4942 oldvdpath = spa_strdup(oldvd->vdev_path);
4943 newvdpath = spa_strdup(newvd->vdev_path);
4944 newvd_isspare = newvd->vdev_isspare;
4947 * Mark newvd's DTL dirty in this txg.
4949 vdev_dirty(tvd, VDD_DTL, newvd, txg);
4952 * Schedule the resilver to restart in the future. We do this to
4953 * ensure that dmu_sync-ed blocks have been stitched into the
4954 * respective datasets.
4956 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4958 if (spa->spa_bootfs)
4959 spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
4961 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_ATTACH);
4966 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4968 spa_history_log_internal(spa, "vdev attach", NULL,
4969 "%s vdev=%s %s vdev=%s",
4970 replacing && newvd_isspare ? "spare in" :
4971 replacing ? "replace" : "attach", newvdpath,
4972 replacing ? "for" : "to", oldvdpath);
4974 spa_strfree(oldvdpath);
4975 spa_strfree(newvdpath);
4981 * Detach a device from a mirror or replacing vdev.
4983 * If 'replace_done' is specified, only detach if the parent
4984 * is a replacing vdev.
4987 spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4991 vdev_t *rvd = spa->spa_root_vdev;
4992 vdev_t *vd, *pvd, *cvd, *tvd;
4993 boolean_t unspare = B_FALSE;
4994 uint64_t unspare_guid = 0;
4997 ASSERT(spa_writeable(spa));
4999 txg = spa_vdev_enter(spa);
5001 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5004 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
5006 if (!vd->vdev_ops->vdev_op_leaf)
5007 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
5009 pvd = vd->vdev_parent;
5012 * If the parent/child relationship is not as expected, don't do it.
5013 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
5014 * vdev that's replacing B with C. The user's intent in replacing
5015 * is to go from M(A,B) to M(A,C). If the user decides to cancel
5016 * the replace by detaching C, the expected behavior is to end up
5017 * M(A,B). But suppose that right after deciding to detach C,
5018 * the replacement of B completes. We would have M(A,C), and then
5019 * ask to detach C, which would leave us with just A -- not what
5020 * the user wanted. To prevent this, we make sure that the
5021 * parent/child relationship hasn't changed -- in this example,
5022 * that C's parent is still the replacing vdev R.
5024 if (pvd->vdev_guid != pguid && pguid != 0)
5025 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
5028 * Only 'replacing' or 'spare' vdevs can be replaced.
5030 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
5031 pvd->vdev_ops != &vdev_spare_ops)
5032 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
5034 ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
5035 spa_version(spa) >= SPA_VERSION_SPARES);
5038 * Only mirror, replacing, and spare vdevs support detach.
5040 if (pvd->vdev_ops != &vdev_replacing_ops &&
5041 pvd->vdev_ops != &vdev_mirror_ops &&
5042 pvd->vdev_ops != &vdev_spare_ops)
5043 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
5046 * If this device has the only valid copy of some data,
5047 * we cannot safely detach it.
5049 if (vdev_dtl_required(vd))
5050 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
5052 ASSERT(pvd->vdev_children >= 2);
5055 * If we are detaching the second disk from a replacing vdev, then
5056 * check to see if we changed the original vdev's path to have "/old"
5057 * at the end in spa_vdev_attach(). If so, undo that change now.
5059 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
5060 vd->vdev_path != NULL) {
5061 size_t len = strlen(vd->vdev_path);
5063 for (int c = 0; c < pvd->vdev_children; c++) {
5064 cvd = pvd->vdev_child[c];
5066 if (cvd == vd || cvd->vdev_path == NULL)
5069 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
5070 strcmp(cvd->vdev_path + len, "/old") == 0) {
5071 spa_strfree(cvd->vdev_path);
5072 cvd->vdev_path = spa_strdup(vd->vdev_path);
5079 * If we are detaching the original disk from a spare, then it implies
5080 * that the spare should become a real disk, and be removed from the
5081 * active spare list for the pool.
5083 if (pvd->vdev_ops == &vdev_spare_ops &&
5085 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
5089 * Erase the disk labels so the disk can be used for other things.
5090 * This must be done after all other error cases are handled,
5091 * but before we disembowel vd (so we can still do I/O to it).
5092 * But if we can't do it, don't treat the error as fatal --
5093 * it may be that the unwritability of the disk is the reason
5094 * it's being detached!
5096 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5099 * Remove vd from its parent and compact the parent's children.
5101 vdev_remove_child(pvd, vd);
5102 vdev_compact_children(pvd);
5105 * Remember one of the remaining children so we can get tvd below.
5107 cvd = pvd->vdev_child[pvd->vdev_children - 1];
5110 * If we need to remove the remaining child from the list of hot spares,
5111 * do it now, marking the vdev as no longer a spare in the process.
5112 * We must do this before vdev_remove_parent(), because that can
5113 * change the GUID if it creates a new toplevel GUID. For a similar
5114 * reason, we must remove the spare now, in the same txg as the detach;
5115 * otherwise someone could attach a new sibling, change the GUID, and
5116 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
5119 ASSERT(cvd->vdev_isspare);
5120 spa_spare_remove(cvd);
5121 unspare_guid = cvd->vdev_guid;
5122 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
5123 cvd->vdev_unspare = B_TRUE;
5127 * If the parent mirror/replacing vdev only has one child,
5128 * the parent is no longer needed. Remove it from the tree.
5130 if (pvd->vdev_children == 1) {
5131 if (pvd->vdev_ops == &vdev_spare_ops)
5132 cvd->vdev_unspare = B_FALSE;
5133 vdev_remove_parent(cvd);
5138 * We don't set tvd until now because the parent we just removed
5139 * may have been the previous top-level vdev.
5141 tvd = cvd->vdev_top;
5142 ASSERT(tvd->vdev_parent == rvd);
5145 * Reevaluate the parent vdev state.
5147 vdev_propagate_state(cvd);
5150 * If the 'autoexpand' property is set on the pool then automatically
5151 * try to expand the size of the pool. For example if the device we
5152 * just detached was smaller than the others, it may be possible to
5153 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
5154 * first so that we can obtain the updated sizes of the leaf vdevs.
5156 if (spa->spa_autoexpand) {
5158 vdev_expand(tvd, txg);
5161 vdev_config_dirty(tvd);
5164 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
5165 * vd->vdev_detached is set and free vd's DTL object in syncing context.
5166 * But first make sure we're not on any *other* txg's DTL list, to
5167 * prevent vd from being accessed after it's freed.
5169 vdpath = spa_strdup(vd->vdev_path);
5170 for (int t = 0; t < TXG_SIZE; t++)
5171 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
5172 vd->vdev_detached = B_TRUE;
5173 vdev_dirty(tvd, VDD_DTL, vd, txg);
5175 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
5177 /* hang on to the spa before we release the lock */
5178 spa_open_ref(spa, FTAG);
5180 error = spa_vdev_exit(spa, vd, txg, 0);
5182 spa_history_log_internal(spa, "detach", NULL,
5184 spa_strfree(vdpath);
5187 * If this was the removal of the original device in a hot spare vdev,
5188 * then we want to go through and remove the device from the hot spare
5189 * list of every other pool.
