4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2011, 2014 by Delphix. All rights reserved.
25 * Copyright (c) 2015, Nexenta Systems, Inc. All rights reserved.
26 * Copyright (c) 2013 Martin Matuska <mm@FreeBSD.org>. All rights reserved.
27 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
28 * Copyright 2013 Saso Kiselkov. All rights reserved.
29 * Copyright (c) 2014 Integros [integros.com]
33 * SPA: Storage Pool Allocator
35 * This file contains all the routines used when modifying on-disk SPA state.
36 * This includes opening, importing, destroying, exporting a pool, and syncing a
40 #include <sys/zfs_context.h>
41 #include <sys/fm/fs/zfs.h>
42 #include <sys/spa_impl.h>
44 #include <sys/zio_checksum.h>
46 #include <sys/dmu_tx.h>
50 #include <sys/vdev_impl.h>
51 #include <sys/metaslab.h>
52 #include <sys/metaslab_impl.h>
53 #include <sys/uberblock_impl.h>
56 #include <sys/dmu_traverse.h>
57 #include <sys/dmu_objset.h>
58 #include <sys/unique.h>
59 #include <sys/dsl_pool.h>
60 #include <sys/dsl_dataset.h>
61 #include <sys/dsl_dir.h>
62 #include <sys/dsl_prop.h>
63 #include <sys/dsl_synctask.h>
64 #include <sys/fs/zfs.h>
66 #include <sys/callb.h>
67 #include <sys/spa_boot.h>
68 #include <sys/zfs_ioctl.h>
69 #include <sys/dsl_scan.h>
70 #include <sys/dmu_send.h>
71 #include <sys/dsl_destroy.h>
72 #include <sys/dsl_userhold.h>
73 #include <sys/zfeature.h>
75 #include <sys/trim_map.h>
78 #include <sys/callb.h>
79 #include <sys/cpupart.h>
84 #include "zfs_comutil.h"
86 /* Check hostid on import? */
87 static int check_hostid = 1;
90 * The interval, in seconds, at which failed configuration cache file writes
93 static int zfs_ccw_retry_interval = 300;
95 SYSCTL_DECL(_vfs_zfs);
96 TUNABLE_INT("vfs.zfs.check_hostid", &check_hostid);
97 SYSCTL_INT(_vfs_zfs, OID_AUTO, check_hostid, CTLFLAG_RWTUN, &check_hostid, 0,
98 "Check hostid on import?");
99 TUNABLE_INT("vfs.zfs.ccw_retry_interval", &zfs_ccw_retry_interval);
100 SYSCTL_INT(_vfs_zfs, OID_AUTO, ccw_retry_interval, CTLFLAG_RW,
101 &zfs_ccw_retry_interval, 0,
102 "Configuration cache file write, retry after failure, interval (seconds)");
104 typedef enum zti_modes {
105 ZTI_MODE_FIXED, /* value is # of threads (min 1) */
106 ZTI_MODE_BATCH, /* cpu-intensive; value is ignored */
107 ZTI_MODE_NULL, /* don't create a taskq */
111 #define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) }
112 #define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 }
113 #define ZTI_NULL { ZTI_MODE_NULL, 0, 0 }
115 #define ZTI_N(n) ZTI_P(n, 1)
116 #define ZTI_ONE ZTI_N(1)
118 typedef struct zio_taskq_info {
119 zti_modes_t zti_mode;
124 static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
125 "issue", "issue_high", "intr", "intr_high"
129 * This table defines the taskq settings for each ZFS I/O type. When
130 * initializing a pool, we use this table to create an appropriately sized
131 * taskq. Some operations are low volume and therefore have a small, static
132 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE
133 * macros. Other operations process a large amount of data; the ZTI_BATCH
134 * macro causes us to create a taskq oriented for throughput. Some operations
135 * are so high frequency and short-lived that the taskq itself can become a a
136 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an
137 * additional degree of parallelism specified by the number of threads per-
138 * taskq and the number of taskqs; when dispatching an event in this case, the
139 * particular taskq is chosen at random.
141 * The different taskq priorities are to handle the different contexts (issue
142 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that
143 * need to be handled with minimum delay.
145 const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
146 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
147 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* NULL */
148 { ZTI_N(8), ZTI_NULL, ZTI_P(12, 8), ZTI_NULL }, /* READ */
149 { ZTI_BATCH, ZTI_N(5), ZTI_N(8), ZTI_N(5) }, /* WRITE */
150 { ZTI_P(12, 8), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FREE */
151 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* CLAIM */
152 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* IOCTL */
155 static sysevent_t *spa_event_create(spa_t *spa, vdev_t *vd, const char *name);
156 static void spa_event_post(sysevent_t *ev);
157 static void spa_sync_version(void *arg, dmu_tx_t *tx);
158 static void spa_sync_props(void *arg, dmu_tx_t *tx);
159 static boolean_t spa_has_active_shared_spare(spa_t *spa);
160 static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
161 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
163 static void spa_vdev_resilver_done(spa_t *spa);
165 uint_t zio_taskq_batch_pct = 75; /* 1 thread per cpu in pset */
167 id_t zio_taskq_psrset_bind = PS_NONE;
170 boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */
171 uint_t zio_taskq_basedc = 80; /* base duty cycle */
174 boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */
175 extern int zfs_sync_pass_deferred_free;
178 * This (illegal) pool name is used when temporarily importing a spa_t in order
179 * to get the vdev stats associated with the imported devices.
181 #define TRYIMPORT_NAME "$import"
184 * ==========================================================================
185 * SPA properties routines
186 * ==========================================================================
190 * Add a (source=src, propname=propval) list to an nvlist.
193 spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
194 uint64_t intval, zprop_source_t src)
196 const char *propname = zpool_prop_to_name(prop);
199 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
200 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
203 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
205 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
207 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
208 nvlist_free(propval);
212 * Get property values from the spa configuration.
215 spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
217 vdev_t *rvd = spa->spa_root_vdev;
218 dsl_pool_t *pool = spa->spa_dsl_pool;
219 uint64_t size, alloc, cap, version;
220 zprop_source_t src = ZPROP_SRC_NONE;
221 spa_config_dirent_t *dp;
222 metaslab_class_t *mc = spa_normal_class(spa);
224 ASSERT(MUTEX_HELD(&spa->spa_props_lock));
227 alloc = metaslab_class_get_alloc(spa_normal_class(spa));
228 size = metaslab_class_get_space(spa_normal_class(spa));
229 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
230 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
231 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
232 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
235 spa_prop_add_list(*nvp, ZPOOL_PROP_FRAGMENTATION, NULL,
236 metaslab_class_fragmentation(mc), src);
237 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL,
238 metaslab_class_expandable_space(mc), src);
239 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
240 (spa_mode(spa) == FREAD), src);
242 cap = (size == 0) ? 0 : (alloc * 100 / size);
243 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
245 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
246 ddt_get_pool_dedup_ratio(spa), src);
248 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
249 rvd->vdev_state, src);
251 version = spa_version(spa);
252 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
253 src = ZPROP_SRC_DEFAULT;
255 src = ZPROP_SRC_LOCAL;
256 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
261 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
262 * when opening pools before this version freedir will be NULL.
264 if (pool->dp_free_dir != NULL) {
265 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
266 dsl_dir_phys(pool->dp_free_dir)->dd_used_bytes,
269 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
273 if (pool->dp_leak_dir != NULL) {
274 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, NULL,
275 dsl_dir_phys(pool->dp_leak_dir)->dd_used_bytes,
278 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED,
283 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
285 if (spa->spa_comment != NULL) {
286 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
290 if (spa->spa_root != NULL)
291 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
294 if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) {
295 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL,
296 MIN(zfs_max_recordsize, SPA_MAXBLOCKSIZE), ZPROP_SRC_NONE);
298 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL,
299 SPA_OLD_MAXBLOCKSIZE, ZPROP_SRC_NONE);
302 if ((dp = list_head(&spa->spa_config_list)) != NULL) {
303 if (dp->scd_path == NULL) {
304 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
305 "none", 0, ZPROP_SRC_LOCAL);
306 } else if (strcmp(dp->scd_path, spa_config_path) != 0) {
307 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
308 dp->scd_path, 0, ZPROP_SRC_LOCAL);
314 * Get zpool property values.
317 spa_prop_get(spa_t *spa, nvlist_t **nvp)
319 objset_t *mos = spa->spa_meta_objset;
324 VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
326 mutex_enter(&spa->spa_props_lock);
329 * Get properties from the spa config.
331 spa_prop_get_config(spa, nvp);
333 /* If no pool property object, no more prop to get. */
334 if (mos == NULL || spa->spa_pool_props_object == 0) {
335 mutex_exit(&spa->spa_props_lock);
340 * Get properties from the MOS pool property object.
342 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
343 (err = zap_cursor_retrieve(&zc, &za)) == 0;
344 zap_cursor_advance(&zc)) {
347 zprop_source_t src = ZPROP_SRC_DEFAULT;
350 if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
353 switch (za.za_integer_length) {
355 /* integer property */
356 if (za.za_first_integer !=
357 zpool_prop_default_numeric(prop))
358 src = ZPROP_SRC_LOCAL;
360 if (prop == ZPOOL_PROP_BOOTFS) {
362 dsl_dataset_t *ds = NULL;
364 dp = spa_get_dsl(spa);
365 dsl_pool_config_enter(dp, FTAG);
366 if (err = dsl_dataset_hold_obj(dp,
367 za.za_first_integer, FTAG, &ds)) {
368 dsl_pool_config_exit(dp, FTAG);
372 strval = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN,
374 dsl_dataset_name(ds, strval);
375 dsl_dataset_rele(ds, FTAG);
376 dsl_pool_config_exit(dp, FTAG);
379 intval = za.za_first_integer;
382 spa_prop_add_list(*nvp, prop, strval, intval, src);
385 kmem_free(strval, ZFS_MAX_DATASET_NAME_LEN);
390 /* string property */
391 strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
392 err = zap_lookup(mos, spa->spa_pool_props_object,
393 za.za_name, 1, za.za_num_integers, strval);
395 kmem_free(strval, za.za_num_integers);
398 spa_prop_add_list(*nvp, prop, strval, 0, src);
399 kmem_free(strval, za.za_num_integers);
406 zap_cursor_fini(&zc);
407 mutex_exit(&spa->spa_props_lock);
409 if (err && err != ENOENT) {
419 * Validate the given pool properties nvlist and modify the list
420 * for the property values to be set.
423 spa_prop_validate(spa_t *spa, nvlist_t *props)
426 int error = 0, reset_bootfs = 0;
428 boolean_t has_feature = B_FALSE;
431 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
433 char *strval, *slash, *check, *fname;
434 const char *propname = nvpair_name(elem);
435 zpool_prop_t prop = zpool_name_to_prop(propname);
439 if (!zpool_prop_feature(propname)) {
440 error = SET_ERROR(EINVAL);
445 * Sanitize the input.
447 if (nvpair_type(elem) != DATA_TYPE_UINT64) {
448 error = SET_ERROR(EINVAL);
452 if (nvpair_value_uint64(elem, &intval) != 0) {
453 error = SET_ERROR(EINVAL);
458 error = SET_ERROR(EINVAL);
462 fname = strchr(propname, '@') + 1;
463 if (zfeature_lookup_name(fname, NULL) != 0) {
464 error = SET_ERROR(EINVAL);
468 has_feature = B_TRUE;
471 case ZPOOL_PROP_VERSION:
472 error = nvpair_value_uint64(elem, &intval);
474 (intval < spa_version(spa) ||
475 intval > SPA_VERSION_BEFORE_FEATURES ||
477 error = SET_ERROR(EINVAL);
480 case ZPOOL_PROP_DELEGATION:
481 case ZPOOL_PROP_AUTOREPLACE:
482 case ZPOOL_PROP_LISTSNAPS:
483 case ZPOOL_PROP_AUTOEXPAND:
484 error = nvpair_value_uint64(elem, &intval);
485 if (!error && intval > 1)
486 error = SET_ERROR(EINVAL);
489 case ZPOOL_PROP_BOOTFS:
491 * If the pool version is less than SPA_VERSION_BOOTFS,
492 * or the pool is still being created (version == 0),
493 * the bootfs property cannot be set.
495 if (spa_version(spa) < SPA_VERSION_BOOTFS) {
496 error = SET_ERROR(ENOTSUP);
501 * Make sure the vdev config is bootable
503 if (!vdev_is_bootable(spa->spa_root_vdev)) {
504 error = SET_ERROR(ENOTSUP);
510 error = nvpair_value_string(elem, &strval);
516 if (strval == NULL || strval[0] == '\0') {
517 objnum = zpool_prop_default_numeric(
522 if (error = dmu_objset_hold(strval, FTAG, &os))
526 * Must be ZPL, and its property settings
527 * must be supported by GRUB (compression
528 * is not gzip, and large blocks are not used).
531 if (dmu_objset_type(os) != DMU_OST_ZFS) {
532 error = SET_ERROR(ENOTSUP);
534 dsl_prop_get_int_ds(dmu_objset_ds(os),
535 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
537 !BOOTFS_COMPRESS_VALID(propval)) {
538 error = SET_ERROR(ENOTSUP);
540 dsl_prop_get_int_ds(dmu_objset_ds(os),
541 zfs_prop_to_name(ZFS_PROP_RECORDSIZE),
543 propval > SPA_OLD_MAXBLOCKSIZE) {
544 error = SET_ERROR(ENOTSUP);
546 objnum = dmu_objset_id(os);
548 dmu_objset_rele(os, FTAG);
552 case ZPOOL_PROP_FAILUREMODE:
553 error = nvpair_value_uint64(elem, &intval);
554 if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
555 intval > ZIO_FAILURE_MODE_PANIC))
556 error = SET_ERROR(EINVAL);
559 * This is a special case which only occurs when
560 * the pool has completely failed. This allows
561 * the user to change the in-core failmode property
562 * without syncing it out to disk (I/Os might
563 * currently be blocked). We do this by returning
564 * EIO to the caller (spa_prop_set) to trick it
565 * into thinking we encountered a property validation
568 if (!error && spa_suspended(spa)) {
569 spa->spa_failmode = intval;
570 error = SET_ERROR(EIO);
574 case ZPOOL_PROP_CACHEFILE:
575 if ((error = nvpair_value_string(elem, &strval)) != 0)
578 if (strval[0] == '\0')
581 if (strcmp(strval, "none") == 0)
584 if (strval[0] != '/') {
585 error = SET_ERROR(EINVAL);
589 slash = strrchr(strval, '/');
590 ASSERT(slash != NULL);
592 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
593 strcmp(slash, "/..") == 0)
594 error = SET_ERROR(EINVAL);
597 case ZPOOL_PROP_COMMENT:
598 if ((error = nvpair_value_string(elem, &strval)) != 0)
600 for (check = strval; *check != '\0'; check++) {
602 * The kernel doesn't have an easy isprint()
603 * check. For this kernel check, we merely
604 * check ASCII apart from DEL. Fix this if
605 * there is an easy-to-use kernel isprint().
607 if (*check >= 0x7f) {
608 error = SET_ERROR(EINVAL);
612 if (strlen(strval) > ZPROP_MAX_COMMENT)
616 case ZPOOL_PROP_DEDUPDITTO:
617 if (spa_version(spa) < SPA_VERSION_DEDUP)
618 error = SET_ERROR(ENOTSUP);
620 error = nvpair_value_uint64(elem, &intval);
622 intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
623 error = SET_ERROR(EINVAL);
631 if (!error && reset_bootfs) {
632 error = nvlist_remove(props,
633 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
636 error = nvlist_add_uint64(props,
637 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
645 spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
648 spa_config_dirent_t *dp;
650 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
654 dp = kmem_alloc(sizeof (spa_config_dirent_t),
657 if (cachefile[0] == '\0')
658 dp->scd_path = spa_strdup(spa_config_path);
659 else if (strcmp(cachefile, "none") == 0)
662 dp->scd_path = spa_strdup(cachefile);
664 list_insert_head(&spa->spa_config_list, dp);
666 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
670 spa_prop_set(spa_t *spa, nvlist_t *nvp)
673 nvpair_t *elem = NULL;
674 boolean_t need_sync = B_FALSE;
676 if ((error = spa_prop_validate(spa, nvp)) != 0)
679 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
680 zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem));
682 if (prop == ZPOOL_PROP_CACHEFILE ||
683 prop == ZPOOL_PROP_ALTROOT ||
684 prop == ZPOOL_PROP_READONLY)
687 if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) {
690 if (prop == ZPOOL_PROP_VERSION) {
691 VERIFY(nvpair_value_uint64(elem, &ver) == 0);
693 ASSERT(zpool_prop_feature(nvpair_name(elem)));
694 ver = SPA_VERSION_FEATURES;
698 /* Save time if the version is already set. */
699 if (ver == spa_version(spa))
703 * In addition to the pool directory object, we might
704 * create the pool properties object, the features for
705 * read object, the features for write object, or the
706 * feature descriptions object.
708 error = dsl_sync_task(spa->spa_name, NULL,
709 spa_sync_version, &ver,
710 6, ZFS_SPACE_CHECK_RESERVED);
721 return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props,
722 nvp, 6, ZFS_SPACE_CHECK_RESERVED));
729 * If the bootfs property value is dsobj, clear it.
732 spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
734 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
735 VERIFY(zap_remove(spa->spa_meta_objset,
736 spa->spa_pool_props_object,
737 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
744 spa_change_guid_check(void *arg, dmu_tx_t *tx)
746 uint64_t *newguid = arg;
747 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
748 vdev_t *rvd = spa->spa_root_vdev;
751 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
752 vdev_state = rvd->vdev_state;
753 spa_config_exit(spa, SCL_STATE, FTAG);
755 if (vdev_state != VDEV_STATE_HEALTHY)
756 return (SET_ERROR(ENXIO));
758 ASSERT3U(spa_guid(spa), !=, *newguid);
764 spa_change_guid_sync(void *arg, dmu_tx_t *tx)
766 uint64_t *newguid = arg;
767 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
769 vdev_t *rvd = spa->spa_root_vdev;
771 oldguid = spa_guid(spa);
773 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
774 rvd->vdev_guid = *newguid;
775 rvd->vdev_guid_sum += (*newguid - oldguid);
776 vdev_config_dirty(rvd);
777 spa_config_exit(spa, SCL_STATE, FTAG);
779 spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu",
784 * Change the GUID for the pool. This is done so that we can later
785 * re-import a pool built from a clone of our own vdevs. We will modify
786 * the root vdev's guid, our own pool guid, and then mark all of our
787 * vdevs dirty. Note that we must make sure that all our vdevs are
788 * online when we do this, or else any vdevs that weren't present
789 * would be orphaned from our pool. We are also going to issue a
790 * sysevent to update any watchers.
793 spa_change_guid(spa_t *spa)
798 mutex_enter(&spa->spa_vdev_top_lock);
799 mutex_enter(&spa_namespace_lock);
800 guid = spa_generate_guid(NULL);
802 error = dsl_sync_task(spa->spa_name, spa_change_guid_check,
803 spa_change_guid_sync, &guid, 5, ZFS_SPACE_CHECK_RESERVED);
806 spa_config_sync(spa, B_FALSE, B_TRUE);
807 spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID);
810 mutex_exit(&spa_namespace_lock);
811 mutex_exit(&spa->spa_vdev_top_lock);
817 * ==========================================================================
818 * SPA state manipulation (open/create/destroy/import/export)
819 * ==========================================================================
823 spa_error_entry_compare(const void *a, const void *b)
825 spa_error_entry_t *sa = (spa_error_entry_t *)a;
826 spa_error_entry_t *sb = (spa_error_entry_t *)b;
829 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
830 sizeof (zbookmark_phys_t));
841 * Utility function which retrieves copies of the current logs and
842 * re-initializes them in the process.
845 spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
847 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
849 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
850 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
852 avl_create(&spa->spa_errlist_scrub,
853 spa_error_entry_compare, sizeof (spa_error_entry_t),
854 offsetof(spa_error_entry_t, se_avl));
855 avl_create(&spa->spa_errlist_last,
856 spa_error_entry_compare, sizeof (spa_error_entry_t),
857 offsetof(spa_error_entry_t, se_avl));
861 spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
863 const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
864 enum zti_modes mode = ztip->zti_mode;
865 uint_t value = ztip->zti_value;
866 uint_t count = ztip->zti_count;
867 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
870 boolean_t batch = B_FALSE;
872 if (mode == ZTI_MODE_NULL) {
874 tqs->stqs_taskq = NULL;
878 ASSERT3U(count, >, 0);
880 tqs->stqs_count = count;
881 tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP);
885 ASSERT3U(value, >=, 1);
886 value = MAX(value, 1);
891 flags |= TASKQ_THREADS_CPU_PCT;
892 value = zio_taskq_batch_pct;
896 panic("unrecognized mode for %s_%s taskq (%u:%u) in "
898 zio_type_name[t], zio_taskq_types[q], mode, value);
902 for (uint_t i = 0; i < count; i++) {
906 (void) snprintf(name, sizeof (name), "%s_%s_%u",
907 zio_type_name[t], zio_taskq_types[q], i);
909 (void) snprintf(name, sizeof (name), "%s_%s",
910 zio_type_name[t], zio_taskq_types[q]);
914 if (zio_taskq_sysdc && spa->spa_proc != &p0) {
916 flags |= TASKQ_DC_BATCH;
918 tq = taskq_create_sysdc(name, value, 50, INT_MAX,
919 spa->spa_proc, zio_taskq_basedc, flags);
922 pri_t pri = maxclsyspri;
924 * The write issue taskq can be extremely CPU
925 * intensive. Run it at slightly lower priority
926 * than the other taskqs.
