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, 2020 by Delphix. All rights reserved.
25 * Copyright (c) 2018, Nexenta Systems, Inc. All rights reserved.
26 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
27 * Copyright 2013 Saso Kiselkov. All rights reserved.
28 * Copyright (c) 2014 Integros [integros.com]
29 * Copyright 2016 Toomas Soome <tsoome@me.com>
30 * Copyright (c) 2016 Actifio, Inc. All rights reserved.
31 * Copyright 2018 Joyent, Inc.
32 * Copyright (c) 2017, 2019, Datto Inc. All rights reserved.
33 * Copyright 2017 Joyent, Inc.
34 * Copyright (c) 2017, Intel Corporation.
38 * SPA: Storage Pool Allocator
40 * This file contains all the routines used when modifying on-disk SPA state.
41 * This includes opening, importing, destroying, exporting a pool, and syncing a
45 #include <sys/zfs_context.h>
46 #include <sys/fm/fs/zfs.h>
47 #include <sys/spa_impl.h>
49 #include <sys/zio_checksum.h>
51 #include <sys/dmu_tx.h>
55 #include <sys/vdev_impl.h>
56 #include <sys/vdev_removal.h>
57 #include <sys/vdev_indirect_mapping.h>
58 #include <sys/vdev_indirect_births.h>
59 #include <sys/vdev_initialize.h>
60 #include <sys/vdev_rebuild.h>
61 #include <sys/vdev_trim.h>
62 #include <sys/vdev_disk.h>
63 #include <sys/metaslab.h>
64 #include <sys/metaslab_impl.h>
66 #include <sys/uberblock_impl.h>
69 #include <sys/bpobj.h>
70 #include <sys/dmu_traverse.h>
71 #include <sys/dmu_objset.h>
72 #include <sys/unique.h>
73 #include <sys/dsl_pool.h>
74 #include <sys/dsl_dataset.h>
75 #include <sys/dsl_dir.h>
76 #include <sys/dsl_prop.h>
77 #include <sys/dsl_synctask.h>
78 #include <sys/fs/zfs.h>
80 #include <sys/callb.h>
81 #include <sys/systeminfo.h>
82 #include <sys/spa_boot.h>
83 #include <sys/zfs_ioctl.h>
84 #include <sys/dsl_scan.h>
85 #include <sys/zfeature.h>
86 #include <sys/dsl_destroy.h>
90 #include <sys/fm/protocol.h>
91 #include <sys/fm/util.h>
92 #include <sys/callb.h>
94 #include <sys/vmsystm.h>
98 #include "zfs_comutil.h"
101 * The interval, in seconds, at which failed configuration cache file writes
104 int zfs_ccw_retry_interval = 300;
106 typedef enum zti_modes {
107 ZTI_MODE_FIXED, /* value is # of threads (min 1) */
108 ZTI_MODE_BATCH, /* cpu-intensive; value is ignored */
109 ZTI_MODE_NULL, /* don't create a taskq */
113 #define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) }
114 #define ZTI_PCT(n) { ZTI_MODE_ONLINE_PERCENT, (n), 1 }
115 #define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 }
116 #define ZTI_NULL { ZTI_MODE_NULL, 0, 0 }
118 #define ZTI_N(n) ZTI_P(n, 1)
119 #define ZTI_ONE ZTI_N(1)
121 typedef struct zio_taskq_info {
122 zti_modes_t zti_mode;
127 static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
128 "iss", "iss_h", "int", "int_h"
132 * This table defines the taskq settings for each ZFS I/O type. When
133 * initializing a pool, we use this table to create an appropriately sized
134 * taskq. Some operations are low volume and therefore have a small, static
135 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE
136 * macros. Other operations process a large amount of data; the ZTI_BATCH
137 * macro causes us to create a taskq oriented for throughput. Some operations
138 * are so high frequency and short-lived that the taskq itself can become a
139 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an
140 * additional degree of parallelism specified by the number of threads per-
141 * taskq and the number of taskqs; when dispatching an event in this case, the
142 * particular taskq is chosen at random.
144 * The different taskq priorities are to handle the different contexts (issue
145 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that
146 * need to be handled with minimum delay.
148 const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
149 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
150 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* NULL */
151 { ZTI_N(8), ZTI_NULL, ZTI_P(12, 8), ZTI_NULL }, /* READ */
152 { ZTI_BATCH, ZTI_N(5), ZTI_P(12, 8), ZTI_N(5) }, /* WRITE */
153 { ZTI_P(12, 8), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FREE */
154 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* CLAIM */
155 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* IOCTL */
156 { ZTI_N(4), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* TRIM */
159 static void spa_sync_version(void *arg, dmu_tx_t *tx);
160 static void spa_sync_props(void *arg, dmu_tx_t *tx);
161 static boolean_t spa_has_active_shared_spare(spa_t *spa);
162 static int spa_load_impl(spa_t *spa, spa_import_type_t type, char **ereport);
163 static void spa_vdev_resilver_done(spa_t *spa);
165 uint_t zio_taskq_batch_pct = 75; /* 1 thread per cpu in pset */
166 boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */
167 uint_t zio_taskq_basedc = 80; /* base duty cycle */
169 boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */
172 * Report any spa_load_verify errors found, but do not fail spa_load.
173 * This is used by zdb to analyze non-idle pools.
175 boolean_t spa_load_verify_dryrun = B_FALSE;
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 * For debugging purposes: print out vdev tree during pool import.
186 int spa_load_print_vdev_tree = B_FALSE;
189 * A non-zero value for zfs_max_missing_tvds means that we allow importing
190 * pools with missing top-level vdevs. This is strictly intended for advanced
191 * pool recovery cases since missing data is almost inevitable. Pools with
192 * missing devices can only be imported read-only for safety reasons, and their
193 * fail-mode will be automatically set to "continue".
195 * With 1 missing vdev we should be able to import the pool and mount all
196 * datasets. User data that was not modified after the missing device has been
197 * added should be recoverable. This means that snapshots created prior to the
198 * addition of that device should be completely intact.
200 * With 2 missing vdevs, some datasets may fail to mount since there are
201 * dataset statistics that are stored as regular metadata. Some data might be
202 * recoverable if those vdevs were added recently.
204 * With 3 or more missing vdevs, the pool is severely damaged and MOS entries
205 * may be missing entirely. Chances of data recovery are very low. Note that
206 * there are also risks of performing an inadvertent rewind as we might be
207 * missing all the vdevs with the latest uberblocks.
209 unsigned long zfs_max_missing_tvds = 0;
212 * The parameters below are similar to zfs_max_missing_tvds but are only
213 * intended for a preliminary open of the pool with an untrusted config which
214 * might be incomplete or out-dated.
216 * We are more tolerant for pools opened from a cachefile since we could have
217 * an out-dated cachefile where a device removal was not registered.
218 * We could have set the limit arbitrarily high but in the case where devices
219 * are really missing we would want to return the proper error codes; we chose
220 * SPA_DVAS_PER_BP - 1 so that some copies of the MOS would still be available
221 * and we get a chance to retrieve the trusted config.
223 uint64_t zfs_max_missing_tvds_cachefile = SPA_DVAS_PER_BP - 1;
226 * In the case where config was assembled by scanning device paths (/dev/dsks
227 * by default) we are less tolerant since all the existing devices should have
228 * been detected and we want spa_load to return the right error codes.
230 uint64_t zfs_max_missing_tvds_scan = 0;
233 * Debugging aid that pauses spa_sync() towards the end.
235 boolean_t zfs_pause_spa_sync = B_FALSE;
238 * Variables to indicate the livelist condense zthr func should wait at certain
239 * points for the livelist to be removed - used to test condense/destroy races
241 int zfs_livelist_condense_zthr_pause = 0;
242 int zfs_livelist_condense_sync_pause = 0;
245 * Variables to track whether or not condense cancellation has been
246 * triggered in testing.
248 int zfs_livelist_condense_sync_cancel = 0;
249 int zfs_livelist_condense_zthr_cancel = 0;
252 * Variable to track whether or not extra ALLOC blkptrs were added to a
253 * livelist entry while it was being condensed (caused by the way we track
254 * remapped blkptrs in dbuf_remap_impl)
256 int zfs_livelist_condense_new_alloc = 0;
259 * ==========================================================================
260 * SPA properties routines
261 * ==========================================================================
265 * Add a (source=src, propname=propval) list to an nvlist.
268 spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
269 uint64_t intval, zprop_source_t src)
271 const char *propname = zpool_prop_to_name(prop);
274 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
275 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
278 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
280 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
282 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
283 nvlist_free(propval);
287 * Get property values from the spa configuration.
290 spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
292 vdev_t *rvd = spa->spa_root_vdev;
293 dsl_pool_t *pool = spa->spa_dsl_pool;
294 uint64_t size, alloc, cap, version;
295 const zprop_source_t src = ZPROP_SRC_NONE;
296 spa_config_dirent_t *dp;
297 metaslab_class_t *mc = spa_normal_class(spa);
299 ASSERT(MUTEX_HELD(&spa->spa_props_lock));
302 alloc = metaslab_class_get_alloc(mc);
303 alloc += metaslab_class_get_alloc(spa_special_class(spa));
304 alloc += metaslab_class_get_alloc(spa_dedup_class(spa));
306 size = metaslab_class_get_space(mc);
307 size += metaslab_class_get_space(spa_special_class(spa));
308 size += metaslab_class_get_space(spa_dedup_class(spa));
310 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
311 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
312 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
313 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
315 spa_prop_add_list(*nvp, ZPOOL_PROP_CHECKPOINT, NULL,
316 spa->spa_checkpoint_info.sci_dspace, src);
318 spa_prop_add_list(*nvp, ZPOOL_PROP_FRAGMENTATION, NULL,
319 metaslab_class_fragmentation(mc), src);
320 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL,
321 metaslab_class_expandable_space(mc), src);
322 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
323 (spa_mode(spa) == SPA_MODE_READ), src);
325 cap = (size == 0) ? 0 : (alloc * 100 / size);
326 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
328 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
329 ddt_get_pool_dedup_ratio(spa), src);
331 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
332 rvd->vdev_state, src);
334 version = spa_version(spa);
335 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION)) {
336 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL,
337 version, ZPROP_SRC_DEFAULT);
339 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL,
340 version, ZPROP_SRC_LOCAL);
342 spa_prop_add_list(*nvp, ZPOOL_PROP_LOAD_GUID,
343 NULL, spa_load_guid(spa), src);
348 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
349 * when opening pools before this version freedir will be NULL.
351 if (pool->dp_free_dir != NULL) {
352 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
353 dsl_dir_phys(pool->dp_free_dir)->dd_used_bytes,
356 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
360 if (pool->dp_leak_dir != NULL) {
361 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, NULL,
362 dsl_dir_phys(pool->dp_leak_dir)->dd_used_bytes,
365 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED,
370 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
372 if (spa->spa_comment != NULL) {
373 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
377 if (spa->spa_root != NULL)
378 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
381 if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) {
382 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL,
383 MIN(zfs_max_recordsize, SPA_MAXBLOCKSIZE), ZPROP_SRC_NONE);
385 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL,
386 SPA_OLD_MAXBLOCKSIZE, ZPROP_SRC_NONE);
389 if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_DNODE)) {
390 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXDNODESIZE, NULL,
391 DNODE_MAX_SIZE, ZPROP_SRC_NONE);
393 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXDNODESIZE, NULL,
394 DNODE_MIN_SIZE, ZPROP_SRC_NONE);
397 if ((dp = list_head(&spa->spa_config_list)) != NULL) {
398 if (dp->scd_path == NULL) {
399 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
400 "none", 0, ZPROP_SRC_LOCAL);
401 } else if (strcmp(dp->scd_path, spa_config_path) != 0) {
402 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
403 dp->scd_path, 0, ZPROP_SRC_LOCAL);
409 * Get zpool property values.
412 spa_prop_get(spa_t *spa, nvlist_t **nvp)
414 objset_t *mos = spa->spa_meta_objset;
420 err = nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP);
424 dp = spa_get_dsl(spa);
425 dsl_pool_config_enter(dp, FTAG);
426 mutex_enter(&spa->spa_props_lock);
429 * Get properties from the spa config.
431 spa_prop_get_config(spa, nvp);
433 /* If no pool property object, no more prop to get. */
434 if (mos == NULL || spa->spa_pool_props_object == 0)
438 * Get properties from the MOS pool property object.
440 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
441 (err = zap_cursor_retrieve(&zc, &za)) == 0;
442 zap_cursor_advance(&zc)) {
445 zprop_source_t src = ZPROP_SRC_DEFAULT;
448 if ((prop = zpool_name_to_prop(za.za_name)) == ZPOOL_PROP_INVAL)
451 switch (za.za_integer_length) {
453 /* integer property */
454 if (za.za_first_integer !=
455 zpool_prop_default_numeric(prop))
456 src = ZPROP_SRC_LOCAL;
458 if (prop == ZPOOL_PROP_BOOTFS) {
459 dsl_dataset_t *ds = NULL;
461 err = dsl_dataset_hold_obj(dp,
462 za.za_first_integer, FTAG, &ds);
466 strval = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN,
468 dsl_dataset_name(ds, strval);
469 dsl_dataset_rele(ds, FTAG);
472 intval = za.za_first_integer;
475 spa_prop_add_list(*nvp, prop, strval, intval, src);
478 kmem_free(strval, ZFS_MAX_DATASET_NAME_LEN);
483 /* string property */
484 strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
485 err = zap_lookup(mos, spa->spa_pool_props_object,
486 za.za_name, 1, za.za_num_integers, strval);
488 kmem_free(strval, za.za_num_integers);
491 spa_prop_add_list(*nvp, prop, strval, 0, src);
492 kmem_free(strval, za.za_num_integers);
499 zap_cursor_fini(&zc);
501 mutex_exit(&spa->spa_props_lock);
502 dsl_pool_config_exit(dp, FTAG);
503 if (err && err != ENOENT) {
513 * Validate the given pool properties nvlist and modify the list
514 * for the property values to be set.
517 spa_prop_validate(spa_t *spa, nvlist_t *props)
520 int error = 0, reset_bootfs = 0;
522 boolean_t has_feature = B_FALSE;
525 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
527 char *strval, *slash, *check, *fname;
528 const char *propname = nvpair_name(elem);
529 zpool_prop_t prop = zpool_name_to_prop(propname);
532 case ZPOOL_PROP_INVAL:
533 if (!zpool_prop_feature(propname)) {
534 error = SET_ERROR(EINVAL);
539 * Sanitize the input.
541 if (nvpair_type(elem) != DATA_TYPE_UINT64) {
542 error = SET_ERROR(EINVAL);
546 if (nvpair_value_uint64(elem, &intval) != 0) {
547 error = SET_ERROR(EINVAL);
552 error = SET_ERROR(EINVAL);
556 fname = strchr(propname, '@') + 1;
557 if (zfeature_lookup_name(fname, NULL) != 0) {
558 error = SET_ERROR(EINVAL);
562 has_feature = B_TRUE;
565 case ZPOOL_PROP_VERSION:
566 error = nvpair_value_uint64(elem, &intval);
568 (intval < spa_version(spa) ||
569 intval > SPA_VERSION_BEFORE_FEATURES ||
571 error = SET_ERROR(EINVAL);
574 case ZPOOL_PROP_DELEGATION:
575 case ZPOOL_PROP_AUTOREPLACE:
576 case ZPOOL_PROP_LISTSNAPS:
577 case ZPOOL_PROP_AUTOEXPAND:
578 case ZPOOL_PROP_AUTOTRIM:
579 error = nvpair_value_uint64(elem, &intval);
580 if (!error && intval > 1)
581 error = SET_ERROR(EINVAL);
584 case ZPOOL_PROP_MULTIHOST:
585 error = nvpair_value_uint64(elem, &intval);
586 if (!error && intval > 1)
587 error = SET_ERROR(EINVAL);
590 uint32_t hostid = zone_get_hostid(NULL);
592 spa->spa_hostid = hostid;
594 error = SET_ERROR(ENOTSUP);
599 case ZPOOL_PROP_BOOTFS:
601 * If the pool version is less than SPA_VERSION_BOOTFS,
602 * or the pool is still being created (version == 0),
603 * the bootfs property cannot be set.
605 if (spa_version(spa) < SPA_VERSION_BOOTFS) {
606 error = SET_ERROR(ENOTSUP);
611 * Make sure the vdev config is bootable
613 if (!vdev_is_bootable(spa->spa_root_vdev)) {
614 error = SET_ERROR(ENOTSUP);
620 error = nvpair_value_string(elem, &strval);
625 if (strval == NULL || strval[0] == '\0') {
626 objnum = zpool_prop_default_numeric(
631 error = dmu_objset_hold(strval, FTAG, &os);
636 if (dmu_objset_type(os) != DMU_OST_ZFS) {
637 error = SET_ERROR(ENOTSUP);
639 objnum = dmu_objset_id(os);
641 dmu_objset_rele(os, FTAG);
645 case ZPOOL_PROP_FAILUREMODE:
646 error = nvpair_value_uint64(elem, &intval);
647 if (!error && intval > ZIO_FAILURE_MODE_PANIC)
648 error = SET_ERROR(EINVAL);
651 * This is a special case which only occurs when
652 * the pool has completely failed. This allows
653 * the user to change the in-core failmode property
654 * without syncing it out to disk (I/Os might
655 * currently be blocked). We do this by returning
656 * EIO to the caller (spa_prop_set) to trick it
657 * into thinking we encountered a property validation
660 if (!error && spa_suspended(spa)) {
661 spa->spa_failmode = intval;
662 error = SET_ERROR(EIO);
666 case ZPOOL_PROP_CACHEFILE:
667 if ((error = nvpair_value_string(elem, &strval)) != 0)
670 if (strval[0] == '\0')
673 if (strcmp(strval, "none") == 0)
676 if (strval[0] != '/') {
677 error = SET_ERROR(EINVAL);
681 slash = strrchr(strval, '/');
682 ASSERT(slash != NULL);
684 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
685 strcmp(slash, "/..") == 0)
686 error = SET_ERROR(EINVAL);
689 case ZPOOL_PROP_COMMENT:
690 if ((error = nvpair_value_string(elem, &strval)) != 0)
692 for (check = strval; *check != '\0'; check++) {
693 if (!isprint(*check)) {
694 error = SET_ERROR(EINVAL);
698 if (strlen(strval) > ZPROP_MAX_COMMENT)
699 error = SET_ERROR(E2BIG);
710 (void) nvlist_remove_all(props,
711 zpool_prop_to_name(ZPOOL_PROP_DEDUPDITTO));
713 if (!error && reset_bootfs) {
714 error = nvlist_remove(props,
715 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
718 error = nvlist_add_uint64(props,
719 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
727 spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
730 spa_config_dirent_t *dp;
732 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
736 dp = kmem_alloc(sizeof (spa_config_dirent_t),
739 if (cachefile[0] == '\0')
740 dp->scd_path = spa_strdup(spa_config_path);
741 else if (strcmp(cachefile, "none") == 0)
744 dp->scd_path = spa_strdup(cachefile);
746 list_insert_head(&spa->spa_config_list, dp);
748 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
752 spa_prop_set(spa_t *spa, nvlist_t *nvp)
755 nvpair_t *elem = NULL;
756 boolean_t need_sync = B_FALSE;
758 if ((error = spa_prop_validate(spa, nvp)) != 0)
761 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
762 zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem));
764 if (prop == ZPOOL_PROP_CACHEFILE ||
765 prop == ZPOOL_PROP_ALTROOT ||
766 prop == ZPOOL_PROP_READONLY)
769 if (prop == ZPOOL_PROP_VERSION || prop == ZPOOL_PROP_INVAL) {
772 if (prop == ZPOOL_PROP_VERSION) {
773 VERIFY(nvpair_value_uint64(elem, &ver) == 0);
775 ASSERT(zpool_prop_feature(nvpair_name(elem)));
776 ver = SPA_VERSION_FEATURES;
780 /* Save time if the version is already set. */
781 if (ver == spa_version(spa))
785 * In addition to the pool directory object, we might
786 * create the pool properties object, the features for
787 * read object, the features for write object, or the
788 * feature descriptions object.
790 error = dsl_sync_task(spa->spa_name, NULL,
791 spa_sync_version, &ver,
792 6, ZFS_SPACE_CHECK_RESERVED);
803 return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props,
804 nvp, 6, ZFS_SPACE_CHECK_RESERVED));
811 * If the bootfs property value is dsobj, clear it.
814 spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
816 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
817 VERIFY(zap_remove(spa->spa_meta_objset,
818 spa->spa_pool_props_object,
819 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
826 spa_change_guid_check(void *arg, dmu_tx_t *tx)
828 uint64_t *newguid __maybe_unused = arg;
829 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
830 vdev_t *rvd = spa->spa_root_vdev;
833 if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) {
834 int error = (spa_has_checkpoint(spa)) ?
835 ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT;
836 return (SET_ERROR(error));
839 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
840 vdev_state = rvd->vdev_state;
841 spa_config_exit(spa, SCL_STATE, FTAG);
843 if (vdev_state != VDEV_STATE_HEALTHY)
844 return (SET_ERROR(ENXIO));
846 ASSERT3U(spa_guid(spa), !=, *newguid);
852 spa_change_guid_sync(void *arg, dmu_tx_t *tx)
854 uint64_t *newguid = arg;
855 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
857 vdev_t *rvd = spa->spa_root_vdev;
859 oldguid = spa_guid(spa);
861 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
862 rvd->vdev_guid = *newguid;
863 rvd->vdev_guid_sum += (*newguid - oldguid);
864 vdev_config_dirty(rvd);
865 spa_config_exit(spa, SCL_STATE, FTAG);
867 spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu",
868 (u_longlong_t)oldguid, (u_longlong_t)*newguid);
872 * Change the GUID for the pool. This is done so that we can later
873 * re-import a pool built from a clone of our own vdevs. We will modify
874 * the root vdev's guid, our own pool guid, and then mark all of our
875 * vdevs dirty. Note that we must make sure that all our vdevs are
876 * online when we do this, or else any vdevs that weren't present
877 * would be orphaned from our pool. We are also going to issue a
878 * sysevent to update any watchers.
881 spa_change_guid(spa_t *spa)
886 mutex_enter(&spa->spa_vdev_top_lock);
887 mutex_enter(&spa_namespace_lock);
888 guid = spa_generate_guid(NULL);
890 error = dsl_sync_task(spa->spa_name, spa_change_guid_check,
891 spa_change_guid_sync, &guid, 5, ZFS_SPACE_CHECK_RESERVED);
894 spa_write_cachefile(spa, B_FALSE, B_TRUE);
895 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_REGUID);
898 mutex_exit(&spa_namespace_lock);
899 mutex_exit(&spa->spa_vdev_top_lock);
905 * ==========================================================================
906 * SPA state manipulation (open/create/destroy/import/export)
907 * ==========================================================================
911 spa_error_entry_compare(const void *a, const void *b)
913 const spa_error_entry_t *sa = (const spa_error_entry_t *)a;
914 const spa_error_entry_t *sb = (const spa_error_entry_t *)b;
917 ret = memcmp(&sa->se_bookmark, &sb->se_bookmark,
918 sizeof (zbookmark_phys_t));
920 return (TREE_ISIGN(ret));
924 * Utility function which retrieves copies of the current logs and
925 * re-initializes them in the process.
928 spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
930 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
932 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
933 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
935 avl_create(&spa->spa_errlist_scrub,
936 spa_error_entry_compare, sizeof (spa_error_entry_t),
937 offsetof(spa_error_entry_t, se_avl));
938 avl_create(&spa->spa_errlist_last,
939 spa_error_entry_compare, sizeof (spa_error_entry_t),
940 offsetof(spa_error_entry_t, se_avl));
944 spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
946 const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
947 enum zti_modes mode = ztip->zti_mode;
948 uint_t value = ztip->zti_value;
949 uint_t count = ztip->zti_count;
950 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
952 boolean_t batch = B_FALSE;
954 if (mode == ZTI_MODE_NULL) {
956 tqs->stqs_taskq = NULL;
960 ASSERT3U(count, >, 0);
962 tqs->stqs_count = count;
963 tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP);
967 ASSERT3U(value, >=, 1);
968 value = MAX(value, 1);
969 flags |= TASKQ_DYNAMIC;
974 flags |= TASKQ_THREADS_CPU_PCT;
975 value = MIN(zio_taskq_batch_pct, 100);
979 panic("unrecognized mode for %s_%s taskq (%u:%u) in "
981 zio_type_name[t], zio_taskq_types[q], mode, value);
985 for (uint_t i = 0; i < count; i++) {
989 (void) snprintf(name, sizeof (name), "%s_%s",
990 zio_type_name[t], zio_taskq_types[q]);
992 if (zio_taskq_sysdc && spa->spa_proc != &p0) {
994 flags |= TASKQ_DC_BATCH;
996 tq = taskq_create_sysdc(name, value, 50, INT_MAX,
997 spa->spa_proc, zio_taskq_basedc, flags);
999 pri_t pri = maxclsyspri;
1001 * The write issue taskq can be extremely CPU
1002 * intensive. Run it at slightly less important
1003 * priority than the other taskqs.
1005 * Under Linux and FreeBSD this means incrementing
1006 * the priority value as opposed to platforms like
1007 * illumos where it should be decremented.
1009 * On FreeBSD, if priorities divided by four (RQ_PPQ)
1010 * are equal then a difference between them is
1013 if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE) {
1014 #if defined(__linux__)
1016 #elif defined(__FreeBSD__)
1022 tq = taskq_create_proc(name, value, pri, 50,
1023 INT_MAX, spa->spa_proc, flags);
1026 tqs->stqs_taskq[i] = tq;
1031 spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
1033 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
1035 if (tqs->stqs_taskq == NULL) {
1036 ASSERT3U(tqs->stqs_count, ==, 0);
1040 for (uint_t i = 0; i < tqs->stqs_count; i++) {
1041 ASSERT3P(tqs->stqs_taskq[i], !=, NULL);
1042 taskq_destroy(tqs->stqs_taskq[i]);
1045 kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *));
1046 tqs->stqs_taskq = NULL;
1050 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
1051 * Note that a type may have multiple discrete taskqs to avoid lock contention
1052 * on the taskq itself. In that case we choose which taskq at random by using
1053 * the low bits of gethrtime().
1056 spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q,
1057 task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent)
1059 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
1062 ASSERT3P(tqs->stqs_taskq, !=, NULL);
1063 ASSERT3U(tqs->stqs_count, !=, 0);
1065 if (tqs->stqs_count == 1) {
1066 tq = tqs->stqs_taskq[0];
1068 tq = tqs->stqs_taskq[((uint64_t)gethrtime()) % tqs->stqs_count];
1071 taskq_dispatch_ent(tq, func, arg, flags, ent);
1075 * Same as spa_taskq_dispatch_ent() but block on the task until completion.
1078 spa_taskq_dispatch_sync(spa_t *spa, zio_type_t t, zio_taskq_type_t q,
1079 task_func_t *func, void *arg, uint_t flags)
1081 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
1085 ASSERT3P(tqs->stqs_taskq, !=, NULL);
1086 ASSERT3U(tqs->stqs_count, !=, 0);
1088 if (tqs->stqs_count == 1) {
1089 tq = tqs->stqs_taskq[0];
1091 tq = tqs->stqs_taskq[((uint64_t)gethrtime()) % tqs->stqs_count];
1094 id = taskq_dispatch(tq, func, arg, flags);
1096 taskq_wait_id(tq, id);
1100 spa_create_zio_taskqs(spa_t *spa)
1102 for (int t = 0; t < ZIO_TYPES; t++) {
1103 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1104 spa_taskqs_init(spa, t, q);
1110 * Disabled until spa_thread() can be adapted for Linux.
1112 #undef HAVE_SPA_THREAD
1114 #if defined(_KERNEL) && defined(HAVE_SPA_THREAD)
1116 spa_thread(void *arg)
1118 psetid_t zio_taskq_psrset_bind = PS_NONE;
1119 callb_cpr_t cprinfo;
1122 user_t *pu = PTOU(curproc);
1124 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
1127 ASSERT(curproc != &p0);
1128 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
1129 "zpool-%s", spa->spa_name);
1130 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
1132 /* bind this thread to the requested psrset */
1133 if (zio_taskq_psrset_bind != PS_NONE) {
1135 mutex_enter(&cpu_lock);
1136 mutex_enter(&pidlock);
1137 mutex_enter(&curproc->p_lock);
1139 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
1140 0, NULL, NULL) == 0) {
1141 curthread->t_bind_pset = zio_taskq_psrset_bind;
1144 "Couldn't bind process for zfs pool \"%s\" to "
1145 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
1148 mutex_exit(&curproc->p_lock);
1149 mutex_exit(&pidlock);
1150 mutex_exit(&cpu_lock);
1154 if (zio_taskq_sysdc) {
1155 sysdc_thread_enter(curthread, 100, 0);
1158 spa->spa_proc = curproc;
1159 spa->spa_did = curthread->t_did;
1161 spa_create_zio_taskqs(spa);
1163 mutex_enter(&spa->spa_proc_lock);
1164 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
1166 spa->spa_proc_state = SPA_PROC_ACTIVE;
1167 cv_broadcast(&spa->spa_proc_cv);
1169 CALLB_CPR_SAFE_BEGIN(&cprinfo);
1170 while (spa->spa_proc_state == SPA_PROC_ACTIVE)
1171 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1172 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
1174 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
1175 spa->spa_proc_state = SPA_PROC_GONE;
1176 spa->spa_proc = &p0;
1177 cv_broadcast(&spa->spa_proc_cv);
1178 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */
1180 mutex_enter(&curproc->p_lock);
1186 * Activate an uninitialized pool.
1189 spa_activate(spa_t *spa, spa_mode_t mode)
1191 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
1193 spa->spa_state = POOL_STATE_ACTIVE;
1194 spa->spa_mode = mode;
1196 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
1197 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
1198 spa->spa_special_class = metaslab_class_create(spa, zfs_metaslab_ops);
1199 spa->spa_dedup_class = metaslab_class_create(spa, zfs_metaslab_ops);
1201 /* Try to create a covering process */
1202 mutex_enter(&spa->spa_proc_lock);
1203 ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
1204 ASSERT(spa->spa_proc == &p0);
1207 #ifdef HAVE_SPA_THREAD
1208 /* Only create a process if we're going to be around a while. */
1209 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
1210 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
1212 spa->spa_proc_state = SPA_PROC_CREATED;
1213 while (spa->spa_proc_state == SPA_PROC_CREATED) {
1214 cv_wait(&spa->spa_proc_cv,
1215 &spa->spa_proc_lock);
1217 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1218 ASSERT(spa->spa_proc != &p0);
1219 ASSERT(spa->spa_did != 0);
1223 "Couldn't create process for zfs pool \"%s\"\n",
1228 #endif /* HAVE_SPA_THREAD */
1229 mutex_exit(&spa->spa_proc_lock);
1231 /* If we didn't create a process, we need to create our taskqs. */
1232 if (spa->spa_proc == &p0) {
1233 spa_create_zio_taskqs(spa);
1236 for (size_t i = 0; i < TXG_SIZE; i++) {
1237 spa->spa_txg_zio[i] = zio_root(spa, NULL, NULL,
1241 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
1242 offsetof(vdev_t, vdev_config_dirty_node));
1243 list_create(&spa->spa_evicting_os_list, sizeof (objset_t),
1244 offsetof(objset_t, os_evicting_node));
1245 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
1246 offsetof(vdev_t, vdev_state_dirty_node));
1248 txg_list_create(&spa->spa_vdev_txg_list, spa,
1249 offsetof(struct vdev, vdev_txg_node));
1251 avl_create(&spa->spa_errlist_scrub,
1252 spa_error_entry_compare, sizeof (spa_error_entry_t),
1253 offsetof(spa_error_entry_t, se_avl));
1254 avl_create(&spa->spa_errlist_last,
1255 spa_error_entry_compare, sizeof (spa_error_entry_t),
1256 offsetof(spa_error_entry_t, se_avl));
1258 spa_keystore_init(&spa->spa_keystore);
1261 * This taskq is used to perform zvol-minor-related tasks
1262 * asynchronously. This has several advantages, including easy
1263 * resolution of various deadlocks (zfsonlinux bug #3681).
1265 * The taskq must be single threaded to ensure tasks are always
1266 * processed in the order in which they were dispatched.
1268 * A taskq per pool allows one to keep the pools independent.
1269 * This way if one pool is suspended, it will not impact another.
1271 * The preferred location to dispatch a zvol minor task is a sync
1272 * task. In this context, there is easy access to the spa_t and minimal
1273 * error handling is required because the sync task must succeed.
1275 spa->spa_zvol_taskq = taskq_create("z_zvol", 1, defclsyspri,
1279 * Taskq dedicated to prefetcher threads: this is used to prevent the
1280 * pool traverse code from monopolizing the global (and limited)
1281 * system_taskq by inappropriately scheduling long running tasks on it.
1283 spa->spa_prefetch_taskq = taskq_create("z_prefetch", boot_ncpus,
1284 defclsyspri, 1, INT_MAX, TASKQ_DYNAMIC);
1287 * The taskq to upgrade datasets in this pool. Currently used by
1288 * feature SPA_FEATURE_USEROBJ_ACCOUNTING/SPA_FEATURE_PROJECT_QUOTA.
1290 spa->spa_upgrade_taskq = taskq_create("z_upgrade", boot_ncpus,
1291 defclsyspri, 1, INT_MAX, TASKQ_DYNAMIC);
1295 * Opposite of spa_activate().
1298 spa_deactivate(spa_t *spa)
1300 ASSERT(spa->spa_sync_on == B_FALSE);
1301 ASSERT(spa->spa_dsl_pool == NULL);
1302 ASSERT(spa->spa_root_vdev == NULL);
1303 ASSERT(spa->spa_async_zio_root == NULL);
1304 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
1306 spa_evicting_os_wait(spa);
1308 if (spa->spa_zvol_taskq) {
1309 taskq_destroy(spa->spa_zvol_taskq);
1310 spa->spa_zvol_taskq = NULL;
1313 if (spa->spa_prefetch_taskq) {
1314 taskq_destroy(spa->spa_prefetch_taskq);
1315 spa->spa_prefetch_taskq = NULL;
1318 if (spa->spa_upgrade_taskq) {
1319 taskq_destroy(spa->spa_upgrade_taskq);
1320 spa->spa_upgrade_taskq = NULL;
1323 txg_list_destroy(&spa->spa_vdev_txg_list);
1325 list_destroy(&spa->spa_config_dirty_list);
1326 list_destroy(&spa->spa_evicting_os_list);
1327 list_destroy(&spa->spa_state_dirty_list);
1329 taskq_cancel_id(system_delay_taskq, spa->spa_deadman_tqid);
1331 for (int t = 0; t < ZIO_TYPES; t++) {
1332 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1333 spa_taskqs_fini(spa, t, q);
1337 for (size_t i = 0; i < TXG_SIZE; i++) {
1338 ASSERT3P(spa->spa_txg_zio[i], !=, NULL);
1339 VERIFY0(zio_wait(spa->spa_txg_zio[i]));
1340 spa->spa_txg_zio[i] = NULL;
1343 metaslab_class_destroy(spa->spa_normal_class);
1344 spa->spa_normal_class = NULL;
1346 metaslab_class_destroy(spa->spa_log_class);
1347 spa->spa_log_class = NULL;
1349 metaslab_class_destroy(spa->spa_special_class);
1350 spa->spa_special_class = NULL;
1352 metaslab_class_destroy(spa->spa_dedup_class);
1353 spa->spa_dedup_class = NULL;
1356 * If this was part of an import or the open otherwise failed, we may
1357 * still have errors left in the queues. Empty them just in case.