5192 spa_t *altspa = NULL;
5194 mutex_enter(&spa_namespace_lock);
5195 while ((altspa = spa_next(altspa)) != NULL) {
5196 if (altspa->spa_state != POOL_STATE_ACTIVE ||
5200 spa_open_ref(altspa, FTAG);
5201 mutex_exit(&spa_namespace_lock);
5202 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
5203 mutex_enter(&spa_namespace_lock);
5204 spa_close(altspa, FTAG);
5206 mutex_exit(&spa_namespace_lock);
5208 /* search the rest of the vdevs for spares to remove */
5209 spa_vdev_resilver_done(spa);
5212 /* all done with the spa; OK to release */
5213 mutex_enter(&spa_namespace_lock);
5214 spa_close(spa, FTAG);
5215 mutex_exit(&spa_namespace_lock);
5221 * Split a set of devices from their mirrors, and create a new pool from them.
5224 spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
5225 nvlist_t *props, boolean_t exp)
5228 uint64_t txg, *glist;
5230 uint_t c, children, lastlog;
5231 nvlist_t **child, *nvl, *tmp;
5233 char *altroot = NULL;
5234 vdev_t *rvd, **vml = NULL; /* vdev modify list */
5235 boolean_t activate_slog;
5237 ASSERT(spa_writeable(spa));
5239 txg = spa_vdev_enter(spa);
5241 /* clear the log and flush everything up to now */
5242 activate_slog = spa_passivate_log(spa);
5243 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5244 error = spa_offline_log(spa);
5245 txg = spa_vdev_config_enter(spa);
5248 spa_activate_log(spa);
5251 return (spa_vdev_exit(spa, NULL, txg, error));
5253 /* check new spa name before going any further */
5254 if (spa_lookup(newname) != NULL)
5255 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
5258 * scan through all the children to ensure they're all mirrors
5260 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
5261 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
5263 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5265 /* first, check to ensure we've got the right child count */
5266 rvd = spa->spa_root_vdev;
5268 for (c = 0; c < rvd->vdev_children; c++) {
5269 vdev_t *vd = rvd->vdev_child[c];
5271 /* don't count the holes & logs as children */
5272 if (vd->vdev_islog || vd->vdev_ishole) {
5280 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
5281 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5283 /* next, ensure no spare or cache devices are part of the split */
5284 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
5285 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
5286 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5288 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
5289 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
5291 /* then, loop over each vdev and validate it */
5292 for (c = 0; c < children; c++) {
5293 uint64_t is_hole = 0;
5295 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
5299 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
5300 spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
5303 error = SET_ERROR(EINVAL);
5308 /* which disk is going to be split? */
5309 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
5311 error = SET_ERROR(EINVAL);
5315 /* look it up in the spa */
5316 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
5317 if (vml[c] == NULL) {
5318 error = SET_ERROR(ENODEV);
5322 /* make sure there's nothing stopping the split */
5323 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
5324 vml[c]->vdev_islog ||
5325 vml[c]->vdev_ishole ||
5326 vml[c]->vdev_isspare ||
5327 vml[c]->vdev_isl2cache ||
5328 !vdev_writeable(vml[c]) ||
5329 vml[c]->vdev_children != 0 ||
5330 vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
5331 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
5332 error = SET_ERROR(EINVAL);
5336 if (vdev_dtl_required(vml[c])) {
5337 error = SET_ERROR(EBUSY);
5341 /* we need certain info from the top level */
5342 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
5343 vml[c]->vdev_top->vdev_ms_array) == 0);
5344 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
5345 vml[c]->vdev_top->vdev_ms_shift) == 0);
5346 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
5347 vml[c]->vdev_top->vdev_asize) == 0);
5348 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
5349 vml[c]->vdev_top->vdev_ashift) == 0);
5353 kmem_free(vml, children * sizeof (vdev_t *));
5354 kmem_free(glist, children * sizeof (uint64_t));
5355 return (spa_vdev_exit(spa, NULL, txg, error));
5358 /* stop writers from using the disks */
5359 for (c = 0; c < children; c++) {
5361 vml[c]->vdev_offline = B_TRUE;
5363 vdev_reopen(spa->spa_root_vdev);
5366 * Temporarily record the splitting vdevs in the spa config. This
5367 * will disappear once the config is regenerated.
5369 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5370 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
5371 glist, children) == 0);
5372 kmem_free(glist, children * sizeof (uint64_t));
5374 mutex_enter(&spa->spa_props_lock);
5375 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
5377 mutex_exit(&spa->spa_props_lock);
5378 spa->spa_config_splitting = nvl;
5379 vdev_config_dirty(spa->spa_root_vdev);
5381 /* configure and create the new pool */
5382 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
5383 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
5384 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
5385 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5386 spa_version(spa)) == 0);
5387 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
5388 spa->spa_config_txg) == 0);
5389 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
5390 spa_generate_guid(NULL)) == 0);
5391 (void) nvlist_lookup_string(props,
5392 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
5394 /* add the new pool to the namespace */
5395 newspa = spa_add(newname, config, altroot);
5396 newspa->spa_config_txg = spa->spa_config_txg;
5397 spa_set_log_state(newspa, SPA_LOG_CLEAR);
5399 /* release the spa config lock, retaining the namespace lock */
5400 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5402 if (zio_injection_enabled)
5403 zio_handle_panic_injection(spa, FTAG, 1);
5405 spa_activate(newspa, spa_mode_global);
5406 spa_async_suspend(newspa);
5409 /* mark that we are creating new spa by splitting */
5410 newspa->spa_splitting_newspa = B_TRUE;
5412 /* create the new pool from the disks of the original pool */
5413 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
5415 newspa->spa_splitting_newspa = B_FALSE;
5420 /* if that worked, generate a real config for the new pool */
5421 if (newspa->spa_root_vdev != NULL) {
5422 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
5423 NV_UNIQUE_NAME, KM_SLEEP) == 0);
5424 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
5425 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
5426 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
5431 if (props != NULL) {
5432 spa_configfile_set(newspa, props, B_FALSE);
5433 error = spa_prop_set(newspa, props);
5438 /* flush everything */
5439 txg = spa_vdev_config_enter(newspa);
5440 vdev_config_dirty(newspa->spa_root_vdev);
5441 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
5443 if (zio_injection_enabled)
5444 zio_handle_panic_injection(spa, FTAG, 2);
5446 spa_async_resume(newspa);
5448 /* finally, update the original pool's config */
5449 txg = spa_vdev_config_enter(spa);
5450 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
5451 error = dmu_tx_assign(tx, TXG_WAIT);
5454 for (c = 0; c < children; c++) {
5455 if (vml[c] != NULL) {
5458 spa_history_log_internal(spa, "detach", tx,
5459 "vdev=%s", vml[c]->vdev_path);
5463 vdev_config_dirty(spa->spa_root_vdev);
5464 spa->spa_config_splitting = NULL;
5468 (void) spa_vdev_exit(spa, NULL, txg, 0);
5470 if (zio_injection_enabled)
5471 zio_handle_panic_injection(spa, FTAG, 3);
5473 /* split is complete; log a history record */
5474 spa_history_log_internal(newspa, "split", NULL,
5475 "from pool %s", spa_name(spa));
5477 kmem_free(vml, children * sizeof (vdev_t *));
5479 /* if we're not going to mount the filesystems in userland, export */
5481 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
5488 spa_deactivate(newspa);
5491 txg = spa_vdev_config_enter(spa);
5493 /* re-online all offlined disks */
5494 for (c = 0; c < children; c++) {
5496 vml[c]->vdev_offline = B_FALSE;
5498 vdev_reopen(spa->spa_root_vdev);
5500 nvlist_free(spa->spa_config_splitting);
5501 spa->spa_config_splitting = NULL;
5502 (void) spa_vdev_exit(spa, NULL, txg, error);
5504 kmem_free(vml, children * sizeof (vdev_t *));
5509 spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
5511 for (int i = 0; i < count; i++) {
5514 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
5517 if (guid == target_guid)
5525 spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
5526 nvlist_t *dev_to_remove)
5528 nvlist_t **newdev = NULL;
5531 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
5533 for (int i = 0, j = 0; i < count; i++) {
5534 if (dev[i] == dev_to_remove)
5536 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
5539 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
5540 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
5542 for (int i = 0; i < count - 1; i++)
5543 nvlist_free(newdev[i]);
5546 kmem_free(newdev, (count - 1) * sizeof (void *));
5550 * Evacuate the device.
