4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2011, 2017 by Delphix. All rights reserved.
25 * Copyright (c) 2015, Nexenta Systems, Inc. All rights reserved.
26 * Copyright (c) 2013 Martin Matuska <mm@FreeBSD.org>. All rights reserved.
27 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
28 * Copyright 2013 Saso Kiselkov. All rights reserved.
29 * Copyright (c) 2014 Integros [integros.com]
30 * Copyright 2016 Toomas Soome <tsoome@me.com>
34 * SPA: Storage Pool Allocator
36 * This file contains all the routines used when modifying on-disk SPA state.
37 * This includes opening, importing, destroying, exporting a pool, and syncing a
41 #include <sys/zfs_context.h>
42 #include <sys/fm/fs/zfs.h>
43 #include <sys/spa_impl.h>
45 #include <sys/zio_checksum.h>
47 #include <sys/dmu_tx.h>
51 #include <sys/vdev_impl.h>
52 #include <sys/metaslab.h>
53 #include <sys/metaslab_impl.h>
54 #include <sys/uberblock_impl.h>
57 #include <sys/dmu_traverse.h>
58 #include <sys/dmu_objset.h>
59 #include <sys/unique.h>
60 #include <sys/dsl_pool.h>
61 #include <sys/dsl_dataset.h>
62 #include <sys/dsl_dir.h>
63 #include <sys/dsl_prop.h>
64 #include <sys/dsl_synctask.h>
65 #include <sys/fs/zfs.h>
67 #include <sys/callb.h>
68 #include <sys/spa_boot.h>
69 #include <sys/zfs_ioctl.h>
70 #include <sys/dsl_scan.h>
71 #include <sys/dmu_send.h>
72 #include <sys/dsl_destroy.h>
73 #include <sys/dsl_userhold.h>
74 #include <sys/zfeature.h>
76 #include <sys/trim_map.h>
80 #include <sys/callb.h>
81 #include <sys/cpupart.h>
86 #include "zfs_comutil.h"
88 /* Check hostid on import? */
89 static int check_hostid = 1;
92 * The interval, in seconds, at which failed configuration cache file writes
95 static int zfs_ccw_retry_interval = 300;
97 SYSCTL_DECL(_vfs_zfs);
98 SYSCTL_INT(_vfs_zfs, OID_AUTO, check_hostid, CTLFLAG_RWTUN, &check_hostid, 0,
99 "Check hostid on import?");
100 TUNABLE_INT("vfs.zfs.ccw_retry_interval", &zfs_ccw_retry_interval);
101 SYSCTL_INT(_vfs_zfs, OID_AUTO, ccw_retry_interval, CTLFLAG_RW,
102 &zfs_ccw_retry_interval, 0,
103 "Configuration cache file write, retry after failure, interval (seconds)");
105 typedef enum zti_modes {
106 ZTI_MODE_FIXED, /* value is # of threads (min 1) */
107 ZTI_MODE_BATCH, /* cpu-intensive; value is ignored */
108 ZTI_MODE_NULL, /* don't create a taskq */
112 #define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) }
113 #define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 }
114 #define ZTI_NULL { ZTI_MODE_NULL, 0, 0 }
116 #define ZTI_N(n) ZTI_P(n, 1)
117 #define ZTI_ONE ZTI_N(1)
119 typedef struct zio_taskq_info {
120 zti_modes_t zti_mode;
125 static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
126 "issue", "issue_high", "intr", "intr_high"
130 * This table defines the taskq settings for each ZFS I/O type. When
131 * initializing a pool, we use this table to create an appropriately sized
132 * taskq. Some operations are low volume and therefore have a small, static
133 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE
134 * macros. Other operations process a large amount of data; the ZTI_BATCH
135 * macro causes us to create a taskq oriented for throughput. Some operations
136 * are so high frequency and short-lived that the taskq itself can become a a
137 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an
138 * additional degree of parallelism specified by the number of threads per-
139 * taskq and the number of taskqs; when dispatching an event in this case, the
140 * particular taskq is chosen at random.
142 * The different taskq priorities are to handle the different contexts (issue
143 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that
144 * need to be handled with minimum delay.
146 const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
147 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
148 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* NULL */
149 { ZTI_N(8), ZTI_NULL, ZTI_P(12, 8), ZTI_NULL }, /* READ */
150 { ZTI_BATCH, ZTI_N(5), ZTI_N(8), ZTI_N(5) }, /* WRITE */
151 { ZTI_P(12, 8), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FREE */
152 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* CLAIM */
153 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* IOCTL */
156 static sysevent_t *spa_event_create(spa_t *spa, vdev_t *vd, const char *name);
157 static void spa_event_post(sysevent_t *ev);
158 static void spa_sync_version(void *arg, dmu_tx_t *tx);
159 static void spa_sync_props(void *arg, dmu_tx_t *tx);
160 static boolean_t spa_has_active_shared_spare(spa_t *spa);
161 static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
162 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
164 static void spa_vdev_resilver_done(spa_t *spa);
166 uint_t zio_taskq_batch_pct = 75; /* 1 thread per cpu in pset */
168 id_t zio_taskq_psrset_bind = PS_NONE;
171 boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */
172 uint_t zio_taskq_basedc = 80; /* base duty cycle */
175 boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */
176 extern int zfs_sync_pass_deferred_free;
179 * This (illegal) pool name is used when temporarily importing a spa_t in order
180 * to get the vdev stats associated with the imported devices.
182 #define TRYIMPORT_NAME "$import"
185 * ==========================================================================
186 * SPA properties routines
187 * ==========================================================================
191 * Add a (source=src, propname=propval) list to an nvlist.
194 spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
195 uint64_t intval, zprop_source_t src)
197 const char *propname = zpool_prop_to_name(prop);
200 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
201 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
204 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
206 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
208 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
209 nvlist_free(propval);
213 * Get property values from the spa configuration.
216 spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
218 vdev_t *rvd = spa->spa_root_vdev;
219 dsl_pool_t *pool = spa->spa_dsl_pool;
220 uint64_t size, alloc, cap, version;
221 zprop_source_t src = ZPROP_SRC_NONE;
222 spa_config_dirent_t *dp;
223 metaslab_class_t *mc = spa_normal_class(spa);
225 ASSERT(MUTEX_HELD(&spa->spa_props_lock));
228 alloc = metaslab_class_get_alloc(spa_normal_class(spa));
229 size = metaslab_class_get_space(spa_normal_class(spa));
230 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
231 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
232 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
233 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
236 spa_prop_add_list(*nvp, ZPOOL_PROP_FRAGMENTATION, NULL,
237 metaslab_class_fragmentation(mc), src);
238 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL,
239 metaslab_class_expandable_space(mc), src);
240 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
241 (spa_mode(spa) == FREAD), src);
243 cap = (size == 0) ? 0 : (alloc * 100 / size);
244 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
246 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
247 ddt_get_pool_dedup_ratio(spa), src);
249 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
250 rvd->vdev_state, src);
252 version = spa_version(spa);
253 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
254 src = ZPROP_SRC_DEFAULT;
256 src = ZPROP_SRC_LOCAL;
257 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
262 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
263 * when opening pools before this version freedir will be NULL.
265 if (pool->dp_free_dir != NULL) {
266 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
267 dsl_dir_phys(pool->dp_free_dir)->dd_used_bytes,
270 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
274 if (pool->dp_leak_dir != NULL) {
275 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, NULL,
276 dsl_dir_phys(pool->dp_leak_dir)->dd_used_bytes,
279 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED,
284 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
286 if (spa->spa_comment != NULL) {
287 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
291 if (spa->spa_root != NULL)
292 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
295 if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) {
296 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL,
297 MIN(zfs_max_recordsize, SPA_MAXBLOCKSIZE), ZPROP_SRC_NONE);
299 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL,
300 SPA_OLD_MAXBLOCKSIZE, ZPROP_SRC_NONE);
303 if ((dp = list_head(&spa->spa_config_list)) != NULL) {
304 if (dp->scd_path == NULL) {
305 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
306 "none", 0, ZPROP_SRC_LOCAL);
307 } else if (strcmp(dp->scd_path, spa_config_path) != 0) {
308 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
309 dp->scd_path, 0, ZPROP_SRC_LOCAL);
315 * Get zpool property values.
318 spa_prop_get(spa_t *spa, nvlist_t **nvp)
320 objset_t *mos = spa->spa_meta_objset;
325 VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
327 mutex_enter(&spa->spa_props_lock);
330 * Get properties from the spa config.
332 spa_prop_get_config(spa, nvp);
334 /* If no pool property object, no more prop to get. */
335 if (mos == NULL || spa->spa_pool_props_object == 0) {
336 mutex_exit(&spa->spa_props_lock);
341 * Get properties from the MOS pool property object.
343 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
344 (err = zap_cursor_retrieve(&zc, &za)) == 0;
345 zap_cursor_advance(&zc)) {
348 zprop_source_t src = ZPROP_SRC_DEFAULT;
351 if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
354 switch (za.za_integer_length) {
356 /* integer property */
357 if (za.za_first_integer !=
358 zpool_prop_default_numeric(prop))
359 src = ZPROP_SRC_LOCAL;
361 if (prop == ZPOOL_PROP_BOOTFS) {
363 dsl_dataset_t *ds = NULL;
365 dp = spa_get_dsl(spa);
366 dsl_pool_config_enter(dp, FTAG);
367 if (err = dsl_dataset_hold_obj(dp,
368 za.za_first_integer, FTAG, &ds)) {
369 dsl_pool_config_exit(dp, FTAG);
373 strval = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN,
375 dsl_dataset_name(ds, strval);
376 dsl_dataset_rele(ds, FTAG);
377 dsl_pool_config_exit(dp, FTAG);
380 intval = za.za_first_integer;
383 spa_prop_add_list(*nvp, prop, strval, intval, src);
386 kmem_free(strval, ZFS_MAX_DATASET_NAME_LEN);
391 /* string property */
392 strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
393 err = zap_lookup(mos, spa->spa_pool_props_object,
394 za.za_name, 1, za.za_num_integers, strval);
396 kmem_free(strval, za.za_num_integers);
399 spa_prop_add_list(*nvp, prop, strval, 0, src);
400 kmem_free(strval, za.za_num_integers);
407 zap_cursor_fini(&zc);
408 mutex_exit(&spa->spa_props_lock);
410 if (err && err != ENOENT) {
420 * Validate the given pool properties nvlist and modify the list
421 * for the property values to be set.
424 spa_prop_validate(spa_t *spa, nvlist_t *props)
427 int error = 0, reset_bootfs = 0;
429 boolean_t has_feature = B_FALSE;
432 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
434 char *strval, *slash, *check, *fname;
435 const char *propname = nvpair_name(elem);
436 zpool_prop_t prop = zpool_name_to_prop(propname);
440 if (!zpool_prop_feature(propname)) {
441 error = SET_ERROR(EINVAL);
446 * Sanitize the input.
448 if (nvpair_type(elem) != DATA_TYPE_UINT64) {
449 error = SET_ERROR(EINVAL);
453 if (nvpair_value_uint64(elem, &intval) != 0) {
454 error = SET_ERROR(EINVAL);
459 error = SET_ERROR(EINVAL);
463 fname = strchr(propname, '@') + 1;
464 if (zfeature_lookup_name(fname, NULL) != 0) {
465 error = SET_ERROR(EINVAL);
469 has_feature = B_TRUE;
472 case ZPOOL_PROP_VERSION:
473 error = nvpair_value_uint64(elem, &intval);
475 (intval < spa_version(spa) ||
476 intval > SPA_VERSION_BEFORE_FEATURES ||
478 error = SET_ERROR(EINVAL);
481 case ZPOOL_PROP_DELEGATION:
482 case ZPOOL_PROP_AUTOREPLACE:
483 case ZPOOL_PROP_LISTSNAPS:
484 case ZPOOL_PROP_AUTOEXPAND:
485 error = nvpair_value_uint64(elem, &intval);
486 if (!error && intval > 1)
487 error = SET_ERROR(EINVAL);
490 case ZPOOL_PROP_BOOTFS:
492 * If the pool version is less than SPA_VERSION_BOOTFS,
493 * or the pool is still being created (version == 0),
494 * the bootfs property cannot be set.
496 if (spa_version(spa) < SPA_VERSION_BOOTFS) {
497 error = SET_ERROR(ENOTSUP);
502 * Make sure the vdev config is bootable
504 if (!vdev_is_bootable(spa->spa_root_vdev)) {
505 error = SET_ERROR(ENOTSUP);
511 error = nvpair_value_string(elem, &strval);
517 if (strval == NULL || strval[0] == '\0') {
518 objnum = zpool_prop_default_numeric(
523 if (error = dmu_objset_hold(strval, FTAG, &os))
527 * Must be ZPL, and its property settings
528 * must be supported by GRUB (compression
529 * is not gzip, and large blocks are not used).
532 if (dmu_objset_type(os) != DMU_OST_ZFS) {
533 error = SET_ERROR(ENOTSUP);
535 dsl_prop_get_int_ds(dmu_objset_ds(os),
536 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
538 !BOOTFS_COMPRESS_VALID(propval)) {
539 error = SET_ERROR(ENOTSUP);
541 objnum = dmu_objset_id(os);
543 dmu_objset_rele(os, FTAG);
547 case ZPOOL_PROP_FAILUREMODE:
548 error = nvpair_value_uint64(elem, &intval);
549 if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
550 intval > ZIO_FAILURE_MODE_PANIC))
551 error = SET_ERROR(EINVAL);
554 * This is a special case which only occurs when
555 * the pool has completely failed. This allows
556 * the user to change the in-core failmode property
557 * without syncing it out to disk (I/Os might
558 * currently be blocked). We do this by returning
559 * EIO to the caller (spa_prop_set) to trick it
560 * into thinking we encountered a property validation
563 if (!error && spa_suspended(spa)) {
564 spa->spa_failmode = intval;
565 error = SET_ERROR(EIO);
569 case ZPOOL_PROP_CACHEFILE:
570 if ((error = nvpair_value_string(elem, &strval)) != 0)
573 if (strval[0] == '\0')
576 if (strcmp(strval, "none") == 0)
579 if (strval[0] != '/') {
580 error = SET_ERROR(EINVAL);
584 slash = strrchr(strval, '/');
585 ASSERT(slash != NULL);
587 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
588 strcmp(slash, "/..") == 0)
589 error = SET_ERROR(EINVAL);
592 case ZPOOL_PROP_COMMENT:
593 if ((error = nvpair_value_string(elem, &strval)) != 0)
595 for (check = strval; *check != '\0'; check++) {
597 * The kernel doesn't have an easy isprint()
598 * check. For this kernel check, we merely
599 * check ASCII apart from DEL. Fix this if
600 * there is an easy-to-use kernel isprint().
602 if (*check >= 0x7f) {
603 error = SET_ERROR(EINVAL);
607 if (strlen(strval) > ZPROP_MAX_COMMENT)
611 case ZPOOL_PROP_DEDUPDITTO:
612 if (spa_version(spa) < SPA_VERSION_DEDUP)
613 error = SET_ERROR(ENOTSUP);
615 error = nvpair_value_uint64(elem, &intval);
617 intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
618 error = SET_ERROR(EINVAL);
626 if (!error && reset_bootfs) {
627 error = nvlist_remove(props,
628 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
631 error = nvlist_add_uint64(props,
632 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
640 spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
643 spa_config_dirent_t *dp;
645 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
649 dp = kmem_alloc(sizeof (spa_config_dirent_t),
652 if (cachefile[0] == '\0')
653 dp->scd_path = spa_strdup(spa_config_path);
654 else if (strcmp(cachefile, "none") == 0)
657 dp->scd_path = spa_strdup(cachefile);
659 list_insert_head(&spa->spa_config_list, dp);
661 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
665 spa_prop_set(spa_t *spa, nvlist_t *nvp)
668 nvpair_t *elem = NULL;
669 boolean_t need_sync = B_FALSE;
671 if ((error = spa_prop_validate(spa, nvp)) != 0)
674 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
675 zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem));
677 if (prop == ZPOOL_PROP_CACHEFILE ||
678 prop == ZPOOL_PROP_ALTROOT ||
679 prop == ZPOOL_PROP_READONLY)
682 if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) {
685 if (prop == ZPOOL_PROP_VERSION) {
686 VERIFY(nvpair_value_uint64(elem, &ver) == 0);
688 ASSERT(zpool_prop_feature(nvpair_name(elem)));
689 ver = SPA_VERSION_FEATURES;
693 /* Save time if the version is already set. */
694 if (ver == spa_version(spa))
698 * In addition to the pool directory object, we might
699 * create the pool properties object, the features for
700 * read object, the features for write object, or the
701 * feature descriptions object.
703 error = dsl_sync_task(spa->spa_name, NULL,
704 spa_sync_version, &ver,
705 6, ZFS_SPACE_CHECK_RESERVED);
716 return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props,
717 nvp, 6, ZFS_SPACE_CHECK_RESERVED));
724 * If the bootfs property value is dsobj, clear it.
727 spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
729 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
730 VERIFY(zap_remove(spa->spa_meta_objset,
731 spa->spa_pool_props_object,
732 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
739 spa_change_guid_check(void *arg, dmu_tx_t *tx)
741 uint64_t *newguid = arg;
742 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
743 vdev_t *rvd = spa->spa_root_vdev;
746 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
747 vdev_state = rvd->vdev_state;
748 spa_config_exit(spa, SCL_STATE, FTAG);
750 if (vdev_state != VDEV_STATE_HEALTHY)
751 return (SET_ERROR(ENXIO));
753 ASSERT3U(spa_guid(spa), !=, *newguid);
759 spa_change_guid_sync(void *arg, dmu_tx_t *tx)
761 uint64_t *newguid = arg;
762 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
764 vdev_t *rvd = spa->spa_root_vdev;
766 oldguid = spa_guid(spa);
768 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
769 rvd->vdev_guid = *newguid;
770 rvd->vdev_guid_sum += (*newguid - oldguid);
771 vdev_config_dirty(rvd);
772 spa_config_exit(spa, SCL_STATE, FTAG);
774 spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu",
779 * Change the GUID for the pool. This is done so that we can later
780 * re-import a pool built from a clone of our own vdevs. We will modify
781 * the root vdev's guid, our own pool guid, and then mark all of our
782 * vdevs dirty. Note that we must make sure that all our vdevs are
783 * online when we do this, or else any vdevs that weren't present
784 * would be orphaned from our pool. We are also going to issue a
785 * sysevent to update any watchers.
788 spa_change_guid(spa_t *spa)
793 mutex_enter(&spa->spa_vdev_top_lock);
794 mutex_enter(&spa_namespace_lock);
795 guid = spa_generate_guid(NULL);
797 error = dsl_sync_task(spa->spa_name, spa_change_guid_check,
798 spa_change_guid_sync, &guid, 5, ZFS_SPACE_CHECK_RESERVED);
801 spa_config_sync(spa, B_FALSE, B_TRUE);
802 spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID);
805 mutex_exit(&spa_namespace_lock);
806 mutex_exit(&spa->spa_vdev_top_lock);
812 * ==========================================================================
813 * SPA state manipulation (open/create/destroy/import/export)
814 * ==========================================================================
818 spa_error_entry_compare(const void *a, const void *b)
820 spa_error_entry_t *sa = (spa_error_entry_t *)a;
821 spa_error_entry_t *sb = (spa_error_entry_t *)b;
824 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
825 sizeof (zbookmark_phys_t));
836 * Utility function which retrieves copies of the current logs and
837 * re-initializes them in the process.
840 spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
842 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
844 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
845 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
847 avl_create(&spa->spa_errlist_scrub,
848 spa_error_entry_compare, sizeof (spa_error_entry_t),
849 offsetof(spa_error_entry_t, se_avl));
850 avl_create(&spa->spa_errlist_last,
851 spa_error_entry_compare, sizeof (spa_error_entry_t),
852 offsetof(spa_error_entry_t, se_avl));
856 spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
858 const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
859 enum zti_modes mode = ztip->zti_mode;
860 uint_t value = ztip->zti_value;
861 uint_t count = ztip->zti_count;
862 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
865 boolean_t batch = B_FALSE;
867 if (mode == ZTI_MODE_NULL) {
869 tqs->stqs_taskq = NULL;
873 ASSERT3U(count, >, 0);
875 tqs->stqs_count = count;
876 tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP);
880 ASSERT3U(value, >=, 1);
881 value = MAX(value, 1);
886 flags |= TASKQ_THREADS_CPU_PCT;
887 value = zio_taskq_batch_pct;
891 panic("unrecognized mode for %s_%s taskq (%u:%u) in "
893 zio_type_name[t], zio_taskq_types[q], mode, value);
897 for (uint_t i = 0; i < count; i++) {
901 (void) snprintf(name, sizeof (name), "%s_%s_%u",
902 zio_type_name[t], zio_taskq_types[q], i);
904 (void) snprintf(name, sizeof (name), "%s_%s",
905 zio_type_name[t], zio_taskq_types[q]);
909 if (zio_taskq_sysdc && spa->spa_proc != &p0) {
911 flags |= TASKQ_DC_BATCH;
913 tq = taskq_create_sysdc(name, value, 50, INT_MAX,
914 spa->spa_proc, zio_taskq_basedc, flags);
917 pri_t pri = maxclsyspri;
919 * The write issue taskq can be extremely CPU
920 * intensive. Run it at slightly lower priority
921 * than the other taskqs.
923 * - numerically higher priorities are lower priorities;
924 * - if priorities divided by four (RQ_PPQ) are equal
925 * then a difference between them is insignificant.
927 if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE)
934 tq = taskq_create_proc(name, value, pri, 50,
935 INT_MAX, spa->spa_proc, flags);
940 tqs->stqs_taskq[i] = tq;
945 spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
947 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
949 if (tqs->stqs_taskq == NULL) {
950 ASSERT0(tqs->stqs_count);
954 for (uint_t i = 0; i < tqs->stqs_count; i++) {
955 ASSERT3P(tqs->stqs_taskq[i], !=, NULL);
956 taskq_destroy(tqs->stqs_taskq[i]);
959 kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *));
960 tqs->stqs_taskq = NULL;
964 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
965 * Note that a type may have multiple discrete taskqs to avoid lock contention
966 * on the taskq itself. In that case we choose which taskq at random by using
967 * the low bits of gethrtime().