928 * - numerically higher priorities are lower priorities;
929 * - if priorities divided by four (RQ_PPQ) are equal
930 * then a difference between them is insignificant.
932 if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE)
939 tq = taskq_create_proc(name, value, pri, 50,
940 INT_MAX, spa->spa_proc, flags);
945 tqs->stqs_taskq[i] = tq;
950 spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
952 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
954 if (tqs->stqs_taskq == NULL) {
955 ASSERT0(tqs->stqs_count);
959 for (uint_t i = 0; i < tqs->stqs_count; i++) {
960 ASSERT3P(tqs->stqs_taskq[i], !=, NULL);
961 taskq_destroy(tqs->stqs_taskq[i]);
964 kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *));
965 tqs->stqs_taskq = NULL;
969 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
970 * Note that a type may have multiple discrete taskqs to avoid lock contention
971 * on the taskq itself. In that case we choose which taskq at random by using
972 * the low bits of gethrtime().
975 spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q,
976 task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent)
978 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
981 ASSERT3P(tqs->stqs_taskq, !=, NULL);
982 ASSERT3U(tqs->stqs_count, !=, 0);
984 if (tqs->stqs_count == 1) {
985 tq = tqs->stqs_taskq[0];
988 tq = tqs->stqs_taskq[cpu_ticks() % tqs->stqs_count];
990 tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count];
994 taskq_dispatch_ent(tq, func, arg, flags, ent);
998 spa_create_zio_taskqs(spa_t *spa)
1000 for (int t = 0; t < ZIO_TYPES; t++) {
1001 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1002 spa_taskqs_init(spa, t, q);
1010 spa_thread(void *arg)
1012 callb_cpr_t cprinfo;
1015 user_t *pu = PTOU(curproc);
1017 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
1020 ASSERT(curproc != &p0);
1021 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
1022 "zpool-%s", spa->spa_name);
1023 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
1026 /* bind this thread to the requested psrset */
1027 if (zio_taskq_psrset_bind != PS_NONE) {
1029 mutex_enter(&cpu_lock);
1030 mutex_enter(&pidlock);
1031 mutex_enter(&curproc->p_lock);
1033 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
1034 0, NULL, NULL) == 0) {
1035 curthread->t_bind_pset = zio_taskq_psrset_bind;
1038 "Couldn't bind process for zfs pool \"%s\" to "
1039 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
1042 mutex_exit(&curproc->p_lock);
1043 mutex_exit(&pidlock);
1044 mutex_exit(&cpu_lock);
1050 if (zio_taskq_sysdc) {
1051 sysdc_thread_enter(curthread, 100, 0);
1055 spa->spa_proc = curproc;
1056 spa->spa_did = curthread->t_did;
1058 spa_create_zio_taskqs(spa);
1060 mutex_enter(&spa->spa_proc_lock);
1061 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
1063 spa->spa_proc_state = SPA_PROC_ACTIVE;
1064 cv_broadcast(&spa->spa_proc_cv);
1066 CALLB_CPR_SAFE_BEGIN(&cprinfo);
1067 while (spa->spa_proc_state == SPA_PROC_ACTIVE)
1068 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1069 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
1071 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
1072 spa->spa_proc_state = SPA_PROC_GONE;
1073 spa->spa_proc = &p0;
1074 cv_broadcast(&spa->spa_proc_cv);
1075 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */
1077 mutex_enter(&curproc->p_lock);
1080 #endif /* SPA_PROCESS */
1084 * Activate an uninitialized pool.
1087 spa_activate(spa_t *spa, int mode)
1089 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
1091 spa->spa_state = POOL_STATE_ACTIVE;
1092 spa->spa_mode = mode;
1094 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
1095 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
1097 /* Try to create a covering process */
1098 mutex_enter(&spa->spa_proc_lock);
1099 ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
1100 ASSERT(spa->spa_proc == &p0);
1104 /* Only create a process if we're going to be around a while. */
1105 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
1106 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
1108 spa->spa_proc_state = SPA_PROC_CREATED;
1109 while (spa->spa_proc_state == SPA_PROC_CREATED) {
1110 cv_wait(&spa->spa_proc_cv,
1111 &spa->spa_proc_lock);
1113 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1114 ASSERT(spa->spa_proc != &p0);
1115 ASSERT(spa->spa_did != 0);
1119 "Couldn't create process for zfs pool \"%s\"\n",
1124 #endif /* SPA_PROCESS */
1125 mutex_exit(&spa->spa_proc_lock);
1127 /* If we didn't create a process, we need to create our taskqs. */
1128 ASSERT(spa->spa_proc == &p0);
1129 if (spa->spa_proc == &p0) {
1130 spa_create_zio_taskqs(spa);
1134 * Start TRIM thread.
1136 trim_thread_create(spa);
1138 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
1139 offsetof(vdev_t, vdev_config_dirty_node));
1140 list_create(&spa->spa_evicting_os_list, sizeof (objset_t),
1141 offsetof(objset_t, os_evicting_node));
1142 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
1143 offsetof(vdev_t, vdev_state_dirty_node));
1145 txg_list_create(&spa->spa_vdev_txg_list,
1146 offsetof(struct vdev, vdev_txg_node));
1148 avl_create(&spa->spa_errlist_scrub,
1149 spa_error_entry_compare, sizeof (spa_error_entry_t),
1150 offsetof(spa_error_entry_t, se_avl));
1151 avl_create(&spa->spa_errlist_last,
1152 spa_error_entry_compare, sizeof (spa_error_entry_t),
1153 offsetof(spa_error_entry_t, se_avl));
1157 * Opposite of spa_activate().
1160 spa_deactivate(spa_t *spa)
1162 ASSERT(spa->spa_sync_on == B_FALSE);
1163 ASSERT(spa->spa_dsl_pool == NULL);
1164 ASSERT(spa->spa_root_vdev == NULL);
1165 ASSERT(spa->spa_async_zio_root == NULL);
1166 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
1169 * Stop TRIM thread in case spa_unload() wasn't called directly
1170 * before spa_deactivate().
1172 trim_thread_destroy(spa);
1174 spa_evicting_os_wait(spa);
1176 txg_list_destroy(&spa->spa_vdev_txg_list);
1178 list_destroy(&spa->spa_config_dirty_list);
1179 list_destroy(&spa->spa_evicting_os_list);
1180 list_destroy(&spa->spa_state_dirty_list);
1182 for (int t = 0; t < ZIO_TYPES; t++) {
1183 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1184 spa_taskqs_fini(spa, t, q);
1188 metaslab_class_destroy(spa->spa_normal_class);
1189 spa->spa_normal_class = NULL;
1191 metaslab_class_destroy(spa->spa_log_class);
1192 spa->spa_log_class = NULL;
1195 * If this was part of an import or the open otherwise failed, we may
1196 * still have errors left in the queues. Empty them just in case.
1198 spa_errlog_drain(spa);
1200 avl_destroy(&spa->spa_errlist_scrub);
1201 avl_destroy(&spa->spa_errlist_last);
1203 spa->spa_state = POOL_STATE_UNINITIALIZED;
1205 mutex_enter(&spa->spa_proc_lock);
1206 if (spa->spa_proc_state != SPA_PROC_NONE) {
1207 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1208 spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1209 cv_broadcast(&spa->spa_proc_cv);
1210 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1211 ASSERT(spa->spa_proc != &p0);
1212 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1214 ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1215 spa->spa_proc_state = SPA_PROC_NONE;
1217 ASSERT(spa->spa_proc == &p0);
1218 mutex_exit(&spa->spa_proc_lock);
1222 * We want to make sure spa_thread() has actually exited the ZFS
1223 * module, so that the module can't be unloaded out from underneath
1226 if (spa->spa_did != 0) {
1227 thread_join(spa->spa_did);
1230 #endif /* SPA_PROCESS */
1234 * Verify a pool configuration, and construct the vdev tree appropriately. This
1235 * will create all the necessary vdevs in the appropriate layout, with each vdev
1236 * in the CLOSED state. This will prep the pool before open/creation/import.
1237 * All vdev validation is done by the vdev_alloc() routine.
1240 spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1241 uint_t id, int atype)
1247 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1250 if ((*vdp)->vdev_ops->vdev_op_leaf)
1253 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1256 if (error == ENOENT)
1262 return (SET_ERROR(EINVAL));
1265 for (int c = 0; c < children; c++) {
1267 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1275 ASSERT(*vdp != NULL);
1281 * Opposite of spa_load().
1284 spa_unload(spa_t *spa)
1288 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1293 trim_thread_destroy(spa);
1298 spa_async_suspend(spa);
1303 if (spa->spa_sync_on) {
1304 txg_sync_stop(spa->spa_dsl_pool);
1305 spa->spa_sync_on = B_FALSE;
1309 * Wait for any outstanding async I/O to complete.
1311 if (spa->spa_async_zio_root != NULL) {
1312 for (int i = 0; i < max_ncpus; i++)
1313 (void) zio_wait(spa->spa_async_zio_root[i]);
1314 kmem_free(spa->spa_async_zio_root, max_ncpus * sizeof (void *));
1315 spa->spa_async_zio_root = NULL;
1318 bpobj_close(&spa->spa_deferred_bpobj);
1320 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1325 if (spa->spa_root_vdev)
1326 vdev_free(spa->spa_root_vdev);
1327 ASSERT(spa->spa_root_vdev == NULL);
1330 * Close the dsl pool.
1332 if (spa->spa_dsl_pool) {
1333 dsl_pool_close(spa->spa_dsl_pool);
1334 spa->spa_dsl_pool = NULL;
1335 spa->spa_meta_objset = NULL;
1341 * Drop and purge level 2 cache
1343 spa_l2cache_drop(spa);
1345 for (i = 0; i < spa->spa_spares.sav_count; i++)
1346 vdev_free(spa->spa_spares.sav_vdevs[i]);
1347 if (spa->spa_spares.sav_vdevs) {
1348 kmem_free(spa->spa_spares.sav_vdevs,
1349 spa->spa_spares.sav_count * sizeof (void *));
1350 spa->spa_spares.sav_vdevs = NULL;
1352 if (spa->spa_spares.sav_config) {
1353 nvlist_free(spa->spa_spares.sav_config);
1354 spa->spa_spares.sav_config = NULL;
1356 spa->spa_spares.sav_count = 0;
1358 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1359 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1360 vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1362 if (spa->spa_l2cache.sav_vdevs) {
1363 kmem_free(spa->spa_l2cache.sav_vdevs,
1364 spa->spa_l2cache.sav_count * sizeof (void *));
1365 spa->spa_l2cache.sav_vdevs = NULL;
1367 if (spa->spa_l2cache.sav_config) {
1368 nvlist_free(spa->spa_l2cache.sav_config);
1369 spa->spa_l2cache.sav_config = NULL;
1371 spa->spa_l2cache.sav_count = 0;
1373 spa->spa_async_suspended = 0;
1375 if (spa->spa_comment != NULL) {
1376 spa_strfree(spa->spa_comment);
1377 spa->spa_comment = NULL;
1380 spa_config_exit(spa, SCL_ALL, FTAG);
1384 * Load (or re-load) the current list of vdevs describing the active spares for
1385 * this pool. When this is called, we have some form of basic information in
1386 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1387 * then re-generate a more complete list including status information.
1390 spa_load_spares(spa_t *spa)
1397 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1400 * First, close and free any existing spare vdevs.
1402 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1403 vd = spa->spa_spares.sav_vdevs[i];
1405 /* Undo the call to spa_activate() below */
1406 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1407 B_FALSE)) != NULL && tvd->vdev_isspare)
1408 spa_spare_remove(tvd);
1413 if (spa->spa_spares.sav_vdevs)
1414 kmem_free(spa->spa_spares.sav_vdevs,
1415 spa->spa_spares.sav_count * sizeof (void *));
1417 if (spa->spa_spares.sav_config == NULL)
1420 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1421 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1423 spa->spa_spares.sav_count = (int)nspares;
1424 spa->spa_spares.sav_vdevs = NULL;
1430 * Construct the array of vdevs, opening them to get status in the
1431 * process. For each spare, there is potentially two different vdev_t
1432 * structures associated with it: one in the list of spares (used only
1433 * for basic validation purposes) and one in the active vdev
1434 * configuration (if it's spared in). During this phase we open and
1435 * validate each vdev on the spare list. If the vdev also exists in the
1436 * active configuration, then we also mark this vdev as an active spare.
1438 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1440 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1441 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1442 VDEV_ALLOC_SPARE) == 0);
1445 spa->spa_spares.sav_vdevs[i] = vd;
1447 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1448 B_FALSE)) != NULL) {
1449 if (!tvd->vdev_isspare)
1453 * We only mark the spare active if we were successfully
1454 * able to load the vdev. Otherwise, importing a pool
1455 * with a bad active spare would result in strange
1456 * behavior, because multiple pool would think the spare
1457 * is actively in use.
1459 * There is a vulnerability here to an equally bizarre
1460 * circumstance, where a dead active spare is later
1461 * brought back to life (onlined or otherwise). Given
1462 * the rarity of this scenario, and the extra complexity
1463 * it adds, we ignore the possibility.
1465 if (!vdev_is_dead(tvd))
1466 spa_spare_activate(tvd);
1470 vd->vdev_aux = &spa->spa_spares;
1472 if (vdev_open(vd) != 0)
1475 if (vdev_validate_aux(vd) == 0)
1480 * Recompute the stashed list of spares, with status information
1483 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1484 DATA_TYPE_NVLIST_ARRAY) == 0);
1486 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1488 for (i = 0; i < spa->spa_spares.sav_count; i++)
1489 spares[i] = vdev_config_generate(spa,
1490 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1491 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1492 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1493 for (i = 0; i < spa->spa_spares.sav_count; i++)
1494 nvlist_free(spares[i]);
1495 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1499 * Load (or re-load) the current list of vdevs describing the active l2cache for
1500 * this pool. When this is called, we have some form of basic information in
1501 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1502 * then re-generate a more complete list including status information.
1503 * Devices which are already active have their details maintained, and are
1507 spa_load_l2cache(spa_t *spa)
1511 int i, j, oldnvdevs;
1513 vdev_t *vd, **oldvdevs, **newvdevs;
1514 spa_aux_vdev_t *sav = &spa->spa_l2cache;
1516 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1518 if (sav->sav_config != NULL) {
1519 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1520 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1521 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1527 oldvdevs = sav->sav_vdevs;
1528 oldnvdevs = sav->sav_count;
1529 sav->sav_vdevs = NULL;
1533 * Process new nvlist of vdevs.
1535 for (i = 0; i < nl2cache; i++) {
1536 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1540 for (j = 0; j < oldnvdevs; j++) {
1542 if (vd != NULL && guid == vd->vdev_guid) {
1544 * Retain previous vdev for add/remove ops.
1552 if (newvdevs[i] == NULL) {
1556 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1557 VDEV_ALLOC_L2CACHE) == 0);
1562 * Commit this vdev as an l2cache device,
1563 * even if it fails to open.
1565 spa_l2cache_add(vd);
1570 spa_l2cache_activate(vd);
1572 if (vdev_open(vd) != 0)
1575 (void) vdev_validate_aux(vd);
1577 if (!vdev_is_dead(vd))
1578 l2arc_add_vdev(spa, vd);
1583 * Purge vdevs that were dropped
1585 for (i = 0; i < oldnvdevs; i++) {
1590 ASSERT(vd->vdev_isl2cache);
1592 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1593 pool != 0ULL && l2arc_vdev_present(vd))
1594 l2arc_remove_vdev(vd);
1595 vdev_clear_stats(vd);
1601 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1603 if (sav->sav_config == NULL)
1606 sav->sav_vdevs = newvdevs;
1607 sav->sav_count = (int)nl2cache;
1610 * Recompute the stashed list of l2cache devices, with status
1611 * information this time.
1613 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1614 DATA_TYPE_NVLIST_ARRAY) == 0);
1616 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1617 for (i = 0; i < sav->sav_count; i++)
1618 l2cache[i] = vdev_config_generate(spa,
1619 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1620 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1621 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1623 for (i = 0; i < sav->sav_count; i++)
1624 nvlist_free(l2cache[i]);
1626 kmem_free(l2cache, sav->sav_count * sizeof (void *));
1630 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1633 char *packed = NULL;
1638 error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db);
1642 nvsize = *(uint64_t *)db->db_data;
1643 dmu_buf_rele(db, FTAG);
1645 packed = kmem_alloc(nvsize, KM_SLEEP);
1646 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1649 error = nvlist_unpack(packed, nvsize, value, 0);
1650 kmem_free(packed, nvsize);
1656 * Checks to see if the given vdev could not be opened, in which case we post a
1657 * sysevent to notify the autoreplace code that the device has been removed.
1660 spa_check_removed(vdev_t *vd)
1662 for (int c = 0; c < vd->vdev_children; c++)
1663 spa_check_removed(vd->vdev_child[c]);
1665 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) &&
1667 zfs_post_autoreplace(vd->vdev_spa, vd);
1668 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1673 * Validate the current config against the MOS config
1676 spa_config_valid(spa_t *spa, nvlist_t *config)
1678 vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1681 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1683 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1684 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1686 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1689 * If we're doing a normal import, then build up any additional
1690 * diagnostic information about missing devices in this config.
1691 * We'll pass this up to the user for further processing.
1693 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1694 nvlist_t **child, *nv;
1697 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1699 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1701 for (int c = 0; c < rvd->vdev_children; c++) {
1702 vdev_t *tvd = rvd->vdev_child[c];
1703 vdev_t *mtvd = mrvd->vdev_child[c];
1705 if (tvd->vdev_ops == &vdev_missing_ops &&
1706 mtvd->vdev_ops != &vdev_missing_ops &&
1708 child[idx++] = vdev_config_generate(spa, mtvd,
1713 VERIFY(nvlist_add_nvlist_array(nv,
1714 ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1715 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1716 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1718 for (int i = 0; i < idx; i++)
1719 nvlist_free(child[i]);
1722 kmem_free(child, rvd->vdev_children * sizeof (char **));
1726 * Compare the root vdev tree with the information we have
1727 * from the MOS config (mrvd). Check each top-level vdev
1728 * with the corresponding MOS config top-level (mtvd).
1730 for (int c = 0; c < rvd->vdev_children; c++) {
1731 vdev_t *tvd = rvd->vdev_child[c];
1732 vdev_t *mtvd = mrvd->vdev_child[c];
1735 * Resolve any "missing" vdevs in the current configuration.
1736 * If we find that the MOS config has more accurate information
1737 * about the top-level vdev then use that vdev instead.
1739 if (tvd->vdev_ops == &vdev_missing_ops &&
1740 mtvd->vdev_ops != &vdev_missing_ops) {
1742 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1746 * Device specific actions.
1748 if (mtvd->vdev_islog) {
1749 spa_set_log_state(spa, SPA_LOG_CLEAR);
1752 * XXX - once we have 'readonly' pool
1753 * support we should be able to handle
1754 * missing data devices by transitioning
1755 * the pool to readonly.
1761 * Swap the missing vdev with the data we were
1762 * able to obtain from the MOS config.
1764 vdev_remove_child(rvd, tvd);
1765 vdev_remove_child(mrvd, mtvd);
1767 vdev_add_child(rvd, mtvd);
1768 vdev_add_child(mrvd, tvd);
1770 spa_config_exit(spa, SCL_ALL, FTAG);
1772 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1775 } else if (mtvd->vdev_islog) {
1777 * Load the slog device's state from the MOS config
1778 * since it's possible that the label does not
1779 * contain the most up-to-date information.
1781 vdev_load_log_state(tvd, mtvd);
1786 spa_config_exit(spa, SCL_ALL, FTAG);
1789 * Ensure we were able to validate the config.