5553 spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
5558 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5559 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5560 ASSERT(vd == vd->vdev_top);
5563 * Evacuate the device. We don't hold the config lock as writer
5564 * since we need to do I/O but we do keep the
5565 * spa_namespace_lock held. Once this completes the device
5566 * should no longer have any blocks allocated on it.
5568 if (vd->vdev_islog) {
5569 if (vd->vdev_stat.vs_alloc != 0)
5570 error = spa_offline_log(spa);
5572 error = SET_ERROR(ENOTSUP);
5579 * The evacuation succeeded. Remove any remaining MOS metadata
5580 * associated with this vdev, and wait for these changes to sync.
5582 ASSERT0(vd->vdev_stat.vs_alloc);
5583 txg = spa_vdev_config_enter(spa);
5584 vd->vdev_removing = B_TRUE;
5585 vdev_dirty_leaves(vd, VDD_DTL, txg);
5586 vdev_config_dirty(vd);
5587 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5593 * Complete the removal by cleaning up the namespace.
5596 spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
5598 vdev_t *rvd = spa->spa_root_vdev;
5599 uint64_t id = vd->vdev_id;
5600 boolean_t last_vdev = (id == (rvd->vdev_children - 1));
5602 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5603 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5604 ASSERT(vd == vd->vdev_top);
5607 * Only remove any devices which are empty.
5609 if (vd->vdev_stat.vs_alloc != 0)
5612 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5614 if (list_link_active(&vd->vdev_state_dirty_node))
5615 vdev_state_clean(vd);
5616 if (list_link_active(&vd->vdev_config_dirty_node))
5617 vdev_config_clean(vd);
5622 vdev_compact_children(rvd);
5624 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
5625 vdev_add_child(rvd, vd);
5627 vdev_config_dirty(rvd);
5630 * Reassess the health of our root vdev.
5636 * Remove a device from the pool -
5638 * Removing a device from the vdev namespace requires several steps
5639 * and can take a significant amount of time. As a result we use
5640 * the spa_vdev_config_[enter/exit] functions which allow us to
5641 * grab and release the spa_config_lock while still holding the namespace
5642 * lock. During each step the configuration is synced out.
5644 * Currently, this supports removing only hot spares, slogs, and level 2 ARC
5648 spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
5651 sysevent_t *ev = NULL;
5652 metaslab_group_t *mg;
5653 nvlist_t **spares, **l2cache, *nv;
5655 uint_t nspares, nl2cache;
5657 boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
5659 ASSERT(spa_writeable(spa));
5662 txg = spa_vdev_enter(spa);
5664 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5666 if (spa->spa_spares.sav_vdevs != NULL &&
5667 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5668 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
5669 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
5671 * Only remove the hot spare if it's not currently in use
5674 if (vd == NULL || unspare) {
5676 vd = spa_lookup_by_guid(spa, guid, B_TRUE);
5677 ev = spa_event_create(spa, vd, ESC_ZFS_VDEV_REMOVE_AUX);
5678 spa_vdev_remove_aux(spa->spa_spares.sav_config,
5679 ZPOOL_CONFIG_SPARES, spares, nspares, nv);
5680 spa_load_spares(spa);
5681 spa->spa_spares.sav_sync = B_TRUE;
5683 error = SET_ERROR(EBUSY);
5685 } else if (spa->spa_l2cache.sav_vdevs != NULL &&
5686 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5687 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
5688 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
5690 * Cache devices can always be removed.
5692 vd = spa_lookup_by_guid(spa, guid, B_TRUE);
5693 ev = spa_event_create(spa, vd, ESC_ZFS_VDEV_REMOVE_AUX);
5694 spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
5695 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
5696 spa_load_l2cache(spa);
5697 spa->spa_l2cache.sav_sync = B_TRUE;
5698 } else if (vd != NULL && vd->vdev_islog) {
5700 ASSERT(vd == vd->vdev_top);
5705 * Stop allocating from this vdev.
5707 metaslab_group_passivate(mg);
5710 * Wait for the youngest allocations and frees to sync,
5711 * and then wait for the deferral of those frees to finish.
5713 spa_vdev_config_exit(spa, NULL,
5714 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
5717 * Attempt to evacuate the vdev.
5719 error = spa_vdev_remove_evacuate(spa, vd);
5721 txg = spa_vdev_config_enter(spa);
5724 * If we couldn't evacuate the vdev, unwind.
5727 metaslab_group_activate(mg);
5728 return (spa_vdev_exit(spa, NULL, txg, error));
5732 * Clean up the vdev namespace.
5734 ev = spa_event_create(spa, vd, ESC_ZFS_VDEV_REMOVE_DEV);
5735 spa_vdev_remove_from_namespace(spa, vd);
5737 } else if (vd != NULL) {
5739 * Normal vdevs cannot be removed (yet).
5741 error = SET_ERROR(ENOTSUP);
5744 * There is no vdev of any kind with the specified guid.
5746 error = SET_ERROR(ENOENT);
5750 error = spa_vdev_exit(spa, NULL, txg, error);
5759 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5760 * currently spared, so we can detach it.
5763 spa_vdev_resilver_done_hunt(vdev_t *vd)
5765 vdev_t *newvd, *oldvd;
5767 for (int c = 0; c < vd->vdev_children; c++) {
5768 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
5774 * Check for a completed replacement. We always consider the first
5775 * vdev in the list to be the oldest vdev, and the last one to be
5776 * the newest (see spa_vdev_attach() for how that works). In
5777 * the case where the newest vdev is faulted, we will not automatically
5778 * remove it after a resilver completes. This is OK as it will require
5779 * user intervention to determine which disk the admin wishes to keep.