970 spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q,
971 task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent)
973 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
976 ASSERT3P(tqs->stqs_taskq, !=, NULL);
977 ASSERT3U(tqs->stqs_count, !=, 0);
979 if (tqs->stqs_count == 1) {
980 tq = tqs->stqs_taskq[0];
983 tq = tqs->stqs_taskq[cpu_ticks() % tqs->stqs_count];
985 tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count];
989 taskq_dispatch_ent(tq, func, arg, flags, ent);
993 spa_create_zio_taskqs(spa_t *spa)
995 for (int t = 0; t < ZIO_TYPES; t++) {
996 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
997 spa_taskqs_init(spa, t, q);
1005 spa_thread(void *arg)
1007 callb_cpr_t cprinfo;
1010 user_t *pu = PTOU(curproc);
1012 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
1015 ASSERT(curproc != &p0);
1016 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
1017 "zpool-%s", spa->spa_name);
1018 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
1021 /* bind this thread to the requested psrset */
1022 if (zio_taskq_psrset_bind != PS_NONE) {
1024 mutex_enter(&cpu_lock);
1025 mutex_enter(&pidlock);
1026 mutex_enter(&curproc->p_lock);
1028 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
1029 0, NULL, NULL) == 0) {
1030 curthread->t_bind_pset = zio_taskq_psrset_bind;
1033 "Couldn't bind process for zfs pool \"%s\" to "
1034 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
1037 mutex_exit(&curproc->p_lock);
1038 mutex_exit(&pidlock);
1039 mutex_exit(&cpu_lock);
1045 if (zio_taskq_sysdc) {
1046 sysdc_thread_enter(curthread, 100, 0);
1050 spa->spa_proc = curproc;
1051 spa->spa_did = curthread->t_did;
1053 spa_create_zio_taskqs(spa);
1055 mutex_enter(&spa->spa_proc_lock);
1056 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
1058 spa->spa_proc_state = SPA_PROC_ACTIVE;
1059 cv_broadcast(&spa->spa_proc_cv);
1061 CALLB_CPR_SAFE_BEGIN(&cprinfo);
1062 while (spa->spa_proc_state == SPA_PROC_ACTIVE)
1063 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1064 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
1066 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
1067 spa->spa_proc_state = SPA_PROC_GONE;
1068 spa->spa_proc = &p0;
1069 cv_broadcast(&spa->spa_proc_cv);
1070 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */
1072 mutex_enter(&curproc->p_lock);
1075 #endif /* SPA_PROCESS */
1079 * Activate an uninitialized pool.
1082 spa_activate(spa_t *spa, int mode)
1084 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
1086 spa->spa_state = POOL_STATE_ACTIVE;
1087 spa->spa_mode = mode;
1089 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
1090 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
1092 /* Try to create a covering process */
1093 mutex_enter(&spa->spa_proc_lock);
1094 ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
1095 ASSERT(spa->spa_proc == &p0);
1099 /* Only create a process if we're going to be around a while. */
1100 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
1101 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
1103 spa->spa_proc_state = SPA_PROC_CREATED;
1104 while (spa->spa_proc_state == SPA_PROC_CREATED) {
1105 cv_wait(&spa->spa_proc_cv,
1106 &spa->spa_proc_lock);
1108 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1109 ASSERT(spa->spa_proc != &p0);
1110 ASSERT(spa->spa_did != 0);
1114 "Couldn't create process for zfs pool \"%s\"\n",
1119 #endif /* SPA_PROCESS */
1120 mutex_exit(&spa->spa_proc_lock);
1122 /* If we didn't create a process, we need to create our taskqs. */
1123 ASSERT(spa->spa_proc == &p0);
1124 if (spa->spa_proc == &p0) {
1125 spa_create_zio_taskqs(spa);
1129 * Start TRIM thread.
1131 trim_thread_create(spa);
1133 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
1134 offsetof(vdev_t, vdev_config_dirty_node));
1135 list_create(&spa->spa_evicting_os_list, sizeof (objset_t),
1136 offsetof(objset_t, os_evicting_node));
1137 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
1138 offsetof(vdev_t, vdev_state_dirty_node));
1140 txg_list_create(&spa->spa_vdev_txg_list, spa,
1141 offsetof(struct vdev, vdev_txg_node));
1143 avl_create(&spa->spa_errlist_scrub,
1144 spa_error_entry_compare, sizeof (spa_error_entry_t),
1145 offsetof(spa_error_entry_t, se_avl));
1146 avl_create(&spa->spa_errlist_last,
1147 spa_error_entry_compare, sizeof (spa_error_entry_t),
1148 offsetof(spa_error_entry_t, se_avl));
1152 * Opposite of spa_activate().
1155 spa_deactivate(spa_t *spa)
1157 ASSERT(spa->spa_sync_on == B_FALSE);
1158 ASSERT(spa->spa_dsl_pool == NULL);
1159 ASSERT(spa->spa_root_vdev == NULL);
1160 ASSERT(spa->spa_async_zio_root == NULL);
1161 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
1164 * Stop TRIM thread in case spa_unload() wasn't called directly
1165 * before spa_deactivate().
1167 trim_thread_destroy(spa);
1169 spa_evicting_os_wait(spa);
1171 txg_list_destroy(&spa->spa_vdev_txg_list);
1173 list_destroy(&spa->spa_config_dirty_list);
1174 list_destroy(&spa->spa_evicting_os_list);
1175 list_destroy(&spa->spa_state_dirty_list);
1177 for (int t = 0; t < ZIO_TYPES; t++) {
1178 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1179 spa_taskqs_fini(spa, t, q);
1183 metaslab_class_destroy(spa->spa_normal_class);
1184 spa->spa_normal_class = NULL;
1186 metaslab_class_destroy(spa->spa_log_class);
1187 spa->spa_log_class = NULL;
1190 * If this was part of an import or the open otherwise failed, we may
1191 * still have errors left in the queues. Empty them just in case.
1193 spa_errlog_drain(spa);
1195 avl_destroy(&spa->spa_errlist_scrub);
1196 avl_destroy(&spa->spa_errlist_last);
1198 spa->spa_state = POOL_STATE_UNINITIALIZED;
1200 mutex_enter(&spa->spa_proc_lock);
1201 if (spa->spa_proc_state != SPA_PROC_NONE) {
1202 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1203 spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1204 cv_broadcast(&spa->spa_proc_cv);
1205 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1206 ASSERT(spa->spa_proc != &p0);
1207 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1209 ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1210 spa->spa_proc_state = SPA_PROC_NONE;
1212 ASSERT(spa->spa_proc == &p0);
1213 mutex_exit(&spa->spa_proc_lock);
1217 * We want to make sure spa_thread() has actually exited the ZFS
1218 * module, so that the module can't be unloaded out from underneath
1221 if (spa->spa_did != 0) {
1222 thread_join(spa->spa_did);
1225 #endif /* SPA_PROCESS */
1229 * Verify a pool configuration, and construct the vdev tree appropriately. This
1230 * will create all the necessary vdevs in the appropriate layout, with each vdev
1231 * in the CLOSED state. This will prep the pool before open/creation/import.
1232 * All vdev validation is done by the vdev_alloc() routine.
1235 spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1236 uint_t id, int atype)
1242 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1245 if ((*vdp)->vdev_ops->vdev_op_leaf)
1248 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1251 if (error == ENOENT)
1257 return (SET_ERROR(EINVAL));
1260 for (int c = 0; c < children; c++) {
1262 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1270 ASSERT(*vdp != NULL);
1276 * Opposite of spa_load().
1279 spa_unload(spa_t *spa)
1283 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1288 trim_thread_destroy(spa);
1293 spa_async_suspend(spa);
1298 if (spa->spa_sync_on) {
1299 txg_sync_stop(spa->spa_dsl_pool);
1300 spa->spa_sync_on = B_FALSE;
1304 * Even though vdev_free() also calls vdev_metaslab_fini, we need
1305 * to call it earlier, before we wait for async i/o to complete.
1306 * This ensures that there is no async metaslab prefetching, by
1307 * calling taskq_wait(mg_taskq).
1309 if (spa->spa_root_vdev != NULL) {
1310 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1311 for (int c = 0; c < spa->spa_root_vdev->vdev_children; c++)
1312 vdev_metaslab_fini(spa->spa_root_vdev->vdev_child[c]);
1313 spa_config_exit(spa, SCL_ALL, FTAG);
1317 * Wait for any outstanding async I/O to complete.
1319 if (spa->spa_async_zio_root != NULL) {
1320 for (int i = 0; i < max_ncpus; i++)
1321 (void) zio_wait(spa->spa_async_zio_root[i]);
1322 kmem_free(spa->spa_async_zio_root, max_ncpus * sizeof (void *));
1323 spa->spa_async_zio_root = NULL;
1326 bpobj_close(&spa->spa_deferred_bpobj);
1328 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1333 if (spa->spa_root_vdev)
1334 vdev_free(spa->spa_root_vdev);
1335 ASSERT(spa->spa_root_vdev == NULL);
1338 * Close the dsl pool.
1340 if (spa->spa_dsl_pool) {
1341 dsl_pool_close(spa->spa_dsl_pool);
1342 spa->spa_dsl_pool = NULL;
1343 spa->spa_meta_objset = NULL;
1349 * Drop and purge level 2 cache
1351 spa_l2cache_drop(spa);
1353 for (i = 0; i < spa->spa_spares.sav_count; i++)
1354 vdev_free(spa->spa_spares.sav_vdevs[i]);
1355 if (spa->spa_spares.sav_vdevs) {
1356 kmem_free(spa->spa_spares.sav_vdevs,
1357 spa->spa_spares.sav_count * sizeof (void *));
1358 spa->spa_spares.sav_vdevs = NULL;
1360 if (spa->spa_spares.sav_config) {
1361 nvlist_free(spa->spa_spares.sav_config);
1362 spa->spa_spares.sav_config = NULL;
1364 spa->spa_spares.sav_count = 0;
1366 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1367 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1368 vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1370 if (spa->spa_l2cache.sav_vdevs) {
1371 kmem_free(spa->spa_l2cache.sav_vdevs,
1372 spa->spa_l2cache.sav_count * sizeof (void *));
1373 spa->spa_l2cache.sav_vdevs = NULL;
1375 if (spa->spa_l2cache.sav_config) {
1376 nvlist_free(spa->spa_l2cache.sav_config);
1377 spa->spa_l2cache.sav_config = NULL;
1379 spa->spa_l2cache.sav_count = 0;
1381 spa->spa_async_suspended = 0;
1383 if (spa->spa_comment != NULL) {
1384 spa_strfree(spa->spa_comment);
1385 spa->spa_comment = NULL;
1388 spa_config_exit(spa, SCL_ALL, FTAG);
1392 * Load (or re-load) the current list of vdevs describing the active spares for
1393 * this pool. When this is called, we have some form of basic information in
1394 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1395 * then re-generate a more complete list including status information.
1398 spa_load_spares(spa_t *spa)
1405 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1408 * First, close and free any existing spare vdevs.
1410 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1411 vd = spa->spa_spares.sav_vdevs[i];
1413 /* Undo the call to spa_activate() below */
1414 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1415 B_FALSE)) != NULL && tvd->vdev_isspare)
1416 spa_spare_remove(tvd);
1421 if (spa->spa_spares.sav_vdevs)
1422 kmem_free(spa->spa_spares.sav_vdevs,
1423 spa->spa_spares.sav_count * sizeof (void *));
1425 if (spa->spa_spares.sav_config == NULL)
1428 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1429 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1431 spa->spa_spares.sav_count = (int)nspares;
1432 spa->spa_spares.sav_vdevs = NULL;
1438 * Construct the array of vdevs, opening them to get status in the
1439 * process. For each spare, there is potentially two different vdev_t
1440 * structures associated with it: one in the list of spares (used only
1441 * for basic validation purposes) and one in the active vdev
1442 * configuration (if it's spared in). During this phase we open and
1443 * validate each vdev on the spare list. If the vdev also exists in the
1444 * active configuration, then we also mark this vdev as an active spare.
1446 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1448 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1449 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1450 VDEV_ALLOC_SPARE) == 0);
1453 spa->spa_spares.sav_vdevs[i] = vd;
1455 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1456 B_FALSE)) != NULL) {
1457 if (!tvd->vdev_isspare)
1461 * We only mark the spare active if we were successfully
1462 * able to load the vdev. Otherwise, importing a pool
1463 * with a bad active spare would result in strange
1464 * behavior, because multiple pool would think the spare
1465 * is actively in use.
1467 * There is a vulnerability here to an equally bizarre
1468 * circumstance, where a dead active spare is later
1469 * brought back to life (onlined or otherwise). Given
1470 * the rarity of this scenario, and the extra complexity
1471 * it adds, we ignore the possibility.
1473 if (!vdev_is_dead(tvd))
1474 spa_spare_activate(tvd);
1478 vd->vdev_aux = &spa->spa_spares;
1480 if (vdev_open(vd) != 0)
1483 if (vdev_validate_aux(vd) == 0)
1488 * Recompute the stashed list of spares, with status information
1491 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1492 DATA_TYPE_NVLIST_ARRAY) == 0);
1494 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1496 for (i = 0; i < spa->spa_spares.sav_count; i++)
1497 spares[i] = vdev_config_generate(spa,
1498 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1499 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1500 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1501 for (i = 0; i < spa->spa_spares.sav_count; i++)
1502 nvlist_free(spares[i]);
1503 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1507 * Load (or re-load) the current list of vdevs describing the active l2cache for
1508 * this pool. When this is called, we have some form of basic information in
1509 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1510 * then re-generate a more complete list including status information.
1511 * Devices which are already active have their details maintained, and are
1515 spa_load_l2cache(spa_t *spa)
1519 int i, j, oldnvdevs;
1521 vdev_t *vd, **oldvdevs, **newvdevs;
1522 spa_aux_vdev_t *sav = &spa->spa_l2cache;
1524 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1526 if (sav->sav_config != NULL) {
1527 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1528 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1529 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1535 oldvdevs = sav->sav_vdevs;
1536 oldnvdevs = sav->sav_count;
1537 sav->sav_vdevs = NULL;
1541 * Process new nvlist of vdevs.
1543 for (i = 0; i < nl2cache; i++) {
1544 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1548 for (j = 0; j < oldnvdevs; j++) {
1550 if (vd != NULL && guid == vd->vdev_guid) {
1552 * Retain previous vdev for add/remove ops.
1560 if (newvdevs[i] == NULL) {
1564 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1565 VDEV_ALLOC_L2CACHE) == 0);
1570 * Commit this vdev as an l2cache device,
1571 * even if it fails to open.
1573 spa_l2cache_add(vd);
1578 spa_l2cache_activate(vd);
1580 if (vdev_open(vd) != 0)
1583 (void) vdev_validate_aux(vd);
1585 if (!vdev_is_dead(vd))
1586 l2arc_add_vdev(spa, vd);
1591 * Purge vdevs that were dropped
1593 for (i = 0; i < oldnvdevs; i++) {
1598 ASSERT(vd->vdev_isl2cache);
1600 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1601 pool != 0ULL && l2arc_vdev_present(vd))
1602 l2arc_remove_vdev(vd);
1603 vdev_clear_stats(vd);
1609 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1611 if (sav->sav_config == NULL)
1614 sav->sav_vdevs = newvdevs;
1615 sav->sav_count = (int)nl2cache;
1618 * Recompute the stashed list of l2cache devices, with status
1619 * information this time.
1621 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1622 DATA_TYPE_NVLIST_ARRAY) == 0);
1624 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1625 for (i = 0; i < sav->sav_count; i++)
1626 l2cache[i] = vdev_config_generate(spa,
1627 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1628 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1629 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1631 for (i = 0; i < sav->sav_count; i++)
1632 nvlist_free(l2cache[i]);
1634 kmem_free(l2cache, sav->sav_count * sizeof (void *));
1638 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1641 char *packed = NULL;
1646 error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db);
1650 nvsize = *(uint64_t *)db->db_data;
1651 dmu_buf_rele(db, FTAG);
1653 packed = kmem_alloc(nvsize, KM_SLEEP);
1654 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1657 error = nvlist_unpack(packed, nvsize, value, 0);
1658 kmem_free(packed, nvsize);
1664 * Checks to see if the given vdev could not be opened, in which case we post a
1665 * sysevent to notify the autoreplace code that the device has been removed.
1668 spa_check_removed(vdev_t *vd)
1670 for (int c = 0; c < vd->vdev_children; c++)
1671 spa_check_removed(vd->vdev_child[c]);
1673 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) &&
1675 zfs_post_autoreplace(vd->vdev_spa, vd);
1676 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1681 spa_config_valid_zaps(vdev_t *vd, vdev_t *mvd)
1683 ASSERT3U(vd->vdev_children, ==, mvd->vdev_children);
1685 vd->vdev_top_zap = mvd->vdev_top_zap;
1686 vd->vdev_leaf_zap = mvd->vdev_leaf_zap;
1688 for (uint64_t i = 0; i < vd->vdev_children; i++) {
1689 spa_config_valid_zaps(vd->vdev_child[i], mvd->vdev_child[i]);
1694 * Validate the current config against the MOS config
1697 spa_config_valid(spa_t *spa, nvlist_t *config)
1699 vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1702 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1704 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1705 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1707 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1710 * If we're doing a normal import, then build up any additional
1711 * diagnostic information about missing devices in this config.
1712 * We'll pass this up to the user for further processing.
1714 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1715 nvlist_t **child, *nv;
1718 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1720 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1722 for (int c = 0; c < rvd->vdev_children; c++) {
1723 vdev_t *tvd = rvd->vdev_child[c];
1724 vdev_t *mtvd = mrvd->vdev_child[c];
1726 if (tvd->vdev_ops == &vdev_missing_ops &&
1727 mtvd->vdev_ops != &vdev_missing_ops &&
1729 child[idx++] = vdev_config_generate(spa, mtvd,
1734 VERIFY(nvlist_add_nvlist_array(nv,
1735 ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1736 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1737 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1739 for (int i = 0; i < idx; i++)
1740 nvlist_free(child[i]);
1743 kmem_free(child, rvd->vdev_children * sizeof (char **));
1747 * Compare the root vdev tree with the information we have
1748 * from the MOS config (mrvd). Check each top-level vdev
1749 * with the corresponding MOS config top-level (mtvd).
1751 for (int c = 0; c < rvd->vdev_children; c++) {
1752 vdev_t *tvd = rvd->vdev_child[c];
1753 vdev_t *mtvd = mrvd->vdev_child[c];
1756 * Resolve any "missing" vdevs in the current configuration.
1757 * If we find that the MOS config has more accurate information
1758 * about the top-level vdev then use that vdev instead.
1760 if (tvd->vdev_ops == &vdev_missing_ops &&
1761 mtvd->vdev_ops != &vdev_missing_ops) {
1763 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1767 * Device specific actions.
1769 if (mtvd->vdev_islog) {
1770 spa_set_log_state(spa, SPA_LOG_CLEAR);
1773 * XXX - once we have 'readonly' pool
1774 * support we should be able to handle
1775 * missing data devices by transitioning
1776 * the pool to readonly.
1782 * Swap the missing vdev with the data we were
1783 * able to obtain from the MOS config.
1785 vdev_remove_child(rvd, tvd);
1786 vdev_remove_child(mrvd, mtvd);
1788 vdev_add_child(rvd, mtvd);
1789 vdev_add_child(mrvd, tvd);
1791 spa_config_exit(spa, SCL_ALL, FTAG);
1793 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1797 if (mtvd->vdev_islog) {
1799 * Load the slog device's state from the MOS
1800 * config since it's possible that the label
1801 * does not contain the most up-to-date
1804 vdev_load_log_state(tvd, mtvd);
1809 * Per-vdev ZAP info is stored exclusively in the MOS.
1811 spa_config_valid_zaps(tvd, mtvd);
1816 spa_config_exit(spa, SCL_ALL, FTAG);
1819 * Ensure we were able to validate the config.
1821 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1825 * Check for missing log devices
1828 spa_check_logs(spa_t *spa)
1830 boolean_t rv = B_FALSE;
1831 dsl_pool_t *dp = spa_get_dsl(spa);
1833 switch (spa->spa_log_state) {
1834 case SPA_LOG_MISSING:
1835 /* need to recheck in case slog has been restored */
1836 case SPA_LOG_UNKNOWN:
1837 rv = (dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
1838 zil_check_log_chain, NULL, DS_FIND_CHILDREN) != 0);
1840 spa_set_log_state(spa, SPA_LOG_MISSING);
1847 spa_passivate_log(spa_t *spa)
1849 vdev_t *rvd = spa->spa_root_vdev;
1850 boolean_t slog_found = B_FALSE;
1852 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1854 if (!spa_has_slogs(spa))
1857 for (int c = 0; c < rvd->vdev_children; c++) {
1858 vdev_t *tvd = rvd->vdev_child[c];
1859 metaslab_group_t *mg = tvd->vdev_mg;
1861 if (tvd->vdev_islog) {
1862 metaslab_group_passivate(mg);
1863 slog_found = B_TRUE;
1867 return (slog_found);
1871 spa_activate_log(spa_t *spa)
1873 vdev_t *rvd = spa->spa_root_vdev;
1875 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1877 for (int c = 0; c < rvd->vdev_children; c++) {
1878 vdev_t *tvd = rvd->vdev_child[c];
1879 metaslab_group_t *mg = tvd->vdev_mg;
1881 if (tvd->vdev_islog)
1882 metaslab_group_activate(mg);
1887 spa_offline_log(spa_t *spa)
1891 error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1892 NULL, DS_FIND_CHILDREN);
1895 * We successfully offlined the log device, sync out the
1896 * current txg so that the "stubby" block can be removed
1899 txg_wait_synced(spa->spa_dsl_pool, 0);
1905 spa_aux_check_removed(spa_aux_vdev_t *sav)
1909 for (i = 0; i < sav->sav_count; i++)
1910 spa_check_removed(sav->sav_vdevs[i]);
1914 spa_claim_notify(zio_t *zio)
1916 spa_t *spa = zio->io_spa;
1921 mutex_enter(&spa->spa_props_lock); /* any mutex will do */
1922 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1923 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1924 mutex_exit(&spa->spa_props_lock);
1927 typedef struct spa_load_error {
1928 uint64_t sle_meta_count;
1929 uint64_t sle_data_count;
1933 spa_load_verify_done(zio_t *zio)
1935 blkptr_t *bp = zio->io_bp;
1936 spa_load_error_t *sle = zio->io_private;
1937 dmu_object_type_t type = BP_GET_TYPE(bp);
1938 int error = zio->io_error;
1939 spa_t *spa = zio->io_spa;
1941 abd_free(zio->io_abd);
1943 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1944 type != DMU_OT_INTENT_LOG)
1945 atomic_inc_64(&sle->sle_meta_count);
1947 atomic_inc_64(&sle->sle_data_count);
1950 mutex_enter(&spa->spa_scrub_lock);
1951 spa->spa_scrub_inflight--;
1952 cv_broadcast(&spa->spa_scrub_io_cv);
1953 mutex_exit(&spa->spa_scrub_lock);
1957 * Maximum number of concurrent scrub i/os to create while verifying
1958 * a pool while importing it.
