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, 2019 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. Under Linux this
1004 * means incrementing the priority value on platforms
1005 * like illumos it should be decremented.
1007 if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE)
1010 tq = taskq_create_proc(name, value, pri, 50,
1011 INT_MAX, spa->spa_proc, flags);
1014 tqs->stqs_taskq[i] = tq;
1019 spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
1021 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
1023 if (tqs->stqs_taskq == NULL) {
1024 ASSERT3U(tqs->stqs_count, ==, 0);
1028 for (uint_t i = 0; i < tqs->stqs_count; i++) {
1029 ASSERT3P(tqs->stqs_taskq[i], !=, NULL);
1030 taskq_destroy(tqs->stqs_taskq[i]);
1033 kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *));
1034 tqs->stqs_taskq = NULL;
1038 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
1039 * Note that a type may have multiple discrete taskqs to avoid lock contention
1040 * on the taskq itself. In that case we choose which taskq at random by using
1041 * the low bits of gethrtime().
1044 spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q,
1045 task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent)
1047 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
1050 ASSERT3P(tqs->stqs_taskq, !=, NULL);
1051 ASSERT3U(tqs->stqs_count, !=, 0);
1053 if (tqs->stqs_count == 1) {
1054 tq = tqs->stqs_taskq[0];
1056 tq = tqs->stqs_taskq[((uint64_t)gethrtime()) % tqs->stqs_count];
1059 taskq_dispatch_ent(tq, func, arg, flags, ent);
1063 * Same as spa_taskq_dispatch_ent() but block on the task until completion.
1066 spa_taskq_dispatch_sync(spa_t *spa, zio_type_t t, zio_taskq_type_t q,
1067 task_func_t *func, void *arg, uint_t flags)
1069 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
1073 ASSERT3P(tqs->stqs_taskq, !=, NULL);
1074 ASSERT3U(tqs->stqs_count, !=, 0);
1076 if (tqs->stqs_count == 1) {
1077 tq = tqs->stqs_taskq[0];
1079 tq = tqs->stqs_taskq[((uint64_t)gethrtime()) % tqs->stqs_count];
1082 id = taskq_dispatch(tq, func, arg, flags);
1084 taskq_wait_id(tq, id);
1088 spa_create_zio_taskqs(spa_t *spa)
1090 for (int t = 0; t < ZIO_TYPES; t++) {
1091 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1092 spa_taskqs_init(spa, t, q);
1098 * Disabled until spa_thread() can be adapted for Linux.
1100 #undef HAVE_SPA_THREAD
1102 #if defined(_KERNEL) && defined(HAVE_SPA_THREAD)
1104 spa_thread(void *arg)
1106 psetid_t zio_taskq_psrset_bind = PS_NONE;
1107 callb_cpr_t cprinfo;
1110 user_t *pu = PTOU(curproc);
1112 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
1115 ASSERT(curproc != &p0);
1116 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
1117 "zpool-%s", spa->spa_name);
1118 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
1120 /* bind this thread to the requested psrset */
1121 if (zio_taskq_psrset_bind != PS_NONE) {
1123 mutex_enter(&cpu_lock);
1124 mutex_enter(&pidlock);
1125 mutex_enter(&curproc->p_lock);
1127 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
1128 0, NULL, NULL) == 0) {
1129 curthread->t_bind_pset = zio_taskq_psrset_bind;
1132 "Couldn't bind process for zfs pool \"%s\" to "
1133 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
1136 mutex_exit(&curproc->p_lock);
1137 mutex_exit(&pidlock);
1138 mutex_exit(&cpu_lock);
1142 if (zio_taskq_sysdc) {
1143 sysdc_thread_enter(curthread, 100, 0);
1146 spa->spa_proc = curproc;
1147 spa->spa_did = curthread->t_did;
1149 spa_create_zio_taskqs(spa);
1151 mutex_enter(&spa->spa_proc_lock);
1152 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
1154 spa->spa_proc_state = SPA_PROC_ACTIVE;
1155 cv_broadcast(&spa->spa_proc_cv);
1157 CALLB_CPR_SAFE_BEGIN(&cprinfo);
1158 while (spa->spa_proc_state == SPA_PROC_ACTIVE)
1159 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1160 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
1162 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
1163 spa->spa_proc_state = SPA_PROC_GONE;
1164 spa->spa_proc = &p0;
1165 cv_broadcast(&spa->spa_proc_cv);
1166 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */
1168 mutex_enter(&curproc->p_lock);
1174 * Activate an uninitialized pool.
1177 spa_activate(spa_t *spa, spa_mode_t mode)
1179 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
1181 spa->spa_state = POOL_STATE_ACTIVE;
1182 spa->spa_mode = mode;
1184 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
1185 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
1186 spa->spa_special_class = metaslab_class_create(spa, zfs_metaslab_ops);
1187 spa->spa_dedup_class = metaslab_class_create(spa, zfs_metaslab_ops);
1189 /* Try to create a covering process */
1190 mutex_enter(&spa->spa_proc_lock);
1191 ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
1192 ASSERT(spa->spa_proc == &p0);
1195 #ifdef HAVE_SPA_THREAD
1196 /* Only create a process if we're going to be around a while. */
1197 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
1198 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
1200 spa->spa_proc_state = SPA_PROC_CREATED;
1201 while (spa->spa_proc_state == SPA_PROC_CREATED) {
1202 cv_wait(&spa->spa_proc_cv,
1203 &spa->spa_proc_lock);
1205 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1206 ASSERT(spa->spa_proc != &p0);
1207 ASSERT(spa->spa_did != 0);
1211 "Couldn't create process for zfs pool \"%s\"\n",
1216 #endif /* HAVE_SPA_THREAD */
1217 mutex_exit(&spa->spa_proc_lock);
1219 /* If we didn't create a process, we need to create our taskqs. */
1220 if (spa->spa_proc == &p0) {
1221 spa_create_zio_taskqs(spa);
1224 for (size_t i = 0; i < TXG_SIZE; i++) {
1225 spa->spa_txg_zio[i] = zio_root(spa, NULL, NULL,
1229 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
1230 offsetof(vdev_t, vdev_config_dirty_node));
1231 list_create(&spa->spa_evicting_os_list, sizeof (objset_t),
1232 offsetof(objset_t, os_evicting_node));
1233 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
1234 offsetof(vdev_t, vdev_state_dirty_node));
1236 txg_list_create(&spa->spa_vdev_txg_list, spa,
1237 offsetof(struct vdev, vdev_txg_node));
1239 avl_create(&spa->spa_errlist_scrub,
1240 spa_error_entry_compare, sizeof (spa_error_entry_t),
1241 offsetof(spa_error_entry_t, se_avl));
1242 avl_create(&spa->spa_errlist_last,
1243 spa_error_entry_compare, sizeof (spa_error_entry_t),
1244 offsetof(spa_error_entry_t, se_avl));
1246 spa_keystore_init(&spa->spa_keystore);
1249 * This taskq is used to perform zvol-minor-related tasks
1250 * asynchronously. This has several advantages, including easy
1251 * resolution of various deadlocks (zfsonlinux bug #3681).
1253 * The taskq must be single threaded to ensure tasks are always
1254 * processed in the order in which they were dispatched.
1256 * A taskq per pool allows one to keep the pools independent.
1257 * This way if one pool is suspended, it will not impact another.
1259 * The preferred location to dispatch a zvol minor task is a sync
1260 * task. In this context, there is easy access to the spa_t and minimal
1261 * error handling is required because the sync task must succeed.
1263 spa->spa_zvol_taskq = taskq_create("z_zvol", 1, defclsyspri,
1267 * Taskq dedicated to prefetcher threads: this is used to prevent the
1268 * pool traverse code from monopolizing the global (and limited)
1269 * system_taskq by inappropriately scheduling long running tasks on it.
1271 spa->spa_prefetch_taskq = taskq_create("z_prefetch", boot_ncpus,
1272 defclsyspri, 1, INT_MAX, TASKQ_DYNAMIC);
1275 * The taskq to upgrade datasets in this pool. Currently used by
1276 * feature SPA_FEATURE_USEROBJ_ACCOUNTING/SPA_FEATURE_PROJECT_QUOTA.
1278 spa->spa_upgrade_taskq = taskq_create("z_upgrade", boot_ncpus,
1279 defclsyspri, 1, INT_MAX, TASKQ_DYNAMIC);
1283 * Opposite of spa_activate().
1286 spa_deactivate(spa_t *spa)
1288 ASSERT(spa->spa_sync_on == B_FALSE);
1289 ASSERT(spa->spa_dsl_pool == NULL);
1290 ASSERT(spa->spa_root_vdev == NULL);
1291 ASSERT(spa->spa_async_zio_root == NULL);
1292 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
1294 spa_evicting_os_wait(spa);
1296 if (spa->spa_zvol_taskq) {
1297 taskq_destroy(spa->spa_zvol_taskq);
1298 spa->spa_zvol_taskq = NULL;
1301 if (spa->spa_prefetch_taskq) {
1302 taskq_destroy(spa->spa_prefetch_taskq);
1303 spa->spa_prefetch_taskq = NULL;
1306 if (spa->spa_upgrade_taskq) {
1307 taskq_destroy(spa->spa_upgrade_taskq);
1308 spa->spa_upgrade_taskq = NULL;
1311 txg_list_destroy(&spa->spa_vdev_txg_list);
1313 list_destroy(&spa->spa_config_dirty_list);
1314 list_destroy(&spa->spa_evicting_os_list);
1315 list_destroy(&spa->spa_state_dirty_list);
1317 taskq_cancel_id(system_delay_taskq, spa->spa_deadman_tqid);
1319 for (int t = 0; t < ZIO_TYPES; t++) {
1320 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1321 spa_taskqs_fini(spa, t, q);
1325 for (size_t i = 0; i < TXG_SIZE; i++) {
1326 ASSERT3P(spa->spa_txg_zio[i], !=, NULL);
1327 VERIFY0(zio_wait(spa->spa_txg_zio[i]));
1328 spa->spa_txg_zio[i] = NULL;
1331 metaslab_class_destroy(spa->spa_normal_class);
1332 spa->spa_normal_class = NULL;
1334 metaslab_class_destroy(spa->spa_log_class);
1335 spa->spa_log_class = NULL;
1337 metaslab_class_destroy(spa->spa_special_class);
1338 spa->spa_special_class = NULL;
1340 metaslab_class_destroy(spa->spa_dedup_class);
1341 spa->spa_dedup_class = NULL;
1344 * If this was part of an import or the open otherwise failed, we may
1345 * still have errors left in the queues. Empty them just in case.
1347 spa_errlog_drain(spa);
1348 avl_destroy(&spa->spa_errlist_scrub);
1349 avl_destroy(&spa->spa_errlist_last);
1351 spa_keystore_fini(&spa->spa_keystore);
1353 spa->spa_state = POOL_STATE_UNINITIALIZED;
1355 mutex_enter(&spa->spa_proc_lock);
1356 if (spa->spa_proc_state != SPA_PROC_NONE) {
1357 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1358 spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1359 cv_broadcast(&spa->spa_proc_cv);
1360 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1361 ASSERT(spa->spa_proc != &p0);
1362 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1364 ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1365 spa->spa_proc_state = SPA_PROC_NONE;
1367 ASSERT(spa->spa_proc == &p0);
1368 mutex_exit(&spa->spa_proc_lock);
1371 * We want to make sure spa_thread() has actually exited the ZFS
1372 * module, so that the module can't be unloaded out from underneath
1375 if (spa->spa_did != 0) {
1376 thread_join(spa->spa_did);
1382 * Verify a pool configuration, and construct the vdev tree appropriately. This
1383 * will create all the necessary vdevs in the appropriate layout, with each vdev
1384 * in the CLOSED state. This will prep the pool before open/creation/import.
1385 * All vdev validation is done by the vdev_alloc() routine.
1388 spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1389 uint_t id, int atype)
1395 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1398 if ((*vdp)->vdev_ops->vdev_op_leaf)
1401 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1404 if (error == ENOENT)
1410 return (SET_ERROR(EINVAL));
1413 for (int c = 0; c < children; c++) {
1415 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1423 ASSERT(*vdp != NULL);
1429 spa_should_flush_logs_on_unload(spa_t *spa)
1431 if (!spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP))
1434 if (!spa_writeable(spa))
1437 if (!spa->spa_sync_on)
1440 if (spa_state(spa) != POOL_STATE_EXPORTED)
1443 if (zfs_keep_log_spacemaps_at_export)
1450 * Opens a transaction that will set the flag that will instruct
1451 * spa_sync to attempt to flush all the metaslabs for that txg.
1454 spa_unload_log_sm_flush_all(spa_t *spa)
1456 dmu_tx_t *tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
1457 VERIFY0(dmu_tx_assign(tx, TXG_WAIT));
1459 ASSERT3U(spa->spa_log_flushall_txg, ==, 0);
1460 spa->spa_log_flushall_txg = dmu_tx_get_txg(tx);
1463 txg_wait_synced(spa_get_dsl(spa), spa->spa_log_flushall_txg);
1467 spa_unload_log_sm_metadata(spa_t *spa)
1469 void *cookie = NULL;
1471 while ((sls = avl_destroy_nodes(&spa->spa_sm_logs_by_txg,
1472 &cookie)) != NULL) {
1473 VERIFY0(sls->sls_mscount);
1474 kmem_free(sls, sizeof (spa_log_sm_t));
1477 for (log_summary_entry_t *e = list_head(&spa->spa_log_summary);
1478 e != NULL; e = list_head(&spa->spa_log_summary)) {
1479 VERIFY0(e->lse_mscount);
1480 list_remove(&spa->spa_log_summary, e);
1481 kmem_free(e, sizeof (log_summary_entry_t));
1484 spa->spa_unflushed_stats.sus_nblocks = 0;
1485 spa->spa_unflushed_stats.sus_memused = 0;
1486 spa->spa_unflushed_stats.sus_blocklimit = 0;
1490 spa_destroy_aux_threads(spa_t *spa)
1492 if (spa->spa_condense_zthr != NULL) {
1493 zthr_destroy(spa->spa_condense_zthr);
1494 spa->spa_condense_zthr = NULL;
1496 if (spa->spa_checkpoint_discard_zthr != NULL) {
1497 zthr_destroy(spa->spa_checkpoint_discard_zthr);
1498 spa->spa_checkpoint_discard_zthr = NULL;
1500 if (spa->spa_livelist_delete_zthr != NULL) {
1501 zthr_destroy(spa->spa_livelist_delete_zthr);
1502 spa->spa_livelist_delete_zthr = NULL;
1504 if (spa->spa_livelist_condense_zthr != NULL) {
1505 zthr_destroy(spa->spa_livelist_condense_zthr);
1506 spa->spa_livelist_condense_zthr = NULL;
1511 * Opposite of spa_load().
1514 spa_unload(spa_t *spa)
1516 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1517 ASSERT(spa_state(spa) != POOL_STATE_UNINITIALIZED);
1519 spa_import_progress_remove(spa_guid(spa));
1520 spa_load_note(spa, "UNLOADING");
1522 spa_wake_waiters(spa);
1525 * If the log space map feature is enabled and the pool is getting
1526 * exported (but not destroyed), we want to spend some time flushing
1527 * as many metaslabs as we can in an attempt to destroy log space
1528 * maps and save import time.
1530 if (spa_should_flush_logs_on_unload(spa))
1531 spa_unload_log_sm_flush_all(spa);
1536 spa_async_suspend(spa);
1538 if (spa->spa_root_vdev) {
1539 vdev_t *root_vdev = spa->spa_root_vdev;
1540 vdev_initialize_stop_all(root_vdev, VDEV_INITIALIZE_ACTIVE);
1541 vdev_trim_stop_all(root_vdev, VDEV_TRIM_ACTIVE);
1542 vdev_autotrim_stop_all(spa);
1543 vdev_rebuild_stop_all(spa);
1549 if (spa->spa_sync_on) {
1550 txg_sync_stop(spa->spa_dsl_pool);
1551 spa->spa_sync_on = B_FALSE;
1555 * This ensures that there is no async metaslab prefetching
1556 * while we attempt to unload the spa.
1558 if (spa->spa_root_vdev != NULL) {
1559 for (int c = 0; c < spa->spa_root_vdev->vdev_children; c++) {
1560 vdev_t *vc = spa->spa_root_vdev->vdev_child[c];
1561 if (vc->vdev_mg != NULL)
1562 taskq_wait(vc->vdev_mg->mg_taskq);
1566 if (spa->spa_mmp.mmp_thread)
1567 mmp_thread_stop(spa);
1570 * Wait for any outstanding async I/O to complete.
1572 if (spa->spa_async_zio_root != NULL) {
1573 for (int i = 0; i < max_ncpus; i++)
1574 (void) zio_wait(spa->spa_async_zio_root[i]);
1575 kmem_free(spa->spa_async_zio_root, max_ncpus * sizeof (void *));
1576 spa->spa_async_zio_root = NULL;
1579 if (spa->spa_vdev_removal != NULL) {
1580 spa_vdev_removal_destroy(spa->spa_vdev_removal);
1581 spa->spa_vdev_removal = NULL;
1584 spa_destroy_aux_threads(spa);
1586 spa_condense_fini(spa);
1588 bpobj_close(&spa->spa_deferred_bpobj);
1590 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
1595 if (spa->spa_root_vdev)
1596 vdev_free(spa->spa_root_vdev);
1597 ASSERT(spa->spa_root_vdev == NULL);
1600 * Close the dsl pool.
1602 if (spa->spa_dsl_pool) {
1603 dsl_pool_close(spa->spa_dsl_pool);
1604 spa->spa_dsl_pool = NULL;
1605 spa->spa_meta_objset = NULL;
1609 spa_unload_log_sm_metadata(spa);
1612 * Drop and purge level 2 cache
1614 spa_l2cache_drop(spa);
1616 for (int i = 0; i < spa->spa_spares.sav_count; i++)
1617 vdev_free(spa->spa_spares.sav_vdevs[i]);
1618 if (spa->spa_spares.sav_vdevs) {
1619 kmem_free(spa->spa_spares.sav_vdevs,
1620 spa->spa_spares.sav_count * sizeof (void *));
1621 spa->spa_spares.sav_vdevs = NULL;
1623 if (spa->spa_spares.sav_config) {
1624 nvlist_free(spa->spa_spares.sav_config);
1625 spa->spa_spares.sav_config = NULL;
1627 spa->spa_spares.sav_count = 0;
1629 for (int i = 0; i < spa->spa_l2cache.sav_count; i++) {
1630 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1631 vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1633 if (spa->spa_l2cache.sav_vdevs) {
1634 kmem_free(spa->spa_l2cache.sav_vdevs,
1635 spa->spa_l2cache.sav_count * sizeof (void *));
1636 spa->spa_l2cache.sav_vdevs = NULL;
1638 if (spa->spa_l2cache.sav_config) {
1639 nvlist_free(spa->spa_l2cache.sav_config);
1640 spa->spa_l2cache.sav_config = NULL;
1642 spa->spa_l2cache.sav_count = 0;
1644 spa->spa_async_suspended = 0;
1646 spa->spa_indirect_vdevs_loaded = B_FALSE;
1648 if (spa->spa_comment != NULL) {
1649 spa_strfree(spa->spa_comment);
1650 spa->spa_comment = NULL;
1653 spa_config_exit(spa, SCL_ALL, spa);
1657 * Load (or re-load) the current list of vdevs describing the active spares for
1658 * this pool. When this is called, we have some form of basic information in
1659 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1660 * then re-generate a more complete list including status information.
1663 spa_load_spares(spa_t *spa)
1672 * zdb opens both the current state of the pool and the
1673 * checkpointed state (if present), with a different spa_t.
1675 * As spare vdevs are shared among open pools, we skip loading
1676 * them when we load the checkpointed state of the pool.
1678 if (!spa_writeable(spa))
1682 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1685 * First, close and free any existing spare vdevs.
1687 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1688 vd = spa->spa_spares.sav_vdevs[i];
1690 /* Undo the call to spa_activate() below */
1691 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1692 B_FALSE)) != NULL && tvd->vdev_isspare)
1693 spa_spare_remove(tvd);
1698 if (spa->spa_spares.sav_vdevs)
1699 kmem_free(spa->spa_spares.sav_vdevs,
1700 spa->spa_spares.sav_count * sizeof (void *));
1702 if (spa->spa_spares.sav_config == NULL)
1705 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1706 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1708 spa->spa_spares.sav_count = (int)nspares;
1709 spa->spa_spares.sav_vdevs = NULL;
1715 * Construct the array of vdevs, opening them to get status in the
1716 * process. For each spare, there is potentially two different vdev_t
1717 * structures associated with it: one in the list of spares (used only
1718 * for basic validation purposes) and one in the active vdev
1719 * configuration (if it's spared in). During this phase we open and
1720 * validate each vdev on the spare list. If the vdev also exists in the
1721 * active configuration, then we also mark this vdev as an active spare.
1723 spa->spa_spares.sav_vdevs = kmem_zalloc(nspares * sizeof (void *),
1725 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1726 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1727 VDEV_ALLOC_SPARE) == 0);
1730 spa->spa_spares.sav_vdevs[i] = vd;
1732 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1733 B_FALSE)) != NULL) {
1734 if (!tvd->vdev_isspare)
1738 * We only mark the spare active if we were successfully
1739 * able to load the vdev. Otherwise, importing a pool
1740 * with a bad active spare would result in strange
1741 * behavior, because multiple pool would think the spare
1742 * is actively in use.
1744 * There is a vulnerability here to an equally bizarre
1745 * circumstance, where a dead active spare is later
1746 * brought back to life (onlined or otherwise). Given
1747 * the rarity of this scenario, and the extra complexity
1748 * it adds, we ignore the possibility.
1750 if (!vdev_is_dead(tvd))
1751 spa_spare_activate(tvd);
1755 vd->vdev_aux = &spa->spa_spares;
1757 if (vdev_open(vd) != 0)
1760 if (vdev_validate_aux(vd) == 0)
1765 * Recompute the stashed list of spares, with status information
1768 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1769 DATA_TYPE_NVLIST_ARRAY) == 0);
1771 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1773 for (i = 0; i < spa->spa_spares.sav_count; i++)
1774 spares[i] = vdev_config_generate(spa,
1775 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1776 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1777 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1778 for (i = 0; i < spa->spa_spares.sav_count; i++)
1779 nvlist_free(spares[i]);
1780 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1784 * Load (or re-load) the current list of vdevs describing the active l2cache for
1785 * this pool. When this is called, we have some form of basic information in
1786 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1787 * then re-generate a more complete list including status information.
1788 * Devices which are already active have their details maintained, and are
1792 spa_load_l2cache(spa_t *spa)
1794 nvlist_t **l2cache = NULL;
1796 int i, j, oldnvdevs;
1798 vdev_t *vd, **oldvdevs, **newvdevs;
1799 spa_aux_vdev_t *sav = &spa->spa_l2cache;
1803 * zdb opens both the current state of the pool and the
1804 * checkpointed state (if present), with a different spa_t.
1806 * As L2 caches are part of the ARC which is shared among open
1807 * pools, we skip loading them when we load the checkpointed
1808 * state of the pool.
1810 if (!spa_writeable(spa))
1814 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1816 oldvdevs = sav->sav_vdevs;
1817 oldnvdevs = sav->sav_count;
1818 sav->sav_vdevs = NULL;
1821 if (sav->sav_config == NULL) {
1827 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1828 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1829 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1832 * Process new nvlist of vdevs.
1834 for (i = 0; i < nl2cache; i++) {
1835 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1839 for (j = 0; j < oldnvdevs; j++) {
1841 if (vd != NULL && guid == vd->vdev_guid) {
1843 * Retain previous vdev for add/remove ops.
1851 if (newvdevs[i] == NULL) {
1855 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1856 VDEV_ALLOC_L2CACHE) == 0);
1861 * Commit this vdev as an l2cache device,
1862 * even if it fails to open.
1864 spa_l2cache_add(vd);
1869 spa_l2cache_activate(vd);
1871 if (vdev_open(vd) != 0)
1874 (void) vdev_validate_aux(vd);
1876 if (!vdev_is_dead(vd))
1877 l2arc_add_vdev(spa, vd);
1880 * Upon cache device addition to a pool or pool
1881 * creation with a cache device or if the header
1882 * of the device is invalid we issue an async
1883 * TRIM command for the whole device which will
1884 * execute if l2arc_trim_ahead > 0.
1886 spa_async_request(spa, SPA_ASYNC_L2CACHE_TRIM);
1890 sav->sav_vdevs = newvdevs;
1891 sav->sav_count = (int)nl2cache;
1894 * Recompute the stashed list of l2cache devices, with status
1895 * information this time.
1897 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1898 DATA_TYPE_NVLIST_ARRAY) == 0);
1900 if (sav->sav_count > 0)
1901 l2cache = kmem_alloc(sav->sav_count * sizeof (void *),
1903 for (i = 0; i < sav->sav_count; i++)
1904 l2cache[i] = vdev_config_generate(spa,
1905 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1906 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1907 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1911 * Purge vdevs that were dropped
1913 for (i = 0; i < oldnvdevs; i++) {
1918 ASSERT(vd->vdev_isl2cache);
1920 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1921 pool != 0ULL && l2arc_vdev_present(vd))
1922 l2arc_remove_vdev(vd);
1923 vdev_clear_stats(vd);
1929 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1931 for (i = 0; i < sav->sav_count; i++)
1932 nvlist_free(l2cache[i]);
1934 kmem_free(l2cache, sav->sav_count * sizeof (void *));
1938 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1941 char *packed = NULL;
1946 error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db);
1950 nvsize = *(uint64_t *)db->db_data;
1951 dmu_buf_rele(db, FTAG);
1953 packed = vmem_alloc(nvsize, KM_SLEEP);
1954 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1957 error = nvlist_unpack(packed, nvsize, value, 0);
1958 vmem_free(packed, nvsize);
1964 * Concrete top-level vdevs that are not missing and are not logs. At every
1965 * spa_sync we write new uberblocks to at least SPA_SYNC_MIN_VDEVS core tvds.
1968 spa_healthy_core_tvds(spa_t *spa)
1970 vdev_t *rvd = spa->spa_root_vdev;
1973 for (uint64_t i = 0; i < rvd->vdev_children; i++) {
1974 vdev_t *vd = rvd->vdev_child[i];
1977 if (vdev_is_concrete(vd) && !vdev_is_dead(vd))
1985 * Checks to see if the given vdev could not be opened, in which case we post a
1986 * sysevent to notify the autoreplace code that the device has been removed.
1989 spa_check_removed(vdev_t *vd)
1991 for (uint64_t c = 0; c < vd->vdev_children; c++)
1992 spa_check_removed(vd->vdev_child[c]);
1994 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) &&
1995 vdev_is_concrete(vd)) {
1996 zfs_post_autoreplace(vd->vdev_spa, vd);
1997 spa_event_notify(vd->vdev_spa, vd, NULL, ESC_ZFS_VDEV_CHECK);
2002 spa_check_for_missing_logs(spa_t *spa)
2004 vdev_t *rvd = spa->spa_root_vdev;
2007 * If we're doing a normal import, then build up any additional
2008 * diagnostic information about missing log devices.
2009 * We'll pass this up to the user for further processing.