5781 if (vd->vdev_ops == &vdev_replacing_ops) {
5782 ASSERT(vd->vdev_children > 1);
5784 newvd = vd->vdev_child[vd->vdev_children - 1];
5785 oldvd = vd->vdev_child[0];
5787 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
5788 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5789 !vdev_dtl_required(oldvd))
5794 * Check for a completed resilver with the 'unspare' flag set.
5796 if (vd->vdev_ops == &vdev_spare_ops) {
5797 vdev_t *first = vd->vdev_child[0];
5798 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
5800 if (last->vdev_unspare) {
5803 } else if (first->vdev_unspare) {
5810 if (oldvd != NULL &&
5811 vdev_dtl_empty(newvd, DTL_MISSING) &&
5812 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5813 !vdev_dtl_required(oldvd))
5817 * If there are more than two spares attached to a disk,
5818 * and those spares are not required, then we want to
5819 * attempt to free them up now so that they can be used
5820 * by other pools. Once we're back down to a single
5821 * disk+spare, we stop removing them.
5823 if (vd->vdev_children > 2) {
5824 newvd = vd->vdev_child[1];
5826 if (newvd->vdev_isspare && last->vdev_isspare &&
5827 vdev_dtl_empty(last, DTL_MISSING) &&
5828 vdev_dtl_empty(last, DTL_OUTAGE) &&
5829 !vdev_dtl_required(newvd))
5838 spa_vdev_resilver_done(spa_t *spa)
5840 vdev_t *vd, *pvd, *ppvd;
5841 uint64_t guid, sguid, pguid, ppguid;
5843 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5845 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
5846 pvd = vd->vdev_parent;
5847 ppvd = pvd->vdev_parent;
5848 guid = vd->vdev_guid;
5849 pguid = pvd->vdev_guid;
5850 ppguid = ppvd->vdev_guid;
5853 * If we have just finished replacing a hot spared device, then
5854 * we need to detach the parent's first child (the original hot
5857 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
5858 ppvd->vdev_children == 2) {
5859 ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
5860 sguid = ppvd->vdev_child[1]->vdev_guid;
5862 ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd));
5864 spa_config_exit(spa, SCL_ALL, FTAG);
5865 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
5867 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
5869 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5872 spa_config_exit(spa, SCL_ALL, FTAG);
5876 * Update the stored path or FRU for this vdev.
5879 spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
5883 boolean_t sync = B_FALSE;
5885 ASSERT(spa_writeable(spa));
5887 spa_vdev_state_enter(spa, SCL_ALL);
5889 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
5890 return (spa_vdev_state_exit(spa, NULL, ENOENT));
5892 if (!vd->vdev_ops->vdev_op_leaf)
5893 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
5896 if (strcmp(value, vd->vdev_path) != 0) {
5897 spa_strfree(vd->vdev_path);
5898 vd->vdev_path = spa_strdup(value);
5902 if (vd->vdev_fru == NULL) {
5903 vd->vdev_fru = spa_strdup(value);
5905 } else if (strcmp(value, vd->vdev_fru) != 0) {
5906 spa_strfree(vd->vdev_fru);
5907 vd->vdev_fru = spa_strdup(value);
5912 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
5916 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
5918 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5922 spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5924 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5928 * ==========================================================================
5930 * ==========================================================================
5934 spa_scan_stop(spa_t *spa)
5936 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5937 if (dsl_scan_resilvering(spa->spa_dsl_pool))
5938 return (SET_ERROR(EBUSY));
5939 return (dsl_scan_cancel(spa->spa_dsl_pool));
5943 spa_scan(spa_t *spa, pool_scan_func_t func)
5945 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5947 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5948 return (SET_ERROR(ENOTSUP));
5951 * If a resilver was requested, but there is no DTL on a
5952 * writeable leaf device, we have nothing to do.
5954 if (func == POOL_SCAN_RESILVER &&
5955 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5956 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5960 return (dsl_scan(spa->spa_dsl_pool, func));
5964 * ==========================================================================
5965 * SPA async task processing
5966 * ==========================================================================
5970 spa_async_remove(spa_t *spa, vdev_t *vd)
5972 if (vd->vdev_remove_wanted) {
5973 vd->vdev_remove_wanted = B_FALSE;
5974 vd->vdev_delayed_close = B_FALSE;
5975 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5978 * We want to clear the stats, but we don't want to do a full
5979 * vdev_clear() as that will cause us to throw away
5980 * degraded/faulted state as well as attempt to reopen the
5981 * device, all of which is a waste.
5983 vd->vdev_stat.vs_read_errors = 0;
5984 vd->vdev_stat.vs_write_errors = 0;
5985 vd->vdev_stat.vs_checksum_errors = 0;
5987 vdev_state_dirty(vd->vdev_top);
5988 /* Tell userspace that the vdev is gone. */
5989 zfs_post_remove(spa, vd);
5992 for (int c = 0; c < vd->vdev_children; c++)
5993 spa_async_remove(spa, vd->vdev_child[c]);
5997 spa_async_probe(spa_t *spa, vdev_t *vd)
5999 if (vd->vdev_probe_wanted) {
6000 vd->vdev_probe_wanted = B_FALSE;
6001 vdev_reopen(vd); /* vdev_open() does the actual probe */
6004 for (int c = 0; c < vd->vdev_children; c++)
6005 spa_async_probe(spa, vd->vdev_child[c]);
6009 spa_async_autoexpand(spa_t *spa, vdev_t *vd)
6015 if (!spa->spa_autoexpand)
6018 for (int c = 0; c < vd->vdev_children; c++) {
6019 vdev_t *cvd = vd->vdev_child[c];
6020 spa_async_autoexpand(spa, cvd);
6023 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
6026 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
6027 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
6029 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
6030 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
6032 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
6033 ESC_ZFS_VDEV_AUTOEXPAND, attr, &eid, DDI_SLEEP);
6036 kmem_free(physpath, MAXPATHLEN);
6040 spa_async_thread(void *arg)
6045 ASSERT(spa->spa_sync_on);
6047 mutex_enter(&spa->spa_async_lock);
6048 tasks = spa->spa_async_tasks;
6049 spa->spa_async_tasks &= SPA_ASYNC_REMOVE;
6050 mutex_exit(&spa->spa_async_lock);
6053 * See if the config needs to be updated.
6055 if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
6056 uint64_t old_space, new_space;
6058 mutex_enter(&spa_namespace_lock);
6059 old_space = metaslab_class_get_space(spa_normal_class(spa));
6060 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
6061 new_space = metaslab_class_get_space(spa_normal_class(spa));
6062 mutex_exit(&spa_namespace_lock);
6065 * If the pool grew as a result of the config update,
6066 * then log an internal history event.
6068 if (new_space != old_space) {
6069 spa_history_log_internal(spa, "vdev online", NULL,
6070 "pool '%s' size: %llu(+%llu)",
6071 spa_name(spa), new_space, new_space - old_space);
6075 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
6076 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
6077 spa_async_autoexpand(spa, spa->spa_root_vdev);
6078 spa_config_exit(spa, SCL_CONFIG, FTAG);
6082 * See if any devices need to be probed.