1791 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1795 * Check for missing log devices
1798 spa_check_logs(spa_t *spa)
1800 boolean_t rv = B_FALSE;
1801 dsl_pool_t *dp = spa_get_dsl(spa);
1803 switch (spa->spa_log_state) {
1804 case SPA_LOG_MISSING:
1805 /* need to recheck in case slog has been restored */
1806 case SPA_LOG_UNKNOWN:
1807 rv = (dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
1808 zil_check_log_chain, NULL, DS_FIND_CHILDREN) != 0);
1810 spa_set_log_state(spa, SPA_LOG_MISSING);
1817 spa_passivate_log(spa_t *spa)
1819 vdev_t *rvd = spa->spa_root_vdev;
1820 boolean_t slog_found = B_FALSE;
1822 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1824 if (!spa_has_slogs(spa))
1827 for (int c = 0; c < rvd->vdev_children; c++) {
1828 vdev_t *tvd = rvd->vdev_child[c];
1829 metaslab_group_t *mg = tvd->vdev_mg;
1831 if (tvd->vdev_islog) {
1832 metaslab_group_passivate(mg);
1833 slog_found = B_TRUE;
1837 return (slog_found);
1841 spa_activate_log(spa_t *spa)
1843 vdev_t *rvd = spa->spa_root_vdev;
1845 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1847 for (int c = 0; c < rvd->vdev_children; c++) {
1848 vdev_t *tvd = rvd->vdev_child[c];
1849 metaslab_group_t *mg = tvd->vdev_mg;
1851 if (tvd->vdev_islog)
1852 metaslab_group_activate(mg);
1857 spa_offline_log(spa_t *spa)
1861 error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1862 NULL, DS_FIND_CHILDREN);
1865 * We successfully offlined the log device, sync out the
1866 * current txg so that the "stubby" block can be removed
1869 txg_wait_synced(spa->spa_dsl_pool, 0);
1875 spa_aux_check_removed(spa_aux_vdev_t *sav)
1879 for (i = 0; i < sav->sav_count; i++)
1880 spa_check_removed(sav->sav_vdevs[i]);
1884 spa_claim_notify(zio_t *zio)
1886 spa_t *spa = zio->io_spa;
1891 mutex_enter(&spa->spa_props_lock); /* any mutex will do */
1892 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1893 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1894 mutex_exit(&spa->spa_props_lock);
1897 typedef struct spa_load_error {
1898 uint64_t sle_meta_count;
1899 uint64_t sle_data_count;
1903 spa_load_verify_done(zio_t *zio)
1905 blkptr_t *bp = zio->io_bp;
1906 spa_load_error_t *sle = zio->io_private;
1907 dmu_object_type_t type = BP_GET_TYPE(bp);
1908 int error = zio->io_error;
1909 spa_t *spa = zio->io_spa;
1912 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1913 type != DMU_OT_INTENT_LOG)
1914 atomic_inc_64(&sle->sle_meta_count);
1916 atomic_inc_64(&sle->sle_data_count);
1918 zio_data_buf_free(zio->io_data, zio->io_size);
1920 mutex_enter(&spa->spa_scrub_lock);
1921 spa->spa_scrub_inflight--;
1922 cv_broadcast(&spa->spa_scrub_io_cv);
1923 mutex_exit(&spa->spa_scrub_lock);
1927 * Maximum number of concurrent scrub i/os to create while verifying
1928 * a pool while importing it.
1930 int spa_load_verify_maxinflight = 10000;
1931 boolean_t spa_load_verify_metadata = B_TRUE;
1932 boolean_t spa_load_verify_data = B_TRUE;
1934 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_maxinflight, CTLFLAG_RWTUN,
1935 &spa_load_verify_maxinflight, 0,
1936 "Maximum number of concurrent scrub I/Os to create while verifying a "
1937 "pool while importing it");
1939 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_metadata, CTLFLAG_RWTUN,
1940 &spa_load_verify_metadata, 0,
1941 "Check metadata on import?");
1943 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_data, CTLFLAG_RWTUN,
1944 &spa_load_verify_data, 0,
1945 "Check user data on import?");
1949 spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1950 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
1952 if (bp == NULL || BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp))
1955 * Note: normally this routine will not be called if
1956 * spa_load_verify_metadata is not set. However, it may be useful
1957 * to manually set the flag after the traversal has begun.
1959 if (!spa_load_verify_metadata)
1961 if (BP_GET_BUFC_TYPE(bp) == ARC_BUFC_DATA && !spa_load_verify_data)
1965 size_t size = BP_GET_PSIZE(bp);
1966 void *data = zio_data_buf_alloc(size);
1968 mutex_enter(&spa->spa_scrub_lock);
1969 while (spa->spa_scrub_inflight >= spa_load_verify_maxinflight)
1970 cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
1971 spa->spa_scrub_inflight++;
1972 mutex_exit(&spa->spa_scrub_lock);
1974 zio_nowait(zio_read(rio, spa, bp, data, size,
1975 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1976 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1977 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1983 verify_dataset_name_len(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg)
1985 if (dsl_dataset_namelen(ds) >= ZFS_MAX_DATASET_NAME_LEN)
1986 return (SET_ERROR(ENAMETOOLONG));
1992 spa_load_verify(spa_t *spa)
1995 spa_load_error_t sle = { 0 };
1996 zpool_rewind_policy_t policy;
1997 boolean_t verify_ok = B_FALSE;
2000 zpool_get_rewind_policy(spa->spa_config, &policy);
2002 if (policy.zrp_request & ZPOOL_NEVER_REWIND)
2005 dsl_pool_config_enter(spa->spa_dsl_pool, FTAG);
2006 error = dmu_objset_find_dp(spa->spa_dsl_pool,
2007 spa->spa_dsl_pool->dp_root_dir_obj, verify_dataset_name_len, NULL,
2009 dsl_pool_config_exit(spa->spa_dsl_pool, FTAG);
2013 rio = zio_root(spa, NULL, &sle,
2014 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
2016 if (spa_load_verify_metadata) {
2017 error = traverse_pool(spa, spa->spa_verify_min_txg,
2018 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA,
2019 spa_load_verify_cb, rio);
2022 (void) zio_wait(rio);
2024 spa->spa_load_meta_errors = sle.sle_meta_count;
2025 spa->spa_load_data_errors = sle.sle_data_count;
2027 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
2028 sle.sle_data_count <= policy.zrp_maxdata) {
2032 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
2033 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
2035 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
2036 VERIFY(nvlist_add_uint64(spa->spa_load_info,
2037 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
2038 VERIFY(nvlist_add_int64(spa->spa_load_info,
2039 ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
2040 VERIFY(nvlist_add_uint64(spa->spa_load_info,
2041 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
2043 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
2047 if (error != ENXIO && error != EIO)
2048 error = SET_ERROR(EIO);
2052 return (verify_ok ? 0 : EIO);
2056 * Find a value in the pool props object.
2059 spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
2061 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
2062 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
2066 * Find a value in the pool directory object.
2069 spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
2071 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
2072 name, sizeof (uint64_t), 1, val));
2076 spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
2078 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
2083 * Fix up config after a partly-completed split. This is done with the
2084 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
2085 * pool have that entry in their config, but only the splitting one contains
2086 * a list of all the guids of the vdevs that are being split off.
2088 * This function determines what to do with that list: either rejoin
2089 * all the disks to the pool, or complete the splitting process. To attempt
2090 * the rejoin, each disk that is offlined is marked online again, and
2091 * we do a reopen() call. If the vdev label for every disk that was
2092 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
2093 * then we call vdev_split() on each disk, and complete the split.
2095 * Otherwise we leave the config alone, with all the vdevs in place in
2096 * the original pool.
2099 spa_try_repair(spa_t *spa, nvlist_t *config)
2106 boolean_t attempt_reopen;
2108 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
2111 /* check that the config is complete */
2112 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
2113 &glist, &gcount) != 0)
2116 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
2118 /* attempt to online all the vdevs & validate */
2119 attempt_reopen = B_TRUE;
2120 for (i = 0; i < gcount; i++) {
2121 if (glist[i] == 0) /* vdev is hole */
2124 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
2125 if (vd[i] == NULL) {
2127 * Don't bother attempting to reopen the disks;
2128 * just do the split.
2130 attempt_reopen = B_FALSE;
2132 /* attempt to re-online it */
2133 vd[i]->vdev_offline = B_FALSE;
2137 if (attempt_reopen) {
2138 vdev_reopen(spa->spa_root_vdev);
2140 /* check each device to see what state it's in */
2141 for (extracted = 0, i = 0; i < gcount; i++) {
2142 if (vd[i] != NULL &&
2143 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
2150 * If every disk has been moved to the new pool, or if we never
2151 * even attempted to look at them, then we split them off for
2154 if (!attempt_reopen || gcount == extracted) {
2155 for (i = 0; i < gcount; i++)
2158 vdev_reopen(spa->spa_root_vdev);
2161 kmem_free(vd, gcount * sizeof (vdev_t *));
2165 spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
2166 boolean_t mosconfig)
2168 nvlist_t *config = spa->spa_config;
2169 char *ereport = FM_EREPORT_ZFS_POOL;
2175 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
2176 return (SET_ERROR(EINVAL));
2178 ASSERT(spa->spa_comment == NULL);
2179 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
2180 spa->spa_comment = spa_strdup(comment);
2183 * Versioning wasn't explicitly added to the label until later, so if
2184 * it's not present treat it as the initial version.
2186 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
2187 &spa->spa_ubsync.ub_version) != 0)
2188 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
2190 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
2191 &spa->spa_config_txg);
2193 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
2194 spa_guid_exists(pool_guid, 0)) {
2195 error = SET_ERROR(EEXIST);
2197 spa->spa_config_guid = pool_guid;
2199 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
2201 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
2205 nvlist_free(spa->spa_load_info);
2206 spa->spa_load_info = fnvlist_alloc();
2208 gethrestime(&spa->spa_loaded_ts);
2209 error = spa_load_impl(spa, pool_guid, config, state, type,
2210 mosconfig, &ereport);
2214 * Don't count references from objsets that are already closed
2215 * and are making their way through the eviction process.
2217 spa_evicting_os_wait(spa);
2218 spa->spa_minref = refcount_count(&spa->spa_refcount);
2220 if (error != EEXIST) {
2221 spa->spa_loaded_ts.tv_sec = 0;
2222 spa->spa_loaded_ts.tv_nsec = 0;
2224 if (error != EBADF) {
2225 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
2228 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2235 * Load an existing storage pool, using the pool's builtin spa_config as a
2236 * source of configuration information.
2239 spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
2240 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
2244 nvlist_t *nvroot = NULL;
2247 uberblock_t *ub = &spa->spa_uberblock;
2248 uint64_t children, config_cache_txg = spa->spa_config_txg;
2249 int orig_mode = spa->spa_mode;
2252 boolean_t missing_feat_write = B_FALSE;
2255 * If this is an untrusted config, access the pool in read-only mode.
2256 * This prevents things like resilvering recently removed devices.
2259 spa->spa_mode = FREAD;
2261 ASSERT(MUTEX_HELD(&spa_namespace_lock));
2263 spa->spa_load_state = state;
2265 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
2266 return (SET_ERROR(EINVAL));
2268 parse = (type == SPA_IMPORT_EXISTING ?
2269 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2272 * Create "The Godfather" zio to hold all async IOs
2274 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
2276 for (int i = 0; i < max_ncpus; i++) {
2277 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
2278 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
2279 ZIO_FLAG_GODFATHER);
2283 * Parse the configuration into a vdev tree. We explicitly set the
2284 * value that will be returned by spa_version() since parsing the
2285 * configuration requires knowing the version number.
2287 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2288 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
2289 spa_config_exit(spa, SCL_ALL, FTAG);
2294 ASSERT(spa->spa_root_vdev == rvd);
2295 ASSERT3U(spa->spa_min_ashift, >=, SPA_MINBLOCKSHIFT);
2296 ASSERT3U(spa->spa_max_ashift, <=, SPA_MAXBLOCKSHIFT);
2298 if (type != SPA_IMPORT_ASSEMBLE) {
2299 ASSERT(spa_guid(spa) == pool_guid);
2303 * Try to open all vdevs, loading each label in the process.
2305 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2306 error = vdev_open(rvd);
2307 spa_config_exit(spa, SCL_ALL, FTAG);
2312 * We need to validate the vdev labels against the configuration that
2313 * we have in hand, which is dependent on the setting of mosconfig. If
2314 * mosconfig is true then we're validating the vdev labels based on
2315 * that config. Otherwise, we're validating against the cached config
2316 * (zpool.cache) that was read when we loaded the zfs module, and then
2317 * later we will recursively call spa_load() and validate against
2320 * If we're assembling a new pool that's been split off from an
2321 * existing pool, the labels haven't yet been updated so we skip
2322 * validation for now.
2324 if (type != SPA_IMPORT_ASSEMBLE) {
2325 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2326 error = vdev_validate(rvd, mosconfig);
2327 spa_config_exit(spa, SCL_ALL, FTAG);
2332 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2333 return (SET_ERROR(ENXIO));
2337 * Find the best uberblock.
2339 vdev_uberblock_load(rvd, ub, &label);
2342 * If we weren't able to find a single valid uberblock, return failure.
2344 if (ub->ub_txg == 0) {
2346 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2350 * If the pool has an unsupported version we can't open it.
2352 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2354 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2357 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2361 * If we weren't able to find what's necessary for reading the
2362 * MOS in the label, return failure.
2364 if (label == NULL || nvlist_lookup_nvlist(label,
2365 ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) {
2367 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2372 * Update our in-core representation with the definitive values
2375 nvlist_free(spa->spa_label_features);
2376 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2382 * Look through entries in the label nvlist's features_for_read. If
2383 * there is a feature listed there which we don't understand then we
2384 * cannot open a pool.
2386 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2387 nvlist_t *unsup_feat;
2389 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2392 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2394 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2395 if (!zfeature_is_supported(nvpair_name(nvp))) {
2396 VERIFY(nvlist_add_string(unsup_feat,
2397 nvpair_name(nvp), "") == 0);
2401 if (!nvlist_empty(unsup_feat)) {
2402 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2403 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2404 nvlist_free(unsup_feat);
2405 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2409 nvlist_free(unsup_feat);
2413 * If the vdev guid sum doesn't match the uberblock, we have an
2414 * incomplete configuration. We first check to see if the pool
2415 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2416 * If it is, defer the vdev_guid_sum check till later so we
2417 * can handle missing vdevs.
2419 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
2420 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
2421 rvd->vdev_guid_sum != ub->ub_guid_sum)
2422 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2424 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2425 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2426 spa_try_repair(spa, config);
2427 spa_config_exit(spa, SCL_ALL, FTAG);
2428 nvlist_free(spa->spa_config_splitting);
2429 spa->spa_config_splitting = NULL;
2433 * Initialize internal SPA structures.
2435 spa->spa_state = POOL_STATE_ACTIVE;
2436 spa->spa_ubsync = spa->spa_uberblock;
2437 spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2438 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2439 spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2440 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2441 spa->spa_claim_max_txg = spa->spa_first_txg;
2442 spa->spa_prev_software_version = ub->ub_software_version;
2444 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2446 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2447 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2449 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2450 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2452 if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2453 boolean_t missing_feat_read = B_FALSE;
2454 nvlist_t *unsup_feat, *enabled_feat;
2456 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2457 &spa->spa_feat_for_read_obj) != 0) {
2458 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2461 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2462 &spa->spa_feat_for_write_obj) != 0) {
2463 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2466 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2467 &spa->spa_feat_desc_obj) != 0) {
2468 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2471 enabled_feat = fnvlist_alloc();
2472 unsup_feat = fnvlist_alloc();
2474 if (!spa_features_check(spa, B_FALSE,
2475 unsup_feat, enabled_feat))
2476 missing_feat_read = B_TRUE;
2478 if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) {
2479 if (!spa_features_check(spa, B_TRUE,
2480 unsup_feat, enabled_feat)) {
2481 missing_feat_write = B_TRUE;
2485 fnvlist_add_nvlist(spa->spa_load_info,
2486 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
2488 if (!nvlist_empty(unsup_feat)) {
2489 fnvlist_add_nvlist(spa->spa_load_info,
2490 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
2493 fnvlist_free(enabled_feat);
2494 fnvlist_free(unsup_feat);
2496 if (!missing_feat_read) {
2497 fnvlist_add_boolean(spa->spa_load_info,
2498 ZPOOL_CONFIG_CAN_RDONLY);
2502 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2503 * twofold: to determine whether the pool is available for
2504 * import in read-write mode and (if it is not) whether the
2505 * pool is available for import in read-only mode. If the pool
2506 * is available for import in read-write mode, it is displayed
2507 * as available in userland; if it is not available for import
2508 * in read-only mode, it is displayed as unavailable in
2509 * userland. If the pool is available for import in read-only
2510 * mode but not read-write mode, it is displayed as unavailable
2511 * in userland with a special note that the pool is actually
2512 * available for open in read-only mode.
2514 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2515 * missing a feature for write, we must first determine whether
2516 * the pool can be opened read-only before returning to
2517 * userland in order to know whether to display the
2518 * abovementioned note.
2520 if (missing_feat_read || (missing_feat_write &&
2521 spa_writeable(spa))) {
2522 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2527 * Load refcounts for ZFS features from disk into an in-memory
2528 * cache during SPA initialization.
2530 for (spa_feature_t i = 0; i < SPA_FEATURES; i++) {
2533 error = feature_get_refcount_from_disk(spa,
2534 &spa_feature_table[i], &refcount);
2536 spa->spa_feat_refcount_cache[i] = refcount;
2537 } else if (error == ENOTSUP) {
2538 spa->spa_feat_refcount_cache[i] =
2539 SPA_FEATURE_DISABLED;
2541 return (spa_vdev_err(rvd,
2542 VDEV_AUX_CORRUPT_DATA, EIO));
2547 if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) {
2548 if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG,
2549 &spa->spa_feat_enabled_txg_obj) != 0)
2550 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2553 spa->spa_is_initializing = B_TRUE;
2554 error = dsl_pool_open(spa->spa_dsl_pool);
2555 spa->spa_is_initializing = B_FALSE;
2557 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2561 nvlist_t *policy = NULL, *nvconfig;
2563 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2564 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2566 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2567 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2569 unsigned long myhostid = 0;
2571 VERIFY(nvlist_lookup_string(nvconfig,
2572 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2575 myhostid = zone_get_hostid(NULL);
2578 * We're emulating the system's hostid in userland, so
2579 * we can't use zone_get_hostid().
2581 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2582 #endif /* _KERNEL */
2583 if (check_hostid && hostid != 0 && myhostid != 0 &&
2584 hostid != myhostid) {
2585 nvlist_free(nvconfig);
2586 cmn_err(CE_WARN, "pool '%s' could not be "
2587 "loaded as it was last accessed by "
2588 "another system (host: %s hostid: 0x%lx). "
2589 "See: http://illumos.org/msg/ZFS-8000-EY",
2590 spa_name(spa), hostname,
2591 (unsigned long)hostid);
2592 return (SET_ERROR(EBADF));
2595 if (nvlist_lookup_nvlist(spa->spa_config,
2596 ZPOOL_REWIND_POLICY, &policy) == 0)
2597 VERIFY(nvlist_add_nvlist(nvconfig,
2598 ZPOOL_REWIND_POLICY, policy) == 0);
2600 spa_config_set(spa, nvconfig);
2602 spa_deactivate(spa);
2603 spa_activate(spa, orig_mode);
2605 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2608 /* Grab the secret checksum salt from the MOS. */
2609 error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
2610 DMU_POOL_CHECKSUM_SALT, 1,
2611 sizeof (spa->spa_cksum_salt.zcs_bytes),
2612 spa->spa_cksum_salt.zcs_bytes);
2613 if (error == ENOENT) {
2614 /* Generate a new salt for subsequent use */
2615 (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes,
2616 sizeof (spa->spa_cksum_salt.zcs_bytes));
2617 } else if (error != 0) {
2618 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2621 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2622 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2623 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2625 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2628 * Load the bit that tells us to use the new accounting function
2629 * (raid-z deflation). If we have an older pool, this will not
2632 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2633 if (error != 0 && error != ENOENT)
2634 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2636 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2637 &spa->spa_creation_version);
2638 if (error != 0 && error != ENOENT)
2639 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2642 * Load the persistent error log. If we have an older pool, this will
2645 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2646 if (error != 0 && error != ENOENT)
2647 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2649 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2650 &spa->spa_errlog_scrub);
2651 if (error != 0 && error != ENOENT)
2652 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2655 * Load the history object. If we have an older pool, this
2656 * will not be present.
2658 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2659 if (error != 0 && error != ENOENT)
2660 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2663 * If we're assembling the pool from the split-off vdevs of
2664 * an existing pool, we don't want to attach the spares & cache
2669 * Load any hot spares for this pool.
2671 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2672 if (error != 0 && error != ENOENT)
2673 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2674 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2675 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2676 if (load_nvlist(spa, spa->spa_spares.sav_object,
2677 &spa->spa_spares.sav_config) != 0)
2678 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2680 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2681 spa_load_spares(spa);
2682 spa_config_exit(spa, SCL_ALL, FTAG);
2683 } else if (error == 0) {
2684 spa->spa_spares.sav_sync = B_TRUE;
2688 * Load any level 2 ARC devices for this pool.
2690 error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2691 &spa->spa_l2cache.sav_object);
2692 if (error != 0 && error != ENOENT)
2693 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2694 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2695 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2696 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2697 &spa->spa_l2cache.sav_config) != 0)
2698 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2700 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2701 spa_load_l2cache(spa);
2702 spa_config_exit(spa, SCL_ALL, FTAG);
2703 } else if (error == 0) {
2704 spa->spa_l2cache.sav_sync = B_TRUE;
2707 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2709 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2710 if (error && error != ENOENT)
2711 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2714 uint64_t autoreplace;
2716 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2717 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2718 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2719 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2720 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2721 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2722 &spa->spa_dedup_ditto);
2724 spa->spa_autoreplace = (autoreplace != 0);
2728 * If the 'autoreplace' property is set, then post a resource notifying
2729 * the ZFS DE that it should not issue any faults for unopenable
2730 * devices. We also iterate over the vdevs, and post a sysevent for any
2731 * unopenable vdevs so that the normal autoreplace handler can take
2734 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2735 spa_check_removed(spa->spa_root_vdev);
2737 * For the import case, this is done in spa_import(), because
2738 * at this point we're using the spare definitions from
2739 * the MOS config, not necessarily from the userland config.