2011 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
2012 nvlist_t **child, *nv;
2015 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t *),
2017 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2019 for (uint64_t c = 0; c < rvd->vdev_children; c++) {
2020 vdev_t *tvd = rvd->vdev_child[c];
2023 * We consider a device as missing only if it failed
2024 * to open (i.e. offline or faulted is not considered
2027 if (tvd->vdev_islog &&
2028 tvd->vdev_state == VDEV_STATE_CANT_OPEN) {
2029 child[idx++] = vdev_config_generate(spa, tvd,
2030 B_FALSE, VDEV_CONFIG_MISSING);
2035 fnvlist_add_nvlist_array(nv,
2036 ZPOOL_CONFIG_CHILDREN, child, idx);
2037 fnvlist_add_nvlist(spa->spa_load_info,
2038 ZPOOL_CONFIG_MISSING_DEVICES, nv);
2040 for (uint64_t i = 0; i < idx; i++)
2041 nvlist_free(child[i]);
2044 kmem_free(child, rvd->vdev_children * sizeof (char **));
2047 spa_load_failed(spa, "some log devices are missing");
2048 vdev_dbgmsg_print_tree(rvd, 2);
2049 return (SET_ERROR(ENXIO));
2052 for (uint64_t c = 0; c < rvd->vdev_children; c++) {
2053 vdev_t *tvd = rvd->vdev_child[c];
2055 if (tvd->vdev_islog &&
2056 tvd->vdev_state == VDEV_STATE_CANT_OPEN) {
2057 spa_set_log_state(spa, SPA_LOG_CLEAR);
2058 spa_load_note(spa, "some log devices are "
2059 "missing, ZIL is dropped.");
2060 vdev_dbgmsg_print_tree(rvd, 2);
2070 * Check for missing log devices
2073 spa_check_logs(spa_t *spa)
2075 boolean_t rv = B_FALSE;
2076 dsl_pool_t *dp = spa_get_dsl(spa);
2078 switch (spa->spa_log_state) {
2081 case SPA_LOG_MISSING:
2082 /* need to recheck in case slog has been restored */
2083 case SPA_LOG_UNKNOWN:
2084 rv = (dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
2085 zil_check_log_chain, NULL, DS_FIND_CHILDREN) != 0);
2087 spa_set_log_state(spa, SPA_LOG_MISSING);
2094 spa_passivate_log(spa_t *spa)
2096 vdev_t *rvd = spa->spa_root_vdev;
2097 boolean_t slog_found = B_FALSE;
2099 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
2101 if (!spa_has_slogs(spa))
2104 for (int c = 0; c < rvd->vdev_children; c++) {
2105 vdev_t *tvd = rvd->vdev_child[c];
2106 metaslab_group_t *mg = tvd->vdev_mg;
2108 if (tvd->vdev_islog) {
2109 metaslab_group_passivate(mg);
2110 slog_found = B_TRUE;
2114 return (slog_found);
2118 spa_activate_log(spa_t *spa)
2120 vdev_t *rvd = spa->spa_root_vdev;
2122 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
2124 for (int c = 0; c < rvd->vdev_children; c++) {
2125 vdev_t *tvd = rvd->vdev_child[c];
2126 metaslab_group_t *mg = tvd->vdev_mg;
2128 if (tvd->vdev_islog)
2129 metaslab_group_activate(mg);
2134 spa_reset_logs(spa_t *spa)
2138 error = dmu_objset_find(spa_name(spa), zil_reset,
2139 NULL, DS_FIND_CHILDREN);
2142 * We successfully offlined the log device, sync out the
2143 * current txg so that the "stubby" block can be removed
2146 txg_wait_synced(spa->spa_dsl_pool, 0);
2152 spa_aux_check_removed(spa_aux_vdev_t *sav)
2154 for (int i = 0; i < sav->sav_count; i++)
2155 spa_check_removed(sav->sav_vdevs[i]);
2159 spa_claim_notify(zio_t *zio)
2161 spa_t *spa = zio->io_spa;
2166 mutex_enter(&spa->spa_props_lock); /* any mutex will do */
2167 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
2168 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
2169 mutex_exit(&spa->spa_props_lock);
2172 typedef struct spa_load_error {
2173 uint64_t sle_meta_count;
2174 uint64_t sle_data_count;
2178 spa_load_verify_done(zio_t *zio)
2180 blkptr_t *bp = zio->io_bp;
2181 spa_load_error_t *sle = zio->io_private;
2182 dmu_object_type_t type = BP_GET_TYPE(bp);
2183 int error = zio->io_error;
2184 spa_t *spa = zio->io_spa;
2186 abd_free(zio->io_abd);
2188 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
2189 type != DMU_OT_INTENT_LOG)
2190 atomic_inc_64(&sle->sle_meta_count);
2192 atomic_inc_64(&sle->sle_data_count);
2195 mutex_enter(&spa->spa_scrub_lock);
2196 spa->spa_load_verify_bytes -= BP_GET_PSIZE(bp);
2197 cv_broadcast(&spa->spa_scrub_io_cv);
2198 mutex_exit(&spa->spa_scrub_lock);
2202 * Maximum number of inflight bytes is the log2 fraction of the arc size.
2203 * By default, we set it to 1/16th of the arc.
2205 int spa_load_verify_shift = 4;
2206 int spa_load_verify_metadata = B_TRUE;
2207 int spa_load_verify_data = B_TRUE;
2211 spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
2212 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
2214 if (zb->zb_level == ZB_DNODE_LEVEL || BP_IS_HOLE(bp) ||
2215 BP_IS_EMBEDDED(bp) || BP_IS_REDACTED(bp))
2218 * Note: normally this routine will not be called if
2219 * spa_load_verify_metadata is not set. However, it may be useful
2220 * to manually set the flag after the traversal has begun.
2222 if (!spa_load_verify_metadata)
2224 if (!BP_IS_METADATA(bp) && !spa_load_verify_data)
2227 uint64_t maxinflight_bytes =
2228 arc_target_bytes() >> spa_load_verify_shift;
2230 size_t size = BP_GET_PSIZE(bp);
2232 mutex_enter(&spa->spa_scrub_lock);
2233 while (spa->spa_load_verify_bytes >= maxinflight_bytes)
2234 cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
2235 spa->spa_load_verify_bytes += size;
2236 mutex_exit(&spa->spa_scrub_lock);
2238 zio_nowait(zio_read(rio, spa, bp, abd_alloc_for_io(size, B_FALSE), size,
2239 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
2240 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
2241 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
2247 verify_dataset_name_len(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg)
2249 if (dsl_dataset_namelen(ds) >= ZFS_MAX_DATASET_NAME_LEN)
2250 return (SET_ERROR(ENAMETOOLONG));
2256 spa_load_verify(spa_t *spa)
2259 spa_load_error_t sle = { 0 };
2260 zpool_load_policy_t policy;
2261 boolean_t verify_ok = B_FALSE;
2264 zpool_get_load_policy(spa->spa_config, &policy);
2266 if (policy.zlp_rewind & ZPOOL_NEVER_REWIND)
2269 dsl_pool_config_enter(spa->spa_dsl_pool, FTAG);
2270 error = dmu_objset_find_dp(spa->spa_dsl_pool,
2271 spa->spa_dsl_pool->dp_root_dir_obj, verify_dataset_name_len, NULL,
2273 dsl_pool_config_exit(spa->spa_dsl_pool, FTAG);
2277 rio = zio_root(spa, NULL, &sle,
2278 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
2280 if (spa_load_verify_metadata) {
2281 if (spa->spa_extreme_rewind) {
2282 spa_load_note(spa, "performing a complete scan of the "
2283 "pool since extreme rewind is on. This may take "
2284 "a very long time.\n (spa_load_verify_data=%u, "
2285 "spa_load_verify_metadata=%u)",
2286 spa_load_verify_data, spa_load_verify_metadata);
2289 error = traverse_pool(spa, spa->spa_verify_min_txg,
2290 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA |
2291 TRAVERSE_NO_DECRYPT, spa_load_verify_cb, rio);
2294 (void) zio_wait(rio);
2295 ASSERT0(spa->spa_load_verify_bytes);
2297 spa->spa_load_meta_errors = sle.sle_meta_count;
2298 spa->spa_load_data_errors = sle.sle_data_count;
2300 if (sle.sle_meta_count != 0 || sle.sle_data_count != 0) {
2301 spa_load_note(spa, "spa_load_verify found %llu metadata errors "
2302 "and %llu data errors", (u_longlong_t)sle.sle_meta_count,
2303 (u_longlong_t)sle.sle_data_count);
2306 if (spa_load_verify_dryrun ||
2307 (!error && sle.sle_meta_count <= policy.zlp_maxmeta &&
2308 sle.sle_data_count <= policy.zlp_maxdata)) {
2312 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
2313 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
2315 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
2316 VERIFY(nvlist_add_uint64(spa->spa_load_info,
2317 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
2318 VERIFY(nvlist_add_int64(spa->spa_load_info,
2319 ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
2320 VERIFY(nvlist_add_uint64(spa->spa_load_info,
2321 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
2323 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
2326 if (spa_load_verify_dryrun)
2330 if (error != ENXIO && error != EIO)
2331 error = SET_ERROR(EIO);
2335 return (verify_ok ? 0 : EIO);
2339 * Find a value in the pool props object.
2342 spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
2344 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
2345 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
2349 * Find a value in the pool directory object.
2352 spa_dir_prop(spa_t *spa, const char *name, uint64_t *val, boolean_t log_enoent)
2354 int error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
2355 name, sizeof (uint64_t), 1, val);
2357 if (error != 0 && (error != ENOENT || log_enoent)) {
2358 spa_load_failed(spa, "couldn't get '%s' value in MOS directory "
2359 "[error=%d]", name, error);
2366 spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
2368 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
2369 return (SET_ERROR(err));
2373 spa_livelist_delete_check(spa_t *spa)
2375 return (spa->spa_livelists_to_delete != 0);
2380 spa_livelist_delete_cb_check(void *arg, zthr_t *z)
2383 return (spa_livelist_delete_check(spa));
2387 delete_blkptr_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
2390 zio_free(spa, tx->tx_txg, bp);
2391 dsl_dir_diduse_space(tx->tx_pool->dp_free_dir, DD_USED_HEAD,
2392 -bp_get_dsize_sync(spa, bp),
2393 -BP_GET_PSIZE(bp), -BP_GET_UCSIZE(bp), tx);
2398 dsl_get_next_livelist_obj(objset_t *os, uint64_t zap_obj, uint64_t *llp)
2403 zap_cursor_init(&zc, os, zap_obj);
2404 err = zap_cursor_retrieve(&zc, &za);
2405 zap_cursor_fini(&zc);
2407 *llp = za.za_first_integer;
2412 * Components of livelist deletion that must be performed in syncing
2413 * context: freeing block pointers and updating the pool-wide data
2414 * structures to indicate how much work is left to do
2416 typedef struct sublist_delete_arg {
2421 } sublist_delete_arg_t;
2424 sublist_delete_sync(void *arg, dmu_tx_t *tx)
2426 sublist_delete_arg_t *sda = arg;
2427 spa_t *spa = sda->spa;
2428 dsl_deadlist_t *ll = sda->ll;
2429 uint64_t key = sda->key;
2430 bplist_t *to_free = sda->to_free;
2432 bplist_iterate(to_free, delete_blkptr_cb, spa, tx);
2433 dsl_deadlist_remove_entry(ll, key, tx);
2436 typedef struct livelist_delete_arg {
2440 } livelist_delete_arg_t;
2443 livelist_delete_sync(void *arg, dmu_tx_t *tx)
2445 livelist_delete_arg_t *lda = arg;
2446 spa_t *spa = lda->spa;
2447 uint64_t ll_obj = lda->ll_obj;
2448 uint64_t zap_obj = lda->zap_obj;
2449 objset_t *mos = spa->spa_meta_objset;
2452 /* free the livelist and decrement the feature count */
2453 VERIFY0(zap_remove_int(mos, zap_obj, ll_obj, tx));
2454 dsl_deadlist_free(mos, ll_obj, tx);
2455 spa_feature_decr(spa, SPA_FEATURE_LIVELIST, tx);
2456 VERIFY0(zap_count(mos, zap_obj, &count));
2458 /* no more livelists to delete */
2459 VERIFY0(zap_remove(mos, DMU_POOL_DIRECTORY_OBJECT,
2460 DMU_POOL_DELETED_CLONES, tx));
2461 VERIFY0(zap_destroy(mos, zap_obj, tx));
2462 spa->spa_livelists_to_delete = 0;
2463 spa_notify_waiters(spa);
2468 * Load in the value for the livelist to be removed and open it. Then,
2469 * load its first sublist and determine which block pointers should actually
2470 * be freed. Then, call a synctask which performs the actual frees and updates
2471 * the pool-wide livelist data.
2475 spa_livelist_delete_cb(void *arg, zthr_t *z)
2478 uint64_t ll_obj = 0, count;
2479 objset_t *mos = spa->spa_meta_objset;
2480 uint64_t zap_obj = spa->spa_livelists_to_delete;
2482 * Determine the next livelist to delete. This function should only
2483 * be called if there is at least one deleted clone.
2485 VERIFY0(dsl_get_next_livelist_obj(mos, zap_obj, &ll_obj));
2486 VERIFY0(zap_count(mos, ll_obj, &count));
2488 dsl_deadlist_t ll = { 0 };
2489 dsl_deadlist_entry_t *dle;
2491 dsl_deadlist_open(&ll, mos, ll_obj);
2492 dle = dsl_deadlist_first(&ll);
2493 ASSERT3P(dle, !=, NULL);
2494 bplist_create(&to_free);
2495 int err = dsl_process_sub_livelist(&dle->dle_bpobj, &to_free,
2498 sublist_delete_arg_t sync_arg = {
2501 .key = dle->dle_mintxg,
2504 zfs_dbgmsg("deleting sublist (id %llu) from"
2505 " livelist %llu, %d remaining",
2506 dle->dle_bpobj.bpo_object, ll_obj, count - 1);
2507 VERIFY0(dsl_sync_task(spa_name(spa), NULL,
2508 sublist_delete_sync, &sync_arg, 0,
2509 ZFS_SPACE_CHECK_DESTROY));
2511 VERIFY3U(err, ==, EINTR);
2513 bplist_clear(&to_free);
2514 bplist_destroy(&to_free);
2515 dsl_deadlist_close(&ll);
2517 livelist_delete_arg_t sync_arg = {
2522 zfs_dbgmsg("deletion of livelist %llu completed", ll_obj);
2523 VERIFY0(dsl_sync_task(spa_name(spa), NULL, livelist_delete_sync,
2524 &sync_arg, 0, ZFS_SPACE_CHECK_DESTROY));
2529 spa_start_livelist_destroy_thread(spa_t *spa)
2531 ASSERT3P(spa->spa_livelist_delete_zthr, ==, NULL);
2532 spa->spa_livelist_delete_zthr =
2533 zthr_create("z_livelist_destroy",
2534 spa_livelist_delete_cb_check, spa_livelist_delete_cb, spa);
2537 typedef struct livelist_new_arg {
2540 } livelist_new_arg_t;
2543 livelist_track_new_cb(void *arg, const blkptr_t *bp, boolean_t bp_freed,
2547 livelist_new_arg_t *lna = arg;
2549 bplist_append(lna->frees, bp);
2551 bplist_append(lna->allocs, bp);
2552 zfs_livelist_condense_new_alloc++;
2557 typedef struct livelist_condense_arg {
2560 uint64_t first_size;
2562 } livelist_condense_arg_t;
2565 spa_livelist_condense_sync(void *arg, dmu_tx_t *tx)
2567 livelist_condense_arg_t *lca = arg;
2568 spa_t *spa = lca->spa;
2570 dsl_dataset_t *ds = spa->spa_to_condense.ds;
2572 /* Have we been cancelled? */
2573 if (spa->spa_to_condense.cancelled) {
2574 zfs_livelist_condense_sync_cancel++;
2578 dsl_deadlist_entry_t *first = spa->spa_to_condense.first;
2579 dsl_deadlist_entry_t *next = spa->spa_to_condense.next;
2580 dsl_deadlist_t *ll = &ds->ds_dir->dd_livelist;
2583 * It's possible that the livelist was changed while the zthr was
2584 * running. Therefore, we need to check for new blkptrs in the two
2585 * entries being condensed and continue to track them in the livelist.
2586 * Because of the way we handle remapped blkptrs (see dbuf_remap_impl),
2587 * it's possible that the newly added blkptrs are FREEs or ALLOCs so
2588 * we need to sort them into two different bplists.
2590 uint64_t first_obj = first->dle_bpobj.bpo_object;
2591 uint64_t next_obj = next->dle_bpobj.bpo_object;
2592 uint64_t cur_first_size = first->dle_bpobj.bpo_phys->bpo_num_blkptrs;
2593 uint64_t cur_next_size = next->dle_bpobj.bpo_phys->bpo_num_blkptrs;
2595 bplist_create(&new_frees);
2596 livelist_new_arg_t new_bps = {
2597 .allocs = &lca->to_keep,
2598 .frees = &new_frees,
2601 if (cur_first_size > lca->first_size) {
2602 VERIFY0(livelist_bpobj_iterate_from_nofree(&first->dle_bpobj,
2603 livelist_track_new_cb, &new_bps, lca->first_size));
2605 if (cur_next_size > lca->next_size) {
2606 VERIFY0(livelist_bpobj_iterate_from_nofree(&next->dle_bpobj,
2607 livelist_track_new_cb, &new_bps, lca->next_size));
2610 dsl_deadlist_clear_entry(first, ll, tx);
2611 ASSERT(bpobj_is_empty(&first->dle_bpobj));
2612 dsl_deadlist_remove_entry(ll, next->dle_mintxg, tx);
2614 bplist_iterate(&lca->to_keep, dsl_deadlist_insert_alloc_cb, ll, tx);
2615 bplist_iterate(&new_frees, dsl_deadlist_insert_free_cb, ll, tx);
2616 bplist_destroy(&new_frees);
2618 char dsname[ZFS_MAX_DATASET_NAME_LEN];
2619 dsl_dataset_name(ds, dsname);
2620 zfs_dbgmsg("txg %llu condensing livelist of %s (id %llu), bpobj %llu "
2621 "(%llu blkptrs) and bpobj %llu (%llu blkptrs) -> bpobj %llu "
2622 "(%llu blkptrs)", tx->tx_txg, dsname, ds->ds_object, first_obj,
2623 cur_first_size, next_obj, cur_next_size,
2624 first->dle_bpobj.bpo_object,
2625 first->dle_bpobj.bpo_phys->bpo_num_blkptrs);
2627 dmu_buf_rele(ds->ds_dbuf, spa);
2628 spa->spa_to_condense.ds = NULL;
2629 bplist_clear(&lca->to_keep);
2630 bplist_destroy(&lca->to_keep);
2631 kmem_free(lca, sizeof (livelist_condense_arg_t));
2632 spa->spa_to_condense.syncing = B_FALSE;
2636 spa_livelist_condense_cb(void *arg, zthr_t *t)
2638 while (zfs_livelist_condense_zthr_pause &&
2639 !(zthr_has_waiters(t) || zthr_iscancelled(t)))
2643 dsl_deadlist_entry_t *first = spa->spa_to_condense.first;
2644 dsl_deadlist_entry_t *next = spa->spa_to_condense.next;
2645 uint64_t first_size, next_size;
2647 livelist_condense_arg_t *lca =
2648 kmem_alloc(sizeof (livelist_condense_arg_t), KM_SLEEP);
2649 bplist_create(&lca->to_keep);
2652 * Process the livelists (matching FREEs and ALLOCs) in open context
2653 * so we have minimal work in syncing context to condense.
2655 * We save bpobj sizes (first_size and next_size) to use later in
2656 * syncing context to determine if entries were added to these sublists
2657 * while in open context. This is possible because the clone is still
2658 * active and open for normal writes and we want to make sure the new,
2659 * unprocessed blockpointers are inserted into the livelist normally.
2661 * Note that dsl_process_sub_livelist() both stores the size number of
2662 * blockpointers and iterates over them while the bpobj's lock held, so
2663 * the sizes returned to us are consistent which what was actually
2666 int err = dsl_process_sub_livelist(&first->dle_bpobj, &lca->to_keep, t,
2669 err = dsl_process_sub_livelist(&next->dle_bpobj, &lca->to_keep,
2673 while (zfs_livelist_condense_sync_pause &&
2674 !(zthr_has_waiters(t) || zthr_iscancelled(t)))
2677 dmu_tx_t *tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
2678 dmu_tx_mark_netfree(tx);
2679 dmu_tx_hold_space(tx, 1);
2680 err = dmu_tx_assign(tx, TXG_NOWAIT | TXG_NOTHROTTLE);
2683 * Prevent the condense zthr restarting before
2684 * the synctask completes.
2686 spa->spa_to_condense.syncing = B_TRUE;
2688 lca->first_size = first_size;
2689 lca->next_size = next_size;
2690 dsl_sync_task_nowait(spa_get_dsl(spa),
2691 spa_livelist_condense_sync, lca, 0,
2692 ZFS_SPACE_CHECK_NONE, tx);
2698 * Condensing can not continue: either it was externally stopped or
2699 * we were unable to assign to a tx because the pool has run out of
2700 * space. In the second case, we'll just end up trying to condense
2701 * again in a later txg.
2704 bplist_clear(&lca->to_keep);
2705 bplist_destroy(&lca->to_keep);
2706 kmem_free(lca, sizeof (livelist_condense_arg_t));
2707 dmu_buf_rele(spa->spa_to_condense.ds->ds_dbuf, spa);
2708 spa->spa_to_condense.ds = NULL;
2710 zfs_livelist_condense_zthr_cancel++;
2715 * Check that there is something to condense but that a condense is not
2716 * already in progress and that condensing has not been cancelled.
2719 spa_livelist_condense_cb_check(void *arg, zthr_t *z)
2722 if ((spa->spa_to_condense.ds != NULL) &&
2723 (spa->spa_to_condense.syncing == B_FALSE) &&
2724 (spa->spa_to_condense.cancelled == B_FALSE)) {
2731 spa_start_livelist_condensing_thread(spa_t *spa)
2733 spa->spa_to_condense.ds = NULL;
2734 spa->spa_to_condense.first = NULL;
2735 spa->spa_to_condense.next = NULL;
2736 spa->spa_to_condense.syncing = B_FALSE;
2737 spa->spa_to_condense.cancelled = B_FALSE;
2739 ASSERT3P(spa->spa_livelist_condense_zthr, ==, NULL);
2740 spa->spa_livelist_condense_zthr =
2741 zthr_create("z_livelist_condense",
2742 spa_livelist_condense_cb_check,
2743 spa_livelist_condense_cb, spa);
2747 spa_spawn_aux_threads(spa_t *spa)
2749 ASSERT(spa_writeable(spa));
2751 ASSERT(MUTEX_HELD(&spa_namespace_lock));
2753 spa_start_indirect_condensing_thread(spa);
2754 spa_start_livelist_destroy_thread(spa);
2755 spa_start_livelist_condensing_thread(spa);
2757 ASSERT3P(spa->spa_checkpoint_discard_zthr, ==, NULL);
2758 spa->spa_checkpoint_discard_zthr =
2759 zthr_create("z_checkpoint_discard",
2760 spa_checkpoint_discard_thread_check,
2761 spa_checkpoint_discard_thread, spa);
2765 * Fix up config after a partly-completed split. This is done with the
2766 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
2767 * pool have that entry in their config, but only the splitting one contains
2768 * a list of all the guids of the vdevs that are being split off.
2770 * This function determines what to do with that list: either rejoin
2771 * all the disks to the pool, or complete the splitting process. To attempt
2772 * the rejoin, each disk that is offlined is marked online again, and
2773 * we do a reopen() call. If the vdev label for every disk that was
2774 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
2775 * then we call vdev_split() on each disk, and complete the split.
2777 * Otherwise we leave the config alone, with all the vdevs in place in
2778 * the original pool.
2781 spa_try_repair(spa_t *spa, nvlist_t *config)
2788 boolean_t attempt_reopen;
2790 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
2793 /* check that the config is complete */
2794 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
2795 &glist, &gcount) != 0)
2798 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
2800 /* attempt to online all the vdevs & validate */
2801 attempt_reopen = B_TRUE;
2802 for (i = 0; i < gcount; i++) {
2803 if (glist[i] == 0) /* vdev is hole */
2806 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
2807 if (vd[i] == NULL) {
2809 * Don't bother attempting to reopen the disks;
2810 * just do the split.
2812 attempt_reopen = B_FALSE;
2814 /* attempt to re-online it */
2815 vd[i]->vdev_offline = B_FALSE;
2819 if (attempt_reopen) {
2820 vdev_reopen(spa->spa_root_vdev);
2822 /* check each device to see what state it's in */
2823 for (extracted = 0, i = 0; i < gcount; i++) {
2824 if (vd[i] != NULL &&
2825 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
2832 * If every disk has been moved to the new pool, or if we never
2833 * even attempted to look at them, then we split them off for
2836 if (!attempt_reopen || gcount == extracted) {
2837 for (i = 0; i < gcount; i++)
2840 vdev_reopen(spa->spa_root_vdev);
2843 kmem_free(vd, gcount * sizeof (vdev_t *));
2847 spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type)
2849 char *ereport = FM_EREPORT_ZFS_POOL;
2852 spa->spa_load_state = state;
2853 (void) spa_import_progress_set_state(spa_guid(spa),
2854 spa_load_state(spa));
2856 gethrestime(&spa->spa_loaded_ts);
2857 error = spa_load_impl(spa, type, &ereport);
2860 * Don't count references from objsets that are already closed
2861 * and are making their way through the eviction process.
2863 spa_evicting_os_wait(spa);
2864 spa->spa_minref = zfs_refcount_count(&spa->spa_refcount);
2866 if (error != EEXIST) {
2867 spa->spa_loaded_ts.tv_sec = 0;
2868 spa->spa_loaded_ts.tv_nsec = 0;
2870 if (error != EBADF) {
2871 zfs_ereport_post(ereport, spa, NULL, NULL, NULL, 0, 0);
2874 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2877 (void) spa_import_progress_set_state(spa_guid(spa),
2878 spa_load_state(spa));
2885 * Count the number of per-vdev ZAPs associated with all of the vdevs in the
2886 * vdev tree rooted in the given vd, and ensure that each ZAP is present in the
2887 * spa's per-vdev ZAP list.
2890 vdev_count_verify_zaps(vdev_t *vd)
2892 spa_t *spa = vd->vdev_spa;
2895 if (vd->vdev_top_zap != 0) {
2897 ASSERT0(zap_lookup_int(spa->spa_meta_objset,
2898 spa->spa_all_vdev_zaps, vd->vdev_top_zap));
2900 if (vd->vdev_leaf_zap != 0) {
2902 ASSERT0(zap_lookup_int(spa->spa_meta_objset,
2903 spa->spa_all_vdev_zaps, vd->vdev_leaf_zap));
2906 for (uint64_t i = 0; i < vd->vdev_children; i++) {
2907 total += vdev_count_verify_zaps(vd->vdev_child[i]);
2915 * Determine whether the activity check is required.
2918 spa_activity_check_required(spa_t *spa, uberblock_t *ub, nvlist_t *label,
2922 uint64_t hostid = 0;
2923 uint64_t tryconfig_txg = 0;
2924 uint64_t tryconfig_timestamp = 0;
2925 uint16_t tryconfig_mmp_seq = 0;
2928 if (nvlist_exists(config, ZPOOL_CONFIG_LOAD_INFO)) {
2929 nvinfo = fnvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO);
2930 (void) nvlist_lookup_uint64(nvinfo, ZPOOL_CONFIG_MMP_TXG,
2932 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
2933 &tryconfig_timestamp);
2934 (void) nvlist_lookup_uint16(nvinfo, ZPOOL_CONFIG_MMP_SEQ,
2935 &tryconfig_mmp_seq);
2938 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE, &state);
2941 * Disable the MMP activity check - This is used by zdb which
2942 * is intended to be used on potentially active pools.