1359 spa_errlog_drain(spa);
1360 avl_destroy(&spa->spa_errlist_scrub);
1361 avl_destroy(&spa->spa_errlist_last);
1363 spa_keystore_fini(&spa->spa_keystore);
1365 spa->spa_state = POOL_STATE_UNINITIALIZED;
1367 mutex_enter(&spa->spa_proc_lock);
1368 if (spa->spa_proc_state != SPA_PROC_NONE) {
1369 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1370 spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1371 cv_broadcast(&spa->spa_proc_cv);
1372 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1373 ASSERT(spa->spa_proc != &p0);
1374 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1376 ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1377 spa->spa_proc_state = SPA_PROC_NONE;
1379 ASSERT(spa->spa_proc == &p0);
1380 mutex_exit(&spa->spa_proc_lock);
1383 * We want to make sure spa_thread() has actually exited the ZFS
1384 * module, so that the module can't be unloaded out from underneath
1387 if (spa->spa_did != 0) {
1388 thread_join(spa->spa_did);
1394 * Verify a pool configuration, and construct the vdev tree appropriately. This
1395 * will create all the necessary vdevs in the appropriate layout, with each vdev
1396 * in the CLOSED state. This will prep the pool before open/creation/import.
1397 * All vdev validation is done by the vdev_alloc() routine.
1400 spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1401 uint_t id, int atype)
1407 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1410 if ((*vdp)->vdev_ops->vdev_op_leaf)
1413 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1416 if (error == ENOENT)
1422 return (SET_ERROR(EINVAL));
1425 for (int c = 0; c < children; c++) {
1427 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1435 ASSERT(*vdp != NULL);
1441 spa_should_flush_logs_on_unload(spa_t *spa)
1443 if (!spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP))
1446 if (!spa_writeable(spa))
1449 if (!spa->spa_sync_on)
1452 if (spa_state(spa) != POOL_STATE_EXPORTED)
1455 if (zfs_keep_log_spacemaps_at_export)
1462 * Opens a transaction that will set the flag that will instruct
1463 * spa_sync to attempt to flush all the metaslabs for that txg.
1466 spa_unload_log_sm_flush_all(spa_t *spa)
1468 dmu_tx_t *tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
1469 VERIFY0(dmu_tx_assign(tx, TXG_WAIT));
1471 ASSERT3U(spa->spa_log_flushall_txg, ==, 0);
1472 spa->spa_log_flushall_txg = dmu_tx_get_txg(tx);
1475 txg_wait_synced(spa_get_dsl(spa), spa->spa_log_flushall_txg);
1479 spa_unload_log_sm_metadata(spa_t *spa)
1481 void *cookie = NULL;
1483 while ((sls = avl_destroy_nodes(&spa->spa_sm_logs_by_txg,
1484 &cookie)) != NULL) {
1485 VERIFY0(sls->sls_mscount);
1486 kmem_free(sls, sizeof (spa_log_sm_t));
1489 for (log_summary_entry_t *e = list_head(&spa->spa_log_summary);
1490 e != NULL; e = list_head(&spa->spa_log_summary)) {
1491 VERIFY0(e->lse_mscount);
1492 list_remove(&spa->spa_log_summary, e);
1493 kmem_free(e, sizeof (log_summary_entry_t));
1496 spa->spa_unflushed_stats.sus_nblocks = 0;
1497 spa->spa_unflushed_stats.sus_memused = 0;
1498 spa->spa_unflushed_stats.sus_blocklimit = 0;
1502 spa_destroy_aux_threads(spa_t *spa)
1504 if (spa->spa_condense_zthr != NULL) {
1505 zthr_destroy(spa->spa_condense_zthr);
1506 spa->spa_condense_zthr = NULL;
1508 if (spa->spa_checkpoint_discard_zthr != NULL) {
1509 zthr_destroy(spa->spa_checkpoint_discard_zthr);
1510 spa->spa_checkpoint_discard_zthr = NULL;
1512 if (spa->spa_livelist_delete_zthr != NULL) {
1513 zthr_destroy(spa->spa_livelist_delete_zthr);
1514 spa->spa_livelist_delete_zthr = NULL;
1516 if (spa->spa_livelist_condense_zthr != NULL) {
1517 zthr_destroy(spa->spa_livelist_condense_zthr);
1518 spa->spa_livelist_condense_zthr = NULL;
1523 * Opposite of spa_load().
1526 spa_unload(spa_t *spa)
1528 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1529 ASSERT(spa_state(spa) != POOL_STATE_UNINITIALIZED);
1531 spa_import_progress_remove(spa_guid(spa));
1532 spa_load_note(spa, "UNLOADING");
1534 spa_wake_waiters(spa);
1537 * If the log space map feature is enabled and the pool is getting
1538 * exported (but not destroyed), we want to spend some time flushing
1539 * as many metaslabs as we can in an attempt to destroy log space
1540 * maps and save import time.
1542 if (spa_should_flush_logs_on_unload(spa))
1543 spa_unload_log_sm_flush_all(spa);
1548 spa_async_suspend(spa);
1550 if (spa->spa_root_vdev) {
1551 vdev_t *root_vdev = spa->spa_root_vdev;
1552 vdev_initialize_stop_all(root_vdev, VDEV_INITIALIZE_ACTIVE);
1553 vdev_trim_stop_all(root_vdev, VDEV_TRIM_ACTIVE);
1554 vdev_autotrim_stop_all(spa);
1555 vdev_rebuild_stop_all(spa);
1561 if (spa->spa_sync_on) {
1562 txg_sync_stop(spa->spa_dsl_pool);
1563 spa->spa_sync_on = B_FALSE;
1567 * This ensures that there is no async metaslab prefetching
1568 * while we attempt to unload the spa.
1570 if (spa->spa_root_vdev != NULL) {
1571 for (int c = 0; c < spa->spa_root_vdev->vdev_children; c++) {
1572 vdev_t *vc = spa->spa_root_vdev->vdev_child[c];
1573 if (vc->vdev_mg != NULL)
1574 taskq_wait(vc->vdev_mg->mg_taskq);
1578 if (spa->spa_mmp.mmp_thread)
1579 mmp_thread_stop(spa);
1582 * Wait for any outstanding async I/O to complete.
1584 if (spa->spa_async_zio_root != NULL) {
1585 for (int i = 0; i < max_ncpus; i++)
1586 (void) zio_wait(spa->spa_async_zio_root[i]);
1587 kmem_free(spa->spa_async_zio_root, max_ncpus * sizeof (void *));
1588 spa->spa_async_zio_root = NULL;
1591 if (spa->spa_vdev_removal != NULL) {
1592 spa_vdev_removal_destroy(spa->spa_vdev_removal);
1593 spa->spa_vdev_removal = NULL;
1596 spa_destroy_aux_threads(spa);
1598 spa_condense_fini(spa);
1600 bpobj_close(&spa->spa_deferred_bpobj);
1602 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
1607 if (spa->spa_root_vdev)
1608 vdev_free(spa->spa_root_vdev);
1609 ASSERT(spa->spa_root_vdev == NULL);
1612 * Close the dsl pool.
1614 if (spa->spa_dsl_pool) {
1615 dsl_pool_close(spa->spa_dsl_pool);
1616 spa->spa_dsl_pool = NULL;
1617 spa->spa_meta_objset = NULL;
1621 spa_unload_log_sm_metadata(spa);
1624 * Drop and purge level 2 cache
1626 spa_l2cache_drop(spa);
1628 for (int i = 0; i < spa->spa_spares.sav_count; i++)
1629 vdev_free(spa->spa_spares.sav_vdevs[i]);
1630 if (spa->spa_spares.sav_vdevs) {
1631 kmem_free(spa->spa_spares.sav_vdevs,
1632 spa->spa_spares.sav_count * sizeof (void *));
1633 spa->spa_spares.sav_vdevs = NULL;
1635 if (spa->spa_spares.sav_config) {
1636 nvlist_free(spa->spa_spares.sav_config);
1637 spa->spa_spares.sav_config = NULL;
1639 spa->spa_spares.sav_count = 0;
1641 for (int i = 0; i < spa->spa_l2cache.sav_count; i++) {
1642 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1643 vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1645 if (spa->spa_l2cache.sav_vdevs) {
1646 kmem_free(spa->spa_l2cache.sav_vdevs,
1647 spa->spa_l2cache.sav_count * sizeof (void *));
1648 spa->spa_l2cache.sav_vdevs = NULL;
1650 if (spa->spa_l2cache.sav_config) {
1651 nvlist_free(spa->spa_l2cache.sav_config);
1652 spa->spa_l2cache.sav_config = NULL;
1654 spa->spa_l2cache.sav_count = 0;
1656 spa->spa_async_suspended = 0;
1658 spa->spa_indirect_vdevs_loaded = B_FALSE;
1660 if (spa->spa_comment != NULL) {
1661 spa_strfree(spa->spa_comment);
1662 spa->spa_comment = NULL;
1665 spa_config_exit(spa, SCL_ALL, spa);
1669 * Load (or re-load) the current list of vdevs describing the active spares for
1670 * this pool. When this is called, we have some form of basic information in
1671 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1672 * then re-generate a more complete list including status information.
1675 spa_load_spares(spa_t *spa)
1684 * zdb opens both the current state of the pool and the
1685 * checkpointed state (if present), with a different spa_t.
1687 * As spare vdevs are shared among open pools, we skip loading
1688 * them when we load the checkpointed state of the pool.
1690 if (!spa_writeable(spa))
1694 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1697 * First, close and free any existing spare vdevs.
1699 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1700 vd = spa->spa_spares.sav_vdevs[i];
1702 /* Undo the call to spa_activate() below */
1703 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1704 B_FALSE)) != NULL && tvd->vdev_isspare)
1705 spa_spare_remove(tvd);
1710 if (spa->spa_spares.sav_vdevs)
1711 kmem_free(spa->spa_spares.sav_vdevs,
1712 spa->spa_spares.sav_count * sizeof (void *));
1714 if (spa->spa_spares.sav_config == NULL)
1717 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1718 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1720 spa->spa_spares.sav_count = (int)nspares;
1721 spa->spa_spares.sav_vdevs = NULL;
1727 * Construct the array of vdevs, opening them to get status in the
1728 * process. For each spare, there is potentially two different vdev_t
1729 * structures associated with it: one in the list of spares (used only
1730 * for basic validation purposes) and one in the active vdev
1731 * configuration (if it's spared in). During this phase we open and
1732 * validate each vdev on the spare list. If the vdev also exists in the
1733 * active configuration, then we also mark this vdev as an active spare.
1735 spa->spa_spares.sav_vdevs = kmem_zalloc(nspares * sizeof (void *),
1737 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1738 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1739 VDEV_ALLOC_SPARE) == 0);
1742 spa->spa_spares.sav_vdevs[i] = vd;
1744 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1745 B_FALSE)) != NULL) {
1746 if (!tvd->vdev_isspare)
1750 * We only mark the spare active if we were successfully
1751 * able to load the vdev. Otherwise, importing a pool
1752 * with a bad active spare would result in strange
1753 * behavior, because multiple pool would think the spare
1754 * is actively in use.
1756 * There is a vulnerability here to an equally bizarre
1757 * circumstance, where a dead active spare is later
1758 * brought back to life (onlined or otherwise). Given
1759 * the rarity of this scenario, and the extra complexity
1760 * it adds, we ignore the possibility.
1762 if (!vdev_is_dead(tvd))
1763 spa_spare_activate(tvd);
1767 vd->vdev_aux = &spa->spa_spares;
1769 if (vdev_open(vd) != 0)
1772 if (vdev_validate_aux(vd) == 0)
1777 * Recompute the stashed list of spares, with status information
1780 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1781 DATA_TYPE_NVLIST_ARRAY) == 0);
1783 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1785 for (i = 0; i < spa->spa_spares.sav_count; i++)
1786 spares[i] = vdev_config_generate(spa,
1787 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1788 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1789 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1790 for (i = 0; i < spa->spa_spares.sav_count; i++)
1791 nvlist_free(spares[i]);
1792 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1796 * Load (or re-load) the current list of vdevs describing the active l2cache for
1797 * this pool. When this is called, we have some form of basic information in
1798 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1799 * then re-generate a more complete list including status information.
1800 * Devices which are already active have their details maintained, and are
1804 spa_load_l2cache(spa_t *spa)
1806 nvlist_t **l2cache = NULL;
1808 int i, j, oldnvdevs;
1810 vdev_t *vd, **oldvdevs, **newvdevs;
1811 spa_aux_vdev_t *sav = &spa->spa_l2cache;
1815 * zdb opens both the current state of the pool and the
1816 * checkpointed state (if present), with a different spa_t.
1818 * As L2 caches are part of the ARC which is shared among open
1819 * pools, we skip loading them when we load the checkpointed
1820 * state of the pool.
1822 if (!spa_writeable(spa))
1826 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1828 oldvdevs = sav->sav_vdevs;
1829 oldnvdevs = sav->sav_count;
1830 sav->sav_vdevs = NULL;
1833 if (sav->sav_config == NULL) {
1839 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1840 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1841 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1844 * Process new nvlist of vdevs.
1846 for (i = 0; i < nl2cache; i++) {
1847 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1851 for (j = 0; j < oldnvdevs; j++) {
1853 if (vd != NULL && guid == vd->vdev_guid) {
1855 * Retain previous vdev for add/remove ops.
1863 if (newvdevs[i] == NULL) {
1867 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1868 VDEV_ALLOC_L2CACHE) == 0);
1873 * Commit this vdev as an l2cache device,
1874 * even if it fails to open.
1876 spa_l2cache_add(vd);
1881 spa_l2cache_activate(vd);
1883 if (vdev_open(vd) != 0)
1886 (void) vdev_validate_aux(vd);
1888 if (!vdev_is_dead(vd))
1889 l2arc_add_vdev(spa, vd);
1892 * Upon cache device addition to a pool or pool
1893 * creation with a cache device or if the header
1894 * of the device is invalid we issue an async
1895 * TRIM command for the whole device which will
1896 * execute if l2arc_trim_ahead > 0.
1898 spa_async_request(spa, SPA_ASYNC_L2CACHE_TRIM);
1902 sav->sav_vdevs = newvdevs;
1903 sav->sav_count = (int)nl2cache;
1906 * Recompute the stashed list of l2cache devices, with status
1907 * information this time.
1909 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1910 DATA_TYPE_NVLIST_ARRAY) == 0);
1912 if (sav->sav_count > 0)
1913 l2cache = kmem_alloc(sav->sav_count * sizeof (void *),
1915 for (i = 0; i < sav->sav_count; i++)
1916 l2cache[i] = vdev_config_generate(spa,
1917 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1918 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1919 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1923 * Purge vdevs that were dropped
1925 for (i = 0; i < oldnvdevs; i++) {
1930 ASSERT(vd->vdev_isl2cache);
1932 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1933 pool != 0ULL && l2arc_vdev_present(vd))
1934 l2arc_remove_vdev(vd);
1935 vdev_clear_stats(vd);
1941 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1943 for (i = 0; i < sav->sav_count; i++)
1944 nvlist_free(l2cache[i]);
1946 kmem_free(l2cache, sav->sav_count * sizeof (void *));
1950 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1953 char *packed = NULL;
1958 error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db);
1962 nvsize = *(uint64_t *)db->db_data;
1963 dmu_buf_rele(db, FTAG);
1965 packed = vmem_alloc(nvsize, KM_SLEEP);
1966 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1969 error = nvlist_unpack(packed, nvsize, value, 0);
1970 vmem_free(packed, nvsize);
1976 * Concrete top-level vdevs that are not missing and are not logs. At every
1977 * spa_sync we write new uberblocks to at least SPA_SYNC_MIN_VDEVS core tvds.
1980 spa_healthy_core_tvds(spa_t *spa)
1982 vdev_t *rvd = spa->spa_root_vdev;
1985 for (uint64_t i = 0; i < rvd->vdev_children; i++) {
1986 vdev_t *vd = rvd->vdev_child[i];
1989 if (vdev_is_concrete(vd) && !vdev_is_dead(vd))
1997 * Checks to see if the given vdev could not be opened, in which case we post a
1998 * sysevent to notify the autoreplace code that the device has been removed.
2001 spa_check_removed(vdev_t *vd)
2003 for (uint64_t c = 0; c < vd->vdev_children; c++)
2004 spa_check_removed(vd->vdev_child[c]);
2006 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) &&
2007 vdev_is_concrete(vd)) {
2008 zfs_post_autoreplace(vd->vdev_spa, vd);
2009 spa_event_notify(vd->vdev_spa, vd, NULL, ESC_ZFS_VDEV_CHECK);
2014 spa_check_for_missing_logs(spa_t *spa)
2016 vdev_t *rvd = spa->spa_root_vdev;
2019 * If we're doing a normal import, then build up any additional
2020 * diagnostic information about missing log devices.
2021 * We'll pass this up to the user for further processing.
2023 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
2024 nvlist_t **child, *nv;
2027 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t *),
2029 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2031 for (uint64_t c = 0; c < rvd->vdev_children; c++) {
2032 vdev_t *tvd = rvd->vdev_child[c];
2035 * We consider a device as missing only if it failed
2036 * to open (i.e. offline or faulted is not considered
2039 if (tvd->vdev_islog &&
2040 tvd->vdev_state == VDEV_STATE_CANT_OPEN) {
2041 child[idx++] = vdev_config_generate(spa, tvd,
2042 B_FALSE, VDEV_CONFIG_MISSING);
2047 fnvlist_add_nvlist_array(nv,
2048 ZPOOL_CONFIG_CHILDREN, child, idx);
2049 fnvlist_add_nvlist(spa->spa_load_info,
2050 ZPOOL_CONFIG_MISSING_DEVICES, nv);
2052 for (uint64_t i = 0; i < idx; i++)
2053 nvlist_free(child[i]);
2056 kmem_free(child, rvd->vdev_children * sizeof (char **));
2059 spa_load_failed(spa, "some log devices are missing");
2060 vdev_dbgmsg_print_tree(rvd, 2);
2061 return (SET_ERROR(ENXIO));
2064 for (uint64_t c = 0; c < rvd->vdev_children; c++) {
2065 vdev_t *tvd = rvd->vdev_child[c];
2067 if (tvd->vdev_islog &&
2068 tvd->vdev_state == VDEV_STATE_CANT_OPEN) {
2069 spa_set_log_state(spa, SPA_LOG_CLEAR);
2070 spa_load_note(spa, "some log devices are "
2071 "missing, ZIL is dropped.");
2072 vdev_dbgmsg_print_tree(rvd, 2);
2082 * Check for missing log devices
2085 spa_check_logs(spa_t *spa)
2087 boolean_t rv = B_FALSE;
2088 dsl_pool_t *dp = spa_get_dsl(spa);
2090 switch (spa->spa_log_state) {
2093 case SPA_LOG_MISSING:
2094 /* need to recheck in case slog has been restored */
2095 case SPA_LOG_UNKNOWN:
2096 rv = (dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
2097 zil_check_log_chain, NULL, DS_FIND_CHILDREN) != 0);
2099 spa_set_log_state(spa, SPA_LOG_MISSING);
2106 spa_passivate_log(spa_t *spa)
2108 vdev_t *rvd = spa->spa_root_vdev;
2109 boolean_t slog_found = B_FALSE;
2111 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
2113 if (!spa_has_slogs(spa))
2116 for (int c = 0; c < rvd->vdev_children; c++) {
2117 vdev_t *tvd = rvd->vdev_child[c];
2118 metaslab_group_t *mg = tvd->vdev_mg;
2120 if (tvd->vdev_islog) {
2121 metaslab_group_passivate(mg);
2122 slog_found = B_TRUE;
2126 return (slog_found);
2130 spa_activate_log(spa_t *spa)
2132 vdev_t *rvd = spa->spa_root_vdev;
2134 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
2136 for (int c = 0; c < rvd->vdev_children; c++) {
2137 vdev_t *tvd = rvd->vdev_child[c];
2138 metaslab_group_t *mg = tvd->vdev_mg;
2140 if (tvd->vdev_islog)
2141 metaslab_group_activate(mg);
2146 spa_reset_logs(spa_t *spa)
2150 error = dmu_objset_find(spa_name(spa), zil_reset,
2151 NULL, DS_FIND_CHILDREN);
2154 * We successfully offlined the log device, sync out the
2155 * current txg so that the "stubby" block can be removed
2158 txg_wait_synced(spa->spa_dsl_pool, 0);
2164 spa_aux_check_removed(spa_aux_vdev_t *sav)
2166 for (int i = 0; i < sav->sav_count; i++)
2167 spa_check_removed(sav->sav_vdevs[i]);
2171 spa_claim_notify(zio_t *zio)
2173 spa_t *spa = zio->io_spa;
2178 mutex_enter(&spa->spa_props_lock); /* any mutex will do */
2179 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
2180 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
2181 mutex_exit(&spa->spa_props_lock);
2184 typedef struct spa_load_error {
2185 uint64_t sle_meta_count;
2186 uint64_t sle_data_count;
2190 spa_load_verify_done(zio_t *zio)
2192 blkptr_t *bp = zio->io_bp;
2193 spa_load_error_t *sle = zio->io_private;
2194 dmu_object_type_t type = BP_GET_TYPE(bp);
2195 int error = zio->io_error;
2196 spa_t *spa = zio->io_spa;
2198 abd_free(zio->io_abd);
2200 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
2201 type != DMU_OT_INTENT_LOG)
2202 atomic_inc_64(&sle->sle_meta_count);
2204 atomic_inc_64(&sle->sle_data_count);
2207 mutex_enter(&spa->spa_scrub_lock);
2208 spa->spa_load_verify_bytes -= BP_GET_PSIZE(bp);
2209 cv_broadcast(&spa->spa_scrub_io_cv);
2210 mutex_exit(&spa->spa_scrub_lock);
2214 * Maximum number of inflight bytes is the log2 fraction of the arc size.
2215 * By default, we set it to 1/16th of the arc.
2217 int spa_load_verify_shift = 4;
2218 int spa_load_verify_metadata = B_TRUE;
2219 int spa_load_verify_data = B_TRUE;
2223 spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
2224 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
2226 if (zb->zb_level == ZB_DNODE_LEVEL || BP_IS_HOLE(bp) ||
2227 BP_IS_EMBEDDED(bp) || BP_IS_REDACTED(bp))
2230 * Note: normally this routine will not be called if
2231 * spa_load_verify_metadata is not set. However, it may be useful
2232 * to manually set the flag after the traversal has begun.
2234 if (!spa_load_verify_metadata)
2236 if (!BP_IS_METADATA(bp) && !spa_load_verify_data)
2239 uint64_t maxinflight_bytes =
2240 arc_target_bytes() >> spa_load_verify_shift;
2242 size_t size = BP_GET_PSIZE(bp);
2244 mutex_enter(&spa->spa_scrub_lock);
2245 while (spa->spa_load_verify_bytes >= maxinflight_bytes)
2246 cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
2247 spa->spa_load_verify_bytes += size;
2248 mutex_exit(&spa->spa_scrub_lock);
2250 zio_nowait(zio_read(rio, spa, bp, abd_alloc_for_io(size, B_FALSE), size,
2251 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
2252 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
2253 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
2259 verify_dataset_name_len(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg)
2261 if (dsl_dataset_namelen(ds) >= ZFS_MAX_DATASET_NAME_LEN)
2262 return (SET_ERROR(ENAMETOOLONG));
2268 spa_load_verify(spa_t *spa)
2271 spa_load_error_t sle = { 0 };
2272 zpool_load_policy_t policy;
2273 boolean_t verify_ok = B_FALSE;
2276 zpool_get_load_policy(spa->spa_config, &policy);
2278 if (policy.zlp_rewind & ZPOOL_NEVER_REWIND)
2281 dsl_pool_config_enter(spa->spa_dsl_pool, FTAG);
2282 error = dmu_objset_find_dp(spa->spa_dsl_pool,
2283 spa->spa_dsl_pool->dp_root_dir_obj, verify_dataset_name_len, NULL,
2285 dsl_pool_config_exit(spa->spa_dsl_pool, FTAG);
2289 rio = zio_root(spa, NULL, &sle,
2290 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
2292 if (spa_load_verify_metadata) {
2293 if (spa->spa_extreme_rewind) {
2294 spa_load_note(spa, "performing a complete scan of the "
2295 "pool since extreme rewind is on. This may take "
2296 "a very long time.\n (spa_load_verify_data=%u, "
2297 "spa_load_verify_metadata=%u)",
2298 spa_load_verify_data, spa_load_verify_metadata);
2301 error = traverse_pool(spa, spa->spa_verify_min_txg,
2302 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA |
2303 TRAVERSE_NO_DECRYPT, spa_load_verify_cb, rio);
2306 (void) zio_wait(rio);
2307 ASSERT0(spa->spa_load_verify_bytes);
2309 spa->spa_load_meta_errors = sle.sle_meta_count;
2310 spa->spa_load_data_errors = sle.sle_data_count;
2312 if (sle.sle_meta_count != 0 || sle.sle_data_count != 0) {
2313 spa_load_note(spa, "spa_load_verify found %llu metadata errors "
2314 "and %llu data errors", (u_longlong_t)sle.sle_meta_count,
2315 (u_longlong_t)sle.sle_data_count);
2318 if (spa_load_verify_dryrun ||
2319 (!error && sle.sle_meta_count <= policy.zlp_maxmeta &&
2320 sle.sle_data_count <= policy.zlp_maxdata)) {
2324 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
2325 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
2327 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
2328 VERIFY(nvlist_add_uint64(spa->spa_load_info,
2329 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
2330 VERIFY(nvlist_add_int64(spa->spa_load_info,
2331 ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
2332 VERIFY(nvlist_add_uint64(spa->spa_load_info,
2333 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
2335 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
2338 if (spa_load_verify_dryrun)
2342 if (error != ENXIO && error != EIO)
2343 error = SET_ERROR(EIO);
2347 return (verify_ok ? 0 : EIO);
2351 * Find a value in the pool props object.
2354 spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
2356 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
2357 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
2361 * Find a value in the pool directory object.
2364 spa_dir_prop(spa_t *spa, const char *name, uint64_t *val, boolean_t log_enoent)
2366 int error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
2367 name, sizeof (uint64_t), 1, val);
2369 if (error != 0 && (error != ENOENT || log_enoent)) {
2370 spa_load_failed(spa, "couldn't get '%s' value in MOS directory "
2371 "[error=%d]", name, error);
2378 spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
2380 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
2381 return (SET_ERROR(err));
2385 spa_livelist_delete_check(spa_t *spa)
2387 return (spa->spa_livelists_to_delete != 0);
2392 spa_livelist_delete_cb_check(void *arg, zthr_t *z)
2395 return (spa_livelist_delete_check(spa));
2399 delete_blkptr_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
2402 zio_free(spa, tx->tx_txg, bp);
2403 dsl_dir_diduse_space(tx->tx_pool->dp_free_dir, DD_USED_HEAD,
2404 -bp_get_dsize_sync(spa, bp),
2405 -BP_GET_PSIZE(bp), -BP_GET_UCSIZE(bp), tx);
2410 dsl_get_next_livelist_obj(objset_t *os, uint64_t zap_obj, uint64_t *llp)
2415 zap_cursor_init(&zc, os, zap_obj);
2416 err = zap_cursor_retrieve(&zc, &za);
2417 zap_cursor_fini(&zc);
2419 *llp = za.za_first_integer;
2424 * Components of livelist deletion that must be performed in syncing
2425 * context: freeing block pointers and updating the pool-wide data
2426 * structures to indicate how much work is left to do
2428 typedef struct sublist_delete_arg {
2433 } sublist_delete_arg_t;
2436 sublist_delete_sync(void *arg, dmu_tx_t *tx)
2438 sublist_delete_arg_t *sda = arg;
2439 spa_t *spa = sda->spa;
2440 dsl_deadlist_t *ll = sda->ll;
2441 uint64_t key = sda->key;
2442 bplist_t *to_free = sda->to_free;
2444 bplist_iterate(to_free, delete_blkptr_cb, spa, tx);
2445 dsl_deadlist_remove_entry(ll, key, tx);
2448 typedef struct livelist_delete_arg {
2452 } livelist_delete_arg_t;
2455 livelist_delete_sync(void *arg, dmu_tx_t *tx)
2457 livelist_delete_arg_t *lda = arg;
2458 spa_t *spa = lda->spa;
2459 uint64_t ll_obj = lda->ll_obj;
2460 uint64_t zap_obj = lda->zap_obj;
2461 objset_t *mos = spa->spa_meta_objset;
2464 /* free the livelist and decrement the feature count */
2465 VERIFY0(zap_remove_int(mos, zap_obj, ll_obj, tx));
2466 dsl_deadlist_free(mos, ll_obj, tx);
2467 spa_feature_decr(spa, SPA_FEATURE_LIVELIST, tx);
2468 VERIFY0(zap_count(mos, zap_obj, &count));
2470 /* no more livelists to delete */
2471 VERIFY0(zap_remove(mos, DMU_POOL_DIRECTORY_OBJECT,
2472 DMU_POOL_DELETED_CLONES, tx));
2473 VERIFY0(zap_destroy(mos, zap_obj, tx));
2474 spa->spa_livelists_to_delete = 0;
2475 spa_notify_waiters(spa);
2480 * Load in the value for the livelist to be removed and open it. Then,
2481 * load its first sublist and determine which block pointers should actually
2482 * be freed. Then, call a synctask which performs the actual frees and updates
2483 * the pool-wide livelist data.
2487 spa_livelist_delete_cb(void *arg, zthr_t *z)
2490 uint64_t ll_obj = 0, count;
2491 objset_t *mos = spa->spa_meta_objset;
2492 uint64_t zap_obj = spa->spa_livelists_to_delete;
2494 * Determine the next livelist to delete. This function should only
2495 * be called if there is at least one deleted clone.
2497 VERIFY0(dsl_get_next_livelist_obj(mos, zap_obj, &ll_obj));
2498 VERIFY0(zap_count(mos, ll_obj, &count));
2501 dsl_deadlist_entry_t *dle;
2503 ll = kmem_zalloc(sizeof (dsl_deadlist_t), KM_SLEEP);
2504 dsl_deadlist_open(ll, mos, ll_obj);
2505 dle = dsl_deadlist_first(ll);
2506 ASSERT3P(dle, !=, NULL);
2507 bplist_create(&to_free);
2508 int err = dsl_process_sub_livelist(&dle->dle_bpobj, &to_free,
2511 sublist_delete_arg_t sync_arg = {
2514 .key = dle->dle_mintxg,
2517 zfs_dbgmsg("deleting sublist (id %llu) from"
2518 " livelist %llu, %d remaining",
2519 dle->dle_bpobj.bpo_object, ll_obj, count - 1);
2520 VERIFY0(dsl_sync_task(spa_name(spa), NULL,
2521 sublist_delete_sync, &sync_arg, 0,
2522 ZFS_SPACE_CHECK_DESTROY));
2524 VERIFY3U(err, ==, EINTR);
2526 bplist_clear(&to_free);
2527 bplist_destroy(&to_free);
2528 dsl_deadlist_close(ll);
2529 kmem_free(ll, sizeof (dsl_deadlist_t));
2531 livelist_delete_arg_t sync_arg = {
2536 zfs_dbgmsg("deletion of livelist %llu completed", ll_obj);
2537 VERIFY0(dsl_sync_task(spa_name(spa), NULL, livelist_delete_sync,
2538 &sync_arg, 0, ZFS_SPACE_CHECK_DESTROY));
2543 spa_start_livelist_destroy_thread(spa_t *spa)
2545 ASSERT3P(spa->spa_livelist_delete_zthr, ==, NULL);
2546 spa->spa_livelist_delete_zthr =
2547 zthr_create("z_livelist_destroy",
2548 spa_livelist_delete_cb_check, spa_livelist_delete_cb, spa);
2551 typedef struct livelist_new_arg {
2554 } livelist_new_arg_t;
2557 livelist_track_new_cb(void *arg, const blkptr_t *bp, boolean_t bp_freed,
2561 livelist_new_arg_t *lna = arg;
2563 bplist_append(lna->frees, bp);
2565 bplist_append(lna->allocs, bp);
2566 zfs_livelist_condense_new_alloc++;
2571 typedef struct livelist_condense_arg {
2574 uint64_t first_size;
2576 } livelist_condense_arg_t;
2579 spa_livelist_condense_sync(void *arg, dmu_tx_t *tx)
2581 livelist_condense_arg_t *lca = arg;
2582 spa_t *spa = lca->spa;
2584 dsl_dataset_t *ds = spa->spa_to_condense.ds;
2586 /* Have we been cancelled? */
2587 if (spa->spa_to_condense.cancelled) {
2588 zfs_livelist_condense_sync_cancel++;
2592 dsl_deadlist_entry_t *first = spa->spa_to_condense.first;
2593 dsl_deadlist_entry_t *next = spa->spa_to_condense.next;
2594 dsl_deadlist_t *ll = &ds->ds_dir->dd_livelist;
2597 * It's possible that the livelist was changed while the zthr was
2598 * running. Therefore, we need to check for new blkptrs in the two
2599 * entries being condensed and continue to track them in the livelist.
2600 * Because of the way we handle remapped blkptrs (see dbuf_remap_impl),
2601 * it's possible that the newly added blkptrs are FREEs or ALLOCs so
2602 * we need to sort them into two different bplists.