1960 int spa_load_verify_maxinflight = 10000;
1961 boolean_t spa_load_verify_metadata = B_TRUE;
1962 boolean_t spa_load_verify_data = B_TRUE;
1964 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_maxinflight, CTLFLAG_RWTUN,
1965 &spa_load_verify_maxinflight, 0,
1966 "Maximum number of concurrent scrub I/Os to create while verifying a "
1967 "pool while importing it");
1969 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_metadata, CTLFLAG_RWTUN,
1970 &spa_load_verify_metadata, 0,
1971 "Check metadata on import?");
1973 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_data, CTLFLAG_RWTUN,
1974 &spa_load_verify_data, 0,
1975 "Check user data on import?");
1979 spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1980 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
1982 if (bp == NULL || BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp))
1985 * Note: normally this routine will not be called if
1986 * spa_load_verify_metadata is not set. However, it may be useful
1987 * to manually set the flag after the traversal has begun.
1989 if (!spa_load_verify_metadata)
1991 if (!BP_IS_METADATA(bp) && !spa_load_verify_data)
1995 size_t size = BP_GET_PSIZE(bp);
1997 mutex_enter(&spa->spa_scrub_lock);
1998 while (spa->spa_scrub_inflight >= spa_load_verify_maxinflight)
1999 cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
2000 spa->spa_scrub_inflight++;
2001 mutex_exit(&spa->spa_scrub_lock);
2003 zio_nowait(zio_read(rio, spa, bp, abd_alloc_for_io(size, B_FALSE), size,
2004 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
2005 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
2006 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
2012 verify_dataset_name_len(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg)
2014 if (dsl_dataset_namelen(ds) >= ZFS_MAX_DATASET_NAME_LEN)
2015 return (SET_ERROR(ENAMETOOLONG));
2021 spa_load_verify(spa_t *spa)
2024 spa_load_error_t sle = { 0 };
2025 zpool_rewind_policy_t policy;
2026 boolean_t verify_ok = B_FALSE;
2029 zpool_get_rewind_policy(spa->spa_config, &policy);
2031 if (policy.zrp_request & ZPOOL_NEVER_REWIND)
2034 dsl_pool_config_enter(spa->spa_dsl_pool, FTAG);
2035 error = dmu_objset_find_dp(spa->spa_dsl_pool,
2036 spa->spa_dsl_pool->dp_root_dir_obj, verify_dataset_name_len, NULL,
2038 dsl_pool_config_exit(spa->spa_dsl_pool, FTAG);
2042 rio = zio_root(spa, NULL, &sle,
2043 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
2045 if (spa_load_verify_metadata) {
2046 error = traverse_pool(spa, spa->spa_verify_min_txg,
2047 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA,
2048 spa_load_verify_cb, rio);
2051 (void) zio_wait(rio);
2053 spa->spa_load_meta_errors = sle.sle_meta_count;
2054 spa->spa_load_data_errors = sle.sle_data_count;
2056 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
2057 sle.sle_data_count <= policy.zrp_maxdata) {
2061 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
2062 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
2064 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
2065 VERIFY(nvlist_add_uint64(spa->spa_load_info,
2066 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
2067 VERIFY(nvlist_add_int64(spa->spa_load_info,
2068 ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
2069 VERIFY(nvlist_add_uint64(spa->spa_load_info,
2070 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
2072 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
2076 if (error != ENXIO && error != EIO)
2077 error = SET_ERROR(EIO);
2081 return (verify_ok ? 0 : EIO);
2085 * Find a value in the pool props object.
2088 spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
2090 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
2091 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
2095 * Find a value in the pool directory object.
2098 spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
2100 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
2101 name, sizeof (uint64_t), 1, val));
2105 spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
2107 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
2112 * Fix up config after a partly-completed split. This is done with the
2113 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
2114 * pool have that entry in their config, but only the splitting one contains
2115 * a list of all the guids of the vdevs that are being split off.
2117 * This function determines what to do with that list: either rejoin
2118 * all the disks to the pool, or complete the splitting process. To attempt
2119 * the rejoin, each disk that is offlined is marked online again, and
2120 * we do a reopen() call. If the vdev label for every disk that was
2121 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
2122 * then we call vdev_split() on each disk, and complete the split.
2124 * Otherwise we leave the config alone, with all the vdevs in place in
2125 * the original pool.
2128 spa_try_repair(spa_t *spa, nvlist_t *config)
2135 boolean_t attempt_reopen;
2137 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
2140 /* check that the config is complete */
2141 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
2142 &glist, &gcount) != 0)
2145 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
2147 /* attempt to online all the vdevs & validate */
2148 attempt_reopen = B_TRUE;
2149 for (i = 0; i < gcount; i++) {
2150 if (glist[i] == 0) /* vdev is hole */
2153 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
2154 if (vd[i] == NULL) {
2156 * Don't bother attempting to reopen the disks;
2157 * just do the split.
2159 attempt_reopen = B_FALSE;
2161 /* attempt to re-online it */
2162 vd[i]->vdev_offline = B_FALSE;
2166 if (attempt_reopen) {
2167 vdev_reopen(spa->spa_root_vdev);
2169 /* check each device to see what state it's in */
2170 for (extracted = 0, i = 0; i < gcount; i++) {
2171 if (vd[i] != NULL &&
2172 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
2179 * If every disk has been moved to the new pool, or if we never
2180 * even attempted to look at them, then we split them off for
2183 if (!attempt_reopen || gcount == extracted) {
2184 for (i = 0; i < gcount; i++)
2187 vdev_reopen(spa->spa_root_vdev);
2190 kmem_free(vd, gcount * sizeof (vdev_t *));
2194 spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
2195 boolean_t mosconfig)
2197 nvlist_t *config = spa->spa_config;
2198 char *ereport = FM_EREPORT_ZFS_POOL;
2204 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
2205 return (SET_ERROR(EINVAL));
2207 ASSERT(spa->spa_comment == NULL);
2208 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
2209 spa->spa_comment = spa_strdup(comment);
2212 * Versioning wasn't explicitly added to the label until later, so if
2213 * it's not present treat it as the initial version.
2215 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
2216 &spa->spa_ubsync.ub_version) != 0)
2217 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
2219 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
2220 &spa->spa_config_txg);
2222 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
2223 spa_guid_exists(pool_guid, 0)) {
2224 error = SET_ERROR(EEXIST);
2226 spa->spa_config_guid = pool_guid;
2228 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
2230 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
2234 nvlist_free(spa->spa_load_info);
2235 spa->spa_load_info = fnvlist_alloc();
2237 gethrestime(&spa->spa_loaded_ts);
2238 error = spa_load_impl(spa, pool_guid, config, state, type,
2239 mosconfig, &ereport);
2243 * Don't count references from objsets that are already closed
2244 * and are making their way through the eviction process.
2246 spa_evicting_os_wait(spa);
2247 spa->spa_minref = refcount_count(&spa->spa_refcount);
2249 if (error != EEXIST) {
2250 spa->spa_loaded_ts.tv_sec = 0;
2251 spa->spa_loaded_ts.tv_nsec = 0;
2253 if (error != EBADF) {
2254 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
2257 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2264 * Count the number of per-vdev ZAPs associated with all of the vdevs in the
2265 * vdev tree rooted in the given vd, and ensure that each ZAP is present in the
2266 * spa's per-vdev ZAP list.
2269 vdev_count_verify_zaps(vdev_t *vd)
2271 spa_t *spa = vd->vdev_spa;
2273 if (vd->vdev_top_zap != 0) {
2275 ASSERT0(zap_lookup_int(spa->spa_meta_objset,
2276 spa->spa_all_vdev_zaps, vd->vdev_top_zap));
2278 if (vd->vdev_leaf_zap != 0) {
2280 ASSERT0(zap_lookup_int(spa->spa_meta_objset,
2281 spa->spa_all_vdev_zaps, vd->vdev_leaf_zap));
2284 for (uint64_t i = 0; i < vd->vdev_children; i++) {
2285 total += vdev_count_verify_zaps(vd->vdev_child[i]);
2292 * Load an existing storage pool, using the pool's builtin spa_config as a
2293 * source of configuration information.
2296 spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
2297 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
2301 nvlist_t *nvroot = NULL;
2304 uberblock_t *ub = &spa->spa_uberblock;
2305 uint64_t children, config_cache_txg = spa->spa_config_txg;
2306 int orig_mode = spa->spa_mode;
2309 boolean_t missing_feat_write = B_FALSE;
2312 * If this is an untrusted config, access the pool in read-only mode.
2313 * This prevents things like resilvering recently removed devices.
2316 spa->spa_mode = FREAD;
2318 ASSERT(MUTEX_HELD(&spa_namespace_lock));
2320 spa->spa_load_state = state;
2322 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
2323 return (SET_ERROR(EINVAL));
2325 parse = (type == SPA_IMPORT_EXISTING ?
2326 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2329 * Create "The Godfather" zio to hold all async IOs
2331 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
2333 for (int i = 0; i < max_ncpus; i++) {
2334 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
2335 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
2336 ZIO_FLAG_GODFATHER);
2340 * Parse the configuration into a vdev tree. We explicitly set the
2341 * value that will be returned by spa_version() since parsing the
2342 * configuration requires knowing the version number.
2344 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2345 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
2346 spa_config_exit(spa, SCL_ALL, FTAG);
2351 ASSERT(spa->spa_root_vdev == rvd);
2352 ASSERT3U(spa->spa_min_ashift, >=, SPA_MINBLOCKSHIFT);
2353 ASSERT3U(spa->spa_max_ashift, <=, SPA_MAXBLOCKSHIFT);
2355 if (type != SPA_IMPORT_ASSEMBLE) {
2356 ASSERT(spa_guid(spa) == pool_guid);
2360 * Try to open all vdevs, loading each label in the process.
2362 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2363 error = vdev_open(rvd);
2364 spa_config_exit(spa, SCL_ALL, FTAG);
2369 * We need to validate the vdev labels against the configuration that
2370 * we have in hand, which is dependent on the setting of mosconfig. If
2371 * mosconfig is true then we're validating the vdev labels based on
2372 * that config. Otherwise, we're validating against the cached config
2373 * (zpool.cache) that was read when we loaded the zfs module, and then
2374 * later we will recursively call spa_load() and validate against
2377 * If we're assembling a new pool that's been split off from an
2378 * existing pool, the labels haven't yet been updated so we skip
2379 * validation for now.
2381 if (type != SPA_IMPORT_ASSEMBLE) {
2382 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2383 error = vdev_validate(rvd, mosconfig);
2384 spa_config_exit(spa, SCL_ALL, FTAG);
2389 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2390 return (SET_ERROR(ENXIO));
2394 * Find the best uberblock.
2396 vdev_uberblock_load(rvd, ub, &label);
2399 * If we weren't able to find a single valid uberblock, return failure.
2401 if (ub->ub_txg == 0) {
2403 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2407 * If the pool has an unsupported version we can't open it.
2409 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2411 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2414 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2418 * If we weren't able to find what's necessary for reading the
2419 * MOS in the label, return failure.
2421 if (label == NULL || nvlist_lookup_nvlist(label,
2422 ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) {
2424 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2429 * Update our in-core representation with the definitive values
2432 nvlist_free(spa->spa_label_features);
2433 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2439 * Look through entries in the label nvlist's features_for_read. If
2440 * there is a feature listed there which we don't understand then we
2441 * cannot open a pool.
2443 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2444 nvlist_t *unsup_feat;
2446 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2449 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2451 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2452 if (!zfeature_is_supported(nvpair_name(nvp))) {
2453 VERIFY(nvlist_add_string(unsup_feat,
2454 nvpair_name(nvp), "") == 0);
2458 if (!nvlist_empty(unsup_feat)) {
2459 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2460 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2461 nvlist_free(unsup_feat);
2462 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2466 nvlist_free(unsup_feat);
2470 * If the vdev guid sum doesn't match the uberblock, we have an
2471 * incomplete configuration. We first check to see if the pool
2472 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2473 * If it is, defer the vdev_guid_sum check till later so we
2474 * can handle missing vdevs.
2476 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
2477 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
2478 rvd->vdev_guid_sum != ub->ub_guid_sum)
2479 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2481 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2482 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2483 spa_try_repair(spa, config);
2484 spa_config_exit(spa, SCL_ALL, FTAG);
2485 nvlist_free(spa->spa_config_splitting);
2486 spa->spa_config_splitting = NULL;
2490 * Initialize internal SPA structures.
2492 spa->spa_state = POOL_STATE_ACTIVE;
2493 spa->spa_ubsync = spa->spa_uberblock;
2494 spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2495 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2496 spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2497 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2498 spa->spa_claim_max_txg = spa->spa_first_txg;
2499 spa->spa_prev_software_version = ub->ub_software_version;
2501 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2503 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2504 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2506 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2507 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2509 if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2510 boolean_t missing_feat_read = B_FALSE;
2511 nvlist_t *unsup_feat, *enabled_feat;
2513 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2514 &spa->spa_feat_for_read_obj) != 0) {
2515 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2518 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2519 &spa->spa_feat_for_write_obj) != 0) {
2520 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2523 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2524 &spa->spa_feat_desc_obj) != 0) {
2525 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2528 enabled_feat = fnvlist_alloc();
2529 unsup_feat = fnvlist_alloc();
2531 if (!spa_features_check(spa, B_FALSE,
2532 unsup_feat, enabled_feat))
2533 missing_feat_read = B_TRUE;
2535 if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) {
2536 if (!spa_features_check(spa, B_TRUE,
2537 unsup_feat, enabled_feat)) {
2538 missing_feat_write = B_TRUE;
2542 fnvlist_add_nvlist(spa->spa_load_info,
2543 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
2545 if (!nvlist_empty(unsup_feat)) {
2546 fnvlist_add_nvlist(spa->spa_load_info,
2547 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
2550 fnvlist_free(enabled_feat);
2551 fnvlist_free(unsup_feat);
2553 if (!missing_feat_read) {
2554 fnvlist_add_boolean(spa->spa_load_info,
2555 ZPOOL_CONFIG_CAN_RDONLY);
2559 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2560 * twofold: to determine whether the pool is available for
2561 * import in read-write mode and (if it is not) whether the
2562 * pool is available for import in read-only mode. If the pool
2563 * is available for import in read-write mode, it is displayed
2564 * as available in userland; if it is not available for import
2565 * in read-only mode, it is displayed as unavailable in
2566 * userland. If the pool is available for import in read-only
2567 * mode but not read-write mode, it is displayed as unavailable
2568 * in userland with a special note that the pool is actually
2569 * available for open in read-only mode.
2571 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2572 * missing a feature for write, we must first determine whether
2573 * the pool can be opened read-only before returning to
2574 * userland in order to know whether to display the
2575 * abovementioned note.
2577 if (missing_feat_read || (missing_feat_write &&
2578 spa_writeable(spa))) {
2579 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2584 * Load refcounts for ZFS features from disk into an in-memory
2585 * cache during SPA initialization.
2587 for (spa_feature_t i = 0; i < SPA_FEATURES; i++) {
2590 error = feature_get_refcount_from_disk(spa,
2591 &spa_feature_table[i], &refcount);
2593 spa->spa_feat_refcount_cache[i] = refcount;
2594 } else if (error == ENOTSUP) {
2595 spa->spa_feat_refcount_cache[i] =
2596 SPA_FEATURE_DISABLED;
2598 return (spa_vdev_err(rvd,
2599 VDEV_AUX_CORRUPT_DATA, EIO));
2604 if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) {
2605 if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG,
2606 &spa->spa_feat_enabled_txg_obj) != 0)
2607 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2610 spa->spa_is_initializing = B_TRUE;
2611 error = dsl_pool_open(spa->spa_dsl_pool);
2612 spa->spa_is_initializing = B_FALSE;
2614 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2618 nvlist_t *policy = NULL, *nvconfig;
2620 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2621 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2623 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2624 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2626 unsigned long myhostid = 0;
2628 VERIFY(nvlist_lookup_string(nvconfig,
2629 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2632 myhostid = zone_get_hostid(NULL);
2635 * We're emulating the system's hostid in userland, so
2636 * we can't use zone_get_hostid().
2638 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2639 #endif /* _KERNEL */
2640 if (check_hostid && hostid != 0 && myhostid != 0 &&
2641 hostid != myhostid) {
2642 nvlist_free(nvconfig);
2643 cmn_err(CE_WARN, "pool '%s' could not be "
2644 "loaded as it was last accessed by "
2645 "another system (host: %s hostid: 0x%lx). "
2646 "See: http://illumos.org/msg/ZFS-8000-EY",
2647 spa_name(spa), hostname,
2648 (unsigned long)hostid);
2649 return (SET_ERROR(EBADF));
2652 if (nvlist_lookup_nvlist(spa->spa_config,
2653 ZPOOL_REWIND_POLICY, &policy) == 0)
2654 VERIFY(nvlist_add_nvlist(nvconfig,
2655 ZPOOL_REWIND_POLICY, policy) == 0);
2657 spa_config_set(spa, nvconfig);
2659 spa_deactivate(spa);
2660 spa_activate(spa, orig_mode);
2662 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2665 /* Grab the secret checksum salt from the MOS. */
2666 error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
2667 DMU_POOL_CHECKSUM_SALT, 1,
2668 sizeof (spa->spa_cksum_salt.zcs_bytes),
2669 spa->spa_cksum_salt.zcs_bytes);
2670 if (error == ENOENT) {
2671 /* Generate a new salt for subsequent use */
2672 (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes,
2673 sizeof (spa->spa_cksum_salt.zcs_bytes));
2674 } else if (error != 0) {
2675 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2678 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2679 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2680 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2682 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2685 * Load the bit that tells us to use the new accounting function
2686 * (raid-z deflation). If we have an older pool, this will not
2689 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2690 if (error != 0 && error != ENOENT)
2691 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2693 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2694 &spa->spa_creation_version);
2695 if (error != 0 && error != ENOENT)
2696 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2699 * Load the persistent error log. If we have an older pool, this will
2702 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2703 if (error != 0 && error != ENOENT)
2704 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2706 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2707 &spa->spa_errlog_scrub);
2708 if (error != 0 && error != ENOENT)
2709 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2712 * Load the history object. If we have an older pool, this
2713 * will not be present.
2715 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2716 if (error != 0 && error != ENOENT)
2717 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2720 * Load the per-vdev ZAP map. If we have an older pool, this will not
2721 * be present; in this case, defer its creation to a later time to
2722 * avoid dirtying the MOS this early / out of sync context. See
2723 * spa_sync_config_object.
2726 /* The sentinel is only available in the MOS config. */
2727 nvlist_t *mos_config;
2728 if (load_nvlist(spa, spa->spa_config_object, &mos_config) != 0)
2729 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2731 error = spa_dir_prop(spa, DMU_POOL_VDEV_ZAP_MAP,
2732 &spa->spa_all_vdev_zaps);
2734 if (error == ENOENT) {
2735 VERIFY(!nvlist_exists(mos_config,
2736 ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS));
2737 spa->spa_avz_action = AVZ_ACTION_INITIALIZE;
2738 ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev));
2739 } else if (error != 0) {
2740 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2741 } else if (!nvlist_exists(mos_config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)) {
2743 * An older version of ZFS overwrote the sentinel value, so
2744 * we have orphaned per-vdev ZAPs in the MOS. Defer their
2745 * destruction to later; see spa_sync_config_object.
2747 spa->spa_avz_action = AVZ_ACTION_DESTROY;
2749 * We're assuming that no vdevs have had their ZAPs created
2750 * before this. Better be sure of it.
2752 ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev));
2754 nvlist_free(mos_config);
2757 * If we're assembling the pool from the split-off vdevs of
2758 * an existing pool, we don't want to attach the spares & cache
2763 * Load any hot spares for this pool.
2765 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2766 if (error != 0 && error != ENOENT)
2767 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2768 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2769 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2770 if (load_nvlist(spa, spa->spa_spares.sav_object,
2771 &spa->spa_spares.sav_config) != 0)
2772 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2774 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2775 spa_load_spares(spa);
2776 spa_config_exit(spa, SCL_ALL, FTAG);
2777 } else if (error == 0) {
2778 spa->spa_spares.sav_sync = B_TRUE;
2782 * Load any level 2 ARC devices for this pool.
2784 error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2785 &spa->spa_l2cache.sav_object);
2786 if (error != 0 && error != ENOENT)
2787 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2788 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2789 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2790 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2791 &spa->spa_l2cache.sav_config) != 0)
2792 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2794 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2795 spa_load_l2cache(spa);
2796 spa_config_exit(spa, SCL_ALL, FTAG);
2797 } else if (error == 0) {
2798 spa->spa_l2cache.sav_sync = B_TRUE;
2801 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2803 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2804 if (error && error != ENOENT)
2805 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2808 uint64_t autoreplace;
2810 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2811 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2812 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2813 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2814 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2815 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2816 &spa->spa_dedup_ditto);
2818 spa->spa_autoreplace = (autoreplace != 0);
2822 * If the 'autoreplace' property is set, then post a resource notifying
2823 * the ZFS DE that it should not issue any faults for unopenable
2824 * devices. We also iterate over the vdevs, and post a sysevent for any
2825 * unopenable vdevs so that the normal autoreplace handler can take
2828 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2829 spa_check_removed(spa->spa_root_vdev);
2831 * For the import case, this is done in spa_import(), because
2832 * at this point we're using the spare definitions from
2833 * the MOS config, not necessarily from the userland config.
2835 if (state != SPA_LOAD_IMPORT) {
2836 spa_aux_check_removed(&spa->spa_spares);
2837 spa_aux_check_removed(&spa->spa_l2cache);
2842 * Load the vdev state for all toplevel vdevs.