2944 if (spa->spa_import_flags & ZFS_IMPORT_SKIP_MMP)
2948 * Skip the activity check when the MMP feature is disabled.
2950 if (ub->ub_mmp_magic == MMP_MAGIC && ub->ub_mmp_delay == 0)
2954 * If the tryconfig_ values are nonzero, they are the results of an
2955 * earlier tryimport. If they all match the uberblock we just found,
2956 * then the pool has not changed and we return false so we do not test
2959 if (tryconfig_txg && tryconfig_txg == ub->ub_txg &&
2960 tryconfig_timestamp && tryconfig_timestamp == ub->ub_timestamp &&
2961 tryconfig_mmp_seq && tryconfig_mmp_seq ==
2962 (MMP_SEQ_VALID(ub) ? MMP_SEQ(ub) : 0))
2966 * Allow the activity check to be skipped when importing the pool
2967 * on the same host which last imported it. Since the hostid from
2968 * configuration may be stale use the one read from the label.
2970 if (nvlist_exists(label, ZPOOL_CONFIG_HOSTID))
2971 hostid = fnvlist_lookup_uint64(label, ZPOOL_CONFIG_HOSTID);
2973 if (hostid == spa_get_hostid(spa))
2977 * Skip the activity test when the pool was cleanly exported.
2979 if (state != POOL_STATE_ACTIVE)
2986 * Nanoseconds the activity check must watch for changes on-disk.
2989 spa_activity_check_duration(spa_t *spa, uberblock_t *ub)
2991 uint64_t import_intervals = MAX(zfs_multihost_import_intervals, 1);
2992 uint64_t multihost_interval = MSEC2NSEC(
2993 MMP_INTERVAL_OK(zfs_multihost_interval));
2994 uint64_t import_delay = MAX(NANOSEC, import_intervals *
2995 multihost_interval);
2998 * Local tunables determine a minimum duration except for the case
2999 * where we know when the remote host will suspend the pool if MMP
3000 * writes do not land.
3002 * See Big Theory comment at the top of mmp.c for the reasoning behind
3003 * these cases and times.
3006 ASSERT(MMP_IMPORT_SAFETY_FACTOR >= 100);
3008 if (MMP_INTERVAL_VALID(ub) && MMP_FAIL_INT_VALID(ub) &&
3009 MMP_FAIL_INT(ub) > 0) {
3011 /* MMP on remote host will suspend pool after failed writes */
3012 import_delay = MMP_FAIL_INT(ub) * MSEC2NSEC(MMP_INTERVAL(ub)) *
3013 MMP_IMPORT_SAFETY_FACTOR / 100;
3015 zfs_dbgmsg("fail_intvals>0 import_delay=%llu ub_mmp "
3016 "mmp_fails=%llu ub_mmp mmp_interval=%llu "
3017 "import_intervals=%u", import_delay, MMP_FAIL_INT(ub),
3018 MMP_INTERVAL(ub), import_intervals);
3020 } else if (MMP_INTERVAL_VALID(ub) && MMP_FAIL_INT_VALID(ub) &&
3021 MMP_FAIL_INT(ub) == 0) {
3023 /* MMP on remote host will never suspend pool */
3024 import_delay = MAX(import_delay, (MSEC2NSEC(MMP_INTERVAL(ub)) +
3025 ub->ub_mmp_delay) * import_intervals);
3027 zfs_dbgmsg("fail_intvals=0 import_delay=%llu ub_mmp "
3028 "mmp_interval=%llu ub_mmp_delay=%llu "
3029 "import_intervals=%u", import_delay, MMP_INTERVAL(ub),
3030 ub->ub_mmp_delay, import_intervals);
3032 } else if (MMP_VALID(ub)) {
3034 * zfs-0.7 compatibility case
3037 import_delay = MAX(import_delay, (multihost_interval +
3038 ub->ub_mmp_delay) * import_intervals);
3040 zfs_dbgmsg("import_delay=%llu ub_mmp_delay=%llu "
3041 "import_intervals=%u leaves=%u", import_delay,
3042 ub->ub_mmp_delay, import_intervals,
3043 vdev_count_leaves(spa));
3045 /* Using local tunings is the only reasonable option */
3046 zfs_dbgmsg("pool last imported on non-MMP aware "
3047 "host using import_delay=%llu multihost_interval=%llu "
3048 "import_intervals=%u", import_delay, multihost_interval,
3052 return (import_delay);
3056 * Perform the import activity check. If the user canceled the import or
3057 * we detected activity then fail.
3060 spa_activity_check(spa_t *spa, uberblock_t *ub, nvlist_t *config)
3062 uint64_t txg = ub->ub_txg;
3063 uint64_t timestamp = ub->ub_timestamp;
3064 uint64_t mmp_config = ub->ub_mmp_config;
3065 uint16_t mmp_seq = MMP_SEQ_VALID(ub) ? MMP_SEQ(ub) : 0;
3066 uint64_t import_delay;
3067 hrtime_t import_expire;
3068 nvlist_t *mmp_label = NULL;
3069 vdev_t *rvd = spa->spa_root_vdev;
3074 cv_init(&cv, NULL, CV_DEFAULT, NULL);
3075 mutex_init(&mtx, NULL, MUTEX_DEFAULT, NULL);
3079 * If ZPOOL_CONFIG_MMP_TXG is present an activity check was performed
3080 * during the earlier tryimport. If the txg recorded there is 0 then
3081 * the pool is known to be active on another host.
3083 * Otherwise, the pool might be in use on another host. Check for
3084 * changes in the uberblocks on disk if necessary.
3086 if (nvlist_exists(config, ZPOOL_CONFIG_LOAD_INFO)) {
3087 nvlist_t *nvinfo = fnvlist_lookup_nvlist(config,
3088 ZPOOL_CONFIG_LOAD_INFO);
3090 if (nvlist_exists(nvinfo, ZPOOL_CONFIG_MMP_TXG) &&
3091 fnvlist_lookup_uint64(nvinfo, ZPOOL_CONFIG_MMP_TXG) == 0) {
3092 vdev_uberblock_load(rvd, ub, &mmp_label);
3093 error = SET_ERROR(EREMOTEIO);
3098 import_delay = spa_activity_check_duration(spa, ub);
3100 /* Add a small random factor in case of simultaneous imports (0-25%) */
3101 import_delay += import_delay * spa_get_random(250) / 1000;
3103 import_expire = gethrtime() + import_delay;
3105 while (gethrtime() < import_expire) {
3106 (void) spa_import_progress_set_mmp_check(spa_guid(spa),
3107 NSEC2SEC(import_expire - gethrtime()));
3109 vdev_uberblock_load(rvd, ub, &mmp_label);
3111 if (txg != ub->ub_txg || timestamp != ub->ub_timestamp ||
3112 mmp_seq != (MMP_SEQ_VALID(ub) ? MMP_SEQ(ub) : 0)) {
3113 zfs_dbgmsg("multihost activity detected "
3114 "txg %llu ub_txg %llu "
3115 "timestamp %llu ub_timestamp %llu "
3116 "mmp_config %#llx ub_mmp_config %#llx",
3117 txg, ub->ub_txg, timestamp, ub->ub_timestamp,
3118 mmp_config, ub->ub_mmp_config);
3120 error = SET_ERROR(EREMOTEIO);
3125 nvlist_free(mmp_label);
3129 error = cv_timedwait_sig(&cv, &mtx, ddi_get_lbolt() + hz);
3131 error = SET_ERROR(EINTR);
3139 mutex_destroy(&mtx);
3143 * If the pool is determined to be active store the status in the
3144 * spa->spa_load_info nvlist. If the remote hostname or hostid are
3145 * available from configuration read from disk store them as well.
3146 * This allows 'zpool import' to generate a more useful message.
3148 * ZPOOL_CONFIG_MMP_STATE - observed pool status (mandatory)
3149 * ZPOOL_CONFIG_MMP_HOSTNAME - hostname from the active pool
3150 * ZPOOL_CONFIG_MMP_HOSTID - hostid from the active pool
3152 if (error == EREMOTEIO) {
3153 char *hostname = "<unknown>";
3154 uint64_t hostid = 0;
3157 if (nvlist_exists(mmp_label, ZPOOL_CONFIG_HOSTNAME)) {
3158 hostname = fnvlist_lookup_string(mmp_label,
3159 ZPOOL_CONFIG_HOSTNAME);
3160 fnvlist_add_string(spa->spa_load_info,
3161 ZPOOL_CONFIG_MMP_HOSTNAME, hostname);
3164 if (nvlist_exists(mmp_label, ZPOOL_CONFIG_HOSTID)) {
3165 hostid = fnvlist_lookup_uint64(mmp_label,
3166 ZPOOL_CONFIG_HOSTID);
3167 fnvlist_add_uint64(spa->spa_load_info,
3168 ZPOOL_CONFIG_MMP_HOSTID, hostid);
3172 fnvlist_add_uint64(spa->spa_load_info,
3173 ZPOOL_CONFIG_MMP_STATE, MMP_STATE_ACTIVE);
3174 fnvlist_add_uint64(spa->spa_load_info,
3175 ZPOOL_CONFIG_MMP_TXG, 0);
3177 error = spa_vdev_err(rvd, VDEV_AUX_ACTIVE, EREMOTEIO);
3181 nvlist_free(mmp_label);
3187 spa_verify_host(spa_t *spa, nvlist_t *mos_config)
3191 uint64_t myhostid = 0;
3193 if (!spa_is_root(spa) && nvlist_lookup_uint64(mos_config,
3194 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
3195 hostname = fnvlist_lookup_string(mos_config,
3196 ZPOOL_CONFIG_HOSTNAME);
3198 myhostid = zone_get_hostid(NULL);
3200 if (hostid != 0 && myhostid != 0 && hostid != myhostid) {
3201 cmn_err(CE_WARN, "pool '%s' could not be "
3202 "loaded as it was last accessed by "
3203 "another system (host: %s hostid: 0x%llx). "
3204 "See: http://illumos.org/msg/ZFS-8000-EY",
3205 spa_name(spa), hostname, (u_longlong_t)hostid);
3206 spa_load_failed(spa, "hostid verification failed: pool "
3207 "last accessed by host: %s (hostid: 0x%llx)",
3208 hostname, (u_longlong_t)hostid);
3209 return (SET_ERROR(EBADF));
3217 spa_ld_parse_config(spa_t *spa, spa_import_type_t type)
3220 nvlist_t *nvtree, *nvl, *config = spa->spa_config;
3227 * Versioning wasn't explicitly added to the label until later, so if
3228 * it's not present treat it as the initial version.
3230 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
3231 &spa->spa_ubsync.ub_version) != 0)
3232 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
3234 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) {
3235 spa_load_failed(spa, "invalid config provided: '%s' missing",
3236 ZPOOL_CONFIG_POOL_GUID);
3237 return (SET_ERROR(EINVAL));
3241 * If we are doing an import, ensure that the pool is not already
3242 * imported by checking if its pool guid already exists in the
3245 * The only case that we allow an already imported pool to be
3246 * imported again, is when the pool is checkpointed and we want to
3247 * look at its checkpointed state from userland tools like zdb.
3250 if ((spa->spa_load_state == SPA_LOAD_IMPORT ||
3251 spa->spa_load_state == SPA_LOAD_TRYIMPORT) &&
3252 spa_guid_exists(pool_guid, 0)) {
3254 if ((spa->spa_load_state == SPA_LOAD_IMPORT ||
3255 spa->spa_load_state == SPA_LOAD_TRYIMPORT) &&
3256 spa_guid_exists(pool_guid, 0) &&
3257 !spa_importing_readonly_checkpoint(spa)) {
3259 spa_load_failed(spa, "a pool with guid %llu is already open",
3260 (u_longlong_t)pool_guid);
3261 return (SET_ERROR(EEXIST));
3264 spa->spa_config_guid = pool_guid;
3266 nvlist_free(spa->spa_load_info);
3267 spa->spa_load_info = fnvlist_alloc();
3269 ASSERT(spa->spa_comment == NULL);
3270 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
3271 spa->spa_comment = spa_strdup(comment);
3273 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
3274 &spa->spa_config_txg);
3276 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) == 0)
3277 spa->spa_config_splitting = fnvlist_dup(nvl);
3279 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvtree)) {
3280 spa_load_failed(spa, "invalid config provided: '%s' missing",
3281 ZPOOL_CONFIG_VDEV_TREE);
3282 return (SET_ERROR(EINVAL));
3286 * Create "The Godfather" zio to hold all async IOs
3288 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
3290 for (int i = 0; i < max_ncpus; i++) {
3291 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
3292 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
3293 ZIO_FLAG_GODFATHER);
3297 * Parse the configuration into a vdev tree. We explicitly set the
3298 * value that will be returned by spa_version() since parsing the
3299 * configuration requires knowing the version number.
3301 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3302 parse = (type == SPA_IMPORT_EXISTING ?
3303 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
3304 error = spa_config_parse(spa, &rvd, nvtree, NULL, 0, parse);
3305 spa_config_exit(spa, SCL_ALL, FTAG);
3308 spa_load_failed(spa, "unable to parse config [error=%d]",
3313 ASSERT(spa->spa_root_vdev == rvd);
3314 ASSERT3U(spa->spa_min_ashift, >=, SPA_MINBLOCKSHIFT);
3315 ASSERT3U(spa->spa_max_ashift, <=, SPA_MAXBLOCKSHIFT);
3317 if (type != SPA_IMPORT_ASSEMBLE) {
3318 ASSERT(spa_guid(spa) == pool_guid);
3325 * Recursively open all vdevs in the vdev tree. This function is called twice:
3326 * first with the untrusted config, then with the trusted config.
3329 spa_ld_open_vdevs(spa_t *spa)
3334 * spa_missing_tvds_allowed defines how many top-level vdevs can be
3335 * missing/unopenable for the root vdev to be still considered openable.
3337 if (spa->spa_trust_config) {
3338 spa->spa_missing_tvds_allowed = zfs_max_missing_tvds;
3339 } else if (spa->spa_config_source == SPA_CONFIG_SRC_CACHEFILE) {
3340 spa->spa_missing_tvds_allowed = zfs_max_missing_tvds_cachefile;
3341 } else if (spa->spa_config_source == SPA_CONFIG_SRC_SCAN) {
3342 spa->spa_missing_tvds_allowed = zfs_max_missing_tvds_scan;
3344 spa->spa_missing_tvds_allowed = 0;
3347 spa->spa_missing_tvds_allowed =
3348 MAX(zfs_max_missing_tvds, spa->spa_missing_tvds_allowed);
3350 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3351 error = vdev_open(spa->spa_root_vdev);
3352 spa_config_exit(spa, SCL_ALL, FTAG);
3354 if (spa->spa_missing_tvds != 0) {
3355 spa_load_note(spa, "vdev tree has %lld missing top-level "
3356 "vdevs.", (u_longlong_t)spa->spa_missing_tvds);
3357 if (spa->spa_trust_config && (spa->spa_mode & SPA_MODE_WRITE)) {
3359 * Although theoretically we could allow users to open
3360 * incomplete pools in RW mode, we'd need to add a lot
3361 * of extra logic (e.g. adjust pool space to account
3362 * for missing vdevs).
3363 * This limitation also prevents users from accidentally
3364 * opening the pool in RW mode during data recovery and
3365 * damaging it further.
3367 spa_load_note(spa, "pools with missing top-level "
3368 "vdevs can only be opened in read-only mode.");
3369 error = SET_ERROR(ENXIO);
3371 spa_load_note(spa, "current settings allow for maximum "
3372 "%lld missing top-level vdevs at this stage.",
3373 (u_longlong_t)spa->spa_missing_tvds_allowed);
3377 spa_load_failed(spa, "unable to open vdev tree [error=%d]",
3380 if (spa->spa_missing_tvds != 0 || error != 0)
3381 vdev_dbgmsg_print_tree(spa->spa_root_vdev, 2);
3387 * We need to validate the vdev labels against the configuration that
3388 * we have in hand. This function is called twice: first with an untrusted
3389 * config, then with a trusted config. The validation is more strict when the
3390 * config is trusted.
3393 spa_ld_validate_vdevs(spa_t *spa)
3396 vdev_t *rvd = spa->spa_root_vdev;
3398 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3399 error = vdev_validate(rvd);
3400 spa_config_exit(spa, SCL_ALL, FTAG);
3403 spa_load_failed(spa, "vdev_validate failed [error=%d]", error);
3407 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) {
3408 spa_load_failed(spa, "cannot open vdev tree after invalidating "
3410 vdev_dbgmsg_print_tree(rvd, 2);
3411 return (SET_ERROR(ENXIO));
3418 spa_ld_select_uberblock_done(spa_t *spa, uberblock_t *ub)
3420 spa->spa_state = POOL_STATE_ACTIVE;
3421 spa->spa_ubsync = spa->spa_uberblock;
3422 spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
3423 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
3424 spa->spa_first_txg = spa->spa_last_ubsync_txg ?
3425 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
3426 spa->spa_claim_max_txg = spa->spa_first_txg;
3427 spa->spa_prev_software_version = ub->ub_software_version;
3431 spa_ld_select_uberblock(spa_t *spa, spa_import_type_t type)
3433 vdev_t *rvd = spa->spa_root_vdev;
3435 uberblock_t *ub = &spa->spa_uberblock;
3436 boolean_t activity_check = B_FALSE;
3439 * If we are opening the checkpointed state of the pool by
3440 * rewinding to it, at this point we will have written the
3441 * checkpointed uberblock to the vdev labels, so searching
3442 * the labels will find the right uberblock. However, if
3443 * we are opening the checkpointed state read-only, we have
3444 * not modified the labels. Therefore, we must ignore the
3445 * labels and continue using the spa_uberblock that was set
3446 * by spa_ld_checkpoint_rewind.
3448 * Note that it would be fine to ignore the labels when
3449 * rewinding (opening writeable) as well. However, if we
3450 * crash just after writing the labels, we will end up
3451 * searching the labels. Doing so in the common case means
3452 * that this code path gets exercised normally, rather than
3453 * just in the edge case.
3455 if (ub->ub_checkpoint_txg != 0 &&
3456 spa_importing_readonly_checkpoint(spa)) {
3457 spa_ld_select_uberblock_done(spa, ub);
3462 * Find the best uberblock.
3464 vdev_uberblock_load(rvd, ub, &label);
3467 * If we weren't able to find a single valid uberblock, return failure.
3469 if (ub->ub_txg == 0) {
3471 spa_load_failed(spa, "no valid uberblock found");
3472 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
3475 if (spa->spa_load_max_txg != UINT64_MAX) {
3476 (void) spa_import_progress_set_max_txg(spa_guid(spa),
3477 (u_longlong_t)spa->spa_load_max_txg);
3479 spa_load_note(spa, "using uberblock with txg=%llu",
3480 (u_longlong_t)ub->ub_txg);
3484 * For pools which have the multihost property on determine if the
3485 * pool is truly inactive and can be safely imported. Prevent
3486 * hosts which don't have a hostid set from importing the pool.
3488 activity_check = spa_activity_check_required(spa, ub, label,
3490 if (activity_check) {
3491 if (ub->ub_mmp_magic == MMP_MAGIC && ub->ub_mmp_delay &&
3492 spa_get_hostid(spa) == 0) {
3494 fnvlist_add_uint64(spa->spa_load_info,
3495 ZPOOL_CONFIG_MMP_STATE, MMP_STATE_NO_HOSTID);
3496 return (spa_vdev_err(rvd, VDEV_AUX_ACTIVE, EREMOTEIO));
3499 int error = spa_activity_check(spa, ub, spa->spa_config);
3505 fnvlist_add_uint64(spa->spa_load_info,
3506 ZPOOL_CONFIG_MMP_STATE, MMP_STATE_INACTIVE);
3507 fnvlist_add_uint64(spa->spa_load_info,
3508 ZPOOL_CONFIG_MMP_TXG, ub->ub_txg);
3509 fnvlist_add_uint16(spa->spa_load_info,
3510 ZPOOL_CONFIG_MMP_SEQ,
3511 (MMP_SEQ_VALID(ub) ? MMP_SEQ(ub) : 0));
3515 * If the pool has an unsupported version we can't open it.
3517 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
3519 spa_load_failed(spa, "version %llu is not supported",
3520 (u_longlong_t)ub->ub_version);
3521 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
3524 if (ub->ub_version >= SPA_VERSION_FEATURES) {
3528 * If we weren't able to find what's necessary for reading the
3529 * MOS in the label, return failure.
3531 if (label == NULL) {
3532 spa_load_failed(spa, "label config unavailable");
3533 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
3537 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_FEATURES_FOR_READ,
3540 spa_load_failed(spa, "invalid label: '%s' missing",
3541 ZPOOL_CONFIG_FEATURES_FOR_READ);
3542 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
3547 * Update our in-core representation with the definitive values
3550 nvlist_free(spa->spa_label_features);
3551 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
3557 * Look through entries in the label nvlist's features_for_read. If
3558 * there is a feature listed there which we don't understand then we
3559 * cannot open a pool.
3561 if (ub->ub_version >= SPA_VERSION_FEATURES) {
3562 nvlist_t *unsup_feat;
3564 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
3567 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
3569 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
3570 if (!zfeature_is_supported(nvpair_name(nvp))) {
3571 VERIFY(nvlist_add_string(unsup_feat,
3572 nvpair_name(nvp), "") == 0);
3576 if (!nvlist_empty(unsup_feat)) {
3577 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
3578 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
3579 nvlist_free(unsup_feat);
3580 spa_load_failed(spa, "some features are unsupported");
3581 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
3585 nvlist_free(unsup_feat);
3588 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
3589 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3590 spa_try_repair(spa, spa->spa_config);
3591 spa_config_exit(spa, SCL_ALL, FTAG);
3592 nvlist_free(spa->spa_config_splitting);
3593 spa->spa_config_splitting = NULL;
3597 * Initialize internal SPA structures.
3599 spa_ld_select_uberblock_done(spa, ub);
3605 spa_ld_open_rootbp(spa_t *spa)
3608 vdev_t *rvd = spa->spa_root_vdev;
3610 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
3612 spa_load_failed(spa, "unable to open rootbp in dsl_pool_init "
3613 "[error=%d]", error);
3614 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3616 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
3622 spa_ld_trusted_config(spa_t *spa, spa_import_type_t type,
3623 boolean_t reloading)
3625 vdev_t *mrvd, *rvd = spa->spa_root_vdev;
3626 nvlist_t *nv, *mos_config, *policy;
3627 int error = 0, copy_error;
3628 uint64_t healthy_tvds, healthy_tvds_mos;
3629 uint64_t mos_config_txg;
3631 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object, B_TRUE)
3633 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3636 * If we're assembling a pool from a split, the config provided is
3637 * already trusted so there is nothing to do.
3639 if (type == SPA_IMPORT_ASSEMBLE)
3642 healthy_tvds = spa_healthy_core_tvds(spa);
3644 if (load_nvlist(spa, spa->spa_config_object, &mos_config)
3646 spa_load_failed(spa, "unable to retrieve MOS config");
3647 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3651 * If we are doing an open, pool owner wasn't verified yet, thus do
3652 * the verification here.
3654 if (spa->spa_load_state == SPA_LOAD_OPEN) {
3655 error = spa_verify_host(spa, mos_config);
3657 nvlist_free(mos_config);
3662 nv = fnvlist_lookup_nvlist(mos_config, ZPOOL_CONFIG_VDEV_TREE);
3664 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3667 * Build a new vdev tree from the trusted config
3669 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
3672 * Vdev paths in the MOS may be obsolete. If the untrusted config was
3673 * obtained by scanning /dev/dsk, then it will have the right vdev
3674 * paths. We update the trusted MOS config with this information.
3675 * We first try to copy the paths with vdev_copy_path_strict, which
3676 * succeeds only when both configs have exactly the same vdev tree.
3677 * If that fails, we fall back to a more flexible method that has a
3678 * best effort policy.
3680 copy_error = vdev_copy_path_strict(rvd, mrvd);
3681 if (copy_error != 0 || spa_load_print_vdev_tree) {
3682 spa_load_note(spa, "provided vdev tree:");
3683 vdev_dbgmsg_print_tree(rvd, 2);
3684 spa_load_note(spa, "MOS vdev tree:");
3685 vdev_dbgmsg_print_tree(mrvd, 2);
3687 if (copy_error != 0) {
3688 spa_load_note(spa, "vdev_copy_path_strict failed, falling "
3689 "back to vdev_copy_path_relaxed");
3690 vdev_copy_path_relaxed(rvd, mrvd);
3695 spa->spa_root_vdev = mrvd;
3697 spa_config_exit(spa, SCL_ALL, FTAG);
3700 * We will use spa_config if we decide to reload the spa or if spa_load
3701 * fails and we rewind. We must thus regenerate the config using the
3702 * MOS information with the updated paths. ZPOOL_LOAD_POLICY is used to
3703 * pass settings on how to load the pool and is not stored in the MOS.
3704 * We copy it over to our new, trusted config.
3706 mos_config_txg = fnvlist_lookup_uint64(mos_config,
3707 ZPOOL_CONFIG_POOL_TXG);
3708 nvlist_free(mos_config);
3709 mos_config = spa_config_generate(spa, NULL, mos_config_txg, B_FALSE);
3710 if (nvlist_lookup_nvlist(spa->spa_config, ZPOOL_LOAD_POLICY,
3712 fnvlist_add_nvlist(mos_config, ZPOOL_LOAD_POLICY, policy);
3713 spa_config_set(spa, mos_config);
3714 spa->spa_config_source = SPA_CONFIG_SRC_MOS;
3717 * Now that we got the config from the MOS, we should be more strict
3718 * in checking blkptrs and can make assumptions about the consistency
3719 * of the vdev tree. spa_trust_config must be set to true before opening
3720 * vdevs in order for them to be writeable.
3722 spa->spa_trust_config = B_TRUE;
3725 * Open and validate the new vdev tree
3727 error = spa_ld_open_vdevs(spa);
3731 error = spa_ld_validate_vdevs(spa);
3735 if (copy_error != 0 || spa_load_print_vdev_tree) {
3736 spa_load_note(spa, "final vdev tree:");
3737 vdev_dbgmsg_print_tree(rvd, 2);
3740 if (spa->spa_load_state != SPA_LOAD_TRYIMPORT &&
3741 !spa->spa_extreme_rewind && zfs_max_missing_tvds == 0) {
3743 * Sanity check to make sure that we are indeed loading the
3744 * latest uberblock. If we missed SPA_SYNC_MIN_VDEVS tvds
3745 * in the config provided and they happened to be the only ones
3746 * to have the latest uberblock, we could involuntarily perform
3747 * an extreme rewind.
3749 healthy_tvds_mos = spa_healthy_core_tvds(spa);
3750 if (healthy_tvds_mos - healthy_tvds >=
3751 SPA_SYNC_MIN_VDEVS) {
3752 spa_load_note(spa, "config provided misses too many "
3753 "top-level vdevs compared to MOS (%lld vs %lld). ",
3754 (u_longlong_t)healthy_tvds,
3755 (u_longlong_t)healthy_tvds_mos);
3756 spa_load_note(spa, "vdev tree:");
3757 vdev_dbgmsg_print_tree(rvd, 2);
3759 spa_load_failed(spa, "config was already "
3760 "provided from MOS. Aborting.");
3761 return (spa_vdev_err(rvd,
3762 VDEV_AUX_CORRUPT_DATA, EIO));
3764 spa_load_note(spa, "spa must be reloaded using MOS "
3766 return (SET_ERROR(EAGAIN));
3770 error = spa_check_for_missing_logs(spa);
3772 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
3774 if (rvd->vdev_guid_sum != spa->spa_uberblock.ub_guid_sum) {
3775 spa_load_failed(spa, "uberblock guid sum doesn't match MOS "
3776 "guid sum (%llu != %llu)",
3777 (u_longlong_t)spa->spa_uberblock.ub_guid_sum,
3778 (u_longlong_t)rvd->vdev_guid_sum);
3779 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
3787 spa_ld_open_indirect_vdev_metadata(spa_t *spa)
3790 vdev_t *rvd = spa->spa_root_vdev;
3793 * Everything that we read before spa_remove_init() must be stored
3794 * on concreted vdevs. Therefore we do this as early as possible.