2604 uint64_t first_obj = first->dle_bpobj.bpo_object;
2605 uint64_t next_obj = next->dle_bpobj.bpo_object;
2606 uint64_t cur_first_size = first->dle_bpobj.bpo_phys->bpo_num_blkptrs;
2607 uint64_t cur_next_size = next->dle_bpobj.bpo_phys->bpo_num_blkptrs;
2609 bplist_create(&new_frees);
2610 livelist_new_arg_t new_bps = {
2611 .allocs = &lca->to_keep,
2612 .frees = &new_frees,
2615 if (cur_first_size > lca->first_size) {
2616 VERIFY0(livelist_bpobj_iterate_from_nofree(&first->dle_bpobj,
2617 livelist_track_new_cb, &new_bps, lca->first_size));
2619 if (cur_next_size > lca->next_size) {
2620 VERIFY0(livelist_bpobj_iterate_from_nofree(&next->dle_bpobj,
2621 livelist_track_new_cb, &new_bps, lca->next_size));
2624 dsl_deadlist_clear_entry(first, ll, tx);
2625 ASSERT(bpobj_is_empty(&first->dle_bpobj));
2626 dsl_deadlist_remove_entry(ll, next->dle_mintxg, tx);
2628 bplist_iterate(&lca->to_keep, dsl_deadlist_insert_alloc_cb, ll, tx);
2629 bplist_iterate(&new_frees, dsl_deadlist_insert_free_cb, ll, tx);
2630 bplist_destroy(&new_frees);
2632 char dsname[ZFS_MAX_DATASET_NAME_LEN];
2633 dsl_dataset_name(ds, dsname);
2634 zfs_dbgmsg("txg %llu condensing livelist of %s (id %llu), bpobj %llu "
2635 "(%llu blkptrs) and bpobj %llu (%llu blkptrs) -> bpobj %llu "
2636 "(%llu blkptrs)", tx->tx_txg, dsname, ds->ds_object, first_obj,
2637 cur_first_size, next_obj, cur_next_size,
2638 first->dle_bpobj.bpo_object,
2639 first->dle_bpobj.bpo_phys->bpo_num_blkptrs);
2641 dmu_buf_rele(ds->ds_dbuf, spa);
2642 spa->spa_to_condense.ds = NULL;
2643 bplist_clear(&lca->to_keep);
2644 bplist_destroy(&lca->to_keep);
2645 kmem_free(lca, sizeof (livelist_condense_arg_t));
2646 spa->spa_to_condense.syncing = B_FALSE;
2650 spa_livelist_condense_cb(void *arg, zthr_t *t)
2652 while (zfs_livelist_condense_zthr_pause &&
2653 !(zthr_has_waiters(t) || zthr_iscancelled(t)))
2657 dsl_deadlist_entry_t *first = spa->spa_to_condense.first;
2658 dsl_deadlist_entry_t *next = spa->spa_to_condense.next;
2659 uint64_t first_size, next_size;
2661 livelist_condense_arg_t *lca =
2662 kmem_alloc(sizeof (livelist_condense_arg_t), KM_SLEEP);
2663 bplist_create(&lca->to_keep);
2666 * Process the livelists (matching FREEs and ALLOCs) in open context
2667 * so we have minimal work in syncing context to condense.
2669 * We save bpobj sizes (first_size and next_size) to use later in
2670 * syncing context to determine if entries were added to these sublists
2671 * while in open context. This is possible because the clone is still
2672 * active and open for normal writes and we want to make sure the new,
2673 * unprocessed blockpointers are inserted into the livelist normally.
2675 * Note that dsl_process_sub_livelist() both stores the size number of
2676 * blockpointers and iterates over them while the bpobj's lock held, so
2677 * the sizes returned to us are consistent which what was actually
2680 int err = dsl_process_sub_livelist(&first->dle_bpobj, &lca->to_keep, t,
2683 err = dsl_process_sub_livelist(&next->dle_bpobj, &lca->to_keep,
2687 while (zfs_livelist_condense_sync_pause &&
2688 !(zthr_has_waiters(t) || zthr_iscancelled(t)))
2691 dmu_tx_t *tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
2692 dmu_tx_mark_netfree(tx);
2693 dmu_tx_hold_space(tx, 1);
2694 err = dmu_tx_assign(tx, TXG_NOWAIT | TXG_NOTHROTTLE);
2697 * Prevent the condense zthr restarting before
2698 * the synctask completes.
2700 spa->spa_to_condense.syncing = B_TRUE;
2702 lca->first_size = first_size;
2703 lca->next_size = next_size;
2704 dsl_sync_task_nowait(spa_get_dsl(spa),
2705 spa_livelist_condense_sync, lca, tx);
2711 * Condensing can not continue: either it was externally stopped or
2712 * we were unable to assign to a tx because the pool has run out of
2713 * space. In the second case, we'll just end up trying to condense
2714 * again in a later txg.
2717 bplist_clear(&lca->to_keep);
2718 bplist_destroy(&lca->to_keep);
2719 kmem_free(lca, sizeof (livelist_condense_arg_t));
2720 dmu_buf_rele(spa->spa_to_condense.ds->ds_dbuf, spa);
2721 spa->spa_to_condense.ds = NULL;
2723 zfs_livelist_condense_zthr_cancel++;
2728 * Check that there is something to condense but that a condense is not
2729 * already in progress and that condensing has not been cancelled.
2732 spa_livelist_condense_cb_check(void *arg, zthr_t *z)
2735 if ((spa->spa_to_condense.ds != NULL) &&
2736 (spa->spa_to_condense.syncing == B_FALSE) &&
2737 (spa->spa_to_condense.cancelled == B_FALSE)) {
2744 spa_start_livelist_condensing_thread(spa_t *spa)
2746 spa->spa_to_condense.ds = NULL;
2747 spa->spa_to_condense.first = NULL;
2748 spa->spa_to_condense.next = NULL;
2749 spa->spa_to_condense.syncing = B_FALSE;
2750 spa->spa_to_condense.cancelled = B_FALSE;
2752 ASSERT3P(spa->spa_livelist_condense_zthr, ==, NULL);
2753 spa->spa_livelist_condense_zthr =
2754 zthr_create("z_livelist_condense",
2755 spa_livelist_condense_cb_check,
2756 spa_livelist_condense_cb, spa);
2760 spa_spawn_aux_threads(spa_t *spa)
2762 ASSERT(spa_writeable(spa));
2764 ASSERT(MUTEX_HELD(&spa_namespace_lock));
2766 spa_start_indirect_condensing_thread(spa);
2767 spa_start_livelist_destroy_thread(spa);
2768 spa_start_livelist_condensing_thread(spa);
2770 ASSERT3P(spa->spa_checkpoint_discard_zthr, ==, NULL);
2771 spa->spa_checkpoint_discard_zthr =
2772 zthr_create("z_checkpoint_discard",
2773 spa_checkpoint_discard_thread_check,
2774 spa_checkpoint_discard_thread, spa);
2778 * Fix up config after a partly-completed split. This is done with the
2779 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
2780 * pool have that entry in their config, but only the splitting one contains
2781 * a list of all the guids of the vdevs that are being split off.
2783 * This function determines what to do with that list: either rejoin
2784 * all the disks to the pool, or complete the splitting process. To attempt
2785 * the rejoin, each disk that is offlined is marked online again, and
2786 * we do a reopen() call. If the vdev label for every disk that was
2787 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
2788 * then we call vdev_split() on each disk, and complete the split.
2790 * Otherwise we leave the config alone, with all the vdevs in place in
2791 * the original pool.
2794 spa_try_repair(spa_t *spa, nvlist_t *config)
2801 boolean_t attempt_reopen;
2803 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
2806 /* check that the config is complete */
2807 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
2808 &glist, &gcount) != 0)
2811 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
2813 /* attempt to online all the vdevs & validate */
2814 attempt_reopen = B_TRUE;
2815 for (i = 0; i < gcount; i++) {
2816 if (glist[i] == 0) /* vdev is hole */
2819 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
2820 if (vd[i] == NULL) {
2822 * Don't bother attempting to reopen the disks;
2823 * just do the split.
2825 attempt_reopen = B_FALSE;
2827 /* attempt to re-online it */
2828 vd[i]->vdev_offline = B_FALSE;
2832 if (attempt_reopen) {
2833 vdev_reopen(spa->spa_root_vdev);
2835 /* check each device to see what state it's in */
2836 for (extracted = 0, i = 0; i < gcount; i++) {
2837 if (vd[i] != NULL &&
2838 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
2845 * If every disk has been moved to the new pool, or if we never
2846 * even attempted to look at them, then we split them off for
2849 if (!attempt_reopen || gcount == extracted) {
2850 for (i = 0; i < gcount; i++)
2853 vdev_reopen(spa->spa_root_vdev);
2856 kmem_free(vd, gcount * sizeof (vdev_t *));
2860 spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type)
2862 char *ereport = FM_EREPORT_ZFS_POOL;
2865 spa->spa_load_state = state;
2866 (void) spa_import_progress_set_state(spa_guid(spa),
2867 spa_load_state(spa));
2869 gethrestime(&spa->spa_loaded_ts);
2870 error = spa_load_impl(spa, type, &ereport);
2873 * Don't count references from objsets that are already closed
2874 * and are making their way through the eviction process.
2876 spa_evicting_os_wait(spa);
2877 spa->spa_minref = zfs_refcount_count(&spa->spa_refcount);
2879 if (error != EEXIST) {
2880 spa->spa_loaded_ts.tv_sec = 0;
2881 spa->spa_loaded_ts.tv_nsec = 0;
2883 if (error != EBADF) {
2884 (void) zfs_ereport_post(ereport, spa,
2885 NULL, NULL, NULL, 0);
2888 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2891 (void) spa_import_progress_set_state(spa_guid(spa),
2892 spa_load_state(spa));
2899 * Count the number of per-vdev ZAPs associated with all of the vdevs in the
2900 * vdev tree rooted in the given vd, and ensure that each ZAP is present in the
2901 * spa's per-vdev ZAP list.
2904 vdev_count_verify_zaps(vdev_t *vd)
2906 spa_t *spa = vd->vdev_spa;
2909 if (vd->vdev_top_zap != 0) {
2911 ASSERT0(zap_lookup_int(spa->spa_meta_objset,
2912 spa->spa_all_vdev_zaps, vd->vdev_top_zap));
2914 if (vd->vdev_leaf_zap != 0) {
2916 ASSERT0(zap_lookup_int(spa->spa_meta_objset,
2917 spa->spa_all_vdev_zaps, vd->vdev_leaf_zap));
2920 for (uint64_t i = 0; i < vd->vdev_children; i++) {
2921 total += vdev_count_verify_zaps(vd->vdev_child[i]);
2929 * Determine whether the activity check is required.
2932 spa_activity_check_required(spa_t *spa, uberblock_t *ub, nvlist_t *label,
2936 uint64_t hostid = 0;
2937 uint64_t tryconfig_txg = 0;
2938 uint64_t tryconfig_timestamp = 0;
2939 uint16_t tryconfig_mmp_seq = 0;
2942 if (nvlist_exists(config, ZPOOL_CONFIG_LOAD_INFO)) {
2943 nvinfo = fnvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO);
2944 (void) nvlist_lookup_uint64(nvinfo, ZPOOL_CONFIG_MMP_TXG,
2946 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
2947 &tryconfig_timestamp);
2948 (void) nvlist_lookup_uint16(nvinfo, ZPOOL_CONFIG_MMP_SEQ,
2949 &tryconfig_mmp_seq);
2952 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE, &state);
2955 * Disable the MMP activity check - This is used by zdb which
2956 * is intended to be used on potentially active pools.
2958 if (spa->spa_import_flags & ZFS_IMPORT_SKIP_MMP)
2962 * Skip the activity check when the MMP feature is disabled.
2964 if (ub->ub_mmp_magic == MMP_MAGIC && ub->ub_mmp_delay == 0)
2968 * If the tryconfig_ values are nonzero, they are the results of an
2969 * earlier tryimport. If they all match the uberblock we just found,
2970 * then the pool has not changed and we return false so we do not test
2973 if (tryconfig_txg && tryconfig_txg == ub->ub_txg &&
2974 tryconfig_timestamp && tryconfig_timestamp == ub->ub_timestamp &&
2975 tryconfig_mmp_seq && tryconfig_mmp_seq ==
2976 (MMP_SEQ_VALID(ub) ? MMP_SEQ(ub) : 0))
2980 * Allow the activity check to be skipped when importing the pool
2981 * on the same host which last imported it. Since the hostid from
2982 * configuration may be stale use the one read from the label.
2984 if (nvlist_exists(label, ZPOOL_CONFIG_HOSTID))
2985 hostid = fnvlist_lookup_uint64(label, ZPOOL_CONFIG_HOSTID);
2987 if (hostid == spa_get_hostid(spa))
2991 * Skip the activity test when the pool was cleanly exported.
2993 if (state != POOL_STATE_ACTIVE)
3000 * Nanoseconds the activity check must watch for changes on-disk.
3003 spa_activity_check_duration(spa_t *spa, uberblock_t *ub)
3005 uint64_t import_intervals = MAX(zfs_multihost_import_intervals, 1);
3006 uint64_t multihost_interval = MSEC2NSEC(
3007 MMP_INTERVAL_OK(zfs_multihost_interval));
3008 uint64_t import_delay = MAX(NANOSEC, import_intervals *
3009 multihost_interval);
3012 * Local tunables determine a minimum duration except for the case
3013 * where we know when the remote host will suspend the pool if MMP
3014 * writes do not land.
3016 * See Big Theory comment at the top of mmp.c for the reasoning behind
3017 * these cases and times.
3020 ASSERT(MMP_IMPORT_SAFETY_FACTOR >= 100);
3022 if (MMP_INTERVAL_VALID(ub) && MMP_FAIL_INT_VALID(ub) &&
3023 MMP_FAIL_INT(ub) > 0) {
3025 /* MMP on remote host will suspend pool after failed writes */
3026 import_delay = MMP_FAIL_INT(ub) * MSEC2NSEC(MMP_INTERVAL(ub)) *
3027 MMP_IMPORT_SAFETY_FACTOR / 100;
3029 zfs_dbgmsg("fail_intvals>0 import_delay=%llu ub_mmp "
3030 "mmp_fails=%llu ub_mmp mmp_interval=%llu "
3031 "import_intervals=%u", import_delay, MMP_FAIL_INT(ub),
3032 MMP_INTERVAL(ub), import_intervals);
3034 } else if (MMP_INTERVAL_VALID(ub) && MMP_FAIL_INT_VALID(ub) &&
3035 MMP_FAIL_INT(ub) == 0) {
3037 /* MMP on remote host will never suspend pool */
3038 import_delay = MAX(import_delay, (MSEC2NSEC(MMP_INTERVAL(ub)) +
3039 ub->ub_mmp_delay) * import_intervals);
3041 zfs_dbgmsg("fail_intvals=0 import_delay=%llu ub_mmp "
3042 "mmp_interval=%llu ub_mmp_delay=%llu "
3043 "import_intervals=%u", import_delay, MMP_INTERVAL(ub),
3044 ub->ub_mmp_delay, import_intervals);
3046 } else if (MMP_VALID(ub)) {
3048 * zfs-0.7 compatibility case
3051 import_delay = MAX(import_delay, (multihost_interval +
3052 ub->ub_mmp_delay) * import_intervals);
3054 zfs_dbgmsg("import_delay=%llu ub_mmp_delay=%llu "
3055 "import_intervals=%u leaves=%u", import_delay,
3056 ub->ub_mmp_delay, import_intervals,
3057 vdev_count_leaves(spa));
3059 /* Using local tunings is the only reasonable option */
3060 zfs_dbgmsg("pool last imported on non-MMP aware "
3061 "host using import_delay=%llu multihost_interval=%llu "
3062 "import_intervals=%u", import_delay, multihost_interval,
3066 return (import_delay);
3070 * Perform the import activity check. If the user canceled the import or
3071 * we detected activity then fail.
3074 spa_activity_check(spa_t *spa, uberblock_t *ub, nvlist_t *config)
3076 uint64_t txg = ub->ub_txg;
3077 uint64_t timestamp = ub->ub_timestamp;
3078 uint64_t mmp_config = ub->ub_mmp_config;
3079 uint16_t mmp_seq = MMP_SEQ_VALID(ub) ? MMP_SEQ(ub) : 0;
3080 uint64_t import_delay;
3081 hrtime_t import_expire;
3082 nvlist_t *mmp_label = NULL;
3083 vdev_t *rvd = spa->spa_root_vdev;
3088 cv_init(&cv, NULL, CV_DEFAULT, NULL);
3089 mutex_init(&mtx, NULL, MUTEX_DEFAULT, NULL);
3093 * If ZPOOL_CONFIG_MMP_TXG is present an activity check was performed
3094 * during the earlier tryimport. If the txg recorded there is 0 then
3095 * the pool is known to be active on another host.
3097 * Otherwise, the pool might be in use on another host. Check for
3098 * changes in the uberblocks on disk if necessary.
3100 if (nvlist_exists(config, ZPOOL_CONFIG_LOAD_INFO)) {
3101 nvlist_t *nvinfo = fnvlist_lookup_nvlist(config,
3102 ZPOOL_CONFIG_LOAD_INFO);
3104 if (nvlist_exists(nvinfo, ZPOOL_CONFIG_MMP_TXG) &&
3105 fnvlist_lookup_uint64(nvinfo, ZPOOL_CONFIG_MMP_TXG) == 0) {
3106 vdev_uberblock_load(rvd, ub, &mmp_label);
3107 error = SET_ERROR(EREMOTEIO);
3112 import_delay = spa_activity_check_duration(spa, ub);
3114 /* Add a small random factor in case of simultaneous imports (0-25%) */
3115 import_delay += import_delay * spa_get_random(250) / 1000;
3117 import_expire = gethrtime() + import_delay;
3119 while (gethrtime() < import_expire) {
3120 (void) spa_import_progress_set_mmp_check(spa_guid(spa),
3121 NSEC2SEC(import_expire - gethrtime()));
3123 vdev_uberblock_load(rvd, ub, &mmp_label);
3125 if (txg != ub->ub_txg || timestamp != ub->ub_timestamp ||
3126 mmp_seq != (MMP_SEQ_VALID(ub) ? MMP_SEQ(ub) : 0)) {
3127 zfs_dbgmsg("multihost activity detected "
3128 "txg %llu ub_txg %llu "
3129 "timestamp %llu ub_timestamp %llu "
3130 "mmp_config %#llx ub_mmp_config %#llx",
3131 txg, ub->ub_txg, timestamp, ub->ub_timestamp,
3132 mmp_config, ub->ub_mmp_config);
3134 error = SET_ERROR(EREMOTEIO);
3139 nvlist_free(mmp_label);
3143 error = cv_timedwait_sig(&cv, &mtx, ddi_get_lbolt() + hz);
3145 error = SET_ERROR(EINTR);
3153 mutex_destroy(&mtx);
3157 * If the pool is determined to be active store the status in the
3158 * spa->spa_load_info nvlist. If the remote hostname or hostid are
3159 * available from configuration read from disk store them as well.
3160 * This allows 'zpool import' to generate a more useful message.
3162 * ZPOOL_CONFIG_MMP_STATE - observed pool status (mandatory)
3163 * ZPOOL_CONFIG_MMP_HOSTNAME - hostname from the active pool
3164 * ZPOOL_CONFIG_MMP_HOSTID - hostid from the active pool
3166 if (error == EREMOTEIO) {
3167 char *hostname = "<unknown>";
3168 uint64_t hostid = 0;
3171 if (nvlist_exists(mmp_label, ZPOOL_CONFIG_HOSTNAME)) {
3172 hostname = fnvlist_lookup_string(mmp_label,
3173 ZPOOL_CONFIG_HOSTNAME);
3174 fnvlist_add_string(spa->spa_load_info,
3175 ZPOOL_CONFIG_MMP_HOSTNAME, hostname);
3178 if (nvlist_exists(mmp_label, ZPOOL_CONFIG_HOSTID)) {
3179 hostid = fnvlist_lookup_uint64(mmp_label,
3180 ZPOOL_CONFIG_HOSTID);
3181 fnvlist_add_uint64(spa->spa_load_info,
3182 ZPOOL_CONFIG_MMP_HOSTID, hostid);
3186 fnvlist_add_uint64(spa->spa_load_info,
3187 ZPOOL_CONFIG_MMP_STATE, MMP_STATE_ACTIVE);
3188 fnvlist_add_uint64(spa->spa_load_info,
3189 ZPOOL_CONFIG_MMP_TXG, 0);
3191 error = spa_vdev_err(rvd, VDEV_AUX_ACTIVE, EREMOTEIO);
3195 nvlist_free(mmp_label);
3201 spa_verify_host(spa_t *spa, nvlist_t *mos_config)
3205 uint64_t myhostid = 0;
3207 if (!spa_is_root(spa) && nvlist_lookup_uint64(mos_config,
3208 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
3209 hostname = fnvlist_lookup_string(mos_config,
3210 ZPOOL_CONFIG_HOSTNAME);
3212 myhostid = zone_get_hostid(NULL);
3214 if (hostid != 0 && myhostid != 0 && hostid != myhostid) {
3215 cmn_err(CE_WARN, "pool '%s' could not be "
3216 "loaded as it was last accessed by "
3217 "another system (host: %s hostid: 0x%llx). "
3218 "See: https://openzfs.github.io/openzfs-docs/msg/"
3220 spa_name(spa), hostname, (u_longlong_t)hostid);
3221 spa_load_failed(spa, "hostid verification failed: pool "
3222 "last accessed by host: %s (hostid: 0x%llx)",
3223 hostname, (u_longlong_t)hostid);
3224 return (SET_ERROR(EBADF));
3232 spa_ld_parse_config(spa_t *spa, spa_import_type_t type)
3235 nvlist_t *nvtree, *nvl, *config = spa->spa_config;
3242 * Versioning wasn't explicitly added to the label until later, so if
3243 * it's not present treat it as the initial version.
3245 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
3246 &spa->spa_ubsync.ub_version) != 0)
3247 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
3249 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) {
3250 spa_load_failed(spa, "invalid config provided: '%s' missing",
3251 ZPOOL_CONFIG_POOL_GUID);
3252 return (SET_ERROR(EINVAL));
3256 * If we are doing an import, ensure that the pool is not already
3257 * imported by checking if its pool guid already exists in the
3260 * The only case that we allow an already imported pool to be
3261 * imported again, is when the pool is checkpointed and we want to
3262 * look at its checkpointed state from userland tools like zdb.
3265 if ((spa->spa_load_state == SPA_LOAD_IMPORT ||
3266 spa->spa_load_state == SPA_LOAD_TRYIMPORT) &&
3267 spa_guid_exists(pool_guid, 0)) {
3269 if ((spa->spa_load_state == SPA_LOAD_IMPORT ||
3270 spa->spa_load_state == SPA_LOAD_TRYIMPORT) &&
3271 spa_guid_exists(pool_guid, 0) &&
3272 !spa_importing_readonly_checkpoint(spa)) {
3274 spa_load_failed(spa, "a pool with guid %llu is already open",
3275 (u_longlong_t)pool_guid);
3276 return (SET_ERROR(EEXIST));
3279 spa->spa_config_guid = pool_guid;
3281 nvlist_free(spa->spa_load_info);
3282 spa->spa_load_info = fnvlist_alloc();
3284 ASSERT(spa->spa_comment == NULL);
3285 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
3286 spa->spa_comment = spa_strdup(comment);
3288 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
3289 &spa->spa_config_txg);
3291 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) == 0)
3292 spa->spa_config_splitting = fnvlist_dup(nvl);
3294 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvtree)) {
3295 spa_load_failed(spa, "invalid config provided: '%s' missing",
3296 ZPOOL_CONFIG_VDEV_TREE);
3297 return (SET_ERROR(EINVAL));
3301 * Create "The Godfather" zio to hold all async IOs
3303 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
3305 for (int i = 0; i < max_ncpus; i++) {
3306 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
3307 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
3308 ZIO_FLAG_GODFATHER);
3312 * Parse the configuration into a vdev tree. We explicitly set the
3313 * value that will be returned by spa_version() since parsing the
3314 * configuration requires knowing the version number.
3316 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3317 parse = (type == SPA_IMPORT_EXISTING ?
3318 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
3319 error = spa_config_parse(spa, &rvd, nvtree, NULL, 0, parse);
3320 spa_config_exit(spa, SCL_ALL, FTAG);
3323 spa_load_failed(spa, "unable to parse config [error=%d]",
3328 ASSERT(spa->spa_root_vdev == rvd);
3329 ASSERT3U(spa->spa_min_ashift, >=, SPA_MINBLOCKSHIFT);
3330 ASSERT3U(spa->spa_max_ashift, <=, SPA_MAXBLOCKSHIFT);
3332 if (type != SPA_IMPORT_ASSEMBLE) {
3333 ASSERT(spa_guid(spa) == pool_guid);
3340 * Recursively open all vdevs in the vdev tree. This function is called twice:
3341 * first with the untrusted config, then with the trusted config.
3344 spa_ld_open_vdevs(spa_t *spa)
3349 * spa_missing_tvds_allowed defines how many top-level vdevs can be
3350 * missing/unopenable for the root vdev to be still considered openable.
3352 if (spa->spa_trust_config) {
3353 spa->spa_missing_tvds_allowed = zfs_max_missing_tvds;
3354 } else if (spa->spa_config_source == SPA_CONFIG_SRC_CACHEFILE) {
3355 spa->spa_missing_tvds_allowed = zfs_max_missing_tvds_cachefile;
3356 } else if (spa->spa_config_source == SPA_CONFIG_SRC_SCAN) {
3357 spa->spa_missing_tvds_allowed = zfs_max_missing_tvds_scan;
3359 spa->spa_missing_tvds_allowed = 0;
3362 spa->spa_missing_tvds_allowed =
3363 MAX(zfs_max_missing_tvds, spa->spa_missing_tvds_allowed);
3365 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3366 error = vdev_open(spa->spa_root_vdev);
3367 spa_config_exit(spa, SCL_ALL, FTAG);
3369 if (spa->spa_missing_tvds != 0) {
3370 spa_load_note(spa, "vdev tree has %lld missing top-level "
3371 "vdevs.", (u_longlong_t)spa->spa_missing_tvds);
3372 if (spa->spa_trust_config && (spa->spa_mode & SPA_MODE_WRITE)) {
3374 * Although theoretically we could allow users to open
3375 * incomplete pools in RW mode, we'd need to add a lot
3376 * of extra logic (e.g. adjust pool space to account
3377 * for missing vdevs).
3378 * This limitation also prevents users from accidentally
3379 * opening the pool in RW mode during data recovery and
3380 * damaging it further.
3382 spa_load_note(spa, "pools with missing top-level "
3383 "vdevs can only be opened in read-only mode.");
3384 error = SET_ERROR(ENXIO);
3386 spa_load_note(spa, "current settings allow for maximum "
3387 "%lld missing top-level vdevs at this stage.",
3388 (u_longlong_t)spa->spa_missing_tvds_allowed);
3392 spa_load_failed(spa, "unable to open vdev tree [error=%d]",
3395 if (spa->spa_missing_tvds != 0 || error != 0)
3396 vdev_dbgmsg_print_tree(spa->spa_root_vdev, 2);
3402 * We need to validate the vdev labels against the configuration that
3403 * we have in hand. This function is called twice: first with an untrusted
3404 * config, then with a trusted config. The validation is more strict when the
3405 * config is trusted.
3408 spa_ld_validate_vdevs(spa_t *spa)
3411 vdev_t *rvd = spa->spa_root_vdev;
3413 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3414 error = vdev_validate(rvd);
3415 spa_config_exit(spa, SCL_ALL, FTAG);
3418 spa_load_failed(spa, "vdev_validate failed [error=%d]", error);
3422 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) {
3423 spa_load_failed(spa, "cannot open vdev tree after invalidating "
3425 vdev_dbgmsg_print_tree(rvd, 2);
3426 return (SET_ERROR(ENXIO));
3433 spa_ld_select_uberblock_done(spa_t *spa, uberblock_t *ub)
3435 spa->spa_state = POOL_STATE_ACTIVE;
3436 spa->spa_ubsync = spa->spa_uberblock;
3437 spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
3438 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
3439 spa->spa_first_txg = spa->spa_last_ubsync_txg ?
3440 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
3441 spa->spa_claim_max_txg = spa->spa_first_txg;
3442 spa->spa_prev_software_version = ub->ub_software_version;
3446 spa_ld_select_uberblock(spa_t *spa, spa_import_type_t type)
3448 vdev_t *rvd = spa->spa_root_vdev;
3450 uberblock_t *ub = &spa->spa_uberblock;
3451 boolean_t activity_check = B_FALSE;
3454 * If we are opening the checkpointed state of the pool by
3455 * rewinding to it, at this point we will have written the
3456 * checkpointed uberblock to the vdev labels, so searching
3457 * the labels will find the right uberblock. However, if
3458 * we are opening the checkpointed state read-only, we have
3459 * not modified the labels. Therefore, we must ignore the
3460 * labels and continue using the spa_uberblock that was set
3461 * by spa_ld_checkpoint_rewind.
3463 * Note that it would be fine to ignore the labels when
3464 * rewinding (opening writeable) as well. However, if we
3465 * crash just after writing the labels, we will end up
3466 * searching the labels. Doing so in the common case means
3467 * that this code path gets exercised normally, rather than
3468 * just in the edge case.
3470 if (ub->ub_checkpoint_txg != 0 &&
3471 spa_importing_readonly_checkpoint(spa)) {
3472 spa_ld_select_uberblock_done(spa, ub);
3477 * Find the best uberblock.
3479 vdev_uberblock_load(rvd, ub, &label);
3482 * If we weren't able to find a single valid uberblock, return failure.
3484 if (ub->ub_txg == 0) {
3486 spa_load_failed(spa, "no valid uberblock found");
3487 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
3490 if (spa->spa_load_max_txg != UINT64_MAX) {
3491 (void) spa_import_progress_set_max_txg(spa_guid(spa),
3492 (u_longlong_t)spa->spa_load_max_txg);
3494 spa_load_note(spa, "using uberblock with txg=%llu",
3495 (u_longlong_t)ub->ub_txg);
3499 * For pools which have the multihost property on determine if the
3500 * pool is truly inactive and can be safely imported. Prevent
3501 * hosts which don't have a hostid set from importing the pool.
3503 activity_check = spa_activity_check_required(spa, ub, label,
3505 if (activity_check) {
3506 if (ub->ub_mmp_magic == MMP_MAGIC && ub->ub_mmp_delay &&
3507 spa_get_hostid(spa) == 0) {
3509 fnvlist_add_uint64(spa->spa_load_info,
3510 ZPOOL_CONFIG_MMP_STATE, MMP_STATE_NO_HOSTID);
3511 return (spa_vdev_err(rvd, VDEV_AUX_ACTIVE, EREMOTEIO));
3514 int error = spa_activity_check(spa, ub, spa->spa_config);
3520 fnvlist_add_uint64(spa->spa_load_info,
3521 ZPOOL_CONFIG_MMP_STATE, MMP_STATE_INACTIVE);
3522 fnvlist_add_uint64(spa->spa_load_info,
3523 ZPOOL_CONFIG_MMP_TXG, ub->ub_txg);
3524 fnvlist_add_uint16(spa->spa_load_info,
3525 ZPOOL_CONFIG_MMP_SEQ,
3526 (MMP_SEQ_VALID(ub) ? MMP_SEQ(ub) : 0));
3530 * If the pool has an unsupported version we can't open it.
3532 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
3534 spa_load_failed(spa, "version %llu is not supported",
3535 (u_longlong_t)ub->ub_version);
3536 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
3539 if (ub->ub_version >= SPA_VERSION_FEATURES) {
3543 * If we weren't able to find what's necessary for reading the
3544 * MOS in the label, return failure.
3546 if (label == NULL) {
3547 spa_load_failed(spa, "label config unavailable");
3548 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
3552 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_FEATURES_FOR_READ,
3555 spa_load_failed(spa, "invalid label: '%s' missing",
3556 ZPOOL_CONFIG_FEATURES_FOR_READ);
3557 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
3562 * Update our in-core representation with the definitive values
3565 nvlist_free(spa->spa_label_features);
3566 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
3572 * Look through entries in the label nvlist's features_for_read. If
3573 * there is a feature listed there which we don't understand then we
3574 * cannot open a pool.
3576 if (ub->ub_version >= SPA_VERSION_FEATURES) {
3577 nvlist_t *unsup_feat;
3579 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
3582 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
3584 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
3585 if (!zfeature_is_supported(nvpair_name(nvp))) {
3586 VERIFY(nvlist_add_string(unsup_feat,
3587 nvpair_name(nvp), "") == 0);
3591 if (!nvlist_empty(unsup_feat)) {
3592 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
3593 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
3594 nvlist_free(unsup_feat);
3595 spa_load_failed(spa, "some features are unsupported");
3596 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
3600 nvlist_free(unsup_feat);
3603 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
3604 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3605 spa_try_repair(spa, spa->spa_config);
3606 spa_config_exit(spa, SCL_ALL, FTAG);
3607 nvlist_free(spa->spa_config_splitting);
3608 spa->spa_config_splitting = NULL;
3612 * Initialize internal SPA structures.
3614 spa_ld_select_uberblock_done(spa, ub);
3620 spa_ld_open_rootbp(spa_t *spa)
3623 vdev_t *rvd = spa->spa_root_vdev;
3625 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
3627 spa_load_failed(spa, "unable to open rootbp in dsl_pool_init "
3628 "[error=%d]", error);
3629 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3631 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
3637 spa_ld_trusted_config(spa_t *spa, spa_import_type_t type,
3638 boolean_t reloading)
3640 vdev_t *mrvd, *rvd = spa->spa_root_vdev;
3641 nvlist_t *nv, *mos_config, *policy;
3642 int error = 0, copy_error;
3643 uint64_t healthy_tvds, healthy_tvds_mos;
3644 uint64_t mos_config_txg;
3646 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object, B_TRUE)
3648 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3651 * If we're assembling a pool from a split, the config provided is
3652 * already trusted so there is nothing to do.
3654 if (type == SPA_IMPORT_ASSEMBLE)
3657 healthy_tvds = spa_healthy_core_tvds(spa);
3659 if (load_nvlist(spa, spa->spa_config_object, &mos_config)
3661 spa_load_failed(spa, "unable to retrieve MOS config");
3662 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3666 * If we are doing an open, pool owner wasn't verified yet, thus do
3667 * the verification here.
3669 if (spa->spa_load_state == SPA_LOAD_OPEN) {
3670 error = spa_verify_host(spa, mos_config);
3672 nvlist_free(mos_config);
3677 nv = fnvlist_lookup_nvlist(mos_config, ZPOOL_CONFIG_VDEV_TREE);
3679 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3682 * Build a new vdev tree from the trusted config
3684 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
3687 * Vdev paths in the MOS may be obsolete. If the untrusted config was
3688 * obtained by scanning /dev/dsk, then it will have the right vdev
3689 * paths. We update the trusted MOS config with this information.
3690 * We first try to copy the paths with vdev_copy_path_strict, which
3691 * succeeds only when both configs have exactly the same vdev tree.
3692 * If that fails, we fall back to a more flexible method that has a
3693 * best effort policy.
3695 copy_error = vdev_copy_path_strict(rvd, mrvd);
3696 if (copy_error != 0 || spa_load_print_vdev_tree) {
3697 spa_load_note(spa, "provided vdev tree:");
3698 vdev_dbgmsg_print_tree(rvd, 2);
3699 spa_load_note(spa, "MOS vdev tree:");
3700 vdev_dbgmsg_print_tree(mrvd, 2);
3702 if (copy_error != 0) {
3703 spa_load_note(spa, "vdev_copy_path_strict failed, falling "
3704 "back to vdev_copy_path_relaxed");
3705 vdev_copy_path_relaxed(rvd, mrvd);
3710 spa->spa_root_vdev = mrvd;
3712 spa_config_exit(spa, SCL_ALL, FTAG);
3715 * We will use spa_config if we decide to reload the spa or if spa_load
3716 * fails and we rewind. We must thus regenerate the config using the
3717 * MOS information with the updated paths. ZPOOL_LOAD_POLICY is used to
3718 * pass settings on how to load the pool and is not stored in the MOS.