2847 * Propagate the leaf DTLs we just loaded all the way up the tree.
2849 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2850 vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2851 spa_config_exit(spa, SCL_ALL, FTAG);
2854 * Load the DDTs (dedup tables).
2856 error = ddt_load(spa);
2858 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2860 spa_update_dspace(spa);
2863 * Validate the config, using the MOS config to fill in any
2864 * information which might be missing. If we fail to validate
2865 * the config then declare the pool unfit for use. If we're
2866 * assembling a pool from a split, the log is not transferred
2869 if (type != SPA_IMPORT_ASSEMBLE) {
2872 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2873 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2875 if (!spa_config_valid(spa, nvconfig)) {
2876 nvlist_free(nvconfig);
2877 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2880 nvlist_free(nvconfig);
2883 * Now that we've validated the config, check the state of the
2884 * root vdev. If it can't be opened, it indicates one or
2885 * more toplevel vdevs are faulted.
2887 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2888 return (SET_ERROR(ENXIO));
2890 if (spa_writeable(spa) && spa_check_logs(spa)) {
2891 *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2892 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2896 if (missing_feat_write) {
2897 ASSERT(state == SPA_LOAD_TRYIMPORT);
2900 * At this point, we know that we can open the pool in
2901 * read-only mode but not read-write mode. We now have enough
2902 * information and can return to userland.
2904 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
2908 * We've successfully opened the pool, verify that we're ready
2909 * to start pushing transactions.
2911 if (state != SPA_LOAD_TRYIMPORT) {
2912 if (error = spa_load_verify(spa))
2913 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2917 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2918 spa->spa_load_max_txg == UINT64_MAX)) {
2920 int need_update = B_FALSE;
2921 dsl_pool_t *dp = spa_get_dsl(spa);
2923 ASSERT(state != SPA_LOAD_TRYIMPORT);
2926 * Claim log blocks that haven't been committed yet.
2927 * This must all happen in a single txg.
2928 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2929 * invoked from zil_claim_log_block()'s i/o done callback.
2930 * Price of rollback is that we abandon the log.
2932 spa->spa_claiming = B_TRUE;
2934 tx = dmu_tx_create_assigned(dp, spa_first_txg(spa));
2935 (void) dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
2936 zil_claim, tx, DS_FIND_CHILDREN);
2939 spa->spa_claiming = B_FALSE;
2941 spa_set_log_state(spa, SPA_LOG_GOOD);
2942 spa->spa_sync_on = B_TRUE;
2943 txg_sync_start(spa->spa_dsl_pool);
2946 * Wait for all claims to sync. We sync up to the highest
2947 * claimed log block birth time so that claimed log blocks
2948 * don't appear to be from the future. spa_claim_max_txg
2949 * will have been set for us by either zil_check_log_chain()
2950 * (invoked from spa_check_logs()) or zil_claim() above.
2952 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2955 * If the config cache is stale, or we have uninitialized
2956 * metaslabs (see spa_vdev_add()), then update the config.
2958 * If this is a verbatim import, trust the current
2959 * in-core spa_config and update the disk labels.
2961 if (config_cache_txg != spa->spa_config_txg ||
2962 state == SPA_LOAD_IMPORT ||
2963 state == SPA_LOAD_RECOVER ||
2964 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2965 need_update = B_TRUE;
2967 for (int c = 0; c < rvd->vdev_children; c++)
2968 if (rvd->vdev_child[c]->vdev_ms_array == 0)
2969 need_update = B_TRUE;
2972 * Update the config cache asychronously in case we're the
2973 * root pool, in which case the config cache isn't writable yet.
2976 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2979 * Check all DTLs to see if anything needs resilvering.
2981 if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2982 vdev_resilver_needed(rvd, NULL, NULL))
2983 spa_async_request(spa, SPA_ASYNC_RESILVER);
2986 * Log the fact that we booted up (so that we can detect if
2987 * we rebooted in the middle of an operation).
2989 spa_history_log_version(spa, "open");
2992 * Delete any inconsistent datasets.
2994 (void) dmu_objset_find(spa_name(spa),
2995 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2998 * Clean up any stale temporary dataset userrefs.
3000 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
3007 spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
3009 int mode = spa->spa_mode;
3012 spa_deactivate(spa);
3014 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg - 1;
3016 spa_activate(spa, mode);
3017 spa_async_suspend(spa);
3019 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
3023 * If spa_load() fails this function will try loading prior txg's. If
3024 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
3025 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
3026 * function will not rewind the pool and will return the same error as
3030 spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
3031 uint64_t max_request, int rewind_flags)
3033 nvlist_t *loadinfo = NULL;
3034 nvlist_t *config = NULL;
3035 int load_error, rewind_error;
3036 uint64_t safe_rewind_txg;
3039 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
3040 spa->spa_load_max_txg = spa->spa_load_txg;
3041 spa_set_log_state(spa, SPA_LOG_CLEAR);
3043 spa->spa_load_max_txg = max_request;
3044 if (max_request != UINT64_MAX)
3045 spa->spa_extreme_rewind = B_TRUE;
3048 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
3050 if (load_error == 0)
3053 if (spa->spa_root_vdev != NULL)
3054 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
3056 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
3057 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
3059 if (rewind_flags & ZPOOL_NEVER_REWIND) {
3060 nvlist_free(config);
3061 return (load_error);
3064 if (state == SPA_LOAD_RECOVER) {
3065 /* Price of rolling back is discarding txgs, including log */
3066 spa_set_log_state(spa, SPA_LOG_CLEAR);
3069 * If we aren't rolling back save the load info from our first
3070 * import attempt so that we can restore it after attempting
3073 loadinfo = spa->spa_load_info;
3074 spa->spa_load_info = fnvlist_alloc();
3077 spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
3078 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
3079 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
3080 TXG_INITIAL : safe_rewind_txg;
3083 * Continue as long as we're finding errors, we're still within
3084 * the acceptable rewind range, and we're still finding uberblocks
3086 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
3087 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
3088 if (spa->spa_load_max_txg < safe_rewind_txg)
3089 spa->spa_extreme_rewind = B_TRUE;
3090 rewind_error = spa_load_retry(spa, state, mosconfig);
3093 spa->spa_extreme_rewind = B_FALSE;
3094 spa->spa_load_max_txg = UINT64_MAX;
3096 if (config && (rewind_error || state != SPA_LOAD_RECOVER))
3097 spa_config_set(spa, config);
3099 if (state == SPA_LOAD_RECOVER) {
3100 ASSERT3P(loadinfo, ==, NULL);
3101 return (rewind_error);
3103 /* Store the rewind info as part of the initial load info */
3104 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
3105 spa->spa_load_info);
3107 /* Restore the initial load info */
3108 fnvlist_free(spa->spa_load_info);
3109 spa->spa_load_info = loadinfo;
3111 return (load_error);
3118 * The import case is identical to an open except that the configuration is sent
3119 * down from userland, instead of grabbed from the configuration cache. For the
3120 * case of an open, the pool configuration will exist in the
3121 * POOL_STATE_UNINITIALIZED state.
3123 * The stats information (gen/count/ustats) is used to gather vdev statistics at
3124 * the same time open the pool, without having to keep around the spa_t in some
3128 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
3132 spa_load_state_t state = SPA_LOAD_OPEN;
3134 int locked = B_FALSE;
3135 int firstopen = B_FALSE;
3140 * As disgusting as this is, we need to support recursive calls to this
3141 * function because dsl_dir_open() is called during spa_load(), and ends
3142 * up calling spa_open() again. The real fix is to figure out how to
3143 * avoid dsl_dir_open() calling this in the first place.
3145 if (mutex_owner(&spa_namespace_lock) != curthread) {
3146 mutex_enter(&spa_namespace_lock);
3150 if ((spa = spa_lookup(pool)) == NULL) {
3152 mutex_exit(&spa_namespace_lock);
3153 return (SET_ERROR(ENOENT));
3156 if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
3157 zpool_rewind_policy_t policy;
3161 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
3163 if (policy.zrp_request & ZPOOL_DO_REWIND)
3164 state = SPA_LOAD_RECOVER;
3166 spa_activate(spa, spa_mode_global);
3168 if (state != SPA_LOAD_RECOVER)
3169 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3171 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
3172 policy.zrp_request);
3174 if (error == EBADF) {
3176 * If vdev_validate() returns failure (indicated by
3177 * EBADF), it indicates that one of the vdevs indicates
3178 * that the pool has been exported or destroyed. If
3179 * this is the case, the config cache is out of sync and
3180 * we should remove the pool from the namespace.
3183 spa_deactivate(spa);
3184 spa_config_sync(spa, B_TRUE, B_TRUE);
3187 mutex_exit(&spa_namespace_lock);
3188 return (SET_ERROR(ENOENT));
3193 * We can't open the pool, but we still have useful
3194 * information: the state of each vdev after the
3195 * attempted vdev_open(). Return this to the user.
3197 if (config != NULL && spa->spa_config) {
3198 VERIFY(nvlist_dup(spa->spa_config, config,
3200 VERIFY(nvlist_add_nvlist(*config,
3201 ZPOOL_CONFIG_LOAD_INFO,
3202 spa->spa_load_info) == 0);
3205 spa_deactivate(spa);
3206 spa->spa_last_open_failed = error;
3208 mutex_exit(&spa_namespace_lock);
3214 spa_open_ref(spa, tag);
3217 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
3220 * If we've recovered the pool, pass back any information we
3221 * gathered while doing the load.
3223 if (state == SPA_LOAD_RECOVER) {
3224 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
3225 spa->spa_load_info) == 0);
3229 spa->spa_last_open_failed = 0;
3230 spa->spa_last_ubsync_txg = 0;
3231 spa->spa_load_txg = 0;
3232 mutex_exit(&spa_namespace_lock);
3236 zvol_create_minors(spa->spa_name);
3247 spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
3250 return (spa_open_common(name, spapp, tag, policy, config));
3254 spa_open(const char *name, spa_t **spapp, void *tag)
3256 return (spa_open_common(name, spapp, tag, NULL, NULL));
3260 * Lookup the given spa_t, incrementing the inject count in the process,
3261 * preventing it from being exported or destroyed.
3264 spa_inject_addref(char *name)
3268 mutex_enter(&spa_namespace_lock);
3269 if ((spa = spa_lookup(name)) == NULL) {
3270 mutex_exit(&spa_namespace_lock);
3273 spa->spa_inject_ref++;
3274 mutex_exit(&spa_namespace_lock);
3280 spa_inject_delref(spa_t *spa)
3282 mutex_enter(&spa_namespace_lock);
3283 spa->spa_inject_ref--;
3284 mutex_exit(&spa_namespace_lock);
3288 * Add spares device information to the nvlist.
3291 spa_add_spares(spa_t *spa, nvlist_t *config)
3301 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3303 if (spa->spa_spares.sav_count == 0)
3306 VERIFY(nvlist_lookup_nvlist(config,
3307 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3308 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
3309 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3311 VERIFY(nvlist_add_nvlist_array(nvroot,
3312 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3313 VERIFY(nvlist_lookup_nvlist_array(nvroot,
3314 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3317 * Go through and find any spares which have since been
3318 * repurposed as an active spare. If this is the case, update
3319 * their status appropriately.
3321 for (i = 0; i < nspares; i++) {
3322 VERIFY(nvlist_lookup_uint64(spares[i],
3323 ZPOOL_CONFIG_GUID, &guid) == 0);
3324 if (spa_spare_exists(guid, &pool, NULL) &&
3326 VERIFY(nvlist_lookup_uint64_array(
3327 spares[i], ZPOOL_CONFIG_VDEV_STATS,
3328 (uint64_t **)&vs, &vsc) == 0);
3329 vs->vs_state = VDEV_STATE_CANT_OPEN;
3330 vs->vs_aux = VDEV_AUX_SPARED;
3337 * Add l2cache device information to the nvlist, including vdev stats.
3340 spa_add_l2cache(spa_t *spa, nvlist_t *config)
3343 uint_t i, j, nl2cache;
3350 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3352 if (spa->spa_l2cache.sav_count == 0)
3355 VERIFY(nvlist_lookup_nvlist(config,
3356 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3357 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
3358 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3359 if (nl2cache != 0) {
3360 VERIFY(nvlist_add_nvlist_array(nvroot,
3361 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3362 VERIFY(nvlist_lookup_nvlist_array(nvroot,
3363 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3366 * Update level 2 cache device stats.
3369 for (i = 0; i < nl2cache; i++) {
3370 VERIFY(nvlist_lookup_uint64(l2cache[i],
3371 ZPOOL_CONFIG_GUID, &guid) == 0);
3374 for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
3376 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
3377 vd = spa->spa_l2cache.sav_vdevs[j];
3383 VERIFY(nvlist_lookup_uint64_array(l2cache[i],
3384 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
3386 vdev_get_stats(vd, vs);
3392 spa_add_feature_stats(spa_t *spa, nvlist_t *config)
3398 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3399 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3401 /* We may be unable to read features if pool is suspended. */
3402 if (spa_suspended(spa))
3405 if (spa->spa_feat_for_read_obj != 0) {
3406 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3407 spa->spa_feat_for_read_obj);
3408 zap_cursor_retrieve(&zc, &za) == 0;
3409 zap_cursor_advance(&zc)) {
3410 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3411 za.za_num_integers == 1);
3412 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3413 za.za_first_integer));
3415 zap_cursor_fini(&zc);
3418 if (spa->spa_feat_for_write_obj != 0) {
3419 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3420 spa->spa_feat_for_write_obj);
3421 zap_cursor_retrieve(&zc, &za) == 0;
3422 zap_cursor_advance(&zc)) {
3423 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3424 za.za_num_integers == 1);
3425 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3426 za.za_first_integer));
3428 zap_cursor_fini(&zc);
3432 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
3434 nvlist_free(features);
3438 spa_get_stats(const char *name, nvlist_t **config,
3439 char *altroot, size_t buflen)
3445 error = spa_open_common(name, &spa, FTAG, NULL, config);
3449 * This still leaves a window of inconsistency where the spares
3450 * or l2cache devices could change and the config would be
3451 * self-inconsistent.
3453 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3455 if (*config != NULL) {
3456 uint64_t loadtimes[2];
3458 loadtimes[0] = spa->spa_loaded_ts.tv_sec;
3459 loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
3460 VERIFY(nvlist_add_uint64_array(*config,
3461 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
3463 VERIFY(nvlist_add_uint64(*config,
3464 ZPOOL_CONFIG_ERRCOUNT,
3465 spa_get_errlog_size(spa)) == 0);
3467 if (spa_suspended(spa))
3468 VERIFY(nvlist_add_uint64(*config,
3469 ZPOOL_CONFIG_SUSPENDED,
3470 spa->spa_failmode) == 0);
3472 spa_add_spares(spa, *config);
3473 spa_add_l2cache(spa, *config);
3474 spa_add_feature_stats(spa, *config);
3479 * We want to get the alternate root even for faulted pools, so we cheat
3480 * and call spa_lookup() directly.
3484 mutex_enter(&spa_namespace_lock);
3485 spa = spa_lookup(name);
3487 spa_altroot(spa, altroot, buflen);
3491 mutex_exit(&spa_namespace_lock);
3493 spa_altroot(spa, altroot, buflen);
3498 spa_config_exit(spa, SCL_CONFIG, FTAG);
3499 spa_close(spa, FTAG);
3506 * Validate that the auxiliary device array is well formed. We must have an
3507 * array of nvlists, each which describes a valid leaf vdev. If this is an
3508 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3509 * specified, as long as they are well-formed.
3512 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
3513 spa_aux_vdev_t *sav, const char *config, uint64_t version,
3514 vdev_labeltype_t label)
3521 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3524 * It's acceptable to have no devs specified.
3526 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
3530 return (SET_ERROR(EINVAL));
3533 * Make sure the pool is formatted with a version that supports this
3536 if (spa_version(spa) < version)
3537 return (SET_ERROR(ENOTSUP));
3540 * Set the pending device list so we correctly handle device in-use
3543 sav->sav_pending = dev;
3544 sav->sav_npending = ndev;
3546 for (i = 0; i < ndev; i++) {
3547 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
3551 if (!vd->vdev_ops->vdev_op_leaf) {
3553 error = SET_ERROR(EINVAL);
3558 * The L2ARC currently only supports disk devices in
3559 * kernel context. For user-level testing, we allow it.
3562 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
3563 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
3564 error = SET_ERROR(ENOTBLK);
3571 if ((error = vdev_open(vd)) == 0 &&
3572 (error = vdev_label_init(vd, crtxg, label)) == 0) {
3573 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
3574 vd->vdev_guid) == 0);
3580 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
3587 sav->sav_pending = NULL;
3588 sav->sav_npending = 0;
3593 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
3597 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3599 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3600 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
3601 VDEV_LABEL_SPARE)) != 0) {
3605 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3606 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
3607 VDEV_LABEL_L2CACHE));
3611 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
3616 if (sav->sav_config != NULL) {
3622 * Generate new dev list by concatentating with the
3625 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
3626 &olddevs, &oldndevs) == 0);
3628 newdevs = kmem_alloc(sizeof (void *) *
3629 (ndevs + oldndevs), KM_SLEEP);
3630 for (i = 0; i < oldndevs; i++)
3631 VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
3633 for (i = 0; i < ndevs; i++)
3634 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
3637 VERIFY(nvlist_remove(sav->sav_config, config,
3638 DATA_TYPE_NVLIST_ARRAY) == 0);
3640 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3641 config, newdevs, ndevs + oldndevs) == 0);
3642 for (i = 0; i < oldndevs + ndevs; i++)
3643 nvlist_free(newdevs[i]);
3644 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
3647 * Generate a new dev list.
3649 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
3651 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
3657 * Stop and drop level 2 ARC devices
3660 spa_l2cache_drop(spa_t *spa)
3664 spa_aux_vdev_t *sav = &spa->spa_l2cache;
3666 for (i = 0; i < sav->sav_count; i++) {
3669 vd = sav->sav_vdevs[i];
3672 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
3673 pool != 0ULL && l2arc_vdev_present(vd))
3674 l2arc_remove_vdev(vd);
3682 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
3686 char *altroot = NULL;
3691 uint64_t txg = TXG_INITIAL;
3692 nvlist_t **spares, **l2cache;
3693 uint_t nspares, nl2cache;
3694 uint64_t version, obj;
3695 boolean_t has_features;
3698 * If this pool already exists, return failure.
3700 mutex_enter(&spa_namespace_lock);
3701 if (spa_lookup(pool) != NULL) {
3702 mutex_exit(&spa_namespace_lock);
3703 return (SET_ERROR(EEXIST));
3707 * Allocate a new spa_t structure.
3709 (void) nvlist_lookup_string(props,
3710 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3711 spa = spa_add(pool, NULL, altroot);
3712 spa_activate(spa, spa_mode_global);
3714 if (props && (error = spa_prop_validate(spa, props))) {
3715 spa_deactivate(spa);
3717 mutex_exit(&spa_namespace_lock);
3721 has_features = B_FALSE;
3722 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
3723 elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
3724 if (zpool_prop_feature(nvpair_name(elem)))
3725 has_features = B_TRUE;
3728 if (has_features || nvlist_lookup_uint64(props,
3729 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
3730 version = SPA_VERSION;
3732 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
3734 spa->spa_first_txg = txg;
3735 spa->spa_uberblock.ub_txg = txg - 1;
3736 spa->spa_uberblock.ub_version = version;
3737 spa->spa_ubsync = spa->spa_uberblock;
3738 spa->spa_load_state = SPA_LOAD_CREATE;
3741 * Create "The Godfather" zio to hold all async IOs
3743 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
3745 for (int i = 0; i < max_ncpus; i++) {
3746 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
3747 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
3748 ZIO_FLAG_GODFATHER);
3752 * Create the root vdev.
3754 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3756 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3758 ASSERT(error != 0 || rvd != NULL);
3759 ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3761 if (error == 0 && !zfs_allocatable_devs(nvroot))
3762 error = SET_ERROR(EINVAL);
3765 (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3766 (error = spa_validate_aux(spa, nvroot, txg,
3767 VDEV_ALLOC_ADD)) == 0) {
3768 for (int c = 0; c < rvd->vdev_children; c++) {
3769 vdev_ashift_optimize(rvd->vdev_child[c]);
3770 vdev_metaslab_set_size(rvd->vdev_child[c]);
3771 vdev_expand(rvd->vdev_child[c], txg);
3775 spa_config_exit(spa, SCL_ALL, FTAG);
3779 spa_deactivate(spa);
3781 mutex_exit(&spa_namespace_lock);
3786 * Get the list of spares, if specified.
3788 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3789 &spares, &nspares) == 0) {
3790 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3792 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3793 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3794 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3795 spa_load_spares(spa);
3796 spa_config_exit(spa, SCL_ALL, FTAG);
3797 spa->spa_spares.sav_sync = B_TRUE;
3801 * Get the list of level 2 cache devices, if specified.
3803 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3804 &l2cache, &nl2cache) == 0) {
3805 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3806 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3807 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3808 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3809 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3810 spa_load_l2cache(spa);
3811 spa_config_exit(spa, SCL_ALL, FTAG);
3812 spa->spa_l2cache.sav_sync = B_TRUE;
3815 spa->spa_is_initializing = B_TRUE;
3816 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3817 spa->spa_meta_objset = dp->dp_meta_objset;
3818 spa->spa_is_initializing = B_FALSE;
3821 * Create DDTs (dedup tables).
3825 spa_update_dspace(spa);
3827 tx = dmu_tx_create_assigned(dp, txg);
3830 * Create the pool config object.
3832 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3833 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3834 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3836 if (zap_add(spa->spa_meta_objset,
3837 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3838 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3839 cmn_err(CE_PANIC, "failed to add pool config");
3842 if (spa_version(spa) >= SPA_VERSION_FEATURES)
3843 spa_feature_create_zap_objects(spa, tx);
3845 if (zap_add(spa->spa_meta_objset,
3846 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3847 sizeof (uint64_t), 1, &version, tx) != 0) {
3848 cmn_err(CE_PANIC, "failed to add pool version");
3851 /* Newly created pools with the right version are always deflated. */
3852 if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3853 spa->spa_deflate = TRUE;
3854 if (zap_add(spa->spa_meta_objset,
3855 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3856 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3857 cmn_err(CE_PANIC, "failed to add deflate");
3862 * Create the deferred-free bpobj. Turn off compression
3863 * because sync-to-convergence takes longer if the blocksize
3866 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3867 dmu_object_set_compress(spa->spa_meta_objset, obj,
3868 ZIO_COMPRESS_OFF, tx);
3869 if (zap_add(spa->spa_meta_objset,
3870 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3871 sizeof (uint64_t), 1, &obj, tx) != 0) {
3872 cmn_err(CE_PANIC, "failed to add bpobj");
3874 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3875 spa->spa_meta_objset, obj));
3878 * Create the pool's history object.