3796 error = spa_remove_init(spa);
3798 spa_load_failed(spa, "spa_remove_init failed [error=%d]",
3800 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3804 * Retrieve information needed to condense indirect vdev mappings.
3806 error = spa_condense_init(spa);
3808 spa_load_failed(spa, "spa_condense_init failed [error=%d]",
3810 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, error));
3817 spa_ld_check_features(spa_t *spa, boolean_t *missing_feat_writep)
3820 vdev_t *rvd = spa->spa_root_vdev;
3822 if (spa_version(spa) >= SPA_VERSION_FEATURES) {
3823 boolean_t missing_feat_read = B_FALSE;
3824 nvlist_t *unsup_feat, *enabled_feat;
3826 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
3827 &spa->spa_feat_for_read_obj, B_TRUE) != 0) {
3828 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3831 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
3832 &spa->spa_feat_for_write_obj, B_TRUE) != 0) {
3833 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3836 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
3837 &spa->spa_feat_desc_obj, B_TRUE) != 0) {
3838 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3841 enabled_feat = fnvlist_alloc();
3842 unsup_feat = fnvlist_alloc();
3844 if (!spa_features_check(spa, B_FALSE,
3845 unsup_feat, enabled_feat))
3846 missing_feat_read = B_TRUE;
3848 if (spa_writeable(spa) ||
3849 spa->spa_load_state == SPA_LOAD_TRYIMPORT) {
3850 if (!spa_features_check(spa, B_TRUE,
3851 unsup_feat, enabled_feat)) {
3852 *missing_feat_writep = B_TRUE;
3856 fnvlist_add_nvlist(spa->spa_load_info,
3857 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
3859 if (!nvlist_empty(unsup_feat)) {
3860 fnvlist_add_nvlist(spa->spa_load_info,
3861 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
3864 fnvlist_free(enabled_feat);
3865 fnvlist_free(unsup_feat);
3867 if (!missing_feat_read) {
3868 fnvlist_add_boolean(spa->spa_load_info,
3869 ZPOOL_CONFIG_CAN_RDONLY);
3873 * If the state is SPA_LOAD_TRYIMPORT, our objective is
3874 * twofold: to determine whether the pool is available for
3875 * import in read-write mode and (if it is not) whether the
3876 * pool is available for import in read-only mode. If the pool
3877 * is available for import in read-write mode, it is displayed
3878 * as available in userland; if it is not available for import
3879 * in read-only mode, it is displayed as unavailable in
3880 * userland. If the pool is available for import in read-only
3881 * mode but not read-write mode, it is displayed as unavailable
3882 * in userland with a special note that the pool is actually
3883 * available for open in read-only mode.
3885 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
3886 * missing a feature for write, we must first determine whether
3887 * the pool can be opened read-only before returning to
3888 * userland in order to know whether to display the
3889 * abovementioned note.
3891 if (missing_feat_read || (*missing_feat_writep &&
3892 spa_writeable(spa))) {
3893 spa_load_failed(spa, "pool uses unsupported features");
3894 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
3899 * Load refcounts for ZFS features from disk into an in-memory
3900 * cache during SPA initialization.
3902 for (spa_feature_t i = 0; i < SPA_FEATURES; i++) {
3905 error = feature_get_refcount_from_disk(spa,
3906 &spa_feature_table[i], &refcount);
3908 spa->spa_feat_refcount_cache[i] = refcount;
3909 } else if (error == ENOTSUP) {
3910 spa->spa_feat_refcount_cache[i] =
3911 SPA_FEATURE_DISABLED;
3913 spa_load_failed(spa, "error getting refcount "
3914 "for feature %s [error=%d]",
3915 spa_feature_table[i].fi_guid, error);
3916 return (spa_vdev_err(rvd,
3917 VDEV_AUX_CORRUPT_DATA, EIO));
3922 if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) {
3923 if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG,
3924 &spa->spa_feat_enabled_txg_obj, B_TRUE) != 0)
3925 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3929 * Encryption was added before bookmark_v2, even though bookmark_v2
3930 * is now a dependency. If this pool has encryption enabled without
3931 * bookmark_v2, trigger an errata message.
3933 if (spa_feature_is_enabled(spa, SPA_FEATURE_ENCRYPTION) &&
3934 !spa_feature_is_enabled(spa, SPA_FEATURE_BOOKMARK_V2)) {
3935 spa->spa_errata = ZPOOL_ERRATA_ZOL_8308_ENCRYPTION;
3942 spa_ld_load_special_directories(spa_t *spa)
3945 vdev_t *rvd = spa->spa_root_vdev;
3947 spa->spa_is_initializing = B_TRUE;
3948 error = dsl_pool_open(spa->spa_dsl_pool);
3949 spa->spa_is_initializing = B_FALSE;
3951 spa_load_failed(spa, "dsl_pool_open failed [error=%d]", error);
3952 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3959 spa_ld_get_props(spa_t *spa)
3963 vdev_t *rvd = spa->spa_root_vdev;
3965 /* Grab the checksum salt from the MOS. */
3966 error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
3967 DMU_POOL_CHECKSUM_SALT, 1,
3968 sizeof (spa->spa_cksum_salt.zcs_bytes),
3969 spa->spa_cksum_salt.zcs_bytes);
3970 if (error == ENOENT) {
3971 /* Generate a new salt for subsequent use */
3972 (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes,
3973 sizeof (spa->spa_cksum_salt.zcs_bytes));
3974 } else if (error != 0) {
3975 spa_load_failed(spa, "unable to retrieve checksum salt from "
3976 "MOS [error=%d]", error);
3977 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3980 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj, B_TRUE) != 0)
3981 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3982 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
3984 spa_load_failed(spa, "error opening deferred-frees bpobj "
3985 "[error=%d]", error);
3986 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3990 * Load the bit that tells us to use the new accounting function
3991 * (raid-z deflation). If we have an older pool, this will not
3994 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate, B_FALSE);
3995 if (error != 0 && error != ENOENT)
3996 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3998 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
3999 &spa->spa_creation_version, B_FALSE);
4000 if (error != 0 && error != ENOENT)
4001 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4004 * Load the persistent error log. If we have an older pool, this will
4007 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last,
4009 if (error != 0 && error != ENOENT)
4010 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4012 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
4013 &spa->spa_errlog_scrub, B_FALSE);
4014 if (error != 0 && error != ENOENT)
4015 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4018 * Load the livelist deletion field. If a livelist is queued for
4019 * deletion, indicate that in the spa
4021 error = spa_dir_prop(spa, DMU_POOL_DELETED_CLONES,
4022 &spa->spa_livelists_to_delete, B_FALSE);
4023 if (error != 0 && error != ENOENT)
4024 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4027 * Load the history object. If we have an older pool, this
4028 * will not be present.
4030 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history, B_FALSE);
4031 if (error != 0 && error != ENOENT)
4032 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4035 * Load the per-vdev ZAP map. If we have an older pool, this will not
4036 * be present; in this case, defer its creation to a later time to
4037 * avoid dirtying the MOS this early / out of sync context. See
4038 * spa_sync_config_object.
4041 /* The sentinel is only available in the MOS config. */
4042 nvlist_t *mos_config;
4043 if (load_nvlist(spa, spa->spa_config_object, &mos_config) != 0) {
4044 spa_load_failed(spa, "unable to retrieve MOS config");
4045 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4048 error = spa_dir_prop(spa, DMU_POOL_VDEV_ZAP_MAP,
4049 &spa->spa_all_vdev_zaps, B_FALSE);
4051 if (error == ENOENT) {
4052 VERIFY(!nvlist_exists(mos_config,
4053 ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS));
4054 spa->spa_avz_action = AVZ_ACTION_INITIALIZE;
4055 ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev));
4056 } else if (error != 0) {
4057 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4058 } else if (!nvlist_exists(mos_config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)) {
4060 * An older version of ZFS overwrote the sentinel value, so
4061 * we have orphaned per-vdev ZAPs in the MOS. Defer their
4062 * destruction to later; see spa_sync_config_object.
4064 spa->spa_avz_action = AVZ_ACTION_DESTROY;
4066 * We're assuming that no vdevs have had their ZAPs created
4067 * before this. Better be sure of it.
4069 ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev));
4071 nvlist_free(mos_config);
4073 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
4075 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object,
4077 if (error && error != ENOENT)
4078 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4081 uint64_t autoreplace;
4083 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
4084 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
4085 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
4086 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
4087 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
4088 spa_prop_find(spa, ZPOOL_PROP_MULTIHOST, &spa->spa_multihost);
4089 spa_prop_find(spa, ZPOOL_PROP_AUTOTRIM, &spa->spa_autotrim);
4090 spa->spa_autoreplace = (autoreplace != 0);
4094 * If we are importing a pool with missing top-level vdevs,
4095 * we enforce that the pool doesn't panic or get suspended on
4096 * error since the likelihood of missing data is extremely high.
4098 if (spa->spa_missing_tvds > 0 &&
4099 spa->spa_failmode != ZIO_FAILURE_MODE_CONTINUE &&
4100 spa->spa_load_state != SPA_LOAD_TRYIMPORT) {
4101 spa_load_note(spa, "forcing failmode to 'continue' "
4102 "as some top level vdevs are missing");
4103 spa->spa_failmode = ZIO_FAILURE_MODE_CONTINUE;
4110 spa_ld_open_aux_vdevs(spa_t *spa, spa_import_type_t type)
4113 vdev_t *rvd = spa->spa_root_vdev;
4116 * If we're assembling the pool from the split-off vdevs of
4117 * an existing pool, we don't want to attach the spares & cache
4122 * Load any hot spares for this pool.
4124 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object,
4126 if (error != 0 && error != ENOENT)
4127 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4128 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
4129 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
4130 if (load_nvlist(spa, spa->spa_spares.sav_object,
4131 &spa->spa_spares.sav_config) != 0) {
4132 spa_load_failed(spa, "error loading spares nvlist");
4133 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4136 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4137 spa_load_spares(spa);
4138 spa_config_exit(spa, SCL_ALL, FTAG);
4139 } else if (error == 0) {
4140 spa->spa_spares.sav_sync = B_TRUE;
4144 * Load any level 2 ARC devices for this pool.
4146 error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
4147 &spa->spa_l2cache.sav_object, B_FALSE);
4148 if (error != 0 && error != ENOENT)
4149 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4150 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
4151 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
4152 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
4153 &spa->spa_l2cache.sav_config) != 0) {
4154 spa_load_failed(spa, "error loading l2cache nvlist");
4155 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4158 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4159 spa_load_l2cache(spa);
4160 spa_config_exit(spa, SCL_ALL, FTAG);
4161 } else if (error == 0) {
4162 spa->spa_l2cache.sav_sync = B_TRUE;
4169 spa_ld_load_vdev_metadata(spa_t *spa)
4172 vdev_t *rvd = spa->spa_root_vdev;
4175 * If the 'multihost' property is set, then never allow a pool to
4176 * be imported when the system hostid is zero. The exception to
4177 * this rule is zdb which is always allowed to access pools.
4179 if (spa_multihost(spa) && spa_get_hostid(spa) == 0 &&
4180 (spa->spa_import_flags & ZFS_IMPORT_SKIP_MMP) == 0) {
4181 fnvlist_add_uint64(spa->spa_load_info,
4182 ZPOOL_CONFIG_MMP_STATE, MMP_STATE_NO_HOSTID);
4183 return (spa_vdev_err(rvd, VDEV_AUX_ACTIVE, EREMOTEIO));
4187 * If the 'autoreplace' property is set, then post a resource notifying
4188 * the ZFS DE that it should not issue any faults for unopenable
4189 * devices. We also iterate over the vdevs, and post a sysevent for any
4190 * unopenable vdevs so that the normal autoreplace handler can take
4193 if (spa->spa_autoreplace && spa->spa_load_state != SPA_LOAD_TRYIMPORT) {
4194 spa_check_removed(spa->spa_root_vdev);
4196 * For the import case, this is done in spa_import(), because
4197 * at this point we're using the spare definitions from
4198 * the MOS config, not necessarily from the userland config.
4200 if (spa->spa_load_state != SPA_LOAD_IMPORT) {
4201 spa_aux_check_removed(&spa->spa_spares);
4202 spa_aux_check_removed(&spa->spa_l2cache);
4207 * Load the vdev metadata such as metaslabs, DTLs, spacemap object, etc.
4209 error = vdev_load(rvd);
4211 spa_load_failed(spa, "vdev_load failed [error=%d]", error);
4212 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, error));
4215 error = spa_ld_log_spacemaps(spa);
4217 spa_load_failed(spa, "spa_ld_log_sm_data failed [error=%d]",
4219 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, error));
4223 * Propagate the leaf DTLs we just loaded all the way up the vdev tree.
4225 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4226 vdev_dtl_reassess(rvd, 0, 0, B_FALSE, B_FALSE);
4227 spa_config_exit(spa, SCL_ALL, FTAG);
4233 spa_ld_load_dedup_tables(spa_t *spa)
4236 vdev_t *rvd = spa->spa_root_vdev;
4238 error = ddt_load(spa);
4240 spa_load_failed(spa, "ddt_load failed [error=%d]", error);
4241 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4248 spa_ld_verify_logs(spa_t *spa, spa_import_type_t type, char **ereport)
4250 vdev_t *rvd = spa->spa_root_vdev;
4252 if (type != SPA_IMPORT_ASSEMBLE && spa_writeable(spa)) {
4253 boolean_t missing = spa_check_logs(spa);
4255 if (spa->spa_missing_tvds != 0) {
4256 spa_load_note(spa, "spa_check_logs failed "
4257 "so dropping the logs");
4259 *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
4260 spa_load_failed(spa, "spa_check_logs failed");
4261 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG,
4271 spa_ld_verify_pool_data(spa_t *spa)
4274 vdev_t *rvd = spa->spa_root_vdev;
4277 * We've successfully opened the pool, verify that we're ready
4278 * to start pushing transactions.
4280 if (spa->spa_load_state != SPA_LOAD_TRYIMPORT) {
4281 error = spa_load_verify(spa);
4283 spa_load_failed(spa, "spa_load_verify failed "
4284 "[error=%d]", error);
4285 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
4294 spa_ld_claim_log_blocks(spa_t *spa)
4297 dsl_pool_t *dp = spa_get_dsl(spa);
4300 * Claim log blocks that haven't been committed yet.
4301 * This must all happen in a single txg.
4302 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
4303 * invoked from zil_claim_log_block()'s i/o done callback.
4304 * Price of rollback is that we abandon the log.
4306 spa->spa_claiming = B_TRUE;
4308 tx = dmu_tx_create_assigned(dp, spa_first_txg(spa));
4309 (void) dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
4310 zil_claim, tx, DS_FIND_CHILDREN);
4313 spa->spa_claiming = B_FALSE;
4315 spa_set_log_state(spa, SPA_LOG_GOOD);
4319 spa_ld_check_for_config_update(spa_t *spa, uint64_t config_cache_txg,
4320 boolean_t update_config_cache)
4322 vdev_t *rvd = spa->spa_root_vdev;
4323 int need_update = B_FALSE;
4326 * If the config cache is stale, or we have uninitialized
4327 * metaslabs (see spa_vdev_add()), then update the config.
4329 * If this is a verbatim import, trust the current
4330 * in-core spa_config and update the disk labels.
4332 if (update_config_cache || config_cache_txg != spa->spa_config_txg ||
4333 spa->spa_load_state == SPA_LOAD_IMPORT ||
4334 spa->spa_load_state == SPA_LOAD_RECOVER ||
4335 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
4336 need_update = B_TRUE;
4338 for (int c = 0; c < rvd->vdev_children; c++)
4339 if (rvd->vdev_child[c]->vdev_ms_array == 0)
4340 need_update = B_TRUE;
4343 * Update the config cache asynchronously in case we're the
4344 * root pool, in which case the config cache isn't writable yet.
4347 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
4351 spa_ld_prepare_for_reload(spa_t *spa)
4353 spa_mode_t mode = spa->spa_mode;
4354 int async_suspended = spa->spa_async_suspended;
4357 spa_deactivate(spa);
4358 spa_activate(spa, mode);
4361 * We save the value of spa_async_suspended as it gets reset to 0 by
4362 * spa_unload(). We want to restore it back to the original value before
4363 * returning as we might be calling spa_async_resume() later.
4365 spa->spa_async_suspended = async_suspended;
4369 spa_ld_read_checkpoint_txg(spa_t *spa)
4371 uberblock_t checkpoint;
4374 ASSERT0(spa->spa_checkpoint_txg);
4375 ASSERT(MUTEX_HELD(&spa_namespace_lock));
4377 error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
4378 DMU_POOL_ZPOOL_CHECKPOINT, sizeof (uint64_t),
4379 sizeof (uberblock_t) / sizeof (uint64_t), &checkpoint);
4381 if (error == ENOENT)
4387 ASSERT3U(checkpoint.ub_txg, !=, 0);
4388 ASSERT3U(checkpoint.ub_checkpoint_txg, !=, 0);
4389 ASSERT3U(checkpoint.ub_timestamp, !=, 0);
4390 spa->spa_checkpoint_txg = checkpoint.ub_txg;
4391 spa->spa_checkpoint_info.sci_timestamp = checkpoint.ub_timestamp;
4397 spa_ld_mos_init(spa_t *spa, spa_import_type_t type)
4401 ASSERT(MUTEX_HELD(&spa_namespace_lock));
4402 ASSERT(spa->spa_config_source != SPA_CONFIG_SRC_NONE);
4405 * Never trust the config that is provided unless we are assembling
4406 * a pool following a split.
4407 * This means don't trust blkptrs and the vdev tree in general. This
4408 * also effectively puts the spa in read-only mode since
4409 * spa_writeable() checks for spa_trust_config to be true.
4410 * We will later load a trusted config from the MOS.
4412 if (type != SPA_IMPORT_ASSEMBLE)
4413 spa->spa_trust_config = B_FALSE;
4416 * Parse the config provided to create a vdev tree.
4418 error = spa_ld_parse_config(spa, type);
4422 spa_import_progress_add(spa);
4425 * Now that we have the vdev tree, try to open each vdev. This involves
4426 * opening the underlying physical device, retrieving its geometry and
4427 * probing the vdev with a dummy I/O. The state of each vdev will be set
4428 * based on the success of those operations. After this we'll be ready
4429 * to read from the vdevs.
4431 error = spa_ld_open_vdevs(spa);
4436 * Read the label of each vdev and make sure that the GUIDs stored
4437 * there match the GUIDs in the config provided.
4438 * If we're assembling a new pool that's been split off from an
4439 * existing pool, the labels haven't yet been updated so we skip
4440 * validation for now.
4442 if (type != SPA_IMPORT_ASSEMBLE) {
4443 error = spa_ld_validate_vdevs(spa);
4449 * Read all vdev labels to find the best uberblock (i.e. latest,
4450 * unless spa_load_max_txg is set) and store it in spa_uberblock. We
4451 * get the list of features required to read blkptrs in the MOS from
4452 * the vdev label with the best uberblock and verify that our version
4453 * of zfs supports them all.
4455 error = spa_ld_select_uberblock(spa, type);
4460 * Pass that uberblock to the dsl_pool layer which will open the root
4461 * blkptr. This blkptr points to the latest version of the MOS and will
4462 * allow us to read its contents.
4464 error = spa_ld_open_rootbp(spa);
4472 spa_ld_checkpoint_rewind(spa_t *spa)
4474 uberblock_t checkpoint;
4477 ASSERT(MUTEX_HELD(&spa_namespace_lock));
4478 ASSERT(spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT);
4480 error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
4481 DMU_POOL_ZPOOL_CHECKPOINT, sizeof (uint64_t),
4482 sizeof (uberblock_t) / sizeof (uint64_t), &checkpoint);
4485 spa_load_failed(spa, "unable to retrieve checkpointed "
4486 "uberblock from the MOS config [error=%d]", error);
4488 if (error == ENOENT)
4489 error = ZFS_ERR_NO_CHECKPOINT;
4494 ASSERT3U(checkpoint.ub_txg, <, spa->spa_uberblock.ub_txg);
4495 ASSERT3U(checkpoint.ub_txg, ==, checkpoint.ub_checkpoint_txg);
4498 * We need to update the txg and timestamp of the checkpointed
4499 * uberblock to be higher than the latest one. This ensures that
4500 * the checkpointed uberblock is selected if we were to close and
4501 * reopen the pool right after we've written it in the vdev labels.
4502 * (also see block comment in vdev_uberblock_compare)
4504 checkpoint.ub_txg = spa->spa_uberblock.ub_txg + 1;
4505 checkpoint.ub_timestamp = gethrestime_sec();
4508 * Set current uberblock to be the checkpointed uberblock.
4510 spa->spa_uberblock = checkpoint;
4513 * If we are doing a normal rewind, then the pool is open for
4514 * writing and we sync the "updated" checkpointed uberblock to
4515 * disk. Once this is done, we've basically rewound the whole
4516 * pool and there is no way back.
4518 * There are cases when we don't want to attempt and sync the
4519 * checkpointed uberblock to disk because we are opening a
4520 * pool as read-only. Specifically, verifying the checkpointed
4521 * state with zdb, and importing the checkpointed state to get
4522 * a "preview" of its content.
4524 if (spa_writeable(spa)) {
4525 vdev_t *rvd = spa->spa_root_vdev;
4527 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4528 vdev_t *svd[SPA_SYNC_MIN_VDEVS] = { NULL };
4530 int children = rvd->vdev_children;
4531 int c0 = spa_get_random(children);
4533 for (int c = 0; c < children; c++) {
4534 vdev_t *vd = rvd->vdev_child[(c0 + c) % children];
4536 /* Stop when revisiting the first vdev */
4537 if (c > 0 && svd[0] == vd)
4540 if (vd->vdev_ms_array == 0 || vd->vdev_islog ||
4541 !vdev_is_concrete(vd))
4544 svd[svdcount++] = vd;
4545 if (svdcount == SPA_SYNC_MIN_VDEVS)
4548 error = vdev_config_sync(svd, svdcount, spa->spa_first_txg);
4550 spa->spa_last_synced_guid = rvd->vdev_guid;
4551 spa_config_exit(spa, SCL_ALL, FTAG);
4554 spa_load_failed(spa, "failed to write checkpointed "
4555 "uberblock to the vdev labels [error=%d]", error);
4564 spa_ld_mos_with_trusted_config(spa_t *spa, spa_import_type_t type,
4565 boolean_t *update_config_cache)
4570 * Parse the config for pool, open and validate vdevs,
4571 * select an uberblock, and use that uberblock to open
4574 error = spa_ld_mos_init(spa, type);
4579 * Retrieve the trusted config stored in the MOS and use it to create
4580 * a new, exact version of the vdev tree, then reopen all vdevs.
4582 error = spa_ld_trusted_config(spa, type, B_FALSE);
4583 if (error == EAGAIN) {
4584 if (update_config_cache != NULL)
4585 *update_config_cache = B_TRUE;
4588 * Redo the loading process with the trusted config if it is
4589 * too different from the untrusted config.
4591 spa_ld_prepare_for_reload(spa);
4592 spa_load_note(spa, "RELOADING");
4593 error = spa_ld_mos_init(spa, type);
4597 error = spa_ld_trusted_config(spa, type, B_TRUE);
4601 } else if (error != 0) {
4609 * Load an existing storage pool, using the config provided. This config
4610 * describes which vdevs are part of the pool and is later validated against
4611 * partial configs present in each vdev's label and an entire copy of the
4612 * config stored in the MOS.
4615 spa_load_impl(spa_t *spa, spa_import_type_t type, char **ereport)
4618 boolean_t missing_feat_write = B_FALSE;
4619 boolean_t checkpoint_rewind =
4620 (spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT);
4621 boolean_t update_config_cache = B_FALSE;
4623 ASSERT(MUTEX_HELD(&spa_namespace_lock));
4624 ASSERT(spa->spa_config_source != SPA_CONFIG_SRC_NONE);
4626 spa_load_note(spa, "LOADING");
4628 error = spa_ld_mos_with_trusted_config(spa, type, &update_config_cache);
4633 * If we are rewinding to the checkpoint then we need to repeat
4634 * everything we've done so far in this function but this time
4635 * selecting the checkpointed uberblock and using that to open
4638 if (checkpoint_rewind) {
4640 * If we are rewinding to the checkpoint update config cache
4643 update_config_cache = B_TRUE;
4646 * Extract the checkpointed uberblock from the current MOS
4647 * and use this as the pool's uberblock from now on. If the
4648 * pool is imported as writeable we also write the checkpoint
4649 * uberblock to the labels, making the rewind permanent.
4651 error = spa_ld_checkpoint_rewind(spa);
4656 * Redo the loading process again with the
4657 * checkpointed uberblock.
4659 spa_ld_prepare_for_reload(spa);
4660 spa_load_note(spa, "LOADING checkpointed uberblock");
4661 error = spa_ld_mos_with_trusted_config(spa, type, NULL);
4667 * Retrieve the checkpoint txg if the pool has a checkpoint.
4669 error = spa_ld_read_checkpoint_txg(spa);
4674 * Retrieve the mapping of indirect vdevs. Those vdevs were removed
4675 * from the pool and their contents were re-mapped to other vdevs. Note
4676 * that everything that we read before this step must have been
4677 * rewritten on concrete vdevs after the last device removal was
4678 * initiated. Otherwise we could be reading from indirect vdevs before
4679 * we have loaded their mappings.
4681 error = spa_ld_open_indirect_vdev_metadata(spa);
4686 * Retrieve the full list of active features from the MOS and check if
4687 * they are all supported.
4689 error = spa_ld_check_features(spa, &missing_feat_write);
4694 * Load several special directories from the MOS needed by the dsl_pool
4697 error = spa_ld_load_special_directories(spa);
4702 * Retrieve pool properties from the MOS.
4704 error = spa_ld_get_props(spa);
4709 * Retrieve the list of auxiliary devices - cache devices and spares -
4712 error = spa_ld_open_aux_vdevs(spa, type);
4717 * Load the metadata for all vdevs. Also check if unopenable devices
4718 * should be autoreplaced.
4720 error = spa_ld_load_vdev_metadata(spa);
4724 error = spa_ld_load_dedup_tables(spa);
4729 * Verify the logs now to make sure we don't have any unexpected errors
4730 * when we claim log blocks later.
4732 error = spa_ld_verify_logs(spa, type, ereport);
4736 if (missing_feat_write) {
4737 ASSERT(spa->spa_load_state == SPA_LOAD_TRYIMPORT);
4740 * At this point, we know that we can open the pool in
4741 * read-only mode but not read-write mode. We now have enough
4742 * information and can return to userland.
4744 return (spa_vdev_err(spa->spa_root_vdev, VDEV_AUX_UNSUP_FEAT,
4749 * Traverse the last txgs to make sure the pool was left off in a safe
4750 * state. When performing an extreme rewind, we verify the whole pool,
4751 * which can take a very long time.
4753 error = spa_ld_verify_pool_data(spa);
4758 * Calculate the deflated space for the pool. This must be done before
4759 * we write anything to the pool because we'd need to update the space
4760 * accounting using the deflated sizes.