3719 * We copy it over to our new, trusted config.
3721 mos_config_txg = fnvlist_lookup_uint64(mos_config,
3722 ZPOOL_CONFIG_POOL_TXG);
3723 nvlist_free(mos_config);
3724 mos_config = spa_config_generate(spa, NULL, mos_config_txg, B_FALSE);
3725 if (nvlist_lookup_nvlist(spa->spa_config, ZPOOL_LOAD_POLICY,
3727 fnvlist_add_nvlist(mos_config, ZPOOL_LOAD_POLICY, policy);
3728 spa_config_set(spa, mos_config);
3729 spa->spa_config_source = SPA_CONFIG_SRC_MOS;
3732 * Now that we got the config from the MOS, we should be more strict
3733 * in checking blkptrs and can make assumptions about the consistency
3734 * of the vdev tree. spa_trust_config must be set to true before opening
3735 * vdevs in order for them to be writeable.
3737 spa->spa_trust_config = B_TRUE;
3740 * Open and validate the new vdev tree
3742 error = spa_ld_open_vdevs(spa);
3746 error = spa_ld_validate_vdevs(spa);
3750 if (copy_error != 0 || spa_load_print_vdev_tree) {
3751 spa_load_note(spa, "final vdev tree:");
3752 vdev_dbgmsg_print_tree(rvd, 2);
3755 if (spa->spa_load_state != SPA_LOAD_TRYIMPORT &&
3756 !spa->spa_extreme_rewind && zfs_max_missing_tvds == 0) {
3758 * Sanity check to make sure that we are indeed loading the
3759 * latest uberblock. If we missed SPA_SYNC_MIN_VDEVS tvds
3760 * in the config provided and they happened to be the only ones
3761 * to have the latest uberblock, we could involuntarily perform
3762 * an extreme rewind.
3764 healthy_tvds_mos = spa_healthy_core_tvds(spa);
3765 if (healthy_tvds_mos - healthy_tvds >=
3766 SPA_SYNC_MIN_VDEVS) {
3767 spa_load_note(spa, "config provided misses too many "
3768 "top-level vdevs compared to MOS (%lld vs %lld). ",
3769 (u_longlong_t)healthy_tvds,
3770 (u_longlong_t)healthy_tvds_mos);
3771 spa_load_note(spa, "vdev tree:");
3772 vdev_dbgmsg_print_tree(rvd, 2);
3774 spa_load_failed(spa, "config was already "
3775 "provided from MOS. Aborting.");
3776 return (spa_vdev_err(rvd,
3777 VDEV_AUX_CORRUPT_DATA, EIO));
3779 spa_load_note(spa, "spa must be reloaded using MOS "
3781 return (SET_ERROR(EAGAIN));
3785 error = spa_check_for_missing_logs(spa);
3787 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
3789 if (rvd->vdev_guid_sum != spa->spa_uberblock.ub_guid_sum) {
3790 spa_load_failed(spa, "uberblock guid sum doesn't match MOS "
3791 "guid sum (%llu != %llu)",
3792 (u_longlong_t)spa->spa_uberblock.ub_guid_sum,
3793 (u_longlong_t)rvd->vdev_guid_sum);
3794 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
3802 spa_ld_open_indirect_vdev_metadata(spa_t *spa)
3805 vdev_t *rvd = spa->spa_root_vdev;
3808 * Everything that we read before spa_remove_init() must be stored
3809 * on concreted vdevs. Therefore we do this as early as possible.
3811 error = spa_remove_init(spa);
3813 spa_load_failed(spa, "spa_remove_init failed [error=%d]",
3815 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3819 * Retrieve information needed to condense indirect vdev mappings.
3821 error = spa_condense_init(spa);
3823 spa_load_failed(spa, "spa_condense_init failed [error=%d]",
3825 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, error));
3832 spa_ld_check_features(spa_t *spa, boolean_t *missing_feat_writep)
3835 vdev_t *rvd = spa->spa_root_vdev;
3837 if (spa_version(spa) >= SPA_VERSION_FEATURES) {
3838 boolean_t missing_feat_read = B_FALSE;
3839 nvlist_t *unsup_feat, *enabled_feat;
3841 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
3842 &spa->spa_feat_for_read_obj, B_TRUE) != 0) {
3843 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3846 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
3847 &spa->spa_feat_for_write_obj, B_TRUE) != 0) {
3848 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3851 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
3852 &spa->spa_feat_desc_obj, B_TRUE) != 0) {
3853 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3856 enabled_feat = fnvlist_alloc();
3857 unsup_feat = fnvlist_alloc();
3859 if (!spa_features_check(spa, B_FALSE,
3860 unsup_feat, enabled_feat))
3861 missing_feat_read = B_TRUE;
3863 if (spa_writeable(spa) ||
3864 spa->spa_load_state == SPA_LOAD_TRYIMPORT) {
3865 if (!spa_features_check(spa, B_TRUE,
3866 unsup_feat, enabled_feat)) {
3867 *missing_feat_writep = B_TRUE;
3871 fnvlist_add_nvlist(spa->spa_load_info,
3872 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
3874 if (!nvlist_empty(unsup_feat)) {
3875 fnvlist_add_nvlist(spa->spa_load_info,
3876 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
3879 fnvlist_free(enabled_feat);
3880 fnvlist_free(unsup_feat);
3882 if (!missing_feat_read) {
3883 fnvlist_add_boolean(spa->spa_load_info,
3884 ZPOOL_CONFIG_CAN_RDONLY);
3888 * If the state is SPA_LOAD_TRYIMPORT, our objective is
3889 * twofold: to determine whether the pool is available for
3890 * import in read-write mode and (if it is not) whether the
3891 * pool is available for import in read-only mode. If the pool
3892 * is available for import in read-write mode, it is displayed
3893 * as available in userland; if it is not available for import
3894 * in read-only mode, it is displayed as unavailable in
3895 * userland. If the pool is available for import in read-only
3896 * mode but not read-write mode, it is displayed as unavailable
3897 * in userland with a special note that the pool is actually
3898 * available for open in read-only mode.
3900 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
3901 * missing a feature for write, we must first determine whether
3902 * the pool can be opened read-only before returning to
3903 * userland in order to know whether to display the
3904 * abovementioned note.
3906 if (missing_feat_read || (*missing_feat_writep &&
3907 spa_writeable(spa))) {
3908 spa_load_failed(spa, "pool uses unsupported features");
3909 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
3914 * Load refcounts for ZFS features from disk into an in-memory
3915 * cache during SPA initialization.
3917 for (spa_feature_t i = 0; i < SPA_FEATURES; i++) {
3920 error = feature_get_refcount_from_disk(spa,
3921 &spa_feature_table[i], &refcount);
3923 spa->spa_feat_refcount_cache[i] = refcount;
3924 } else if (error == ENOTSUP) {
3925 spa->spa_feat_refcount_cache[i] =
3926 SPA_FEATURE_DISABLED;
3928 spa_load_failed(spa, "error getting refcount "
3929 "for feature %s [error=%d]",
3930 spa_feature_table[i].fi_guid, error);
3931 return (spa_vdev_err(rvd,
3932 VDEV_AUX_CORRUPT_DATA, EIO));
3937 if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) {
3938 if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG,
3939 &spa->spa_feat_enabled_txg_obj, B_TRUE) != 0)
3940 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3944 * Encryption was added before bookmark_v2, even though bookmark_v2
3945 * is now a dependency. If this pool has encryption enabled without
3946 * bookmark_v2, trigger an errata message.
3948 if (spa_feature_is_enabled(spa, SPA_FEATURE_ENCRYPTION) &&
3949 !spa_feature_is_enabled(spa, SPA_FEATURE_BOOKMARK_V2)) {
3950 spa->spa_errata = ZPOOL_ERRATA_ZOL_8308_ENCRYPTION;
3957 spa_ld_load_special_directories(spa_t *spa)
3960 vdev_t *rvd = spa->spa_root_vdev;
3962 spa->spa_is_initializing = B_TRUE;
3963 error = dsl_pool_open(spa->spa_dsl_pool);
3964 spa->spa_is_initializing = B_FALSE;
3966 spa_load_failed(spa, "dsl_pool_open failed [error=%d]", error);
3967 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3974 spa_ld_get_props(spa_t *spa)
3978 vdev_t *rvd = spa->spa_root_vdev;
3980 /* Grab the checksum salt from the MOS. */
3981 error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
3982 DMU_POOL_CHECKSUM_SALT, 1,
3983 sizeof (spa->spa_cksum_salt.zcs_bytes),
3984 spa->spa_cksum_salt.zcs_bytes);
3985 if (error == ENOENT) {
3986 /* Generate a new salt for subsequent use */
3987 (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes,
3988 sizeof (spa->spa_cksum_salt.zcs_bytes));
3989 } else if (error != 0) {
3990 spa_load_failed(spa, "unable to retrieve checksum salt from "
3991 "MOS [error=%d]", error);
3992 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3995 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj, B_TRUE) != 0)
3996 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3997 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
3999 spa_load_failed(spa, "error opening deferred-frees bpobj "
4000 "[error=%d]", error);
4001 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4005 * Load the bit that tells us to use the new accounting function
4006 * (raid-z deflation). If we have an older pool, this will not
4009 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate, B_FALSE);
4010 if (error != 0 && error != ENOENT)
4011 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4013 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
4014 &spa->spa_creation_version, B_FALSE);
4015 if (error != 0 && error != ENOENT)
4016 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4019 * Load the persistent error log. If we have an older pool, this will
4022 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last,
4024 if (error != 0 && error != ENOENT)
4025 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4027 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
4028 &spa->spa_errlog_scrub, B_FALSE);
4029 if (error != 0 && error != ENOENT)
4030 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4033 * Load the livelist deletion field. If a livelist is queued for
4034 * deletion, indicate that in the spa
4036 error = spa_dir_prop(spa, DMU_POOL_DELETED_CLONES,
4037 &spa->spa_livelists_to_delete, B_FALSE);
4038 if (error != 0 && error != ENOENT)
4039 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4042 * Load the history object. If we have an older pool, this
4043 * will not be present.
4045 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history, B_FALSE);
4046 if (error != 0 && error != ENOENT)
4047 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4050 * Load the per-vdev ZAP map. If we have an older pool, this will not
4051 * be present; in this case, defer its creation to a later time to
4052 * avoid dirtying the MOS this early / out of sync context. See
4053 * spa_sync_config_object.
4056 /* The sentinel is only available in the MOS config. */
4057 nvlist_t *mos_config;
4058 if (load_nvlist(spa, spa->spa_config_object, &mos_config) != 0) {
4059 spa_load_failed(spa, "unable to retrieve MOS config");
4060 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4063 error = spa_dir_prop(spa, DMU_POOL_VDEV_ZAP_MAP,
4064 &spa->spa_all_vdev_zaps, B_FALSE);
4066 if (error == ENOENT) {
4067 VERIFY(!nvlist_exists(mos_config,
4068 ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS));
4069 spa->spa_avz_action = AVZ_ACTION_INITIALIZE;
4070 ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev));
4071 } else if (error != 0) {
4072 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4073 } else if (!nvlist_exists(mos_config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)) {
4075 * An older version of ZFS overwrote the sentinel value, so
4076 * we have orphaned per-vdev ZAPs in the MOS. Defer their
4077 * destruction to later; see spa_sync_config_object.
4079 spa->spa_avz_action = AVZ_ACTION_DESTROY;
4081 * We're assuming that no vdevs have had their ZAPs created
4082 * before this. Better be sure of it.
4084 ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev));
4086 nvlist_free(mos_config);
4088 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
4090 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object,
4092 if (error && error != ENOENT)
4093 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4096 uint64_t autoreplace;
4098 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
4099 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
4100 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
4101 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
4102 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
4103 spa_prop_find(spa, ZPOOL_PROP_MULTIHOST, &spa->spa_multihost);
4104 spa_prop_find(spa, ZPOOL_PROP_AUTOTRIM, &spa->spa_autotrim);
4105 spa->spa_autoreplace = (autoreplace != 0);
4109 * If we are importing a pool with missing top-level vdevs,
4110 * we enforce that the pool doesn't panic or get suspended on
4111 * error since the likelihood of missing data is extremely high.
4113 if (spa->spa_missing_tvds > 0 &&
4114 spa->spa_failmode != ZIO_FAILURE_MODE_CONTINUE &&
4115 spa->spa_load_state != SPA_LOAD_TRYIMPORT) {
4116 spa_load_note(spa, "forcing failmode to 'continue' "
4117 "as some top level vdevs are missing");
4118 spa->spa_failmode = ZIO_FAILURE_MODE_CONTINUE;
4125 spa_ld_open_aux_vdevs(spa_t *spa, spa_import_type_t type)
4128 vdev_t *rvd = spa->spa_root_vdev;
4131 * If we're assembling the pool from the split-off vdevs of
4132 * an existing pool, we don't want to attach the spares & cache
4137 * Load any hot spares for this pool.
4139 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object,
4141 if (error != 0 && error != ENOENT)
4142 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4143 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
4144 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
4145 if (load_nvlist(spa, spa->spa_spares.sav_object,
4146 &spa->spa_spares.sav_config) != 0) {
4147 spa_load_failed(spa, "error loading spares nvlist");
4148 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4151 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4152 spa_load_spares(spa);
4153 spa_config_exit(spa, SCL_ALL, FTAG);
4154 } else if (error == 0) {
4155 spa->spa_spares.sav_sync = B_TRUE;
4159 * Load any level 2 ARC devices for this pool.
4161 error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
4162 &spa->spa_l2cache.sav_object, B_FALSE);
4163 if (error != 0 && error != ENOENT)
4164 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4165 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
4166 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
4167 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
4168 &spa->spa_l2cache.sav_config) != 0) {
4169 spa_load_failed(spa, "error loading l2cache nvlist");
4170 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4173 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4174 spa_load_l2cache(spa);
4175 spa_config_exit(spa, SCL_ALL, FTAG);
4176 } else if (error == 0) {
4177 spa->spa_l2cache.sav_sync = B_TRUE;
4184 spa_ld_load_vdev_metadata(spa_t *spa)
4187 vdev_t *rvd = spa->spa_root_vdev;
4190 * If the 'multihost' property is set, then never allow a pool to
4191 * be imported when the system hostid is zero. The exception to
4192 * this rule is zdb which is always allowed to access pools.
4194 if (spa_multihost(spa) && spa_get_hostid(spa) == 0 &&
4195 (spa->spa_import_flags & ZFS_IMPORT_SKIP_MMP) == 0) {
4196 fnvlist_add_uint64(spa->spa_load_info,
4197 ZPOOL_CONFIG_MMP_STATE, MMP_STATE_NO_HOSTID);
4198 return (spa_vdev_err(rvd, VDEV_AUX_ACTIVE, EREMOTEIO));
4202 * If the 'autoreplace' property is set, then post a resource notifying
4203 * the ZFS DE that it should not issue any faults for unopenable
4204 * devices. We also iterate over the vdevs, and post a sysevent for any
4205 * unopenable vdevs so that the normal autoreplace handler can take
4208 if (spa->spa_autoreplace && spa->spa_load_state != SPA_LOAD_TRYIMPORT) {
4209 spa_check_removed(spa->spa_root_vdev);
4211 * For the import case, this is done in spa_import(), because
4212 * at this point we're using the spare definitions from
4213 * the MOS config, not necessarily from the userland config.
4215 if (spa->spa_load_state != SPA_LOAD_IMPORT) {
4216 spa_aux_check_removed(&spa->spa_spares);
4217 spa_aux_check_removed(&spa->spa_l2cache);
4222 * Load the vdev metadata such as metaslabs, DTLs, spacemap object, etc.
4224 error = vdev_load(rvd);
4226 spa_load_failed(spa, "vdev_load failed [error=%d]", error);
4227 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, error));
4230 error = spa_ld_log_spacemaps(spa);
4232 spa_load_failed(spa, "spa_ld_log_sm_data failed [error=%d]",
4234 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, error));
4238 * Propagate the leaf DTLs we just loaded all the way up the vdev tree.
4240 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4241 vdev_dtl_reassess(rvd, 0, 0, B_FALSE, B_FALSE);
4242 spa_config_exit(spa, SCL_ALL, FTAG);
4248 spa_ld_load_dedup_tables(spa_t *spa)
4251 vdev_t *rvd = spa->spa_root_vdev;
4253 error = ddt_load(spa);
4255 spa_load_failed(spa, "ddt_load failed [error=%d]", error);
4256 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4263 spa_ld_verify_logs(spa_t *spa, spa_import_type_t type, char **ereport)
4265 vdev_t *rvd = spa->spa_root_vdev;
4267 if (type != SPA_IMPORT_ASSEMBLE && spa_writeable(spa)) {
4268 boolean_t missing = spa_check_logs(spa);
4270 if (spa->spa_missing_tvds != 0) {
4271 spa_load_note(spa, "spa_check_logs failed "
4272 "so dropping the logs");
4274 *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
4275 spa_load_failed(spa, "spa_check_logs failed");
4276 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG,
4286 spa_ld_verify_pool_data(spa_t *spa)
4289 vdev_t *rvd = spa->spa_root_vdev;
4292 * We've successfully opened the pool, verify that we're ready
4293 * to start pushing transactions.
4295 if (spa->spa_load_state != SPA_LOAD_TRYIMPORT) {
4296 error = spa_load_verify(spa);
4298 spa_load_failed(spa, "spa_load_verify failed "
4299 "[error=%d]", error);
4300 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
4309 spa_ld_claim_log_blocks(spa_t *spa)
4312 dsl_pool_t *dp = spa_get_dsl(spa);
4315 * Claim log blocks that haven't been committed yet.
4316 * This must all happen in a single txg.
4317 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
4318 * invoked from zil_claim_log_block()'s i/o done callback.
4319 * Price of rollback is that we abandon the log.
4321 spa->spa_claiming = B_TRUE;
4323 tx = dmu_tx_create_assigned(dp, spa_first_txg(spa));
4324 (void) dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
4325 zil_claim, tx, DS_FIND_CHILDREN);
4328 spa->spa_claiming = B_FALSE;
4330 spa_set_log_state(spa, SPA_LOG_GOOD);
4334 spa_ld_check_for_config_update(spa_t *spa, uint64_t config_cache_txg,
4335 boolean_t update_config_cache)
4337 vdev_t *rvd = spa->spa_root_vdev;
4338 int need_update = B_FALSE;
4341 * If the config cache is stale, or we have uninitialized
4342 * metaslabs (see spa_vdev_add()), then update the config.
4344 * If this is a verbatim import, trust the current
4345 * in-core spa_config and update the disk labels.
4347 if (update_config_cache || config_cache_txg != spa->spa_config_txg ||
4348 spa->spa_load_state == SPA_LOAD_IMPORT ||
4349 spa->spa_load_state == SPA_LOAD_RECOVER ||
4350 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
4351 need_update = B_TRUE;
4353 for (int c = 0; c < rvd->vdev_children; c++)
4354 if (rvd->vdev_child[c]->vdev_ms_array == 0)
4355 need_update = B_TRUE;
4358 * Update the config cache asynchronously in case we're the
4359 * root pool, in which case the config cache isn't writable yet.
4362 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
4366 spa_ld_prepare_for_reload(spa_t *spa)
4368 spa_mode_t mode = spa->spa_mode;
4369 int async_suspended = spa->spa_async_suspended;
4372 spa_deactivate(spa);
4373 spa_activate(spa, mode);
4376 * We save the value of spa_async_suspended as it gets reset to 0 by
4377 * spa_unload(). We want to restore it back to the original value before
4378 * returning as we might be calling spa_async_resume() later.
4380 spa->spa_async_suspended = async_suspended;
4384 spa_ld_read_checkpoint_txg(spa_t *spa)
4386 uberblock_t checkpoint;
4389 ASSERT0(spa->spa_checkpoint_txg);
4390 ASSERT(MUTEX_HELD(&spa_namespace_lock));
4392 error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
4393 DMU_POOL_ZPOOL_CHECKPOINT, sizeof (uint64_t),
4394 sizeof (uberblock_t) / sizeof (uint64_t), &checkpoint);
4396 if (error == ENOENT)
4402 ASSERT3U(checkpoint.ub_txg, !=, 0);
4403 ASSERT3U(checkpoint.ub_checkpoint_txg, !=, 0);
4404 ASSERT3U(checkpoint.ub_timestamp, !=, 0);
4405 spa->spa_checkpoint_txg = checkpoint.ub_txg;
4406 spa->spa_checkpoint_info.sci_timestamp = checkpoint.ub_timestamp;
4412 spa_ld_mos_init(spa_t *spa, spa_import_type_t type)
4416 ASSERT(MUTEX_HELD(&spa_namespace_lock));
4417 ASSERT(spa->spa_config_source != SPA_CONFIG_SRC_NONE);
4420 * Never trust the config that is provided unless we are assembling
4421 * a pool following a split.
4422 * This means don't trust blkptrs and the vdev tree in general. This
4423 * also effectively puts the spa in read-only mode since
4424 * spa_writeable() checks for spa_trust_config to be true.
4425 * We will later load a trusted config from the MOS.
4427 if (type != SPA_IMPORT_ASSEMBLE)
4428 spa->spa_trust_config = B_FALSE;
4431 * Parse the config provided to create a vdev tree.
4433 error = spa_ld_parse_config(spa, type);
4437 spa_import_progress_add(spa);
4440 * Now that we have the vdev tree, try to open each vdev. This involves
4441 * opening the underlying physical device, retrieving its geometry and
4442 * probing the vdev with a dummy I/O. The state of each vdev will be set
4443 * based on the success of those operations. After this we'll be ready
4444 * to read from the vdevs.
4446 error = spa_ld_open_vdevs(spa);
4451 * Read the label of each vdev and make sure that the GUIDs stored
4452 * there match the GUIDs in the config provided.
4453 * If we're assembling a new pool that's been split off from an
4454 * existing pool, the labels haven't yet been updated so we skip
4455 * validation for now.
4457 if (type != SPA_IMPORT_ASSEMBLE) {
4458 error = spa_ld_validate_vdevs(spa);
4464 * Read all vdev labels to find the best uberblock (i.e. latest,
4465 * unless spa_load_max_txg is set) and store it in spa_uberblock. We
4466 * get the list of features required to read blkptrs in the MOS from
4467 * the vdev label with the best uberblock and verify that our version
4468 * of zfs supports them all.
4470 error = spa_ld_select_uberblock(spa, type);
4475 * Pass that uberblock to the dsl_pool layer which will open the root
4476 * blkptr. This blkptr points to the latest version of the MOS and will
4477 * allow us to read its contents.
4479 error = spa_ld_open_rootbp(spa);
4487 spa_ld_checkpoint_rewind(spa_t *spa)
4489 uberblock_t checkpoint;
4492 ASSERT(MUTEX_HELD(&spa_namespace_lock));
4493 ASSERT(spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT);
4495 error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
4496 DMU_POOL_ZPOOL_CHECKPOINT, sizeof (uint64_t),
4497 sizeof (uberblock_t) / sizeof (uint64_t), &checkpoint);
4500 spa_load_failed(spa, "unable to retrieve checkpointed "
4501 "uberblock from the MOS config [error=%d]", error);
4503 if (error == ENOENT)
4504 error = ZFS_ERR_NO_CHECKPOINT;
4509 ASSERT3U(checkpoint.ub_txg, <, spa->spa_uberblock.ub_txg);
4510 ASSERT3U(checkpoint.ub_txg, ==, checkpoint.ub_checkpoint_txg);
4513 * We need to update the txg and timestamp of the checkpointed
4514 * uberblock to be higher than the latest one. This ensures that
4515 * the checkpointed uberblock is selected if we were to close and
4516 * reopen the pool right after we've written it in the vdev labels.
4517 * (also see block comment in vdev_uberblock_compare)
4519 checkpoint.ub_txg = spa->spa_uberblock.ub_txg + 1;
4520 checkpoint.ub_timestamp = gethrestime_sec();
4523 * Set current uberblock to be the checkpointed uberblock.
4525 spa->spa_uberblock = checkpoint;
4528 * If we are doing a normal rewind, then the pool is open for
4529 * writing and we sync the "updated" checkpointed uberblock to
4530 * disk. Once this is done, we've basically rewound the whole
4531 * pool and there is no way back.
4533 * There are cases when we don't want to attempt and sync the
4534 * checkpointed uberblock to disk because we are opening a
4535 * pool as read-only. Specifically, verifying the checkpointed
4536 * state with zdb, and importing the checkpointed state to get
4537 * a "preview" of its content.
4539 if (spa_writeable(spa)) {
4540 vdev_t *rvd = spa->spa_root_vdev;
4542 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4543 vdev_t *svd[SPA_SYNC_MIN_VDEVS] = { NULL };
4545 int children = rvd->vdev_children;
4546 int c0 = spa_get_random(children);
4548 for (int c = 0; c < children; c++) {
4549 vdev_t *vd = rvd->vdev_child[(c0 + c) % children];
4551 /* Stop when revisiting the first vdev */
4552 if (c > 0 && svd[0] == vd)
4555 if (vd->vdev_ms_array == 0 || vd->vdev_islog ||
4556 !vdev_is_concrete(vd))
4559 svd[svdcount++] = vd;
4560 if (svdcount == SPA_SYNC_MIN_VDEVS)
4563 error = vdev_config_sync(svd, svdcount, spa->spa_first_txg);
4565 spa->spa_last_synced_guid = rvd->vdev_guid;
4566 spa_config_exit(spa, SCL_ALL, FTAG);
4569 spa_load_failed(spa, "failed to write checkpointed "
4570 "uberblock to the vdev labels [error=%d]", error);
4579 spa_ld_mos_with_trusted_config(spa_t *spa, spa_import_type_t type,
4580 boolean_t *update_config_cache)
4585 * Parse the config for pool, open and validate vdevs,
4586 * select an uberblock, and use that uberblock to open
4589 error = spa_ld_mos_init(spa, type);
4594 * Retrieve the trusted config stored in the MOS and use it to create
4595 * a new, exact version of the vdev tree, then reopen all vdevs.
4597 error = spa_ld_trusted_config(spa, type, B_FALSE);
4598 if (error == EAGAIN) {
4599 if (update_config_cache != NULL)
4600 *update_config_cache = B_TRUE;
4603 * Redo the loading process with the trusted config if it is
4604 * too different from the untrusted config.
4606 spa_ld_prepare_for_reload(spa);
4607 spa_load_note(spa, "RELOADING");
4608 error = spa_ld_mos_init(spa, type);
4612 error = spa_ld_trusted_config(spa, type, B_TRUE);
4616 } else if (error != 0) {
4624 * Load an existing storage pool, using the config provided. This config
4625 * describes which vdevs are part of the pool and is later validated against
4626 * partial configs present in each vdev's label and an entire copy of the
4627 * config stored in the MOS.
4630 spa_load_impl(spa_t *spa, spa_import_type_t type, char **ereport)
4633 boolean_t missing_feat_write = B_FALSE;
4634 boolean_t checkpoint_rewind =
4635 (spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT);
4636 boolean_t update_config_cache = B_FALSE;
4638 ASSERT(MUTEX_HELD(&spa_namespace_lock));
4639 ASSERT(spa->spa_config_source != SPA_CONFIG_SRC_NONE);
4641 spa_load_note(spa, "LOADING");
4643 error = spa_ld_mos_with_trusted_config(spa, type, &update_config_cache);
4648 * If we are rewinding to the checkpoint then we need to repeat
4649 * everything we've done so far in this function but this time
4650 * selecting the checkpointed uberblock and using that to open
4653 if (checkpoint_rewind) {
4655 * If we are rewinding to the checkpoint update config cache
4658 update_config_cache = B_TRUE;
4661 * Extract the checkpointed uberblock from the current MOS
4662 * and use this as the pool's uberblock from now on. If the
4663 * pool is imported as writeable we also write the checkpoint
4664 * uberblock to the labels, making the rewind permanent.
4666 error = spa_ld_checkpoint_rewind(spa);
4671 * Redo the loading process again with the
4672 * checkpointed uberblock.
4674 spa_ld_prepare_for_reload(spa);
4675 spa_load_note(spa, "LOADING checkpointed uberblock");
4676 error = spa_ld_mos_with_trusted_config(spa, type, NULL);
4682 * Retrieve the checkpoint txg if the pool has a checkpoint.
4684 error = spa_ld_read_checkpoint_txg(spa);
4689 * Retrieve the mapping of indirect vdevs. Those vdevs were removed
4690 * from the pool and their contents were re-mapped to other vdevs. Note
4691 * that everything that we read before this step must have been
4692 * rewritten on concrete vdevs after the last device removal was
4693 * initiated. Otherwise we could be reading from indirect vdevs before
4694 * we have loaded their mappings.
4696 error = spa_ld_open_indirect_vdev_metadata(spa);
4701 * Retrieve the full list of active features from the MOS and check if
4702 * they are all supported.
4704 error = spa_ld_check_features(spa, &missing_feat_write);
4709 * Load several special directories from the MOS needed by the dsl_pool
4712 error = spa_ld_load_special_directories(spa);
4717 * Retrieve pool properties from the MOS.
4719 error = spa_ld_get_props(spa);
4724 * Retrieve the list of auxiliary devices - cache devices and spares -
4727 error = spa_ld_open_aux_vdevs(spa, type);
4732 * Load the metadata for all vdevs. Also check if unopenable devices
4733 * should be autoreplaced.
4735 error = spa_ld_load_vdev_metadata(spa);
4739 error = spa_ld_load_dedup_tables(spa);
4744 * Verify the logs now to make sure we don't have any unexpected errors
4745 * when we claim log blocks later.
4747 error = spa_ld_verify_logs(spa, type, ereport);
4751 if (missing_feat_write) {
4752 ASSERT(spa->spa_load_state == SPA_LOAD_TRYIMPORT);
4755 * At this point, we know that we can open the pool in
4756 * read-only mode but not read-write mode. We now have enough
4757 * information and can return to userland.
4759 return (spa_vdev_err(spa->spa_root_vdev, VDEV_AUX_UNSUP_FEAT,
4764 * Traverse the last txgs to make sure the pool was left off in a safe
4765 * state. When performing an extreme rewind, we verify the whole pool,
4766 * which can take a very long time.
4768 error = spa_ld_verify_pool_data(spa);
4773 * Calculate the deflated space for the pool. This must be done before
4774 * we write anything to the pool because we'd need to update the space
4775 * accounting using the deflated sizes.
4777 spa_update_dspace(spa);
4780 * We have now retrieved all the information we needed to open the
4781 * pool. If we are importing the pool in read-write mode, a few
4782 * additional steps must be performed to finish the import.
4784 if (spa_writeable(spa) && (spa->spa_load_state == SPA_LOAD_RECOVER ||
4785 spa->spa_load_max_txg == UINT64_MAX)) {
4786 uint64_t config_cache_txg = spa->spa_config_txg;
4788 ASSERT(spa->spa_load_state != SPA_LOAD_TRYIMPORT);
4791 * In case of a checkpoint rewind, log the original txg
4792 * of the checkpointed uberblock.
4794 if (checkpoint_rewind) {
4795 spa_history_log_internal(spa, "checkpoint rewind",
4796 NULL, "rewound state to txg=%llu",
4797 (u_longlong_t)spa->spa_uberblock.ub_checkpoint_txg);
4801 * Traverse the ZIL and claim all blocks.
4803 spa_ld_claim_log_blocks(spa);
4806 * Kick-off the syncing thread.
4808 spa->spa_sync_on = B_TRUE;
4809 txg_sync_start(spa->spa_dsl_pool);
4810 mmp_thread_start(spa);
4813 * Wait for all claims to sync. We sync up to the highest
4814 * claimed log block birth time so that claimed log blocks
4815 * don't appear to be from the future. spa_claim_max_txg
4816 * will have been set for us by ZIL traversal operations
4819 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
4822 * Check if we need to request an update of the config. On the
4823 * next sync, we would update the config stored in vdev labels
4824 * and the cachefile (by default /etc/zfs/zpool.cache).
4826 spa_ld_check_for_config_update(spa, config_cache_txg,
4827 update_config_cache);
4830 * Check if a rebuild was in progress and if so resume it.
4831 * Then check all DTLs to see if anything needs resilvering.
4832 * The resilver will be deferred if a rebuild was started.
4834 if (vdev_rebuild_active(spa->spa_root_vdev)) {
4835 vdev_rebuild_restart(spa);
4836 } else if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
4837 vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
4838 spa_async_request(spa, SPA_ASYNC_RESILVER);
4842 * Log the fact that we booted up (so that we can detect if
4843 * we rebooted in the middle of an operation).
4845 spa_history_log_version(spa, "open", NULL);
4847 spa_restart_removal(spa);
4848 spa_spawn_aux_threads(spa);
4851 * Delete any inconsistent datasets.
4854 * Since we may be issuing deletes for clones here,
4855 * we make sure to do so after we've spawned all the
4856 * auxiliary threads above (from which the livelist
4857 * deletion zthr is part of).
4859 (void) dmu_objset_find(spa_name(spa),
4860 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
4863 * Clean up any stale temporary dataset userrefs.
4865 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
4867 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4868 vdev_initialize_restart(spa->spa_root_vdev);
4869 vdev_trim_restart(spa->spa_root_vdev);
4870 vdev_autotrim_restart(spa);
4871 spa_config_exit(spa, SCL_CONFIG, FTAG);
4874 spa_import_progress_remove(spa_guid(spa));
4875 spa_async_request(spa, SPA_ASYNC_L2CACHE_REBUILD);
4877 spa_load_note(spa, "LOADED");
4883 spa_load_retry(spa_t *spa, spa_load_state_t state)
4885 spa_mode_t mode = spa->spa_mode;
4888 spa_deactivate(spa);
4890 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg - 1;
4892 spa_activate(spa, mode);
4893 spa_async_suspend(spa);
4895 spa_load_note(spa, "spa_load_retry: rewind, max txg: %llu",
4896 (u_longlong_t)spa->spa_load_max_txg);
4898 return (spa_load(spa, state, SPA_IMPORT_EXISTING));
4902 * If spa_load() fails this function will try loading prior txg's. If
4903 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
4904 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
4905 * function will not rewind the pool and will return the same error as
4909 spa_load_best(spa_t *spa, spa_load_state_t state, uint64_t max_request,
4912 nvlist_t *loadinfo = NULL;
4913 nvlist_t *config = NULL;
4914 int load_error, rewind_error;
4915 uint64_t safe_rewind_txg;
4918 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
4919 spa->spa_load_max_txg = spa->spa_load_txg;
4920 spa_set_log_state(spa, SPA_LOG_CLEAR);
4922 spa->spa_load_max_txg = max_request;
4923 if (max_request != UINT64_MAX)
4924 spa->spa_extreme_rewind = B_TRUE;
4927 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING);
4928 if (load_error == 0)
4930 if (load_error == ZFS_ERR_NO_CHECKPOINT) {
4932 * When attempting checkpoint-rewind on a pool with no
4933 * checkpoint, we should not attempt to load uberblocks
4934 * from previous txgs when spa_load fails.