3880 if (version >= SPA_VERSION_ZPOOL_HISTORY)
3881 spa_history_create_obj(spa, tx);
3884 * Generate some random noise for salted checksums to operate on.
3886 (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes,
3887 sizeof (spa->spa_cksum_salt.zcs_bytes));
3890 * Set pool properties.
3892 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3893 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3894 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3895 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3897 if (props != NULL) {
3898 spa_configfile_set(spa, props, B_FALSE);
3899 spa_sync_props(props, tx);
3904 spa->spa_sync_on = B_TRUE;
3905 txg_sync_start(spa->spa_dsl_pool);
3908 * We explicitly wait for the first transaction to complete so that our
3909 * bean counters are appropriately updated.
3911 txg_wait_synced(spa->spa_dsl_pool, txg);
3913 spa_config_sync(spa, B_FALSE, B_TRUE);
3914 spa_event_notify(spa, NULL, ESC_ZFS_POOL_CREATE);
3916 spa_history_log_version(spa, "create");
3919 * Don't count references from objsets that are already closed
3920 * and are making their way through the eviction process.
3922 spa_evicting_os_wait(spa);
3923 spa->spa_minref = refcount_count(&spa->spa_refcount);
3924 spa->spa_load_state = SPA_LOAD_NONE;
3926 mutex_exit(&spa_namespace_lock);
3934 * Get the root pool information from the root disk, then import the root pool
3935 * during the system boot up time.
3937 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3940 spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3943 nvlist_t *nvtop, *nvroot;
3946 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3950 * Add this top-level vdev to the child array.
3952 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3954 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3956 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3959 * Put this pool's top-level vdevs into a root vdev.
3961 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3962 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3963 VDEV_TYPE_ROOT) == 0);
3964 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3965 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3966 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3970 * Replace the existing vdev_tree with the new root vdev in
3971 * this pool's configuration (remove the old, add the new).
3973 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3974 nvlist_free(nvroot);
3979 * Walk the vdev tree and see if we can find a device with "better"
3980 * configuration. A configuration is "better" if the label on that
3981 * device has a more recent txg.
3984 spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3986 for (int c = 0; c < vd->vdev_children; c++)
3987 spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3989 if (vd->vdev_ops->vdev_op_leaf) {
3993 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3997 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
4001 * Do we have a better boot device?
4003 if (label_txg > *txg) {
4012 * Import a root pool.
4014 * For x86. devpath_list will consist of devid and/or physpath name of
4015 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
4016 * The GRUB "findroot" command will return the vdev we should boot.
4018 * For Sparc, devpath_list consists the physpath name of the booting device
4019 * no matter the rootpool is a single device pool or a mirrored pool.
4021 * "/pci@1f,0/ide@d/disk@0,0:a"
4024 spa_import_rootpool(char *devpath, char *devid)
4027 vdev_t *rvd, *bvd, *avd = NULL;
4028 nvlist_t *config, *nvtop;
4034 * Read the label from the boot device and generate a configuration.
4036 config = spa_generate_rootconf(devpath, devid, &guid);
4037 #if defined(_OBP) && defined(_KERNEL)
4038 if (config == NULL) {
4039 if (strstr(devpath, "/iscsi/ssd") != NULL) {
4041 get_iscsi_bootpath_phy(devpath);
4042 config = spa_generate_rootconf(devpath, devid, &guid);
4046 if (config == NULL) {
4047 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
4049 return (SET_ERROR(EIO));
4052 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
4054 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
4056 mutex_enter(&spa_namespace_lock);
4057 if ((spa = spa_lookup(pname)) != NULL) {
4059 * Remove the existing root pool from the namespace so that we
4060 * can replace it with the correct config we just read in.
4065 spa = spa_add(pname, config, NULL);
4066 spa->spa_is_root = B_TRUE;
4067 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
4070 * Build up a vdev tree based on the boot device's label config.
4072 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4074 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4075 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
4076 VDEV_ALLOC_ROOTPOOL);
4077 spa_config_exit(spa, SCL_ALL, FTAG);
4079 mutex_exit(&spa_namespace_lock);
4080 nvlist_free(config);
4081 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
4087 * Get the boot vdev.
4089 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
4090 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
4091 (u_longlong_t)guid);
4092 error = SET_ERROR(ENOENT);
4097 * Determine if there is a better boot device.
4100 spa_alt_rootvdev(rvd, &avd, &txg);
4102 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
4103 "try booting from '%s'", avd->vdev_path);
4104 error = SET_ERROR(EINVAL);
4109 * If the boot device is part of a spare vdev then ensure that
4110 * we're booting off the active spare.
4112 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
4113 !bvd->vdev_isspare) {
4114 cmn_err(CE_NOTE, "The boot device is currently spared. Please "
4115 "try booting from '%s'",
4117 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
4118 error = SET_ERROR(EINVAL);
4124 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4126 spa_config_exit(spa, SCL_ALL, FTAG);
4127 mutex_exit(&spa_namespace_lock);
4129 nvlist_free(config);
4133 #else /* !illumos */
4135 extern int vdev_geom_read_pool_label(const char *name, nvlist_t ***configs,
4139 spa_generate_rootconf(const char *name)
4141 nvlist_t **configs, **tops;
4143 nvlist_t *best_cfg, *nvtop, *nvroot;
4152 if (vdev_geom_read_pool_label(name, &configs, &count) != 0)
4155 ASSERT3U(count, !=, 0);
4157 for (i = 0; i < count; i++) {
4160 VERIFY(nvlist_lookup_uint64(configs[i], ZPOOL_CONFIG_POOL_TXG,
4162 if (txg > best_txg) {
4164 best_cfg = configs[i];
4169 nvlist_lookup_uint64(best_cfg, ZPOOL_CONFIG_VDEV_CHILDREN, &nchildren);
4171 nvlist_lookup_uint64_array(best_cfg, ZPOOL_CONFIG_HOLE_ARRAY,
4174 tops = kmem_zalloc(nchildren * sizeof(void *), KM_SLEEP);
4175 for (i = 0; i < nchildren; i++) {
4178 if (configs[i] == NULL)
4180 VERIFY(nvlist_lookup_nvlist(configs[i], ZPOOL_CONFIG_VDEV_TREE,
4182 nvlist_dup(nvtop, &tops[i], KM_SLEEP);
4184 for (i = 0; holes != NULL && i < nholes; i++) {
4187 if (tops[holes[i]] != NULL)
4189 nvlist_alloc(&tops[holes[i]], NV_UNIQUE_NAME, KM_SLEEP);
4190 VERIFY(nvlist_add_string(tops[holes[i]], ZPOOL_CONFIG_TYPE,
4191 VDEV_TYPE_HOLE) == 0);
4192 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_ID,
4194 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_GUID,
4197 for (i = 0; i < nchildren; i++) {
4198 if (tops[i] != NULL)
4200 nvlist_alloc(&tops[i], NV_UNIQUE_NAME, KM_SLEEP);
4201 VERIFY(nvlist_add_string(tops[i], ZPOOL_CONFIG_TYPE,
4202 VDEV_TYPE_MISSING) == 0);
4203 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_ID,
4205 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_GUID,
4210 * Create pool config based on the best vdev config.
4212 nvlist_dup(best_cfg, &config, KM_SLEEP);
4215 * Put this pool's top-level vdevs into a root vdev.
4217 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4219 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4220 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
4221 VDEV_TYPE_ROOT) == 0);
4222 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
4223 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
4224 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
4225 tops, nchildren) == 0);
4228 * Replace the existing vdev_tree with the new root vdev in
4229 * this pool's configuration (remove the old, add the new).
4231 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
4234 * Drop vdev config elements that should not be present at pool level.
4236 nvlist_remove(config, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64);
4237 nvlist_remove(config, ZPOOL_CONFIG_TOP_GUID, DATA_TYPE_UINT64);
4239 for (i = 0; i < count; i++)
4240 nvlist_free(configs[i]);
4241 kmem_free(configs, count * sizeof(void *));
4242 for (i = 0; i < nchildren; i++)
4243 nvlist_free(tops[i]);
4244 kmem_free(tops, nchildren * sizeof(void *));
4245 nvlist_free(nvroot);
4250 spa_import_rootpool(const char *name)
4253 vdev_t *rvd, *bvd, *avd = NULL;
4254 nvlist_t *config, *nvtop;
4260 * Read the label from the boot device and generate a configuration.
4262 config = spa_generate_rootconf(name);
4264 mutex_enter(&spa_namespace_lock);
4265 if (config != NULL) {
4266 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
4267 &pname) == 0 && strcmp(name, pname) == 0);
4268 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg)
4271 if ((spa = spa_lookup(pname)) != NULL) {
4273 * Remove the existing root pool from the namespace so
4274 * that we can replace it with the correct config
4279 spa = spa_add(pname, config, NULL);
4282 * Set spa_ubsync.ub_version as it can be used in vdev_alloc()
4283 * via spa_version().
4285 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
4286 &spa->spa_ubsync.ub_version) != 0)
4287 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
4288 } else if ((spa = spa_lookup(name)) == NULL) {
4289 mutex_exit(&spa_namespace_lock);
4290 nvlist_free(config);
4291 cmn_err(CE_NOTE, "Cannot find the pool label for '%s'",
4295 VERIFY(nvlist_dup(spa->spa_config, &config, KM_SLEEP) == 0);
4297 spa->spa_is_root = B_TRUE;
4298 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
4301 * Build up a vdev tree based on the boot device's label config.
4303 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4305 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4306 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
4307 VDEV_ALLOC_ROOTPOOL);
4308 spa_config_exit(spa, SCL_ALL, FTAG);
4310 mutex_exit(&spa_namespace_lock);
4311 nvlist_free(config);
4312 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
4317 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4319 spa_config_exit(spa, SCL_ALL, FTAG);
4320 mutex_exit(&spa_namespace_lock);
4322 nvlist_free(config);
4326 #endif /* illumos */
4327 #endif /* _KERNEL */
4330 * Import a non-root pool into the system.
4333 spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
4336 char *altroot = NULL;
4337 spa_load_state_t state = SPA_LOAD_IMPORT;
4338 zpool_rewind_policy_t policy;
4339 uint64_t mode = spa_mode_global;
4340 uint64_t readonly = B_FALSE;
4343 nvlist_t **spares, **l2cache;
4344 uint_t nspares, nl2cache;
4347 * If a pool with this name exists, return failure.
4349 mutex_enter(&spa_namespace_lock);
4350 if (spa_lookup(pool) != NULL) {
4351 mutex_exit(&spa_namespace_lock);
4352 return (SET_ERROR(EEXIST));
4356 * Create and initialize the spa structure.
4358 (void) nvlist_lookup_string(props,
4359 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4360 (void) nvlist_lookup_uint64(props,
4361 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
4364 spa = spa_add(pool, config, altroot);
4365 spa->spa_import_flags = flags;
4368 * Verbatim import - Take a pool and insert it into the namespace
4369 * as if it had been loaded at boot.
4371 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
4373 spa_configfile_set(spa, props, B_FALSE);
4375 spa_config_sync(spa, B_FALSE, B_TRUE);
4376 spa_event_notify(spa, NULL, ESC_ZFS_POOL_IMPORT);
4378 mutex_exit(&spa_namespace_lock);
4382 spa_activate(spa, mode);
4385 * Don't start async tasks until we know everything is healthy.
4387 spa_async_suspend(spa);
4389 zpool_get_rewind_policy(config, &policy);
4390 if (policy.zrp_request & ZPOOL_DO_REWIND)
4391 state = SPA_LOAD_RECOVER;
4394 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
4395 * because the user-supplied config is actually the one to trust when
4398 if (state != SPA_LOAD_RECOVER)
4399 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
4401 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
4402 policy.zrp_request);
4405 * Propagate anything learned while loading the pool and pass it
4406 * back to caller (i.e. rewind info, missing devices, etc).
4408 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4409 spa->spa_load_info) == 0);
4411 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4413 * Toss any existing sparelist, as it doesn't have any validity
4414 * anymore, and conflicts with spa_has_spare().
4416 if (spa->spa_spares.sav_config) {
4417 nvlist_free(spa->spa_spares.sav_config);
4418 spa->spa_spares.sav_config = NULL;
4419 spa_load_spares(spa);
4421 if (spa->spa_l2cache.sav_config) {
4422 nvlist_free(spa->spa_l2cache.sav_config);
4423 spa->spa_l2cache.sav_config = NULL;
4424 spa_load_l2cache(spa);
4427 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4430 error = spa_validate_aux(spa, nvroot, -1ULL,
4433 error = spa_validate_aux(spa, nvroot, -1ULL,
4434 VDEV_ALLOC_L2CACHE);
4435 spa_config_exit(spa, SCL_ALL, FTAG);
4438 spa_configfile_set(spa, props, B_FALSE);
4440 if (error != 0 || (props && spa_writeable(spa) &&
4441 (error = spa_prop_set(spa, props)))) {
4443 spa_deactivate(spa);
4445 mutex_exit(&spa_namespace_lock);
4449 spa_async_resume(spa);
4452 * Override any spares and level 2 cache devices as specified by
4453 * the user, as these may have correct device names/devids, etc.
4455 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
4456 &spares, &nspares) == 0) {
4457 if (spa->spa_spares.sav_config)
4458 VERIFY(nvlist_remove(spa->spa_spares.sav_config,
4459 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
4461 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
4462 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4463 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
4464 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
4465 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4466 spa_load_spares(spa);
4467 spa_config_exit(spa, SCL_ALL, FTAG);
4468 spa->spa_spares.sav_sync = B_TRUE;
4470 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
4471 &l2cache, &nl2cache) == 0) {
4472 if (spa->spa_l2cache.sav_config)
4473 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
4474 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
4476 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
4477 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4478 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
4479 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
4480 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4481 spa_load_l2cache(spa);
4482 spa_config_exit(spa, SCL_ALL, FTAG);
4483 spa->spa_l2cache.sav_sync = B_TRUE;
4487 * Check for any removed devices.
4489 if (spa->spa_autoreplace) {
4490 spa_aux_check_removed(&spa->spa_spares);
4491 spa_aux_check_removed(&spa->spa_l2cache);
4494 if (spa_writeable(spa)) {
4496 * Update the config cache to include the newly-imported pool.
4498 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4502 * It's possible that the pool was expanded while it was exported.
4503 * We kick off an async task to handle this for us.
4505 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
4507 spa_history_log_version(spa, "import");
4509 spa_event_notify(spa, NULL, ESC_ZFS_POOL_IMPORT);
4511 mutex_exit(&spa_namespace_lock);
4515 zvol_create_minors(pool);
4522 spa_tryimport(nvlist_t *tryconfig)
4524 nvlist_t *config = NULL;
4530 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
4533 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
4537 * Create and initialize the spa structure.
4539 mutex_enter(&spa_namespace_lock);
4540 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
4541 spa_activate(spa, FREAD);
4544 * Pass off the heavy lifting to spa_load().
4545 * Pass TRUE for mosconfig because the user-supplied config
4546 * is actually the one to trust when doing an import.
4548 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
4551 * If 'tryconfig' was at least parsable, return the current config.
4553 if (spa->spa_root_vdev != NULL) {
4554 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4555 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
4557 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4559 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
4560 spa->spa_uberblock.ub_timestamp) == 0);
4561 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4562 spa->spa_load_info) == 0);
4565 * If the bootfs property exists on this pool then we
4566 * copy it out so that external consumers can tell which
4567 * pools are bootable.
4569 if ((!error || error == EEXIST) && spa->spa_bootfs) {
4570 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4573 * We have to play games with the name since the
4574 * pool was opened as TRYIMPORT_NAME.
4576 if (dsl_dsobj_to_dsname(spa_name(spa),
4577 spa->spa_bootfs, tmpname) == 0) {
4579 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4581 cp = strchr(tmpname, '/');
4583 (void) strlcpy(dsname, tmpname,
4586 (void) snprintf(dsname, MAXPATHLEN,
4587 "%s/%s", poolname, ++cp);
4589 VERIFY(nvlist_add_string(config,
4590 ZPOOL_CONFIG_BOOTFS, dsname) == 0);
4591 kmem_free(dsname, MAXPATHLEN);
4593 kmem_free(tmpname, MAXPATHLEN);
4597 * Add the list of hot spares and level 2 cache devices.
4599 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4600 spa_add_spares(spa, config);
4601 spa_add_l2cache(spa, config);
4602 spa_config_exit(spa, SCL_CONFIG, FTAG);
4606 spa_deactivate(spa);
4608 mutex_exit(&spa_namespace_lock);
4614 * Pool export/destroy
4616 * The act of destroying or exporting a pool is very simple. We make sure there
4617 * is no more pending I/O and any references to the pool are gone. Then, we
4618 * update the pool state and sync all the labels to disk, removing the
4619 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4620 * we don't sync the labels or remove the configuration cache.
4623 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
4624 boolean_t force, boolean_t hardforce)
4631 if (!(spa_mode_global & FWRITE))
4632 return (SET_ERROR(EROFS));
4634 mutex_enter(&spa_namespace_lock);
4635 if ((spa = spa_lookup(pool)) == NULL) {
4636 mutex_exit(&spa_namespace_lock);
4637 return (SET_ERROR(ENOENT));
4641 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4642 * reacquire the namespace lock, and see if we can export.
4644 spa_open_ref(spa, FTAG);
4645 mutex_exit(&spa_namespace_lock);
4646 spa_async_suspend(spa);
4647 mutex_enter(&spa_namespace_lock);
4648 spa_close(spa, FTAG);
4651 * The pool will be in core if it's openable,
4652 * in which case we can modify its state.
4654 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
4656 * Objsets may be open only because they're dirty, so we
4657 * have to force it to sync before checking spa_refcnt.
4659 txg_wait_synced(spa->spa_dsl_pool, 0);
4660 spa_evicting_os_wait(spa);
4663 * A pool cannot be exported or destroyed if there are active
4664 * references. If we are resetting a pool, allow references by
4665 * fault injection handlers.
4667 if (!spa_refcount_zero(spa) ||
4668 (spa->spa_inject_ref != 0 &&
4669 new_state != POOL_STATE_UNINITIALIZED)) {
4670 spa_async_resume(spa);
4671 mutex_exit(&spa_namespace_lock);
4672 return (SET_ERROR(EBUSY));
4676 * A pool cannot be exported if it has an active shared spare.
4677 * This is to prevent other pools stealing the active spare
4678 * from an exported pool. At user's own will, such pool can
4679 * be forcedly exported.
4681 if (!force && new_state == POOL_STATE_EXPORTED &&
4682 spa_has_active_shared_spare(spa)) {
4683 spa_async_resume(spa);
4684 mutex_exit(&spa_namespace_lock);
4685 return (SET_ERROR(EXDEV));
4689 * We want this to be reflected on every label,
4690 * so mark them all dirty. spa_unload() will do the
4691 * final sync that pushes these changes out.
4693 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
4694 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4695 spa->spa_state = new_state;
4696 spa->spa_final_txg = spa_last_synced_txg(spa) +
4698 vdev_config_dirty(spa->spa_root_vdev);
4699 spa_config_exit(spa, SCL_ALL, FTAG);
4703 spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
4705 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4707 spa_deactivate(spa);
4710 if (oldconfig && spa->spa_config)
4711 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
4713 if (new_state != POOL_STATE_UNINITIALIZED) {
4715 spa_config_sync(spa, B_TRUE, B_TRUE);
4718 mutex_exit(&spa_namespace_lock);
4724 * Destroy a storage pool.
4727 spa_destroy(char *pool)
4729 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
4734 * Export a storage pool.
4737 spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
4738 boolean_t hardforce)
4740 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
4745 * Similar to spa_export(), this unloads the spa_t without actually removing it
4746 * from the namespace in any way.
4749 spa_reset(char *pool)
4751 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
4756 * ==========================================================================
4757 * Device manipulation
4758 * ==========================================================================
4762 * Add a device to a storage pool.
4765 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
4769 vdev_t *rvd = spa->spa_root_vdev;
4771 nvlist_t **spares, **l2cache;
4772 uint_t nspares, nl2cache;
4774 ASSERT(spa_writeable(spa));
4776 txg = spa_vdev_enter(spa);
4778 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
4779 VDEV_ALLOC_ADD)) != 0)
4780 return (spa_vdev_exit(spa, NULL, txg, error));
4782 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */
4784 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
4788 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
4792 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
4793 return (spa_vdev_exit(spa, vd, txg, EINVAL));
4795 if (vd->vdev_children != 0 &&
4796 (error = vdev_create(vd, txg, B_FALSE)) != 0)
4797 return (spa_vdev_exit(spa, vd, txg, error));
4800 * We must validate the spares and l2cache devices after checking the
4801 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
4803 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
4804 return (spa_vdev_exit(spa, vd, txg, error));
4807 * Transfer each new top-level vdev from vd to rvd.
4809 for (int c = 0; c < vd->vdev_children; c++) {
4812 * Set the vdev id to the first hole, if one exists.
4814 for (id = 0; id < rvd->vdev_children; id++) {
4815 if (rvd->vdev_child[id]->vdev_ishole) {
4816 vdev_free(rvd->vdev_child[id]);
4820 tvd = vd->vdev_child[c];
4821 vdev_remove_child(vd, tvd);
4823 vdev_add_child(rvd, tvd);
4824 vdev_config_dirty(tvd);
4828 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
4829 ZPOOL_CONFIG_SPARES);
4830 spa_load_spares(spa);
4831 spa->spa_spares.sav_sync = B_TRUE;
4834 if (nl2cache != 0) {
4835 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
4836 ZPOOL_CONFIG_L2CACHE);
4837 spa_load_l2cache(spa);
4838 spa->spa_l2cache.sav_sync = B_TRUE;
4842 * We have to be careful when adding new vdevs to an existing pool.