2741 if (state != SPA_LOAD_IMPORT) {
2742 spa_aux_check_removed(&spa->spa_spares);
2743 spa_aux_check_removed(&spa->spa_l2cache);
2748 * Load the vdev state for all toplevel vdevs.
2753 * Propagate the leaf DTLs we just loaded all the way up the tree.
2755 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2756 vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2757 spa_config_exit(spa, SCL_ALL, FTAG);
2760 * Load the DDTs (dedup tables).
2762 error = ddt_load(spa);
2764 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2766 spa_update_dspace(spa);
2769 * Validate the config, using the MOS config to fill in any
2770 * information which might be missing. If we fail to validate
2771 * the config then declare the pool unfit for use. If we're
2772 * assembling a pool from a split, the log is not transferred
2775 if (type != SPA_IMPORT_ASSEMBLE) {
2778 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2779 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2781 if (!spa_config_valid(spa, nvconfig)) {
2782 nvlist_free(nvconfig);
2783 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2786 nvlist_free(nvconfig);
2789 * Now that we've validated the config, check the state of the
2790 * root vdev. If it can't be opened, it indicates one or
2791 * more toplevel vdevs are faulted.
2793 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2794 return (SET_ERROR(ENXIO));
2796 if (spa_writeable(spa) && spa_check_logs(spa)) {
2797 *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2798 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2802 if (missing_feat_write) {
2803 ASSERT(state == SPA_LOAD_TRYIMPORT);
2806 * At this point, we know that we can open the pool in
2807 * read-only mode but not read-write mode. We now have enough
2808 * information and can return to userland.
2810 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
2814 * We've successfully opened the pool, verify that we're ready
2815 * to start pushing transactions.
2817 if (state != SPA_LOAD_TRYIMPORT) {
2818 if (error = spa_load_verify(spa))
2819 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2823 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2824 spa->spa_load_max_txg == UINT64_MAX)) {
2826 int need_update = B_FALSE;
2827 dsl_pool_t *dp = spa_get_dsl(spa);
2829 ASSERT(state != SPA_LOAD_TRYIMPORT);
2832 * Claim log blocks that haven't been committed yet.
2833 * This must all happen in a single txg.
2834 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2835 * invoked from zil_claim_log_block()'s i/o done callback.
2836 * Price of rollback is that we abandon the log.
2838 spa->spa_claiming = B_TRUE;
2840 tx = dmu_tx_create_assigned(dp, spa_first_txg(spa));
2841 (void) dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
2842 zil_claim, tx, DS_FIND_CHILDREN);
2845 spa->spa_claiming = B_FALSE;
2847 spa_set_log_state(spa, SPA_LOG_GOOD);
2848 spa->spa_sync_on = B_TRUE;
2849 txg_sync_start(spa->spa_dsl_pool);
2852 * Wait for all claims to sync. We sync up to the highest
2853 * claimed log block birth time so that claimed log blocks
2854 * don't appear to be from the future. spa_claim_max_txg
2855 * will have been set for us by either zil_check_log_chain()
2856 * (invoked from spa_check_logs()) or zil_claim() above.
2858 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2861 * If the config cache is stale, or we have uninitialized
2862 * metaslabs (see spa_vdev_add()), then update the config.
2864 * If this is a verbatim import, trust the current
2865 * in-core spa_config and update the disk labels.
2867 if (config_cache_txg != spa->spa_config_txg ||
2868 state == SPA_LOAD_IMPORT ||
2869 state == SPA_LOAD_RECOVER ||
2870 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2871 need_update = B_TRUE;
2873 for (int c = 0; c < rvd->vdev_children; c++)
2874 if (rvd->vdev_child[c]->vdev_ms_array == 0)
2875 need_update = B_TRUE;
2878 * Update the config cache asychronously in case we're the
2879 * root pool, in which case the config cache isn't writable yet.
2882 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2885 * Check all DTLs to see if anything needs resilvering.
2887 if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2888 vdev_resilver_needed(rvd, NULL, NULL))
2889 spa_async_request(spa, SPA_ASYNC_RESILVER);
2892 * Log the fact that we booted up (so that we can detect if
2893 * we rebooted in the middle of an operation).
2895 spa_history_log_version(spa, "open");
2898 * Delete any inconsistent datasets.
2900 (void) dmu_objset_find(spa_name(spa),
2901 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2904 * Clean up any stale temporary dataset userrefs.
2906 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2913 spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2915 int mode = spa->spa_mode;
2918 spa_deactivate(spa);
2920 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg - 1;
2922 spa_activate(spa, mode);
2923 spa_async_suspend(spa);
2925 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2929 * If spa_load() fails this function will try loading prior txg's. If
2930 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2931 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2932 * function will not rewind the pool and will return the same error as
2936 spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2937 uint64_t max_request, int rewind_flags)
2939 nvlist_t *loadinfo = NULL;
2940 nvlist_t *config = NULL;
2941 int load_error, rewind_error;
2942 uint64_t safe_rewind_txg;
2945 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2946 spa->spa_load_max_txg = spa->spa_load_txg;
2947 spa_set_log_state(spa, SPA_LOG_CLEAR);
2949 spa->spa_load_max_txg = max_request;
2950 if (max_request != UINT64_MAX)
2951 spa->spa_extreme_rewind = B_TRUE;
2954 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2956 if (load_error == 0)
2959 if (spa->spa_root_vdev != NULL)
2960 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2962 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2963 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2965 if (rewind_flags & ZPOOL_NEVER_REWIND) {
2966 nvlist_free(config);
2967 return (load_error);
2970 if (state == SPA_LOAD_RECOVER) {
2971 /* Price of rolling back is discarding txgs, including log */
2972 spa_set_log_state(spa, SPA_LOG_CLEAR);
2975 * If we aren't rolling back save the load info from our first
2976 * import attempt so that we can restore it after attempting
2979 loadinfo = spa->spa_load_info;
2980 spa->spa_load_info = fnvlist_alloc();
2983 spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2984 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2985 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2986 TXG_INITIAL : safe_rewind_txg;
2989 * Continue as long as we're finding errors, we're still within
2990 * the acceptable rewind range, and we're still finding uberblocks
2992 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2993 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2994 if (spa->spa_load_max_txg < safe_rewind_txg)
2995 spa->spa_extreme_rewind = B_TRUE;
2996 rewind_error = spa_load_retry(spa, state, mosconfig);
2999 spa->spa_extreme_rewind = B_FALSE;
3000 spa->spa_load_max_txg = UINT64_MAX;
3002 if (config && (rewind_error || state != SPA_LOAD_RECOVER))
3003 spa_config_set(spa, config);
3005 if (state == SPA_LOAD_RECOVER) {
3006 ASSERT3P(loadinfo, ==, NULL);
3007 return (rewind_error);
3009 /* Store the rewind info as part of the initial load info */
3010 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
3011 spa->spa_load_info);
3013 /* Restore the initial load info */
3014 fnvlist_free(spa->spa_load_info);
3015 spa->spa_load_info = loadinfo;
3017 return (load_error);
3024 * The import case is identical to an open except that the configuration is sent
3025 * down from userland, instead of grabbed from the configuration cache. For the
3026 * case of an open, the pool configuration will exist in the
3027 * POOL_STATE_UNINITIALIZED state.
3029 * The stats information (gen/count/ustats) is used to gather vdev statistics at
3030 * the same time open the pool, without having to keep around the spa_t in some
3034 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
3038 spa_load_state_t state = SPA_LOAD_OPEN;
3040 int locked = B_FALSE;
3041 int firstopen = B_FALSE;
3046 * As disgusting as this is, we need to support recursive calls to this
3047 * function because dsl_dir_open() is called during spa_load(), and ends
3048 * up calling spa_open() again. The real fix is to figure out how to
3049 * avoid dsl_dir_open() calling this in the first place.
3051 if (mutex_owner(&spa_namespace_lock) != curthread) {
3052 mutex_enter(&spa_namespace_lock);
3056 if ((spa = spa_lookup(pool)) == NULL) {
3058 mutex_exit(&spa_namespace_lock);
3059 return (SET_ERROR(ENOENT));
3062 if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
3063 zpool_rewind_policy_t policy;
3067 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
3069 if (policy.zrp_request & ZPOOL_DO_REWIND)
3070 state = SPA_LOAD_RECOVER;
3072 spa_activate(spa, spa_mode_global);
3074 if (state != SPA_LOAD_RECOVER)
3075 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3077 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
3078 policy.zrp_request);
3080 if (error == EBADF) {
3082 * If vdev_validate() returns failure (indicated by
3083 * EBADF), it indicates that one of the vdevs indicates
3084 * that the pool has been exported or destroyed. If
3085 * this is the case, the config cache is out of sync and
3086 * we should remove the pool from the namespace.
3089 spa_deactivate(spa);
3090 spa_config_sync(spa, B_TRUE, B_TRUE);
3093 mutex_exit(&spa_namespace_lock);
3094 return (SET_ERROR(ENOENT));
3099 * We can't open the pool, but we still have useful
3100 * information: the state of each vdev after the
3101 * attempted vdev_open(). Return this to the user.
3103 if (config != NULL && spa->spa_config) {
3104 VERIFY(nvlist_dup(spa->spa_config, config,
3106 VERIFY(nvlist_add_nvlist(*config,
3107 ZPOOL_CONFIG_LOAD_INFO,
3108 spa->spa_load_info) == 0);
3111 spa_deactivate(spa);
3112 spa->spa_last_open_failed = error;
3114 mutex_exit(&spa_namespace_lock);
3120 spa_open_ref(spa, tag);
3123 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
3126 * If we've recovered the pool, pass back any information we
3127 * gathered while doing the load.
3129 if (state == SPA_LOAD_RECOVER) {
3130 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
3131 spa->spa_load_info) == 0);
3135 spa->spa_last_open_failed = 0;
3136 spa->spa_last_ubsync_txg = 0;
3137 spa->spa_load_txg = 0;
3138 mutex_exit(&spa_namespace_lock);
3142 zvol_create_minors(spa->spa_name);
3153 spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
3156 return (spa_open_common(name, spapp, tag, policy, config));
3160 spa_open(const char *name, spa_t **spapp, void *tag)
3162 return (spa_open_common(name, spapp, tag, NULL, NULL));
3166 * Lookup the given spa_t, incrementing the inject count in the process,
3167 * preventing it from being exported or destroyed.
3170 spa_inject_addref(char *name)
3174 mutex_enter(&spa_namespace_lock);
3175 if ((spa = spa_lookup(name)) == NULL) {
3176 mutex_exit(&spa_namespace_lock);
3179 spa->spa_inject_ref++;
3180 mutex_exit(&spa_namespace_lock);
3186 spa_inject_delref(spa_t *spa)
3188 mutex_enter(&spa_namespace_lock);
3189 spa->spa_inject_ref--;
3190 mutex_exit(&spa_namespace_lock);
3194 * Add spares device information to the nvlist.
3197 spa_add_spares(spa_t *spa, nvlist_t *config)
3207 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3209 if (spa->spa_spares.sav_count == 0)
3212 VERIFY(nvlist_lookup_nvlist(config,
3213 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3214 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
3215 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3217 VERIFY(nvlist_add_nvlist_array(nvroot,
3218 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3219 VERIFY(nvlist_lookup_nvlist_array(nvroot,
3220 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3223 * Go through and find any spares which have since been
3224 * repurposed as an active spare. If this is the case, update
3225 * their status appropriately.
3227 for (i = 0; i < nspares; i++) {
3228 VERIFY(nvlist_lookup_uint64(spares[i],
3229 ZPOOL_CONFIG_GUID, &guid) == 0);
3230 if (spa_spare_exists(guid, &pool, NULL) &&
3232 VERIFY(nvlist_lookup_uint64_array(
3233 spares[i], ZPOOL_CONFIG_VDEV_STATS,
3234 (uint64_t **)&vs, &vsc) == 0);
3235 vs->vs_state = VDEV_STATE_CANT_OPEN;
3236 vs->vs_aux = VDEV_AUX_SPARED;
3243 * Add l2cache device information to the nvlist, including vdev stats.
3246 spa_add_l2cache(spa_t *spa, nvlist_t *config)
3249 uint_t i, j, nl2cache;
3256 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3258 if (spa->spa_l2cache.sav_count == 0)
3261 VERIFY(nvlist_lookup_nvlist(config,
3262 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3263 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
3264 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3265 if (nl2cache != 0) {
3266 VERIFY(nvlist_add_nvlist_array(nvroot,
3267 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3268 VERIFY(nvlist_lookup_nvlist_array(nvroot,
3269 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3272 * Update level 2 cache device stats.
3275 for (i = 0; i < nl2cache; i++) {
3276 VERIFY(nvlist_lookup_uint64(l2cache[i],
3277 ZPOOL_CONFIG_GUID, &guid) == 0);
3280 for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
3282 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
3283 vd = spa->spa_l2cache.sav_vdevs[j];
3289 VERIFY(nvlist_lookup_uint64_array(l2cache[i],
3290 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
3292 vdev_get_stats(vd, vs);
3298 spa_add_feature_stats(spa_t *spa, nvlist_t *config)
3304 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3305 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3307 /* We may be unable to read features if pool is suspended. */
3308 if (spa_suspended(spa))
3311 if (spa->spa_feat_for_read_obj != 0) {
3312 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3313 spa->spa_feat_for_read_obj);
3314 zap_cursor_retrieve(&zc, &za) == 0;
3315 zap_cursor_advance(&zc)) {
3316 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3317 za.za_num_integers == 1);
3318 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3319 za.za_first_integer));
3321 zap_cursor_fini(&zc);
3324 if (spa->spa_feat_for_write_obj != 0) {
3325 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3326 spa->spa_feat_for_write_obj);
3327 zap_cursor_retrieve(&zc, &za) == 0;
3328 zap_cursor_advance(&zc)) {
3329 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3330 za.za_num_integers == 1);
3331 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3332 za.za_first_integer));
3334 zap_cursor_fini(&zc);
3338 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
3340 nvlist_free(features);
3344 spa_get_stats(const char *name, nvlist_t **config,
3345 char *altroot, size_t buflen)
3351 error = spa_open_common(name, &spa, FTAG, NULL, config);
3355 * This still leaves a window of inconsistency where the spares
3356 * or l2cache devices could change and the config would be
3357 * self-inconsistent.
3359 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3361 if (*config != NULL) {
3362 uint64_t loadtimes[2];
3364 loadtimes[0] = spa->spa_loaded_ts.tv_sec;
3365 loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
3366 VERIFY(nvlist_add_uint64_array(*config,
3367 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
3369 VERIFY(nvlist_add_uint64(*config,
3370 ZPOOL_CONFIG_ERRCOUNT,
3371 spa_get_errlog_size(spa)) == 0);
3373 if (spa_suspended(spa))
3374 VERIFY(nvlist_add_uint64(*config,
3375 ZPOOL_CONFIG_SUSPENDED,
3376 spa->spa_failmode) == 0);
3378 spa_add_spares(spa, *config);
3379 spa_add_l2cache(spa, *config);
3380 spa_add_feature_stats(spa, *config);
3385 * We want to get the alternate root even for faulted pools, so we cheat
3386 * and call spa_lookup() directly.
3390 mutex_enter(&spa_namespace_lock);
3391 spa = spa_lookup(name);
3393 spa_altroot(spa, altroot, buflen);
3397 mutex_exit(&spa_namespace_lock);
3399 spa_altroot(spa, altroot, buflen);
3404 spa_config_exit(spa, SCL_CONFIG, FTAG);
3405 spa_close(spa, FTAG);
3412 * Validate that the auxiliary device array is well formed. We must have an
3413 * array of nvlists, each which describes a valid leaf vdev. If this is an
3414 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3415 * specified, as long as they are well-formed.
3418 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
3419 spa_aux_vdev_t *sav, const char *config, uint64_t version,
3420 vdev_labeltype_t label)
3427 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3430 * It's acceptable to have no devs specified.
3432 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
3436 return (SET_ERROR(EINVAL));
3439 * Make sure the pool is formatted with a version that supports this
3442 if (spa_version(spa) < version)
3443 return (SET_ERROR(ENOTSUP));
3446 * Set the pending device list so we correctly handle device in-use
3449 sav->sav_pending = dev;
3450 sav->sav_npending = ndev;
3452 for (i = 0; i < ndev; i++) {
3453 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
3457 if (!vd->vdev_ops->vdev_op_leaf) {
3459 error = SET_ERROR(EINVAL);
3464 * The L2ARC currently only supports disk devices in
3465 * kernel context. For user-level testing, we allow it.
3468 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
3469 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
3470 error = SET_ERROR(ENOTBLK);
3477 if ((error = vdev_open(vd)) == 0 &&
3478 (error = vdev_label_init(vd, crtxg, label)) == 0) {
3479 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
3480 vd->vdev_guid) == 0);
3486 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
3493 sav->sav_pending = NULL;
3494 sav->sav_npending = 0;
3499 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
3503 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3505 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3506 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
3507 VDEV_LABEL_SPARE)) != 0) {
3511 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3512 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
3513 VDEV_LABEL_L2CACHE));
3517 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
3522 if (sav->sav_config != NULL) {
3528 * Generate new dev list by concatentating with the
3531 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
3532 &olddevs, &oldndevs) == 0);
3534 newdevs = kmem_alloc(sizeof (void *) *
3535 (ndevs + oldndevs), KM_SLEEP);
3536 for (i = 0; i < oldndevs; i++)
3537 VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
3539 for (i = 0; i < ndevs; i++)
3540 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
3543 VERIFY(nvlist_remove(sav->sav_config, config,
3544 DATA_TYPE_NVLIST_ARRAY) == 0);
3546 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3547 config, newdevs, ndevs + oldndevs) == 0);
3548 for (i = 0; i < oldndevs + ndevs; i++)
3549 nvlist_free(newdevs[i]);
3550 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
3553 * Generate a new dev list.
3555 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
3557 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
3563 * Stop and drop level 2 ARC devices
3566 spa_l2cache_drop(spa_t *spa)
3570 spa_aux_vdev_t *sav = &spa->spa_l2cache;
3572 for (i = 0; i < sav->sav_count; i++) {
3575 vd = sav->sav_vdevs[i];
3578 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
3579 pool != 0ULL && l2arc_vdev_present(vd))
3580 l2arc_remove_vdev(vd);
3588 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
3592 char *altroot = NULL;
3597 uint64_t txg = TXG_INITIAL;
3598 nvlist_t **spares, **l2cache;
3599 uint_t nspares, nl2cache;
3600 uint64_t version, obj;
3601 boolean_t has_features;
3605 if (nvlist_lookup_string(props,
3606 zpool_prop_to_name(ZPOOL_PROP_TNAME), &poolname) != 0)
3607 poolname = (char *)pool;
3610 * If this pool already exists, return failure.
3612 mutex_enter(&spa_namespace_lock);
3613 if (spa_lookup(poolname) != NULL) {
3614 mutex_exit(&spa_namespace_lock);
3615 return (SET_ERROR(EEXIST));
3619 * Allocate a new spa_t structure.
3621 nvl = fnvlist_alloc();
3622 fnvlist_add_string(nvl, ZPOOL_CONFIG_POOL_NAME, pool);
3623 (void) nvlist_lookup_string(props,
3624 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3625 spa = spa_add(poolname, nvl, altroot);
3627 spa_activate(spa, spa_mode_global);
3629 if (props && (error = spa_prop_validate(spa, props))) {
3630 spa_deactivate(spa);
3632 mutex_exit(&spa_namespace_lock);
3637 * Temporary pool names should never be written to disk.
3639 if (poolname != pool)
3640 spa->spa_import_flags |= ZFS_IMPORT_TEMP_NAME;
3642 has_features = B_FALSE;
3643 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
3644 elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
3645 if (zpool_prop_feature(nvpair_name(elem)))
3646 has_features = B_TRUE;
3649 if (has_features || nvlist_lookup_uint64(props,
3650 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
3651 version = SPA_VERSION;
3653 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
3655 spa->spa_first_txg = txg;
3656 spa->spa_uberblock.ub_txg = txg - 1;
3657 spa->spa_uberblock.ub_version = version;
3658 spa->spa_ubsync = spa->spa_uberblock;
3659 spa->spa_load_state = SPA_LOAD_CREATE;
3662 * Create "The Godfather" zio to hold all async IOs
3664 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
3666 for (int i = 0; i < max_ncpus; i++) {
3667 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
3668 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
3669 ZIO_FLAG_GODFATHER);
3673 * Create the root vdev.
3675 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3677 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3679 ASSERT(error != 0 || rvd != NULL);
3680 ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3682 if (error == 0 && !zfs_allocatable_devs(nvroot))
3683 error = SET_ERROR(EINVAL);
3686 (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3687 (error = spa_validate_aux(spa, nvroot, txg,
3688 VDEV_ALLOC_ADD)) == 0) {
3689 for (int c = 0; c < rvd->vdev_children; c++) {
3690 vdev_ashift_optimize(rvd->vdev_child[c]);
3691 vdev_metaslab_set_size(rvd->vdev_child[c]);
3692 vdev_expand(rvd->vdev_child[c], txg);
3696 spa_config_exit(spa, SCL_ALL, FTAG);
3700 spa_deactivate(spa);
3702 mutex_exit(&spa_namespace_lock);
3707 * Get the list of spares, if specified.