4762 spa_update_dspace(spa);
4765 * We have now retrieved all the information we needed to open the
4766 * pool. If we are importing the pool in read-write mode, a few
4767 * additional steps must be performed to finish the import.
4769 if (spa_writeable(spa) && (spa->spa_load_state == SPA_LOAD_RECOVER ||
4770 spa->spa_load_max_txg == UINT64_MAX)) {
4771 uint64_t config_cache_txg = spa->spa_config_txg;
4773 ASSERT(spa->spa_load_state != SPA_LOAD_TRYIMPORT);
4776 * In case of a checkpoint rewind, log the original txg
4777 * of the checkpointed uberblock.
4779 if (checkpoint_rewind) {
4780 spa_history_log_internal(spa, "checkpoint rewind",
4781 NULL, "rewound state to txg=%llu",
4782 (u_longlong_t)spa->spa_uberblock.ub_checkpoint_txg);
4786 * Traverse the ZIL and claim all blocks.
4788 spa_ld_claim_log_blocks(spa);
4791 * Kick-off the syncing thread.
4793 spa->spa_sync_on = B_TRUE;
4794 txg_sync_start(spa->spa_dsl_pool);
4795 mmp_thread_start(spa);
4798 * Wait for all claims to sync. We sync up to the highest
4799 * claimed log block birth time so that claimed log blocks
4800 * don't appear to be from the future. spa_claim_max_txg
4801 * will have been set for us by ZIL traversal operations
4804 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
4807 * Check if we need to request an update of the config. On the
4808 * next sync, we would update the config stored in vdev labels
4809 * and the cachefile (by default /etc/zfs/zpool.cache).
4811 spa_ld_check_for_config_update(spa, config_cache_txg,
4812 update_config_cache);
4815 * Check if a rebuild was in progress and if so resume it.
4816 * Then check all DTLs to see if anything needs resilvering.
4817 * The resilver will be deferred if a rebuild was started.
4819 if (vdev_rebuild_active(spa->spa_root_vdev)) {
4820 vdev_rebuild_restart(spa);
4821 } else if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
4822 vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
4823 spa_async_request(spa, SPA_ASYNC_RESILVER);
4827 * Log the fact that we booted up (so that we can detect if
4828 * we rebooted in the middle of an operation).
4830 spa_history_log_version(spa, "open", NULL);
4832 spa_restart_removal(spa);
4833 spa_spawn_aux_threads(spa);
4836 * Delete any inconsistent datasets.
4839 * Since we may be issuing deletes for clones here,
4840 * we make sure to do so after we've spawned all the
4841 * auxiliary threads above (from which the livelist
4842 * deletion zthr is part of).
4844 (void) dmu_objset_find(spa_name(spa),
4845 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
4848 * Clean up any stale temporary dataset userrefs.
4850 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
4852 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4853 vdev_initialize_restart(spa->spa_root_vdev);
4854 vdev_trim_restart(spa->spa_root_vdev);
4855 vdev_autotrim_restart(spa);
4856 spa_config_exit(spa, SCL_CONFIG, FTAG);
4859 spa_import_progress_remove(spa_guid(spa));
4860 spa_async_request(spa, SPA_ASYNC_L2CACHE_REBUILD);
4862 spa_load_note(spa, "LOADED");
4868 spa_load_retry(spa_t *spa, spa_load_state_t state)
4870 spa_mode_t mode = spa->spa_mode;
4873 spa_deactivate(spa);
4875 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg - 1;
4877 spa_activate(spa, mode);
4878 spa_async_suspend(spa);
4880 spa_load_note(spa, "spa_load_retry: rewind, max txg: %llu",
4881 (u_longlong_t)spa->spa_load_max_txg);
4883 return (spa_load(spa, state, SPA_IMPORT_EXISTING));
4887 * If spa_load() fails this function will try loading prior txg's. If
4888 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
4889 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
4890 * function will not rewind the pool and will return the same error as
4894 spa_load_best(spa_t *spa, spa_load_state_t state, uint64_t max_request,
4897 nvlist_t *loadinfo = NULL;
4898 nvlist_t *config = NULL;
4899 int load_error, rewind_error;
4900 uint64_t safe_rewind_txg;
4903 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
4904 spa->spa_load_max_txg = spa->spa_load_txg;
4905 spa_set_log_state(spa, SPA_LOG_CLEAR);
4907 spa->spa_load_max_txg = max_request;
4908 if (max_request != UINT64_MAX)
4909 spa->spa_extreme_rewind = B_TRUE;
4912 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING);
4913 if (load_error == 0)
4915 if (load_error == ZFS_ERR_NO_CHECKPOINT) {
4917 * When attempting checkpoint-rewind on a pool with no
4918 * checkpoint, we should not attempt to load uberblocks
4919 * from previous txgs when spa_load fails.
4921 ASSERT(spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT);
4922 spa_import_progress_remove(spa_guid(spa));
4923 return (load_error);
4926 if (spa->spa_root_vdev != NULL)
4927 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4929 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
4930 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
4932 if (rewind_flags & ZPOOL_NEVER_REWIND) {
4933 nvlist_free(config);
4934 spa_import_progress_remove(spa_guid(spa));
4935 return (load_error);
4938 if (state == SPA_LOAD_RECOVER) {
4939 /* Price of rolling back is discarding txgs, including log */
4940 spa_set_log_state(spa, SPA_LOG_CLEAR);
4943 * If we aren't rolling back save the load info from our first
4944 * import attempt so that we can restore it after attempting
4947 loadinfo = spa->spa_load_info;
4948 spa->spa_load_info = fnvlist_alloc();
4951 spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
4952 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
4953 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
4954 TXG_INITIAL : safe_rewind_txg;
4957 * Continue as long as we're finding errors, we're still within
4958 * the acceptable rewind range, and we're still finding uberblocks
4960 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
4961 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
4962 if (spa->spa_load_max_txg < safe_rewind_txg)
4963 spa->spa_extreme_rewind = B_TRUE;
4964 rewind_error = spa_load_retry(spa, state);
4967 spa->spa_extreme_rewind = B_FALSE;
4968 spa->spa_load_max_txg = UINT64_MAX;
4970 if (config && (rewind_error || state != SPA_LOAD_RECOVER))
4971 spa_config_set(spa, config);
4973 nvlist_free(config);
4975 if (state == SPA_LOAD_RECOVER) {
4976 ASSERT3P(loadinfo, ==, NULL);
4977 spa_import_progress_remove(spa_guid(spa));
4978 return (rewind_error);
4980 /* Store the rewind info as part of the initial load info */
4981 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
4982 spa->spa_load_info);
4984 /* Restore the initial load info */
4985 fnvlist_free(spa->spa_load_info);
4986 spa->spa_load_info = loadinfo;
4988 spa_import_progress_remove(spa_guid(spa));
4989 return (load_error);
4996 * The import case is identical to an open except that the configuration is sent
4997 * down from userland, instead of grabbed from the configuration cache. For the
4998 * case of an open, the pool configuration will exist in the
4999 * POOL_STATE_UNINITIALIZED state.
5001 * The stats information (gen/count/ustats) is used to gather vdev statistics at
5002 * the same time open the pool, without having to keep around the spa_t in some
5006 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
5010 spa_load_state_t state = SPA_LOAD_OPEN;
5012 int locked = B_FALSE;
5013 int firstopen = B_FALSE;
5018 * As disgusting as this is, we need to support recursive calls to this
5019 * function because dsl_dir_open() is called during spa_load(), and ends
5020 * up calling spa_open() again. The real fix is to figure out how to
5021 * avoid dsl_dir_open() calling this in the first place.
5023 if (MUTEX_NOT_HELD(&spa_namespace_lock)) {
5024 mutex_enter(&spa_namespace_lock);
5028 if ((spa = spa_lookup(pool)) == NULL) {
5030 mutex_exit(&spa_namespace_lock);
5031 return (SET_ERROR(ENOENT));
5034 if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
5035 zpool_load_policy_t policy;
5039 zpool_get_load_policy(nvpolicy ? nvpolicy : spa->spa_config,
5041 if (policy.zlp_rewind & ZPOOL_DO_REWIND)
5042 state = SPA_LOAD_RECOVER;
5044 spa_activate(spa, spa_mode_global);
5046 if (state != SPA_LOAD_RECOVER)
5047 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
5048 spa->spa_config_source = SPA_CONFIG_SRC_CACHEFILE;
5050 zfs_dbgmsg("spa_open_common: opening %s", pool);
5051 error = spa_load_best(spa, state, policy.zlp_txg,
5054 if (error == EBADF) {
5056 * If vdev_validate() returns failure (indicated by
5057 * EBADF), it indicates that one of the vdevs indicates
5058 * that the pool has been exported or destroyed. If
5059 * this is the case, the config cache is out of sync and
5060 * we should remove the pool from the namespace.
5063 spa_deactivate(spa);
5064 spa_write_cachefile(spa, B_TRUE, B_TRUE);
5067 mutex_exit(&spa_namespace_lock);
5068 return (SET_ERROR(ENOENT));
5073 * We can't open the pool, but we still have useful
5074 * information: the state of each vdev after the
5075 * attempted vdev_open(). Return this to the user.
5077 if (config != NULL && spa->spa_config) {
5078 VERIFY(nvlist_dup(spa->spa_config, config,
5080 VERIFY(nvlist_add_nvlist(*config,
5081 ZPOOL_CONFIG_LOAD_INFO,
5082 spa->spa_load_info) == 0);
5085 spa_deactivate(spa);
5086 spa->spa_last_open_failed = error;
5088 mutex_exit(&spa_namespace_lock);
5094 spa_open_ref(spa, tag);
5097 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
5100 * If we've recovered the pool, pass back any information we
5101 * gathered while doing the load.
5103 if (state == SPA_LOAD_RECOVER) {
5104 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
5105 spa->spa_load_info) == 0);
5109 spa->spa_last_open_failed = 0;
5110 spa->spa_last_ubsync_txg = 0;
5111 spa->spa_load_txg = 0;
5112 mutex_exit(&spa_namespace_lock);
5116 zvol_create_minors_recursive(spa_name(spa));
5124 spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
5127 return (spa_open_common(name, spapp, tag, policy, config));
5131 spa_open(const char *name, spa_t **spapp, void *tag)
5133 return (spa_open_common(name, spapp, tag, NULL, NULL));
5137 * Lookup the given spa_t, incrementing the inject count in the process,
5138 * preventing it from being exported or destroyed.
5141 spa_inject_addref(char *name)
5145 mutex_enter(&spa_namespace_lock);
5146 if ((spa = spa_lookup(name)) == NULL) {
5147 mutex_exit(&spa_namespace_lock);
5150 spa->spa_inject_ref++;
5151 mutex_exit(&spa_namespace_lock);
5157 spa_inject_delref(spa_t *spa)
5159 mutex_enter(&spa_namespace_lock);
5160 spa->spa_inject_ref--;
5161 mutex_exit(&spa_namespace_lock);
5165 * Add spares device information to the nvlist.
5168 spa_add_spares(spa_t *spa, nvlist_t *config)
5178 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
5180 if (spa->spa_spares.sav_count == 0)
5183 VERIFY(nvlist_lookup_nvlist(config,
5184 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
5185 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5186 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
5188 VERIFY(nvlist_add_nvlist_array(nvroot,
5189 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
5190 VERIFY(nvlist_lookup_nvlist_array(nvroot,
5191 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
5194 * Go through and find any spares which have since been
5195 * repurposed as an active spare. If this is the case, update
5196 * their status appropriately.
5198 for (i = 0; i < nspares; i++) {
5199 VERIFY(nvlist_lookup_uint64(spares[i],
5200 ZPOOL_CONFIG_GUID, &guid) == 0);
5201 if (spa_spare_exists(guid, &pool, NULL) &&
5203 VERIFY(nvlist_lookup_uint64_array(
5204 spares[i], ZPOOL_CONFIG_VDEV_STATS,
5205 (uint64_t **)&vs, &vsc) == 0);
5206 vs->vs_state = VDEV_STATE_CANT_OPEN;
5207 vs->vs_aux = VDEV_AUX_SPARED;
5214 * Add l2cache device information to the nvlist, including vdev stats.
5217 spa_add_l2cache(spa_t *spa, nvlist_t *config)
5220 uint_t i, j, nl2cache;
5227 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
5229 if (spa->spa_l2cache.sav_count == 0)
5232 VERIFY(nvlist_lookup_nvlist(config,
5233 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
5234 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5235 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
5236 if (nl2cache != 0) {
5237 VERIFY(nvlist_add_nvlist_array(nvroot,
5238 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
5239 VERIFY(nvlist_lookup_nvlist_array(nvroot,
5240 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
5243 * Update level 2 cache device stats.
5246 for (i = 0; i < nl2cache; i++) {
5247 VERIFY(nvlist_lookup_uint64(l2cache[i],
5248 ZPOOL_CONFIG_GUID, &guid) == 0);
5251 for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
5253 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
5254 vd = spa->spa_l2cache.sav_vdevs[j];
5260 VERIFY(nvlist_lookup_uint64_array(l2cache[i],
5261 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
5263 vdev_get_stats(vd, vs);
5264 vdev_config_generate_stats(vd, l2cache[i]);
5271 spa_feature_stats_from_disk(spa_t *spa, nvlist_t *features)
5276 if (spa->spa_feat_for_read_obj != 0) {
5277 for (zap_cursor_init(&zc, spa->spa_meta_objset,
5278 spa->spa_feat_for_read_obj);
5279 zap_cursor_retrieve(&zc, &za) == 0;
5280 zap_cursor_advance(&zc)) {
5281 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
5282 za.za_num_integers == 1);
5283 VERIFY0(nvlist_add_uint64(features, za.za_name,
5284 za.za_first_integer));
5286 zap_cursor_fini(&zc);
5289 if (spa->spa_feat_for_write_obj != 0) {
5290 for (zap_cursor_init(&zc, spa->spa_meta_objset,
5291 spa->spa_feat_for_write_obj);
5292 zap_cursor_retrieve(&zc, &za) == 0;
5293 zap_cursor_advance(&zc)) {
5294 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
5295 za.za_num_integers == 1);
5296 VERIFY0(nvlist_add_uint64(features, za.za_name,
5297 za.za_first_integer));
5299 zap_cursor_fini(&zc);
5304 spa_feature_stats_from_cache(spa_t *spa, nvlist_t *features)
5308 for (i = 0; i < SPA_FEATURES; i++) {
5309 zfeature_info_t feature = spa_feature_table[i];
5312 if (feature_get_refcount(spa, &feature, &refcount) != 0)
5315 VERIFY0(nvlist_add_uint64(features, feature.fi_guid, refcount));
5320 * Store a list of pool features and their reference counts in the
5323 * The first time this is called on a spa, allocate a new nvlist, fetch
5324 * the pool features and reference counts from disk, then save the list
5325 * in the spa. In subsequent calls on the same spa use the saved nvlist
5326 * and refresh its values from the cached reference counts. This
5327 * ensures we don't block here on I/O on a suspended pool so 'zpool
5328 * clear' can resume the pool.
5331 spa_add_feature_stats(spa_t *spa, nvlist_t *config)
5335 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
5337 mutex_enter(&spa->spa_feat_stats_lock);
5338 features = spa->spa_feat_stats;
5340 if (features != NULL) {
5341 spa_feature_stats_from_cache(spa, features);
5343 VERIFY0(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP));
5344 spa->spa_feat_stats = features;
5345 spa_feature_stats_from_disk(spa, features);
5348 VERIFY0(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
5351 mutex_exit(&spa->spa_feat_stats_lock);
5355 spa_get_stats(const char *name, nvlist_t **config,
5356 char *altroot, size_t buflen)
5362 error = spa_open_common(name, &spa, FTAG, NULL, config);
5366 * This still leaves a window of inconsistency where the spares
5367 * or l2cache devices could change and the config would be
5368 * self-inconsistent.
5370 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5372 if (*config != NULL) {
5373 uint64_t loadtimes[2];
5375 loadtimes[0] = spa->spa_loaded_ts.tv_sec;
5376 loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
5377 VERIFY(nvlist_add_uint64_array(*config,
5378 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
5380 VERIFY(nvlist_add_uint64(*config,
5381 ZPOOL_CONFIG_ERRCOUNT,
5382 spa_get_errlog_size(spa)) == 0);
5384 if (spa_suspended(spa)) {
5385 VERIFY(nvlist_add_uint64(*config,
5386 ZPOOL_CONFIG_SUSPENDED,
5387 spa->spa_failmode) == 0);
5388 VERIFY(nvlist_add_uint64(*config,
5389 ZPOOL_CONFIG_SUSPENDED_REASON,
5390 spa->spa_suspended) == 0);
5393 spa_add_spares(spa, *config);
5394 spa_add_l2cache(spa, *config);
5395 spa_add_feature_stats(spa, *config);
5400 * We want to get the alternate root even for faulted pools, so we cheat
5401 * and call spa_lookup() directly.
5405 mutex_enter(&spa_namespace_lock);
5406 spa = spa_lookup(name);
5408 spa_altroot(spa, altroot, buflen);
5412 mutex_exit(&spa_namespace_lock);
5414 spa_altroot(spa, altroot, buflen);
5419 spa_config_exit(spa, SCL_CONFIG, FTAG);
5420 spa_close(spa, FTAG);
5427 * Validate that the auxiliary device array is well formed. We must have an
5428 * array of nvlists, each which describes a valid leaf vdev. If this is an
5429 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
5430 * specified, as long as they are well-formed.
5433 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
5434 spa_aux_vdev_t *sav, const char *config, uint64_t version,
5435 vdev_labeltype_t label)
5442 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5445 * It's acceptable to have no devs specified.
5447 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
5451 return (SET_ERROR(EINVAL));
5454 * Make sure the pool is formatted with a version that supports this
5457 if (spa_version(spa) < version)
5458 return (SET_ERROR(ENOTSUP));
5461 * Set the pending device list so we correctly handle device in-use
5464 sav->sav_pending = dev;
5465 sav->sav_npending = ndev;
5467 for (i = 0; i < ndev; i++) {
5468 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
5472 if (!vd->vdev_ops->vdev_op_leaf) {
5474 error = SET_ERROR(EINVAL);
5480 if ((error = vdev_open(vd)) == 0 &&
5481 (error = vdev_label_init(vd, crtxg, label)) == 0) {
5482 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
5483 vd->vdev_guid) == 0);
5489 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
5496 sav->sav_pending = NULL;
5497 sav->sav_npending = 0;
5502 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
5506 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5508 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
5509 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
5510 VDEV_LABEL_SPARE)) != 0) {
5514 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
5515 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
5516 VDEV_LABEL_L2CACHE));
5520 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
5525 if (sav->sav_config != NULL) {
5531 * Generate new dev list by concatenating with the
5534 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
5535 &olddevs, &oldndevs) == 0);
5537 newdevs = kmem_alloc(sizeof (void *) *
5538 (ndevs + oldndevs), KM_SLEEP);
5539 for (i = 0; i < oldndevs; i++)
5540 VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
5542 for (i = 0; i < ndevs; i++)
5543 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
5546 VERIFY(nvlist_remove(sav->sav_config, config,
5547 DATA_TYPE_NVLIST_ARRAY) == 0);
5549 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
5550 config, newdevs, ndevs + oldndevs) == 0);
5551 for (i = 0; i < oldndevs + ndevs; i++)
5552 nvlist_free(newdevs[i]);
5553 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
5556 * Generate a new dev list.
5558 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
5560 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
5566 * Stop and drop level 2 ARC devices
5569 spa_l2cache_drop(spa_t *spa)
5573 spa_aux_vdev_t *sav = &spa->spa_l2cache;
5575 for (i = 0; i < sav->sav_count; i++) {
5578 vd = sav->sav_vdevs[i];
5581 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
5582 pool != 0ULL && l2arc_vdev_present(vd))
5583 l2arc_remove_vdev(vd);
5588 * Verify encryption parameters for spa creation. If we are encrypting, we must
5589 * have the encryption feature flag enabled.
5592 spa_create_check_encryption_params(dsl_crypto_params_t *dcp,
5593 boolean_t has_encryption)
5595 if (dcp->cp_crypt != ZIO_CRYPT_OFF &&
5596 dcp->cp_crypt != ZIO_CRYPT_INHERIT &&
5598 return (SET_ERROR(ENOTSUP));
5600 return (dmu_objset_create_crypt_check(NULL, dcp, NULL));
5607 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
5608 nvlist_t *zplprops, dsl_crypto_params_t *dcp)
5611 char *altroot = NULL;
5616 uint64_t txg = TXG_INITIAL;
5617 nvlist_t **spares, **l2cache;
5618 uint_t nspares, nl2cache;
5619 uint64_t version, obj;
5620 boolean_t has_features;
5621 boolean_t has_encryption;
5622 boolean_t has_allocclass;
5628 if (props == NULL ||
5629 nvlist_lookup_string(props, "tname", &poolname) != 0)
5630 poolname = (char *)pool;
5633 * If this pool already exists, return failure.
5635 mutex_enter(&spa_namespace_lock);
5636 if (spa_lookup(poolname) != NULL) {
5637 mutex_exit(&spa_namespace_lock);
5638 return (SET_ERROR(EEXIST));
5642 * Allocate a new spa_t structure.
5644 nvl = fnvlist_alloc();
5645 fnvlist_add_string(nvl, ZPOOL_CONFIG_POOL_NAME, pool);
5646 (void) nvlist_lookup_string(props,
5647 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
5648 spa = spa_add(poolname, nvl, altroot);
5650 spa_activate(spa, spa_mode_global);
5652 if (props && (error = spa_prop_validate(spa, props))) {
5653 spa_deactivate(spa);
5655 mutex_exit(&spa_namespace_lock);
5660 * Temporary pool names should never be written to disk.
5662 if (poolname != pool)
5663 spa->spa_import_flags |= ZFS_IMPORT_TEMP_NAME;
5665 has_features = B_FALSE;
5666 has_encryption = B_FALSE;
5667 has_allocclass = B_FALSE;
5668 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
5669 elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
5670 if (zpool_prop_feature(nvpair_name(elem))) {
5671 has_features = B_TRUE;
5673 feat_name = strchr(nvpair_name(elem), '@') + 1;
5674 VERIFY0(zfeature_lookup_name(feat_name, &feat));
5675 if (feat == SPA_FEATURE_ENCRYPTION)
5676 has_encryption = B_TRUE;
5677 if (feat == SPA_FEATURE_ALLOCATION_CLASSES)
5678 has_allocclass = B_TRUE;
5682 /* verify encryption params, if they were provided */
5684 error = spa_create_check_encryption_params(dcp, has_encryption);
5686 spa_deactivate(spa);
5688 mutex_exit(&spa_namespace_lock);
5692 if (!has_allocclass && zfs_special_devs(nvroot, NULL)) {
5693 spa_deactivate(spa);
5695 mutex_exit(&spa_namespace_lock);
5699 if (has_features || nvlist_lookup_uint64(props,
5700 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
5701 version = SPA_VERSION;
5703 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
5705 spa->spa_first_txg = txg;
5706 spa->spa_uberblock.ub_txg = txg - 1;
5707 spa->spa_uberblock.ub_version = version;
5708 spa->spa_ubsync = spa->spa_uberblock;
5709 spa->spa_load_state = SPA_LOAD_CREATE;
5710 spa->spa_removing_phys.sr_state = DSS_NONE;
5711 spa->spa_removing_phys.sr_removing_vdev = -1;
5712 spa->spa_removing_phys.sr_prev_indirect_vdev = -1;
5713 spa->spa_indirect_vdevs_loaded = B_TRUE;
5716 * Create "The Godfather" zio to hold all async IOs
5718 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
5720 for (int i = 0; i < max_ncpus; i++) {
5721 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
5722 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
5723 ZIO_FLAG_GODFATHER);
5727 * Create the root vdev.
5729 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5731 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
5733 ASSERT(error != 0 || rvd != NULL);
5734 ASSERT(error != 0 || spa->spa_root_vdev == rvd);
5736 if (error == 0 && !zfs_allocatable_devs(nvroot))
5737 error = SET_ERROR(EINVAL);
5740 (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
5741 (error = spa_validate_aux(spa, nvroot, txg,
5742 VDEV_ALLOC_ADD)) == 0) {
5744 * instantiate the metaslab groups (this will dirty the vdevs)
5745 * we can no longer error exit past this point
5747 for (int c = 0; error == 0 && c < rvd->vdev_children; c++) {
5748 vdev_t *vd = rvd->vdev_child[c];
5750 vdev_ashift_optimize(vd);
5751 vdev_metaslab_set_size(vd);
5752 vdev_expand(vd, txg);
5756 spa_config_exit(spa, SCL_ALL, FTAG);
5760 spa_deactivate(spa);
5762 mutex_exit(&spa_namespace_lock);
5767 * Get the list of spares, if specified.
5769 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
5770 &spares, &nspares) == 0) {
5771 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
5773 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
5774 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
5775 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5776 spa_load_spares(spa);
5777 spa_config_exit(spa, SCL_ALL, FTAG);
5778 spa->spa_spares.sav_sync = B_TRUE;
5782 * Get the list of level 2 cache devices, if specified.
5784 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
5785 &l2cache, &nl2cache) == 0) {
5786 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
5787 NV_UNIQUE_NAME, KM_SLEEP) == 0);
5788 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
5789 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
5790 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5791 spa_load_l2cache(spa);
5792 spa_config_exit(spa, SCL_ALL, FTAG);
5793 spa->spa_l2cache.sav_sync = B_TRUE;
5796 spa->spa_is_initializing = B_TRUE;
5797 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, dcp, txg);
5798 spa->spa_is_initializing = B_FALSE;
5801 * Create DDTs (dedup tables).
5805 spa_update_dspace(spa);
5807 tx = dmu_tx_create_assigned(dp, txg);
5810 * Create the pool's history object.
5812 if (version >= SPA_VERSION_ZPOOL_HISTORY && !spa->spa_history)
5813 spa_history_create_obj(spa, tx);
5815 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_CREATE);
5816 spa_history_log_version(spa, "create", tx);
5819 * Create the pool config object.
5821 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
5822 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
5823 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
5825 if (zap_add(spa->spa_meta_objset,
5826 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
5827 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
5828 cmn_err(CE_PANIC, "failed to add pool config");
5831 if (zap_add(spa->spa_meta_objset,
5832 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
5833 sizeof (uint64_t), 1, &version, tx) != 0) {
5834 cmn_err(CE_PANIC, "failed to add pool version");
5837 /* Newly created pools with the right version are always deflated. */
5838 if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
5839 spa->spa_deflate = TRUE;
5840 if (zap_add(spa->spa_meta_objset,
5841 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
5842 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
5843 cmn_err(CE_PANIC, "failed to add deflate");
5848 * Create the deferred-free bpobj. Turn off compression
5849 * because sync-to-convergence takes longer if the blocksize
5852 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
5853 dmu_object_set_compress(spa->spa_meta_objset, obj,
5854 ZIO_COMPRESS_OFF, tx);
5855 if (zap_add(spa->spa_meta_objset,
5856 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
5857 sizeof (uint64_t), 1, &obj, tx) != 0) {
5858 cmn_err(CE_PANIC, "failed to add bpobj");
5860 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
5861 spa->spa_meta_objset, obj));
5864 * Generate some random noise for salted checksums to operate on.
5866 (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes,
5867 sizeof (spa->spa_cksum_salt.zcs_bytes));
5870 * Set pool properties.