4936 ASSERT(spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT);
4937 spa_import_progress_remove(spa_guid(spa));
4938 return (load_error);
4941 if (spa->spa_root_vdev != NULL)
4942 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4944 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
4945 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
4947 if (rewind_flags & ZPOOL_NEVER_REWIND) {
4948 nvlist_free(config);
4949 spa_import_progress_remove(spa_guid(spa));
4950 return (load_error);
4953 if (state == SPA_LOAD_RECOVER) {
4954 /* Price of rolling back is discarding txgs, including log */
4955 spa_set_log_state(spa, SPA_LOG_CLEAR);
4958 * If we aren't rolling back save the load info from our first
4959 * import attempt so that we can restore it after attempting
4962 loadinfo = spa->spa_load_info;
4963 spa->spa_load_info = fnvlist_alloc();
4966 spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
4967 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
4968 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
4969 TXG_INITIAL : safe_rewind_txg;
4972 * Continue as long as we're finding errors, we're still within
4973 * the acceptable rewind range, and we're still finding uberblocks
4975 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
4976 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
4977 if (spa->spa_load_max_txg < safe_rewind_txg)
4978 spa->spa_extreme_rewind = B_TRUE;
4979 rewind_error = spa_load_retry(spa, state);
4982 spa->spa_extreme_rewind = B_FALSE;
4983 spa->spa_load_max_txg = UINT64_MAX;
4985 if (config && (rewind_error || state != SPA_LOAD_RECOVER))
4986 spa_config_set(spa, config);
4988 nvlist_free(config);
4990 if (state == SPA_LOAD_RECOVER) {
4991 ASSERT3P(loadinfo, ==, NULL);
4992 spa_import_progress_remove(spa_guid(spa));
4993 return (rewind_error);
4995 /* Store the rewind info as part of the initial load info */
4996 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
4997 spa->spa_load_info);
4999 /* Restore the initial load info */
5000 fnvlist_free(spa->spa_load_info);
5001 spa->spa_load_info = loadinfo;
5003 spa_import_progress_remove(spa_guid(spa));
5004 return (load_error);
5011 * The import case is identical to an open except that the configuration is sent
5012 * down from userland, instead of grabbed from the configuration cache. For the
5013 * case of an open, the pool configuration will exist in the
5014 * POOL_STATE_UNINITIALIZED state.
5016 * The stats information (gen/count/ustats) is used to gather vdev statistics at
5017 * the same time open the pool, without having to keep around the spa_t in some
5021 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
5025 spa_load_state_t state = SPA_LOAD_OPEN;
5027 int locked = B_FALSE;
5028 int firstopen = B_FALSE;
5033 * As disgusting as this is, we need to support recursive calls to this
5034 * function because dsl_dir_open() is called during spa_load(), and ends
5035 * up calling spa_open() again. The real fix is to figure out how to
5036 * avoid dsl_dir_open() calling this in the first place.
5038 if (MUTEX_NOT_HELD(&spa_namespace_lock)) {
5039 mutex_enter(&spa_namespace_lock);
5043 if ((spa = spa_lookup(pool)) == NULL) {
5045 mutex_exit(&spa_namespace_lock);
5046 return (SET_ERROR(ENOENT));
5049 if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
5050 zpool_load_policy_t policy;
5054 zpool_get_load_policy(nvpolicy ? nvpolicy : spa->spa_config,
5056 if (policy.zlp_rewind & ZPOOL_DO_REWIND)
5057 state = SPA_LOAD_RECOVER;
5059 spa_activate(spa, spa_mode_global);
5061 if (state != SPA_LOAD_RECOVER)
5062 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
5063 spa->spa_config_source = SPA_CONFIG_SRC_CACHEFILE;
5065 zfs_dbgmsg("spa_open_common: opening %s", pool);
5066 error = spa_load_best(spa, state, policy.zlp_txg,
5069 if (error == EBADF) {
5071 * If vdev_validate() returns failure (indicated by
5072 * EBADF), it indicates that one of the vdevs indicates
5073 * that the pool has been exported or destroyed. If
5074 * this is the case, the config cache is out of sync and
5075 * we should remove the pool from the namespace.
5078 spa_deactivate(spa);
5079 spa_write_cachefile(spa, B_TRUE, B_TRUE);
5082 mutex_exit(&spa_namespace_lock);
5083 return (SET_ERROR(ENOENT));
5088 * We can't open the pool, but we still have useful
5089 * information: the state of each vdev after the
5090 * attempted vdev_open(). Return this to the user.
5092 if (config != NULL && spa->spa_config) {
5093 VERIFY(nvlist_dup(spa->spa_config, config,
5095 VERIFY(nvlist_add_nvlist(*config,
5096 ZPOOL_CONFIG_LOAD_INFO,
5097 spa->spa_load_info) == 0);
5100 spa_deactivate(spa);
5101 spa->spa_last_open_failed = error;
5103 mutex_exit(&spa_namespace_lock);
5109 spa_open_ref(spa, tag);
5112 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
5115 * If we've recovered the pool, pass back any information we
5116 * gathered while doing the load.
5118 if (state == SPA_LOAD_RECOVER) {
5119 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
5120 spa->spa_load_info) == 0);
5124 spa->spa_last_open_failed = 0;
5125 spa->spa_last_ubsync_txg = 0;
5126 spa->spa_load_txg = 0;
5127 mutex_exit(&spa_namespace_lock);
5131 zvol_create_minors_recursive(spa_name(spa));
5139 spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
5142 return (spa_open_common(name, spapp, tag, policy, config));
5146 spa_open(const char *name, spa_t **spapp, void *tag)
5148 return (spa_open_common(name, spapp, tag, NULL, NULL));
5152 * Lookup the given spa_t, incrementing the inject count in the process,
5153 * preventing it from being exported or destroyed.
5156 spa_inject_addref(char *name)
5160 mutex_enter(&spa_namespace_lock);
5161 if ((spa = spa_lookup(name)) == NULL) {
5162 mutex_exit(&spa_namespace_lock);
5165 spa->spa_inject_ref++;
5166 mutex_exit(&spa_namespace_lock);
5172 spa_inject_delref(spa_t *spa)
5174 mutex_enter(&spa_namespace_lock);
5175 spa->spa_inject_ref--;
5176 mutex_exit(&spa_namespace_lock);
5180 * Add spares device information to the nvlist.
5183 spa_add_spares(spa_t *spa, nvlist_t *config)
5193 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
5195 if (spa->spa_spares.sav_count == 0)
5198 VERIFY(nvlist_lookup_nvlist(config,
5199 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
5200 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5201 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
5203 VERIFY(nvlist_add_nvlist_array(nvroot,
5204 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
5205 VERIFY(nvlist_lookup_nvlist_array(nvroot,
5206 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
5209 * Go through and find any spares which have since been
5210 * repurposed as an active spare. If this is the case, update
5211 * their status appropriately.
5213 for (i = 0; i < nspares; i++) {
5214 VERIFY(nvlist_lookup_uint64(spares[i],
5215 ZPOOL_CONFIG_GUID, &guid) == 0);
5216 if (spa_spare_exists(guid, &pool, NULL) &&
5218 VERIFY(nvlist_lookup_uint64_array(
5219 spares[i], ZPOOL_CONFIG_VDEV_STATS,
5220 (uint64_t **)&vs, &vsc) == 0);
5221 vs->vs_state = VDEV_STATE_CANT_OPEN;
5222 vs->vs_aux = VDEV_AUX_SPARED;
5229 * Add l2cache device information to the nvlist, including vdev stats.
5232 spa_add_l2cache(spa_t *spa, nvlist_t *config)
5235 uint_t i, j, nl2cache;
5242 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
5244 if (spa->spa_l2cache.sav_count == 0)
5247 VERIFY(nvlist_lookup_nvlist(config,
5248 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
5249 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5250 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
5251 if (nl2cache != 0) {
5252 VERIFY(nvlist_add_nvlist_array(nvroot,
5253 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
5254 VERIFY(nvlist_lookup_nvlist_array(nvroot,
5255 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
5258 * Update level 2 cache device stats.
5261 for (i = 0; i < nl2cache; i++) {
5262 VERIFY(nvlist_lookup_uint64(l2cache[i],
5263 ZPOOL_CONFIG_GUID, &guid) == 0);
5266 for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
5268 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
5269 vd = spa->spa_l2cache.sav_vdevs[j];
5275 VERIFY(nvlist_lookup_uint64_array(l2cache[i],
5276 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
5278 vdev_get_stats(vd, vs);
5279 vdev_config_generate_stats(vd, l2cache[i]);
5286 spa_feature_stats_from_disk(spa_t *spa, nvlist_t *features)
5291 if (spa->spa_feat_for_read_obj != 0) {
5292 for (zap_cursor_init(&zc, spa->spa_meta_objset,
5293 spa->spa_feat_for_read_obj);
5294 zap_cursor_retrieve(&zc, &za) == 0;
5295 zap_cursor_advance(&zc)) {
5296 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
5297 za.za_num_integers == 1);
5298 VERIFY0(nvlist_add_uint64(features, za.za_name,
5299 za.za_first_integer));
5301 zap_cursor_fini(&zc);
5304 if (spa->spa_feat_for_write_obj != 0) {
5305 for (zap_cursor_init(&zc, spa->spa_meta_objset,
5306 spa->spa_feat_for_write_obj);
5307 zap_cursor_retrieve(&zc, &za) == 0;
5308 zap_cursor_advance(&zc)) {
5309 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
5310 za.za_num_integers == 1);
5311 VERIFY0(nvlist_add_uint64(features, za.za_name,
5312 za.za_first_integer));
5314 zap_cursor_fini(&zc);
5319 spa_feature_stats_from_cache(spa_t *spa, nvlist_t *features)
5323 for (i = 0; i < SPA_FEATURES; i++) {
5324 zfeature_info_t feature = spa_feature_table[i];
5327 if (feature_get_refcount(spa, &feature, &refcount) != 0)
5330 VERIFY0(nvlist_add_uint64(features, feature.fi_guid, refcount));
5335 * Store a list of pool features and their reference counts in the
5338 * The first time this is called on a spa, allocate a new nvlist, fetch
5339 * the pool features and reference counts from disk, then save the list
5340 * in the spa. In subsequent calls on the same spa use the saved nvlist
5341 * and refresh its values from the cached reference counts. This
5342 * ensures we don't block here on I/O on a suspended pool so 'zpool
5343 * clear' can resume the pool.
5346 spa_add_feature_stats(spa_t *spa, nvlist_t *config)
5350 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
5352 mutex_enter(&spa->spa_feat_stats_lock);
5353 features = spa->spa_feat_stats;
5355 if (features != NULL) {
5356 spa_feature_stats_from_cache(spa, features);
5358 VERIFY0(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP));
5359 spa->spa_feat_stats = features;
5360 spa_feature_stats_from_disk(spa, features);
5363 VERIFY0(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
5366 mutex_exit(&spa->spa_feat_stats_lock);
5370 spa_get_stats(const char *name, nvlist_t **config,
5371 char *altroot, size_t buflen)
5377 error = spa_open_common(name, &spa, FTAG, NULL, config);
5381 * This still leaves a window of inconsistency where the spares
5382 * or l2cache devices could change and the config would be
5383 * self-inconsistent.
5385 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5387 if (*config != NULL) {
5388 uint64_t loadtimes[2];
5390 loadtimes[0] = spa->spa_loaded_ts.tv_sec;
5391 loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
5392 VERIFY(nvlist_add_uint64_array(*config,
5393 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
5395 VERIFY(nvlist_add_uint64(*config,
5396 ZPOOL_CONFIG_ERRCOUNT,
5397 spa_get_errlog_size(spa)) == 0);
5399 if (spa_suspended(spa)) {
5400 VERIFY(nvlist_add_uint64(*config,
5401 ZPOOL_CONFIG_SUSPENDED,
5402 spa->spa_failmode) == 0);
5403 VERIFY(nvlist_add_uint64(*config,
5404 ZPOOL_CONFIG_SUSPENDED_REASON,
5405 spa->spa_suspended) == 0);
5408 spa_add_spares(spa, *config);
5409 spa_add_l2cache(spa, *config);
5410 spa_add_feature_stats(spa, *config);
5415 * We want to get the alternate root even for faulted pools, so we cheat
5416 * and call spa_lookup() directly.
5420 mutex_enter(&spa_namespace_lock);
5421 spa = spa_lookup(name);
5423 spa_altroot(spa, altroot, buflen);
5427 mutex_exit(&spa_namespace_lock);
5429 spa_altroot(spa, altroot, buflen);
5434 spa_config_exit(spa, SCL_CONFIG, FTAG);
5435 spa_close(spa, FTAG);
5442 * Validate that the auxiliary device array is well formed. We must have an
5443 * array of nvlists, each which describes a valid leaf vdev. If this is an
5444 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
5445 * specified, as long as they are well-formed.
5448 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
5449 spa_aux_vdev_t *sav, const char *config, uint64_t version,
5450 vdev_labeltype_t label)
5457 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5460 * It's acceptable to have no devs specified.
5462 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
5466 return (SET_ERROR(EINVAL));
5469 * Make sure the pool is formatted with a version that supports this
5472 if (spa_version(spa) < version)
5473 return (SET_ERROR(ENOTSUP));
5476 * Set the pending device list so we correctly handle device in-use
5479 sav->sav_pending = dev;
5480 sav->sav_npending = ndev;
5482 for (i = 0; i < ndev; i++) {
5483 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
5487 if (!vd->vdev_ops->vdev_op_leaf) {
5489 error = SET_ERROR(EINVAL);
5495 if ((error = vdev_open(vd)) == 0 &&
5496 (error = vdev_label_init(vd, crtxg, label)) == 0) {
5497 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
5498 vd->vdev_guid) == 0);
5504 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
5511 sav->sav_pending = NULL;
5512 sav->sav_npending = 0;
5517 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
5521 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5523 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
5524 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
5525 VDEV_LABEL_SPARE)) != 0) {
5529 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
5530 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
5531 VDEV_LABEL_L2CACHE));
5535 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
5540 if (sav->sav_config != NULL) {
5546 * Generate new dev list by concatenating with the
5549 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
5550 &olddevs, &oldndevs) == 0);
5552 newdevs = kmem_alloc(sizeof (void *) *
5553 (ndevs + oldndevs), KM_SLEEP);
5554 for (i = 0; i < oldndevs; i++)
5555 VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
5557 for (i = 0; i < ndevs; i++)
5558 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
5561 VERIFY(nvlist_remove(sav->sav_config, config,
5562 DATA_TYPE_NVLIST_ARRAY) == 0);
5564 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
5565 config, newdevs, ndevs + oldndevs) == 0);
5566 for (i = 0; i < oldndevs + ndevs; i++)
5567 nvlist_free(newdevs[i]);
5568 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
5571 * Generate a new dev list.
5573 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
5575 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
5581 * Stop and drop level 2 ARC devices
5584 spa_l2cache_drop(spa_t *spa)
5588 spa_aux_vdev_t *sav = &spa->spa_l2cache;
5590 for (i = 0; i < sav->sav_count; i++) {
5593 vd = sav->sav_vdevs[i];
5596 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
5597 pool != 0ULL && l2arc_vdev_present(vd))
5598 l2arc_remove_vdev(vd);
5603 * Verify encryption parameters for spa creation. If we are encrypting, we must
5604 * have the encryption feature flag enabled.
5607 spa_create_check_encryption_params(dsl_crypto_params_t *dcp,
5608 boolean_t has_encryption)
5610 if (dcp->cp_crypt != ZIO_CRYPT_OFF &&
5611 dcp->cp_crypt != ZIO_CRYPT_INHERIT &&
5613 return (SET_ERROR(ENOTSUP));
5615 return (dmu_objset_create_crypt_check(NULL, dcp, NULL));
5622 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
5623 nvlist_t *zplprops, dsl_crypto_params_t *dcp)
5626 char *altroot = NULL;
5631 uint64_t txg = TXG_INITIAL;
5632 nvlist_t **spares, **l2cache;
5633 uint_t nspares, nl2cache;
5634 uint64_t version, obj;
5635 boolean_t has_features;
5636 boolean_t has_encryption;
5637 boolean_t has_allocclass;
5643 if (props == NULL ||
5644 nvlist_lookup_string(props, "tname", &poolname) != 0)
5645 poolname = (char *)pool;
5648 * If this pool already exists, return failure.
5650 mutex_enter(&spa_namespace_lock);
5651 if (spa_lookup(poolname) != NULL) {
5652 mutex_exit(&spa_namespace_lock);
5653 return (SET_ERROR(EEXIST));
5657 * Allocate a new spa_t structure.
5659 nvl = fnvlist_alloc();
5660 fnvlist_add_string(nvl, ZPOOL_CONFIG_POOL_NAME, pool);
5661 (void) nvlist_lookup_string(props,
5662 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
5663 spa = spa_add(poolname, nvl, altroot);
5665 spa_activate(spa, spa_mode_global);
5667 if (props && (error = spa_prop_validate(spa, props))) {
5668 spa_deactivate(spa);
5670 mutex_exit(&spa_namespace_lock);
5675 * Temporary pool names should never be written to disk.
5677 if (poolname != pool)
5678 spa->spa_import_flags |= ZFS_IMPORT_TEMP_NAME;
5680 has_features = B_FALSE;
5681 has_encryption = B_FALSE;
5682 has_allocclass = B_FALSE;
5683 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
5684 elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
5685 if (zpool_prop_feature(nvpair_name(elem))) {
5686 has_features = B_TRUE;
5688 feat_name = strchr(nvpair_name(elem), '@') + 1;
5689 VERIFY0(zfeature_lookup_name(feat_name, &feat));
5690 if (feat == SPA_FEATURE_ENCRYPTION)
5691 has_encryption = B_TRUE;
5692 if (feat == SPA_FEATURE_ALLOCATION_CLASSES)
5693 has_allocclass = B_TRUE;
5697 /* verify encryption params, if they were provided */
5699 error = spa_create_check_encryption_params(dcp, has_encryption);
5701 spa_deactivate(spa);
5703 mutex_exit(&spa_namespace_lock);
5707 if (!has_allocclass && zfs_special_devs(nvroot, NULL)) {
5708 spa_deactivate(spa);
5710 mutex_exit(&spa_namespace_lock);
5714 if (has_features || nvlist_lookup_uint64(props,
5715 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
5716 version = SPA_VERSION;
5718 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
5720 spa->spa_first_txg = txg;
5721 spa->spa_uberblock.ub_txg = txg - 1;
5722 spa->spa_uberblock.ub_version = version;
5723 spa->spa_ubsync = spa->spa_uberblock;
5724 spa->spa_load_state = SPA_LOAD_CREATE;
5725 spa->spa_removing_phys.sr_state = DSS_NONE;
5726 spa->spa_removing_phys.sr_removing_vdev = -1;
5727 spa->spa_removing_phys.sr_prev_indirect_vdev = -1;
5728 spa->spa_indirect_vdevs_loaded = B_TRUE;
5731 * Create "The Godfather" zio to hold all async IOs
5733 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
5735 for (int i = 0; i < max_ncpus; i++) {
5736 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
5737 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
5738 ZIO_FLAG_GODFATHER);
5742 * Create the root vdev.
5744 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5746 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
5748 ASSERT(error != 0 || rvd != NULL);
5749 ASSERT(error != 0 || spa->spa_root_vdev == rvd);
5751 if (error == 0 && !zfs_allocatable_devs(nvroot))
5752 error = SET_ERROR(EINVAL);
5755 (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
5756 (error = spa_validate_aux(spa, nvroot, txg,
5757 VDEV_ALLOC_ADD)) == 0) {
5759 * instantiate the metaslab groups (this will dirty the vdevs)
5760 * we can no longer error exit past this point
5762 for (int c = 0; error == 0 && c < rvd->vdev_children; c++) {
5763 vdev_t *vd = rvd->vdev_child[c];
5765 vdev_metaslab_set_size(vd);
5766 vdev_expand(vd, txg);
5770 spa_config_exit(spa, SCL_ALL, FTAG);
5774 spa_deactivate(spa);
5776 mutex_exit(&spa_namespace_lock);
5781 * Get the list of spares, if specified.
5783 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
5784 &spares, &nspares) == 0) {
5785 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
5787 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
5788 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
5789 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5790 spa_load_spares(spa);
5791 spa_config_exit(spa, SCL_ALL, FTAG);
5792 spa->spa_spares.sav_sync = B_TRUE;
5796 * Get the list of level 2 cache devices, if specified.
5798 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
5799 &l2cache, &nl2cache) == 0) {
5800 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
5801 NV_UNIQUE_NAME, KM_SLEEP) == 0);
5802 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
5803 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
5804 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5805 spa_load_l2cache(spa);
5806 spa_config_exit(spa, SCL_ALL, FTAG);
5807 spa->spa_l2cache.sav_sync = B_TRUE;
5810 spa->spa_is_initializing = B_TRUE;
5811 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, dcp, txg);
5812 spa->spa_is_initializing = B_FALSE;
5815 * Create DDTs (dedup tables).
5819 spa_update_dspace(spa);
5821 tx = dmu_tx_create_assigned(dp, txg);
5824 * Create the pool's history object.
5826 if (version >= SPA_VERSION_ZPOOL_HISTORY && !spa->spa_history)
5827 spa_history_create_obj(spa, tx);
5829 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_CREATE);
5830 spa_history_log_version(spa, "create", tx);
5833 * Create the pool config object.
5835 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
5836 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
5837 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
5839 if (zap_add(spa->spa_meta_objset,
5840 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
5841 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
5842 cmn_err(CE_PANIC, "failed to add pool config");
5845 if (zap_add(spa->spa_meta_objset,
5846 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
5847 sizeof (uint64_t), 1, &version, tx) != 0) {
5848 cmn_err(CE_PANIC, "failed to add pool version");
5851 /* Newly created pools with the right version are always deflated. */
5852 if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
5853 spa->spa_deflate = TRUE;
5854 if (zap_add(spa->spa_meta_objset,
5855 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
5856 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
5857 cmn_err(CE_PANIC, "failed to add deflate");
5862 * Create the deferred-free bpobj. Turn off compression
5863 * because sync-to-convergence takes longer if the blocksize
5866 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
5867 dmu_object_set_compress(spa->spa_meta_objset, obj,
5868 ZIO_COMPRESS_OFF, tx);
5869 if (zap_add(spa->spa_meta_objset,
5870 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
5871 sizeof (uint64_t), 1, &obj, tx) != 0) {
5872 cmn_err(CE_PANIC, "failed to add bpobj");
5874 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
5875 spa->spa_meta_objset, obj));
5878 * Generate some random noise for salted checksums to operate on.
5880 (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes,
5881 sizeof (spa->spa_cksum_salt.zcs_bytes));
5884 * Set pool properties.
5886 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
5887 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
5888 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
5889 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
5890 spa->spa_multihost = zpool_prop_default_numeric(ZPOOL_PROP_MULTIHOST);
5891 spa->spa_autotrim = zpool_prop_default_numeric(ZPOOL_PROP_AUTOTRIM);
5893 if (props != NULL) {
5894 spa_configfile_set(spa, props, B_FALSE);
5895 spa_sync_props(props, tx);
5900 spa->spa_sync_on = B_TRUE;
5902 mmp_thread_start(spa);
5903 txg_wait_synced(dp, txg);
5905 spa_spawn_aux_threads(spa);
5907 spa_write_cachefile(spa, B_FALSE, B_TRUE);
5910 * Don't count references from objsets that are already closed
5911 * and are making their way through the eviction process.
5913 spa_evicting_os_wait(spa);
5914 spa->spa_minref = zfs_refcount_count(&spa->spa_refcount);
5915 spa->spa_load_state = SPA_LOAD_NONE;
5917 mutex_exit(&spa_namespace_lock);
5923 * Import a non-root pool into the system.
5926 spa_import(char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
5929 char *altroot = NULL;
5930 spa_load_state_t state = SPA_LOAD_IMPORT;
5931 zpool_load_policy_t policy;
5932 spa_mode_t mode = spa_mode_global;
5933 uint64_t readonly = B_FALSE;
5936 nvlist_t **spares, **l2cache;
5937 uint_t nspares, nl2cache;
5940 * If a pool with this name exists, return failure.
5942 mutex_enter(&spa_namespace_lock);
5943 if (spa_lookup(pool) != NULL) {
5944 mutex_exit(&spa_namespace_lock);
5945 return (SET_ERROR(EEXIST));
5949 * Create and initialize the spa structure.
5951 (void) nvlist_lookup_string(props,
5952 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
5953 (void) nvlist_lookup_uint64(props,
5954 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
5956 mode = SPA_MODE_READ;
5957 spa = spa_add(pool, config, altroot);
5958 spa->spa_import_flags = flags;
5961 * Verbatim import - Take a pool and insert it into the namespace
5962 * as if it had been loaded at boot.
5964 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
5966 spa_configfile_set(spa, props, B_FALSE);
5968 spa_write_cachefile(spa, B_FALSE, B_TRUE);
5969 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_IMPORT);
5970 zfs_dbgmsg("spa_import: verbatim import of %s", pool);
5971 mutex_exit(&spa_namespace_lock);
5975 spa_activate(spa, mode);
5978 * Don't start async tasks until we know everything is healthy.
5980 spa_async_suspend(spa);
5982 zpool_get_load_policy(config, &policy);
5983 if (policy.zlp_rewind & ZPOOL_DO_REWIND)
5984 state = SPA_LOAD_RECOVER;
5986 spa->spa_config_source = SPA_CONFIG_SRC_TRYIMPORT;
5988 if (state != SPA_LOAD_RECOVER) {
5989 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
5990 zfs_dbgmsg("spa_import: importing %s", pool);
5992 zfs_dbgmsg("spa_import: importing %s, max_txg=%lld "
5993 "(RECOVERY MODE)", pool, (longlong_t)policy.zlp_txg);
5995 error = spa_load_best(spa, state, policy.zlp_txg, policy.zlp_rewind);
5998 * Propagate anything learned while loading the pool and pass it
5999 * back to caller (i.e. rewind info, missing devices, etc).
6001 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
6002 spa->spa_load_info) == 0);
6004 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6006 * Toss any existing sparelist, as it doesn't have any validity
6007 * anymore, and conflicts with spa_has_spare().
6009 if (spa->spa_spares.sav_config) {
6010 nvlist_free(spa->spa_spares.sav_config);
6011 spa->spa_spares.sav_config = NULL;
6012 spa_load_spares(spa);
6014 if (spa->spa_l2cache.sav_config) {
6015 nvlist_free(spa->spa_l2cache.sav_config);
6016 spa->spa_l2cache.sav_config = NULL;
6017 spa_load_l2cache(spa);
6020 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
6022 spa_config_exit(spa, SCL_ALL, FTAG);
6025 spa_configfile_set(spa, props, B_FALSE);
6027 if (error != 0 || (props && spa_writeable(spa) &&
6028 (error = spa_prop_set(spa, props)))) {
6030 spa_deactivate(spa);
6032 mutex_exit(&spa_namespace_lock);
6036 spa_async_resume(spa);
6039 * Override any spares and level 2 cache devices as specified by
6040 * the user, as these may have correct device names/devids, etc.
6042 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
6043 &spares, &nspares) == 0) {
6044 if (spa->spa_spares.sav_config)
6045 VERIFY(nvlist_remove(spa->spa_spares.sav_config,
6046 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
6048 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
6049 NV_UNIQUE_NAME, KM_SLEEP) == 0);
6050 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
6051 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
6052 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6053 spa_load_spares(spa);
6054 spa_config_exit(spa, SCL_ALL, FTAG);
6055 spa->spa_spares.sav_sync = B_TRUE;
6057 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
6058 &l2cache, &nl2cache) == 0) {
6059 if (spa->spa_l2cache.sav_config)
6060 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
6061 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
6063 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
6064 NV_UNIQUE_NAME, KM_SLEEP) == 0);
6065 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
6066 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
6067 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6068 spa_load_l2cache(spa);
6069 spa_config_exit(spa, SCL_ALL, FTAG);
6070 spa->spa_l2cache.sav_sync = B_TRUE;
6074 * Check for any removed devices.
6076 if (spa->spa_autoreplace) {
6077 spa_aux_check_removed(&spa->spa_spares);
6078 spa_aux_check_removed(&spa->spa_l2cache);
6081 if (spa_writeable(spa)) {
6083 * Update the config cache to include the newly-imported pool.
6085 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
6089 * It's possible that the pool was expanded while it was exported.
6090 * We kick off an async task to handle this for us.
6092 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
6094 spa_history_log_version(spa, "import", NULL);
6096 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_IMPORT);
6098 mutex_exit(&spa_namespace_lock);
6100 zvol_create_minors_recursive(pool);
6106 spa_tryimport(nvlist_t *tryconfig)
6108 nvlist_t *config = NULL;
6109 char *poolname, *cachefile;
6113 zpool_load_policy_t policy;
6115 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
6118 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
6122 * Create and initialize the spa structure.
6124 mutex_enter(&spa_namespace_lock);
6125 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
6126 spa_activate(spa, SPA_MODE_READ);
6129 * Rewind pool if a max txg was provided.
6131 zpool_get_load_policy(spa->spa_config, &policy);
6132 if (policy.zlp_txg != UINT64_MAX) {
6133 spa->spa_load_max_txg = policy.zlp_txg;
6134 spa->spa_extreme_rewind = B_TRUE;
6135 zfs_dbgmsg("spa_tryimport: importing %s, max_txg=%lld",
6136 poolname, (longlong_t)policy.zlp_txg);
6138 zfs_dbgmsg("spa_tryimport: importing %s", poolname);
6141 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_CACHEFILE, &cachefile)
6143 zfs_dbgmsg("spa_tryimport: using cachefile '%s'", cachefile);
6144 spa->spa_config_source = SPA_CONFIG_SRC_CACHEFILE;
6146 spa->spa_config_source = SPA_CONFIG_SRC_SCAN;
6149 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING);
6152 * If 'tryconfig' was at least parsable, return the current config.
6154 if (spa->spa_root_vdev != NULL) {
6155 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
6156 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
6158 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
6160 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
6161 spa->spa_uberblock.ub_timestamp) == 0);
6162 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
6163 spa->spa_load_info) == 0);
6164 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_ERRATA,
6165 spa->spa_errata) == 0);
6168 * If the bootfs property exists on this pool then we
6169 * copy it out so that external consumers can tell which
6170 * pools are bootable.
6172 if ((!error || error == EEXIST) && spa->spa_bootfs) {
6173 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
6176 * We have to play games with the name since the
6177 * pool was opened as TRYIMPORT_NAME.
6179 if (dsl_dsobj_to_dsname(spa_name(spa),
6180 spa->spa_bootfs, tmpname) == 0) {
6184 dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
6186 cp = strchr(tmpname, '/');
6188 (void) strlcpy(dsname, tmpname,
6191 (void) snprintf(dsname, MAXPATHLEN,
6192 "%s/%s", poolname, ++cp);
6194 VERIFY(nvlist_add_string(config,
6195 ZPOOL_CONFIG_BOOTFS, dsname) == 0);
6196 kmem_free(dsname, MAXPATHLEN);
6198 kmem_free(tmpname, MAXPATHLEN);
6202 * Add the list of hot spares and level 2 cache devices.
6204 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
6205 spa_add_spares(spa, config);
6206 spa_add_l2cache(spa, config);
6207 spa_config_exit(spa, SCL_CONFIG, FTAG);
6211 spa_deactivate(spa);
6213 mutex_exit(&spa_namespace_lock);
6219 * Pool export/destroy
6221 * The act of destroying or exporting a pool is very simple. We make sure there
6222 * is no more pending I/O and any references to the pool are gone. Then, we
6223 * update the pool state and sync all the labels to disk, removing the
6224 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
6225 * we don't sync the labels or remove the configuration cache.
6228 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
6229 boolean_t force, boolean_t hardforce)
6236 if (!(spa_mode_global & SPA_MODE_WRITE))
6237 return (SET_ERROR(EROFS));
6239 mutex_enter(&spa_namespace_lock);
6240 if ((spa = spa_lookup(pool)) == NULL) {
6241 mutex_exit(&spa_namespace_lock);
6242 return (SET_ERROR(ENOENT));
6245 if (spa->spa_is_exporting) {
6246 /* the pool is being exported by another thread */
6247 mutex_exit(&spa_namespace_lock);
6248 return (SET_ERROR(ZFS_ERR_EXPORT_IN_PROGRESS));
6250 spa->spa_is_exporting = B_TRUE;
6253 * Put a hold on the pool, drop the namespace lock, stop async tasks,
6254 * reacquire the namespace lock, and see if we can export.
6256 spa_open_ref(spa, FTAG);
6257 mutex_exit(&spa_namespace_lock);
6258 spa_async_suspend(spa);
6259 if (spa->spa_zvol_taskq) {
6260 zvol_remove_minors(spa, spa_name(spa), B_TRUE);
6261 taskq_wait(spa->spa_zvol_taskq);
6263 mutex_enter(&spa_namespace_lock);
6264 spa_close(spa, FTAG);
6266 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
6269 * The pool will be in core if it's openable, in which case we can
6270 * modify its state. Objsets may be open only because they're dirty,
6271 * so we have to force it to sync before checking spa_refcnt.
6273 if (spa->spa_sync_on) {
6274 txg_wait_synced(spa->spa_dsl_pool, 0);
6275 spa_evicting_os_wait(spa);
6279 * A pool cannot be exported or destroyed if there are active
6280 * references. If we are resetting a pool, allow references by
6281 * fault injection handlers.
6283 if (!spa_refcount_zero(spa) ||
6284 (spa->spa_inject_ref != 0 &&
6285 new_state != POOL_STATE_UNINITIALIZED)) {
6286 spa_async_resume(spa);
6287 spa->spa_is_exporting = B_FALSE;
6288 mutex_exit(&spa_namespace_lock);
6289 return (SET_ERROR(EBUSY));
6292 if (spa->spa_sync_on) {
6294 * A pool cannot be exported if it has an active shared spare.
6295 * This is to prevent other pools stealing the active spare
6296 * from an exported pool. At user's own will, such pool can
6297 * be forcedly exported.
6299 if (!force && new_state == POOL_STATE_EXPORTED &&
6300 spa_has_active_shared_spare(spa)) {
6301 spa_async_resume(spa);
6302 spa->spa_is_exporting = B_FALSE;
6303 mutex_exit(&spa_namespace_lock);
6304 return (SET_ERROR(EXDEV));
6308 * We're about to export or destroy this pool. Make sure
6309 * we stop all initialization and trim activity here before
6310 * we set the spa_final_txg. This will ensure that all
6311 * dirty data resulting from the initialization is
6312 * committed to disk before we unload the pool.
6314 if (spa->spa_root_vdev != NULL) {
6315 vdev_t *rvd = spa->spa_root_vdev;
6316 vdev_initialize_stop_all(rvd, VDEV_INITIALIZE_ACTIVE);
6317 vdev_trim_stop_all(rvd, VDEV_TRIM_ACTIVE);
6318 vdev_autotrim_stop_all(spa);
6319 vdev_rebuild_stop_all(spa);
6323 * We want this to be reflected on every label,
6324 * so mark them all dirty. spa_unload() will do the
6325 * final sync that pushes these changes out.