3709 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3710 &spares, &nspares) == 0) {
3711 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3713 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3714 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3715 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3716 spa_load_spares(spa);
3717 spa_config_exit(spa, SCL_ALL, FTAG);
3718 spa->spa_spares.sav_sync = B_TRUE;
3722 * Get the list of level 2 cache devices, if specified.
3724 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3725 &l2cache, &nl2cache) == 0) {
3726 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3727 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3728 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3729 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3730 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3731 spa_load_l2cache(spa);
3732 spa_config_exit(spa, SCL_ALL, FTAG);
3733 spa->spa_l2cache.sav_sync = B_TRUE;
3736 spa->spa_is_initializing = B_TRUE;
3737 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3738 spa->spa_meta_objset = dp->dp_meta_objset;
3739 spa->spa_is_initializing = B_FALSE;
3742 * Create DDTs (dedup tables).
3746 spa_update_dspace(spa);
3748 tx = dmu_tx_create_assigned(dp, txg);
3751 * Create the pool config object.
3753 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3754 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3755 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3757 if (zap_add(spa->spa_meta_objset,
3758 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3759 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3760 cmn_err(CE_PANIC, "failed to add pool config");
3763 if (spa_version(spa) >= SPA_VERSION_FEATURES)
3764 spa_feature_create_zap_objects(spa, tx);
3766 if (zap_add(spa->spa_meta_objset,
3767 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3768 sizeof (uint64_t), 1, &version, tx) != 0) {
3769 cmn_err(CE_PANIC, "failed to add pool version");
3772 /* Newly created pools with the right version are always deflated. */
3773 if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3774 spa->spa_deflate = TRUE;
3775 if (zap_add(spa->spa_meta_objset,
3776 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3777 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3778 cmn_err(CE_PANIC, "failed to add deflate");
3783 * Create the deferred-free bpobj. Turn off compression
3784 * because sync-to-convergence takes longer if the blocksize
3787 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3788 dmu_object_set_compress(spa->spa_meta_objset, obj,
3789 ZIO_COMPRESS_OFF, tx);
3790 if (zap_add(spa->spa_meta_objset,
3791 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3792 sizeof (uint64_t), 1, &obj, tx) != 0) {
3793 cmn_err(CE_PANIC, "failed to add bpobj");
3795 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3796 spa->spa_meta_objset, obj));
3799 * Create the pool's history object.
3801 if (version >= SPA_VERSION_ZPOOL_HISTORY)
3802 spa_history_create_obj(spa, tx);
3805 * Generate some random noise for salted checksums to operate on.
3807 (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes,
3808 sizeof (spa->spa_cksum_salt.zcs_bytes));
3811 * Set pool properties.
3813 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3814 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3815 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3816 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3818 if (props != NULL) {
3819 spa_configfile_set(spa, props, B_FALSE);
3820 spa_sync_props(props, tx);
3825 spa->spa_sync_on = B_TRUE;
3826 txg_sync_start(spa->spa_dsl_pool);
3829 * We explicitly wait for the first transaction to complete so that our
3830 * bean counters are appropriately updated.
3832 txg_wait_synced(spa->spa_dsl_pool, txg);
3834 spa_config_sync(spa, B_FALSE, B_TRUE);
3835 spa_event_notify(spa, NULL, ESC_ZFS_POOL_CREATE);
3837 spa_history_log_version(spa, "create");
3840 * Don't count references from objsets that are already closed
3841 * and are making their way through the eviction process.
3843 spa_evicting_os_wait(spa);
3844 spa->spa_minref = refcount_count(&spa->spa_refcount);
3845 spa->spa_load_state = SPA_LOAD_NONE;
3847 mutex_exit(&spa_namespace_lock);
3855 * Get the root pool information from the root disk, then import the root pool
3856 * during the system boot up time.
3858 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3861 spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3864 nvlist_t *nvtop, *nvroot;
3867 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3871 * Add this top-level vdev to the child array.
3873 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3875 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3877 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3880 * Put this pool's top-level vdevs into a root vdev.
3882 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3883 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3884 VDEV_TYPE_ROOT) == 0);
3885 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3886 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3887 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3891 * Replace the existing vdev_tree with the new root vdev in
3892 * this pool's configuration (remove the old, add the new).
3894 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3895 nvlist_free(nvroot);
3900 * Walk the vdev tree and see if we can find a device with "better"
3901 * configuration. A configuration is "better" if the label on that
3902 * device has a more recent txg.
3905 spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3907 for (int c = 0; c < vd->vdev_children; c++)
3908 spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3910 if (vd->vdev_ops->vdev_op_leaf) {
3914 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3918 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3922 * Do we have a better boot device?
3924 if (label_txg > *txg) {
3933 * Import a root pool.
3935 * For x86. devpath_list will consist of devid and/or physpath name of
3936 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3937 * The GRUB "findroot" command will return the vdev we should boot.
3939 * For Sparc, devpath_list consists the physpath name of the booting device
3940 * no matter the rootpool is a single device pool or a mirrored pool.
3942 * "/pci@1f,0/ide@d/disk@0,0:a"
3945 spa_import_rootpool(char *devpath, char *devid)
3948 vdev_t *rvd, *bvd, *avd = NULL;
3949 nvlist_t *config, *nvtop;
3955 * Read the label from the boot device and generate a configuration.
3957 config = spa_generate_rootconf(devpath, devid, &guid);
3958 #if defined(_OBP) && defined(_KERNEL)
3959 if (config == NULL) {
3960 if (strstr(devpath, "/iscsi/ssd") != NULL) {
3962 get_iscsi_bootpath_phy(devpath);
3963 config = spa_generate_rootconf(devpath, devid, &guid);
3967 if (config == NULL) {
3968 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
3970 return (SET_ERROR(EIO));
3973 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3975 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3977 mutex_enter(&spa_namespace_lock);
3978 if ((spa = spa_lookup(pname)) != NULL) {
3980 * Remove the existing root pool from the namespace so that we
3981 * can replace it with the correct config we just read in.
3986 spa = spa_add(pname, config, NULL);
3987 spa->spa_is_root = B_TRUE;
3988 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3991 * Build up a vdev tree based on the boot device's label config.
3993 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3995 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3996 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3997 VDEV_ALLOC_ROOTPOOL);
3998 spa_config_exit(spa, SCL_ALL, FTAG);
4000 mutex_exit(&spa_namespace_lock);
4001 nvlist_free(config);
4002 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
4008 * Get the boot vdev.
4010 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
4011 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
4012 (u_longlong_t)guid);
4013 error = SET_ERROR(ENOENT);
4018 * Determine if there is a better boot device.
4021 spa_alt_rootvdev(rvd, &avd, &txg);
4023 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
4024 "try booting from '%s'", avd->vdev_path);
4025 error = SET_ERROR(EINVAL);
4030 * If the boot device is part of a spare vdev then ensure that
4031 * we're booting off the active spare.
4033 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
4034 !bvd->vdev_isspare) {
4035 cmn_err(CE_NOTE, "The boot device is currently spared. Please "
4036 "try booting from '%s'",
4038 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
4039 error = SET_ERROR(EINVAL);
4045 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4047 spa_config_exit(spa, SCL_ALL, FTAG);
4048 mutex_exit(&spa_namespace_lock);
4050 nvlist_free(config);
4054 #else /* !illumos */
4056 extern int vdev_geom_read_pool_label(const char *name, nvlist_t ***configs,
4060 spa_generate_rootconf(const char *name)
4062 nvlist_t **configs, **tops;
4064 nvlist_t *best_cfg, *nvtop, *nvroot;
4073 if (vdev_geom_read_pool_label(name, &configs, &count) != 0)
4076 ASSERT3U(count, !=, 0);
4078 for (i = 0; i < count; i++) {
4081 VERIFY(nvlist_lookup_uint64(configs[i], ZPOOL_CONFIG_POOL_TXG,
4083 if (txg > best_txg) {
4085 best_cfg = configs[i];
4090 * Multi-vdev root pool configuration discovery is not supported yet.
4093 nvlist_lookup_uint64(best_cfg, ZPOOL_CONFIG_VDEV_CHILDREN, &nchildren);
4095 nvlist_lookup_uint64_array(best_cfg, ZPOOL_CONFIG_HOLE_ARRAY,
4098 tops = kmem_zalloc(nchildren * sizeof(void *), KM_SLEEP);
4099 for (i = 0; i < nchildren; i++) {
4102 if (configs[i] == NULL)
4104 VERIFY(nvlist_lookup_nvlist(configs[i], ZPOOL_CONFIG_VDEV_TREE,
4106 nvlist_dup(nvtop, &tops[i], KM_SLEEP);
4108 for (i = 0; holes != NULL && i < nholes; i++) {
4111 if (tops[holes[i]] != NULL)
4113 nvlist_alloc(&tops[holes[i]], NV_UNIQUE_NAME, KM_SLEEP);
4114 VERIFY(nvlist_add_string(tops[holes[i]], ZPOOL_CONFIG_TYPE,
4115 VDEV_TYPE_HOLE) == 0);
4116 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_ID,
4118 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_GUID,
4121 for (i = 0; i < nchildren; i++) {
4122 if (tops[i] != NULL)
4124 nvlist_alloc(&tops[i], NV_UNIQUE_NAME, KM_SLEEP);
4125 VERIFY(nvlist_add_string(tops[i], ZPOOL_CONFIG_TYPE,
4126 VDEV_TYPE_MISSING) == 0);
4127 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_ID,
4129 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_GUID,
4134 * Create pool config based on the best vdev config.
4136 nvlist_dup(best_cfg, &config, KM_SLEEP);
4139 * Put this pool's top-level vdevs into a root vdev.
4141 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4143 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4144 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
4145 VDEV_TYPE_ROOT) == 0);
4146 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
4147 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
4148 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
4149 tops, nchildren) == 0);
4152 * Replace the existing vdev_tree with the new root vdev in
4153 * this pool's configuration (remove the old, add the new).
4155 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
4158 * Drop vdev config elements that should not be present at pool level.
4160 nvlist_remove(config, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64);
4161 nvlist_remove(config, ZPOOL_CONFIG_TOP_GUID, DATA_TYPE_UINT64);
4163 for (i = 0; i < count; i++)
4164 nvlist_free(configs[i]);
4165 kmem_free(configs, count * sizeof(void *));
4166 for (i = 0; i < nchildren; i++)
4167 nvlist_free(tops[i]);
4168 kmem_free(tops, nchildren * sizeof(void *));
4169 nvlist_free(nvroot);
4174 spa_import_rootpool(const char *name)
4177 vdev_t *rvd, *bvd, *avd = NULL;
4178 nvlist_t *config, *nvtop;
4184 * Read the label from the boot device and generate a configuration.
4186 config = spa_generate_rootconf(name);
4188 mutex_enter(&spa_namespace_lock);
4189 if (config != NULL) {
4190 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
4191 &pname) == 0 && strcmp(name, pname) == 0);
4192 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg)
4195 if ((spa = spa_lookup(pname)) != NULL) {
4197 * The pool could already be imported,
4198 * e.g., after reboot -r.
4200 if (spa->spa_state == POOL_STATE_ACTIVE) {
4201 mutex_exit(&spa_namespace_lock);
4202 nvlist_free(config);
4207 * Remove the existing root pool from the namespace so
4208 * that we can replace it with the correct config
4213 spa = spa_add(pname, config, NULL);
4216 * Set spa_ubsync.ub_version as it can be used in vdev_alloc()
4217 * via spa_version().
4219 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
4220 &spa->spa_ubsync.ub_version) != 0)
4221 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
4222 } else if ((spa = spa_lookup(name)) == NULL) {
4223 mutex_exit(&spa_namespace_lock);
4224 nvlist_free(config);
4225 cmn_err(CE_NOTE, "Cannot find the pool label for '%s'",
4229 VERIFY(nvlist_dup(spa->spa_config, &config, KM_SLEEP) == 0);
4231 spa->spa_is_root = B_TRUE;
4232 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
4235 * Build up a vdev tree based on the boot device's label config.
4237 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4239 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4240 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
4241 VDEV_ALLOC_ROOTPOOL);
4242 spa_config_exit(spa, SCL_ALL, FTAG);
4244 mutex_exit(&spa_namespace_lock);
4245 nvlist_free(config);
4246 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
4251 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4253 spa_config_exit(spa, SCL_ALL, FTAG);
4254 mutex_exit(&spa_namespace_lock);
4256 nvlist_free(config);
4260 #endif /* illumos */
4261 #endif /* _KERNEL */
4264 * Import a non-root pool into the system.
4267 spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
4270 char *altroot = NULL;
4271 spa_load_state_t state = SPA_LOAD_IMPORT;
4272 zpool_rewind_policy_t policy;
4273 uint64_t mode = spa_mode_global;
4274 uint64_t readonly = B_FALSE;
4277 nvlist_t **spares, **l2cache;
4278 uint_t nspares, nl2cache;
4281 * If a pool with this name exists, return failure.
4283 mutex_enter(&spa_namespace_lock);
4284 if (spa_lookup(pool) != NULL) {
4285 mutex_exit(&spa_namespace_lock);
4286 return (SET_ERROR(EEXIST));
4290 * Create and initialize the spa structure.
4292 (void) nvlist_lookup_string(props,
4293 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4294 (void) nvlist_lookup_uint64(props,
4295 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
4298 spa = spa_add(pool, config, altroot);
4299 spa->spa_import_flags = flags;
4302 * Verbatim import - Take a pool and insert it into the namespace
4303 * as if it had been loaded at boot.
4305 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
4307 spa_configfile_set(spa, props, B_FALSE);
4309 spa_config_sync(spa, B_FALSE, B_TRUE);
4310 spa_event_notify(spa, NULL, ESC_ZFS_POOL_IMPORT);
4312 mutex_exit(&spa_namespace_lock);
4316 spa_activate(spa, mode);
4319 * Don't start async tasks until we know everything is healthy.
4321 spa_async_suspend(spa);
4323 zpool_get_rewind_policy(config, &policy);
4324 if (policy.zrp_request & ZPOOL_DO_REWIND)
4325 state = SPA_LOAD_RECOVER;
4328 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
4329 * because the user-supplied config is actually the one to trust when
4332 if (state != SPA_LOAD_RECOVER)
4333 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
4335 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
4336 policy.zrp_request);
4339 * Propagate anything learned while loading the pool and pass it
4340 * back to caller (i.e. rewind info, missing devices, etc).
4342 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4343 spa->spa_load_info) == 0);
4345 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4347 * Toss any existing sparelist, as it doesn't have any validity
4348 * anymore, and conflicts with spa_has_spare().
4350 if (spa->spa_spares.sav_config) {
4351 nvlist_free(spa->spa_spares.sav_config);
4352 spa->spa_spares.sav_config = NULL;
4353 spa_load_spares(spa);
4355 if (spa->spa_l2cache.sav_config) {
4356 nvlist_free(spa->spa_l2cache.sav_config);
4357 spa->spa_l2cache.sav_config = NULL;
4358 spa_load_l2cache(spa);
4361 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4364 error = spa_validate_aux(spa, nvroot, -1ULL,
4367 error = spa_validate_aux(spa, nvroot, -1ULL,
4368 VDEV_ALLOC_L2CACHE);
4369 spa_config_exit(spa, SCL_ALL, FTAG);
4372 spa_configfile_set(spa, props, B_FALSE);
4374 if (error != 0 || (props && spa_writeable(spa) &&
4375 (error = spa_prop_set(spa, props)))) {
4377 spa_deactivate(spa);
4379 mutex_exit(&spa_namespace_lock);
4383 spa_async_resume(spa);
4386 * Override any spares and level 2 cache devices as specified by
4387 * the user, as these may have correct device names/devids, etc.
4389 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
4390 &spares, &nspares) == 0) {
4391 if (spa->spa_spares.sav_config)
4392 VERIFY(nvlist_remove(spa->spa_spares.sav_config,
4393 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
4395 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
4396 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4397 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
4398 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
4399 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4400 spa_load_spares(spa);
4401 spa_config_exit(spa, SCL_ALL, FTAG);
4402 spa->spa_spares.sav_sync = B_TRUE;
4404 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
4405 &l2cache, &nl2cache) == 0) {
4406 if (spa->spa_l2cache.sav_config)
4407 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
4408 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
4410 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
4411 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4412 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
4413 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
4414 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4415 spa_load_l2cache(spa);
4416 spa_config_exit(spa, SCL_ALL, FTAG);
4417 spa->spa_l2cache.sav_sync = B_TRUE;
4421 * Check for any removed devices.
4423 if (spa->spa_autoreplace) {
4424 spa_aux_check_removed(&spa->spa_spares);
4425 spa_aux_check_removed(&spa->spa_l2cache);
4428 if (spa_writeable(spa)) {
4430 * Update the config cache to include the newly-imported pool.
4432 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4436 * It's possible that the pool was expanded while it was exported.
4437 * We kick off an async task to handle this for us.
4439 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
4441 spa_history_log_version(spa, "import");
4443 spa_event_notify(spa, NULL, ESC_ZFS_POOL_IMPORT);
4445 mutex_exit(&spa_namespace_lock);
4449 zvol_create_minors(pool);
4456 spa_tryimport(nvlist_t *tryconfig)
4458 nvlist_t *config = NULL;
4464 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
4467 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
4471 * Create and initialize the spa structure.
4473 mutex_enter(&spa_namespace_lock);
4474 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
4475 spa_activate(spa, FREAD);
4478 * Pass off the heavy lifting to spa_load().
4479 * Pass TRUE for mosconfig because the user-supplied config
4480 * is actually the one to trust when doing an import.
4482 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
4485 * If 'tryconfig' was at least parsable, return the current config.
4487 if (spa->spa_root_vdev != NULL) {
4488 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4489 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
4491 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4493 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
4494 spa->spa_uberblock.ub_timestamp) == 0);
4495 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4496 spa->spa_load_info) == 0);
4499 * If the bootfs property exists on this pool then we
4500 * copy it out so that external consumers can tell which
4501 * pools are bootable.
4503 if ((!error || error == EEXIST) && spa->spa_bootfs) {
4504 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4507 * We have to play games with the name since the
4508 * pool was opened as TRYIMPORT_NAME.
4510 if (dsl_dsobj_to_dsname(spa_name(spa),
4511 spa->spa_bootfs, tmpname) == 0) {
4513 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4515 cp = strchr(tmpname, '/');
4517 (void) strlcpy(dsname, tmpname,
4520 (void) snprintf(dsname, MAXPATHLEN,
4521 "%s/%s", poolname, ++cp);
4523 VERIFY(nvlist_add_string(config,
4524 ZPOOL_CONFIG_BOOTFS, dsname) == 0);
4525 kmem_free(dsname, MAXPATHLEN);
4527 kmem_free(tmpname, MAXPATHLEN);
4531 * Add the list of hot spares and level 2 cache devices.
4533 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4534 spa_add_spares(spa, config);
4535 spa_add_l2cache(spa, config);
4536 spa_config_exit(spa, SCL_CONFIG, FTAG);
4540 spa_deactivate(spa);
4542 mutex_exit(&spa_namespace_lock);
4548 * Pool export/destroy
4550 * The act of destroying or exporting a pool is very simple. We make sure there
4551 * is no more pending I/O and any references to the pool are gone. Then, we
4552 * update the pool state and sync all the labels to disk, removing the
4553 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4554 * we don't sync the labels or remove the configuration cache.
4557 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
4558 boolean_t force, boolean_t hardforce)
4565 if (!(spa_mode_global & FWRITE))
4566 return (SET_ERROR(EROFS));
4568 mutex_enter(&spa_namespace_lock);
4569 if ((spa = spa_lookup(pool)) == NULL) {
4570 mutex_exit(&spa_namespace_lock);
4571 return (SET_ERROR(ENOENT));
4575 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4576 * reacquire the namespace lock, and see if we can export.
4578 spa_open_ref(spa, FTAG);
4579 mutex_exit(&spa_namespace_lock);
4580 spa_async_suspend(spa);
4581 mutex_enter(&spa_namespace_lock);
4582 spa_close(spa, FTAG);
4585 * The pool will be in core if it's openable,
4586 * in which case we can modify its state.
4588 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
4590 * Objsets may be open only because they're dirty, so we
4591 * have to force it to sync before checking spa_refcnt.
4593 txg_wait_synced(spa->spa_dsl_pool, 0);
4594 spa_evicting_os_wait(spa);
4597 * A pool cannot be exported or destroyed if there are active
4598 * references. If we are resetting a pool, allow references by
4599 * fault injection handlers.
4601 if (!spa_refcount_zero(spa) ||
4602 (spa->spa_inject_ref != 0 &&
4603 new_state != POOL_STATE_UNINITIALIZED)) {
4604 spa_async_resume(spa);
4605 mutex_exit(&spa_namespace_lock);
4606 return (SET_ERROR(EBUSY));
4610 * A pool cannot be exported if it has an active shared spare.
4611 * This is to prevent other pools stealing the active spare
4612 * from an exported pool. At user's own will, such pool can
4613 * be forcedly exported.
4615 if (!force && new_state == POOL_STATE_EXPORTED &&
4616 spa_has_active_shared_spare(spa)) {
4617 spa_async_resume(spa);
4618 mutex_exit(&spa_namespace_lock);
4619 return (SET_ERROR(EXDEV));
4623 * We want this to be reflected on every label,
4624 * so mark them all dirty. spa_unload() will do the
4625 * final sync that pushes these changes out.