6084 if (tasks & SPA_ASYNC_PROBE) {
6085 spa_vdev_state_enter(spa, SCL_NONE);
6086 spa_async_probe(spa, spa->spa_root_vdev);
6087 (void) spa_vdev_state_exit(spa, NULL, 0);
6091 * If any devices are done replacing, detach them.
6093 if (tasks & SPA_ASYNC_RESILVER_DONE)
6094 spa_vdev_resilver_done(spa);
6097 * Kick off a resilver.
6099 if (tasks & SPA_ASYNC_RESILVER)
6100 dsl_resilver_restart(spa->spa_dsl_pool, 0);
6103 * Let the world know that we're done.
6105 mutex_enter(&spa->spa_async_lock);
6106 spa->spa_async_thread = NULL;
6107 cv_broadcast(&spa->spa_async_cv);
6108 mutex_exit(&spa->spa_async_lock);
6113 spa_async_thread_vd(void *arg)
6118 ASSERT(spa->spa_sync_on);
6120 mutex_enter(&spa->spa_async_lock);
6121 tasks = spa->spa_async_tasks;
6123 spa->spa_async_tasks &= ~SPA_ASYNC_REMOVE;
6124 mutex_exit(&spa->spa_async_lock);
6127 * See if any devices need to be marked REMOVED.
6129 if (tasks & SPA_ASYNC_REMOVE) {
6130 spa_vdev_state_enter(spa, SCL_NONE);
6131 spa_async_remove(spa, spa->spa_root_vdev);
6132 for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
6133 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
6134 for (int i = 0; i < spa->spa_spares.sav_count; i++)
6135 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
6136 (void) spa_vdev_state_exit(spa, NULL, 0);
6140 * Let the world know that we're done.
6142 mutex_enter(&spa->spa_async_lock);
6143 tasks = spa->spa_async_tasks;
6144 if ((tasks & SPA_ASYNC_REMOVE) != 0)
6146 spa->spa_async_thread_vd = NULL;
6147 cv_broadcast(&spa->spa_async_cv);
6148 mutex_exit(&spa->spa_async_lock);
6153 spa_async_suspend(spa_t *spa)
6155 mutex_enter(&spa->spa_async_lock);
6156 spa->spa_async_suspended++;
6157 while (spa->spa_async_thread != NULL &&
6158 spa->spa_async_thread_vd != NULL)
6159 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
6160 mutex_exit(&spa->spa_async_lock);
6164 spa_async_resume(spa_t *spa)
6166 mutex_enter(&spa->spa_async_lock);
6167 ASSERT(spa->spa_async_suspended != 0);
6168 spa->spa_async_suspended--;
6169 mutex_exit(&spa->spa_async_lock);
6173 spa_async_tasks_pending(spa_t *spa)
6175 uint_t non_config_tasks;
6177 boolean_t config_task_suspended;
6179 non_config_tasks = spa->spa_async_tasks & ~(SPA_ASYNC_CONFIG_UPDATE |
6181 config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE;
6182 if (spa->spa_ccw_fail_time == 0) {
6183 config_task_suspended = B_FALSE;
6185 config_task_suspended =
6186 (gethrtime() - spa->spa_ccw_fail_time) <
6187 (zfs_ccw_retry_interval * NANOSEC);
6190 return (non_config_tasks || (config_task && !config_task_suspended));
6194 spa_async_dispatch(spa_t *spa)
6196 mutex_enter(&spa->spa_async_lock);
6197 if (spa_async_tasks_pending(spa) &&
6198 !spa->spa_async_suspended &&
6199 spa->spa_async_thread == NULL &&
6201 spa->spa_async_thread = thread_create(NULL, 0,
6202 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
6203 mutex_exit(&spa->spa_async_lock);
6207 spa_async_dispatch_vd(spa_t *spa)
6209 mutex_enter(&spa->spa_async_lock);
6210 if ((spa->spa_async_tasks & SPA_ASYNC_REMOVE) != 0 &&
6211 !spa->spa_async_suspended &&
6212 spa->spa_async_thread_vd == NULL &&
6214 spa->spa_async_thread_vd = thread_create(NULL, 0,
6215 spa_async_thread_vd, spa, 0, &p0, TS_RUN, maxclsyspri);
6216 mutex_exit(&spa->spa_async_lock);
6220 spa_async_request(spa_t *spa, int task)
6222 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
6223 mutex_enter(&spa->spa_async_lock);
6224 spa->spa_async_tasks |= task;
6225 mutex_exit(&spa->spa_async_lock);
6226 spa_async_dispatch_vd(spa);
6230 * ==========================================================================
6231 * SPA syncing routines
6232 * ==========================================================================
6236 bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
6239 bpobj_enqueue(bpo, bp, tx);
6244 spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
6248 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
6249 BP_GET_PSIZE(bp), zio->io_flags));
6254 * Note: this simple function is not inlined to make it easier to dtrace the
6255 * amount of time spent syncing frees.
6258 spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx)
6260 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6261 bplist_iterate(bpl, spa_free_sync_cb, zio, tx);
6262 VERIFY(zio_wait(zio) == 0);
6266 * Note: this simple function is not inlined to make it easier to dtrace the
6267 * amount of time spent syncing deferred frees.
6270 spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx)
6272 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6273 VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj,
6274 spa_free_sync_cb, zio, tx), ==, 0);
6275 VERIFY0(zio_wait(zio));
6280 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
6282 char *packed = NULL;
6287 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
6290 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
6291 * information. This avoids the dmu_buf_will_dirty() path and
6292 * saves us a pre-read to get data we don't actually care about.
6294 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
6295 packed = kmem_alloc(bufsize, KM_SLEEP);
6297 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
6299 bzero(packed + nvsize, bufsize - nvsize);
6301 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
6303 kmem_free(packed, bufsize);
6305 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
6306 dmu_buf_will_dirty(db, tx);
6307 *(uint64_t *)db->db_data = nvsize;
6308 dmu_buf_rele(db, FTAG);
6312 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
6313 const char *config, const char *entry)
6323 * Update the MOS nvlist describing the list of available devices.
6324 * spa_validate_aux() will have already made sure this nvlist is
6325 * valid and the vdevs are labeled appropriately.
6327 if (sav->sav_object == 0) {
6328 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
6329 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
6330 sizeof (uint64_t), tx);
6331 VERIFY(zap_update(spa->spa_meta_objset,
6332 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
6333 &sav->sav_object, tx) == 0);
6336 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
6337 if (sav->sav_count == 0) {
6338 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
6340 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
6341 for (i = 0; i < sav->sav_count; i++)
6342 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
6343 B_FALSE, VDEV_CONFIG_L2CACHE);
6344 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
6345 sav->sav_count) == 0);
6346 for (i = 0; i < sav->sav_count; i++)
6347 nvlist_free(list[i]);
6348 kmem_free(list, sav->sav_count * sizeof (void *));
6351 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
6352 nvlist_free(nvroot);
6354 sav->sav_sync = B_FALSE;
6358 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
6362 if (list_is_empty(&spa->spa_config_dirty_list))
6365 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6367 config = spa_config_generate(spa, spa->spa_root_vdev,
6368 dmu_tx_get_txg(tx), B_FALSE);
6371 * If we're upgrading the spa version then make sure that
6372 * the config object gets updated with the correct version.