6327 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
6328 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6329 spa->spa_state = new_state;
6330 spa->spa_final_txg = spa_last_synced_txg(spa) +
6332 vdev_config_dirty(spa->spa_root_vdev);
6333 spa_config_exit(spa, SCL_ALL, FTAG);
6338 if (new_state == POOL_STATE_DESTROYED)
6339 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_DESTROY);
6340 else if (new_state == POOL_STATE_EXPORTED)
6341 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_EXPORT);
6343 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
6345 spa_deactivate(spa);
6348 if (oldconfig && spa->spa_config)
6349 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
6351 if (new_state != POOL_STATE_UNINITIALIZED) {
6353 spa_write_cachefile(spa, B_TRUE, B_TRUE);
6357 * If spa_remove() is not called for this spa_t and
6358 * there is any possibility that it can be reused,
6359 * we make sure to reset the exporting flag.
6361 spa->spa_is_exporting = B_FALSE;
6364 mutex_exit(&spa_namespace_lock);
6369 * Destroy a storage pool.
6372 spa_destroy(char *pool)
6374 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
6379 * Export a storage pool.
6382 spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
6383 boolean_t hardforce)
6385 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
6390 * Similar to spa_export(), this unloads the spa_t without actually removing it
6391 * from the namespace in any way.
6394 spa_reset(char *pool)
6396 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
6401 * ==========================================================================
6402 * Device manipulation
6403 * ==========================================================================
6407 * Add a device to a storage pool.
6410 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
6414 vdev_t *rvd = spa->spa_root_vdev;
6416 nvlist_t **spares, **l2cache;
6417 uint_t nspares, nl2cache;
6419 ASSERT(spa_writeable(spa));
6421 txg = spa_vdev_enter(spa);
6423 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
6424 VDEV_ALLOC_ADD)) != 0)
6425 return (spa_vdev_exit(spa, NULL, txg, error));
6427 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */
6429 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
6433 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
6437 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
6438 return (spa_vdev_exit(spa, vd, txg, EINVAL));
6440 if (vd->vdev_children != 0 &&
6441 (error = vdev_create(vd, txg, B_FALSE)) != 0)
6442 return (spa_vdev_exit(spa, vd, txg, error));
6445 * We must validate the spares and l2cache devices after checking the
6446 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
6448 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
6449 return (spa_vdev_exit(spa, vd, txg, error));
6452 * If we are in the middle of a device removal, we can only add
6453 * devices which match the existing devices in the pool.
6454 * If we are in the middle of a removal, or have some indirect
6455 * vdevs, we can not add raidz toplevels.
6457 if (spa->spa_vdev_removal != NULL ||
6458 spa->spa_removing_phys.sr_prev_indirect_vdev != -1) {
6459 for (int c = 0; c < vd->vdev_children; c++) {
6460 tvd = vd->vdev_child[c];
6461 if (spa->spa_vdev_removal != NULL &&
6462 tvd->vdev_ashift != spa->spa_max_ashift) {
6463 return (spa_vdev_exit(spa, vd, txg, EINVAL));
6465 /* Fail if top level vdev is raidz */
6466 if (tvd->vdev_ops == &vdev_raidz_ops) {
6467 return (spa_vdev_exit(spa, vd, txg, EINVAL));
6470 * Need the top level mirror to be
6471 * a mirror of leaf vdevs only
6473 if (tvd->vdev_ops == &vdev_mirror_ops) {
6474 for (uint64_t cid = 0;
6475 cid < tvd->vdev_children; cid++) {
6476 vdev_t *cvd = tvd->vdev_child[cid];
6477 if (!cvd->vdev_ops->vdev_op_leaf) {
6478 return (spa_vdev_exit(spa, vd,
6486 for (int c = 0; c < vd->vdev_children; c++) {
6487 tvd = vd->vdev_child[c];
6488 vdev_remove_child(vd, tvd);
6489 tvd->vdev_id = rvd->vdev_children;
6490 vdev_add_child(rvd, tvd);
6491 vdev_config_dirty(tvd);
6495 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
6496 ZPOOL_CONFIG_SPARES);
6497 spa_load_spares(spa);
6498 spa->spa_spares.sav_sync = B_TRUE;
6501 if (nl2cache != 0) {
6502 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
6503 ZPOOL_CONFIG_L2CACHE);
6504 spa_load_l2cache(spa);
6505 spa->spa_l2cache.sav_sync = B_TRUE;
6509 * We have to be careful when adding new vdevs to an existing pool.
6510 * If other threads start allocating from these vdevs before we
6511 * sync the config cache, and we lose power, then upon reboot we may
6512 * fail to open the pool because there are DVAs that the config cache
6513 * can't translate. Therefore, we first add the vdevs without
6514 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
6515 * and then let spa_config_update() initialize the new metaslabs.
6517 * spa_load() checks for added-but-not-initialized vdevs, so that
6518 * if we lose power at any point in this sequence, the remaining
6519 * steps will be completed the next time we load the pool.
6521 (void) spa_vdev_exit(spa, vd, txg, 0);
6523 mutex_enter(&spa_namespace_lock);
6524 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
6525 spa_event_notify(spa, NULL, NULL, ESC_ZFS_VDEV_ADD);
6526 mutex_exit(&spa_namespace_lock);
6532 * Attach a device to a mirror. The arguments are the path to any device
6533 * in the mirror, and the nvroot for the new device. If the path specifies
6534 * a device that is not mirrored, we automatically insert the mirror vdev.
6536 * If 'replacing' is specified, the new device is intended to replace the
6537 * existing device; in this case the two devices are made into their own
6538 * mirror using the 'replacing' vdev, which is functionally identical to
6539 * the mirror vdev (it actually reuses all the same ops) but has a few
6540 * extra rules: you can't attach to it after it's been created, and upon
6541 * completion of resilvering, the first disk (the one being replaced)
6542 * is automatically detached.
6544 * If 'rebuild' is specified, then sequential reconstruction (a.ka. rebuild)
6545 * should be performed instead of traditional healing reconstruction. From
6546 * an administrators perspective these are both resilver operations.
6549 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing,
6552 uint64_t txg, dtl_max_txg;
6553 vdev_t *rvd = spa->spa_root_vdev;
6554 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
6556 char *oldvdpath, *newvdpath;
6560 ASSERT(spa_writeable(spa));
6562 txg = spa_vdev_enter(spa);
6564 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
6566 ASSERT(MUTEX_HELD(&spa_namespace_lock));
6567 if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) {
6568 error = (spa_has_checkpoint(spa)) ?
6569 ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT;
6570 return (spa_vdev_exit(spa, NULL, txg, error));
6574 if (!spa_feature_is_enabled(spa, SPA_FEATURE_DEVICE_REBUILD))
6575 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
6577 if (dsl_scan_resilvering(spa_get_dsl(spa)))
6578 return (spa_vdev_exit(spa, NULL, txg,
6579 ZFS_ERR_RESILVER_IN_PROGRESS));
6581 if (vdev_rebuild_active(rvd))
6582 return (spa_vdev_exit(spa, NULL, txg,
6583 ZFS_ERR_REBUILD_IN_PROGRESS));
6586 if (spa->spa_vdev_removal != NULL)
6587 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
6590 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
6592 if (!oldvd->vdev_ops->vdev_op_leaf)
6593 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
6595 pvd = oldvd->vdev_parent;
6597 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
6598 VDEV_ALLOC_ATTACH)) != 0)
6599 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
6601 if (newrootvd->vdev_children != 1)
6602 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
6604 newvd = newrootvd->vdev_child[0];
6606 if (!newvd->vdev_ops->vdev_op_leaf)
6607 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
6609 if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
6610 return (spa_vdev_exit(spa, newrootvd, txg, error));
6613 * Spares can't replace logs
6615 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
6616 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
6620 * For rebuilds, the parent vdev must support reconstruction
6621 * using only space maps. This means the only allowable
6622 * parents are the root vdev or a mirror vdev.
6624 if (pvd->vdev_ops != &vdev_mirror_ops &&
6625 pvd->vdev_ops != &vdev_root_ops) {
6626 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
6632 * For attach, the only allowable parent is a mirror or the root
6635 if (pvd->vdev_ops != &vdev_mirror_ops &&
6636 pvd->vdev_ops != &vdev_root_ops)
6637 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
6639 pvops = &vdev_mirror_ops;
6642 * Active hot spares can only be replaced by inactive hot
6645 if (pvd->vdev_ops == &vdev_spare_ops &&
6646 oldvd->vdev_isspare &&
6647 !spa_has_spare(spa, newvd->vdev_guid))
6648 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
6651 * If the source is a hot spare, and the parent isn't already a
6652 * spare, then we want to create a new hot spare. Otherwise, we
6653 * want to create a replacing vdev. The user is not allowed to
6654 * attach to a spared vdev child unless the 'isspare' state is
6655 * the same (spare replaces spare, non-spare replaces
6658 if (pvd->vdev_ops == &vdev_replacing_ops &&
6659 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
6660 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
6661 } else if (pvd->vdev_ops == &vdev_spare_ops &&
6662 newvd->vdev_isspare != oldvd->vdev_isspare) {
6663 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
6666 if (newvd->vdev_isspare)
6667 pvops = &vdev_spare_ops;
6669 pvops = &vdev_replacing_ops;
6673 * Make sure the new device is big enough.
6675 if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
6676 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
6679 * The new device cannot have a higher alignment requirement
6680 * than the top-level vdev.
6682 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
6683 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
6686 * If this is an in-place replacement, update oldvd's path and devid
6687 * to make it distinguishable from newvd, and unopenable from now on.
6689 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
6690 spa_strfree(oldvd->vdev_path);
6691 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
6693 (void) snprintf(oldvd->vdev_path, strlen(newvd->vdev_path) + 5,
6694 "%s/%s", newvd->vdev_path, "old");
6695 if (oldvd->vdev_devid != NULL) {
6696 spa_strfree(oldvd->vdev_devid);
6697 oldvd->vdev_devid = NULL;
6702 * If the parent is not a mirror, or if we're replacing, insert the new
6703 * mirror/replacing/spare vdev above oldvd.
6705 if (pvd->vdev_ops != pvops)
6706 pvd = vdev_add_parent(oldvd, pvops);
6708 ASSERT(pvd->vdev_top->vdev_parent == rvd);
6709 ASSERT(pvd->vdev_ops == pvops);
6710 ASSERT(oldvd->vdev_parent == pvd);
6713 * Extract the new device from its root and add it to pvd.
6715 vdev_remove_child(newrootvd, newvd);
6716 newvd->vdev_id = pvd->vdev_children;
6717 newvd->vdev_crtxg = oldvd->vdev_crtxg;
6718 vdev_add_child(pvd, newvd);
6721 * Reevaluate the parent vdev state.
6723 vdev_propagate_state(pvd);
6725 tvd = newvd->vdev_top;
6726 ASSERT(pvd->vdev_top == tvd);
6727 ASSERT(tvd->vdev_parent == rvd);
6729 vdev_config_dirty(tvd);
6732 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
6733 * for any dmu_sync-ed blocks. It will propagate upward when
6734 * spa_vdev_exit() calls vdev_dtl_reassess().
6736 dtl_max_txg = txg + TXG_CONCURRENT_STATES;
6738 vdev_dtl_dirty(newvd, DTL_MISSING,
6739 TXG_INITIAL, dtl_max_txg - TXG_INITIAL);
6741 if (newvd->vdev_isspare) {
6742 spa_spare_activate(newvd);
6743 spa_event_notify(spa, newvd, NULL, ESC_ZFS_VDEV_SPARE);
6746 oldvdpath = spa_strdup(oldvd->vdev_path);
6747 newvdpath = spa_strdup(newvd->vdev_path);
6748 newvd_isspare = newvd->vdev_isspare;
6751 * Mark newvd's DTL dirty in this txg.
6753 vdev_dirty(tvd, VDD_DTL, newvd, txg);
6756 * Schedule the resilver or rebuild to restart in the future. We do
6757 * this to ensure that dmu_sync-ed blocks have been stitched into the
6758 * respective datasets.
6761 newvd->vdev_rebuild_txg = txg;
6765 newvd->vdev_resilver_txg = txg;
6767 if (dsl_scan_resilvering(spa_get_dsl(spa)) &&
6768 spa_feature_is_enabled(spa, SPA_FEATURE_RESILVER_DEFER)) {
6769 vdev_defer_resilver(newvd);
6771 dsl_scan_restart_resilver(spa->spa_dsl_pool,
6776 if (spa->spa_bootfs)
6777 spa_event_notify(spa, newvd, NULL, ESC_ZFS_BOOTFS_VDEV_ATTACH);
6779 spa_event_notify(spa, newvd, NULL, ESC_ZFS_VDEV_ATTACH);
6784 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
6786 spa_history_log_internal(spa, "vdev attach", NULL,
6787 "%s vdev=%s %s vdev=%s",
6788 replacing && newvd_isspare ? "spare in" :
6789 replacing ? "replace" : "attach", newvdpath,
6790 replacing ? "for" : "to", oldvdpath);
6792 spa_strfree(oldvdpath);
6793 spa_strfree(newvdpath);
6799 * Detach a device from a mirror or replacing vdev.
6801 * If 'replace_done' is specified, only detach if the parent
6802 * is a replacing vdev.
6805 spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
6809 vdev_t *rvd __maybe_unused = spa->spa_root_vdev;
6810 vdev_t *vd, *pvd, *cvd, *tvd;
6811 boolean_t unspare = B_FALSE;
6812 uint64_t unspare_guid = 0;
6815 ASSERT(spa_writeable(spa));
6817 txg = spa_vdev_detach_enter(spa, guid);
6819 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
6822 * Besides being called directly from the userland through the
6823 * ioctl interface, spa_vdev_detach() can be potentially called
6824 * at the end of spa_vdev_resilver_done().
6826 * In the regular case, when we have a checkpoint this shouldn't
6827 * happen as we never empty the DTLs of a vdev during the scrub
6828 * [see comment in dsl_scan_done()]. Thus spa_vdev_resilvering_done()
6829 * should never get here when we have a checkpoint.
6831 * That said, even in a case when we checkpoint the pool exactly
6832 * as spa_vdev_resilver_done() calls this function everything
6833 * should be fine as the resilver will return right away.
6835 ASSERT(MUTEX_HELD(&spa_namespace_lock));
6836 if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) {
6837 error = (spa_has_checkpoint(spa)) ?
6838 ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT;
6839 return (spa_vdev_exit(spa, NULL, txg, error));
6843 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
6845 if (!vd->vdev_ops->vdev_op_leaf)
6846 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
6848 pvd = vd->vdev_parent;
6851 * If the parent/child relationship is not as expected, don't do it.
6852 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
6853 * vdev that's replacing B with C. The user's intent in replacing
6854 * is to go from M(A,B) to M(A,C). If the user decides to cancel
6855 * the replace by detaching C, the expected behavior is to end up
6856 * M(A,B). But suppose that right after deciding to detach C,
6857 * the replacement of B completes. We would have M(A,C), and then
6858 * ask to detach C, which would leave us with just A -- not what
6859 * the user wanted. To prevent this, we make sure that the
6860 * parent/child relationship hasn't changed -- in this example,
6861 * that C's parent is still the replacing vdev R.
6863 if (pvd->vdev_guid != pguid && pguid != 0)
6864 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
6867 * Only 'replacing' or 'spare' vdevs can be replaced.
6869 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
6870 pvd->vdev_ops != &vdev_spare_ops)
6871 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
6873 ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
6874 spa_version(spa) >= SPA_VERSION_SPARES);
6877 * Only mirror, replacing, and spare vdevs support detach.
6879 if (pvd->vdev_ops != &vdev_replacing_ops &&
6880 pvd->vdev_ops != &vdev_mirror_ops &&
6881 pvd->vdev_ops != &vdev_spare_ops)
6882 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
6885 * If this device has the only valid copy of some data,
6886 * we cannot safely detach it.
6888 if (vdev_dtl_required(vd))
6889 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
6891 ASSERT(pvd->vdev_children >= 2);
6894 * If we are detaching the second disk from a replacing vdev, then
6895 * check to see if we changed the original vdev's path to have "/old"
6896 * at the end in spa_vdev_attach(). If so, undo that change now.
6898 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
6899 vd->vdev_path != NULL) {
6900 size_t len = strlen(vd->vdev_path);
6902 for (int c = 0; c < pvd->vdev_children; c++) {
6903 cvd = pvd->vdev_child[c];
6905 if (cvd == vd || cvd->vdev_path == NULL)
6908 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
6909 strcmp(cvd->vdev_path + len, "/old") == 0) {
6910 spa_strfree(cvd->vdev_path);
6911 cvd->vdev_path = spa_strdup(vd->vdev_path);
6918 * If we are detaching the original disk from a spare, then it implies
6919 * that the spare should become a real disk, and be removed from the
6920 * active spare list for the pool.
6922 if (pvd->vdev_ops == &vdev_spare_ops &&
6924 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
6928 * Erase the disk labels so the disk can be used for other things.
6929 * This must be done after all other error cases are handled,
6930 * but before we disembowel vd (so we can still do I/O to it).
6931 * But if we can't do it, don't treat the error as fatal --
6932 * it may be that the unwritability of the disk is the reason
6933 * it's being detached!
6935 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
6938 * Remove vd from its parent and compact the parent's children.
6940 vdev_remove_child(pvd, vd);
6941 vdev_compact_children(pvd);
6944 * Remember one of the remaining children so we can get tvd below.
6946 cvd = pvd->vdev_child[pvd->vdev_children - 1];
6949 * If we need to remove the remaining child from the list of hot spares,
6950 * do it now, marking the vdev as no longer a spare in the process.
6951 * We must do this before vdev_remove_parent(), because that can
6952 * change the GUID if it creates a new toplevel GUID. For a similar
6953 * reason, we must remove the spare now, in the same txg as the detach;
6954 * otherwise someone could attach a new sibling, change the GUID, and
6955 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
6958 ASSERT(cvd->vdev_isspare);
6959 spa_spare_remove(cvd);
6960 unspare_guid = cvd->vdev_guid;
6961 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
6962 cvd->vdev_unspare = B_TRUE;
6966 * If the parent mirror/replacing vdev only has one child,
6967 * the parent is no longer needed. Remove it from the tree.
6969 if (pvd->vdev_children == 1) {
6970 if (pvd->vdev_ops == &vdev_spare_ops)
6971 cvd->vdev_unspare = B_FALSE;
6972 vdev_remove_parent(cvd);
6976 * We don't set tvd until now because the parent we just removed
6977 * may have been the previous top-level vdev.
6979 tvd = cvd->vdev_top;
6980 ASSERT(tvd->vdev_parent == rvd);
6983 * Reevaluate the parent vdev state.
6985 vdev_propagate_state(cvd);
6988 * If the 'autoexpand' property is set on the pool then automatically
6989 * try to expand the size of the pool. For example if the device we
6990 * just detached was smaller than the others, it may be possible to
6991 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
6992 * first so that we can obtain the updated sizes of the leaf vdevs.
6994 if (spa->spa_autoexpand) {
6996 vdev_expand(tvd, txg);
6999 vdev_config_dirty(tvd);
7002 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
7003 * vd->vdev_detached is set and free vd's DTL object in syncing context.
7004 * But first make sure we're not on any *other* txg's DTL list, to
7005 * prevent vd from being accessed after it's freed.
7007 vdpath = spa_strdup(vd->vdev_path ? vd->vdev_path : "none");
7008 for (int t = 0; t < TXG_SIZE; t++)
7009 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
7010 vd->vdev_detached = B_TRUE;
7011 vdev_dirty(tvd, VDD_DTL, vd, txg);
7013 spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_REMOVE);
7014 spa_notify_waiters(spa);
7016 /* hang on to the spa before we release the lock */
7017 spa_open_ref(spa, FTAG);
7019 error = spa_vdev_exit(spa, vd, txg, 0);
7021 spa_history_log_internal(spa, "detach", NULL,
7023 spa_strfree(vdpath);
7026 * If this was the removal of the original device in a hot spare vdev,
7027 * then we want to go through and remove the device from the hot spare
7028 * list of every other pool.
7031 spa_t *altspa = NULL;
7033 mutex_enter(&spa_namespace_lock);
7034 while ((altspa = spa_next(altspa)) != NULL) {
7035 if (altspa->spa_state != POOL_STATE_ACTIVE ||
7039 spa_open_ref(altspa, FTAG);
7040 mutex_exit(&spa_namespace_lock);
7041 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
7042 mutex_enter(&spa_namespace_lock);
7043 spa_close(altspa, FTAG);
7045 mutex_exit(&spa_namespace_lock);
7047 /* search the rest of the vdevs for spares to remove */
7048 spa_vdev_resilver_done(spa);
7051 /* all done with the spa; OK to release */
7052 mutex_enter(&spa_namespace_lock);
7053 spa_close(spa, FTAG);
7054 mutex_exit(&spa_namespace_lock);
7060 spa_vdev_initialize_impl(spa_t *spa, uint64_t guid, uint64_t cmd_type,
7063 ASSERT(MUTEX_HELD(&spa_namespace_lock));
7065 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
7067 /* Look up vdev and ensure it's a leaf. */
7068 vdev_t *vd = spa_lookup_by_guid(spa, guid, B_FALSE);
7069 if (vd == NULL || vd->vdev_detached) {
7070 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
7071 return (SET_ERROR(ENODEV));
7072 } else if (!vd->vdev_ops->vdev_op_leaf || !vdev_is_concrete(vd)) {
7073 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
7074 return (SET_ERROR(EINVAL));
7075 } else if (!vdev_writeable(vd)) {
7076 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
7077 return (SET_ERROR(EROFS));
7079 mutex_enter(&vd->vdev_initialize_lock);
7080 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
7083 * When we activate an initialize action we check to see
7084 * if the vdev_initialize_thread is NULL. We do this instead
7085 * of using the vdev_initialize_state since there might be
7086 * a previous initialization process which has completed but
7087 * the thread is not exited.
7089 if (cmd_type == POOL_INITIALIZE_START &&
7090 (vd->vdev_initialize_thread != NULL ||
7091 vd->vdev_top->vdev_removing)) {
7092 mutex_exit(&vd->vdev_initialize_lock);
7093 return (SET_ERROR(EBUSY));
7094 } else if (cmd_type == POOL_INITIALIZE_CANCEL &&
7095 (vd->vdev_initialize_state != VDEV_INITIALIZE_ACTIVE &&
7096 vd->vdev_initialize_state != VDEV_INITIALIZE_SUSPENDED)) {
7097 mutex_exit(&vd->vdev_initialize_lock);
7098 return (SET_ERROR(ESRCH));
7099 } else if (cmd_type == POOL_INITIALIZE_SUSPEND &&
7100 vd->vdev_initialize_state != VDEV_INITIALIZE_ACTIVE) {
7101 mutex_exit(&vd->vdev_initialize_lock);
7102 return (SET_ERROR(ESRCH));
7106 case POOL_INITIALIZE_START:
7107 vdev_initialize(vd);
7109 case POOL_INITIALIZE_CANCEL:
7110 vdev_initialize_stop(vd, VDEV_INITIALIZE_CANCELED, vd_list);
7112 case POOL_INITIALIZE_SUSPEND:
7113 vdev_initialize_stop(vd, VDEV_INITIALIZE_SUSPENDED, vd_list);
7116 panic("invalid cmd_type %llu", (unsigned long long)cmd_type);
7118 mutex_exit(&vd->vdev_initialize_lock);
7124 spa_vdev_initialize(spa_t *spa, nvlist_t *nv, uint64_t cmd_type,
7125 nvlist_t *vdev_errlist)
7127 int total_errors = 0;
7130 list_create(&vd_list, sizeof (vdev_t),
7131 offsetof(vdev_t, vdev_initialize_node));
7134 * We hold the namespace lock through the whole function
7135 * to prevent any changes to the pool while we're starting or
7136 * stopping initialization. The config and state locks are held so that
7137 * we can properly assess the vdev state before we commit to
7138 * the initializing operation.
7140 mutex_enter(&spa_namespace_lock);
7142 for (nvpair_t *pair = nvlist_next_nvpair(nv, NULL);
7143 pair != NULL; pair = nvlist_next_nvpair(nv, pair)) {
7144 uint64_t vdev_guid = fnvpair_value_uint64(pair);
7146 int error = spa_vdev_initialize_impl(spa, vdev_guid, cmd_type,
7149 char guid_as_str[MAXNAMELEN];
7151 (void) snprintf(guid_as_str, sizeof (guid_as_str),
7152 "%llu", (unsigned long long)vdev_guid);
7153 fnvlist_add_int64(vdev_errlist, guid_as_str, error);
7158 /* Wait for all initialize threads to stop. */
7159 vdev_initialize_stop_wait(spa, &vd_list);
7161 /* Sync out the initializing state */
7162 txg_wait_synced(spa->spa_dsl_pool, 0);
7163 mutex_exit(&spa_namespace_lock);
7165 list_destroy(&vd_list);
7167 return (total_errors);
7171 spa_vdev_trim_impl(spa_t *spa, uint64_t guid, uint64_t cmd_type,
7172 uint64_t rate, boolean_t partial, boolean_t secure, list_t *vd_list)
7174 ASSERT(MUTEX_HELD(&spa_namespace_lock));
7176 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
7178 /* Look up vdev and ensure it's a leaf. */
7179 vdev_t *vd = spa_lookup_by_guid(spa, guid, B_FALSE);
7180 if (vd == NULL || vd->vdev_detached) {
7181 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
7182 return (SET_ERROR(ENODEV));
7183 } else if (!vd->vdev_ops->vdev_op_leaf || !vdev_is_concrete(vd)) {
7184 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
7185 return (SET_ERROR(EINVAL));
7186 } else if (!vdev_writeable(vd)) {
7187 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
7188 return (SET_ERROR(EROFS));
7189 } else if (!vd->vdev_has_trim) {
7190 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
7191 return (SET_ERROR(EOPNOTSUPP));
7192 } else if (secure && !vd->vdev_has_securetrim) {
7193 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
7194 return (SET_ERROR(EOPNOTSUPP));
7196 mutex_enter(&vd->vdev_trim_lock);
7197 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
7200 * When we activate a TRIM action we check to see if the
7201 * vdev_trim_thread is NULL. We do this instead of using the
7202 * vdev_trim_state since there might be a previous TRIM process
7203 * which has completed but the thread is not exited.
7205 if (cmd_type == POOL_TRIM_START &&
7206 (vd->vdev_trim_thread != NULL || vd->vdev_top->vdev_removing)) {
7207 mutex_exit(&vd->vdev_trim_lock);
7208 return (SET_ERROR(EBUSY));
7209 } else if (cmd_type == POOL_TRIM_CANCEL &&
7210 (vd->vdev_trim_state != VDEV_TRIM_ACTIVE &&
7211 vd->vdev_trim_state != VDEV_TRIM_SUSPENDED)) {
7212 mutex_exit(&vd->vdev_trim_lock);
7213 return (SET_ERROR(ESRCH));
7214 } else if (cmd_type == POOL_TRIM_SUSPEND &&
7215 vd->vdev_trim_state != VDEV_TRIM_ACTIVE) {
7216 mutex_exit(&vd->vdev_trim_lock);
7217 return (SET_ERROR(ESRCH));
7221 case POOL_TRIM_START:
7222 vdev_trim(vd, rate, partial, secure);
7224 case POOL_TRIM_CANCEL:
7225 vdev_trim_stop(vd, VDEV_TRIM_CANCELED, vd_list);
7227 case POOL_TRIM_SUSPEND:
7228 vdev_trim_stop(vd, VDEV_TRIM_SUSPENDED, vd_list);
7231 panic("invalid cmd_type %llu", (unsigned long long)cmd_type);
7233 mutex_exit(&vd->vdev_trim_lock);
7239 * Initiates a manual TRIM for the requested vdevs. This kicks off individual
7240 * TRIM threads for each child vdev. These threads pass over all of the free
7241 * space in the vdev's metaslabs and issues TRIM commands for that space.
7244 spa_vdev_trim(spa_t *spa, nvlist_t *nv, uint64_t cmd_type, uint64_t rate,
7245 boolean_t partial, boolean_t secure, nvlist_t *vdev_errlist)
7247 int total_errors = 0;
7250 list_create(&vd_list, sizeof (vdev_t),
7251 offsetof(vdev_t, vdev_trim_node));
7254 * We hold the namespace lock through the whole function
7255 * to prevent any changes to the pool while we're starting or
7256 * stopping TRIM. The config and state locks are held so that
7257 * we can properly assess the vdev state before we commit to
7258 * the TRIM operation.
7260 mutex_enter(&spa_namespace_lock);
7262 for (nvpair_t *pair = nvlist_next_nvpair(nv, NULL);
7263 pair != NULL; pair = nvlist_next_nvpair(nv, pair)) {
7264 uint64_t vdev_guid = fnvpair_value_uint64(pair);
7266 int error = spa_vdev_trim_impl(spa, vdev_guid, cmd_type,
7267 rate, partial, secure, &vd_list);
7269 char guid_as_str[MAXNAMELEN];
7271 (void) snprintf(guid_as_str, sizeof (guid_as_str),
7272 "%llu", (unsigned long long)vdev_guid);
7273 fnvlist_add_int64(vdev_errlist, guid_as_str, error);
7278 /* Wait for all TRIM threads to stop. */
7279 vdev_trim_stop_wait(spa, &vd_list);
7281 /* Sync out the TRIM state */
7282 txg_wait_synced(spa->spa_dsl_pool, 0);
7283 mutex_exit(&spa_namespace_lock);
7285 list_destroy(&vd_list);
7287 return (total_errors);
7291 * Split a set of devices from their mirrors, and create a new pool from them.
7294 spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
7295 nvlist_t *props, boolean_t exp)
7298 uint64_t txg, *glist;
7300 uint_t c, children, lastlog;
7301 nvlist_t **child, *nvl, *tmp;
7303 char *altroot = NULL;
7304 vdev_t *rvd, **vml = NULL; /* vdev modify list */
7305 boolean_t activate_slog;
7307 ASSERT(spa_writeable(spa));
7309 txg = spa_vdev_enter(spa);
7311 ASSERT(MUTEX_HELD(&spa_namespace_lock));
7312 if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) {
7313 error = (spa_has_checkpoint(spa)) ?
7314 ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT;
7315 return (spa_vdev_exit(spa, NULL, txg, error));
7318 /* clear the log and flush everything up to now */
7319 activate_slog = spa_passivate_log(spa);
7320 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
7321 error = spa_reset_logs(spa);
7322 txg = spa_vdev_config_enter(spa);
7325 spa_activate_log(spa);
7328 return (spa_vdev_exit(spa, NULL, txg, error));
7330 /* check new spa name before going any further */
7331 if (spa_lookup(newname) != NULL)
7332 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
7335 * scan through all the children to ensure they're all mirrors
7337 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
7338 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
7340 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
7342 /* first, check to ensure we've got the right child count */
7343 rvd = spa->spa_root_vdev;
7345 for (c = 0; c < rvd->vdev_children; c++) {
7346 vdev_t *vd = rvd->vdev_child[c];
7348 /* don't count the holes & logs as children */
7349 if (vd->vdev_islog || (vd->vdev_ops != &vdev_indirect_ops &&
7350 !vdev_is_concrete(vd))) {
7358 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
7359 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
7361 /* next, ensure no spare or cache devices are part of the split */
7362 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
7363 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
7364 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
7366 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
7367 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
7369 /* then, loop over each vdev and validate it */
7370 for (c = 0; c < children; c++) {
7371 uint64_t is_hole = 0;
7373 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
7377 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
7378 spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
7381 error = SET_ERROR(EINVAL);
7386 /* deal with indirect vdevs */
7387 if (spa->spa_root_vdev->vdev_child[c]->vdev_ops ==
7391 /* which disk is going to be split? */
7392 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
7394 error = SET_ERROR(EINVAL);
7398 /* look it up in the spa */
7399 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
7400 if (vml[c] == NULL) {
7401 error = SET_ERROR(ENODEV);
7405 /* make sure there's nothing stopping the split */
7406 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
7407 vml[c]->vdev_islog ||
7408 !vdev_is_concrete(vml[c]) ||
7409 vml[c]->vdev_isspare ||
7410 vml[c]->vdev_isl2cache ||
7411 !vdev_writeable(vml[c]) ||
7412 vml[c]->vdev_children != 0 ||
7413 vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
7414 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
7415 error = SET_ERROR(EINVAL);
7419 if (vdev_dtl_required(vml[c]) ||
7420 vdev_resilver_needed(vml[c], NULL, NULL)) {
7421 error = SET_ERROR(EBUSY);
7425 /* we need certain info from the top level */
7426 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
7427 vml[c]->vdev_top->vdev_ms_array) == 0);
7428 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
7429 vml[c]->vdev_top->vdev_ms_shift) == 0);
7430 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
7431 vml[c]->vdev_top->vdev_asize) == 0);
7432 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
7433 vml[c]->vdev_top->vdev_ashift) == 0);
7435 /* transfer per-vdev ZAPs */
7436 ASSERT3U(vml[c]->vdev_leaf_zap, !=, 0);
7437 VERIFY0(nvlist_add_uint64(child[c],
7438 ZPOOL_CONFIG_VDEV_LEAF_ZAP, vml[c]->vdev_leaf_zap));
7440 ASSERT3U(vml[c]->vdev_top->vdev_top_zap, !=, 0);
7441 VERIFY0(nvlist_add_uint64(child[c],
7442 ZPOOL_CONFIG_VDEV_TOP_ZAP,
7443 vml[c]->vdev_parent->vdev_top_zap));
7447 kmem_free(vml, children * sizeof (vdev_t *));
7448 kmem_free(glist, children * sizeof (uint64_t));
7449 return (spa_vdev_exit(spa, NULL, txg, error));
7452 /* stop writers from using the disks */
7453 for (c = 0; c < children; c++) {
7455 vml[c]->vdev_offline = B_TRUE;
7457 vdev_reopen(spa->spa_root_vdev);
7460 * Temporarily record the splitting vdevs in the spa config. This
7461 * will disappear once the config is regenerated.
7463 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
7464 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
7465 glist, children) == 0);
7466 kmem_free(glist, children * sizeof (uint64_t));
7468 mutex_enter(&spa->spa_props_lock);
7469 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
7471 mutex_exit(&spa->spa_props_lock);
7472 spa->spa_config_splitting = nvl;
7473 vdev_config_dirty(spa->spa_root_vdev);
7475 /* configure and create the new pool */
7476 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
7477 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
7478 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
7479 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
7480 spa_version(spa)) == 0);
7481 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
7482 spa->spa_config_txg) == 0);
7483 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
7484 spa_generate_guid(NULL)) == 0);
7485 VERIFY0(nvlist_add_boolean(config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS));
7486 (void) nvlist_lookup_string(props,
7487 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
7489 /* add the new pool to the namespace */
7490 newspa = spa_add(newname, config, altroot);
7491 newspa->spa_avz_action = AVZ_ACTION_REBUILD;
7492 newspa->spa_config_txg = spa->spa_config_txg;
7493 spa_set_log_state(newspa, SPA_LOG_CLEAR);
7495 /* release the spa config lock, retaining the namespace lock */
7496 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
7498 if (zio_injection_enabled)
7499 zio_handle_panic_injection(spa, FTAG, 1);
7501 spa_activate(newspa, spa_mode_global);
7502 spa_async_suspend(newspa);
7505 * Temporarily stop the initializing and TRIM activity. We set the
7506 * state to ACTIVE so that we know to resume initializing or TRIM
7507 * once the split has completed.