4843 * If other threads start allocating from these vdevs before we
4844 * sync the config cache, and we lose power, then upon reboot we may
4845 * fail to open the pool because there are DVAs that the config cache
4846 * can't translate. Therefore, we first add the vdevs without
4847 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4848 * and then let spa_config_update() initialize the new metaslabs.
4850 * spa_load() checks for added-but-not-initialized vdevs, so that
4851 * if we lose power at any point in this sequence, the remaining
4852 * steps will be completed the next time we load the pool.
4854 (void) spa_vdev_exit(spa, vd, txg, 0);
4856 mutex_enter(&spa_namespace_lock);
4857 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4858 spa_event_notify(spa, NULL, ESC_ZFS_VDEV_ADD);
4859 mutex_exit(&spa_namespace_lock);
4865 * Attach a device to a mirror. The arguments are the path to any device
4866 * in the mirror, and the nvroot for the new device. If the path specifies
4867 * a device that is not mirrored, we automatically insert the mirror vdev.
4869 * If 'replacing' is specified, the new device is intended to replace the
4870 * existing device; in this case the two devices are made into their own
4871 * mirror using the 'replacing' vdev, which is functionally identical to
4872 * the mirror vdev (it actually reuses all the same ops) but has a few
4873 * extra rules: you can't attach to it after it's been created, and upon
4874 * completion of resilvering, the first disk (the one being replaced)
4875 * is automatically detached.
4878 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
4880 uint64_t txg, dtl_max_txg;
4881 vdev_t *rvd = spa->spa_root_vdev;
4882 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
4884 char *oldvdpath, *newvdpath;
4888 ASSERT(spa_writeable(spa));
4890 txg = spa_vdev_enter(spa);
4892 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
4895 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4897 if (!oldvd->vdev_ops->vdev_op_leaf)
4898 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4900 pvd = oldvd->vdev_parent;
4902 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
4903 VDEV_ALLOC_ATTACH)) != 0)
4904 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4906 if (newrootvd->vdev_children != 1)
4907 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4909 newvd = newrootvd->vdev_child[0];
4911 if (!newvd->vdev_ops->vdev_op_leaf)
4912 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4914 if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
4915 return (spa_vdev_exit(spa, newrootvd, txg, error));
4918 * Spares can't replace logs
4920 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
4921 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4925 * For attach, the only allowable parent is a mirror or the root
4928 if (pvd->vdev_ops != &vdev_mirror_ops &&
4929 pvd->vdev_ops != &vdev_root_ops)
4930 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4932 pvops = &vdev_mirror_ops;
4935 * Active hot spares can only be replaced by inactive hot
4938 if (pvd->vdev_ops == &vdev_spare_ops &&
4939 oldvd->vdev_isspare &&
4940 !spa_has_spare(spa, newvd->vdev_guid))
4941 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4944 * If the source is a hot spare, and the parent isn't already a
4945 * spare, then we want to create a new hot spare. Otherwise, we
4946 * want to create a replacing vdev. The user is not allowed to
4947 * attach to a spared vdev child unless the 'isspare' state is
4948 * the same (spare replaces spare, non-spare replaces
4951 if (pvd->vdev_ops == &vdev_replacing_ops &&
4952 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
4953 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4954 } else if (pvd->vdev_ops == &vdev_spare_ops &&
4955 newvd->vdev_isspare != oldvd->vdev_isspare) {
4956 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4959 if (newvd->vdev_isspare)
4960 pvops = &vdev_spare_ops;
4962 pvops = &vdev_replacing_ops;
4966 * Make sure the new device is big enough.
4968 if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
4969 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
4972 * The new device cannot have a higher alignment requirement
4973 * than the top-level vdev.
4975 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4976 return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4979 * If this is an in-place replacement, update oldvd's path and devid
4980 * to make it distinguishable from newvd, and unopenable from now on.
4982 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4983 spa_strfree(oldvd->vdev_path);
4984 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4986 (void) sprintf(oldvd->vdev_path, "%s/%s",
4987 newvd->vdev_path, "old");
4988 if (oldvd->vdev_devid != NULL) {
4989 spa_strfree(oldvd->vdev_devid);
4990 oldvd->vdev_devid = NULL;
4994 /* mark the device being resilvered */
4995 newvd->vdev_resilver_txg = txg;
4998 * If the parent is not a mirror, or if we're replacing, insert the new
4999 * mirror/replacing/spare vdev above oldvd.
5001 if (pvd->vdev_ops != pvops)
5002 pvd = vdev_add_parent(oldvd, pvops);
5004 ASSERT(pvd->vdev_top->vdev_parent == rvd);
5005 ASSERT(pvd->vdev_ops == pvops);
5006 ASSERT(oldvd->vdev_parent == pvd);
5009 * Extract the new device from its root and add it to pvd.
5011 vdev_remove_child(newrootvd, newvd);
5012 newvd->vdev_id = pvd->vdev_children;
5013 newvd->vdev_crtxg = oldvd->vdev_crtxg;
5014 vdev_add_child(pvd, newvd);
5016 tvd = newvd->vdev_top;
5017 ASSERT(pvd->vdev_top == tvd);
5018 ASSERT(tvd->vdev_parent == rvd);
5020 vdev_config_dirty(tvd);
5023 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
5024 * for any dmu_sync-ed blocks. It will propagate upward when
5025 * spa_vdev_exit() calls vdev_dtl_reassess().
5027 dtl_max_txg = txg + TXG_CONCURRENT_STATES;
5029 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
5030 dtl_max_txg - TXG_INITIAL);
5032 if (newvd->vdev_isspare) {
5033 spa_spare_activate(newvd);
5034 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
5037 oldvdpath = spa_strdup(oldvd->vdev_path);
5038 newvdpath = spa_strdup(newvd->vdev_path);
5039 newvd_isspare = newvd->vdev_isspare;
5042 * Mark newvd's DTL dirty in this txg.
5044 vdev_dirty(tvd, VDD_DTL, newvd, txg);
5047 * Schedule the resilver to restart in the future. We do this to
5048 * ensure that dmu_sync-ed blocks have been stitched into the
5049 * respective datasets.
5051 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
5053 if (spa->spa_bootfs)
5054 spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
5056 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_ATTACH);
5061 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
5063 spa_history_log_internal(spa, "vdev attach", NULL,
5064 "%s vdev=%s %s vdev=%s",
5065 replacing && newvd_isspare ? "spare in" :
5066 replacing ? "replace" : "attach", newvdpath,
5067 replacing ? "for" : "to", oldvdpath);
5069 spa_strfree(oldvdpath);
5070 spa_strfree(newvdpath);
5076 * Detach a device from a mirror or replacing vdev.
5078 * If 'replace_done' is specified, only detach if the parent
5079 * is a replacing vdev.
5082 spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
5086 vdev_t *rvd = spa->spa_root_vdev;
5087 vdev_t *vd, *pvd, *cvd, *tvd;
5088 boolean_t unspare = B_FALSE;
5089 uint64_t unspare_guid = 0;
5092 ASSERT(spa_writeable(spa));
5094 txg = spa_vdev_enter(spa);
5096 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5099 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
5101 if (!vd->vdev_ops->vdev_op_leaf)
5102 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
5104 pvd = vd->vdev_parent;
5107 * If the parent/child relationship is not as expected, don't do it.
5108 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
5109 * vdev that's replacing B with C. The user's intent in replacing
5110 * is to go from M(A,B) to M(A,C). If the user decides to cancel
5111 * the replace by detaching C, the expected behavior is to end up
5112 * M(A,B). But suppose that right after deciding to detach C,
5113 * the replacement of B completes. We would have M(A,C), and then
5114 * ask to detach C, which would leave us with just A -- not what
5115 * the user wanted. To prevent this, we make sure that the
5116 * parent/child relationship hasn't changed -- in this example,
5117 * that C's parent is still the replacing vdev R.
5119 if (pvd->vdev_guid != pguid && pguid != 0)
5120 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
5123 * Only 'replacing' or 'spare' vdevs can be replaced.
5125 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
5126 pvd->vdev_ops != &vdev_spare_ops)
5127 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
5129 ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
5130 spa_version(spa) >= SPA_VERSION_SPARES);
5133 * Only mirror, replacing, and spare vdevs support detach.
5135 if (pvd->vdev_ops != &vdev_replacing_ops &&
5136 pvd->vdev_ops != &vdev_mirror_ops &&
5137 pvd->vdev_ops != &vdev_spare_ops)
5138 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
5141 * If this device has the only valid copy of some data,
5142 * we cannot safely detach it.
5144 if (vdev_dtl_required(vd))
5145 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
5147 ASSERT(pvd->vdev_children >= 2);
5150 * If we are detaching the second disk from a replacing vdev, then
5151 * check to see if we changed the original vdev's path to have "/old"
5152 * at the end in spa_vdev_attach(). If so, undo that change now.
5154 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
5155 vd->vdev_path != NULL) {
5156 size_t len = strlen(vd->vdev_path);
5158 for (int c = 0; c < pvd->vdev_children; c++) {
5159 cvd = pvd->vdev_child[c];
5161 if (cvd == vd || cvd->vdev_path == NULL)
5164 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
5165 strcmp(cvd->vdev_path + len, "/old") == 0) {
5166 spa_strfree(cvd->vdev_path);
5167 cvd->vdev_path = spa_strdup(vd->vdev_path);
5174 * If we are detaching the original disk from a spare, then it implies
5175 * that the spare should become a real disk, and be removed from the
5176 * active spare list for the pool.
5178 if (pvd->vdev_ops == &vdev_spare_ops &&
5180 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
5184 * Erase the disk labels so the disk can be used for other things.
5185 * This must be done after all other error cases are handled,
5186 * but before we disembowel vd (so we can still do I/O to it).
5187 * But if we can't do it, don't treat the error as fatal --
5188 * it may be that the unwritability of the disk is the reason
5189 * it's being detached!
5191 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5194 * Remove vd from its parent and compact the parent's children.
5196 vdev_remove_child(pvd, vd);
5197 vdev_compact_children(pvd);
5200 * Remember one of the remaining children so we can get tvd below.
5202 cvd = pvd->vdev_child[pvd->vdev_children - 1];
5205 * If we need to remove the remaining child from the list of hot spares,
5206 * do it now, marking the vdev as no longer a spare in the process.
5207 * We must do this before vdev_remove_parent(), because that can
5208 * change the GUID if it creates a new toplevel GUID. For a similar
5209 * reason, we must remove the spare now, in the same txg as the detach;
5210 * otherwise someone could attach a new sibling, change the GUID, and
5211 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
5214 ASSERT(cvd->vdev_isspare);
5215 spa_spare_remove(cvd);
5216 unspare_guid = cvd->vdev_guid;
5217 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
5218 cvd->vdev_unspare = B_TRUE;
5222 * If the parent mirror/replacing vdev only has one child,
5223 * the parent is no longer needed. Remove it from the tree.
5225 if (pvd->vdev_children == 1) {
5226 if (pvd->vdev_ops == &vdev_spare_ops)
5227 cvd->vdev_unspare = B_FALSE;
5228 vdev_remove_parent(cvd);
5233 * We don't set tvd until now because the parent we just removed
5234 * may have been the previous top-level vdev.
5236 tvd = cvd->vdev_top;
5237 ASSERT(tvd->vdev_parent == rvd);
5240 * Reevaluate the parent vdev state.
5242 vdev_propagate_state(cvd);
5245 * If the 'autoexpand' property is set on the pool then automatically
5246 * try to expand the size of the pool. For example if the device we
5247 * just detached was smaller than the others, it may be possible to
5248 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
5249 * first so that we can obtain the updated sizes of the leaf vdevs.
5251 if (spa->spa_autoexpand) {
5253 vdev_expand(tvd, txg);
5256 vdev_config_dirty(tvd);
5259 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
5260 * vd->vdev_detached is set and free vd's DTL object in syncing context.
5261 * But first make sure we're not on any *other* txg's DTL list, to
5262 * prevent vd from being accessed after it's freed.
5264 vdpath = spa_strdup(vd->vdev_path);
5265 for (int t = 0; t < TXG_SIZE; t++)
5266 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
5267 vd->vdev_detached = B_TRUE;
5268 vdev_dirty(tvd, VDD_DTL, vd, txg);
5270 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
5272 /* hang on to the spa before we release the lock */
5273 spa_open_ref(spa, FTAG);
5275 error = spa_vdev_exit(spa, vd, txg, 0);
5277 spa_history_log_internal(spa, "detach", NULL,
5279 spa_strfree(vdpath);
5282 * If this was the removal of the original device in a hot spare vdev,
5283 * then we want to go through and remove the device from the hot spare
5284 * list of every other pool.
5287 spa_t *altspa = NULL;
5289 mutex_enter(&spa_namespace_lock);
5290 while ((altspa = spa_next(altspa)) != NULL) {
5291 if (altspa->spa_state != POOL_STATE_ACTIVE ||
5295 spa_open_ref(altspa, FTAG);
5296 mutex_exit(&spa_namespace_lock);
5297 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
5298 mutex_enter(&spa_namespace_lock);
5299 spa_close(altspa, FTAG);
5301 mutex_exit(&spa_namespace_lock);
5303 /* search the rest of the vdevs for spares to remove */
5304 spa_vdev_resilver_done(spa);
5307 /* all done with the spa; OK to release */
5308 mutex_enter(&spa_namespace_lock);
5309 spa_close(spa, FTAG);
5310 mutex_exit(&spa_namespace_lock);
5316 * Split a set of devices from their mirrors, and create a new pool from them.
5319 spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
5320 nvlist_t *props, boolean_t exp)
5323 uint64_t txg, *glist;
5325 uint_t c, children, lastlog;
5326 nvlist_t **child, *nvl, *tmp;
5328 char *altroot = NULL;
5329 vdev_t *rvd, **vml = NULL; /* vdev modify list */
5330 boolean_t activate_slog;
5332 ASSERT(spa_writeable(spa));
5334 txg = spa_vdev_enter(spa);
5336 /* clear the log and flush everything up to now */
5337 activate_slog = spa_passivate_log(spa);
5338 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5339 error = spa_offline_log(spa);
5340 txg = spa_vdev_config_enter(spa);
5343 spa_activate_log(spa);
5346 return (spa_vdev_exit(spa, NULL, txg, error));
5348 /* check new spa name before going any further */
5349 if (spa_lookup(newname) != NULL)
5350 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
5353 * scan through all the children to ensure they're all mirrors
5355 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
5356 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
5358 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5360 /* first, check to ensure we've got the right child count */
5361 rvd = spa->spa_root_vdev;
5363 for (c = 0; c < rvd->vdev_children; c++) {
5364 vdev_t *vd = rvd->vdev_child[c];
5366 /* don't count the holes & logs as children */
5367 if (vd->vdev_islog || vd->vdev_ishole) {
5375 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
5376 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5378 /* next, ensure no spare or cache devices are part of the split */
5379 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
5380 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
5381 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5383 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
5384 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
5386 /* then, loop over each vdev and validate it */
5387 for (c = 0; c < children; c++) {
5388 uint64_t is_hole = 0;
5390 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
5394 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
5395 spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
5398 error = SET_ERROR(EINVAL);
5403 /* which disk is going to be split? */
5404 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
5406 error = SET_ERROR(EINVAL);
5410 /* look it up in the spa */
5411 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
5412 if (vml[c] == NULL) {
5413 error = SET_ERROR(ENODEV);
5417 /* make sure there's nothing stopping the split */
5418 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
5419 vml[c]->vdev_islog ||
5420 vml[c]->vdev_ishole ||
5421 vml[c]->vdev_isspare ||
5422 vml[c]->vdev_isl2cache ||
5423 !vdev_writeable(vml[c]) ||
5424 vml[c]->vdev_children != 0 ||
5425 vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
5426 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
5427 error = SET_ERROR(EINVAL);
5431 if (vdev_dtl_required(vml[c])) {
5432 error = SET_ERROR(EBUSY);
5436 /* we need certain info from the top level */
5437 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
5438 vml[c]->vdev_top->vdev_ms_array) == 0);
5439 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
5440 vml[c]->vdev_top->vdev_ms_shift) == 0);
5441 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
5442 vml[c]->vdev_top->vdev_asize) == 0);
5443 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
5444 vml[c]->vdev_top->vdev_ashift) == 0);
5446 /* transfer per-vdev ZAPs */
5447 ASSERT3U(vml[c]->vdev_leaf_zap, !=, 0);
5448 VERIFY0(nvlist_add_uint64(child[c],
5449 ZPOOL_CONFIG_VDEV_LEAF_ZAP, vml[c]->vdev_leaf_zap));
5451 ASSERT3U(vml[c]->vdev_top->vdev_top_zap, !=, 0);
5452 VERIFY0(nvlist_add_uint64(child[c],
5453 ZPOOL_CONFIG_VDEV_TOP_ZAP,
5454 vml[c]->vdev_parent->vdev_top_zap));
5458 kmem_free(vml, children * sizeof (vdev_t *));
5459 kmem_free(glist, children * sizeof (uint64_t));
5460 return (spa_vdev_exit(spa, NULL, txg, error));
5463 /* stop writers from using the disks */
5464 for (c = 0; c < children; c++) {
5466 vml[c]->vdev_offline = B_TRUE;
5468 vdev_reopen(spa->spa_root_vdev);
5471 * Temporarily record the splitting vdevs in the spa config. This
5472 * will disappear once the config is regenerated.
5474 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5475 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
5476 glist, children) == 0);
5477 kmem_free(glist, children * sizeof (uint64_t));
5479 mutex_enter(&spa->spa_props_lock);
5480 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
5482 mutex_exit(&spa->spa_props_lock);
5483 spa->spa_config_splitting = nvl;
5484 vdev_config_dirty(spa->spa_root_vdev);
5486 /* configure and create the new pool */
5487 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
5488 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
5489 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
5490 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5491 spa_version(spa)) == 0);
5492 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
5493 spa->spa_config_txg) == 0);
5494 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
5495 spa_generate_guid(NULL)) == 0);
5496 VERIFY0(nvlist_add_boolean(config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS));
5497 (void) nvlist_lookup_string(props,
5498 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
5500 /* add the new pool to the namespace */
5501 newspa = spa_add(newname, config, altroot);
5502 newspa->spa_avz_action = AVZ_ACTION_REBUILD;
5503 newspa->spa_config_txg = spa->spa_config_txg;
5504 spa_set_log_state(newspa, SPA_LOG_CLEAR);
5506 /* release the spa config lock, retaining the namespace lock */
5507 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5509 if (zio_injection_enabled)
5510 zio_handle_panic_injection(spa, FTAG, 1);
5512 spa_activate(newspa, spa_mode_global);
5513 spa_async_suspend(newspa);
5516 /* mark that we are creating new spa by splitting */
5517 newspa->spa_splitting_newspa = B_TRUE;
5519 /* create the new pool from the disks of the original pool */
5520 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
5522 newspa->spa_splitting_newspa = B_FALSE;
5527 /* if that worked, generate a real config for the new pool */
5528 if (newspa->spa_root_vdev != NULL) {
5529 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
5530 NV_UNIQUE_NAME, KM_SLEEP) == 0);
5531 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
5532 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
5533 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
5538 if (props != NULL) {
5539 spa_configfile_set(newspa, props, B_FALSE);
5540 error = spa_prop_set(newspa, props);
5545 /* flush everything */
5546 txg = spa_vdev_config_enter(newspa);
5547 vdev_config_dirty(newspa->spa_root_vdev);
5548 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
5550 if (zio_injection_enabled)
5551 zio_handle_panic_injection(spa, FTAG, 2);
5553 spa_async_resume(newspa);
5555 /* finally, update the original pool's config */
5556 txg = spa_vdev_config_enter(spa);
5557 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
5558 error = dmu_tx_assign(tx, TXG_WAIT);
5561 for (c = 0; c < children; c++) {
5562 if (vml[c] != NULL) {
5565 spa_history_log_internal(spa, "detach", tx,
5566 "vdev=%s", vml[c]->vdev_path);
5571 spa->spa_avz_action = AVZ_ACTION_REBUILD;
5572 vdev_config_dirty(spa->spa_root_vdev);
5573 spa->spa_config_splitting = NULL;
5577 (void) spa_vdev_exit(spa, NULL, txg, 0);
5579 if (zio_injection_enabled)
5580 zio_handle_panic_injection(spa, FTAG, 3);
5582 /* split is complete; log a history record */
5583 spa_history_log_internal(newspa, "split", NULL,
5584 "from pool %s", spa_name(spa));
5586 kmem_free(vml, children * sizeof (vdev_t *));
5588 /* if we're not going to mount the filesystems in userland, export */
5590 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
5597 spa_deactivate(newspa);
5600 txg = spa_vdev_config_enter(spa);
5602 /* re-online all offlined disks */
5603 for (c = 0; c < children; c++) {
5605 vml[c]->vdev_offline = B_FALSE;
5607 vdev_reopen(spa->spa_root_vdev);
5609 nvlist_free(spa->spa_config_splitting);
5610 spa->spa_config_splitting = NULL;
5611 (void) spa_vdev_exit(spa, NULL, txg, error);
5613 kmem_free(vml, children * sizeof (vdev_t *));
5618 spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
5620 for (int i = 0; i < count; i++) {
5623 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
5626 if (guid == target_guid)
5634 spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
5635 nvlist_t *dev_to_remove)
5637 nvlist_t **newdev = NULL;
5640 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
5642 for (int i = 0, j = 0; i < count; i++) {
5643 if (dev[i] == dev_to_remove)
5645 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
5648 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
5649 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
5651 for (int i = 0; i < count - 1; i++)
5652 nvlist_free(newdev[i]);
5655 kmem_free(newdev, (count - 1) * sizeof (void *));
5659 * Evacuate the device.
5662 spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
5667 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5668 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5669 ASSERT(vd == vd->vdev_top);
5672 * Evacuate the device. We don't hold the config lock as writer
5673 * since we need to do I/O but we do keep the
5674 * spa_namespace_lock held. Once this completes the device
5675 * should no longer have any blocks allocated on it.
5677 if (vd->vdev_islog) {
5678 if (vd->vdev_stat.vs_alloc != 0)
5679 error = spa_offline_log(spa);
5681 error = SET_ERROR(ENOTSUP);
5688 * The evacuation succeeded. Remove any remaining MOS metadata
5689 * associated with this vdev, and wait for these changes to sync.