6374 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
6375 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
6376 spa->spa_uberblock.ub_version);
6378 spa_config_exit(spa, SCL_STATE, FTAG);
6380 nvlist_free(spa->spa_config_syncing);
6381 spa->spa_config_syncing = config;
6383 spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
6387 spa_sync_version(void *arg, dmu_tx_t *tx)
6389 uint64_t *versionp = arg;
6390 uint64_t version = *versionp;
6391 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
6394 * Setting the version is special cased when first creating the pool.
6396 ASSERT(tx->tx_txg != TXG_INITIAL);
6398 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
6399 ASSERT(version >= spa_version(spa));
6401 spa->spa_uberblock.ub_version = version;
6402 vdev_config_dirty(spa->spa_root_vdev);
6403 spa_history_log_internal(spa, "set", tx, "version=%lld", version);
6407 * Set zpool properties.
6410 spa_sync_props(void *arg, dmu_tx_t *tx)
6412 nvlist_t *nvp = arg;
6413 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
6414 objset_t *mos = spa->spa_meta_objset;
6415 nvpair_t *elem = NULL;
6417 mutex_enter(&spa->spa_props_lock);
6419 while ((elem = nvlist_next_nvpair(nvp, elem))) {
6421 char *strval, *fname;
6423 const char *propname;
6424 zprop_type_t proptype;
6427 switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
6430 * We checked this earlier in spa_prop_validate().
6432 ASSERT(zpool_prop_feature(nvpair_name(elem)));
6434 fname = strchr(nvpair_name(elem), '@') + 1;
6435 VERIFY0(zfeature_lookup_name(fname, &fid));
6437 spa_feature_enable(spa, fid, tx);
6438 spa_history_log_internal(spa, "set", tx,
6439 "%s=enabled", nvpair_name(elem));
6442 case ZPOOL_PROP_VERSION:
6443 intval = fnvpair_value_uint64(elem);
6445 * The version is synced seperatly before other
6446 * properties and should be correct by now.
6448 ASSERT3U(spa_version(spa), >=, intval);
6451 case ZPOOL_PROP_ALTROOT:
6453 * 'altroot' is a non-persistent property. It should
6454 * have been set temporarily at creation or import time.
6456 ASSERT(spa->spa_root != NULL);
6459 case ZPOOL_PROP_READONLY:
6460 case ZPOOL_PROP_CACHEFILE:
6462 * 'readonly' and 'cachefile' are also non-persisitent
6466 case ZPOOL_PROP_COMMENT:
6467 strval = fnvpair_value_string(elem);
6468 if (spa->spa_comment != NULL)
6469 spa_strfree(spa->spa_comment);
6470 spa->spa_comment = spa_strdup(strval);
6472 * We need to dirty the configuration on all the vdevs
6473 * so that their labels get updated. It's unnecessary
6474 * to do this for pool creation since the vdev's
6475 * configuratoin has already been dirtied.
6477 if (tx->tx_txg != TXG_INITIAL)
6478 vdev_config_dirty(spa->spa_root_vdev);
6479 spa_history_log_internal(spa, "set", tx,
6480 "%s=%s", nvpair_name(elem), strval);
6484 * Set pool property values in the poolprops mos object.
6486 if (spa->spa_pool_props_object == 0) {
6487 spa->spa_pool_props_object =
6488 zap_create_link(mos, DMU_OT_POOL_PROPS,
6489 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
6493 /* normalize the property name */
6494 propname = zpool_prop_to_name(prop);
6495 proptype = zpool_prop_get_type(prop);
6497 if (nvpair_type(elem) == DATA_TYPE_STRING) {
6498 ASSERT(proptype == PROP_TYPE_STRING);
6499 strval = fnvpair_value_string(elem);
6500 VERIFY0(zap_update(mos,
6501 spa->spa_pool_props_object, propname,
6502 1, strlen(strval) + 1, strval, tx));
6503 spa_history_log_internal(spa, "set", tx,
6504 "%s=%s", nvpair_name(elem), strval);
6505 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
6506 intval = fnvpair_value_uint64(elem);
6508 if (proptype == PROP_TYPE_INDEX) {
6510 VERIFY0(zpool_prop_index_to_string(
6511 prop, intval, &unused));
6513 VERIFY0(zap_update(mos,
6514 spa->spa_pool_props_object, propname,
6515 8, 1, &intval, tx));
6516 spa_history_log_internal(spa, "set", tx,
6517 "%s=%lld", nvpair_name(elem), intval);
6519 ASSERT(0); /* not allowed */
6523 case ZPOOL_PROP_DELEGATION:
6524 spa->spa_delegation = intval;
6526 case ZPOOL_PROP_BOOTFS:
6527 spa->spa_bootfs = intval;
6529 case ZPOOL_PROP_FAILUREMODE:
6530 spa->spa_failmode = intval;
6532 case ZPOOL_PROP_AUTOEXPAND:
6533 spa->spa_autoexpand = intval;
6534 if (tx->tx_txg != TXG_INITIAL)
6535 spa_async_request(spa,
6536 SPA_ASYNC_AUTOEXPAND);
6538 case ZPOOL_PROP_DEDUPDITTO:
6539 spa->spa_dedup_ditto = intval;
6548 mutex_exit(&spa->spa_props_lock);
6552 * Perform one-time upgrade on-disk changes. spa_version() does not
6553 * reflect the new version this txg, so there must be no changes this
6554 * txg to anything that the upgrade code depends on after it executes.
6555 * Therefore this must be called after dsl_pool_sync() does the sync
6559 spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
6561 dsl_pool_t *dp = spa->spa_dsl_pool;
6563 ASSERT(spa->spa_sync_pass == 1);
6565 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
6567 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
6568 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
6569 dsl_pool_create_origin(dp, tx);
6571 /* Keeping the origin open increases spa_minref */
6572 spa->spa_minref += 3;
6575 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
6576 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
6577 dsl_pool_upgrade_clones(dp, tx);
6580 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
6581 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
6582 dsl_pool_upgrade_dir_clones(dp, tx);
6584 /* Keeping the freedir open increases spa_minref */
6585 spa->spa_minref += 3;
6588 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES &&
6589 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
6590 spa_feature_create_zap_objects(spa, tx);
6594 * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable
6595 * when possibility to use lz4 compression for metadata was added
6596 * Old pools that have this feature enabled must be upgraded to have
6597 * this feature active
6599 if (spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
6600 boolean_t lz4_en = spa_feature_is_enabled(spa,
6601 SPA_FEATURE_LZ4_COMPRESS);
6602 boolean_t lz4_ac = spa_feature_is_active(spa,
6603 SPA_FEATURE_LZ4_COMPRESS);
6605 if (lz4_en && !lz4_ac)
6606 spa_feature_incr(spa, SPA_FEATURE_LZ4_COMPRESS, tx);
6610 * If we haven't written the salt, do so now. Note that the
6611 * feature may not be activated yet, but that's fine since
6612 * the presence of this ZAP entry is backwards compatible.