7509 list_t vd_initialize_list;
7510 list_create(&vd_initialize_list, sizeof (vdev_t),
7511 offsetof(vdev_t, vdev_initialize_node));
7513 list_t vd_trim_list;
7514 list_create(&vd_trim_list, sizeof (vdev_t),
7515 offsetof(vdev_t, vdev_trim_node));
7517 for (c = 0; c < children; c++) {
7518 if (vml[c] != NULL && vml[c]->vdev_ops != &vdev_indirect_ops) {
7519 mutex_enter(&vml[c]->vdev_initialize_lock);
7520 vdev_initialize_stop(vml[c],
7521 VDEV_INITIALIZE_ACTIVE, &vd_initialize_list);
7522 mutex_exit(&vml[c]->vdev_initialize_lock);
7524 mutex_enter(&vml[c]->vdev_trim_lock);
7525 vdev_trim_stop(vml[c], VDEV_TRIM_ACTIVE, &vd_trim_list);
7526 mutex_exit(&vml[c]->vdev_trim_lock);
7530 vdev_initialize_stop_wait(spa, &vd_initialize_list);
7531 vdev_trim_stop_wait(spa, &vd_trim_list);
7533 list_destroy(&vd_initialize_list);
7534 list_destroy(&vd_trim_list);
7536 newspa->spa_config_source = SPA_CONFIG_SRC_SPLIT;
7537 newspa->spa_is_splitting = B_TRUE;
7539 /* create the new pool from the disks of the original pool */
7540 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE);
7544 /* if that worked, generate a real config for the new pool */
7545 if (newspa->spa_root_vdev != NULL) {
7546 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
7547 NV_UNIQUE_NAME, KM_SLEEP) == 0);
7548 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
7549 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
7550 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
7555 if (props != NULL) {
7556 spa_configfile_set(newspa, props, B_FALSE);
7557 error = spa_prop_set(newspa, props);
7562 /* flush everything */
7563 txg = spa_vdev_config_enter(newspa);
7564 vdev_config_dirty(newspa->spa_root_vdev);
7565 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
7567 if (zio_injection_enabled)
7568 zio_handle_panic_injection(spa, FTAG, 2);
7570 spa_async_resume(newspa);
7572 /* finally, update the original pool's config */
7573 txg = spa_vdev_config_enter(spa);
7574 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
7575 error = dmu_tx_assign(tx, TXG_WAIT);
7578 for (c = 0; c < children; c++) {
7579 if (vml[c] != NULL && vml[c]->vdev_ops != &vdev_indirect_ops) {
7580 vdev_t *tvd = vml[c]->vdev_top;
7583 * Need to be sure the detachable VDEV is not
7584 * on any *other* txg's DTL list to prevent it
7585 * from being accessed after it's freed.
7587 for (int t = 0; t < TXG_SIZE; t++) {
7588 (void) txg_list_remove_this(
7589 &tvd->vdev_dtl_list, vml[c], t);
7594 spa_history_log_internal(spa, "detach", tx,
7595 "vdev=%s", vml[c]->vdev_path);
7600 spa->spa_avz_action = AVZ_ACTION_REBUILD;
7601 vdev_config_dirty(spa->spa_root_vdev);
7602 spa->spa_config_splitting = NULL;
7606 (void) spa_vdev_exit(spa, NULL, txg, 0);
7608 if (zio_injection_enabled)
7609 zio_handle_panic_injection(spa, FTAG, 3);
7611 /* split is complete; log a history record */
7612 spa_history_log_internal(newspa, "split", NULL,
7613 "from pool %s", spa_name(spa));
7615 newspa->spa_is_splitting = B_FALSE;
7616 kmem_free(vml, children * sizeof (vdev_t *));
7618 /* if we're not going to mount the filesystems in userland, export */
7620 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
7627 spa_deactivate(newspa);
7630 txg = spa_vdev_config_enter(spa);
7632 /* re-online all offlined disks */
7633 for (c = 0; c < children; c++) {
7635 vml[c]->vdev_offline = B_FALSE;
7638 /* restart initializing or trimming disks as necessary */
7639 spa_async_request(spa, SPA_ASYNC_INITIALIZE_RESTART);
7640 spa_async_request(spa, SPA_ASYNC_TRIM_RESTART);
7641 spa_async_request(spa, SPA_ASYNC_AUTOTRIM_RESTART);
7643 vdev_reopen(spa->spa_root_vdev);
7645 nvlist_free(spa->spa_config_splitting);
7646 spa->spa_config_splitting = NULL;
7647 (void) spa_vdev_exit(spa, NULL, txg, error);
7649 kmem_free(vml, children * sizeof (vdev_t *));
7654 * Find any device that's done replacing, or a vdev marked 'unspare' that's
7655 * currently spared, so we can detach it.
7658 spa_vdev_resilver_done_hunt(vdev_t *vd)
7660 vdev_t *newvd, *oldvd;
7662 for (int c = 0; c < vd->vdev_children; c++) {
7663 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
7669 * Check for a completed replacement. We always consider the first
7670 * vdev in the list to be the oldest vdev, and the last one to be
7671 * the newest (see spa_vdev_attach() for how that works). In
7672 * the case where the newest vdev is faulted, we will not automatically
7673 * remove it after a resilver completes. This is OK as it will require
7674 * user intervention to determine which disk the admin wishes to keep.
7676 if (vd->vdev_ops == &vdev_replacing_ops) {
7677 ASSERT(vd->vdev_children > 1);
7679 newvd = vd->vdev_child[vd->vdev_children - 1];
7680 oldvd = vd->vdev_child[0];
7682 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
7683 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
7684 !vdev_dtl_required(oldvd))
7689 * Check for a completed resilver with the 'unspare' flag set.
7690 * Also potentially update faulted state.
7692 if (vd->vdev_ops == &vdev_spare_ops) {
7693 vdev_t *first = vd->vdev_child[0];
7694 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
7696 if (last->vdev_unspare) {
7699 } else if (first->vdev_unspare) {
7706 if (oldvd != NULL &&
7707 vdev_dtl_empty(newvd, DTL_MISSING) &&
7708 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
7709 !vdev_dtl_required(oldvd))
7712 vdev_propagate_state(vd);
7715 * If there are more than two spares attached to a disk,
7716 * and those spares are not required, then we want to
7717 * attempt to free them up now so that they can be used
7718 * by other pools. Once we're back down to a single
7719 * disk+spare, we stop removing them.
7721 if (vd->vdev_children > 2) {
7722 newvd = vd->vdev_child[1];
7724 if (newvd->vdev_isspare && last->vdev_isspare &&
7725 vdev_dtl_empty(last, DTL_MISSING) &&
7726 vdev_dtl_empty(last, DTL_OUTAGE) &&
7727 !vdev_dtl_required(newvd))
7736 spa_vdev_resilver_done(spa_t *spa)
7738 vdev_t *vd, *pvd, *ppvd;
7739 uint64_t guid, sguid, pguid, ppguid;
7741 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
7743 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
7744 pvd = vd->vdev_parent;
7745 ppvd = pvd->vdev_parent;
7746 guid = vd->vdev_guid;
7747 pguid = pvd->vdev_guid;
7748 ppguid = ppvd->vdev_guid;
7751 * If we have just finished replacing a hot spared device, then
7752 * we need to detach the parent's first child (the original hot
7755 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
7756 ppvd->vdev_children == 2) {
7757 ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
7758 sguid = ppvd->vdev_child[1]->vdev_guid;
7760 ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd));
7762 spa_config_exit(spa, SCL_ALL, FTAG);
7763 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
7765 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
7767 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
7770 spa_config_exit(spa, SCL_ALL, FTAG);
7773 * If a detach was not performed above replace waiters will not have
7774 * been notified. In which case we must do so now.
7776 spa_notify_waiters(spa);
7780 * Update the stored path or FRU for this vdev.
7783 spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
7787 boolean_t sync = B_FALSE;
7789 ASSERT(spa_writeable(spa));
7791 spa_vdev_state_enter(spa, SCL_ALL);
7793 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
7794 return (spa_vdev_state_exit(spa, NULL, ENOENT));
7796 if (!vd->vdev_ops->vdev_op_leaf)
7797 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
7800 if (strcmp(value, vd->vdev_path) != 0) {
7801 spa_strfree(vd->vdev_path);
7802 vd->vdev_path = spa_strdup(value);
7806 if (vd->vdev_fru == NULL) {
7807 vd->vdev_fru = spa_strdup(value);
7809 } else if (strcmp(value, vd->vdev_fru) != 0) {
7810 spa_strfree(vd->vdev_fru);
7811 vd->vdev_fru = spa_strdup(value);
7816 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
7820 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
7822 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
7826 spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
7828 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
7832 * ==========================================================================
7834 * ==========================================================================
7837 spa_scrub_pause_resume(spa_t *spa, pool_scrub_cmd_t cmd)
7839 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
7841 if (dsl_scan_resilvering(spa->spa_dsl_pool))
7842 return (SET_ERROR(EBUSY));
7844 return (dsl_scrub_set_pause_resume(spa->spa_dsl_pool, cmd));
7848 spa_scan_stop(spa_t *spa)
7850 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
7851 if (dsl_scan_resilvering(spa->spa_dsl_pool))
7852 return (SET_ERROR(EBUSY));
7853 return (dsl_scan_cancel(spa->spa_dsl_pool));
7857 spa_scan(spa_t *spa, pool_scan_func_t func)
7859 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
7861 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
7862 return (SET_ERROR(ENOTSUP));
7864 if (func == POOL_SCAN_RESILVER &&
7865 !spa_feature_is_enabled(spa, SPA_FEATURE_RESILVER_DEFER))
7866 return (SET_ERROR(ENOTSUP));
7869 * If a resilver was requested, but there is no DTL on a
7870 * writeable leaf device, we have nothing to do.
7872 if (func == POOL_SCAN_RESILVER &&
7873 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
7874 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
7878 return (dsl_scan(spa->spa_dsl_pool, func));
7882 * ==========================================================================
7883 * SPA async task processing
7884 * ==========================================================================
7888 spa_async_remove(spa_t *spa, vdev_t *vd)
7890 if (vd->vdev_remove_wanted) {
7891 vd->vdev_remove_wanted = B_FALSE;
7892 vd->vdev_delayed_close = B_FALSE;
7893 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
7896 * We want to clear the stats, but we don't want to do a full
7897 * vdev_clear() as that will cause us to throw away
7898 * degraded/faulted state as well as attempt to reopen the
7899 * device, all of which is a waste.
7901 vd->vdev_stat.vs_read_errors = 0;
7902 vd->vdev_stat.vs_write_errors = 0;
7903 vd->vdev_stat.vs_checksum_errors = 0;
7905 vdev_state_dirty(vd->vdev_top);
7908 for (int c = 0; c < vd->vdev_children; c++)
7909 spa_async_remove(spa, vd->vdev_child[c]);
7913 spa_async_probe(spa_t *spa, vdev_t *vd)
7915 if (vd->vdev_probe_wanted) {
7916 vd->vdev_probe_wanted = B_FALSE;
7917 vdev_reopen(vd); /* vdev_open() does the actual probe */
7920 for (int c = 0; c < vd->vdev_children; c++)
7921 spa_async_probe(spa, vd->vdev_child[c]);
7925 spa_async_autoexpand(spa_t *spa, vdev_t *vd)
7927 if (!spa->spa_autoexpand)
7930 for (int c = 0; c < vd->vdev_children; c++) {
7931 vdev_t *cvd = vd->vdev_child[c];
7932 spa_async_autoexpand(spa, cvd);
7935 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
7938 spa_event_notify(vd->vdev_spa, vd, NULL, ESC_ZFS_VDEV_AUTOEXPAND);
7942 spa_async_thread(void *arg)
7944 spa_t *spa = (spa_t *)arg;
7945 dsl_pool_t *dp = spa->spa_dsl_pool;
7948 ASSERT(spa->spa_sync_on);
7950 mutex_enter(&spa->spa_async_lock);
7951 tasks = spa->spa_async_tasks;
7952 spa->spa_async_tasks = 0;
7953 mutex_exit(&spa->spa_async_lock);
7956 * See if the config needs to be updated.
7958 if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
7959 uint64_t old_space, new_space;
7961 mutex_enter(&spa_namespace_lock);
7962 old_space = metaslab_class_get_space(spa_normal_class(spa));
7963 old_space += metaslab_class_get_space(spa_special_class(spa));
7964 old_space += metaslab_class_get_space(spa_dedup_class(spa));
7966 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
7968 new_space = metaslab_class_get_space(spa_normal_class(spa));
7969 new_space += metaslab_class_get_space(spa_special_class(spa));
7970 new_space += metaslab_class_get_space(spa_dedup_class(spa));
7971 mutex_exit(&spa_namespace_lock);
7974 * If the pool grew as a result of the config update,
7975 * then log an internal history event.
7977 if (new_space != old_space) {
7978 spa_history_log_internal(spa, "vdev online", NULL,
7979 "pool '%s' size: %llu(+%llu)",
7980 spa_name(spa), (u_longlong_t)new_space,
7981 (u_longlong_t)(new_space - old_space));
7986 * See if any devices need to be marked REMOVED.
7988 if (tasks & SPA_ASYNC_REMOVE) {
7989 spa_vdev_state_enter(spa, SCL_NONE);
7990 spa_async_remove(spa, spa->spa_root_vdev);
7991 for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
7992 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
7993 for (int i = 0; i < spa->spa_spares.sav_count; i++)
7994 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
7995 (void) spa_vdev_state_exit(spa, NULL, 0);
7998 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
7999 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
8000 spa_async_autoexpand(spa, spa->spa_root_vdev);
8001 spa_config_exit(spa, SCL_CONFIG, FTAG);
8005 * See if any devices need to be probed.
8007 if (tasks & SPA_ASYNC_PROBE) {
8008 spa_vdev_state_enter(spa, SCL_NONE);
8009 spa_async_probe(spa, spa->spa_root_vdev);
8010 (void) spa_vdev_state_exit(spa, NULL, 0);
8014 * If any devices are done replacing, detach them.
8016 if (tasks & SPA_ASYNC_RESILVER_DONE)
8017 spa_vdev_resilver_done(spa);
8020 * If any devices are done replacing, detach them. Then if no
8021 * top-level vdevs are rebuilding attempt to kick off a scrub.
8023 if (tasks & SPA_ASYNC_REBUILD_DONE) {
8024 spa_vdev_resilver_done(spa);
8026 if (!vdev_rebuild_active(spa->spa_root_vdev))
8027 (void) dsl_scan(spa->spa_dsl_pool, POOL_SCAN_SCRUB);
8031 * Kick off a resilver.
8033 if (tasks & SPA_ASYNC_RESILVER &&
8034 !vdev_rebuild_active(spa->spa_root_vdev) &&
8035 (!dsl_scan_resilvering(dp) ||
8036 !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_RESILVER_DEFER)))
8037 dsl_scan_restart_resilver(dp, 0);
8039 if (tasks & SPA_ASYNC_INITIALIZE_RESTART) {
8040 mutex_enter(&spa_namespace_lock);
8041 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
8042 vdev_initialize_restart(spa->spa_root_vdev);
8043 spa_config_exit(spa, SCL_CONFIG, FTAG);
8044 mutex_exit(&spa_namespace_lock);
8047 if (tasks & SPA_ASYNC_TRIM_RESTART) {
8048 mutex_enter(&spa_namespace_lock);
8049 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
8050 vdev_trim_restart(spa->spa_root_vdev);
8051 spa_config_exit(spa, SCL_CONFIG, FTAG);
8052 mutex_exit(&spa_namespace_lock);
8055 if (tasks & SPA_ASYNC_AUTOTRIM_RESTART) {
8056 mutex_enter(&spa_namespace_lock);
8057 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
8058 vdev_autotrim_restart(spa);
8059 spa_config_exit(spa, SCL_CONFIG, FTAG);
8060 mutex_exit(&spa_namespace_lock);
8064 * Kick off L2 cache whole device TRIM.
8066 if (tasks & SPA_ASYNC_L2CACHE_TRIM) {
8067 mutex_enter(&spa_namespace_lock);
8068 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
8069 vdev_trim_l2arc(spa);
8070 spa_config_exit(spa, SCL_CONFIG, FTAG);
8071 mutex_exit(&spa_namespace_lock);
8075 * Kick off L2 cache rebuilding.
8077 if (tasks & SPA_ASYNC_L2CACHE_REBUILD) {
8078 mutex_enter(&spa_namespace_lock);
8079 spa_config_enter(spa, SCL_L2ARC, FTAG, RW_READER);
8080 l2arc_spa_rebuild_start(spa);
8081 spa_config_exit(spa, SCL_L2ARC, FTAG);
8082 mutex_exit(&spa_namespace_lock);
8086 * Let the world know that we're done.
8088 mutex_enter(&spa->spa_async_lock);
8089 spa->spa_async_thread = NULL;
8090 cv_broadcast(&spa->spa_async_cv);
8091 mutex_exit(&spa->spa_async_lock);
8096 spa_async_suspend(spa_t *spa)
8098 mutex_enter(&spa->spa_async_lock);
8099 spa->spa_async_suspended++;
8100 while (spa->spa_async_thread != NULL)
8101 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
8102 mutex_exit(&spa->spa_async_lock);
8104 spa_vdev_remove_suspend(spa);
8106 zthr_t *condense_thread = spa->spa_condense_zthr;
8107 if (condense_thread != NULL)
8108 zthr_cancel(condense_thread);
8110 zthr_t *discard_thread = spa->spa_checkpoint_discard_zthr;
8111 if (discard_thread != NULL)
8112 zthr_cancel(discard_thread);
8114 zthr_t *ll_delete_thread = spa->spa_livelist_delete_zthr;
8115 if (ll_delete_thread != NULL)
8116 zthr_cancel(ll_delete_thread);
8118 zthr_t *ll_condense_thread = spa->spa_livelist_condense_zthr;
8119 if (ll_condense_thread != NULL)
8120 zthr_cancel(ll_condense_thread);
8124 spa_async_resume(spa_t *spa)
8126 mutex_enter(&spa->spa_async_lock);
8127 ASSERT(spa->spa_async_suspended != 0);
8128 spa->spa_async_suspended--;
8129 mutex_exit(&spa->spa_async_lock);
8130 spa_restart_removal(spa);
8132 zthr_t *condense_thread = spa->spa_condense_zthr;
8133 if (condense_thread != NULL)
8134 zthr_resume(condense_thread);
8136 zthr_t *discard_thread = spa->spa_checkpoint_discard_zthr;
8137 if (discard_thread != NULL)
8138 zthr_resume(discard_thread);
8140 zthr_t *ll_delete_thread = spa->spa_livelist_delete_zthr;
8141 if (ll_delete_thread != NULL)
8142 zthr_resume(ll_delete_thread);
8144 zthr_t *ll_condense_thread = spa->spa_livelist_condense_zthr;
8145 if (ll_condense_thread != NULL)
8146 zthr_resume(ll_condense_thread);
8150 spa_async_tasks_pending(spa_t *spa)
8152 uint_t non_config_tasks;
8154 boolean_t config_task_suspended;
8156 non_config_tasks = spa->spa_async_tasks & ~SPA_ASYNC_CONFIG_UPDATE;
8157 config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE;
8158 if (spa->spa_ccw_fail_time == 0) {
8159 config_task_suspended = B_FALSE;
8161 config_task_suspended =
8162 (gethrtime() - spa->spa_ccw_fail_time) <
8163 ((hrtime_t)zfs_ccw_retry_interval * NANOSEC);
8166 return (non_config_tasks || (config_task && !config_task_suspended));
8170 spa_async_dispatch(spa_t *spa)
8172 mutex_enter(&spa->spa_async_lock);
8173 if (spa_async_tasks_pending(spa) &&
8174 !spa->spa_async_suspended &&
8175 spa->spa_async_thread == NULL)
8176 spa->spa_async_thread = thread_create(NULL, 0,
8177 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
8178 mutex_exit(&spa->spa_async_lock);
8182 spa_async_request(spa_t *spa, int task)
8184 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
8185 mutex_enter(&spa->spa_async_lock);
8186 spa->spa_async_tasks |= task;
8187 mutex_exit(&spa->spa_async_lock);
8191 spa_async_tasks(spa_t *spa)
8193 return (spa->spa_async_tasks);
8197 * ==========================================================================
8198 * SPA syncing routines
8199 * ==========================================================================
8204 bpobj_enqueue_cb(void *arg, const blkptr_t *bp, boolean_t bp_freed,
8208 bpobj_enqueue(bpo, bp, bp_freed, tx);
8213 bpobj_enqueue_alloc_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
8215 return (bpobj_enqueue_cb(arg, bp, B_FALSE, tx));
8219 bpobj_enqueue_free_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
8221 return (bpobj_enqueue_cb(arg, bp, B_TRUE, tx));
8225 spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
8229 zio_nowait(zio_free_sync(pio, pio->io_spa, dmu_tx_get_txg(tx), bp,
8235 bpobj_spa_free_sync_cb(void *arg, const blkptr_t *bp, boolean_t bp_freed,
8239 return (spa_free_sync_cb(arg, bp, tx));
8243 * Note: this simple function is not inlined to make it easier to dtrace the
8244 * amount of time spent syncing frees.
8247 spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx)
8249 zio_t *zio = zio_root(spa, NULL, NULL, 0);
8250 bplist_iterate(bpl, spa_free_sync_cb, zio, tx);
8251 VERIFY(zio_wait(zio) == 0);
8255 * Note: this simple function is not inlined to make it easier to dtrace the
8256 * amount of time spent syncing deferred frees.
8259 spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx)
8261 if (spa_sync_pass(spa) != 1)
8266 * If the log space map feature is active, we stop deferring
8267 * frees to the next TXG and therefore running this function
8268 * would be considered a no-op as spa_deferred_bpobj should
8269 * not have any entries.
8271 * That said we run this function anyway (instead of returning
8272 * immediately) for the edge-case scenario where we just
8273 * activated the log space map feature in this TXG but we have
8274 * deferred frees from the previous TXG.
8276 zio_t *zio = zio_root(spa, NULL, NULL, 0);
8277 VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj,
8278 bpobj_spa_free_sync_cb, zio, tx), ==, 0);
8279 VERIFY0(zio_wait(zio));
8283 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
8285 char *packed = NULL;
8290 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
8293 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
8294 * information. This avoids the dmu_buf_will_dirty() path and
8295 * saves us a pre-read to get data we don't actually care about.
8297 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
8298 packed = vmem_alloc(bufsize, KM_SLEEP);
8300 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
8302 bzero(packed + nvsize, bufsize - nvsize);
8304 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
8306 vmem_free(packed, bufsize);
8308 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
8309 dmu_buf_will_dirty(db, tx);
8310 *(uint64_t *)db->db_data = nvsize;
8311 dmu_buf_rele(db, FTAG);
8315 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
8316 const char *config, const char *entry)
8326 * Update the MOS nvlist describing the list of available devices.
8327 * spa_validate_aux() will have already made sure this nvlist is
8328 * valid and the vdevs are labeled appropriately.
8330 if (sav->sav_object == 0) {
8331 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
8332 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
8333 sizeof (uint64_t), tx);
8334 VERIFY(zap_update(spa->spa_meta_objset,
8335 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
8336 &sav->sav_object, tx) == 0);
8339 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
8340 if (sav->sav_count == 0) {
8341 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
8343 list = kmem_alloc(sav->sav_count*sizeof (void *), KM_SLEEP);
8344 for (i = 0; i < sav->sav_count; i++)
8345 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
8346 B_FALSE, VDEV_CONFIG_L2CACHE);
8347 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
8348 sav->sav_count) == 0);
8349 for (i = 0; i < sav->sav_count; i++)
8350 nvlist_free(list[i]);
8351 kmem_free(list, sav->sav_count * sizeof (void *));
8354 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
8355 nvlist_free(nvroot);
8357 sav->sav_sync = B_FALSE;
8361 * Rebuild spa's all-vdev ZAP from the vdev ZAPs indicated in each vdev_t.
8362 * The all-vdev ZAP must be empty.
8365 spa_avz_build(vdev_t *vd, uint64_t avz, dmu_tx_t *tx)
8367 spa_t *spa = vd->vdev_spa;
8369 if (vd->vdev_top_zap != 0) {
8370 VERIFY0(zap_add_int(spa->spa_meta_objset, avz,
8371 vd->vdev_top_zap, tx));
8373 if (vd->vdev_leaf_zap != 0) {
8374 VERIFY0(zap_add_int(spa->spa_meta_objset, avz,
8375 vd->vdev_leaf_zap, tx));
8377 for (uint64_t i = 0; i < vd->vdev_children; i++) {
8378 spa_avz_build(vd->vdev_child[i], avz, tx);
8383 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
8388 * If the pool is being imported from a pre-per-vdev-ZAP version of ZFS,
8389 * its config may not be dirty but we still need to build per-vdev ZAPs.
8390 * Similarly, if the pool is being assembled (e.g. after a split), we
8391 * need to rebuild the AVZ although the config may not be dirty.
8393 if (list_is_empty(&spa->spa_config_dirty_list) &&
8394 spa->spa_avz_action == AVZ_ACTION_NONE)
8397 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
8399 ASSERT(spa->spa_avz_action == AVZ_ACTION_NONE ||
8400 spa->spa_avz_action == AVZ_ACTION_INITIALIZE ||
8401 spa->spa_all_vdev_zaps != 0);
8403 if (spa->spa_avz_action == AVZ_ACTION_REBUILD) {
8404 /* Make and build the new AVZ */
8405 uint64_t new_avz = zap_create(spa->spa_meta_objset,
8406 DMU_OTN_ZAP_METADATA, DMU_OT_NONE, 0, tx);
8407 spa_avz_build(spa->spa_root_vdev, new_avz, tx);
8409 /* Diff old AVZ with new one */
8413 for (zap_cursor_init(&zc, spa->spa_meta_objset,
8414 spa->spa_all_vdev_zaps);
8415 zap_cursor_retrieve(&zc, &za) == 0;
8416 zap_cursor_advance(&zc)) {
8417 uint64_t vdzap = za.za_first_integer;
8418 if (zap_lookup_int(spa->spa_meta_objset, new_avz,
8421 * ZAP is listed in old AVZ but not in new one;
8424 VERIFY0(zap_destroy(spa->spa_meta_objset, vdzap,
8429 zap_cursor_fini(&zc);
8431 /* Destroy the old AVZ */
8432 VERIFY0(zap_destroy(spa->spa_meta_objset,
8433 spa->spa_all_vdev_zaps, tx));
8435 /* Replace the old AVZ in the dir obj with the new one */
8436 VERIFY0(zap_update(spa->spa_meta_objset,
8437 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP,
8438 sizeof (new_avz), 1, &new_avz, tx));
8440 spa->spa_all_vdev_zaps = new_avz;
8441 } else if (spa->spa_avz_action == AVZ_ACTION_DESTROY) {
8445 /* Walk through the AVZ and destroy all listed ZAPs */
8446 for (zap_cursor_init(&zc, spa->spa_meta_objset,
8447 spa->spa_all_vdev_zaps);
8448 zap_cursor_retrieve(&zc, &za) == 0;
8449 zap_cursor_advance(&zc)) {
8450 uint64_t zap = za.za_first_integer;
8451 VERIFY0(zap_destroy(spa->spa_meta_objset, zap, tx));
8454 zap_cursor_fini(&zc);
8456 /* Destroy and unlink the AVZ itself */
8457 VERIFY0(zap_destroy(spa->spa_meta_objset,
8458 spa->spa_all_vdev_zaps, tx));
8459 VERIFY0(zap_remove(spa->spa_meta_objset,
8460 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP, tx));
8461 spa->spa_all_vdev_zaps = 0;
8464 if (spa->spa_all_vdev_zaps == 0) {
8465 spa->spa_all_vdev_zaps = zap_create_link(spa->spa_meta_objset,
8466 DMU_OTN_ZAP_METADATA, DMU_POOL_DIRECTORY_OBJECT,
8467 DMU_POOL_VDEV_ZAP_MAP, tx);
8469 spa->spa_avz_action = AVZ_ACTION_NONE;
8471 /* Create ZAPs for vdevs that don't have them. */
8472 vdev_construct_zaps(spa->spa_root_vdev, tx);
8474 config = spa_config_generate(spa, spa->spa_root_vdev,
8475 dmu_tx_get_txg(tx), B_FALSE);
8478 * If we're upgrading the spa version then make sure that
8479 * the config object gets updated with the correct version.
8481 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
8482 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
8483 spa->spa_uberblock.ub_version);
8485 spa_config_exit(spa, SCL_STATE, FTAG);
8487 nvlist_free(spa->spa_config_syncing);
8488 spa->spa_config_syncing = config;
8490 spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
8494 spa_sync_version(void *arg, dmu_tx_t *tx)
8496 uint64_t *versionp = arg;
8497 uint64_t version = *versionp;
8498 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
8501 * Setting the version is special cased when first creating the pool.
8503 ASSERT(tx->tx_txg != TXG_INITIAL);
8505 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
8506 ASSERT(version >= spa_version(spa));
8508 spa->spa_uberblock.ub_version = version;
8509 vdev_config_dirty(spa->spa_root_vdev);
8510 spa_history_log_internal(spa, "set", tx, "version=%lld",
8511 (longlong_t)version);
8515 * Set zpool properties.
8518 spa_sync_props(void *arg, dmu_tx_t *tx)
8520 nvlist_t *nvp = arg;
8521 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
8522 objset_t *mos = spa->spa_meta_objset;
8523 nvpair_t *elem = NULL;
8525 mutex_enter(&spa->spa_props_lock);
8527 while ((elem = nvlist_next_nvpair(nvp, elem))) {
8529 char *strval, *fname;
8531 const char *propname;
8532 zprop_type_t proptype;
8535 switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
8536 case ZPOOL_PROP_INVAL:
8538 * We checked this earlier in spa_prop_validate().
8540 ASSERT(zpool_prop_feature(nvpair_name(elem)));
8542 fname = strchr(nvpair_name(elem), '@') + 1;
8543 VERIFY0(zfeature_lookup_name(fname, &fid));
8545 spa_feature_enable(spa, fid, tx);
8546 spa_history_log_internal(spa, "set", tx,
8547 "%s=enabled", nvpair_name(elem));
8550 case ZPOOL_PROP_VERSION:
8551 intval = fnvpair_value_uint64(elem);
8553 * The version is synced separately before other
8554 * properties and should be correct by now.
8556 ASSERT3U(spa_version(spa), >=, intval);
8559 case ZPOOL_PROP_ALTROOT:
8561 * 'altroot' is a non-persistent property. It should
8562 * have been set temporarily at creation or import time.
8564 ASSERT(spa->spa_root != NULL);
8567 case ZPOOL_PROP_READONLY:
8568 case ZPOOL_PROP_CACHEFILE:
8570 * 'readonly' and 'cachefile' are also non-persistent
8574 case ZPOOL_PROP_COMMENT:
8575 strval = fnvpair_value_string(elem);
8576 if (spa->spa_comment != NULL)
8577 spa_strfree(spa->spa_comment);
8578 spa->spa_comment = spa_strdup(strval);
8580 * We need to dirty the configuration on all the vdevs
8581 * so that their labels get updated. It's unnecessary
8582 * to do this for pool creation since the vdev's
8583 * configuration has already been dirtied.
8585 if (tx->tx_txg != TXG_INITIAL)
8586 vdev_config_dirty(spa->spa_root_vdev);
8587 spa_history_log_internal(spa, "set", tx,
8588 "%s=%s", nvpair_name(elem), strval);
8592 * Set pool property values in the poolprops mos object.
8594 if (spa->spa_pool_props_object == 0) {
8595 spa->spa_pool_props_object =
8596 zap_create_link(mos, DMU_OT_POOL_PROPS,
8597 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
8601 /* normalize the property name */
8602 propname = zpool_prop_to_name(prop);
8603 proptype = zpool_prop_get_type(prop);
8605 if (nvpair_type(elem) == DATA_TYPE_STRING) {
8606 ASSERT(proptype == PROP_TYPE_STRING);
8607 strval = fnvpair_value_string(elem);
8608 VERIFY0(zap_update(mos,
8609 spa->spa_pool_props_object, propname,
8610 1, strlen(strval) + 1, strval, tx));
8611 spa_history_log_internal(spa, "set", tx,
8612 "%s=%s", nvpair_name(elem), strval);
8613 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
8614 intval = fnvpair_value_uint64(elem);
8616 if (proptype == PROP_TYPE_INDEX) {
8618 VERIFY0(zpool_prop_index_to_string(
8619 prop, intval, &unused));
8621 VERIFY0(zap_update(mos,
8622 spa->spa_pool_props_object, propname,
8623 8, 1, &intval, tx));
8624 spa_history_log_internal(spa, "set", tx,
8625 "%s=%lld", nvpair_name(elem),
8626 (longlong_t)intval);
8628 ASSERT(0); /* not allowed */
8632 case ZPOOL_PROP_DELEGATION:
8633 spa->spa_delegation = intval;
8635 case ZPOOL_PROP_BOOTFS:
8636 spa->spa_bootfs = intval;
8638 case ZPOOL_PROP_FAILUREMODE:
8639 spa->spa_failmode = intval;
8641 case ZPOOL_PROP_AUTOTRIM:
8642 spa->spa_autotrim = intval;
8643 spa_async_request(spa,
8644 SPA_ASYNC_AUTOTRIM_RESTART);
8646 case ZPOOL_PROP_AUTOEXPAND:
8647 spa->spa_autoexpand = intval;
8648 if (tx->tx_txg != TXG_INITIAL)
8649 spa_async_request(spa,
8650 SPA_ASYNC_AUTOEXPAND);
8652 case ZPOOL_PROP_MULTIHOST:
8653 spa->spa_multihost = intval;
8662 mutex_exit(&spa->spa_props_lock);
8666 * Perform one-time upgrade on-disk changes. spa_version() does not
8667 * reflect the new version this txg, so there must be no changes this
8668 * txg to anything that the upgrade code depends on after it executes.