5872 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
5873 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
5874 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
5875 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
5876 spa->spa_multihost = zpool_prop_default_numeric(ZPOOL_PROP_MULTIHOST);
5877 spa->spa_autotrim = zpool_prop_default_numeric(ZPOOL_PROP_AUTOTRIM);
5879 if (props != NULL) {
5880 spa_configfile_set(spa, props, B_FALSE);
5881 spa_sync_props(props, tx);
5886 spa->spa_sync_on = B_TRUE;
5888 mmp_thread_start(spa);
5889 txg_wait_synced(dp, txg);
5891 spa_spawn_aux_threads(spa);
5893 spa_write_cachefile(spa, B_FALSE, B_TRUE);
5896 * Don't count references from objsets that are already closed
5897 * and are making their way through the eviction process.
5899 spa_evicting_os_wait(spa);
5900 spa->spa_minref = zfs_refcount_count(&spa->spa_refcount);
5901 spa->spa_load_state = SPA_LOAD_NONE;
5903 mutex_exit(&spa_namespace_lock);
5909 * Import a non-root pool into the system.
5912 spa_import(char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
5915 char *altroot = NULL;
5916 spa_load_state_t state = SPA_LOAD_IMPORT;
5917 zpool_load_policy_t policy;
5918 spa_mode_t mode = spa_mode_global;
5919 uint64_t readonly = B_FALSE;
5922 nvlist_t **spares, **l2cache;
5923 uint_t nspares, nl2cache;
5926 * If a pool with this name exists, return failure.
5928 mutex_enter(&spa_namespace_lock);
5929 if (spa_lookup(pool) != NULL) {
5930 mutex_exit(&spa_namespace_lock);
5931 return (SET_ERROR(EEXIST));
5935 * Create and initialize the spa structure.
5937 (void) nvlist_lookup_string(props,
5938 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
5939 (void) nvlist_lookup_uint64(props,
5940 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
5942 mode = SPA_MODE_READ;
5943 spa = spa_add(pool, config, altroot);
5944 spa->spa_import_flags = flags;
5947 * Verbatim import - Take a pool and insert it into the namespace
5948 * as if it had been loaded at boot.
5950 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
5952 spa_configfile_set(spa, props, B_FALSE);
5954 spa_write_cachefile(spa, B_FALSE, B_TRUE);
5955 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_IMPORT);
5956 zfs_dbgmsg("spa_import: verbatim import of %s", pool);
5957 mutex_exit(&spa_namespace_lock);
5961 spa_activate(spa, mode);
5964 * Don't start async tasks until we know everything is healthy.
5966 spa_async_suspend(spa);
5968 zpool_get_load_policy(config, &policy);
5969 if (policy.zlp_rewind & ZPOOL_DO_REWIND)
5970 state = SPA_LOAD_RECOVER;
5972 spa->spa_config_source = SPA_CONFIG_SRC_TRYIMPORT;
5974 if (state != SPA_LOAD_RECOVER) {
5975 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
5976 zfs_dbgmsg("spa_import: importing %s", pool);
5978 zfs_dbgmsg("spa_import: importing %s, max_txg=%lld "
5979 "(RECOVERY MODE)", pool, (longlong_t)policy.zlp_txg);
5981 error = spa_load_best(spa, state, policy.zlp_txg, policy.zlp_rewind);
5984 * Propagate anything learned while loading the pool and pass it
5985 * back to caller (i.e. rewind info, missing devices, etc).
5987 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
5988 spa->spa_load_info) == 0);
5990 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5992 * Toss any existing sparelist, as it doesn't have any validity
5993 * anymore, and conflicts with spa_has_spare().
5995 if (spa->spa_spares.sav_config) {
5996 nvlist_free(spa->spa_spares.sav_config);
5997 spa->spa_spares.sav_config = NULL;
5998 spa_load_spares(spa);
6000 if (spa->spa_l2cache.sav_config) {
6001 nvlist_free(spa->spa_l2cache.sav_config);
6002 spa->spa_l2cache.sav_config = NULL;
6003 spa_load_l2cache(spa);
6006 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
6008 spa_config_exit(spa, SCL_ALL, FTAG);
6011 spa_configfile_set(spa, props, B_FALSE);
6013 if (error != 0 || (props && spa_writeable(spa) &&
6014 (error = spa_prop_set(spa, props)))) {
6016 spa_deactivate(spa);
6018 mutex_exit(&spa_namespace_lock);
6022 spa_async_resume(spa);
6025 * Override any spares and level 2 cache devices as specified by
6026 * the user, as these may have correct device names/devids, etc.
6028 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
6029 &spares, &nspares) == 0) {
6030 if (spa->spa_spares.sav_config)
6031 VERIFY(nvlist_remove(spa->spa_spares.sav_config,
6032 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
6034 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
6035 NV_UNIQUE_NAME, KM_SLEEP) == 0);
6036 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
6037 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
6038 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6039 spa_load_spares(spa);
6040 spa_config_exit(spa, SCL_ALL, FTAG);
6041 spa->spa_spares.sav_sync = B_TRUE;
6043 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
6044 &l2cache, &nl2cache) == 0) {
6045 if (spa->spa_l2cache.sav_config)
6046 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
6047 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
6049 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
6050 NV_UNIQUE_NAME, KM_SLEEP) == 0);
6051 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
6052 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
6053 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6054 spa_load_l2cache(spa);
6055 spa_config_exit(spa, SCL_ALL, FTAG);
6056 spa->spa_l2cache.sav_sync = B_TRUE;
6060 * Check for any removed devices.
6062 if (spa->spa_autoreplace) {
6063 spa_aux_check_removed(&spa->spa_spares);
6064 spa_aux_check_removed(&spa->spa_l2cache);
6067 if (spa_writeable(spa)) {
6069 * Update the config cache to include the newly-imported pool.
6071 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
6075 * It's possible that the pool was expanded while it was exported.
6076 * We kick off an async task to handle this for us.
6078 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
6080 spa_history_log_version(spa, "import", NULL);
6082 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_IMPORT);
6084 mutex_exit(&spa_namespace_lock);
6086 zvol_create_minors_recursive(pool);
6092 spa_tryimport(nvlist_t *tryconfig)
6094 nvlist_t *config = NULL;
6095 char *poolname, *cachefile;
6099 zpool_load_policy_t policy;
6101 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
6104 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
6108 * Create and initialize the spa structure.
6110 mutex_enter(&spa_namespace_lock);
6111 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
6112 spa_activate(spa, SPA_MODE_READ);
6115 * Rewind pool if a max txg was provided.
6117 zpool_get_load_policy(spa->spa_config, &policy);
6118 if (policy.zlp_txg != UINT64_MAX) {
6119 spa->spa_load_max_txg = policy.zlp_txg;
6120 spa->spa_extreme_rewind = B_TRUE;
6121 zfs_dbgmsg("spa_tryimport: importing %s, max_txg=%lld",
6122 poolname, (longlong_t)policy.zlp_txg);
6124 zfs_dbgmsg("spa_tryimport: importing %s", poolname);
6127 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_CACHEFILE, &cachefile)
6129 zfs_dbgmsg("spa_tryimport: using cachefile '%s'", cachefile);
6130 spa->spa_config_source = SPA_CONFIG_SRC_CACHEFILE;
6132 spa->spa_config_source = SPA_CONFIG_SRC_SCAN;
6135 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING);
6138 * If 'tryconfig' was at least parsable, return the current config.
6140 if (spa->spa_root_vdev != NULL) {
6141 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
6142 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
6144 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
6146 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
6147 spa->spa_uberblock.ub_timestamp) == 0);
6148 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
6149 spa->spa_load_info) == 0);
6150 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_ERRATA,
6151 spa->spa_errata) == 0);
6154 * If the bootfs property exists on this pool then we
6155 * copy it out so that external consumers can tell which
6156 * pools are bootable.
6158 if ((!error || error == EEXIST) && spa->spa_bootfs) {
6159 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
6162 * We have to play games with the name since the
6163 * pool was opened as TRYIMPORT_NAME.
6165 if (dsl_dsobj_to_dsname(spa_name(spa),
6166 spa->spa_bootfs, tmpname) == 0) {
6170 dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
6172 cp = strchr(tmpname, '/');
6174 (void) strlcpy(dsname, tmpname,
6177 (void) snprintf(dsname, MAXPATHLEN,
6178 "%s/%s", poolname, ++cp);
6180 VERIFY(nvlist_add_string(config,
6181 ZPOOL_CONFIG_BOOTFS, dsname) == 0);
6182 kmem_free(dsname, MAXPATHLEN);
6184 kmem_free(tmpname, MAXPATHLEN);
6188 * Add the list of hot spares and level 2 cache devices.
6190 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
6191 spa_add_spares(spa, config);
6192 spa_add_l2cache(spa, config);
6193 spa_config_exit(spa, SCL_CONFIG, FTAG);
6197 spa_deactivate(spa);
6199 mutex_exit(&spa_namespace_lock);
6205 * Pool export/destroy
6207 * The act of destroying or exporting a pool is very simple. We make sure there
6208 * is no more pending I/O and any references to the pool are gone. Then, we
6209 * update the pool state and sync all the labels to disk, removing the
6210 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
6211 * we don't sync the labels or remove the configuration cache.
6214 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
6215 boolean_t force, boolean_t hardforce)
6222 if (!(spa_mode_global & SPA_MODE_WRITE))
6223 return (SET_ERROR(EROFS));
6225 mutex_enter(&spa_namespace_lock);
6226 if ((spa = spa_lookup(pool)) == NULL) {
6227 mutex_exit(&spa_namespace_lock);
6228 return (SET_ERROR(ENOENT));
6231 if (spa->spa_is_exporting) {
6232 /* the pool is being exported by another thread */
6233 mutex_exit(&spa_namespace_lock);
6234 return (SET_ERROR(ZFS_ERR_EXPORT_IN_PROGRESS));
6236 spa->spa_is_exporting = B_TRUE;
6239 * Put a hold on the pool, drop the namespace lock, stop async tasks,
6240 * reacquire the namespace lock, and see if we can export.
6242 spa_open_ref(spa, FTAG);
6243 mutex_exit(&spa_namespace_lock);
6244 spa_async_suspend(spa);
6245 if (spa->spa_zvol_taskq) {
6246 zvol_remove_minors(spa, spa_name(spa), B_TRUE);
6247 taskq_wait(spa->spa_zvol_taskq);
6249 mutex_enter(&spa_namespace_lock);
6250 spa_close(spa, FTAG);
6252 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
6255 * The pool will be in core if it's openable, in which case we can
6256 * modify its state. Objsets may be open only because they're dirty,
6257 * so we have to force it to sync before checking spa_refcnt.
6259 if (spa->spa_sync_on) {
6260 txg_wait_synced(spa->spa_dsl_pool, 0);
6261 spa_evicting_os_wait(spa);
6265 * A pool cannot be exported or destroyed if there are active
6266 * references. If we are resetting a pool, allow references by
6267 * fault injection handlers.
6269 if (!spa_refcount_zero(spa) ||
6270 (spa->spa_inject_ref != 0 &&
6271 new_state != POOL_STATE_UNINITIALIZED)) {
6272 spa_async_resume(spa);
6273 spa->spa_is_exporting = B_FALSE;
6274 mutex_exit(&spa_namespace_lock);
6275 return (SET_ERROR(EBUSY));
6278 if (spa->spa_sync_on) {
6280 * A pool cannot be exported if it has an active shared spare.
6281 * This is to prevent other pools stealing the active spare
6282 * from an exported pool. At user's own will, such pool can
6283 * be forcedly exported.
6285 if (!force && new_state == POOL_STATE_EXPORTED &&
6286 spa_has_active_shared_spare(spa)) {
6287 spa_async_resume(spa);
6288 spa->spa_is_exporting = B_FALSE;
6289 mutex_exit(&spa_namespace_lock);
6290 return (SET_ERROR(EXDEV));
6294 * We're about to export or destroy this pool. Make sure
6295 * we stop all initialization and trim activity here before
6296 * we set the spa_final_txg. This will ensure that all
6297 * dirty data resulting from the initialization is
6298 * committed to disk before we unload the pool.
6300 if (spa->spa_root_vdev != NULL) {
6301 vdev_t *rvd = spa->spa_root_vdev;
6302 vdev_initialize_stop_all(rvd, VDEV_INITIALIZE_ACTIVE);
6303 vdev_trim_stop_all(rvd, VDEV_TRIM_ACTIVE);
6304 vdev_autotrim_stop_all(spa);
6305 vdev_rebuild_stop_all(spa);
6309 * We want this to be reflected on every label,
6310 * so mark them all dirty. spa_unload() will do the
6311 * final sync that pushes these changes out.
6313 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
6314 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6315 spa->spa_state = new_state;
6316 spa->spa_final_txg = spa_last_synced_txg(spa) +
6318 vdev_config_dirty(spa->spa_root_vdev);
6319 spa_config_exit(spa, SCL_ALL, FTAG);
6324 if (new_state == POOL_STATE_DESTROYED)
6325 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_DESTROY);
6326 else if (new_state == POOL_STATE_EXPORTED)
6327 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_EXPORT);
6329 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
6331 spa_deactivate(spa);
6334 if (oldconfig && spa->spa_config)
6335 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
6337 if (new_state != POOL_STATE_UNINITIALIZED) {
6339 spa_write_cachefile(spa, B_TRUE, B_TRUE);
6343 * If spa_remove() is not called for this spa_t and
6344 * there is any possibility that it can be reused,
6345 * we make sure to reset the exporting flag.
6347 spa->spa_is_exporting = B_FALSE;
6350 mutex_exit(&spa_namespace_lock);
6355 * Destroy a storage pool.
6358 spa_destroy(char *pool)
6360 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
6365 * Export a storage pool.
6368 spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
6369 boolean_t hardforce)
6371 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
6376 * Similar to spa_export(), this unloads the spa_t without actually removing it
6377 * from the namespace in any way.
6380 spa_reset(char *pool)
6382 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
6387 * ==========================================================================
6388 * Device manipulation
6389 * ==========================================================================
6393 * Add a device to a storage pool.
6396 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
6400 vdev_t *rvd = spa->spa_root_vdev;
6402 nvlist_t **spares, **l2cache;
6403 uint_t nspares, nl2cache;
6405 ASSERT(spa_writeable(spa));
6407 txg = spa_vdev_enter(spa);
6409 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
6410 VDEV_ALLOC_ADD)) != 0)
6411 return (spa_vdev_exit(spa, NULL, txg, error));
6413 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */
6415 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
6419 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
6423 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
6424 return (spa_vdev_exit(spa, vd, txg, EINVAL));
6426 if (vd->vdev_children != 0 &&
6427 (error = vdev_create(vd, txg, B_FALSE)) != 0)
6428 return (spa_vdev_exit(spa, vd, txg, error));
6431 * We must validate the spares and l2cache devices after checking the
6432 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
6434 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
6435 return (spa_vdev_exit(spa, vd, txg, error));
6438 * If we are in the middle of a device removal, we can only add
6439 * devices which match the existing devices in the pool.
6440 * If we are in the middle of a removal, or have some indirect
6441 * vdevs, we can not add raidz toplevels.
6443 if (spa->spa_vdev_removal != NULL ||
6444 spa->spa_removing_phys.sr_prev_indirect_vdev != -1) {
6445 for (int c = 0; c < vd->vdev_children; c++) {
6446 tvd = vd->vdev_child[c];
6447 if (spa->spa_vdev_removal != NULL &&
6448 tvd->vdev_ashift != spa->spa_max_ashift) {
6449 return (spa_vdev_exit(spa, vd, txg, EINVAL));
6451 /* Fail if top level vdev is raidz */
6452 if (tvd->vdev_ops == &vdev_raidz_ops) {
6453 return (spa_vdev_exit(spa, vd, txg, EINVAL));
6456 * Need the top level mirror to be
6457 * a mirror of leaf vdevs only
6459 if (tvd->vdev_ops == &vdev_mirror_ops) {
6460 for (uint64_t cid = 0;
6461 cid < tvd->vdev_children; cid++) {
6462 vdev_t *cvd = tvd->vdev_child[cid];
6463 if (!cvd->vdev_ops->vdev_op_leaf) {
6464 return (spa_vdev_exit(spa, vd,
6472 for (int c = 0; c < vd->vdev_children; c++) {
6473 tvd = vd->vdev_child[c];
6474 vdev_remove_child(vd, tvd);
6475 tvd->vdev_id = rvd->vdev_children;
6476 vdev_add_child(rvd, tvd);
6477 vdev_config_dirty(tvd);
6481 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
6482 ZPOOL_CONFIG_SPARES);
6483 spa_load_spares(spa);
6484 spa->spa_spares.sav_sync = B_TRUE;
6487 if (nl2cache != 0) {
6488 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
6489 ZPOOL_CONFIG_L2CACHE);
6490 spa_load_l2cache(spa);
6491 spa->spa_l2cache.sav_sync = B_TRUE;
6495 * We have to be careful when adding new vdevs to an existing pool.
6496 * If other threads start allocating from these vdevs before we
6497 * sync the config cache, and we lose power, then upon reboot we may
6498 * fail to open the pool because there are DVAs that the config cache
6499 * can't translate. Therefore, we first add the vdevs without
6500 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
6501 * and then let spa_config_update() initialize the new metaslabs.
6503 * spa_load() checks for added-but-not-initialized vdevs, so that
6504 * if we lose power at any point in this sequence, the remaining
6505 * steps will be completed the next time we load the pool.
6507 (void) spa_vdev_exit(spa, vd, txg, 0);
6509 mutex_enter(&spa_namespace_lock);
6510 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
6511 spa_event_notify(spa, NULL, NULL, ESC_ZFS_VDEV_ADD);
6512 mutex_exit(&spa_namespace_lock);
6518 * Attach a device to a mirror. The arguments are the path to any device
6519 * in the mirror, and the nvroot for the new device. If the path specifies
6520 * a device that is not mirrored, we automatically insert the mirror vdev.
6522 * If 'replacing' is specified, the new device is intended to replace the
6523 * existing device; in this case the two devices are made into their own
6524 * mirror using the 'replacing' vdev, which is functionally identical to
6525 * the mirror vdev (it actually reuses all the same ops) but has a few
6526 * extra rules: you can't attach to it after it's been created, and upon
6527 * completion of resilvering, the first disk (the one being replaced)
6528 * is automatically detached.
6530 * If 'rebuild' is specified, then sequential reconstruction (a.ka. rebuild)
6531 * should be performed instead of traditional healing reconstruction. From
6532 * an administrators perspective these are both resilver operations.
6535 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing,
6538 uint64_t txg, dtl_max_txg;
6539 vdev_t *rvd = spa->spa_root_vdev;
6540 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
6542 char *oldvdpath, *newvdpath;
6546 ASSERT(spa_writeable(spa));
6548 txg = spa_vdev_enter(spa);
6550 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
6552 ASSERT(MUTEX_HELD(&spa_namespace_lock));
6553 if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) {
6554 error = (spa_has_checkpoint(spa)) ?
6555 ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT;
6556 return (spa_vdev_exit(spa, NULL, txg, error));
6560 if (!spa_feature_is_enabled(spa, SPA_FEATURE_DEVICE_REBUILD))
6561 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
6563 if (dsl_scan_resilvering(spa_get_dsl(spa)))
6564 return (spa_vdev_exit(spa, NULL, txg,
6565 ZFS_ERR_RESILVER_IN_PROGRESS));
6567 if (vdev_rebuild_active(rvd))
6568 return (spa_vdev_exit(spa, NULL, txg,
6569 ZFS_ERR_REBUILD_IN_PROGRESS));
6572 if (spa->spa_vdev_removal != NULL)
6573 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
6576 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
6578 if (!oldvd->vdev_ops->vdev_op_leaf)
6579 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
6581 pvd = oldvd->vdev_parent;
6583 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
6584 VDEV_ALLOC_ATTACH)) != 0)
6585 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
6587 if (newrootvd->vdev_children != 1)
6588 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
6590 newvd = newrootvd->vdev_child[0];
6592 if (!newvd->vdev_ops->vdev_op_leaf)
6593 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
6595 if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
6596 return (spa_vdev_exit(spa, newrootvd, txg, error));
6599 * Spares can't replace logs
6601 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
6602 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
6606 * For rebuilds, the parent vdev must support reconstruction
6607 * using only space maps. This means the only allowable
6608 * parents are the root vdev or a mirror vdev.
6610 if (pvd->vdev_ops != &vdev_mirror_ops &&
6611 pvd->vdev_ops != &vdev_root_ops) {
6612 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
6618 * For attach, the only allowable parent is a mirror or the root
6621 if (pvd->vdev_ops != &vdev_mirror_ops &&
6622 pvd->vdev_ops != &vdev_root_ops)
6623 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
6625 pvops = &vdev_mirror_ops;
6628 * Active hot spares can only be replaced by inactive hot
6631 if (pvd->vdev_ops == &vdev_spare_ops &&
6632 oldvd->vdev_isspare &&
6633 !spa_has_spare(spa, newvd->vdev_guid))
6634 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
6637 * If the source is a hot spare, and the parent isn't already a
6638 * spare, then we want to create a new hot spare. Otherwise, we
6639 * want to create a replacing vdev. The user is not allowed to
6640 * attach to a spared vdev child unless the 'isspare' state is
6641 * the same (spare replaces spare, non-spare replaces
6644 if (pvd->vdev_ops == &vdev_replacing_ops &&
6645 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
6646 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
6647 } else if (pvd->vdev_ops == &vdev_spare_ops &&
6648 newvd->vdev_isspare != oldvd->vdev_isspare) {
6649 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
6652 if (newvd->vdev_isspare)
6653 pvops = &vdev_spare_ops;
6655 pvops = &vdev_replacing_ops;
6659 * Make sure the new device is big enough.
6661 if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
6662 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
6665 * The new device cannot have a higher alignment requirement
6666 * than the top-level vdev.
6668 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
6669 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
6672 * If this is an in-place replacement, update oldvd's path and devid
6673 * to make it distinguishable from newvd, and unopenable from now on.
6675 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
6676 spa_strfree(oldvd->vdev_path);
6677 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
6679 (void) snprintf(oldvd->vdev_path, strlen(newvd->vdev_path) + 5,
6680 "%s/%s", newvd->vdev_path, "old");
6681 if (oldvd->vdev_devid != NULL) {
6682 spa_strfree(oldvd->vdev_devid);
6683 oldvd->vdev_devid = NULL;
6688 * If the parent is not a mirror, or if we're replacing, insert the new
6689 * mirror/replacing/spare vdev above oldvd.
6691 if (pvd->vdev_ops != pvops)
6692 pvd = vdev_add_parent(oldvd, pvops);
6694 ASSERT(pvd->vdev_top->vdev_parent == rvd);
6695 ASSERT(pvd->vdev_ops == pvops);
6696 ASSERT(oldvd->vdev_parent == pvd);
6699 * Extract the new device from its root and add it to pvd.
6701 vdev_remove_child(newrootvd, newvd);
6702 newvd->vdev_id = pvd->vdev_children;
6703 newvd->vdev_crtxg = oldvd->vdev_crtxg;
6704 vdev_add_child(pvd, newvd);
6707 * Reevaluate the parent vdev state.
6709 vdev_propagate_state(pvd);
6711 tvd = newvd->vdev_top;
6712 ASSERT(pvd->vdev_top == tvd);
6713 ASSERT(tvd->vdev_parent == rvd);
6715 vdev_config_dirty(tvd);
6718 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
6719 * for any dmu_sync-ed blocks. It will propagate upward when
6720 * spa_vdev_exit() calls vdev_dtl_reassess().
6722 dtl_max_txg = txg + TXG_CONCURRENT_STATES;
6724 vdev_dtl_dirty(newvd, DTL_MISSING,
6725 TXG_INITIAL, dtl_max_txg - TXG_INITIAL);
6727 if (newvd->vdev_isspare) {
6728 spa_spare_activate(newvd);
6729 spa_event_notify(spa, newvd, NULL, ESC_ZFS_VDEV_SPARE);
6732 oldvdpath = spa_strdup(oldvd->vdev_path);
6733 newvdpath = spa_strdup(newvd->vdev_path);
6734 newvd_isspare = newvd->vdev_isspare;
6737 * Mark newvd's DTL dirty in this txg.
6739 vdev_dirty(tvd, VDD_DTL, newvd, txg);
6742 * Schedule the resilver or rebuild to restart in the future. We do
6743 * this to ensure that dmu_sync-ed blocks have been stitched into the
6744 * respective datasets.
6747 newvd->vdev_rebuild_txg = txg;
6751 newvd->vdev_resilver_txg = txg;
6753 if (dsl_scan_resilvering(spa_get_dsl(spa)) &&
6754 spa_feature_is_enabled(spa, SPA_FEATURE_RESILVER_DEFER)) {
6755 vdev_defer_resilver(newvd);
6757 dsl_scan_restart_resilver(spa->spa_dsl_pool,
6762 if (spa->spa_bootfs)
6763 spa_event_notify(spa, newvd, NULL, ESC_ZFS_BOOTFS_VDEV_ATTACH);
6765 spa_event_notify(spa, newvd, NULL, ESC_ZFS_VDEV_ATTACH);
6770 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
6772 spa_history_log_internal(spa, "vdev attach", NULL,
6773 "%s vdev=%s %s vdev=%s",
6774 replacing && newvd_isspare ? "spare in" :
6775 replacing ? "replace" : "attach", newvdpath,
6776 replacing ? "for" : "to", oldvdpath);
6778 spa_strfree(oldvdpath);
6779 spa_strfree(newvdpath);
6785 * Detach a device from a mirror or replacing vdev.
6787 * If 'replace_done' is specified, only detach if the parent
6788 * is a replacing vdev.
6791 spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
6795 vdev_t *rvd __maybe_unused = spa->spa_root_vdev;
6796 vdev_t *vd, *pvd, *cvd, *tvd;
6797 boolean_t unspare = B_FALSE;
6798 uint64_t unspare_guid = 0;
6801 ASSERT(spa_writeable(spa));
6803 txg = spa_vdev_detach_enter(spa, guid);
6805 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
6808 * Besides being called directly from the userland through the
6809 * ioctl interface, spa_vdev_detach() can be potentially called
6810 * at the end of spa_vdev_resilver_done().
6812 * In the regular case, when we have a checkpoint this shouldn't
6813 * happen as we never empty the DTLs of a vdev during the scrub
6814 * [see comment in dsl_scan_done()]. Thus spa_vdev_resilvering_done()
6815 * should never get here when we have a checkpoint.
6817 * That said, even in a case when we checkpoint the pool exactly
6818 * as spa_vdev_resilver_done() calls this function everything
6819 * should be fine as the resilver will return right away.
6821 ASSERT(MUTEX_HELD(&spa_namespace_lock));
6822 if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) {
6823 error = (spa_has_checkpoint(spa)) ?
6824 ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT;
6825 return (spa_vdev_exit(spa, NULL, txg, error));
6829 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
6831 if (!vd->vdev_ops->vdev_op_leaf)
6832 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
6834 pvd = vd->vdev_parent;
6837 * If the parent/child relationship is not as expected, don't do it.
6838 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
6839 * vdev that's replacing B with C. The user's intent in replacing
6840 * is to go from M(A,B) to M(A,C). If the user decides to cancel
6841 * the replace by detaching C, the expected behavior is to end up
6842 * M(A,B). But suppose that right after deciding to detach C,
6843 * the replacement of B completes. We would have M(A,C), and then
6844 * ask to detach C, which would leave us with just A -- not what
6845 * the user wanted. To prevent this, we make sure that the
6846 * parent/child relationship hasn't changed -- in this example,
6847 * that C's parent is still the replacing vdev R.
6849 if (pvd->vdev_guid != pguid && pguid != 0)
6850 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
6853 * Only 'replacing' or 'spare' vdevs can be replaced.
6855 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
6856 pvd->vdev_ops != &vdev_spare_ops)
6857 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
6859 ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
6860 spa_version(spa) >= SPA_VERSION_SPARES);
6863 * Only mirror, replacing, and spare vdevs support detach.