4627 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
4628 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4629 spa->spa_state = new_state;
4630 spa->spa_final_txg = spa_last_synced_txg(spa) +
4632 vdev_config_dirty(spa->spa_root_vdev);
4633 spa_config_exit(spa, SCL_ALL, FTAG);
4637 spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
4639 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4641 spa_deactivate(spa);
4644 if (oldconfig && spa->spa_config)
4645 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
4647 if (new_state != POOL_STATE_UNINITIALIZED) {
4649 spa_config_sync(spa, B_TRUE, B_TRUE);
4652 mutex_exit(&spa_namespace_lock);
4658 * Destroy a storage pool.
4661 spa_destroy(char *pool)
4663 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
4668 * Export a storage pool.
4671 spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
4672 boolean_t hardforce)
4674 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
4679 * Similar to spa_export(), this unloads the spa_t without actually removing it
4680 * from the namespace in any way.
4683 spa_reset(char *pool)
4685 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
4690 * ==========================================================================
4691 * Device manipulation
4692 * ==========================================================================
4696 * Add a device to a storage pool.
4699 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
4703 vdev_t *rvd = spa->spa_root_vdev;
4705 nvlist_t **spares, **l2cache;
4706 uint_t nspares, nl2cache;
4708 ASSERT(spa_writeable(spa));
4710 txg = spa_vdev_enter(spa);
4712 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
4713 VDEV_ALLOC_ADD)) != 0)
4714 return (spa_vdev_exit(spa, NULL, txg, error));
4716 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */
4718 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
4722 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
4726 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
4727 return (spa_vdev_exit(spa, vd, txg, EINVAL));
4729 if (vd->vdev_children != 0 &&
4730 (error = vdev_create(vd, txg, B_FALSE)) != 0)
4731 return (spa_vdev_exit(spa, vd, txg, error));
4734 * We must validate the spares and l2cache devices after checking the
4735 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
4737 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
4738 return (spa_vdev_exit(spa, vd, txg, error));
4741 * Transfer each new top-level vdev from vd to rvd.
4743 for (int c = 0; c < vd->vdev_children; c++) {
4746 * Set the vdev id to the first hole, if one exists.
4748 for (id = 0; id < rvd->vdev_children; id++) {
4749 if (rvd->vdev_child[id]->vdev_ishole) {
4750 vdev_free(rvd->vdev_child[id]);
4754 tvd = vd->vdev_child[c];
4755 vdev_remove_child(vd, tvd);
4757 vdev_add_child(rvd, tvd);
4758 vdev_config_dirty(tvd);
4762 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
4763 ZPOOL_CONFIG_SPARES);
4764 spa_load_spares(spa);
4765 spa->spa_spares.sav_sync = B_TRUE;
4768 if (nl2cache != 0) {
4769 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
4770 ZPOOL_CONFIG_L2CACHE);
4771 spa_load_l2cache(spa);
4772 spa->spa_l2cache.sav_sync = B_TRUE;
4776 * We have to be careful when adding new vdevs to an existing pool.
4777 * If other threads start allocating from these vdevs before we
4778 * sync the config cache, and we lose power, then upon reboot we may
4779 * fail to open the pool because there are DVAs that the config cache
4780 * can't translate. Therefore, we first add the vdevs without
4781 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4782 * and then let spa_config_update() initialize the new metaslabs.
4784 * spa_load() checks for added-but-not-initialized vdevs, so that
4785 * if we lose power at any point in this sequence, the remaining
4786 * steps will be completed the next time we load the pool.
4788 (void) spa_vdev_exit(spa, vd, txg, 0);
4790 mutex_enter(&spa_namespace_lock);
4791 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4792 spa_event_notify(spa, NULL, ESC_ZFS_VDEV_ADD);
4793 mutex_exit(&spa_namespace_lock);
4799 * Attach a device to a mirror. The arguments are the path to any device
4800 * in the mirror, and the nvroot for the new device. If the path specifies
4801 * a device that is not mirrored, we automatically insert the mirror vdev.
4803 * If 'replacing' is specified, the new device is intended to replace the
4804 * existing device; in this case the two devices are made into their own
4805 * mirror using the 'replacing' vdev, which is functionally identical to
4806 * the mirror vdev (it actually reuses all the same ops) but has a few
4807 * extra rules: you can't attach to it after it's been created, and upon
4808 * completion of resilvering, the first disk (the one being replaced)
4809 * is automatically detached.
4812 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
4814 uint64_t txg, dtl_max_txg;
4815 vdev_t *rvd = spa->spa_root_vdev;
4816 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
4818 char *oldvdpath, *newvdpath;
4822 ASSERT(spa_writeable(spa));
4824 txg = spa_vdev_enter(spa);
4826 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
4829 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4831 if (!oldvd->vdev_ops->vdev_op_leaf)
4832 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4834 pvd = oldvd->vdev_parent;
4836 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
4837 VDEV_ALLOC_ATTACH)) != 0)
4838 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4840 if (newrootvd->vdev_children != 1)
4841 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4843 newvd = newrootvd->vdev_child[0];
4845 if (!newvd->vdev_ops->vdev_op_leaf)
4846 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4848 if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
4849 return (spa_vdev_exit(spa, newrootvd, txg, error));
4852 * Spares can't replace logs
4854 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
4855 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4859 * For attach, the only allowable parent is a mirror or the root
4862 if (pvd->vdev_ops != &vdev_mirror_ops &&
4863 pvd->vdev_ops != &vdev_root_ops)
4864 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4866 pvops = &vdev_mirror_ops;
4869 * Active hot spares can only be replaced by inactive hot
4872 if (pvd->vdev_ops == &vdev_spare_ops &&
4873 oldvd->vdev_isspare &&
4874 !spa_has_spare(spa, newvd->vdev_guid))
4875 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4878 * If the source is a hot spare, and the parent isn't already a
4879 * spare, then we want to create a new hot spare. Otherwise, we
4880 * want to create a replacing vdev. The user is not allowed to
4881 * attach to a spared vdev child unless the 'isspare' state is
4882 * the same (spare replaces spare, non-spare replaces
4885 if (pvd->vdev_ops == &vdev_replacing_ops &&
4886 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
4887 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4888 } else if (pvd->vdev_ops == &vdev_spare_ops &&
4889 newvd->vdev_isspare != oldvd->vdev_isspare) {
4890 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4893 if (newvd->vdev_isspare)
4894 pvops = &vdev_spare_ops;
4896 pvops = &vdev_replacing_ops;
4900 * Make sure the new device is big enough.
4902 if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
4903 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
4906 * The new device cannot have a higher alignment requirement
4907 * than the top-level vdev.
4909 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4910 return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4913 * If this is an in-place replacement, update oldvd's path and devid
4914 * to make it distinguishable from newvd, and unopenable from now on.
4916 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4917 spa_strfree(oldvd->vdev_path);
4918 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4920 (void) sprintf(oldvd->vdev_path, "%s/%s",
4921 newvd->vdev_path, "old");
4922 if (oldvd->vdev_devid != NULL) {
4923 spa_strfree(oldvd->vdev_devid);
4924 oldvd->vdev_devid = NULL;
4928 /* mark the device being resilvered */
4929 newvd->vdev_resilver_txg = txg;
4932 * If the parent is not a mirror, or if we're replacing, insert the new
4933 * mirror/replacing/spare vdev above oldvd.
4935 if (pvd->vdev_ops != pvops)
4936 pvd = vdev_add_parent(oldvd, pvops);
4938 ASSERT(pvd->vdev_top->vdev_parent == rvd);
4939 ASSERT(pvd->vdev_ops == pvops);
4940 ASSERT(oldvd->vdev_parent == pvd);
4943 * Extract the new device from its root and add it to pvd.
4945 vdev_remove_child(newrootvd, newvd);
4946 newvd->vdev_id = pvd->vdev_children;
4947 newvd->vdev_crtxg = oldvd->vdev_crtxg;
4948 vdev_add_child(pvd, newvd);
4950 tvd = newvd->vdev_top;
4951 ASSERT(pvd->vdev_top == tvd);
4952 ASSERT(tvd->vdev_parent == rvd);
4954 vdev_config_dirty(tvd);
4957 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4958 * for any dmu_sync-ed blocks. It will propagate upward when
4959 * spa_vdev_exit() calls vdev_dtl_reassess().
4961 dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4963 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4964 dtl_max_txg - TXG_INITIAL);
4966 if (newvd->vdev_isspare) {
4967 spa_spare_activate(newvd);
4968 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
4971 oldvdpath = spa_strdup(oldvd->vdev_path);
4972 newvdpath = spa_strdup(newvd->vdev_path);
4973 newvd_isspare = newvd->vdev_isspare;
4976 * Mark newvd's DTL dirty in this txg.
4978 vdev_dirty(tvd, VDD_DTL, newvd, txg);
4981 * Schedule the resilver to restart in the future. We do this to
4982 * ensure that dmu_sync-ed blocks have been stitched into the
4983 * respective datasets.
4985 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4987 if (spa->spa_bootfs)
4988 spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
4990 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_ATTACH);
4995 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4997 spa_history_log_internal(spa, "vdev attach", NULL,
4998 "%s vdev=%s %s vdev=%s",
4999 replacing && newvd_isspare ? "spare in" :
5000 replacing ? "replace" : "attach", newvdpath,
5001 replacing ? "for" : "to", oldvdpath);
5003 spa_strfree(oldvdpath);
5004 spa_strfree(newvdpath);
5010 * Detach a device from a mirror or replacing vdev.
5012 * If 'replace_done' is specified, only detach if the parent
5013 * is a replacing vdev.
5016 spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
5020 vdev_t *rvd = spa->spa_root_vdev;
5021 vdev_t *vd, *pvd, *cvd, *tvd;
5022 boolean_t unspare = B_FALSE;
5023 uint64_t unspare_guid = 0;
5026 ASSERT(spa_writeable(spa));
5028 txg = spa_vdev_enter(spa);
5030 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5033 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
5035 if (!vd->vdev_ops->vdev_op_leaf)
5036 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
5038 pvd = vd->vdev_parent;
5041 * If the parent/child relationship is not as expected, don't do it.
5042 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
5043 * vdev that's replacing B with C. The user's intent in replacing
5044 * is to go from M(A,B) to M(A,C). If the user decides to cancel
5045 * the replace by detaching C, the expected behavior is to end up
5046 * M(A,B). But suppose that right after deciding to detach C,
5047 * the replacement of B completes. We would have M(A,C), and then
5048 * ask to detach C, which would leave us with just A -- not what
5049 * the user wanted. To prevent this, we make sure that the
5050 * parent/child relationship hasn't changed -- in this example,
5051 * that C's parent is still the replacing vdev R.
5053 if (pvd->vdev_guid != pguid && pguid != 0)
5054 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
5057 * Only 'replacing' or 'spare' vdevs can be replaced.
5059 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
5060 pvd->vdev_ops != &vdev_spare_ops)
5061 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
5063 ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
5064 spa_version(spa) >= SPA_VERSION_SPARES);
5067 * Only mirror, replacing, and spare vdevs support detach.
5069 if (pvd->vdev_ops != &vdev_replacing_ops &&
5070 pvd->vdev_ops != &vdev_mirror_ops &&
5071 pvd->vdev_ops != &vdev_spare_ops)
5072 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
5075 * If this device has the only valid copy of some data,
5076 * we cannot safely detach it.
5078 if (vdev_dtl_required(vd))
5079 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
5081 ASSERT(pvd->vdev_children >= 2);
5084 * If we are detaching the second disk from a replacing vdev, then
5085 * check to see if we changed the original vdev's path to have "/old"
5086 * at the end in spa_vdev_attach(). If so, undo that change now.
5088 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
5089 vd->vdev_path != NULL) {
5090 size_t len = strlen(vd->vdev_path);
5092 for (int c = 0; c < pvd->vdev_children; c++) {
5093 cvd = pvd->vdev_child[c];
5095 if (cvd == vd || cvd->vdev_path == NULL)
5098 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
5099 strcmp(cvd->vdev_path + len, "/old") == 0) {
5100 spa_strfree(cvd->vdev_path);
5101 cvd->vdev_path = spa_strdup(vd->vdev_path);
5108 * If we are detaching the original disk from a spare, then it implies
5109 * that the spare should become a real disk, and be removed from the
5110 * active spare list for the pool.
5112 if (pvd->vdev_ops == &vdev_spare_ops &&
5114 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
5118 * Erase the disk labels so the disk can be used for other things.
5119 * This must be done after all other error cases are handled,
5120 * but before we disembowel vd (so we can still do I/O to it).
5121 * But if we can't do it, don't treat the error as fatal --
5122 * it may be that the unwritability of the disk is the reason
5123 * it's being detached!
5125 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5128 * Remove vd from its parent and compact the parent's children.
5130 vdev_remove_child(pvd, vd);
5131 vdev_compact_children(pvd);
5134 * Remember one of the remaining children so we can get tvd below.
5136 cvd = pvd->vdev_child[pvd->vdev_children - 1];
5139 * If we need to remove the remaining child from the list of hot spares,
5140 * do it now, marking the vdev as no longer a spare in the process.
5141 * We must do this before vdev_remove_parent(), because that can
5142 * change the GUID if it creates a new toplevel GUID. For a similar
5143 * reason, we must remove the spare now, in the same txg as the detach;
5144 * otherwise someone could attach a new sibling, change the GUID, and
5145 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
5148 ASSERT(cvd->vdev_isspare);
5149 spa_spare_remove(cvd);
5150 unspare_guid = cvd->vdev_guid;
5151 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
5152 cvd->vdev_unspare = B_TRUE;
5156 * If the parent mirror/replacing vdev only has one child,
5157 * the parent is no longer needed. Remove it from the tree.
5159 if (pvd->vdev_children == 1) {
5160 if (pvd->vdev_ops == &vdev_spare_ops)
5161 cvd->vdev_unspare = B_FALSE;
5162 vdev_remove_parent(cvd);
5167 * We don't set tvd until now because the parent we just removed
5168 * may have been the previous top-level vdev.
5170 tvd = cvd->vdev_top;
5171 ASSERT(tvd->vdev_parent == rvd);
5174 * Reevaluate the parent vdev state.
5176 vdev_propagate_state(cvd);
5179 * If the 'autoexpand' property is set on the pool then automatically
5180 * try to expand the size of the pool. For example if the device we
5181 * just detached was smaller than the others, it may be possible to
5182 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
5183 * first so that we can obtain the updated sizes of the leaf vdevs.
5185 if (spa->spa_autoexpand) {
5187 vdev_expand(tvd, txg);
5190 vdev_config_dirty(tvd);
5193 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
5194 * vd->vdev_detached is set and free vd's DTL object in syncing context.
5195 * But first make sure we're not on any *other* txg's DTL list, to
5196 * prevent vd from being accessed after it's freed.
5198 vdpath = spa_strdup(vd->vdev_path);
5199 for (int t = 0; t < TXG_SIZE; t++)
5200 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
5201 vd->vdev_detached = B_TRUE;
5202 vdev_dirty(tvd, VDD_DTL, vd, txg);
5204 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
5206 /* hang on to the spa before we release the lock */
5207 spa_open_ref(spa, FTAG);
5209 error = spa_vdev_exit(spa, vd, txg, 0);
5211 spa_history_log_internal(spa, "detach", NULL,
5213 spa_strfree(vdpath);
5216 * If this was the removal of the original device in a hot spare vdev,
5217 * then we want to go through and remove the device from the hot spare
5218 * list of every other pool.
5221 spa_t *altspa = NULL;
5223 mutex_enter(&spa_namespace_lock);
5224 while ((altspa = spa_next(altspa)) != NULL) {
5225 if (altspa->spa_state != POOL_STATE_ACTIVE ||
5229 spa_open_ref(altspa, FTAG);
5230 mutex_exit(&spa_namespace_lock);
5231 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
5232 mutex_enter(&spa_namespace_lock);
5233 spa_close(altspa, FTAG);
5235 mutex_exit(&spa_namespace_lock);
5237 /* search the rest of the vdevs for spares to remove */
5238 spa_vdev_resilver_done(spa);
5241 /* all done with the spa; OK to release */
5242 mutex_enter(&spa_namespace_lock);
5243 spa_close(spa, FTAG);
5244 mutex_exit(&spa_namespace_lock);
5250 * Split a set of devices from their mirrors, and create a new pool from them.
5253 spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
5254 nvlist_t *props, boolean_t exp)
5257 uint64_t txg, *glist;
5259 uint_t c, children, lastlog;
5260 nvlist_t **child, *nvl, *tmp;
5262 char *altroot = NULL;
5263 vdev_t *rvd, **vml = NULL; /* vdev modify list */
5264 boolean_t activate_slog;
5266 ASSERT(spa_writeable(spa));
5268 txg = spa_vdev_enter(spa);
5270 /* clear the log and flush everything up to now */
5271 activate_slog = spa_passivate_log(spa);
5272 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5273 error = spa_offline_log(spa);
5274 txg = spa_vdev_config_enter(spa);
5277 spa_activate_log(spa);
5280 return (spa_vdev_exit(spa, NULL, txg, error));
5282 /* check new spa name before going any further */
5283 if (spa_lookup(newname) != NULL)
5284 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
5287 * scan through all the children to ensure they're all mirrors
5289 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
5290 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
5292 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5294 /* first, check to ensure we've got the right child count */
5295 rvd = spa->spa_root_vdev;
5297 for (c = 0; c < rvd->vdev_children; c++) {
5298 vdev_t *vd = rvd->vdev_child[c];
5300 /* don't count the holes & logs as children */
5301 if (vd->vdev_islog || vd->vdev_ishole) {
5309 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
5310 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5312 /* next, ensure no spare or cache devices are part of the split */
5313 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
5314 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
5315 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5317 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
5318 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
5320 /* then, loop over each vdev and validate it */
5321 for (c = 0; c < children; c++) {
5322 uint64_t is_hole = 0;
5324 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
5328 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
5329 spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
5332 error = SET_ERROR(EINVAL);
5337 /* which disk is going to be split? */
5338 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
5340 error = SET_ERROR(EINVAL);
5344 /* look it up in the spa */
5345 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
5346 if (vml[c] == NULL) {
5347 error = SET_ERROR(ENODEV);
5351 /* make sure there's nothing stopping the split */
5352 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
5353 vml[c]->vdev_islog ||
5354 vml[c]->vdev_ishole ||
5355 vml[c]->vdev_isspare ||
5356 vml[c]->vdev_isl2cache ||
5357 !vdev_writeable(vml[c]) ||
5358 vml[c]->vdev_children != 0 ||
5359 vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
5360 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
5361 error = SET_ERROR(EINVAL);
5365 if (vdev_dtl_required(vml[c])) {
5366 error = SET_ERROR(EBUSY);
5370 /* we need certain info from the top level */
5371 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
5372 vml[c]->vdev_top->vdev_ms_array) == 0);
5373 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
5374 vml[c]->vdev_top->vdev_ms_shift) == 0);
5375 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
5376 vml[c]->vdev_top->vdev_asize) == 0);
5377 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
5378 vml[c]->vdev_top->vdev_ashift) == 0);
5382 kmem_free(vml, children * sizeof (vdev_t *));
5383 kmem_free(glist, children * sizeof (uint64_t));
5384 return (spa_vdev_exit(spa, NULL, txg, error));
5387 /* stop writers from using the disks */
5388 for (c = 0; c < children; c++) {
5390 vml[c]->vdev_offline = B_TRUE;
5392 vdev_reopen(spa->spa_root_vdev);
5395 * Temporarily record the splitting vdevs in the spa config. This
5396 * will disappear once the config is regenerated.