5691 ASSERT0(vd->vdev_stat.vs_alloc);
5692 txg = spa_vdev_config_enter(spa);
5693 vd->vdev_removing = B_TRUE;
5694 vdev_dirty_leaves(vd, VDD_DTL, txg);
5695 vdev_config_dirty(vd);
5696 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5702 * Complete the removal by cleaning up the namespace.
5705 spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
5707 vdev_t *rvd = spa->spa_root_vdev;
5708 uint64_t id = vd->vdev_id;
5709 boolean_t last_vdev = (id == (rvd->vdev_children - 1));
5711 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5712 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5713 ASSERT(vd == vd->vdev_top);
5716 * Only remove any devices which are empty.
5718 if (vd->vdev_stat.vs_alloc != 0)
5721 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5723 if (list_link_active(&vd->vdev_state_dirty_node))
5724 vdev_state_clean(vd);
5725 if (list_link_active(&vd->vdev_config_dirty_node))
5726 vdev_config_clean(vd);
5731 vdev_compact_children(rvd);
5733 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
5734 vdev_add_child(rvd, vd);
5736 vdev_config_dirty(rvd);
5739 * Reassess the health of our root vdev.
5745 * Remove a device from the pool -
5747 * Removing a device from the vdev namespace requires several steps
5748 * and can take a significant amount of time. As a result we use
5749 * the spa_vdev_config_[enter/exit] functions which allow us to
5750 * grab and release the spa_config_lock while still holding the namespace
5751 * lock. During each step the configuration is synced out.
5753 * Currently, this supports removing only hot spares, slogs, and level 2 ARC
5757 spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
5760 sysevent_t *ev = NULL;
5761 metaslab_group_t *mg;
5762 nvlist_t **spares, **l2cache, *nv;
5764 uint_t nspares, nl2cache;
5766 boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
5768 ASSERT(spa_writeable(spa));
5771 txg = spa_vdev_enter(spa);
5773 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5775 if (spa->spa_spares.sav_vdevs != NULL &&
5776 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5777 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
5778 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
5780 * Only remove the hot spare if it's not currently in use
5783 if (vd == NULL || unspare) {
5785 vd = spa_lookup_by_guid(spa, guid, B_TRUE);
5786 ev = spa_event_create(spa, vd, ESC_ZFS_VDEV_REMOVE_AUX);
5787 spa_vdev_remove_aux(spa->spa_spares.sav_config,
5788 ZPOOL_CONFIG_SPARES, spares, nspares, nv);
5789 spa_load_spares(spa);
5790 spa->spa_spares.sav_sync = B_TRUE;
5792 error = SET_ERROR(EBUSY);
5794 } else if (spa->spa_l2cache.sav_vdevs != NULL &&
5795 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5796 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
5797 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
5799 * Cache devices can always be removed.
5801 vd = spa_lookup_by_guid(spa, guid, B_TRUE);
5802 ev = spa_event_create(spa, vd, ESC_ZFS_VDEV_REMOVE_AUX);
5803 spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
5804 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
5805 spa_load_l2cache(spa);
5806 spa->spa_l2cache.sav_sync = B_TRUE;
5807 } else if (vd != NULL && vd->vdev_islog) {
5809 ASSERT(vd == vd->vdev_top);
5814 * Stop allocating from this vdev.
5816 metaslab_group_passivate(mg);
5819 * Wait for the youngest allocations and frees to sync,
5820 * and then wait for the deferral of those frees to finish.
5822 spa_vdev_config_exit(spa, NULL,
5823 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
5826 * Attempt to evacuate the vdev.
5828 error = spa_vdev_remove_evacuate(spa, vd);
5830 txg = spa_vdev_config_enter(spa);
5833 * If we couldn't evacuate the vdev, unwind.
5836 metaslab_group_activate(mg);
5837 return (spa_vdev_exit(spa, NULL, txg, error));
5841 * Clean up the vdev namespace.
5843 ev = spa_event_create(spa, vd, ESC_ZFS_VDEV_REMOVE_DEV);
5844 spa_vdev_remove_from_namespace(spa, vd);
5846 } else if (vd != NULL) {
5848 * Normal vdevs cannot be removed (yet).
5850 error = SET_ERROR(ENOTSUP);
5853 * There is no vdev of any kind with the specified guid.
5855 error = SET_ERROR(ENOENT);
5859 error = spa_vdev_exit(spa, NULL, txg, error);
5868 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5869 * currently spared, so we can detach it.
5872 spa_vdev_resilver_done_hunt(vdev_t *vd)
5874 vdev_t *newvd, *oldvd;
5876 for (int c = 0; c < vd->vdev_children; c++) {
5877 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
5883 * Check for a completed replacement. We always consider the first
5884 * vdev in the list to be the oldest vdev, and the last one to be
5885 * the newest (see spa_vdev_attach() for how that works). In
5886 * the case where the newest vdev is faulted, we will not automatically
5887 * remove it after a resilver completes. This is OK as it will require
5888 * user intervention to determine which disk the admin wishes to keep.
5890 if (vd->vdev_ops == &vdev_replacing_ops) {
5891 ASSERT(vd->vdev_children > 1);
5893 newvd = vd->vdev_child[vd->vdev_children - 1];
5894 oldvd = vd->vdev_child[0];
5896 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
5897 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5898 !vdev_dtl_required(oldvd))
5903 * Check for a completed resilver with the 'unspare' flag set.
5905 if (vd->vdev_ops == &vdev_spare_ops) {
5906 vdev_t *first = vd->vdev_child[0];
5907 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
5909 if (last->vdev_unspare) {
5912 } else if (first->vdev_unspare) {
5919 if (oldvd != NULL &&
5920 vdev_dtl_empty(newvd, DTL_MISSING) &&
5921 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5922 !vdev_dtl_required(oldvd))
5926 * If there are more than two spares attached to a disk,
5927 * and those spares are not required, then we want to
5928 * attempt to free them up now so that they can be used
5929 * by other pools. Once we're back down to a single
5930 * disk+spare, we stop removing them.
5932 if (vd->vdev_children > 2) {
5933 newvd = vd->vdev_child[1];
5935 if (newvd->vdev_isspare && last->vdev_isspare &&
5936 vdev_dtl_empty(last, DTL_MISSING) &&
5937 vdev_dtl_empty(last, DTL_OUTAGE) &&
5938 !vdev_dtl_required(newvd))
5947 spa_vdev_resilver_done(spa_t *spa)
5949 vdev_t *vd, *pvd, *ppvd;
5950 uint64_t guid, sguid, pguid, ppguid;
5952 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5954 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
5955 pvd = vd->vdev_parent;
5956 ppvd = pvd->vdev_parent;
5957 guid = vd->vdev_guid;
5958 pguid = pvd->vdev_guid;
5959 ppguid = ppvd->vdev_guid;
5962 * If we have just finished replacing a hot spared device, then
5963 * we need to detach the parent's first child (the original hot
5966 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
5967 ppvd->vdev_children == 2) {
5968 ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
5969 sguid = ppvd->vdev_child[1]->vdev_guid;
5971 ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd));
5973 spa_config_exit(spa, SCL_ALL, FTAG);
5974 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
5976 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
5978 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5981 spa_config_exit(spa, SCL_ALL, FTAG);
5985 * Update the stored path or FRU for this vdev.
5988 spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
5992 boolean_t sync = B_FALSE;
5994 ASSERT(spa_writeable(spa));
5996 spa_vdev_state_enter(spa, SCL_ALL);
5998 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
5999 return (spa_vdev_state_exit(spa, NULL, ENOENT));
6001 if (!vd->vdev_ops->vdev_op_leaf)
6002 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
6005 if (strcmp(value, vd->vdev_path) != 0) {
6006 spa_strfree(vd->vdev_path);
6007 vd->vdev_path = spa_strdup(value);
6011 if (vd->vdev_fru == NULL) {
6012 vd->vdev_fru = spa_strdup(value);
6014 } else if (strcmp(value, vd->vdev_fru) != 0) {
6015 spa_strfree(vd->vdev_fru);
6016 vd->vdev_fru = spa_strdup(value);
6021 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
6025 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
6027 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
6031 spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
6033 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
6037 * ==========================================================================
6039 * ==========================================================================
6043 spa_scan_stop(spa_t *spa)
6045 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
6046 if (dsl_scan_resilvering(spa->spa_dsl_pool))
6047 return (SET_ERROR(EBUSY));
6048 return (dsl_scan_cancel(spa->spa_dsl_pool));
6052 spa_scan(spa_t *spa, pool_scan_func_t func)
6054 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
6056 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
6057 return (SET_ERROR(ENOTSUP));
6060 * If a resilver was requested, but there is no DTL on a
6061 * writeable leaf device, we have nothing to do.
6063 if (func == POOL_SCAN_RESILVER &&
6064 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
6065 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
6069 return (dsl_scan(spa->spa_dsl_pool, func));
6073 * ==========================================================================
6074 * SPA async task processing
6075 * ==========================================================================
6079 spa_async_remove(spa_t *spa, vdev_t *vd)
6081 if (vd->vdev_remove_wanted) {
6082 vd->vdev_remove_wanted = B_FALSE;
6083 vd->vdev_delayed_close = B_FALSE;
6084 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
6087 * We want to clear the stats, but we don't want to do a full
6088 * vdev_clear() as that will cause us to throw away
6089 * degraded/faulted state as well as attempt to reopen the
6090 * device, all of which is a waste.
6092 vd->vdev_stat.vs_read_errors = 0;
6093 vd->vdev_stat.vs_write_errors = 0;
6094 vd->vdev_stat.vs_checksum_errors = 0;
6096 vdev_state_dirty(vd->vdev_top);
6097 /* Tell userspace that the vdev is gone. */
6098 zfs_post_remove(spa, vd);
6101 for (int c = 0; c < vd->vdev_children; c++)
6102 spa_async_remove(spa, vd->vdev_child[c]);
6106 spa_async_probe(spa_t *spa, vdev_t *vd)
6108 if (vd->vdev_probe_wanted) {
6109 vd->vdev_probe_wanted = B_FALSE;
6110 vdev_reopen(vd); /* vdev_open() does the actual probe */
6113 for (int c = 0; c < vd->vdev_children; c++)
6114 spa_async_probe(spa, vd->vdev_child[c]);
6118 spa_async_autoexpand(spa_t *spa, vdev_t *vd)
6124 if (!spa->spa_autoexpand)
6127 for (int c = 0; c < vd->vdev_children; c++) {
6128 vdev_t *cvd = vd->vdev_child[c];
6129 spa_async_autoexpand(spa, cvd);
6132 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
6135 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
6136 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
6138 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
6139 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
6141 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
6142 ESC_ZFS_VDEV_AUTOEXPAND, attr, &eid, DDI_SLEEP);
6145 kmem_free(physpath, MAXPATHLEN);
6149 spa_async_thread(void *arg)
6154 ASSERT(spa->spa_sync_on);
6156 mutex_enter(&spa->spa_async_lock);
6157 tasks = spa->spa_async_tasks;
6158 spa->spa_async_tasks &= SPA_ASYNC_REMOVE;
6159 mutex_exit(&spa->spa_async_lock);
6162 * See if the config needs to be updated.
6164 if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
6165 uint64_t old_space, new_space;
6167 mutex_enter(&spa_namespace_lock);
6168 old_space = metaslab_class_get_space(spa_normal_class(spa));
6169 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
6170 new_space = metaslab_class_get_space(spa_normal_class(spa));
6171 mutex_exit(&spa_namespace_lock);
6174 * If the pool grew as a result of the config update,
6175 * then log an internal history event.
6177 if (new_space != old_space) {
6178 spa_history_log_internal(spa, "vdev online", NULL,
6179 "pool '%s' size: %llu(+%llu)",
6180 spa_name(spa), new_space, new_space - old_space);
6184 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
6185 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
6186 spa_async_autoexpand(spa, spa->spa_root_vdev);
6187 spa_config_exit(spa, SCL_CONFIG, FTAG);
6191 * See if any devices need to be probed.
6193 if (tasks & SPA_ASYNC_PROBE) {
6194 spa_vdev_state_enter(spa, SCL_NONE);
6195 spa_async_probe(spa, spa->spa_root_vdev);
6196 (void) spa_vdev_state_exit(spa, NULL, 0);
6200 * If any devices are done replacing, detach them.
6202 if (tasks & SPA_ASYNC_RESILVER_DONE)
6203 spa_vdev_resilver_done(spa);
6206 * Kick off a resilver.
6208 if (tasks & SPA_ASYNC_RESILVER)
6209 dsl_resilver_restart(spa->spa_dsl_pool, 0);
6212 * Let the world know that we're done.
6214 mutex_enter(&spa->spa_async_lock);
6215 spa->spa_async_thread = NULL;
6216 cv_broadcast(&spa->spa_async_cv);
6217 mutex_exit(&spa->spa_async_lock);
6222 spa_async_thread_vd(void *arg)
6227 ASSERT(spa->spa_sync_on);
6229 mutex_enter(&spa->spa_async_lock);
6230 tasks = spa->spa_async_tasks;
6232 spa->spa_async_tasks &= ~SPA_ASYNC_REMOVE;
6233 mutex_exit(&spa->spa_async_lock);
6236 * See if any devices need to be marked REMOVED.
6238 if (tasks & SPA_ASYNC_REMOVE) {
6239 spa_vdev_state_enter(spa, SCL_NONE);
6240 spa_async_remove(spa, spa->spa_root_vdev);
6241 for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
6242 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
6243 for (int i = 0; i < spa->spa_spares.sav_count; i++)
6244 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
6245 (void) spa_vdev_state_exit(spa, NULL, 0);
6249 * Let the world know that we're done.
6251 mutex_enter(&spa->spa_async_lock);
6252 tasks = spa->spa_async_tasks;
6253 if ((tasks & SPA_ASYNC_REMOVE) != 0)
6255 spa->spa_async_thread_vd = NULL;
6256 cv_broadcast(&spa->spa_async_cv);
6257 mutex_exit(&spa->spa_async_lock);
6262 spa_async_suspend(spa_t *spa)
6264 mutex_enter(&spa->spa_async_lock);
6265 spa->spa_async_suspended++;
6266 while (spa->spa_async_thread != NULL &&
6267 spa->spa_async_thread_vd != NULL)
6268 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
6269 mutex_exit(&spa->spa_async_lock);
6273 spa_async_resume(spa_t *spa)
6275 mutex_enter(&spa->spa_async_lock);
6276 ASSERT(spa->spa_async_suspended != 0);
6277 spa->spa_async_suspended--;
6278 mutex_exit(&spa->spa_async_lock);
6282 spa_async_tasks_pending(spa_t *spa)
6284 uint_t non_config_tasks;
6286 boolean_t config_task_suspended;
6288 non_config_tasks = spa->spa_async_tasks & ~(SPA_ASYNC_CONFIG_UPDATE |
6290 config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE;
6291 if (spa->spa_ccw_fail_time == 0) {
6292 config_task_suspended = B_FALSE;
6294 config_task_suspended =
6295 (gethrtime() - spa->spa_ccw_fail_time) <
6296 (zfs_ccw_retry_interval * NANOSEC);
6299 return (non_config_tasks || (config_task && !config_task_suspended));
6303 spa_async_dispatch(spa_t *spa)
6305 mutex_enter(&spa->spa_async_lock);
6306 if (spa_async_tasks_pending(spa) &&
6307 !spa->spa_async_suspended &&
6308 spa->spa_async_thread == NULL &&
6310 spa->spa_async_thread = thread_create(NULL, 0,
6311 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
6312 mutex_exit(&spa->spa_async_lock);
6316 spa_async_dispatch_vd(spa_t *spa)
6318 mutex_enter(&spa->spa_async_lock);
6319 if ((spa->spa_async_tasks & SPA_ASYNC_REMOVE) != 0 &&
6320 !spa->spa_async_suspended &&
6321 spa->spa_async_thread_vd == NULL &&
6323 spa->spa_async_thread_vd = thread_create(NULL, 0,
6324 spa_async_thread_vd, spa, 0, &p0, TS_RUN, maxclsyspri);
6325 mutex_exit(&spa->spa_async_lock);
6329 spa_async_request(spa_t *spa, int task)
6331 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
6332 mutex_enter(&spa->spa_async_lock);
6333 spa->spa_async_tasks |= task;
6334 mutex_exit(&spa->spa_async_lock);
6335 spa_async_dispatch_vd(spa);
6339 * ==========================================================================
6340 * SPA syncing routines
6341 * ==========================================================================
6345 bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
6348 bpobj_enqueue(bpo, bp, tx);
6353 spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
6357 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
6358 BP_GET_PSIZE(bp), zio->io_flags));
6363 * Note: this simple function is not inlined to make it easier to dtrace the
6364 * amount of time spent syncing frees.
6367 spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx)
6369 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6370 bplist_iterate(bpl, spa_free_sync_cb, zio, tx);
6371 VERIFY(zio_wait(zio) == 0);
6375 * Note: this simple function is not inlined to make it easier to dtrace the
6376 * amount of time spent syncing deferred frees.
6379 spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx)
6381 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6382 VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj,
6383 spa_free_sync_cb, zio, tx), ==, 0);
6384 VERIFY0(zio_wait(zio));
6389 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
6391 char *packed = NULL;
6396 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
6399 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
6400 * information. This avoids the dmu_buf_will_dirty() path and
6401 * saves us a pre-read to get data we don't actually care about.
6403 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
6404 packed = kmem_alloc(bufsize, KM_SLEEP);
6406 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
6408 bzero(packed + nvsize, bufsize - nvsize);
6410 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
6412 kmem_free(packed, bufsize);
6414 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
6415 dmu_buf_will_dirty(db, tx);
6416 *(uint64_t *)db->db_data = nvsize;
6417 dmu_buf_rele(db, FTAG);
6421 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
6422 const char *config, const char *entry)
6432 * Update the MOS nvlist describing the list of available devices.
6433 * spa_validate_aux() will have already made sure this nvlist is
6434 * valid and the vdevs are labeled appropriately.
6436 if (sav->sav_object == 0) {
6437 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
6438 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
6439 sizeof (uint64_t), tx);
6440 VERIFY(zap_update(spa->spa_meta_objset,
6441 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
6442 &sav->sav_object, tx) == 0);
6445 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
6446 if (sav->sav_count == 0) {
6447 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
6449 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
6450 for (i = 0; i < sav->sav_count; i++)
6451 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
6452 B_FALSE, VDEV_CONFIG_L2CACHE);
6453 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
6454 sav->sav_count) == 0);
6455 for (i = 0; i < sav->sav_count; i++)
6456 nvlist_free(list[i]);
6457 kmem_free(list, sav->sav_count * sizeof (void *));
6460 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
6461 nvlist_free(nvroot);
6463 sav->sav_sync = B_FALSE;
6467 * Rebuild spa's all-vdev ZAP from the vdev ZAPs indicated in each vdev_t.
6468 * The all-vdev ZAP must be empty.
6471 spa_avz_build(vdev_t *vd, uint64_t avz, dmu_tx_t *tx)
6473 spa_t *spa = vd->vdev_spa;
6474 if (vd->vdev_top_zap != 0) {
6475 VERIFY0(zap_add_int(spa->spa_meta_objset, avz,
6476 vd->vdev_top_zap, tx));
6478 if (vd->vdev_leaf_zap != 0) {
6479 VERIFY0(zap_add_int(spa->spa_meta_objset, avz,
6480 vd->vdev_leaf_zap, tx));
6482 for (uint64_t i = 0; i < vd->vdev_children; i++) {
6483 spa_avz_build(vd->vdev_child[i], avz, tx);
6488 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
6493 * If the pool is being imported from a pre-per-vdev-ZAP version of ZFS,
6494 * its config may not be dirty but we still need to build per-vdev ZAPs.
6495 * Similarly, if the pool is being assembled (e.g. after a split), we
6496 * need to rebuild the AVZ although the config may not be dirty.
6498 if (list_is_empty(&spa->spa_config_dirty_list) &&
6499 spa->spa_avz_action == AVZ_ACTION_NONE)
6502 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6504 ASSERT(spa->spa_avz_action == AVZ_ACTION_NONE ||
6505 spa->spa_avz_action == AVZ_ACTION_INITIALIZE ||
6506 spa->spa_all_vdev_zaps != 0);
6508 if (spa->spa_avz_action == AVZ_ACTION_REBUILD) {
6509 /* Make and build the new AVZ */
6510 uint64_t new_avz = zap_create(spa->spa_meta_objset,
6511 DMU_OTN_ZAP_METADATA, DMU_OT_NONE, 0, tx);
6512 spa_avz_build(spa->spa_root_vdev, new_avz, tx);
6514 /* Diff old AVZ with new one */
6518 for (zap_cursor_init(&zc, spa->spa_meta_objset,
6519 spa->spa_all_vdev_zaps);
6520 zap_cursor_retrieve(&zc, &za) == 0;
6521 zap_cursor_advance(&zc)) {
6522 uint64_t vdzap = za.za_first_integer;
6523 if (zap_lookup_int(spa->spa_meta_objset, new_avz,
6526 * ZAP is listed in old AVZ but not in new one;
6529 VERIFY0(zap_destroy(spa->spa_meta_objset, vdzap,
6534 zap_cursor_fini(&zc);
6536 /* Destroy the old AVZ */
6537 VERIFY0(zap_destroy(spa->spa_meta_objset,
6538 spa->spa_all_vdev_zaps, tx));
6540 /* Replace the old AVZ in the dir obj with the new one */
6541 VERIFY0(zap_update(spa->spa_meta_objset,
6542 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP,
6543 sizeof (new_avz), 1, &new_avz, tx));
6545 spa->spa_all_vdev_zaps = new_avz;
6546 } else if (spa->spa_avz_action == AVZ_ACTION_DESTROY) {
6550 /* Walk through the AVZ and destroy all listed ZAPs */
6551 for (zap_cursor_init(&zc, spa->spa_meta_objset,
6552 spa->spa_all_vdev_zaps);
6553 zap_cursor_retrieve(&zc, &za) == 0;
6554 zap_cursor_advance(&zc)) {
6555 uint64_t zap = za.za_first_integer;
6556 VERIFY0(zap_destroy(spa->spa_meta_objset, zap, tx));
6559 zap_cursor_fini(&zc);
6561 /* Destroy and unlink the AVZ itself */
6562 VERIFY0(zap_destroy(spa->spa_meta_objset,
6563 spa->spa_all_vdev_zaps, tx));
6564 VERIFY0(zap_remove(spa->spa_meta_objset,
6565 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP, tx));
6566 spa->spa_all_vdev_zaps = 0;
6569 if (spa->spa_all_vdev_zaps == 0) {
6570 spa->spa_all_vdev_zaps = zap_create_link(spa->spa_meta_objset,
6571 DMU_OTN_ZAP_METADATA, DMU_POOL_DIRECTORY_OBJECT,
6572 DMU_POOL_VDEV_ZAP_MAP, tx);
6574 spa->spa_avz_action = AVZ_ACTION_NONE;
6576 /* Create ZAPs for vdevs that don't have them. */
6577 vdev_construct_zaps(spa->spa_root_vdev, tx);
6579 config = spa_config_generate(spa, spa->spa_root_vdev,
6580 dmu_tx_get_txg(tx), B_FALSE);
6583 * If we're upgrading the spa version then make sure that
6584 * the config object gets updated with the correct version.