6865 if (pvd->vdev_ops != &vdev_replacing_ops &&
6866 pvd->vdev_ops != &vdev_mirror_ops &&
6867 pvd->vdev_ops != &vdev_spare_ops)
6868 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
6871 * If this device has the only valid copy of some data,
6872 * we cannot safely detach it.
6874 if (vdev_dtl_required(vd))
6875 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
6877 ASSERT(pvd->vdev_children >= 2);
6880 * If we are detaching the second disk from a replacing vdev, then
6881 * check to see if we changed the original vdev's path to have "/old"
6882 * at the end in spa_vdev_attach(). If so, undo that change now.
6884 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
6885 vd->vdev_path != NULL) {
6886 size_t len = strlen(vd->vdev_path);
6888 for (int c = 0; c < pvd->vdev_children; c++) {
6889 cvd = pvd->vdev_child[c];
6891 if (cvd == vd || cvd->vdev_path == NULL)
6894 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
6895 strcmp(cvd->vdev_path + len, "/old") == 0) {
6896 spa_strfree(cvd->vdev_path);
6897 cvd->vdev_path = spa_strdup(vd->vdev_path);
6904 * If we are detaching the original disk from a spare, then it implies
6905 * that the spare should become a real disk, and be removed from the
6906 * active spare list for the pool.
6908 if (pvd->vdev_ops == &vdev_spare_ops &&
6910 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
6914 * Erase the disk labels so the disk can be used for other things.
6915 * This must be done after all other error cases are handled,
6916 * but before we disembowel vd (so we can still do I/O to it).
6917 * But if we can't do it, don't treat the error as fatal --
6918 * it may be that the unwritability of the disk is the reason
6919 * it's being detached!
6921 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
6924 * Remove vd from its parent and compact the parent's children.
6926 vdev_remove_child(pvd, vd);
6927 vdev_compact_children(pvd);
6930 * Remember one of the remaining children so we can get tvd below.
6932 cvd = pvd->vdev_child[pvd->vdev_children - 1];
6935 * If we need to remove the remaining child from the list of hot spares,
6936 * do it now, marking the vdev as no longer a spare in the process.
6937 * We must do this before vdev_remove_parent(), because that can
6938 * change the GUID if it creates a new toplevel GUID. For a similar
6939 * reason, we must remove the spare now, in the same txg as the detach;
6940 * otherwise someone could attach a new sibling, change the GUID, and
6941 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
6944 ASSERT(cvd->vdev_isspare);
6945 spa_spare_remove(cvd);
6946 unspare_guid = cvd->vdev_guid;
6947 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
6948 cvd->vdev_unspare = B_TRUE;
6952 * If the parent mirror/replacing vdev only has one child,
6953 * the parent is no longer needed. Remove it from the tree.
6955 if (pvd->vdev_children == 1) {
6956 if (pvd->vdev_ops == &vdev_spare_ops)
6957 cvd->vdev_unspare = B_FALSE;
6958 vdev_remove_parent(cvd);
6962 * We don't set tvd until now because the parent we just removed
6963 * may have been the previous top-level vdev.
6965 tvd = cvd->vdev_top;
6966 ASSERT(tvd->vdev_parent == rvd);
6969 * Reevaluate the parent vdev state.
6971 vdev_propagate_state(cvd);
6974 * If the 'autoexpand' property is set on the pool then automatically
6975 * try to expand the size of the pool. For example if the device we
6976 * just detached was smaller than the others, it may be possible to
6977 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
6978 * first so that we can obtain the updated sizes of the leaf vdevs.
6980 if (spa->spa_autoexpand) {
6982 vdev_expand(tvd, txg);
6985 vdev_config_dirty(tvd);
6988 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
6989 * vd->vdev_detached is set and free vd's DTL object in syncing context.
6990 * But first make sure we're not on any *other* txg's DTL list, to
6991 * prevent vd from being accessed after it's freed.
6993 vdpath = spa_strdup(vd->vdev_path ? vd->vdev_path : "none");
6994 for (int t = 0; t < TXG_SIZE; t++)
6995 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
6996 vd->vdev_detached = B_TRUE;
6997 vdev_dirty(tvd, VDD_DTL, vd, txg);
6999 spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_REMOVE);
7000 spa_notify_waiters(spa);
7002 /* hang on to the spa before we release the lock */
7003 spa_open_ref(spa, FTAG);
7005 error = spa_vdev_exit(spa, vd, txg, 0);
7007 spa_history_log_internal(spa, "detach", NULL,
7009 spa_strfree(vdpath);
7012 * If this was the removal of the original device in a hot spare vdev,
7013 * then we want to go through and remove the device from the hot spare
7014 * list of every other pool.
7017 spa_t *altspa = NULL;
7019 mutex_enter(&spa_namespace_lock);
7020 while ((altspa = spa_next(altspa)) != NULL) {
7021 if (altspa->spa_state != POOL_STATE_ACTIVE ||
7025 spa_open_ref(altspa, FTAG);
7026 mutex_exit(&spa_namespace_lock);
7027 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
7028 mutex_enter(&spa_namespace_lock);
7029 spa_close(altspa, FTAG);
7031 mutex_exit(&spa_namespace_lock);
7033 /* search the rest of the vdevs for spares to remove */
7034 spa_vdev_resilver_done(spa);
7037 /* all done with the spa; OK to release */
7038 mutex_enter(&spa_namespace_lock);
7039 spa_close(spa, FTAG);
7040 mutex_exit(&spa_namespace_lock);
7046 spa_vdev_initialize_impl(spa_t *spa, uint64_t guid, uint64_t cmd_type,
7049 ASSERT(MUTEX_HELD(&spa_namespace_lock));
7051 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
7053 /* Look up vdev and ensure it's a leaf. */
7054 vdev_t *vd = spa_lookup_by_guid(spa, guid, B_FALSE);
7055 if (vd == NULL || vd->vdev_detached) {
7056 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
7057 return (SET_ERROR(ENODEV));
7058 } else if (!vd->vdev_ops->vdev_op_leaf || !vdev_is_concrete(vd)) {
7059 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
7060 return (SET_ERROR(EINVAL));
7061 } else if (!vdev_writeable(vd)) {
7062 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
7063 return (SET_ERROR(EROFS));
7065 mutex_enter(&vd->vdev_initialize_lock);
7066 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
7069 * When we activate an initialize action we check to see
7070 * if the vdev_initialize_thread is NULL. We do this instead
7071 * of using the vdev_initialize_state since there might be
7072 * a previous initialization process which has completed but
7073 * the thread is not exited.
7075 if (cmd_type == POOL_INITIALIZE_START &&
7076 (vd->vdev_initialize_thread != NULL ||
7077 vd->vdev_top->vdev_removing)) {
7078 mutex_exit(&vd->vdev_initialize_lock);
7079 return (SET_ERROR(EBUSY));
7080 } else if (cmd_type == POOL_INITIALIZE_CANCEL &&
7081 (vd->vdev_initialize_state != VDEV_INITIALIZE_ACTIVE &&
7082 vd->vdev_initialize_state != VDEV_INITIALIZE_SUSPENDED)) {
7083 mutex_exit(&vd->vdev_initialize_lock);
7084 return (SET_ERROR(ESRCH));
7085 } else if (cmd_type == POOL_INITIALIZE_SUSPEND &&
7086 vd->vdev_initialize_state != VDEV_INITIALIZE_ACTIVE) {
7087 mutex_exit(&vd->vdev_initialize_lock);
7088 return (SET_ERROR(ESRCH));
7092 case POOL_INITIALIZE_START:
7093 vdev_initialize(vd);
7095 case POOL_INITIALIZE_CANCEL:
7096 vdev_initialize_stop(vd, VDEV_INITIALIZE_CANCELED, vd_list);
7098 case POOL_INITIALIZE_SUSPEND:
7099 vdev_initialize_stop(vd, VDEV_INITIALIZE_SUSPENDED, vd_list);
7102 panic("invalid cmd_type %llu", (unsigned long long)cmd_type);
7104 mutex_exit(&vd->vdev_initialize_lock);
7110 spa_vdev_initialize(spa_t *spa, nvlist_t *nv, uint64_t cmd_type,
7111 nvlist_t *vdev_errlist)
7113 int total_errors = 0;
7116 list_create(&vd_list, sizeof (vdev_t),
7117 offsetof(vdev_t, vdev_initialize_node));
7120 * We hold the namespace lock through the whole function
7121 * to prevent any changes to the pool while we're starting or
7122 * stopping initialization. The config and state locks are held so that
7123 * we can properly assess the vdev state before we commit to
7124 * the initializing operation.
7126 mutex_enter(&spa_namespace_lock);
7128 for (nvpair_t *pair = nvlist_next_nvpair(nv, NULL);
7129 pair != NULL; pair = nvlist_next_nvpair(nv, pair)) {
7130 uint64_t vdev_guid = fnvpair_value_uint64(pair);
7132 int error = spa_vdev_initialize_impl(spa, vdev_guid, cmd_type,
7135 char guid_as_str[MAXNAMELEN];
7137 (void) snprintf(guid_as_str, sizeof (guid_as_str),
7138 "%llu", (unsigned long long)vdev_guid);
7139 fnvlist_add_int64(vdev_errlist, guid_as_str, error);
7144 /* Wait for all initialize threads to stop. */
7145 vdev_initialize_stop_wait(spa, &vd_list);
7147 /* Sync out the initializing state */
7148 txg_wait_synced(spa->spa_dsl_pool, 0);
7149 mutex_exit(&spa_namespace_lock);
7151 list_destroy(&vd_list);
7153 return (total_errors);
7157 spa_vdev_trim_impl(spa_t *spa, uint64_t guid, uint64_t cmd_type,
7158 uint64_t rate, boolean_t partial, boolean_t secure, list_t *vd_list)
7160 ASSERT(MUTEX_HELD(&spa_namespace_lock));
7162 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
7164 /* Look up vdev and ensure it's a leaf. */
7165 vdev_t *vd = spa_lookup_by_guid(spa, guid, B_FALSE);
7166 if (vd == NULL || vd->vdev_detached) {
7167 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
7168 return (SET_ERROR(ENODEV));
7169 } else if (!vd->vdev_ops->vdev_op_leaf || !vdev_is_concrete(vd)) {
7170 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
7171 return (SET_ERROR(EINVAL));
7172 } else if (!vdev_writeable(vd)) {
7173 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
7174 return (SET_ERROR(EROFS));
7175 } else if (!vd->vdev_has_trim) {
7176 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
7177 return (SET_ERROR(EOPNOTSUPP));
7178 } else if (secure && !vd->vdev_has_securetrim) {
7179 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
7180 return (SET_ERROR(EOPNOTSUPP));
7182 mutex_enter(&vd->vdev_trim_lock);
7183 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
7186 * When we activate a TRIM action we check to see if the
7187 * vdev_trim_thread is NULL. We do this instead of using the
7188 * vdev_trim_state since there might be a previous TRIM process
7189 * which has completed but the thread is not exited.
7191 if (cmd_type == POOL_TRIM_START &&
7192 (vd->vdev_trim_thread != NULL || vd->vdev_top->vdev_removing)) {
7193 mutex_exit(&vd->vdev_trim_lock);
7194 return (SET_ERROR(EBUSY));
7195 } else if (cmd_type == POOL_TRIM_CANCEL &&
7196 (vd->vdev_trim_state != VDEV_TRIM_ACTIVE &&
7197 vd->vdev_trim_state != VDEV_TRIM_SUSPENDED)) {
7198 mutex_exit(&vd->vdev_trim_lock);
7199 return (SET_ERROR(ESRCH));
7200 } else if (cmd_type == POOL_TRIM_SUSPEND &&
7201 vd->vdev_trim_state != VDEV_TRIM_ACTIVE) {
7202 mutex_exit(&vd->vdev_trim_lock);
7203 return (SET_ERROR(ESRCH));
7207 case POOL_TRIM_START:
7208 vdev_trim(vd, rate, partial, secure);
7210 case POOL_TRIM_CANCEL:
7211 vdev_trim_stop(vd, VDEV_TRIM_CANCELED, vd_list);
7213 case POOL_TRIM_SUSPEND:
7214 vdev_trim_stop(vd, VDEV_TRIM_SUSPENDED, vd_list);
7217 panic("invalid cmd_type %llu", (unsigned long long)cmd_type);
7219 mutex_exit(&vd->vdev_trim_lock);
7225 * Initiates a manual TRIM for the requested vdevs. This kicks off individual
7226 * TRIM threads for each child vdev. These threads pass over all of the free
7227 * space in the vdev's metaslabs and issues TRIM commands for that space.
7230 spa_vdev_trim(spa_t *spa, nvlist_t *nv, uint64_t cmd_type, uint64_t rate,
7231 boolean_t partial, boolean_t secure, nvlist_t *vdev_errlist)
7233 int total_errors = 0;
7236 list_create(&vd_list, sizeof (vdev_t),
7237 offsetof(vdev_t, vdev_trim_node));
7240 * We hold the namespace lock through the whole function
7241 * to prevent any changes to the pool while we're starting or
7242 * stopping TRIM. The config and state locks are held so that
7243 * we can properly assess the vdev state before we commit to
7244 * the TRIM operation.
7246 mutex_enter(&spa_namespace_lock);
7248 for (nvpair_t *pair = nvlist_next_nvpair(nv, NULL);
7249 pair != NULL; pair = nvlist_next_nvpair(nv, pair)) {
7250 uint64_t vdev_guid = fnvpair_value_uint64(pair);
7252 int error = spa_vdev_trim_impl(spa, vdev_guid, cmd_type,
7253 rate, partial, secure, &vd_list);
7255 char guid_as_str[MAXNAMELEN];
7257 (void) snprintf(guid_as_str, sizeof (guid_as_str),
7258 "%llu", (unsigned long long)vdev_guid);
7259 fnvlist_add_int64(vdev_errlist, guid_as_str, error);
7264 /* Wait for all TRIM threads to stop. */
7265 vdev_trim_stop_wait(spa, &vd_list);
7267 /* Sync out the TRIM state */
7268 txg_wait_synced(spa->spa_dsl_pool, 0);
7269 mutex_exit(&spa_namespace_lock);
7271 list_destroy(&vd_list);
7273 return (total_errors);
7277 * Split a set of devices from their mirrors, and create a new pool from them.
7280 spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
7281 nvlist_t *props, boolean_t exp)
7284 uint64_t txg, *glist;
7286 uint_t c, children, lastlog;
7287 nvlist_t **child, *nvl, *tmp;
7289 char *altroot = NULL;
7290 vdev_t *rvd, **vml = NULL; /* vdev modify list */
7291 boolean_t activate_slog;
7293 ASSERT(spa_writeable(spa));
7295 txg = spa_vdev_enter(spa);
7297 ASSERT(MUTEX_HELD(&spa_namespace_lock));
7298 if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) {
7299 error = (spa_has_checkpoint(spa)) ?
7300 ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT;
7301 return (spa_vdev_exit(spa, NULL, txg, error));
7304 /* clear the log and flush everything up to now */
7305 activate_slog = spa_passivate_log(spa);
7306 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
7307 error = spa_reset_logs(spa);
7308 txg = spa_vdev_config_enter(spa);
7311 spa_activate_log(spa);
7314 return (spa_vdev_exit(spa, NULL, txg, error));
7316 /* check new spa name before going any further */
7317 if (spa_lookup(newname) != NULL)
7318 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
7321 * scan through all the children to ensure they're all mirrors
7323 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
7324 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
7326 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
7328 /* first, check to ensure we've got the right child count */
7329 rvd = spa->spa_root_vdev;
7331 for (c = 0; c < rvd->vdev_children; c++) {
7332 vdev_t *vd = rvd->vdev_child[c];
7334 /* don't count the holes & logs as children */
7335 if (vd->vdev_islog || (vd->vdev_ops != &vdev_indirect_ops &&
7336 !vdev_is_concrete(vd))) {
7344 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
7345 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
7347 /* next, ensure no spare or cache devices are part of the split */
7348 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
7349 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
7350 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
7352 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
7353 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
7355 /* then, loop over each vdev and validate it */
7356 for (c = 0; c < children; c++) {
7357 uint64_t is_hole = 0;
7359 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
7363 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
7364 spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
7367 error = SET_ERROR(EINVAL);
7372 /* deal with indirect vdevs */
7373 if (spa->spa_root_vdev->vdev_child[c]->vdev_ops ==
7377 /* which disk is going to be split? */
7378 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
7380 error = SET_ERROR(EINVAL);
7384 /* look it up in the spa */
7385 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
7386 if (vml[c] == NULL) {
7387 error = SET_ERROR(ENODEV);
7391 /* make sure there's nothing stopping the split */
7392 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
7393 vml[c]->vdev_islog ||
7394 !vdev_is_concrete(vml[c]) ||
7395 vml[c]->vdev_isspare ||
7396 vml[c]->vdev_isl2cache ||
7397 !vdev_writeable(vml[c]) ||
7398 vml[c]->vdev_children != 0 ||
7399 vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
7400 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
7401 error = SET_ERROR(EINVAL);
7405 if (vdev_dtl_required(vml[c]) ||
7406 vdev_resilver_needed(vml[c], NULL, NULL)) {
7407 error = SET_ERROR(EBUSY);
7411 /* we need certain info from the top level */
7412 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
7413 vml[c]->vdev_top->vdev_ms_array) == 0);
7414 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
7415 vml[c]->vdev_top->vdev_ms_shift) == 0);
7416 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
7417 vml[c]->vdev_top->vdev_asize) == 0);
7418 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
7419 vml[c]->vdev_top->vdev_ashift) == 0);
7421 /* transfer per-vdev ZAPs */
7422 ASSERT3U(vml[c]->vdev_leaf_zap, !=, 0);
7423 VERIFY0(nvlist_add_uint64(child[c],
7424 ZPOOL_CONFIG_VDEV_LEAF_ZAP, vml[c]->vdev_leaf_zap));
7426 ASSERT3U(vml[c]->vdev_top->vdev_top_zap, !=, 0);
7427 VERIFY0(nvlist_add_uint64(child[c],
7428 ZPOOL_CONFIG_VDEV_TOP_ZAP,
7429 vml[c]->vdev_parent->vdev_top_zap));
7433 kmem_free(vml, children * sizeof (vdev_t *));
7434 kmem_free(glist, children * sizeof (uint64_t));
7435 return (spa_vdev_exit(spa, NULL, txg, error));
7438 /* stop writers from using the disks */
7439 for (c = 0; c < children; c++) {
7441 vml[c]->vdev_offline = B_TRUE;
7443 vdev_reopen(spa->spa_root_vdev);
7446 * Temporarily record the splitting vdevs in the spa config. This
7447 * will disappear once the config is regenerated.
7449 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
7450 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
7451 glist, children) == 0);
7452 kmem_free(glist, children * sizeof (uint64_t));
7454 mutex_enter(&spa->spa_props_lock);
7455 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
7457 mutex_exit(&spa->spa_props_lock);
7458 spa->spa_config_splitting = nvl;
7459 vdev_config_dirty(spa->spa_root_vdev);
7461 /* configure and create the new pool */
7462 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
7463 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
7464 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
7465 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
7466 spa_version(spa)) == 0);
7467 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
7468 spa->spa_config_txg) == 0);
7469 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
7470 spa_generate_guid(NULL)) == 0);
7471 VERIFY0(nvlist_add_boolean(config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS));
7472 (void) nvlist_lookup_string(props,
7473 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
7475 /* add the new pool to the namespace */
7476 newspa = spa_add(newname, config, altroot);
7477 newspa->spa_avz_action = AVZ_ACTION_REBUILD;
7478 newspa->spa_config_txg = spa->spa_config_txg;
7479 spa_set_log_state(newspa, SPA_LOG_CLEAR);
7481 /* release the spa config lock, retaining the namespace lock */
7482 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
7484 if (zio_injection_enabled)
7485 zio_handle_panic_injection(spa, FTAG, 1);
7487 spa_activate(newspa, spa_mode_global);
7488 spa_async_suspend(newspa);
7491 * Temporarily stop the initializing and TRIM activity. We set the
7492 * state to ACTIVE so that we know to resume initializing or TRIM
7493 * once the split has completed.
7495 list_t vd_initialize_list;
7496 list_create(&vd_initialize_list, sizeof (vdev_t),
7497 offsetof(vdev_t, vdev_initialize_node));
7499 list_t vd_trim_list;
7500 list_create(&vd_trim_list, sizeof (vdev_t),
7501 offsetof(vdev_t, vdev_trim_node));
7503 for (c = 0; c < children; c++) {
7504 if (vml[c] != NULL && vml[c]->vdev_ops != &vdev_indirect_ops) {
7505 mutex_enter(&vml[c]->vdev_initialize_lock);
7506 vdev_initialize_stop(vml[c],
7507 VDEV_INITIALIZE_ACTIVE, &vd_initialize_list);
7508 mutex_exit(&vml[c]->vdev_initialize_lock);
7510 mutex_enter(&vml[c]->vdev_trim_lock);
7511 vdev_trim_stop(vml[c], VDEV_TRIM_ACTIVE, &vd_trim_list);
7512 mutex_exit(&vml[c]->vdev_trim_lock);
7516 vdev_initialize_stop_wait(spa, &vd_initialize_list);
7517 vdev_trim_stop_wait(spa, &vd_trim_list);
7519 list_destroy(&vd_initialize_list);
7520 list_destroy(&vd_trim_list);
7522 newspa->spa_config_source = SPA_CONFIG_SRC_SPLIT;
7523 newspa->spa_is_splitting = B_TRUE;
7525 /* create the new pool from the disks of the original pool */
7526 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE);
7530 /* if that worked, generate a real config for the new pool */
7531 if (newspa->spa_root_vdev != NULL) {
7532 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
7533 NV_UNIQUE_NAME, KM_SLEEP) == 0);
7534 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
7535 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
7536 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
7541 if (props != NULL) {
7542 spa_configfile_set(newspa, props, B_FALSE);
7543 error = spa_prop_set(newspa, props);
7548 /* flush everything */
7549 txg = spa_vdev_config_enter(newspa);
7550 vdev_config_dirty(newspa->spa_root_vdev);
7551 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
7553 if (zio_injection_enabled)
7554 zio_handle_panic_injection(spa, FTAG, 2);
7556 spa_async_resume(newspa);
7558 /* finally, update the original pool's config */
7559 txg = spa_vdev_config_enter(spa);
7560 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
7561 error = dmu_tx_assign(tx, TXG_WAIT);
7564 for (c = 0; c < children; c++) {
7565 if (vml[c] != NULL && vml[c]->vdev_ops != &vdev_indirect_ops) {
7566 vdev_t *tvd = vml[c]->vdev_top;
7569 * Need to be sure the detachable VDEV is not
7570 * on any *other* txg's DTL list to prevent it
7571 * from being accessed after it's freed.
7573 for (int t = 0; t < TXG_SIZE; t++) {
7574 (void) txg_list_remove_this(
7575 &tvd->vdev_dtl_list, vml[c], t);
7580 spa_history_log_internal(spa, "detach", tx,
7581 "vdev=%s", vml[c]->vdev_path);
7586 spa->spa_avz_action = AVZ_ACTION_REBUILD;
7587 vdev_config_dirty(spa->spa_root_vdev);
7588 spa->spa_config_splitting = NULL;
7592 (void) spa_vdev_exit(spa, NULL, txg, 0);
7594 if (zio_injection_enabled)
7595 zio_handle_panic_injection(spa, FTAG, 3);
7597 /* split is complete; log a history record */
7598 spa_history_log_internal(newspa, "split", NULL,
7599 "from pool %s", spa_name(spa));
7601 newspa->spa_is_splitting = B_FALSE;
7602 kmem_free(vml, children * sizeof (vdev_t *));
7604 /* if we're not going to mount the filesystems in userland, export */
7606 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
7613 spa_deactivate(newspa);
7616 txg = spa_vdev_config_enter(spa);
7618 /* re-online all offlined disks */
7619 for (c = 0; c < children; c++) {
7621 vml[c]->vdev_offline = B_FALSE;
7624 /* restart initializing or trimming disks as necessary */
7625 spa_async_request(spa, SPA_ASYNC_INITIALIZE_RESTART);
7626 spa_async_request(spa, SPA_ASYNC_TRIM_RESTART);
7627 spa_async_request(spa, SPA_ASYNC_AUTOTRIM_RESTART);
7629 vdev_reopen(spa->spa_root_vdev);
7631 nvlist_free(spa->spa_config_splitting);
7632 spa->spa_config_splitting = NULL;
7633 (void) spa_vdev_exit(spa, NULL, txg, error);
7635 kmem_free(vml, children * sizeof (vdev_t *));
7640 * Find any device that's done replacing, or a vdev marked 'unspare' that's
7641 * currently spared, so we can detach it.
7644 spa_vdev_resilver_done_hunt(vdev_t *vd)
7646 vdev_t *newvd, *oldvd;
7648 for (int c = 0; c < vd->vdev_children; c++) {
7649 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
7655 * Check for a completed replacement. We always consider the first
7656 * vdev in the list to be the oldest vdev, and the last one to be
7657 * the newest (see spa_vdev_attach() for how that works). In
7658 * the case where the newest vdev is faulted, we will not automatically
7659 * remove it after a resilver completes. This is OK as it will require
7660 * user intervention to determine which disk the admin wishes to keep.
7662 if (vd->vdev_ops == &vdev_replacing_ops) {
7663 ASSERT(vd->vdev_children > 1);
7665 newvd = vd->vdev_child[vd->vdev_children - 1];
7666 oldvd = vd->vdev_child[0];
7668 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
7669 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
7670 !vdev_dtl_required(oldvd))
7675 * Check for a completed resilver with the 'unspare' flag set.
7676 * Also potentially update faulted state.
7678 if (vd->vdev_ops == &vdev_spare_ops) {
7679 vdev_t *first = vd->vdev_child[0];
7680 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
7682 if (last->vdev_unspare) {
7685 } else if (first->vdev_unspare) {
7692 if (oldvd != NULL &&
7693 vdev_dtl_empty(newvd, DTL_MISSING) &&
7694 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
7695 !vdev_dtl_required(oldvd))
7698 vdev_propagate_state(vd);
7701 * If there are more than two spares attached to a disk,
7702 * and those spares are not required, then we want to
7703 * attempt to free them up now so that they can be used
7704 * by other pools. Once we're back down to a single
7705 * disk+spare, we stop removing them.
7707 if (vd->vdev_children > 2) {
7708 newvd = vd->vdev_child[1];
7710 if (newvd->vdev_isspare && last->vdev_isspare &&
7711 vdev_dtl_empty(last, DTL_MISSING) &&
7712 vdev_dtl_empty(last, DTL_OUTAGE) &&
7713 !vdev_dtl_required(newvd))
7722 spa_vdev_resilver_done(spa_t *spa)
7724 vdev_t *vd, *pvd, *ppvd;
7725 uint64_t guid, sguid, pguid, ppguid;
7727 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
7729 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
7730 pvd = vd->vdev_parent;
7731 ppvd = pvd->vdev_parent;
7732 guid = vd->vdev_guid;
7733 pguid = pvd->vdev_guid;
7734 ppguid = ppvd->vdev_guid;
7737 * If we have just finished replacing a hot spared device, then
7738 * we need to detach the parent's first child (the original hot
7741 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
7742 ppvd->vdev_children == 2) {
7743 ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
7744 sguid = ppvd->vdev_child[1]->vdev_guid;
7746 ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd));
7748 spa_config_exit(spa, SCL_ALL, FTAG);
7749 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
7751 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
7753 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
7756 spa_config_exit(spa, SCL_ALL, FTAG);
7759 * If a detach was not performed above replace waiters will not have
7760 * been notified. In which case we must do so now.