5398 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5399 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
5400 glist, children) == 0);
5401 kmem_free(glist, children * sizeof (uint64_t));
5403 mutex_enter(&spa->spa_props_lock);
5404 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
5406 mutex_exit(&spa->spa_props_lock);
5407 spa->spa_config_splitting = nvl;
5408 vdev_config_dirty(spa->spa_root_vdev);
5410 /* configure and create the new pool */
5411 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
5412 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
5413 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
5414 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5415 spa_version(spa)) == 0);
5416 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
5417 spa->spa_config_txg) == 0);
5418 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
5419 spa_generate_guid(NULL)) == 0);
5420 (void) nvlist_lookup_string(props,
5421 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
5423 /* add the new pool to the namespace */
5424 newspa = spa_add(newname, config, altroot);
5425 newspa->spa_config_txg = spa->spa_config_txg;
5426 spa_set_log_state(newspa, SPA_LOG_CLEAR);
5428 /* release the spa config lock, retaining the namespace lock */
5429 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5431 if (zio_injection_enabled)
5432 zio_handle_panic_injection(spa, FTAG, 1);
5434 spa_activate(newspa, spa_mode_global);
5435 spa_async_suspend(newspa);
5438 /* mark that we are creating new spa by splitting */
5439 newspa->spa_splitting_newspa = B_TRUE;
5441 /* create the new pool from the disks of the original pool */
5442 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
5444 newspa->spa_splitting_newspa = B_FALSE;
5449 /* if that worked, generate a real config for the new pool */
5450 if (newspa->spa_root_vdev != NULL) {
5451 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
5452 NV_UNIQUE_NAME, KM_SLEEP) == 0);
5453 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
5454 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
5455 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
5460 if (props != NULL) {
5461 spa_configfile_set(newspa, props, B_FALSE);
5462 error = spa_prop_set(newspa, props);
5467 /* flush everything */
5468 txg = spa_vdev_config_enter(newspa);
5469 vdev_config_dirty(newspa->spa_root_vdev);
5470 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
5472 if (zio_injection_enabled)
5473 zio_handle_panic_injection(spa, FTAG, 2);
5475 spa_async_resume(newspa);
5477 /* finally, update the original pool's config */
5478 txg = spa_vdev_config_enter(spa);
5479 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
5480 error = dmu_tx_assign(tx, TXG_WAIT);
5483 for (c = 0; c < children; c++) {
5484 if (vml[c] != NULL) {
5487 spa_history_log_internal(spa, "detach", tx,
5488 "vdev=%s", vml[c]->vdev_path);
5492 vdev_config_dirty(spa->spa_root_vdev);
5493 spa->spa_config_splitting = NULL;
5497 (void) spa_vdev_exit(spa, NULL, txg, 0);
5499 if (zio_injection_enabled)
5500 zio_handle_panic_injection(spa, FTAG, 3);
5502 /* split is complete; log a history record */
5503 spa_history_log_internal(newspa, "split", NULL,
5504 "from pool %s", spa_name(spa));
5506 kmem_free(vml, children * sizeof (vdev_t *));
5508 /* if we're not going to mount the filesystems in userland, export */
5510 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
5517 spa_deactivate(newspa);
5520 txg = spa_vdev_config_enter(spa);
5522 /* re-online all offlined disks */
5523 for (c = 0; c < children; c++) {
5525 vml[c]->vdev_offline = B_FALSE;
5527 vdev_reopen(spa->spa_root_vdev);
5529 nvlist_free(spa->spa_config_splitting);
5530 spa->spa_config_splitting = NULL;
5531 (void) spa_vdev_exit(spa, NULL, txg, error);
5533 kmem_free(vml, children * sizeof (vdev_t *));
5538 spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
5540 for (int i = 0; i < count; i++) {
5543 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
5546 if (guid == target_guid)
5554 spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
5555 nvlist_t *dev_to_remove)
5557 nvlist_t **newdev = NULL;
5560 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
5562 for (int i = 0, j = 0; i < count; i++) {
5563 if (dev[i] == dev_to_remove)
5565 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
5568 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
5569 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
5571 for (int i = 0; i < count - 1; i++)
5572 nvlist_free(newdev[i]);
5575 kmem_free(newdev, (count - 1) * sizeof (void *));
5579 * Evacuate the device.
5582 spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
5587 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5588 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5589 ASSERT(vd == vd->vdev_top);
5592 * Evacuate the device. We don't hold the config lock as writer
5593 * since we need to do I/O but we do keep the
5594 * spa_namespace_lock held. Once this completes the device
5595 * should no longer have any blocks allocated on it.
5597 if (vd->vdev_islog) {
5598 if (vd->vdev_stat.vs_alloc != 0)
5599 error = spa_offline_log(spa);
5601 error = SET_ERROR(ENOTSUP);
5608 * The evacuation succeeded. Remove any remaining MOS metadata
5609 * associated with this vdev, and wait for these changes to sync.
5611 ASSERT0(vd->vdev_stat.vs_alloc);
5612 txg = spa_vdev_config_enter(spa);
5613 vd->vdev_removing = B_TRUE;
5614 vdev_dirty_leaves(vd, VDD_DTL, txg);
5615 vdev_config_dirty(vd);
5616 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5622 * Complete the removal by cleaning up the namespace.
5625 spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
5627 vdev_t *rvd = spa->spa_root_vdev;
5628 uint64_t id = vd->vdev_id;
5629 boolean_t last_vdev = (id == (rvd->vdev_children - 1));
5631 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5632 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5633 ASSERT(vd == vd->vdev_top);
5636 * Only remove any devices which are empty.
5638 if (vd->vdev_stat.vs_alloc != 0)
5641 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5643 if (list_link_active(&vd->vdev_state_dirty_node))
5644 vdev_state_clean(vd);
5645 if (list_link_active(&vd->vdev_config_dirty_node))
5646 vdev_config_clean(vd);
5651 vdev_compact_children(rvd);
5653 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
5654 vdev_add_child(rvd, vd);
5656 vdev_config_dirty(rvd);
5659 * Reassess the health of our root vdev.
5665 * Remove a device from the pool -
5667 * Removing a device from the vdev namespace requires several steps
5668 * and can take a significant amount of time. As a result we use
5669 * the spa_vdev_config_[enter/exit] functions which allow us to
5670 * grab and release the spa_config_lock while still holding the namespace
5671 * lock. During each step the configuration is synced out.
5673 * Currently, this supports removing only hot spares, slogs, and level 2 ARC
5677 spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
5680 sysevent_t *ev = NULL;
5681 metaslab_group_t *mg;
5682 nvlist_t **spares, **l2cache, *nv;
5684 uint_t nspares, nl2cache;
5686 boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
5688 ASSERT(spa_writeable(spa));
5691 txg = spa_vdev_enter(spa);
5693 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5695 if (spa->spa_spares.sav_vdevs != NULL &&
5696 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5697 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
5698 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
5700 * Only remove the hot spare if it's not currently in use
5703 if (vd == NULL || unspare) {
5705 vd = spa_lookup_by_guid(spa, guid, B_TRUE);
5706 ev = spa_event_create(spa, vd, ESC_ZFS_VDEV_REMOVE_AUX);
5707 spa_vdev_remove_aux(spa->spa_spares.sav_config,
5708 ZPOOL_CONFIG_SPARES, spares, nspares, nv);
5709 spa_load_spares(spa);
5710 spa->spa_spares.sav_sync = B_TRUE;
5712 error = SET_ERROR(EBUSY);
5714 } else if (spa->spa_l2cache.sav_vdevs != NULL &&
5715 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5716 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
5717 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
5719 * Cache devices can always be removed.
5721 vd = spa_lookup_by_guid(spa, guid, B_TRUE);
5722 ev = spa_event_create(spa, vd, ESC_ZFS_VDEV_REMOVE_AUX);
5723 spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
5724 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
5725 spa_load_l2cache(spa);
5726 spa->spa_l2cache.sav_sync = B_TRUE;
5727 } else if (vd != NULL && vd->vdev_islog) {
5729 ASSERT(vd == vd->vdev_top);
5734 * Stop allocating from this vdev.
5736 metaslab_group_passivate(mg);
5739 * Wait for the youngest allocations and frees to sync,
5740 * and then wait for the deferral of those frees to finish.
5742 spa_vdev_config_exit(spa, NULL,
5743 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
5746 * Attempt to evacuate the vdev.
5748 error = spa_vdev_remove_evacuate(spa, vd);
5750 txg = spa_vdev_config_enter(spa);
5753 * If we couldn't evacuate the vdev, unwind.
5756 metaslab_group_activate(mg);
5757 return (spa_vdev_exit(spa, NULL, txg, error));
5761 * Clean up the vdev namespace.
5763 ev = spa_event_create(spa, vd, ESC_ZFS_VDEV_REMOVE_DEV);
5764 spa_vdev_remove_from_namespace(spa, vd);
5766 } else if (vd != NULL) {
5768 * Normal vdevs cannot be removed (yet).
5770 error = SET_ERROR(ENOTSUP);
5773 * There is no vdev of any kind with the specified guid.
5775 error = SET_ERROR(ENOENT);
5779 error = spa_vdev_exit(spa, NULL, txg, error);
5788 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5789 * currently spared, so we can detach it.
5792 spa_vdev_resilver_done_hunt(vdev_t *vd)
5794 vdev_t *newvd, *oldvd;
5796 for (int c = 0; c < vd->vdev_children; c++) {
5797 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
5803 * Check for a completed replacement. We always consider the first
5804 * vdev in the list to be the oldest vdev, and the last one to be
5805 * the newest (see spa_vdev_attach() for how that works). In
5806 * the case where the newest vdev is faulted, we will not automatically
5807 * remove it after a resilver completes. This is OK as it will require
5808 * user intervention to determine which disk the admin wishes to keep.
5810 if (vd->vdev_ops == &vdev_replacing_ops) {
5811 ASSERT(vd->vdev_children > 1);
5813 newvd = vd->vdev_child[vd->vdev_children - 1];
5814 oldvd = vd->vdev_child[0];
5816 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
5817 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5818 !vdev_dtl_required(oldvd))
5823 * Check for a completed resilver with the 'unspare' flag set.
5825 if (vd->vdev_ops == &vdev_spare_ops) {
5826 vdev_t *first = vd->vdev_child[0];
5827 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
5829 if (last->vdev_unspare) {
5832 } else if (first->vdev_unspare) {
5839 if (oldvd != NULL &&
5840 vdev_dtl_empty(newvd, DTL_MISSING) &&
5841 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5842 !vdev_dtl_required(oldvd))
5846 * If there are more than two spares attached to a disk,
5847 * and those spares are not required, then we want to
5848 * attempt to free them up now so that they can be used
5849 * by other pools. Once we're back down to a single
5850 * disk+spare, we stop removing them.
5852 if (vd->vdev_children > 2) {
5853 newvd = vd->vdev_child[1];
5855 if (newvd->vdev_isspare && last->vdev_isspare &&
5856 vdev_dtl_empty(last, DTL_MISSING) &&
5857 vdev_dtl_empty(last, DTL_OUTAGE) &&
5858 !vdev_dtl_required(newvd))
5867 spa_vdev_resilver_done(spa_t *spa)
5869 vdev_t *vd, *pvd, *ppvd;
5870 uint64_t guid, sguid, pguid, ppguid;
5872 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5874 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
5875 pvd = vd->vdev_parent;
5876 ppvd = pvd->vdev_parent;
5877 guid = vd->vdev_guid;
5878 pguid = pvd->vdev_guid;
5879 ppguid = ppvd->vdev_guid;
5882 * If we have just finished replacing a hot spared device, then
5883 * we need to detach the parent's first child (the original hot
5886 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
5887 ppvd->vdev_children == 2) {
5888 ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
5889 sguid = ppvd->vdev_child[1]->vdev_guid;
5891 ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd));
5893 spa_config_exit(spa, SCL_ALL, FTAG);
5894 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
5896 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
5898 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5901 spa_config_exit(spa, SCL_ALL, FTAG);
5905 * Update the stored path or FRU for this vdev.
5908 spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
5912 boolean_t sync = B_FALSE;
5914 ASSERT(spa_writeable(spa));
5916 spa_vdev_state_enter(spa, SCL_ALL);
5918 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
5919 return (spa_vdev_state_exit(spa, NULL, ENOENT));
5921 if (!vd->vdev_ops->vdev_op_leaf)
5922 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
5925 if (strcmp(value, vd->vdev_path) != 0) {
5926 spa_strfree(vd->vdev_path);
5927 vd->vdev_path = spa_strdup(value);
5931 if (vd->vdev_fru == NULL) {
5932 vd->vdev_fru = spa_strdup(value);
5934 } else if (strcmp(value, vd->vdev_fru) != 0) {
5935 spa_strfree(vd->vdev_fru);
5936 vd->vdev_fru = spa_strdup(value);
5941 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
5945 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
5947 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5951 spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5953 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5957 * ==========================================================================
5959 * ==========================================================================
5963 spa_scan_stop(spa_t *spa)
5965 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5966 if (dsl_scan_resilvering(spa->spa_dsl_pool))
5967 return (SET_ERROR(EBUSY));
5968 return (dsl_scan_cancel(spa->spa_dsl_pool));
5972 spa_scan(spa_t *spa, pool_scan_func_t func)
5974 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5976 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5977 return (SET_ERROR(ENOTSUP));
5980 * If a resilver was requested, but there is no DTL on a
5981 * writeable leaf device, we have nothing to do.
5983 if (func == POOL_SCAN_RESILVER &&
5984 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5985 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5989 return (dsl_scan(spa->spa_dsl_pool, func));
5993 * ==========================================================================
5994 * SPA async task processing
5995 * ==========================================================================
5999 spa_async_remove(spa_t *spa, vdev_t *vd)
6001 if (vd->vdev_remove_wanted) {
6002 vd->vdev_remove_wanted = B_FALSE;
6003 vd->vdev_delayed_close = B_FALSE;
6004 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
6007 * We want to clear the stats, but we don't want to do a full
6008 * vdev_clear() as that will cause us to throw away
6009 * degraded/faulted state as well as attempt to reopen the
6010 * device, all of which is a waste.
6012 vd->vdev_stat.vs_read_errors = 0;
6013 vd->vdev_stat.vs_write_errors = 0;
6014 vd->vdev_stat.vs_checksum_errors = 0;
6016 vdev_state_dirty(vd->vdev_top);
6017 /* Tell userspace that the vdev is gone. */
6018 zfs_post_remove(spa, vd);
6021 for (int c = 0; c < vd->vdev_children; c++)
6022 spa_async_remove(spa, vd->vdev_child[c]);
6026 spa_async_probe(spa_t *spa, vdev_t *vd)
6028 if (vd->vdev_probe_wanted) {
6029 vd->vdev_probe_wanted = B_FALSE;
6030 vdev_reopen(vd); /* vdev_open() does the actual probe */
6033 for (int c = 0; c < vd->vdev_children; c++)
6034 spa_async_probe(spa, vd->vdev_child[c]);
6038 spa_async_autoexpand(spa_t *spa, vdev_t *vd)
6044 if (!spa->spa_autoexpand)
6047 for (int c = 0; c < vd->vdev_children; c++) {
6048 vdev_t *cvd = vd->vdev_child[c];
6049 spa_async_autoexpand(spa, cvd);
6052 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
6055 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
6056 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
6058 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
6059 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
6061 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
6062 ESC_ZFS_VDEV_AUTOEXPAND, attr, &eid, DDI_SLEEP);
6065 kmem_free(physpath, MAXPATHLEN);
6069 spa_async_thread(void *arg)
6074 ASSERT(spa->spa_sync_on);
6076 mutex_enter(&spa->spa_async_lock);
6077 tasks = spa->spa_async_tasks;
6078 spa->spa_async_tasks &= SPA_ASYNC_REMOVE;
6079 mutex_exit(&spa->spa_async_lock);
6082 * See if the config needs to be updated.
6084 if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
6085 uint64_t old_space, new_space;
6087 mutex_enter(&spa_namespace_lock);
6088 old_space = metaslab_class_get_space(spa_normal_class(spa));
6089 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
6090 new_space = metaslab_class_get_space(spa_normal_class(spa));
6091 mutex_exit(&spa_namespace_lock);
6094 * If the pool grew as a result of the config update,
6095 * then log an internal history event.
6097 if (new_space != old_space) {
6098 spa_history_log_internal(spa, "vdev online", NULL,
6099 "pool '%s' size: %llu(+%llu)",
6100 spa_name(spa), new_space, new_space - old_space);
6104 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
6105 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
6106 spa_async_autoexpand(spa, spa->spa_root_vdev);
6107 spa_config_exit(spa, SCL_CONFIG, FTAG);
6111 * See if any devices need to be probed.
6113 if (tasks & SPA_ASYNC_PROBE) {
6114 spa_vdev_state_enter(spa, SCL_NONE);
6115 spa_async_probe(spa, spa->spa_root_vdev);
6116 (void) spa_vdev_state_exit(spa, NULL, 0);
6120 * If any devices are done replacing, detach them.
6122 if (tasks & SPA_ASYNC_RESILVER_DONE)
6123 spa_vdev_resilver_done(spa);
6126 * Kick off a resilver.
6128 if (tasks & SPA_ASYNC_RESILVER)
6129 dsl_resilver_restart(spa->spa_dsl_pool, 0);
6132 * Let the world know that we're done.
6134 mutex_enter(&spa->spa_async_lock);
6135 spa->spa_async_thread = NULL;
6136 cv_broadcast(&spa->spa_async_cv);
6137 mutex_exit(&spa->spa_async_lock);
6142 spa_async_thread_vd(void *arg)
6147 mutex_enter(&spa->spa_async_lock);
6148 tasks = spa->spa_async_tasks;
6150 spa->spa_async_tasks &= ~SPA_ASYNC_REMOVE;
6151 mutex_exit(&spa->spa_async_lock);
6154 * See if any devices need to be marked REMOVED.
6156 if (tasks & SPA_ASYNC_REMOVE) {
6157 spa_vdev_state_enter(spa, SCL_NONE);
6158 spa_async_remove(spa, spa->spa_root_vdev);
6159 for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
6160 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
6161 for (int i = 0; i < spa->spa_spares.sav_count; i++)
6162 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
6163 (void) spa_vdev_state_exit(spa, NULL, 0);
6167 * Let the world know that we're done.
6169 mutex_enter(&spa->spa_async_lock);
6170 tasks = spa->spa_async_tasks;
6171 if ((tasks & SPA_ASYNC_REMOVE) != 0)
6173 spa->spa_async_thread_vd = NULL;
6174 cv_broadcast(&spa->spa_async_cv);
6175 mutex_exit(&spa->spa_async_lock);
6180 spa_async_suspend(spa_t *spa)
6182 mutex_enter(&spa->spa_async_lock);
6183 spa->spa_async_suspended++;
6184 while (spa->spa_async_thread != NULL &&
6185 spa->spa_async_thread_vd != NULL)
6186 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
6187 mutex_exit(&spa->spa_async_lock);
6191 spa_async_resume(spa_t *spa)
6193 mutex_enter(&spa->spa_async_lock);
6194 ASSERT(spa->spa_async_suspended != 0);
6195 spa->spa_async_suspended--;
6196 mutex_exit(&spa->spa_async_lock);
6200 spa_async_tasks_pending(spa_t *spa)
6202 uint_t non_config_tasks;
6204 boolean_t config_task_suspended;
6206 non_config_tasks = spa->spa_async_tasks & ~(SPA_ASYNC_CONFIG_UPDATE |
6208 config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE;
6209 if (spa->spa_ccw_fail_time == 0) {
6210 config_task_suspended = B_FALSE;
6212 config_task_suspended =
6213 (gethrtime() - spa->spa_ccw_fail_time) <
6214 (zfs_ccw_retry_interval * NANOSEC);
6217 return (non_config_tasks || (config_task && !config_task_suspended));
6221 spa_async_dispatch(spa_t *spa)
6223 mutex_enter(&spa->spa_async_lock);
6224 if (spa_async_tasks_pending(spa) &&
6225 !spa->spa_async_suspended &&
6226 spa->spa_async_thread == NULL &&
6228 spa->spa_async_thread = thread_create(NULL, 0,
6229 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
6230 mutex_exit(&spa->spa_async_lock);
6234 spa_async_dispatch_vd(spa_t *spa)
6236 mutex_enter(&spa->spa_async_lock);
6237 if ((spa->spa_async_tasks & SPA_ASYNC_REMOVE) != 0 &&
6238 !spa->spa_async_suspended &&
6239 spa->spa_async_thread_vd == NULL &&
6241 spa->spa_async_thread_vd = thread_create(NULL, 0,
6242 spa_async_thread_vd, spa, 0, &p0, TS_RUN, maxclsyspri);
6243 mutex_exit(&spa->spa_async_lock);
6247 spa_async_request(spa_t *spa, int task)
6249 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
6250 mutex_enter(&spa->spa_async_lock);
6251 spa->spa_async_tasks |= task;
6252 mutex_exit(&spa->spa_async_lock);
6253 spa_async_dispatch_vd(spa);
6257 * ==========================================================================
6258 * SPA syncing routines
6259 * ==========================================================================
6263 bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
6266 bpobj_enqueue(bpo, bp, tx);
6271 spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
6275 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
6276 BP_GET_PSIZE(bp), zio->io_flags));
6281 * Note: this simple function is not inlined to make it easier to dtrace the
6282 * amount of time spent syncing frees.
6285 spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx)
6287 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6288 bplist_iterate(bpl, spa_free_sync_cb, zio, tx);
6289 VERIFY(zio_wait(zio) == 0);
6293 * Note: this simple function is not inlined to make it easier to dtrace the
6294 * amount of time spent syncing deferred frees.
6297 spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx)
6299 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6300 VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj,
6301 spa_free_sync_cb, zio, tx), ==, 0);
6302 VERIFY0(zio_wait(zio));
6307 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
6309 char *packed = NULL;
6314 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
6317 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
6318 * information. This avoids the dmu_buf_will_dirty() path and
6319 * saves us a pre-read to get data we don't actually care about.
6321 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
6322 packed = kmem_alloc(bufsize, KM_SLEEP);
6324 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
6326 bzero(packed + nvsize, bufsize - nvsize);
6328 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
6330 kmem_free(packed, bufsize);
6332 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
6333 dmu_buf_will_dirty(db, tx);
6334 *(uint64_t *)db->db_data = nvsize;
6335 dmu_buf_rele(db, FTAG);
6339 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
6340 const char *config, const char *entry)
6350 * Update the MOS nvlist describing the list of available devices.
6351 * spa_validate_aux() will have already made sure this nvlist is
6352 * valid and the vdevs are labeled appropriately.