6614 if (zap_contains(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
6615 DMU_POOL_CHECKSUM_SALT) == ENOENT) {
6616 VERIFY0(zap_add(spa->spa_meta_objset,
6617 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CHECKSUM_SALT, 1,
6618 sizeof (spa->spa_cksum_salt.zcs_bytes),
6619 spa->spa_cksum_salt.zcs_bytes, tx));
6622 rrw_exit(&dp->dp_config_rwlock, FTAG);
6626 * Sync the specified transaction group. New blocks may be dirtied as
6627 * part of the process, so we iterate until it converges.
6630 spa_sync(spa_t *spa, uint64_t txg)
6632 dsl_pool_t *dp = spa->spa_dsl_pool;
6633 objset_t *mos = spa->spa_meta_objset;
6634 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
6635 vdev_t *rvd = spa->spa_root_vdev;
6639 uint32_t max_queue_depth = zfs_vdev_async_write_max_active *
6640 zfs_vdev_queue_depth_pct / 100;
6642 VERIFY(spa_writeable(spa));
6645 * Lock out configuration changes.
6647 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
6649 spa->spa_syncing_txg = txg;
6650 spa->spa_sync_pass = 0;
6652 mutex_enter(&spa->spa_alloc_lock);
6653 VERIFY0(avl_numnodes(&spa->spa_alloc_tree));
6654 mutex_exit(&spa->spa_alloc_lock);
6657 * If there are any pending vdev state changes, convert them
6658 * into config changes that go out with this transaction group.
6660 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6661 while (list_head(&spa->spa_state_dirty_list) != NULL) {
6663 * We need the write lock here because, for aux vdevs,
6664 * calling vdev_config_dirty() modifies sav_config.
6665 * This is ugly and will become unnecessary when we
6666 * eliminate the aux vdev wart by integrating all vdevs
6667 * into the root vdev tree.
6669 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6670 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
6671 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
6672 vdev_state_clean(vd);
6673 vdev_config_dirty(vd);
6675 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6676 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
6678 spa_config_exit(spa, SCL_STATE, FTAG);
6680 tx = dmu_tx_create_assigned(dp, txg);
6682 spa->spa_sync_starttime = gethrtime();
6684 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid,
6685 spa->spa_sync_starttime + spa->spa_deadman_synctime));
6686 #else /* !illumos */
6688 callout_reset(&spa->spa_deadman_cycid,
6689 hz * spa->spa_deadman_synctime / NANOSEC, spa_deadman, spa);
6691 #endif /* illumos */
6694 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
6695 * set spa_deflate if we have no raid-z vdevs.
6697 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
6698 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
6701 for (i = 0; i < rvd->vdev_children; i++) {
6702 vd = rvd->vdev_child[i];
6703 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
6706 if (i == rvd->vdev_children) {
6707 spa->spa_deflate = TRUE;
6708 VERIFY(0 == zap_add(spa->spa_meta_objset,
6709 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
6710 sizeof (uint64_t), 1, &spa->spa_deflate, tx));
6715 * Set the top-level vdev's max queue depth. Evaluate each
6716 * top-level's async write queue depth in case it changed.
6717 * The max queue depth will not change in the middle of syncing
6720 uint64_t queue_depth_total = 0;
6721 for (int c = 0; c < rvd->vdev_children; c++) {
6722 vdev_t *tvd = rvd->vdev_child[c];
6723 metaslab_group_t *mg = tvd->vdev_mg;
6725 if (mg == NULL || mg->mg_class != spa_normal_class(spa) ||
6726 !metaslab_group_initialized(mg))
6730 * It is safe to do a lock-free check here because only async
6731 * allocations look at mg_max_alloc_queue_depth, and async
6732 * allocations all happen from spa_sync().
6734 ASSERT0(refcount_count(&mg->mg_alloc_queue_depth));
6735 mg->mg_max_alloc_queue_depth = max_queue_depth;
6736 queue_depth_total += mg->mg_max_alloc_queue_depth;
6738 metaslab_class_t *mc = spa_normal_class(spa);
6739 ASSERT0(refcount_count(&mc->mc_alloc_slots));
6740 mc->mc_alloc_max_slots = queue_depth_total;
6741 mc->mc_alloc_throttle_enabled = zio_dva_throttle_enabled;
6743 ASSERT3U(mc->mc_alloc_max_slots, <=,
6744 max_queue_depth * rvd->vdev_children);
6747 * Iterate to convergence.
6750 int pass = ++spa->spa_sync_pass;
6752 spa_sync_config_object(spa, tx);
6753 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
6754 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
6755 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
6756 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
6757 spa_errlog_sync(spa, txg);
6758 dsl_pool_sync(dp, txg);
6760 if (pass < zfs_sync_pass_deferred_free) {
6761 spa_sync_frees(spa, free_bpl, tx);
6764 * We can not defer frees in pass 1, because
6765 * we sync the deferred frees later in pass 1.
6767 ASSERT3U(pass, >, 1);
6768 bplist_iterate(free_bpl, bpobj_enqueue_cb,
6769 &spa->spa_deferred_bpobj, tx);
6773 dsl_scan_sync(dp, tx);
6775 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
6779 spa_sync_upgrades(spa, tx);
6781 spa->spa_uberblock.ub_rootbp.blk_birth);
6783 * Note: We need to check if the MOS is dirty
6784 * because we could have marked the MOS dirty
6785 * without updating the uberblock (e.g. if we
6786 * have sync tasks but no dirty user data). We
6787 * need to check the uberblock's rootbp because
6788 * it is updated if we have synced out dirty
6789 * data (though in this case the MOS will most
6790 * likely also be dirty due to second order
6791 * effects, we don't want to rely on that here).
6793 if (spa->spa_uberblock.ub_rootbp.blk_birth < txg &&
6794 !dmu_objset_is_dirty(mos, txg)) {
6796 * Nothing changed on the first pass,
6797 * therefore this TXG is a no-op. Avoid
6798 * syncing deferred frees, so that we
6799 * can keep this TXG as a no-op.
6801 ASSERT(txg_list_empty(&dp->dp_dirty_datasets,
6803 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
6804 ASSERT(txg_list_empty(&dp->dp_sync_tasks, txg));
6807 spa_sync_deferred_frees(spa, tx);
6810 } while (dmu_objset_is_dirty(mos, txg));
6813 * Rewrite the vdev configuration (which includes the uberblock)
6814 * to commit the transaction group.
6816 * If there are no dirty vdevs, we sync the uberblock to a few
6817 * random top-level vdevs that are known to be visible in the
6818 * config cache (see spa_vdev_add() for a complete description).
6819 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
6823 * We hold SCL_STATE to prevent vdev open/close/etc.
6824 * while we're attempting to write the vdev labels.