6586 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
6587 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
6588 spa->spa_uberblock.ub_version);
6590 spa_config_exit(spa, SCL_STATE, FTAG);
6592 nvlist_free(spa->spa_config_syncing);
6593 spa->spa_config_syncing = config;
6595 spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
6599 spa_sync_version(void *arg, dmu_tx_t *tx)
6601 uint64_t *versionp = arg;
6602 uint64_t version = *versionp;
6603 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
6606 * Setting the version is special cased when first creating the pool.
6608 ASSERT(tx->tx_txg != TXG_INITIAL);
6610 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
6611 ASSERT(version >= spa_version(spa));
6613 spa->spa_uberblock.ub_version = version;
6614 vdev_config_dirty(spa->spa_root_vdev);
6615 spa_history_log_internal(spa, "set", tx, "version=%lld", version);
6619 * Set zpool properties.
6622 spa_sync_props(void *arg, dmu_tx_t *tx)
6624 nvlist_t *nvp = arg;
6625 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
6626 objset_t *mos = spa->spa_meta_objset;
6627 nvpair_t *elem = NULL;
6629 mutex_enter(&spa->spa_props_lock);
6631 while ((elem = nvlist_next_nvpair(nvp, elem))) {
6633 char *strval, *fname;
6635 const char *propname;
6636 zprop_type_t proptype;
6639 switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
6642 * We checked this earlier in spa_prop_validate().
6644 ASSERT(zpool_prop_feature(nvpair_name(elem)));
6646 fname = strchr(nvpair_name(elem), '@') + 1;
6647 VERIFY0(zfeature_lookup_name(fname, &fid));
6649 spa_feature_enable(spa, fid, tx);
6650 spa_history_log_internal(spa, "set", tx,
6651 "%s=enabled", nvpair_name(elem));
6654 case ZPOOL_PROP_VERSION:
6655 intval = fnvpair_value_uint64(elem);
6657 * The version is synced seperatly before other
6658 * properties and should be correct by now.
6660 ASSERT3U(spa_version(spa), >=, intval);
6663 case ZPOOL_PROP_ALTROOT:
6665 * 'altroot' is a non-persistent property. It should
6666 * have been set temporarily at creation or import time.
6668 ASSERT(spa->spa_root != NULL);
6671 case ZPOOL_PROP_READONLY:
6672 case ZPOOL_PROP_CACHEFILE:
6674 * 'readonly' and 'cachefile' are also non-persisitent
6678 case ZPOOL_PROP_COMMENT:
6679 strval = fnvpair_value_string(elem);
6680 if (spa->spa_comment != NULL)
6681 spa_strfree(spa->spa_comment);
6682 spa->spa_comment = spa_strdup(strval);
6684 * We need to dirty the configuration on all the vdevs
6685 * so that their labels get updated. It's unnecessary
6686 * to do this for pool creation since the vdev's
6687 * configuratoin has already been dirtied.
6689 if (tx->tx_txg != TXG_INITIAL)
6690 vdev_config_dirty(spa->spa_root_vdev);
6691 spa_history_log_internal(spa, "set", tx,
6692 "%s=%s", nvpair_name(elem), strval);
6696 * Set pool property values in the poolprops mos object.
6698 if (spa->spa_pool_props_object == 0) {
6699 spa->spa_pool_props_object =
6700 zap_create_link(mos, DMU_OT_POOL_PROPS,
6701 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
6705 /* normalize the property name */
6706 propname = zpool_prop_to_name(prop);
6707 proptype = zpool_prop_get_type(prop);
6709 if (nvpair_type(elem) == DATA_TYPE_STRING) {
6710 ASSERT(proptype == PROP_TYPE_STRING);
6711 strval = fnvpair_value_string(elem);
6712 VERIFY0(zap_update(mos,
6713 spa->spa_pool_props_object, propname,
6714 1, strlen(strval) + 1, strval, tx));
6715 spa_history_log_internal(spa, "set", tx,
6716 "%s=%s", nvpair_name(elem), strval);
6717 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
6718 intval = fnvpair_value_uint64(elem);
6720 if (proptype == PROP_TYPE_INDEX) {
6722 VERIFY0(zpool_prop_index_to_string(
6723 prop, intval, &unused));
6725 VERIFY0(zap_update(mos,
6726 spa->spa_pool_props_object, propname,
6727 8, 1, &intval, tx));
6728 spa_history_log_internal(spa, "set", tx,
6729 "%s=%lld", nvpair_name(elem), intval);
6731 ASSERT(0); /* not allowed */
6735 case ZPOOL_PROP_DELEGATION:
6736 spa->spa_delegation = intval;
6738 case ZPOOL_PROP_BOOTFS:
6739 spa->spa_bootfs = intval;
6741 case ZPOOL_PROP_FAILUREMODE:
6742 spa->spa_failmode = intval;
6744 case ZPOOL_PROP_AUTOEXPAND:
6745 spa->spa_autoexpand = intval;
6746 if (tx->tx_txg != TXG_INITIAL)
6747 spa_async_request(spa,
6748 SPA_ASYNC_AUTOEXPAND);
6750 case ZPOOL_PROP_DEDUPDITTO:
6751 spa->spa_dedup_ditto = intval;
6760 mutex_exit(&spa->spa_props_lock);
6764 * Perform one-time upgrade on-disk changes. spa_version() does not
6765 * reflect the new version this txg, so there must be no changes this
6766 * txg to anything that the upgrade code depends on after it executes.
6767 * Therefore this must be called after dsl_pool_sync() does the sync
6771 spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
6773 dsl_pool_t *dp = spa->spa_dsl_pool;
6775 ASSERT(spa->spa_sync_pass == 1);
6777 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
6779 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
6780 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
6781 dsl_pool_create_origin(dp, tx);
6783 /* Keeping the origin open increases spa_minref */
6784 spa->spa_minref += 3;
6787 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
6788 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
6789 dsl_pool_upgrade_clones(dp, tx);
6792 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
6793 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
6794 dsl_pool_upgrade_dir_clones(dp, tx);
6796 /* Keeping the freedir open increases spa_minref */
6797 spa->spa_minref += 3;
6800 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES &&
6801 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
6802 spa_feature_create_zap_objects(spa, tx);
6806 * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable
6807 * when possibility to use lz4 compression for metadata was added
6808 * Old pools that have this feature enabled must be upgraded to have
6809 * this feature active
6811 if (spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
6812 boolean_t lz4_en = spa_feature_is_enabled(spa,
6813 SPA_FEATURE_LZ4_COMPRESS);
6814 boolean_t lz4_ac = spa_feature_is_active(spa,
6815 SPA_FEATURE_LZ4_COMPRESS);
6817 if (lz4_en && !lz4_ac)
6818 spa_feature_incr(spa, SPA_FEATURE_LZ4_COMPRESS, tx);
6822 * If we haven't written the salt, do so now. Note that the
6823 * feature may not be activated yet, but that's fine since
6824 * the presence of this ZAP entry is backwards compatible.
6826 if (zap_contains(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
6827 DMU_POOL_CHECKSUM_SALT) == ENOENT) {
6828 VERIFY0(zap_add(spa->spa_meta_objset,
6829 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CHECKSUM_SALT, 1,
6830 sizeof (spa->spa_cksum_salt.zcs_bytes),
6831 spa->spa_cksum_salt.zcs_bytes, tx));
6834 rrw_exit(&dp->dp_config_rwlock, FTAG);
6838 * Sync the specified transaction group. New blocks may be dirtied as
6839 * part of the process, so we iterate until it converges.
6842 spa_sync(spa_t *spa, uint64_t txg)
6844 dsl_pool_t *dp = spa->spa_dsl_pool;
6845 objset_t *mos = spa->spa_meta_objset;
6846 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
6847 vdev_t *rvd = spa->spa_root_vdev;
6851 uint32_t max_queue_depth = zfs_vdev_async_write_max_active *
6852 zfs_vdev_queue_depth_pct / 100;
6854 VERIFY(spa_writeable(spa));
6857 * Lock out configuration changes.
6859 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
6861 spa->spa_syncing_txg = txg;
6862 spa->spa_sync_pass = 0;
6864 mutex_enter(&spa->spa_alloc_lock);
6865 VERIFY0(avl_numnodes(&spa->spa_alloc_tree));
6866 mutex_exit(&spa->spa_alloc_lock);
6869 * If there are any pending vdev state changes, convert them
6870 * into config changes that go out with this transaction group.
6872 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6873 while (list_head(&spa->spa_state_dirty_list) != NULL) {
6875 * We need the write lock here because, for aux vdevs,
6876 * calling vdev_config_dirty() modifies sav_config.
6877 * This is ugly and will become unnecessary when we
6878 * eliminate the aux vdev wart by integrating all vdevs
6879 * into the root vdev tree.
6881 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6882 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
6883 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
6884 vdev_state_clean(vd);
6885 vdev_config_dirty(vd);
6887 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6888 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
6890 spa_config_exit(spa, SCL_STATE, FTAG);
6892 tx = dmu_tx_create_assigned(dp, txg);
6894 spa->spa_sync_starttime = gethrtime();
6896 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid,
6897 spa->spa_sync_starttime + spa->spa_deadman_synctime));
6898 #else /* !illumos */
6900 callout_schedule(&spa->spa_deadman_cycid,
6901 hz * spa->spa_deadman_synctime / NANOSEC);
6903 #endif /* illumos */
6906 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
6907 * set spa_deflate if we have no raid-z vdevs.
6909 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
6910 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
6913 for (i = 0; i < rvd->vdev_children; i++) {
6914 vd = rvd->vdev_child[i];
6915 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
6918 if (i == rvd->vdev_children) {
6919 spa->spa_deflate = TRUE;
6920 VERIFY(0 == zap_add(spa->spa_meta_objset,
6921 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
6922 sizeof (uint64_t), 1, &spa->spa_deflate, tx));
6927 * Set the top-level vdev's max queue depth. Evaluate each
6928 * top-level's async write queue depth in case it changed.
6929 * The max queue depth will not change in the middle of syncing
6932 uint64_t queue_depth_total = 0;
6933 for (int c = 0; c < rvd->vdev_children; c++) {
6934 vdev_t *tvd = rvd->vdev_child[c];
6935 metaslab_group_t *mg = tvd->vdev_mg;
6937 if (mg == NULL || mg->mg_class != spa_normal_class(spa) ||
6938 !metaslab_group_initialized(mg))
6942 * It is safe to do a lock-free check here because only async
6943 * allocations look at mg_max_alloc_queue_depth, and async
6944 * allocations all happen from spa_sync().
6946 ASSERT0(refcount_count(&mg->mg_alloc_queue_depth));
6947 mg->mg_max_alloc_queue_depth = max_queue_depth;
6948 queue_depth_total += mg->mg_max_alloc_queue_depth;
6950 metaslab_class_t *mc = spa_normal_class(spa);
6951 ASSERT0(refcount_count(&mc->mc_alloc_slots));
6952 mc->mc_alloc_max_slots = queue_depth_total;
6953 mc->mc_alloc_throttle_enabled = zio_dva_throttle_enabled;
6955 ASSERT3U(mc->mc_alloc_max_slots, <=,
6956 max_queue_depth * rvd->vdev_children);
6959 * Iterate to convergence.
6962 int pass = ++spa->spa_sync_pass;
6964 spa_sync_config_object(spa, tx);
6965 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
6966 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
6967 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
6968 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
6969 spa_errlog_sync(spa, txg);
6970 dsl_pool_sync(dp, txg);
6972 if (pass < zfs_sync_pass_deferred_free) {
6973 spa_sync_frees(spa, free_bpl, tx);
6976 * We can not defer frees in pass 1, because
6977 * we sync the deferred frees later in pass 1.
6979 ASSERT3U(pass, >, 1);
6980 bplist_iterate(free_bpl, bpobj_enqueue_cb,
6981 &spa->spa_deferred_bpobj, tx);
6985 dsl_scan_sync(dp, tx);
6987 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
6991 spa_sync_upgrades(spa, tx);
6993 spa->spa_uberblock.ub_rootbp.blk_birth);
6995 * Note: We need to check if the MOS is dirty
6996 * because we could have marked the MOS dirty
6997 * without updating the uberblock (e.g. if we
6998 * have sync tasks but no dirty user data). We
6999 * need to check the uberblock's rootbp because
7000 * it is updated if we have synced out dirty
7001 * data (though in this case the MOS will most
7002 * likely also be dirty due to second order
7003 * effects, we don't want to rely on that here).
7005 if (spa->spa_uberblock.ub_rootbp.blk_birth < txg &&
7006 !dmu_objset_is_dirty(mos, txg)) {
7008 * Nothing changed on the first pass,
7009 * therefore this TXG is a no-op. Avoid
7010 * syncing deferred frees, so that we
7011 * can keep this TXG as a no-op.
7013 ASSERT(txg_list_empty(&dp->dp_dirty_datasets,
7015 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
7016 ASSERT(txg_list_empty(&dp->dp_sync_tasks, txg));
7019 spa_sync_deferred_frees(spa, tx);
7022 } while (dmu_objset_is_dirty(mos, txg));
7024 if (!list_is_empty(&spa->spa_config_dirty_list)) {
7026 * Make sure that the number of ZAPs for all the vdevs matches
7027 * the number of ZAPs in the per-vdev ZAP list. This only gets
7028 * called if the config is dirty; otherwise there may be
7029 * outstanding AVZ operations that weren't completed in
7030 * spa_sync_config_object.
7032 uint64_t all_vdev_zap_entry_count;
7033 ASSERT0(zap_count(spa->spa_meta_objset,
7034 spa->spa_all_vdev_zaps, &all_vdev_zap_entry_count));
7035 ASSERT3U(vdev_count_verify_zaps(spa->spa_root_vdev), ==,
7036 all_vdev_zap_entry_count);
7040 * Rewrite the vdev configuration (which includes the uberblock)
7041 * to commit the transaction group.
7043 * If there are no dirty vdevs, we sync the uberblock to a few
7044 * random top-level vdevs that are known to be visible in the
7045 * config cache (see spa_vdev_add() for a complete description).
7046 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
7050 * We hold SCL_STATE to prevent vdev open/close/etc.
7051 * while we're attempting to write the vdev labels.
7053 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
7055 if (list_is_empty(&spa->spa_config_dirty_list)) {
7056 vdev_t *svd[SPA_DVAS_PER_BP];
7058 int children = rvd->vdev_children;
7059 int c0 = spa_get_random(children);
7061 for (int c = 0; c < children; c++) {
7062 vd = rvd->vdev_child[(c0 + c) % children];
7063 if (vd->vdev_ms_array == 0 || vd->vdev_islog)
7065 svd[svdcount++] = vd;
7066 if (svdcount == SPA_DVAS_PER_BP)
7069 error = vdev_config_sync(svd, svdcount, txg);
7071 error = vdev_config_sync(rvd->vdev_child,
7072 rvd->vdev_children, txg);
7076 spa->spa_last_synced_guid = rvd->vdev_guid;
7078 spa_config_exit(spa, SCL_STATE, FTAG);
7082 zio_suspend(spa, NULL);
7083 zio_resume_wait(spa);
7088 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY));
7089 #else /* !illumos */
7091 callout_drain(&spa->spa_deadman_cycid);
7093 #endif /* illumos */
7096 * Clear the dirty config list.
7098 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
7099 vdev_config_clean(vd);
7102 * Now that the new config has synced transactionally,
7103 * let it become visible to the config cache.
7105 if (spa->spa_config_syncing != NULL) {
7106 spa_config_set(spa, spa->spa_config_syncing);
7107 spa->spa_config_txg = txg;
7108 spa->spa_config_syncing = NULL;
7111 dsl_pool_sync_done(dp, txg);
7113 mutex_enter(&spa->spa_alloc_lock);
7114 VERIFY0(avl_numnodes(&spa->spa_alloc_tree));
7115 mutex_exit(&spa->spa_alloc_lock);
7118 * Update usable space statistics.
7120 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
7121 vdev_sync_done(vd, txg);
7123 spa_update_dspace(spa);
7126 * It had better be the case that we didn't dirty anything
7127 * since vdev_config_sync().
7129 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
7130 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
7131 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
7133 spa->spa_sync_pass = 0;
7136 * Update the last synced uberblock here. We want to do this at
7137 * the end of spa_sync() so that consumers of spa_last_synced_txg()
7138 * will be guaranteed that all the processing associated with
7139 * that txg has been completed.
7141 spa->spa_ubsync = spa->spa_uberblock;
7142 spa_config_exit(spa, SCL_CONFIG, FTAG);
7144 spa_handle_ignored_writes(spa);
7147 * If any async tasks have been requested, kick them off.
7149 spa_async_dispatch(spa);
7150 spa_async_dispatch_vd(spa);
7154 * Sync all pools. We don't want to hold the namespace lock across these
7155 * operations, so we take a reference on the spa_t and drop the lock during the
7159 spa_sync_allpools(void)
7162 mutex_enter(&spa_namespace_lock);
7163 while ((spa = spa_next(spa)) != NULL) {
7164 if (spa_state(spa) != POOL_STATE_ACTIVE ||
7165 !spa_writeable(spa) || spa_suspended(spa))
7167 spa_open_ref(spa, FTAG);
7168 mutex_exit(&spa_namespace_lock);
7169 txg_wait_synced(spa_get_dsl(spa), 0);
7170 mutex_enter(&spa_namespace_lock);
7171 spa_close(spa, FTAG);
7173 mutex_exit(&spa_namespace_lock);
7177 * ==========================================================================
7178 * Miscellaneous routines
7179 * ==========================================================================
7183 * Remove all pools in the system.
7191 * Remove all cached state. All pools should be closed now,
7192 * so every spa in the AVL tree should be unreferenced.
7194 mutex_enter(&spa_namespace_lock);
7195 while ((spa = spa_next(NULL)) != NULL) {
7197 * Stop async tasks. The async thread may need to detach
7198 * a device that's been replaced, which requires grabbing
7199 * spa_namespace_lock, so we must drop it here.
7201 spa_open_ref(spa, FTAG);
7202 mutex_exit(&spa_namespace_lock);
7203 spa_async_suspend(spa);
7204 mutex_enter(&spa_namespace_lock);
7205 spa_close(spa, FTAG);
7207 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
7209 spa_deactivate(spa);
7213 mutex_exit(&spa_namespace_lock);
7217 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
7222 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
7226 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
7227 vd = spa->spa_l2cache.sav_vdevs[i];
7228 if (vd->vdev_guid == guid)
7232 for (i = 0; i < spa->spa_spares.sav_count; i++) {
7233 vd = spa->spa_spares.sav_vdevs[i];
7234 if (vd->vdev_guid == guid)
7243 spa_upgrade(spa_t *spa, uint64_t version)
7245 ASSERT(spa_writeable(spa));
7247 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
7250 * This should only be called for a non-faulted pool, and since a
7251 * future version would result in an unopenable pool, this shouldn't be
7254 ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version));
7255 ASSERT3U(version, >=, spa->spa_uberblock.ub_version);
7257 spa->spa_uberblock.ub_version = version;
7258 vdev_config_dirty(spa->spa_root_vdev);
7260 spa_config_exit(spa, SCL_ALL, FTAG);
7262 txg_wait_synced(spa_get_dsl(spa), 0);
7266 spa_has_spare(spa_t *spa, uint64_t guid)
7270 spa_aux_vdev_t *sav = &spa->spa_spares;
7272 for (i = 0; i < sav->sav_count; i++)
7273 if (sav->sav_vdevs[i]->vdev_guid == guid)
7276 for (i = 0; i < sav->sav_npending; i++) {
7277 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
7278 &spareguid) == 0 && spareguid == guid)
7286 * Check if a pool has an active shared spare device.
7287 * Note: reference count of an active spare is 2, as a spare and as a replace
7290 spa_has_active_shared_spare(spa_t *spa)
7294 spa_aux_vdev_t *sav = &spa->spa_spares;
7296 for (i = 0; i < sav->sav_count; i++) {
7297 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
7298 &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
7307 spa_event_create(spa_t *spa, vdev_t *vd, const char *name)
7309 sysevent_t *ev = NULL;
7311 sysevent_attr_list_t *attr = NULL;
7312 sysevent_value_t value;
7314 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
7318 value.value_type = SE_DATA_TYPE_STRING;
7319 value.value.sv_string = spa_name(spa);
7320 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
7323 value.value_type = SE_DATA_TYPE_UINT64;
7324 value.value.sv_uint64 = spa_guid(spa);
7325 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
7329 value.value_type = SE_DATA_TYPE_UINT64;
7330 value.value.sv_uint64 = vd->vdev_guid;
7331 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
7335 if (vd->vdev_path) {
7336 value.value_type = SE_DATA_TYPE_STRING;
7337 value.value.sv_string = vd->vdev_path;
7338 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
7339 &value, SE_SLEEP) != 0)
7344 if (sysevent_attach_attributes(ev, attr) != 0)
7350 sysevent_free_attr(attr);
7357 spa_event_post(sysevent_t *ev)
7362 (void) log_sysevent(ev, SE_SLEEP, &eid);
7368 * Post a sysevent corresponding to the given event. The 'name' must be one of
7369 * the event definitions in sys/sysevent/eventdefs.h. The payload will be
7370 * filled in from the spa and (optionally) the vdev. This doesn't do anything
7371 * in the userland libzpool, as we don't want consumers to misinterpret ztest
7372 * or zdb as real changes.
7375 spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
7377 spa_event_post(spa_event_create(spa, vd, name));