6354 if (sav->sav_object == 0) {
6355 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
6356 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
6357 sizeof (uint64_t), tx);
6358 VERIFY(zap_update(spa->spa_meta_objset,
6359 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
6360 &sav->sav_object, tx) == 0);
6363 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
6364 if (sav->sav_count == 0) {
6365 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
6367 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
6368 for (i = 0; i < sav->sav_count; i++)
6369 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
6370 B_FALSE, VDEV_CONFIG_L2CACHE);
6371 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
6372 sav->sav_count) == 0);
6373 for (i = 0; i < sav->sav_count; i++)
6374 nvlist_free(list[i]);
6375 kmem_free(list, sav->sav_count * sizeof (void *));
6378 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
6379 nvlist_free(nvroot);
6381 sav->sav_sync = B_FALSE;
6385 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
6389 if (list_is_empty(&spa->spa_config_dirty_list))
6392 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6394 config = spa_config_generate(spa, spa->spa_root_vdev,
6395 dmu_tx_get_txg(tx), B_FALSE);
6398 * If we're upgrading the spa version then make sure that
6399 * the config object gets updated with the correct version.
6401 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
6402 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
6403 spa->spa_uberblock.ub_version);
6405 spa_config_exit(spa, SCL_STATE, FTAG);
6407 nvlist_free(spa->spa_config_syncing);
6408 spa->spa_config_syncing = config;
6410 spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
6414 spa_sync_version(void *arg, dmu_tx_t *tx)
6416 uint64_t *versionp = arg;
6417 uint64_t version = *versionp;
6418 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
6421 * Setting the version is special cased when first creating the pool.
6423 ASSERT(tx->tx_txg != TXG_INITIAL);
6425 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
6426 ASSERT(version >= spa_version(spa));
6428 spa->spa_uberblock.ub_version = version;
6429 vdev_config_dirty(spa->spa_root_vdev);
6430 spa_history_log_internal(spa, "set", tx, "version=%lld", version);
6434 * Set zpool properties.
6437 spa_sync_props(void *arg, dmu_tx_t *tx)
6439 nvlist_t *nvp = arg;
6440 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
6441 objset_t *mos = spa->spa_meta_objset;
6442 nvpair_t *elem = NULL;
6444 mutex_enter(&spa->spa_props_lock);
6446 while ((elem = nvlist_next_nvpair(nvp, elem))) {
6448 char *strval, *fname;
6450 const char *propname;
6451 zprop_type_t proptype;
6454 switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
6457 * We checked this earlier in spa_prop_validate().
6459 ASSERT(zpool_prop_feature(nvpair_name(elem)));
6461 fname = strchr(nvpair_name(elem), '@') + 1;
6462 VERIFY0(zfeature_lookup_name(fname, &fid));
6464 spa_feature_enable(spa, fid, tx);
6465 spa_history_log_internal(spa, "set", tx,
6466 "%s=enabled", nvpair_name(elem));
6469 case ZPOOL_PROP_VERSION:
6470 intval = fnvpair_value_uint64(elem);
6472 * The version is synced seperatly before other
6473 * properties and should be correct by now.
6475 ASSERT3U(spa_version(spa), >=, intval);
6478 case ZPOOL_PROP_ALTROOT:
6480 * 'altroot' is a non-persistent property. It should
6481 * have been set temporarily at creation or import time.
6483 ASSERT(spa->spa_root != NULL);
6486 case ZPOOL_PROP_READONLY:
6487 case ZPOOL_PROP_CACHEFILE:
6489 * 'readonly' and 'cachefile' are also non-persisitent
6493 case ZPOOL_PROP_COMMENT:
6494 strval = fnvpair_value_string(elem);
6495 if (spa->spa_comment != NULL)
6496 spa_strfree(spa->spa_comment);
6497 spa->spa_comment = spa_strdup(strval);
6499 * We need to dirty the configuration on all the vdevs
6500 * so that their labels get updated. It's unnecessary
6501 * to do this for pool creation since the vdev's
6502 * configuratoin has already been dirtied.
6504 if (tx->tx_txg != TXG_INITIAL)
6505 vdev_config_dirty(spa->spa_root_vdev);
6506 spa_history_log_internal(spa, "set", tx,
6507 "%s=%s", nvpair_name(elem), strval);
6511 * Set pool property values in the poolprops mos object.
6513 if (spa->spa_pool_props_object == 0) {
6514 spa->spa_pool_props_object =
6515 zap_create_link(mos, DMU_OT_POOL_PROPS,
6516 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
6520 /* normalize the property name */
6521 propname = zpool_prop_to_name(prop);
6522 proptype = zpool_prop_get_type(prop);
6524 if (nvpair_type(elem) == DATA_TYPE_STRING) {
6525 ASSERT(proptype == PROP_TYPE_STRING);
6526 strval = fnvpair_value_string(elem);
6527 VERIFY0(zap_update(mos,
6528 spa->spa_pool_props_object, propname,
6529 1, strlen(strval) + 1, strval, tx));
6530 spa_history_log_internal(spa, "set", tx,
6531 "%s=%s", nvpair_name(elem), strval);
6532 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
6533 intval = fnvpair_value_uint64(elem);
6535 if (proptype == PROP_TYPE_INDEX) {
6537 VERIFY0(zpool_prop_index_to_string(
6538 prop, intval, &unused));
6540 VERIFY0(zap_update(mos,
6541 spa->spa_pool_props_object, propname,
6542 8, 1, &intval, tx));
6543 spa_history_log_internal(spa, "set", tx,
6544 "%s=%lld", nvpair_name(elem), intval);
6546 ASSERT(0); /* not allowed */
6550 case ZPOOL_PROP_DELEGATION:
6551 spa->spa_delegation = intval;
6553 case ZPOOL_PROP_BOOTFS:
6554 spa->spa_bootfs = intval;
6556 case ZPOOL_PROP_FAILUREMODE:
6557 spa->spa_failmode = intval;
6559 case ZPOOL_PROP_AUTOEXPAND:
6560 spa->spa_autoexpand = intval;
6561 if (tx->tx_txg != TXG_INITIAL)
6562 spa_async_request(spa,
6563 SPA_ASYNC_AUTOEXPAND);
6565 case ZPOOL_PROP_DEDUPDITTO:
6566 spa->spa_dedup_ditto = intval;
6575 mutex_exit(&spa->spa_props_lock);
6579 * Perform one-time upgrade on-disk changes. spa_version() does not
6580 * reflect the new version this txg, so there must be no changes this
6581 * txg to anything that the upgrade code depends on after it executes.
6582 * Therefore this must be called after dsl_pool_sync() does the sync
6586 spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
6588 dsl_pool_t *dp = spa->spa_dsl_pool;
6590 ASSERT(spa->spa_sync_pass == 1);
6592 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
6594 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
6595 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
6596 dsl_pool_create_origin(dp, tx);
6598 /* Keeping the origin open increases spa_minref */
6599 spa->spa_minref += 3;
6602 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
6603 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
6604 dsl_pool_upgrade_clones(dp, tx);
6607 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
6608 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
6609 dsl_pool_upgrade_dir_clones(dp, tx);
6611 /* Keeping the freedir open increases spa_minref */
6612 spa->spa_minref += 3;
6615 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES &&
6616 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
6617 spa_feature_create_zap_objects(spa, tx);
6621 * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable
6622 * when possibility to use lz4 compression for metadata was added
6623 * Old pools that have this feature enabled must be upgraded to have
6624 * this feature active
6626 if (spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
6627 boolean_t lz4_en = spa_feature_is_enabled(spa,
6628 SPA_FEATURE_LZ4_COMPRESS);
6629 boolean_t lz4_ac = spa_feature_is_active(spa,
6630 SPA_FEATURE_LZ4_COMPRESS);
6632 if (lz4_en && !lz4_ac)
6633 spa_feature_incr(spa, SPA_FEATURE_LZ4_COMPRESS, tx);
6637 * If we haven't written the salt, do so now. Note that the
6638 * feature may not be activated yet, but that's fine since
6639 * the presence of this ZAP entry is backwards compatible.
6641 if (zap_contains(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
6642 DMU_POOL_CHECKSUM_SALT) == ENOENT) {
6643 VERIFY0(zap_add(spa->spa_meta_objset,
6644 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CHECKSUM_SALT, 1,
6645 sizeof (spa->spa_cksum_salt.zcs_bytes),
6646 spa->spa_cksum_salt.zcs_bytes, tx));
6649 rrw_exit(&dp->dp_config_rwlock, FTAG);
6653 * Sync the specified transaction group. New blocks may be dirtied as
6654 * part of the process, so we iterate until it converges.
6657 spa_sync(spa_t *spa, uint64_t txg)
6659 dsl_pool_t *dp = spa->spa_dsl_pool;
6660 objset_t *mos = spa->spa_meta_objset;
6661 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
6662 vdev_t *rvd = spa->spa_root_vdev;
6666 uint32_t max_queue_depth = zfs_vdev_async_write_max_active *
6667 zfs_vdev_queue_depth_pct / 100;
6669 VERIFY(spa_writeable(spa));
6672 * Lock out configuration changes.
6674 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
6676 spa->spa_syncing_txg = txg;
6677 spa->spa_sync_pass = 0;
6679 mutex_enter(&spa->spa_alloc_lock);
6680 VERIFY0(avl_numnodes(&spa->spa_alloc_tree));
6681 mutex_exit(&spa->spa_alloc_lock);
6684 * If there are any pending vdev state changes, convert them
6685 * into config changes that go out with this transaction group.
6687 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6688 while (list_head(&spa->spa_state_dirty_list) != NULL) {
6690 * We need the write lock here because, for aux vdevs,
6691 * calling vdev_config_dirty() modifies sav_config.
6692 * This is ugly and will become unnecessary when we
6693 * eliminate the aux vdev wart by integrating all vdevs
6694 * into the root vdev tree.
6696 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6697 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
6698 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
6699 vdev_state_clean(vd);
6700 vdev_config_dirty(vd);
6702 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6703 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
6705 spa_config_exit(spa, SCL_STATE, FTAG);
6707 tx = dmu_tx_create_assigned(dp, txg);
6709 spa->spa_sync_starttime = gethrtime();
6711 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid,
6712 spa->spa_sync_starttime + spa->spa_deadman_synctime));
6713 #else /* !illumos */
6715 callout_schedule(&spa->spa_deadman_cycid,
6716 hz * spa->spa_deadman_synctime / NANOSEC);
6718 #endif /* illumos */
6721 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
6722 * set spa_deflate if we have no raid-z vdevs.
6724 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
6725 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
6728 for (i = 0; i < rvd->vdev_children; i++) {
6729 vd = rvd->vdev_child[i];
6730 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
6733 if (i == rvd->vdev_children) {
6734 spa->spa_deflate = TRUE;
6735 VERIFY(0 == zap_add(spa->spa_meta_objset,
6736 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
6737 sizeof (uint64_t), 1, &spa->spa_deflate, tx));
6742 * Set the top-level vdev's max queue depth. Evaluate each
6743 * top-level's async write queue depth in case it changed.
6744 * The max queue depth will not change in the middle of syncing
6747 uint64_t queue_depth_total = 0;
6748 for (int c = 0; c < rvd->vdev_children; c++) {
6749 vdev_t *tvd = rvd->vdev_child[c];
6750 metaslab_group_t *mg = tvd->vdev_mg;
6752 if (mg == NULL || mg->mg_class != spa_normal_class(spa) ||
6753 !metaslab_group_initialized(mg))
6757 * It is safe to do a lock-free check here because only async
6758 * allocations look at mg_max_alloc_queue_depth, and async
6759 * allocations all happen from spa_sync().
6761 ASSERT0(refcount_count(&mg->mg_alloc_queue_depth));
6762 mg->mg_max_alloc_queue_depth = max_queue_depth;
6763 queue_depth_total += mg->mg_max_alloc_queue_depth;
6765 metaslab_class_t *mc = spa_normal_class(spa);
6766 ASSERT0(refcount_count(&mc->mc_alloc_slots));
6767 mc->mc_alloc_max_slots = queue_depth_total;
6768 mc->mc_alloc_throttle_enabled = zio_dva_throttle_enabled;
6770 ASSERT3U(mc->mc_alloc_max_slots, <=,
6771 max_queue_depth * rvd->vdev_children);
6774 * Iterate to convergence.
6777 int pass = ++spa->spa_sync_pass;
6779 spa_sync_config_object(spa, tx);
6780 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
6781 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
6782 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
6783 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
6784 spa_errlog_sync(spa, txg);
6785 dsl_pool_sync(dp, txg);
6787 if (pass < zfs_sync_pass_deferred_free) {
6788 spa_sync_frees(spa, free_bpl, tx);
6791 * We can not defer frees in pass 1, because
6792 * we sync the deferred frees later in pass 1.
6794 ASSERT3U(pass, >, 1);
6795 bplist_iterate(free_bpl, bpobj_enqueue_cb,
6796 &spa->spa_deferred_bpobj, tx);
6800 dsl_scan_sync(dp, tx);
6802 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
6806 spa_sync_upgrades(spa, tx);
6808 spa->spa_uberblock.ub_rootbp.blk_birth);
6810 * Note: We need to check if the MOS is dirty
6811 * because we could have marked the MOS dirty
6812 * without updating the uberblock (e.g. if we
6813 * have sync tasks but no dirty user data). We
6814 * need to check the uberblock's rootbp because
6815 * it is updated if we have synced out dirty
6816 * data (though in this case the MOS will most
6817 * likely also be dirty due to second order
6818 * effects, we don't want to rely on that here).
6820 if (spa->spa_uberblock.ub_rootbp.blk_birth < txg &&
6821 !dmu_objset_is_dirty(mos, txg)) {
6823 * Nothing changed on the first pass,
6824 * therefore this TXG is a no-op. Avoid
6825 * syncing deferred frees, so that we
6826 * can keep this TXG as a no-op.
6828 ASSERT(txg_list_empty(&dp->dp_dirty_datasets,
6830 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
6831 ASSERT(txg_list_empty(&dp->dp_sync_tasks, txg));
6834 spa_sync_deferred_frees(spa, tx);
6837 } while (dmu_objset_is_dirty(mos, txg));
6840 * Rewrite the vdev configuration (which includes the uberblock)
6841 * to commit the transaction group.
6843 * If there are no dirty vdevs, we sync the uberblock to a few
6844 * random top-level vdevs that are known to be visible in the
6845 * config cache (see spa_vdev_add() for a complete description).
6846 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
6850 * We hold SCL_STATE to prevent vdev open/close/etc.
6851 * while we're attempting to write the vdev labels.
6853 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6855 if (list_is_empty(&spa->spa_config_dirty_list)) {
6856 vdev_t *svd[SPA_DVAS_PER_BP];
6858 int children = rvd->vdev_children;
6859 int c0 = spa_get_random(children);
6861 for (int c = 0; c < children; c++) {
6862 vd = rvd->vdev_child[(c0 + c) % children];
6863 if (vd->vdev_ms_array == 0 || vd->vdev_islog)
6865 svd[svdcount++] = vd;
6866 if (svdcount == SPA_DVAS_PER_BP)
6869 error = vdev_config_sync(svd, svdcount, txg);
6871 error = vdev_config_sync(rvd->vdev_child,
6872 rvd->vdev_children, txg);
6876 spa->spa_last_synced_guid = rvd->vdev_guid;
6878 spa_config_exit(spa, SCL_STATE, FTAG);
6882 zio_suspend(spa, NULL);
6883 zio_resume_wait(spa);
6888 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY));
6889 #else /* !illumos */
6891 callout_drain(&spa->spa_deadman_cycid);
6893 #endif /* illumos */
6896 * Clear the dirty config list.
6898 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
6899 vdev_config_clean(vd);
6902 * Now that the new config has synced transactionally,
6903 * let it become visible to the config cache.
6905 if (spa->spa_config_syncing != NULL) {
6906 spa_config_set(spa, spa->spa_config_syncing);
6907 spa->spa_config_txg = txg;
6908 spa->spa_config_syncing = NULL;
6911 dsl_pool_sync_done(dp, txg);
6913 mutex_enter(&spa->spa_alloc_lock);
6914 VERIFY0(avl_numnodes(&spa->spa_alloc_tree));
6915 mutex_exit(&spa->spa_alloc_lock);
6918 * Update usable space statistics.
6920 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
6921 vdev_sync_done(vd, txg);
6923 spa_update_dspace(spa);
6926 * It had better be the case that we didn't dirty anything
6927 * since vdev_config_sync().
6929 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
6930 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
6931 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
6933 spa->spa_sync_pass = 0;
6936 * Update the last synced uberblock here. We want to do this at
6937 * the end of spa_sync() so that consumers of spa_last_synced_txg()
6938 * will be guaranteed that all the processing associated with
6939 * that txg has been completed.
6941 spa->spa_ubsync = spa->spa_uberblock;
6942 spa_config_exit(spa, SCL_CONFIG, FTAG);
6944 spa_handle_ignored_writes(spa);
6947 * If any async tasks have been requested, kick them off.
6949 spa_async_dispatch(spa);
6950 spa_async_dispatch_vd(spa);
6954 * Sync all pools. We don't want to hold the namespace lock across these
6955 * operations, so we take a reference on the spa_t and drop the lock during the
6959 spa_sync_allpools(void)
6962 mutex_enter(&spa_namespace_lock);
6963 while ((spa = spa_next(spa)) != NULL) {
6964 if (spa_state(spa) != POOL_STATE_ACTIVE ||
6965 !spa_writeable(spa) || spa_suspended(spa))
6967 spa_open_ref(spa, FTAG);
6968 mutex_exit(&spa_namespace_lock);
6969 txg_wait_synced(spa_get_dsl(spa), 0);
6970 mutex_enter(&spa_namespace_lock);
6971 spa_close(spa, FTAG);
6973 mutex_exit(&spa_namespace_lock);
6977 * ==========================================================================
6978 * Miscellaneous routines
6979 * ==========================================================================
6983 * Remove all pools in the system.
6991 * Remove all cached state. All pools should be closed now,
6992 * so every spa in the AVL tree should be unreferenced.
6994 mutex_enter(&spa_namespace_lock);
6995 while ((spa = spa_next(NULL)) != NULL) {
6997 * Stop async tasks. The async thread may need to detach
6998 * a device that's been replaced, which requires grabbing
6999 * spa_namespace_lock, so we must drop it here.
7001 spa_open_ref(spa, FTAG);
7002 mutex_exit(&spa_namespace_lock);
7003 spa_async_suspend(spa);
7004 mutex_enter(&spa_namespace_lock);
7005 spa_close(spa, FTAG);
7007 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
7009 spa_deactivate(spa);
7013 mutex_exit(&spa_namespace_lock);
7017 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
7022 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
7026 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
7027 vd = spa->spa_l2cache.sav_vdevs[i];
7028 if (vd->vdev_guid == guid)
7032 for (i = 0; i < spa->spa_spares.sav_count; i++) {
7033 vd = spa->spa_spares.sav_vdevs[i];
7034 if (vd->vdev_guid == guid)
7043 spa_upgrade(spa_t *spa, uint64_t version)
7045 ASSERT(spa_writeable(spa));
7047 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
7050 * This should only be called for a non-faulted pool, and since a
7051 * future version would result in an unopenable pool, this shouldn't be
7054 ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version));
7055 ASSERT3U(version, >=, spa->spa_uberblock.ub_version);
7057 spa->spa_uberblock.ub_version = version;
7058 vdev_config_dirty(spa->spa_root_vdev);
7060 spa_config_exit(spa, SCL_ALL, FTAG);
7062 txg_wait_synced(spa_get_dsl(spa), 0);
7066 spa_has_spare(spa_t *spa, uint64_t guid)
7070 spa_aux_vdev_t *sav = &spa->spa_spares;
7072 for (i = 0; i < sav->sav_count; i++)
7073 if (sav->sav_vdevs[i]->vdev_guid == guid)
7076 for (i = 0; i < sav->sav_npending; i++) {
7077 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
7078 &spareguid) == 0 && spareguid == guid)
7086 * Check if a pool has an active shared spare device.
7087 * Note: reference count of an active spare is 2, as a spare and as a replace
7090 spa_has_active_shared_spare(spa_t *spa)
7094 spa_aux_vdev_t *sav = &spa->spa_spares;
7096 for (i = 0; i < sav->sav_count; i++) {
7097 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
7098 &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
7107 spa_event_create(spa_t *spa, vdev_t *vd, const char *name)
7109 sysevent_t *ev = NULL;
7111 sysevent_attr_list_t *attr = NULL;
7112 sysevent_value_t value;
7114 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
7118 value.value_type = SE_DATA_TYPE_STRING;
7119 value.value.sv_string = spa_name(spa);
7120 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
7123 value.value_type = SE_DATA_TYPE_UINT64;
7124 value.value.sv_uint64 = spa_guid(spa);
7125 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
7129 value.value_type = SE_DATA_TYPE_UINT64;
7130 value.value.sv_uint64 = vd->vdev_guid;
7131 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
7135 if (vd->vdev_path) {
7136 value.value_type = SE_DATA_TYPE_STRING;
7137 value.value.sv_string = vd->vdev_path;
7138 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
7139 &value, SE_SLEEP) != 0)
7144 if (sysevent_attach_attributes(ev, attr) != 0)
7150 sysevent_free_attr(attr);
7157 spa_event_post(sysevent_t *ev)
7162 (void) log_sysevent(ev, SE_SLEEP, &eid);
7168 * Post a sysevent corresponding to the given event. The 'name' must be one of
7169 * the event definitions in sys/sysevent/eventdefs.h. The payload will be
7170 * filled in from the spa and (optionally) the vdev. This doesn't do anything
7171 * in the userland libzpool, as we don't want consumers to misinterpret ztest
7172 * or zdb as real changes.
7175 spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
7177 spa_event_post(spa_event_create(spa, vd, name));