6826 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6828 if (list_is_empty(&spa->spa_config_dirty_list)) {
6829 vdev_t *svd[SPA_DVAS_PER_BP];
6831 int children = rvd->vdev_children;
6832 int c0 = spa_get_random(children);
6834 for (int c = 0; c < children; c++) {
6835 vd = rvd->vdev_child[(c0 + c) % children];
6836 if (vd->vdev_ms_array == 0 || vd->vdev_islog)
6838 svd[svdcount++] = vd;
6839 if (svdcount == SPA_DVAS_PER_BP)
6842 error = vdev_config_sync(svd, svdcount, txg);
6844 error = vdev_config_sync(rvd->vdev_child,
6845 rvd->vdev_children, txg);
6849 spa->spa_last_synced_guid = rvd->vdev_guid;
6851 spa_config_exit(spa, SCL_STATE, FTAG);
6855 zio_suspend(spa, NULL);
6856 zio_resume_wait(spa);
6861 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY));
6862 #else /* !illumos */
6864 callout_drain(&spa->spa_deadman_cycid);
6866 #endif /* illumos */
6869 * Clear the dirty config list.
6871 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
6872 vdev_config_clean(vd);
6875 * Now that the new config has synced transactionally,
6876 * let it become visible to the config cache.
6878 if (spa->spa_config_syncing != NULL) {
6879 spa_config_set(spa, spa->spa_config_syncing);
6880 spa->spa_config_txg = txg;
6881 spa->spa_config_syncing = NULL;
6884 dsl_pool_sync_done(dp, txg);
6886 mutex_enter(&spa->spa_alloc_lock);
6887 VERIFY0(avl_numnodes(&spa->spa_alloc_tree));
6888 mutex_exit(&spa->spa_alloc_lock);
6891 * Update usable space statistics.
6893 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
6894 vdev_sync_done(vd, txg);
6896 spa_update_dspace(spa);
6899 * It had better be the case that we didn't dirty anything
6900 * since vdev_config_sync().
6902 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
6903 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
6904 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
6906 spa->spa_sync_pass = 0;
6909 * Update the last synced uberblock here. We want to do this at
6910 * the end of spa_sync() so that consumers of spa_last_synced_txg()
6911 * will be guaranteed that all the processing associated with
6912 * that txg has been completed.
6914 spa->spa_ubsync = spa->spa_uberblock;
6915 spa_config_exit(spa, SCL_CONFIG, FTAG);
6917 spa_handle_ignored_writes(spa);
6920 * If any async tasks have been requested, kick them off.
6922 spa_async_dispatch(spa);
6923 spa_async_dispatch_vd(spa);
6927 * Sync all pools. We don't want to hold the namespace lock across these
6928 * operations, so we take a reference on the spa_t and drop the lock during the
6932 spa_sync_allpools(void)
6935 mutex_enter(&spa_namespace_lock);
6936 while ((spa = spa_next(spa)) != NULL) {
6937 if (spa_state(spa) != POOL_STATE_ACTIVE ||
6938 !spa_writeable(spa) || spa_suspended(spa))
6940 spa_open_ref(spa, FTAG);
6941 mutex_exit(&spa_namespace_lock);
6942 txg_wait_synced(spa_get_dsl(spa), 0);
6943 mutex_enter(&spa_namespace_lock);
6944 spa_close(spa, FTAG);
6946 mutex_exit(&spa_namespace_lock);
6950 * ==========================================================================
6951 * Miscellaneous routines
6952 * ==========================================================================
6956 * Remove all pools in the system.
6964 * Remove all cached state. All pools should be closed now,
6965 * so every spa in the AVL tree should be unreferenced.
6967 mutex_enter(&spa_namespace_lock);
6968 while ((spa = spa_next(NULL)) != NULL) {
6970 * Stop async tasks. The async thread may need to detach
6971 * a device that's been replaced, which requires grabbing
6972 * spa_namespace_lock, so we must drop it here.
6974 spa_open_ref(spa, FTAG);
6975 mutex_exit(&spa_namespace_lock);
6976 spa_async_suspend(spa);
6977 mutex_enter(&spa_namespace_lock);
6978 spa_close(spa, FTAG);
6980 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
6982 spa_deactivate(spa);
6986 mutex_exit(&spa_namespace_lock);
6990 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
6995 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
6999 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
7000 vd = spa->spa_l2cache.sav_vdevs[i];
7001 if (vd->vdev_guid == guid)
7005 for (i = 0; i < spa->spa_spares.sav_count; i++) {
7006 vd = spa->spa_spares.sav_vdevs[i];
7007 if (vd->vdev_guid == guid)
7016 spa_upgrade(spa_t *spa, uint64_t version)
7018 ASSERT(spa_writeable(spa));
7020 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
7023 * This should only be called for a non-faulted pool, and since a
7024 * future version would result in an unopenable pool, this shouldn't be
7027 ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version));
7028 ASSERT3U(version, >=, spa->spa_uberblock.ub_version);
7030 spa->spa_uberblock.ub_version = version;
7031 vdev_config_dirty(spa->spa_root_vdev);
7033 spa_config_exit(spa, SCL_ALL, FTAG);
7035 txg_wait_synced(spa_get_dsl(spa), 0);
7039 spa_has_spare(spa_t *spa, uint64_t guid)
7043 spa_aux_vdev_t *sav = &spa->spa_spares;
7045 for (i = 0; i < sav->sav_count; i++)
7046 if (sav->sav_vdevs[i]->vdev_guid == guid)
7049 for (i = 0; i < sav->sav_npending; i++) {
7050 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
7051 &spareguid) == 0 && spareguid == guid)
7059 * Check if a pool has an active shared spare device.
7060 * Note: reference count of an active spare is 2, as a spare and as a replace
7063 spa_has_active_shared_spare(spa_t *spa)
7067 spa_aux_vdev_t *sav = &spa->spa_spares;
7069 for (i = 0; i < sav->sav_count; i++) {
7070 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
7071 &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
7080 spa_event_create(spa_t *spa, vdev_t *vd, const char *name)
7082 sysevent_t *ev = NULL;
7084 sysevent_attr_list_t *attr = NULL;
7085 sysevent_value_t value;
7087 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
7091 value.value_type = SE_DATA_TYPE_STRING;
7092 value.value.sv_string = spa_name(spa);
7093 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
7096 value.value_type = SE_DATA_TYPE_UINT64;
7097 value.value.sv_uint64 = spa_guid(spa);
7098 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
7102 value.value_type = SE_DATA_TYPE_UINT64;
7103 value.value.sv_uint64 = vd->vdev_guid;
7104 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
7108 if (vd->vdev_path) {
7109 value.value_type = SE_DATA_TYPE_STRING;
7110 value.value.sv_string = vd->vdev_path;
7111 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
7112 &value, SE_SLEEP) != 0)
7117 if (sysevent_attach_attributes(ev, attr) != 0)
7123 sysevent_free_attr(attr);
7130 spa_event_post(sysevent_t *ev)
7135 (void) log_sysevent(ev, SE_SLEEP, &eid);
7141 * Post a sysevent corresponding to the given event. The 'name' must be one of
7142 * the event definitions in sys/sysevent/eventdefs.h. The payload will be
7143 * filled in from the spa and (optionally) the vdev. This doesn't do anything
7144 * in the userland libzpool, as we don't want consumers to misinterpret ztest
7145 * or zdb as real changes.
7148 spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
7150 spa_event_post(spa_event_create(spa, vd, name));