8669 * Therefore this must be called after dsl_pool_sync() does the sync
8673 spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
8675 if (spa_sync_pass(spa) != 1)
8678 dsl_pool_t *dp = spa->spa_dsl_pool;
8679 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
8681 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
8682 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
8683 dsl_pool_create_origin(dp, tx);
8685 /* Keeping the origin open increases spa_minref */
8686 spa->spa_minref += 3;
8689 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
8690 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
8691 dsl_pool_upgrade_clones(dp, tx);
8694 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
8695 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
8696 dsl_pool_upgrade_dir_clones(dp, tx);
8698 /* Keeping the freedir open increases spa_minref */
8699 spa->spa_minref += 3;
8702 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES &&
8703 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
8704 spa_feature_create_zap_objects(spa, tx);
8708 * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable
8709 * when possibility to use lz4 compression for metadata was added
8710 * Old pools that have this feature enabled must be upgraded to have
8711 * this feature active
8713 if (spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
8714 boolean_t lz4_en = spa_feature_is_enabled(spa,
8715 SPA_FEATURE_LZ4_COMPRESS);
8716 boolean_t lz4_ac = spa_feature_is_active(spa,
8717 SPA_FEATURE_LZ4_COMPRESS);
8719 if (lz4_en && !lz4_ac)
8720 spa_feature_incr(spa, SPA_FEATURE_LZ4_COMPRESS, tx);
8724 * If we haven't written the salt, do so now. Note that the
8725 * feature may not be activated yet, but that's fine since
8726 * the presence of this ZAP entry is backwards compatible.
8728 if (zap_contains(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
8729 DMU_POOL_CHECKSUM_SALT) == ENOENT) {
8730 VERIFY0(zap_add(spa->spa_meta_objset,
8731 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CHECKSUM_SALT, 1,
8732 sizeof (spa->spa_cksum_salt.zcs_bytes),
8733 spa->spa_cksum_salt.zcs_bytes, tx));
8736 rrw_exit(&dp->dp_config_rwlock, FTAG);
8740 vdev_indirect_state_sync_verify(vdev_t *vd)
8742 vdev_indirect_mapping_t *vim __maybe_unused = vd->vdev_indirect_mapping;
8743 vdev_indirect_births_t *vib __maybe_unused = vd->vdev_indirect_births;
8745 if (vd->vdev_ops == &vdev_indirect_ops) {
8746 ASSERT(vim != NULL);
8747 ASSERT(vib != NULL);
8750 uint64_t obsolete_sm_object = 0;
8751 ASSERT0(vdev_obsolete_sm_object(vd, &obsolete_sm_object));
8752 if (obsolete_sm_object != 0) {
8753 ASSERT(vd->vdev_obsolete_sm != NULL);
8754 ASSERT(vd->vdev_removing ||
8755 vd->vdev_ops == &vdev_indirect_ops);
8756 ASSERT(vdev_indirect_mapping_num_entries(vim) > 0);
8757 ASSERT(vdev_indirect_mapping_bytes_mapped(vim) > 0);
8758 ASSERT3U(obsolete_sm_object, ==,
8759 space_map_object(vd->vdev_obsolete_sm));
8760 ASSERT3U(vdev_indirect_mapping_bytes_mapped(vim), >=,
8761 space_map_allocated(vd->vdev_obsolete_sm));
8763 ASSERT(vd->vdev_obsolete_segments != NULL);
8766 * Since frees / remaps to an indirect vdev can only
8767 * happen in syncing context, the obsolete segments
8768 * tree must be empty when we start syncing.
8770 ASSERT0(range_tree_space(vd->vdev_obsolete_segments));
8774 * Set the top-level vdev's max queue depth. Evaluate each top-level's
8775 * async write queue depth in case it changed. The max queue depth will
8776 * not change in the middle of syncing out this txg.
8779 spa_sync_adjust_vdev_max_queue_depth(spa_t *spa)
8781 ASSERT(spa_writeable(spa));
8783 vdev_t *rvd = spa->spa_root_vdev;
8784 uint32_t max_queue_depth = zfs_vdev_async_write_max_active *
8785 zfs_vdev_queue_depth_pct / 100;
8786 metaslab_class_t *normal = spa_normal_class(spa);
8787 metaslab_class_t *special = spa_special_class(spa);
8788 metaslab_class_t *dedup = spa_dedup_class(spa);
8790 uint64_t slots_per_allocator = 0;
8791 for (int c = 0; c < rvd->vdev_children; c++) {
8792 vdev_t *tvd = rvd->vdev_child[c];
8794 metaslab_group_t *mg = tvd->vdev_mg;
8795 if (mg == NULL || !metaslab_group_initialized(mg))
8798 metaslab_class_t *mc = mg->mg_class;
8799 if (mc != normal && mc != special && mc != dedup)
8803 * It is safe to do a lock-free check here because only async
8804 * allocations look at mg_max_alloc_queue_depth, and async
8805 * allocations all happen from spa_sync().
8807 for (int i = 0; i < mg->mg_allocators; i++) {
8808 ASSERT0(zfs_refcount_count(
8809 &(mg->mg_allocator[i].mga_alloc_queue_depth)));
8811 mg->mg_max_alloc_queue_depth = max_queue_depth;
8813 for (int i = 0; i < mg->mg_allocators; i++) {
8814 mg->mg_allocator[i].mga_cur_max_alloc_queue_depth =
8815 zfs_vdev_def_queue_depth;
8817 slots_per_allocator += zfs_vdev_def_queue_depth;
8820 for (int i = 0; i < spa->spa_alloc_count; i++) {
8821 ASSERT0(zfs_refcount_count(&normal->mc_alloc_slots[i]));
8822 ASSERT0(zfs_refcount_count(&special->mc_alloc_slots[i]));
8823 ASSERT0(zfs_refcount_count(&dedup->mc_alloc_slots[i]));
8824 normal->mc_alloc_max_slots[i] = slots_per_allocator;
8825 special->mc_alloc_max_slots[i] = slots_per_allocator;
8826 dedup->mc_alloc_max_slots[i] = slots_per_allocator;
8828 normal->mc_alloc_throttle_enabled = zio_dva_throttle_enabled;
8829 special->mc_alloc_throttle_enabled = zio_dva_throttle_enabled;
8830 dedup->mc_alloc_throttle_enabled = zio_dva_throttle_enabled;
8834 spa_sync_condense_indirect(spa_t *spa, dmu_tx_t *tx)
8836 ASSERT(spa_writeable(spa));
8838 vdev_t *rvd = spa->spa_root_vdev;
8839 for (int c = 0; c < rvd->vdev_children; c++) {
8840 vdev_t *vd = rvd->vdev_child[c];
8841 vdev_indirect_state_sync_verify(vd);
8843 if (vdev_indirect_should_condense(vd)) {
8844 spa_condense_indirect_start_sync(vd, tx);
8851 spa_sync_iterate_to_convergence(spa_t *spa, dmu_tx_t *tx)
8853 objset_t *mos = spa->spa_meta_objset;
8854 dsl_pool_t *dp = spa->spa_dsl_pool;
8855 uint64_t txg = tx->tx_txg;
8856 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
8859 int pass = ++spa->spa_sync_pass;
8861 spa_sync_config_object(spa, tx);
8862 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
8863 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
8864 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
8865 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
8866 spa_errlog_sync(spa, txg);
8867 dsl_pool_sync(dp, txg);
8869 if (pass < zfs_sync_pass_deferred_free ||
8870 spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP)) {
8872 * If the log space map feature is active we don't
8873 * care about deferred frees and the deferred bpobj
8874 * as the log space map should effectively have the
8875 * same results (i.e. appending only to one object).
8877 spa_sync_frees(spa, free_bpl, tx);
8880 * We can not defer frees in pass 1, because
8881 * we sync the deferred frees later in pass 1.
8883 ASSERT3U(pass, >, 1);
8884 bplist_iterate(free_bpl, bpobj_enqueue_alloc_cb,
8885 &spa->spa_deferred_bpobj, tx);
8889 dsl_scan_sync(dp, tx);
8891 spa_sync_upgrades(spa, tx);
8893 spa_flush_metaslabs(spa, tx);
8896 while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
8901 * Note: We need to check if the MOS is dirty because we could
8902 * have marked the MOS dirty without updating the uberblock
8903 * (e.g. if we have sync tasks but no dirty user data). We need
8904 * to check the uberblock's rootbp because it is updated if we
8905 * have synced out dirty data (though in this case the MOS will
8906 * most likely also be dirty due to second order effects, we
8907 * don't want to rely on that here).
8910 spa->spa_uberblock.ub_rootbp.blk_birth < txg &&
8911 !dmu_objset_is_dirty(mos, txg)) {
8913 * Nothing changed on the first pass, therefore this
8914 * TXG is a no-op. Avoid syncing deferred frees, so
8915 * that we can keep this TXG as a no-op.
8917 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
8918 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
8919 ASSERT(txg_list_empty(&dp->dp_sync_tasks, txg));
8920 ASSERT(txg_list_empty(&dp->dp_early_sync_tasks, txg));
8924 spa_sync_deferred_frees(spa, tx);
8925 } while (dmu_objset_is_dirty(mos, txg));
8929 * Rewrite the vdev configuration (which includes the uberblock) to
8930 * commit the transaction group.
8932 * If there are no dirty vdevs, we sync the uberblock to a few random
8933 * top-level vdevs that are known to be visible in the config cache
8934 * (see spa_vdev_add() for a complete description). If there *are* dirty
8935 * vdevs, sync the uberblock to all vdevs.
8938 spa_sync_rewrite_vdev_config(spa_t *spa, dmu_tx_t *tx)
8940 vdev_t *rvd = spa->spa_root_vdev;
8941 uint64_t txg = tx->tx_txg;
8947 * We hold SCL_STATE to prevent vdev open/close/etc.
8948 * while we're attempting to write the vdev labels.
8950 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
8952 if (list_is_empty(&spa->spa_config_dirty_list)) {
8953 vdev_t *svd[SPA_SYNC_MIN_VDEVS] = { NULL };
8955 int children = rvd->vdev_children;
8956 int c0 = spa_get_random(children);
8958 for (int c = 0; c < children; c++) {
8960 rvd->vdev_child[(c0 + c) % children];
8962 /* Stop when revisiting the first vdev */
8963 if (c > 0 && svd[0] == vd)
8966 if (vd->vdev_ms_array == 0 ||
8968 !vdev_is_concrete(vd))
8971 svd[svdcount++] = vd;
8972 if (svdcount == SPA_SYNC_MIN_VDEVS)
8975 error = vdev_config_sync(svd, svdcount, txg);
8977 error = vdev_config_sync(rvd->vdev_child,
8978 rvd->vdev_children, txg);
8982 spa->spa_last_synced_guid = rvd->vdev_guid;
8984 spa_config_exit(spa, SCL_STATE, FTAG);
8988 zio_suspend(spa, NULL, ZIO_SUSPEND_IOERR);
8989 zio_resume_wait(spa);
8994 * Sync the specified transaction group. New blocks may be dirtied as
8995 * part of the process, so we iterate until it converges.
8998 spa_sync(spa_t *spa, uint64_t txg)
9002 VERIFY(spa_writeable(spa));
9005 * Wait for i/os issued in open context that need to complete
9006 * before this txg syncs.
9008 (void) zio_wait(spa->spa_txg_zio[txg & TXG_MASK]);
9009 spa->spa_txg_zio[txg & TXG_MASK] = zio_root(spa, NULL, NULL,
9013 * Lock out configuration changes.
9015 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
9017 spa->spa_syncing_txg = txg;
9018 spa->spa_sync_pass = 0;
9020 for (int i = 0; i < spa->spa_alloc_count; i++) {
9021 mutex_enter(&spa->spa_alloc_locks[i]);
9022 VERIFY0(avl_numnodes(&spa->spa_alloc_trees[i]));
9023 mutex_exit(&spa->spa_alloc_locks[i]);
9027 * If there are any pending vdev state changes, convert them
9028 * into config changes that go out with this transaction group.
9030 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
9031 while (list_head(&spa->spa_state_dirty_list) != NULL) {
9033 * We need the write lock here because, for aux vdevs,
9034 * calling vdev_config_dirty() modifies sav_config.
9035 * This is ugly and will become unnecessary when we
9036 * eliminate the aux vdev wart by integrating all vdevs
9037 * into the root vdev tree.
9039 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
9040 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
9041 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
9042 vdev_state_clean(vd);
9043 vdev_config_dirty(vd);
9045 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
9046 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
9048 spa_config_exit(spa, SCL_STATE, FTAG);
9050 dsl_pool_t *dp = spa->spa_dsl_pool;
9051 dmu_tx_t *tx = dmu_tx_create_assigned(dp, txg);
9053 spa->spa_sync_starttime = gethrtime();
9054 taskq_cancel_id(system_delay_taskq, spa->spa_deadman_tqid);
9055 spa->spa_deadman_tqid = taskq_dispatch_delay(system_delay_taskq,
9056 spa_deadman, spa, TQ_SLEEP, ddi_get_lbolt() +
9057 NSEC_TO_TICK(spa->spa_deadman_synctime));
9060 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
9061 * set spa_deflate if we have no raid-z vdevs.
9063 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
9064 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
9065 vdev_t *rvd = spa->spa_root_vdev;
9068 for (i = 0; i < rvd->vdev_children; i++) {
9069 vd = rvd->vdev_child[i];
9070 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
9073 if (i == rvd->vdev_children) {
9074 spa->spa_deflate = TRUE;
9075 VERIFY0(zap_add(spa->spa_meta_objset,
9076 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
9077 sizeof (uint64_t), 1, &spa->spa_deflate, tx));
9081 spa_sync_adjust_vdev_max_queue_depth(spa);
9083 spa_sync_condense_indirect(spa, tx);
9085 spa_sync_iterate_to_convergence(spa, tx);
9088 if (!list_is_empty(&spa->spa_config_dirty_list)) {
9090 * Make sure that the number of ZAPs for all the vdevs matches
9091 * the number of ZAPs in the per-vdev ZAP list. This only gets
9092 * called if the config is dirty; otherwise there may be
9093 * outstanding AVZ operations that weren't completed in
9094 * spa_sync_config_object.
9096 uint64_t all_vdev_zap_entry_count;
9097 ASSERT0(zap_count(spa->spa_meta_objset,
9098 spa->spa_all_vdev_zaps, &all_vdev_zap_entry_count));
9099 ASSERT3U(vdev_count_verify_zaps(spa->spa_root_vdev), ==,
9100 all_vdev_zap_entry_count);
9104 if (spa->spa_vdev_removal != NULL) {
9105 ASSERT0(spa->spa_vdev_removal->svr_bytes_done[txg & TXG_MASK]);
9108 spa_sync_rewrite_vdev_config(spa, tx);
9111 taskq_cancel_id(system_delay_taskq, spa->spa_deadman_tqid);
9112 spa->spa_deadman_tqid = 0;
9115 * Clear the dirty config list.
9117 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
9118 vdev_config_clean(vd);
9121 * Now that the new config has synced transactionally,
9122 * let it become visible to the config cache.
9124 if (spa->spa_config_syncing != NULL) {
9125 spa_config_set(spa, spa->spa_config_syncing);
9126 spa->spa_config_txg = txg;
9127 spa->spa_config_syncing = NULL;
9130 dsl_pool_sync_done(dp, txg);
9132 for (int i = 0; i < spa->spa_alloc_count; i++) {
9133 mutex_enter(&spa->spa_alloc_locks[i]);
9134 VERIFY0(avl_numnodes(&spa->spa_alloc_trees[i]));
9135 mutex_exit(&spa->spa_alloc_locks[i]);
9139 * Update usable space statistics.
9141 while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
9143 vdev_sync_done(vd, txg);
9145 metaslab_class_evict_old(spa->spa_normal_class, txg);
9146 metaslab_class_evict_old(spa->spa_log_class, txg);
9148 spa_sync_close_syncing_log_sm(spa);
9150 spa_update_dspace(spa);
9153 * It had better be the case that we didn't dirty anything
9154 * since vdev_config_sync().
9156 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
9157 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
9158 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
9160 while (zfs_pause_spa_sync)
9163 spa->spa_sync_pass = 0;
9166 * Update the last synced uberblock here. We want to do this at
9167 * the end of spa_sync() so that consumers of spa_last_synced_txg()
9168 * will be guaranteed that all the processing associated with
9169 * that txg has been completed.
9171 spa->spa_ubsync = spa->spa_uberblock;
9172 spa_config_exit(spa, SCL_CONFIG, FTAG);
9174 spa_handle_ignored_writes(spa);
9177 * If any async tasks have been requested, kick them off.
9179 spa_async_dispatch(spa);
9183 * Sync all pools. We don't want to hold the namespace lock across these
9184 * operations, so we take a reference on the spa_t and drop the lock during the
9188 spa_sync_allpools(void)
9191 mutex_enter(&spa_namespace_lock);
9192 while ((spa = spa_next(spa)) != NULL) {
9193 if (spa_state(spa) != POOL_STATE_ACTIVE ||
9194 !spa_writeable(spa) || spa_suspended(spa))
9196 spa_open_ref(spa, FTAG);
9197 mutex_exit(&spa_namespace_lock);
9198 txg_wait_synced(spa_get_dsl(spa), 0);
9199 mutex_enter(&spa_namespace_lock);
9200 spa_close(spa, FTAG);
9202 mutex_exit(&spa_namespace_lock);
9206 * ==========================================================================
9207 * Miscellaneous routines
9208 * ==========================================================================
9212 * Remove all pools in the system.
9220 * Remove all cached state. All pools should be closed now,
9221 * so every spa in the AVL tree should be unreferenced.
9223 mutex_enter(&spa_namespace_lock);
9224 while ((spa = spa_next(NULL)) != NULL) {
9226 * Stop async tasks. The async thread may need to detach
9227 * a device that's been replaced, which requires grabbing
9228 * spa_namespace_lock, so we must drop it here.
9230 spa_open_ref(spa, FTAG);
9231 mutex_exit(&spa_namespace_lock);
9232 spa_async_suspend(spa);
9233 mutex_enter(&spa_namespace_lock);
9234 spa_close(spa, FTAG);
9236 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
9238 spa_deactivate(spa);
9242 mutex_exit(&spa_namespace_lock);
9246 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
9251 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
9255 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
9256 vd = spa->spa_l2cache.sav_vdevs[i];
9257 if (vd->vdev_guid == guid)
9261 for (i = 0; i < spa->spa_spares.sav_count; i++) {
9262 vd = spa->spa_spares.sav_vdevs[i];
9263 if (vd->vdev_guid == guid)
9272 spa_upgrade(spa_t *spa, uint64_t version)
9274 ASSERT(spa_writeable(spa));
9276 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
9279 * This should only be called for a non-faulted pool, and since a
9280 * future version would result in an unopenable pool, this shouldn't be
9283 ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version));
9284 ASSERT3U(version, >=, spa->spa_uberblock.ub_version);
9286 spa->spa_uberblock.ub_version = version;
9287 vdev_config_dirty(spa->spa_root_vdev);
9289 spa_config_exit(spa, SCL_ALL, FTAG);
9291 txg_wait_synced(spa_get_dsl(spa), 0);
9295 spa_has_spare(spa_t *spa, uint64_t guid)
9299 spa_aux_vdev_t *sav = &spa->spa_spares;
9301 for (i = 0; i < sav->sav_count; i++)
9302 if (sav->sav_vdevs[i]->vdev_guid == guid)
9305 for (i = 0; i < sav->sav_npending; i++) {
9306 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
9307 &spareguid) == 0 && spareguid == guid)
9315 * Check if a pool has an active shared spare device.
9316 * Note: reference count of an active spare is 2, as a spare and as a replace
9319 spa_has_active_shared_spare(spa_t *spa)
9323 spa_aux_vdev_t *sav = &spa->spa_spares;
9325 for (i = 0; i < sav->sav_count; i++) {
9326 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
9327 &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
9336 spa_total_metaslabs(spa_t *spa)
9338 vdev_t *rvd = spa->spa_root_vdev;
9341 for (uint64_t c = 0; c < rvd->vdev_children; c++) {
9342 vdev_t *vd = rvd->vdev_child[c];
9343 if (!vdev_is_concrete(vd))
9345 m += vd->vdev_ms_count;
9351 * Notify any waiting threads that some activity has switched from being in-
9352 * progress to not-in-progress so that the thread can wake up and determine
9353 * whether it is finished waiting.
9356 spa_notify_waiters(spa_t *spa)
9359 * Acquiring spa_activities_lock here prevents the cv_broadcast from
9360 * happening between the waiting thread's check and cv_wait.
9362 mutex_enter(&spa->spa_activities_lock);
9363 cv_broadcast(&spa->spa_activities_cv);
9364 mutex_exit(&spa->spa_activities_lock);
9368 * Notify any waiting threads that the pool is exporting, and then block until
9369 * they are finished using the spa_t.
9372 spa_wake_waiters(spa_t *spa)
9374 mutex_enter(&spa->spa_activities_lock);
9375 spa->spa_waiters_cancel = B_TRUE;
9376 cv_broadcast(&spa->spa_activities_cv);
9377 while (spa->spa_waiters != 0)
9378 cv_wait(&spa->spa_waiters_cv, &spa->spa_activities_lock);
9379 spa->spa_waiters_cancel = B_FALSE;
9380 mutex_exit(&spa->spa_activities_lock);
9383 /* Whether the vdev or any of its descendants are being initialized/trimmed. */
9385 spa_vdev_activity_in_progress_impl(vdev_t *vd, zpool_wait_activity_t activity)
9387 spa_t *spa = vd->vdev_spa;
9389 ASSERT(spa_config_held(spa, SCL_CONFIG | SCL_STATE, RW_READER));
9390 ASSERT(MUTEX_HELD(&spa->spa_activities_lock));
9391 ASSERT(activity == ZPOOL_WAIT_INITIALIZE ||
9392 activity == ZPOOL_WAIT_TRIM);
9394 kmutex_t *lock = activity == ZPOOL_WAIT_INITIALIZE ?
9395 &vd->vdev_initialize_lock : &vd->vdev_trim_lock;
9397 mutex_exit(&spa->spa_activities_lock);
9399 mutex_enter(&spa->spa_activities_lock);
9401 boolean_t in_progress = (activity == ZPOOL_WAIT_INITIALIZE) ?
9402 (vd->vdev_initialize_state == VDEV_INITIALIZE_ACTIVE) :
9403 (vd->vdev_trim_state == VDEV_TRIM_ACTIVE);
9409 for (int i = 0; i < vd->vdev_children; i++) {
9410 if (spa_vdev_activity_in_progress_impl(vd->vdev_child[i],
9419 * If use_guid is true, this checks whether the vdev specified by guid is
9420 * being initialized/trimmed. Otherwise, it checks whether any vdev in the pool
9421 * is being initialized/trimmed. The caller must hold the config lock and
9422 * spa_activities_lock.
9425 spa_vdev_activity_in_progress(spa_t *spa, boolean_t use_guid, uint64_t guid,
9426 zpool_wait_activity_t activity, boolean_t *in_progress)
9428 mutex_exit(&spa->spa_activities_lock);
9429 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
9430 mutex_enter(&spa->spa_activities_lock);
9434 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
9435 if (vd == NULL || !vd->vdev_ops->vdev_op_leaf) {
9436 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
9440 vd = spa->spa_root_vdev;
9443 *in_progress = spa_vdev_activity_in_progress_impl(vd, activity);
9445 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
9450 * Locking for waiting threads
9451 * ---------------------------
9453 * Waiting threads need a way to check whether a given activity is in progress,
9454 * and then, if it is, wait for it to complete. Each activity will have some
9455 * in-memory representation of the relevant on-disk state which can be used to
9456 * determine whether or not the activity is in progress. The in-memory state and
9457 * the locking used to protect it will be different for each activity, and may
9458 * not be suitable for use with a cvar (e.g., some state is protected by the
9459 * config lock). To allow waiting threads to wait without any races, another
9460 * lock, spa_activities_lock, is used.
9462 * When the state is checked, both the activity-specific lock (if there is one)
9463 * and spa_activities_lock are held. In some cases, the activity-specific lock
9464 * is acquired explicitly (e.g. the config lock). In others, the locking is
9465 * internal to some check (e.g. bpobj_is_empty). After checking, the waiting
9466 * thread releases the activity-specific lock and, if the activity is in
9467 * progress, then cv_waits using spa_activities_lock.
9469 * The waiting thread is woken when another thread, one completing some
9470 * activity, updates the state of the activity and then calls
9471 * spa_notify_waiters, which will cv_broadcast. This 'completing' thread only
9472 * needs to hold its activity-specific lock when updating the state, and this
9473 * lock can (but doesn't have to) be dropped before calling spa_notify_waiters.
9475 * Because spa_notify_waiters acquires spa_activities_lock before broadcasting,
9476 * and because it is held when the waiting thread checks the state of the
9477 * activity, it can never be the case that the completing thread both updates
9478 * the activity state and cv_broadcasts in between the waiting thread's check
9479 * and cv_wait. Thus, a waiting thread can never miss a wakeup.
9481 * In order to prevent deadlock, when the waiting thread does its check, in some
9482 * cases it will temporarily drop spa_activities_lock in order to acquire the
9483 * activity-specific lock. The order in which spa_activities_lock and the
9484 * activity specific lock are acquired in the waiting thread is determined by
9485 * the order in which they are acquired in the completing thread; if the
9486 * completing thread calls spa_notify_waiters with the activity-specific lock
9487 * held, then the waiting thread must also acquire the activity-specific lock
9492 spa_activity_in_progress(spa_t *spa, zpool_wait_activity_t activity,
9493 boolean_t use_tag, uint64_t tag, boolean_t *in_progress)
9497 ASSERT(MUTEX_HELD(&spa->spa_activities_lock));
9500 case ZPOOL_WAIT_CKPT_DISCARD:
9502 (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT) &&
9503 zap_contains(spa_meta_objset(spa),
9504 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_ZPOOL_CHECKPOINT) ==
9507 case ZPOOL_WAIT_FREE:
9508 *in_progress = ((spa_version(spa) >= SPA_VERSION_DEADLISTS &&
9509 !bpobj_is_empty(&spa->spa_dsl_pool->dp_free_bpobj)) ||
9510 spa_feature_is_active(spa, SPA_FEATURE_ASYNC_DESTROY) ||
9511 spa_livelist_delete_check(spa));
9513 case ZPOOL_WAIT_INITIALIZE:
9514 case ZPOOL_WAIT_TRIM:
9515 error = spa_vdev_activity_in_progress(spa, use_tag, tag,
9516 activity, in_progress);
9518 case ZPOOL_WAIT_REPLACE:
9519 mutex_exit(&spa->spa_activities_lock);
9520 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
9521 mutex_enter(&spa->spa_activities_lock);
9523 *in_progress = vdev_replace_in_progress(spa->spa_root_vdev);
9524 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
9526 case ZPOOL_WAIT_REMOVE:
9527 *in_progress = (spa->spa_removing_phys.sr_state ==
9530 case ZPOOL_WAIT_RESILVER:
9531 if ((*in_progress = vdev_rebuild_active(spa->spa_root_vdev)))
9534 case ZPOOL_WAIT_SCRUB:
9536 boolean_t scanning, paused, is_scrub;
9537 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
9539 is_scrub = (scn->scn_phys.scn_func == POOL_SCAN_SCRUB);
9540 scanning = (scn->scn_phys.scn_state == DSS_SCANNING);
9541 paused = dsl_scan_is_paused_scrub(scn);
9542 *in_progress = (scanning && !paused &&
9543 is_scrub == (activity == ZPOOL_WAIT_SCRUB));
9547 panic("unrecognized value for activity %d", activity);
9554 spa_wait_common(const char *pool, zpool_wait_activity_t activity,
9555 boolean_t use_tag, uint64_t tag, boolean_t *waited)
9558 * The tag is used to distinguish between instances of an activity.
9559 * 'initialize' and 'trim' are the only activities that we use this for.
9560 * The other activities can only have a single instance in progress in a
9561 * pool at one time, making the tag unnecessary.
9563 * There can be multiple devices being replaced at once, but since they
9564 * all finish once resilvering finishes, we don't bother keeping track
9565 * of them individually, we just wait for them all to finish.
9567 if (use_tag && activity != ZPOOL_WAIT_INITIALIZE &&
9568 activity != ZPOOL_WAIT_TRIM)
9571 if (activity < 0 || activity >= ZPOOL_WAIT_NUM_ACTIVITIES)
9575 int error = spa_open(pool, &spa, FTAG);
9580 * Increment the spa's waiter count so that we can call spa_close and
9581 * still ensure that the spa_t doesn't get freed before this thread is
9582 * finished with it when the pool is exported. We want to call spa_close
9583 * before we start waiting because otherwise the additional ref would
9584 * prevent the pool from being exported or destroyed throughout the
9585 * potentially long wait.
9587 mutex_enter(&spa->spa_activities_lock);
9589 spa_close(spa, FTAG);
9593 boolean_t in_progress;
9594 error = spa_activity_in_progress(spa, activity, use_tag, tag,
9597 if (error || !in_progress || spa->spa_waiters_cancel)
9602 if (cv_wait_sig(&spa->spa_activities_cv,
9603 &spa->spa_activities_lock) == 0) {
9610 cv_signal(&spa->spa_waiters_cv);
9611 mutex_exit(&spa->spa_activities_lock);
9617 * Wait for a particular instance of the specified activity to complete, where
9618 * the instance is identified by 'tag'
9621 spa_wait_tag(const char *pool, zpool_wait_activity_t activity, uint64_t tag,
9624 return (spa_wait_common(pool, activity, B_TRUE, tag, waited));
9628 * Wait for all instances of the specified activity complete
9631 spa_wait(const char *pool, zpool_wait_activity_t activity, boolean_t *waited)
9634 return (spa_wait_common(pool, activity, B_FALSE, 0, waited));
9638 spa_event_create(spa_t *spa, vdev_t *vd, nvlist_t *hist_nvl, const char *name)
9640 sysevent_t *ev = NULL;
9644 resource = zfs_event_create(spa, vd, FM_SYSEVENT_CLASS, name, hist_nvl);
9646 ev = kmem_alloc(sizeof (sysevent_t), KM_SLEEP);
9647 ev->resource = resource;
9654 spa_event_post(sysevent_t *ev)
9658 zfs_zevent_post(ev->resource, NULL, zfs_zevent_post_cb);
9659 kmem_free(ev, sizeof (*ev));
9665 * Post a zevent corresponding to the given sysevent. The 'name' must be one
9666 * of the event definitions in sys/sysevent/eventdefs.h. The payload will be
9667 * filled in from the spa and (optionally) the vdev. This doesn't do anything
9668 * in the userland libzpool, as we don't want consumers to misinterpret ztest
9669 * or zdb as real changes.
9672 spa_event_notify(spa_t *spa, vdev_t *vd, nvlist_t *hist_nvl, const char *name)
9674 spa_event_post(spa_event_create(spa, vd, hist_nvl, name));
9677 /* state manipulation functions */
9678 EXPORT_SYMBOL(spa_open);
9679 EXPORT_SYMBOL(spa_open_rewind);
9680 EXPORT_SYMBOL(spa_get_stats);
9681 EXPORT_SYMBOL(spa_create);
9682 EXPORT_SYMBOL(spa_import);
9683 EXPORT_SYMBOL(spa_tryimport);
9684 EXPORT_SYMBOL(spa_destroy);
9685 EXPORT_SYMBOL(spa_export);
9686 EXPORT_SYMBOL(spa_reset);
9687 EXPORT_SYMBOL(spa_async_request);
9688 EXPORT_SYMBOL(spa_async_suspend);
9689 EXPORT_SYMBOL(spa_async_resume);
9690 EXPORT_SYMBOL(spa_inject_addref);
9691 EXPORT_SYMBOL(spa_inject_delref);
9692 EXPORT_SYMBOL(spa_scan_stat_init);
9693 EXPORT_SYMBOL(spa_scan_get_stats);
9695 /* device manipulation */
9696 EXPORT_SYMBOL(spa_vdev_add);
9697 EXPORT_SYMBOL(spa_vdev_attach);
9698 EXPORT_SYMBOL(spa_vdev_detach);
9699 EXPORT_SYMBOL(spa_vdev_setpath);
9700 EXPORT_SYMBOL(spa_vdev_setfru);
9701 EXPORT_SYMBOL(spa_vdev_split_mirror);
9703 /* spare statech is global across all pools) */
9704 EXPORT_SYMBOL(spa_spare_add);
9705 EXPORT_SYMBOL(spa_spare_remove);
9706 EXPORT_SYMBOL(spa_spare_exists);
9707 EXPORT_SYMBOL(spa_spare_activate);
9709 /* L2ARC statech is global across all pools) */
9710 EXPORT_SYMBOL(spa_l2cache_add);
9711 EXPORT_SYMBOL(spa_l2cache_remove);
9712 EXPORT_SYMBOL(spa_l2cache_exists);
9713 EXPORT_SYMBOL(spa_l2cache_activate);
9714 EXPORT_SYMBOL(spa_l2cache_drop);
9717 EXPORT_SYMBOL(spa_scan);
9718 EXPORT_SYMBOL(spa_scan_stop);
9721 EXPORT_SYMBOL(spa_sync); /* only for DMU use */
9722 EXPORT_SYMBOL(spa_sync_allpools);
9725 EXPORT_SYMBOL(spa_prop_set);
9726 EXPORT_SYMBOL(spa_prop_get);
9727 EXPORT_SYMBOL(spa_prop_clear_bootfs);
9729 /* asynchronous event notification */
9730 EXPORT_SYMBOL(spa_event_notify);
9733 ZFS_MODULE_PARAM(zfs_spa, spa_, load_verify_shift, INT, ZMOD_RW,
9734 "log2(fraction of arc that can be used by inflight I/Os when "
9735 "verifying pool during import");
9737 ZFS_MODULE_PARAM(zfs_spa, spa_, load_verify_metadata, INT, ZMOD_RW,
9738 "Set to traverse metadata on pool import");
9740 ZFS_MODULE_PARAM(zfs_spa, spa_, load_verify_data, INT, ZMOD_RW,
9741 "Set to traverse data on pool import");
9743 ZFS_MODULE_PARAM(zfs_spa, spa_, load_print_vdev_tree, INT, ZMOD_RW,
9744 "Print vdev tree to zfs_dbgmsg during pool import");
9746 ZFS_MODULE_PARAM(zfs_zio, zio_, taskq_batch_pct, UINT, ZMOD_RD,
9747 "Percentage of CPUs to run an IO worker thread");
9749 ZFS_MODULE_PARAM(zfs, zfs_, max_missing_tvds, ULONG, ZMOD_RW,
9750 "Allow importing pool with up to this number of missing top-level "
9751 "vdevs (in read-only mode)");
9753 ZFS_MODULE_PARAM(zfs_livelist_condense, zfs_livelist_condense_, zthr_pause, INT, ZMOD_RW,
9754 "Set the livelist condense zthr to pause");
9756 ZFS_MODULE_PARAM(zfs_livelist_condense, zfs_livelist_condense_, sync_pause, INT, ZMOD_RW,
9757 "Set the livelist condense synctask to pause");
9759 ZFS_MODULE_PARAM(zfs_livelist_condense, zfs_livelist_condense_, sync_cancel, INT, ZMOD_RW,
9760 "Whether livelist condensing was canceled in the synctask");
9762 ZFS_MODULE_PARAM(zfs_livelist_condense, zfs_livelist_condense_, zthr_cancel, INT, ZMOD_RW,
9763 "Whether livelist condensing was canceled in the zthr function");
9765 ZFS_MODULE_PARAM(zfs_livelist_condense, zfs_livelist_condense_, new_alloc, INT, ZMOD_RW,
9766 "Whether extra ALLOC blkptrs were added to a livelist entry while it "
9767 "was being condensed");