7762 spa_notify_waiters(spa);
7766 * Update the stored path or FRU for this vdev.
7769 spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
7773 boolean_t sync = B_FALSE;
7775 ASSERT(spa_writeable(spa));
7777 spa_vdev_state_enter(spa, SCL_ALL);
7779 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
7780 return (spa_vdev_state_exit(spa, NULL, ENOENT));
7782 if (!vd->vdev_ops->vdev_op_leaf)
7783 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
7786 if (strcmp(value, vd->vdev_path) != 0) {
7787 spa_strfree(vd->vdev_path);
7788 vd->vdev_path = spa_strdup(value);
7792 if (vd->vdev_fru == NULL) {
7793 vd->vdev_fru = spa_strdup(value);
7795 } else if (strcmp(value, vd->vdev_fru) != 0) {
7796 spa_strfree(vd->vdev_fru);
7797 vd->vdev_fru = spa_strdup(value);
7802 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
7806 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
7808 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
7812 spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
7814 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
7818 * ==========================================================================
7820 * ==========================================================================
7823 spa_scrub_pause_resume(spa_t *spa, pool_scrub_cmd_t cmd)
7825 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
7827 if (dsl_scan_resilvering(spa->spa_dsl_pool))
7828 return (SET_ERROR(EBUSY));
7830 return (dsl_scrub_set_pause_resume(spa->spa_dsl_pool, cmd));
7834 spa_scan_stop(spa_t *spa)
7836 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
7837 if (dsl_scan_resilvering(spa->spa_dsl_pool))
7838 return (SET_ERROR(EBUSY));
7839 return (dsl_scan_cancel(spa->spa_dsl_pool));
7843 spa_scan(spa_t *spa, pool_scan_func_t func)
7845 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
7847 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
7848 return (SET_ERROR(ENOTSUP));
7850 if (func == POOL_SCAN_RESILVER &&
7851 !spa_feature_is_enabled(spa, SPA_FEATURE_RESILVER_DEFER))
7852 return (SET_ERROR(ENOTSUP));
7855 * If a resilver was requested, but there is no DTL on a
7856 * writeable leaf device, we have nothing to do.
7858 if (func == POOL_SCAN_RESILVER &&
7859 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
7860 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
7864 return (dsl_scan(spa->spa_dsl_pool, func));
7868 * ==========================================================================
7869 * SPA async task processing
7870 * ==========================================================================
7874 spa_async_remove(spa_t *spa, vdev_t *vd)
7876 if (vd->vdev_remove_wanted) {
7877 vd->vdev_remove_wanted = B_FALSE;
7878 vd->vdev_delayed_close = B_FALSE;
7879 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
7882 * We want to clear the stats, but we don't want to do a full
7883 * vdev_clear() as that will cause us to throw away
7884 * degraded/faulted state as well as attempt to reopen the
7885 * device, all of which is a waste.
7887 vd->vdev_stat.vs_read_errors = 0;
7888 vd->vdev_stat.vs_write_errors = 0;
7889 vd->vdev_stat.vs_checksum_errors = 0;
7891 vdev_state_dirty(vd->vdev_top);
7894 for (int c = 0; c < vd->vdev_children; c++)
7895 spa_async_remove(spa, vd->vdev_child[c]);
7899 spa_async_probe(spa_t *spa, vdev_t *vd)
7901 if (vd->vdev_probe_wanted) {
7902 vd->vdev_probe_wanted = B_FALSE;
7903 vdev_reopen(vd); /* vdev_open() does the actual probe */
7906 for (int c = 0; c < vd->vdev_children; c++)
7907 spa_async_probe(spa, vd->vdev_child[c]);
7911 spa_async_autoexpand(spa_t *spa, vdev_t *vd)
7913 if (!spa->spa_autoexpand)
7916 for (int c = 0; c < vd->vdev_children; c++) {
7917 vdev_t *cvd = vd->vdev_child[c];
7918 spa_async_autoexpand(spa, cvd);
7921 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
7924 spa_event_notify(vd->vdev_spa, vd, NULL, ESC_ZFS_VDEV_AUTOEXPAND);
7928 spa_async_thread(void *arg)
7930 spa_t *spa = (spa_t *)arg;
7931 dsl_pool_t *dp = spa->spa_dsl_pool;
7934 ASSERT(spa->spa_sync_on);
7936 mutex_enter(&spa->spa_async_lock);
7937 tasks = spa->spa_async_tasks;
7938 spa->spa_async_tasks = 0;
7939 mutex_exit(&spa->spa_async_lock);
7942 * See if the config needs to be updated.
7944 if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
7945 uint64_t old_space, new_space;
7947 mutex_enter(&spa_namespace_lock);
7948 old_space = metaslab_class_get_space(spa_normal_class(spa));
7949 old_space += metaslab_class_get_space(spa_special_class(spa));
7950 old_space += metaslab_class_get_space(spa_dedup_class(spa));
7952 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
7954 new_space = metaslab_class_get_space(spa_normal_class(spa));
7955 new_space += metaslab_class_get_space(spa_special_class(spa));
7956 new_space += metaslab_class_get_space(spa_dedup_class(spa));
7957 mutex_exit(&spa_namespace_lock);
7960 * If the pool grew as a result of the config update,
7961 * then log an internal history event.
7963 if (new_space != old_space) {
7964 spa_history_log_internal(spa, "vdev online", NULL,
7965 "pool '%s' size: %llu(+%llu)",
7966 spa_name(spa), (u_longlong_t)new_space,
7967 (u_longlong_t)(new_space - old_space));
7972 * See if any devices need to be marked REMOVED.
7974 if (tasks & SPA_ASYNC_REMOVE) {
7975 spa_vdev_state_enter(spa, SCL_NONE);
7976 spa_async_remove(spa, spa->spa_root_vdev);
7977 for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
7978 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
7979 for (int i = 0; i < spa->spa_spares.sav_count; i++)
7980 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
7981 (void) spa_vdev_state_exit(spa, NULL, 0);
7984 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
7985 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
7986 spa_async_autoexpand(spa, spa->spa_root_vdev);
7987 spa_config_exit(spa, SCL_CONFIG, FTAG);
7991 * See if any devices need to be probed.
7993 if (tasks & SPA_ASYNC_PROBE) {
7994 spa_vdev_state_enter(spa, SCL_NONE);
7995 spa_async_probe(spa, spa->spa_root_vdev);
7996 (void) spa_vdev_state_exit(spa, NULL, 0);
8000 * If any devices are done replacing, detach them.
8002 if (tasks & SPA_ASYNC_RESILVER_DONE)
8003 spa_vdev_resilver_done(spa);
8006 * If any devices are done replacing, detach them. Then if no
8007 * top-level vdevs are rebuilding attempt to kick off a scrub.
8009 if (tasks & SPA_ASYNC_REBUILD_DONE) {
8010 spa_vdev_resilver_done(spa);
8012 if (!vdev_rebuild_active(spa->spa_root_vdev))
8013 (void) dsl_scan(spa->spa_dsl_pool, POOL_SCAN_SCRUB);
8017 * Kick off a resilver.
8019 if (tasks & SPA_ASYNC_RESILVER &&
8020 !vdev_rebuild_active(spa->spa_root_vdev) &&
8021 (!dsl_scan_resilvering(dp) ||
8022 !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_RESILVER_DEFER)))
8023 dsl_scan_restart_resilver(dp, 0);
8025 if (tasks & SPA_ASYNC_INITIALIZE_RESTART) {
8026 mutex_enter(&spa_namespace_lock);
8027 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
8028 vdev_initialize_restart(spa->spa_root_vdev);
8029 spa_config_exit(spa, SCL_CONFIG, FTAG);
8030 mutex_exit(&spa_namespace_lock);
8033 if (tasks & SPA_ASYNC_TRIM_RESTART) {
8034 mutex_enter(&spa_namespace_lock);
8035 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
8036 vdev_trim_restart(spa->spa_root_vdev);
8037 spa_config_exit(spa, SCL_CONFIG, FTAG);
8038 mutex_exit(&spa_namespace_lock);
8041 if (tasks & SPA_ASYNC_AUTOTRIM_RESTART) {
8042 mutex_enter(&spa_namespace_lock);
8043 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
8044 vdev_autotrim_restart(spa);
8045 spa_config_exit(spa, SCL_CONFIG, FTAG);
8046 mutex_exit(&spa_namespace_lock);
8050 * Kick off L2 cache whole device TRIM.
8052 if (tasks & SPA_ASYNC_L2CACHE_TRIM) {
8053 mutex_enter(&spa_namespace_lock);
8054 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
8055 vdev_trim_l2arc(spa);
8056 spa_config_exit(spa, SCL_CONFIG, FTAG);
8057 mutex_exit(&spa_namespace_lock);
8061 * Kick off L2 cache rebuilding.
8063 if (tasks & SPA_ASYNC_L2CACHE_REBUILD) {
8064 mutex_enter(&spa_namespace_lock);
8065 spa_config_enter(spa, SCL_L2ARC, FTAG, RW_READER);
8066 l2arc_spa_rebuild_start(spa);
8067 spa_config_exit(spa, SCL_L2ARC, FTAG);
8068 mutex_exit(&spa_namespace_lock);
8072 * Let the world know that we're done.
8074 mutex_enter(&spa->spa_async_lock);
8075 spa->spa_async_thread = NULL;
8076 cv_broadcast(&spa->spa_async_cv);
8077 mutex_exit(&spa->spa_async_lock);
8082 spa_async_suspend(spa_t *spa)
8084 mutex_enter(&spa->spa_async_lock);
8085 spa->spa_async_suspended++;
8086 while (spa->spa_async_thread != NULL)
8087 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
8088 mutex_exit(&spa->spa_async_lock);
8090 spa_vdev_remove_suspend(spa);
8092 zthr_t *condense_thread = spa->spa_condense_zthr;
8093 if (condense_thread != NULL)
8094 zthr_cancel(condense_thread);
8096 zthr_t *discard_thread = spa->spa_checkpoint_discard_zthr;
8097 if (discard_thread != NULL)
8098 zthr_cancel(discard_thread);
8100 zthr_t *ll_delete_thread = spa->spa_livelist_delete_zthr;
8101 if (ll_delete_thread != NULL)
8102 zthr_cancel(ll_delete_thread);
8104 zthr_t *ll_condense_thread = spa->spa_livelist_condense_zthr;
8105 if (ll_condense_thread != NULL)
8106 zthr_cancel(ll_condense_thread);
8110 spa_async_resume(spa_t *spa)
8112 mutex_enter(&spa->spa_async_lock);
8113 ASSERT(spa->spa_async_suspended != 0);
8114 spa->spa_async_suspended--;
8115 mutex_exit(&spa->spa_async_lock);
8116 spa_restart_removal(spa);
8118 zthr_t *condense_thread = spa->spa_condense_zthr;
8119 if (condense_thread != NULL)
8120 zthr_resume(condense_thread);
8122 zthr_t *discard_thread = spa->spa_checkpoint_discard_zthr;
8123 if (discard_thread != NULL)
8124 zthr_resume(discard_thread);
8126 zthr_t *ll_delete_thread = spa->spa_livelist_delete_zthr;
8127 if (ll_delete_thread != NULL)
8128 zthr_resume(ll_delete_thread);
8130 zthr_t *ll_condense_thread = spa->spa_livelist_condense_zthr;
8131 if (ll_condense_thread != NULL)
8132 zthr_resume(ll_condense_thread);
8136 spa_async_tasks_pending(spa_t *spa)
8138 uint_t non_config_tasks;
8140 boolean_t config_task_suspended;
8142 non_config_tasks = spa->spa_async_tasks & ~SPA_ASYNC_CONFIG_UPDATE;
8143 config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE;
8144 if (spa->spa_ccw_fail_time == 0) {
8145 config_task_suspended = B_FALSE;
8147 config_task_suspended =
8148 (gethrtime() - spa->spa_ccw_fail_time) <
8149 ((hrtime_t)zfs_ccw_retry_interval * NANOSEC);
8152 return (non_config_tasks || (config_task && !config_task_suspended));
8156 spa_async_dispatch(spa_t *spa)
8158 mutex_enter(&spa->spa_async_lock);
8159 if (spa_async_tasks_pending(spa) &&
8160 !spa->spa_async_suspended &&
8161 spa->spa_async_thread == NULL)
8162 spa->spa_async_thread = thread_create(NULL, 0,
8163 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
8164 mutex_exit(&spa->spa_async_lock);
8168 spa_async_request(spa_t *spa, int task)
8170 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
8171 mutex_enter(&spa->spa_async_lock);
8172 spa->spa_async_tasks |= task;
8173 mutex_exit(&spa->spa_async_lock);
8177 spa_async_tasks(spa_t *spa)
8179 return (spa->spa_async_tasks);
8183 * ==========================================================================
8184 * SPA syncing routines
8185 * ==========================================================================
8190 bpobj_enqueue_cb(void *arg, const blkptr_t *bp, boolean_t bp_freed,
8194 bpobj_enqueue(bpo, bp, bp_freed, tx);
8199 bpobj_enqueue_alloc_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
8201 return (bpobj_enqueue_cb(arg, bp, B_FALSE, tx));
8205 bpobj_enqueue_free_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
8207 return (bpobj_enqueue_cb(arg, bp, B_TRUE, tx));
8211 spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
8215 zio_nowait(zio_free_sync(pio, pio->io_spa, dmu_tx_get_txg(tx), bp,
8221 bpobj_spa_free_sync_cb(void *arg, const blkptr_t *bp, boolean_t bp_freed,
8225 return (spa_free_sync_cb(arg, bp, tx));
8229 * Note: this simple function is not inlined to make it easier to dtrace the
8230 * amount of time spent syncing frees.
8233 spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx)
8235 zio_t *zio = zio_root(spa, NULL, NULL, 0);
8236 bplist_iterate(bpl, spa_free_sync_cb, zio, tx);
8237 VERIFY(zio_wait(zio) == 0);
8241 * Note: this simple function is not inlined to make it easier to dtrace the
8242 * amount of time spent syncing deferred frees.
8245 spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx)
8247 if (spa_sync_pass(spa) != 1)
8252 * If the log space map feature is active, we stop deferring
8253 * frees to the next TXG and therefore running this function
8254 * would be considered a no-op as spa_deferred_bpobj should
8255 * not have any entries.
8257 * That said we run this function anyway (instead of returning
8258 * immediately) for the edge-case scenario where we just
8259 * activated the log space map feature in this TXG but we have
8260 * deferred frees from the previous TXG.
8262 zio_t *zio = zio_root(spa, NULL, NULL, 0);
8263 VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj,
8264 bpobj_spa_free_sync_cb, zio, tx), ==, 0);
8265 VERIFY0(zio_wait(zio));
8269 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
8271 char *packed = NULL;
8276 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
8279 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
8280 * information. This avoids the dmu_buf_will_dirty() path and
8281 * saves us a pre-read to get data we don't actually care about.
8283 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
8284 packed = vmem_alloc(bufsize, KM_SLEEP);
8286 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
8288 bzero(packed + nvsize, bufsize - nvsize);
8290 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
8292 vmem_free(packed, bufsize);
8294 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
8295 dmu_buf_will_dirty(db, tx);
8296 *(uint64_t *)db->db_data = nvsize;
8297 dmu_buf_rele(db, FTAG);
8301 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
8302 const char *config, const char *entry)
8312 * Update the MOS nvlist describing the list of available devices.
8313 * spa_validate_aux() will have already made sure this nvlist is
8314 * valid and the vdevs are labeled appropriately.
8316 if (sav->sav_object == 0) {
8317 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
8318 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
8319 sizeof (uint64_t), tx);
8320 VERIFY(zap_update(spa->spa_meta_objset,
8321 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
8322 &sav->sav_object, tx) == 0);
8325 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
8326 if (sav->sav_count == 0) {
8327 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
8329 list = kmem_alloc(sav->sav_count*sizeof (void *), KM_SLEEP);
8330 for (i = 0; i < sav->sav_count; i++)
8331 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
8332 B_FALSE, VDEV_CONFIG_L2CACHE);
8333 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
8334 sav->sav_count) == 0);
8335 for (i = 0; i < sav->sav_count; i++)
8336 nvlist_free(list[i]);
8337 kmem_free(list, sav->sav_count * sizeof (void *));
8340 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
8341 nvlist_free(nvroot);
8343 sav->sav_sync = B_FALSE;
8347 * Rebuild spa's all-vdev ZAP from the vdev ZAPs indicated in each vdev_t.
8348 * The all-vdev ZAP must be empty.
8351 spa_avz_build(vdev_t *vd, uint64_t avz, dmu_tx_t *tx)
8353 spa_t *spa = vd->vdev_spa;
8355 if (vd->vdev_top_zap != 0) {
8356 VERIFY0(zap_add_int(spa->spa_meta_objset, avz,
8357 vd->vdev_top_zap, tx));
8359 if (vd->vdev_leaf_zap != 0) {
8360 VERIFY0(zap_add_int(spa->spa_meta_objset, avz,
8361 vd->vdev_leaf_zap, tx));
8363 for (uint64_t i = 0; i < vd->vdev_children; i++) {
8364 spa_avz_build(vd->vdev_child[i], avz, tx);
8369 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
8374 * If the pool is being imported from a pre-per-vdev-ZAP version of ZFS,
8375 * its config may not be dirty but we still need to build per-vdev ZAPs.
8376 * Similarly, if the pool is being assembled (e.g. after a split), we
8377 * need to rebuild the AVZ although the config may not be dirty.
8379 if (list_is_empty(&spa->spa_config_dirty_list) &&
8380 spa->spa_avz_action == AVZ_ACTION_NONE)
8383 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
8385 ASSERT(spa->spa_avz_action == AVZ_ACTION_NONE ||
8386 spa->spa_avz_action == AVZ_ACTION_INITIALIZE ||
8387 spa->spa_all_vdev_zaps != 0);
8389 if (spa->spa_avz_action == AVZ_ACTION_REBUILD) {
8390 /* Make and build the new AVZ */
8391 uint64_t new_avz = zap_create(spa->spa_meta_objset,
8392 DMU_OTN_ZAP_METADATA, DMU_OT_NONE, 0, tx);
8393 spa_avz_build(spa->spa_root_vdev, new_avz, tx);
8395 /* Diff old AVZ with new one */
8399 for (zap_cursor_init(&zc, spa->spa_meta_objset,
8400 spa->spa_all_vdev_zaps);
8401 zap_cursor_retrieve(&zc, &za) == 0;
8402 zap_cursor_advance(&zc)) {
8403 uint64_t vdzap = za.za_first_integer;
8404 if (zap_lookup_int(spa->spa_meta_objset, new_avz,
8407 * ZAP is listed in old AVZ but not in new one;
8410 VERIFY0(zap_destroy(spa->spa_meta_objset, vdzap,
8415 zap_cursor_fini(&zc);
8417 /* Destroy the old AVZ */
8418 VERIFY0(zap_destroy(spa->spa_meta_objset,
8419 spa->spa_all_vdev_zaps, tx));
8421 /* Replace the old AVZ in the dir obj with the new one */
8422 VERIFY0(zap_update(spa->spa_meta_objset,
8423 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP,
8424 sizeof (new_avz), 1, &new_avz, tx));
8426 spa->spa_all_vdev_zaps = new_avz;
8427 } else if (spa->spa_avz_action == AVZ_ACTION_DESTROY) {
8431 /* Walk through the AVZ and destroy all listed ZAPs */
8432 for (zap_cursor_init(&zc, spa->spa_meta_objset,
8433 spa->spa_all_vdev_zaps);
8434 zap_cursor_retrieve(&zc, &za) == 0;
8435 zap_cursor_advance(&zc)) {
8436 uint64_t zap = za.za_first_integer;
8437 VERIFY0(zap_destroy(spa->spa_meta_objset, zap, tx));
8440 zap_cursor_fini(&zc);
8442 /* Destroy and unlink the AVZ itself */
8443 VERIFY0(zap_destroy(spa->spa_meta_objset,
8444 spa->spa_all_vdev_zaps, tx));
8445 VERIFY0(zap_remove(spa->spa_meta_objset,
8446 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP, tx));
8447 spa->spa_all_vdev_zaps = 0;
8450 if (spa->spa_all_vdev_zaps == 0) {
8451 spa->spa_all_vdev_zaps = zap_create_link(spa->spa_meta_objset,
8452 DMU_OTN_ZAP_METADATA, DMU_POOL_DIRECTORY_OBJECT,
8453 DMU_POOL_VDEV_ZAP_MAP, tx);
8455 spa->spa_avz_action = AVZ_ACTION_NONE;
8457 /* Create ZAPs for vdevs that don't have them. */
8458 vdev_construct_zaps(spa->spa_root_vdev, tx);
8460 config = spa_config_generate(spa, spa->spa_root_vdev,
8461 dmu_tx_get_txg(tx), B_FALSE);
8464 * If we're upgrading the spa version then make sure that
8465 * the config object gets updated with the correct version.
8467 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
8468 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
8469 spa->spa_uberblock.ub_version);
8471 spa_config_exit(spa, SCL_STATE, FTAG);
8473 nvlist_free(spa->spa_config_syncing);
8474 spa->spa_config_syncing = config;
8476 spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
8480 spa_sync_version(void *arg, dmu_tx_t *tx)
8482 uint64_t *versionp = arg;
8483 uint64_t version = *versionp;
8484 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
8487 * Setting the version is special cased when first creating the pool.
8489 ASSERT(tx->tx_txg != TXG_INITIAL);
8491 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
8492 ASSERT(version >= spa_version(spa));
8494 spa->spa_uberblock.ub_version = version;
8495 vdev_config_dirty(spa->spa_root_vdev);
8496 spa_history_log_internal(spa, "set", tx, "version=%lld",
8497 (longlong_t)version);
8501 * Set zpool properties.
8504 spa_sync_props(void *arg, dmu_tx_t *tx)
8506 nvlist_t *nvp = arg;
8507 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
8508 objset_t *mos = spa->spa_meta_objset;
8509 nvpair_t *elem = NULL;
8511 mutex_enter(&spa->spa_props_lock);
8513 while ((elem = nvlist_next_nvpair(nvp, elem))) {
8515 char *strval, *fname;
8517 const char *propname;
8518 zprop_type_t proptype;
8521 switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
8522 case ZPOOL_PROP_INVAL:
8524 * We checked this earlier in spa_prop_validate().
8526 ASSERT(zpool_prop_feature(nvpair_name(elem)));
8528 fname = strchr(nvpair_name(elem), '@') + 1;
8529 VERIFY0(zfeature_lookup_name(fname, &fid));
8531 spa_feature_enable(spa, fid, tx);
8532 spa_history_log_internal(spa, "set", tx,
8533 "%s=enabled", nvpair_name(elem));
8536 case ZPOOL_PROP_VERSION:
8537 intval = fnvpair_value_uint64(elem);
8539 * The version is synced separately before other
8540 * properties and should be correct by now.
8542 ASSERT3U(spa_version(spa), >=, intval);
8545 case ZPOOL_PROP_ALTROOT:
8547 * 'altroot' is a non-persistent property. It should
8548 * have been set temporarily at creation or import time.
8550 ASSERT(spa->spa_root != NULL);
8553 case ZPOOL_PROP_READONLY:
8554 case ZPOOL_PROP_CACHEFILE:
8556 * 'readonly' and 'cachefile' are also non-persistent
8560 case ZPOOL_PROP_COMMENT:
8561 strval = fnvpair_value_string(elem);
8562 if (spa->spa_comment != NULL)
8563 spa_strfree(spa->spa_comment);
8564 spa->spa_comment = spa_strdup(strval);
8566 * We need to dirty the configuration on all the vdevs
8567 * so that their labels get updated. It's unnecessary
8568 * to do this for pool creation since the vdev's
8569 * configuration has already been dirtied.
8571 if (tx->tx_txg != TXG_INITIAL)
8572 vdev_config_dirty(spa->spa_root_vdev);
8573 spa_history_log_internal(spa, "set", tx,
8574 "%s=%s", nvpair_name(elem), strval);
8578 * Set pool property values in the poolprops mos object.
8580 if (spa->spa_pool_props_object == 0) {
8581 spa->spa_pool_props_object =
8582 zap_create_link(mos, DMU_OT_POOL_PROPS,
8583 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
8587 /* normalize the property name */
8588 propname = zpool_prop_to_name(prop);
8589 proptype = zpool_prop_get_type(prop);
8591 if (nvpair_type(elem) == DATA_TYPE_STRING) {
8592 ASSERT(proptype == PROP_TYPE_STRING);
8593 strval = fnvpair_value_string(elem);
8594 VERIFY0(zap_update(mos,
8595 spa->spa_pool_props_object, propname,
8596 1, strlen(strval) + 1, strval, tx));
8597 spa_history_log_internal(spa, "set", tx,
8598 "%s=%s", nvpair_name(elem), strval);
8599 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
8600 intval = fnvpair_value_uint64(elem);
8602 if (proptype == PROP_TYPE_INDEX) {
8604 VERIFY0(zpool_prop_index_to_string(
8605 prop, intval, &unused));
8607 VERIFY0(zap_update(mos,
8608 spa->spa_pool_props_object, propname,
8609 8, 1, &intval, tx));
8610 spa_history_log_internal(spa, "set", tx,
8611 "%s=%lld", nvpair_name(elem),
8612 (longlong_t)intval);
8614 ASSERT(0); /* not allowed */
8618 case ZPOOL_PROP_DELEGATION:
8619 spa->spa_delegation = intval;
8621 case ZPOOL_PROP_BOOTFS:
8622 spa->spa_bootfs = intval;
8624 case ZPOOL_PROP_FAILUREMODE:
8625 spa->spa_failmode = intval;
8627 case ZPOOL_PROP_AUTOTRIM:
8628 spa->spa_autotrim = intval;
8629 spa_async_request(spa,
8630 SPA_ASYNC_AUTOTRIM_RESTART);
8632 case ZPOOL_PROP_AUTOEXPAND:
8633 spa->spa_autoexpand = intval;
8634 if (tx->tx_txg != TXG_INITIAL)
8635 spa_async_request(spa,
8636 SPA_ASYNC_AUTOEXPAND);
8638 case ZPOOL_PROP_MULTIHOST:
8639 spa->spa_multihost = intval;
8648 mutex_exit(&spa->spa_props_lock);
8652 * Perform one-time upgrade on-disk changes. spa_version() does not
8653 * reflect the new version this txg, so there must be no changes this
8654 * txg to anything that the upgrade code depends on after it executes.
8655 * Therefore this must be called after dsl_pool_sync() does the sync
8659 spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
8661 if (spa_sync_pass(spa) != 1)
8664 dsl_pool_t *dp = spa->spa_dsl_pool;
8665 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
8667 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
8668 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
8669 dsl_pool_create_origin(dp, tx);
8671 /* Keeping the origin open increases spa_minref */
8672 spa->spa_minref += 3;
8675 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
8676 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
8677 dsl_pool_upgrade_clones(dp, tx);
8680 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
8681 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
8682 dsl_pool_upgrade_dir_clones(dp, tx);
8684 /* Keeping the freedir open increases spa_minref */
8685 spa->spa_minref += 3;
8688 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES &&
8689 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
8690 spa_feature_create_zap_objects(spa, tx);
8694 * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable
8695 * when possibility to use lz4 compression for metadata was added
8696 * Old pools that have this feature enabled must be upgraded to have
8697 * this feature active
8699 if (spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
8700 boolean_t lz4_en = spa_feature_is_enabled(spa,
8701 SPA_FEATURE_LZ4_COMPRESS);
8702 boolean_t lz4_ac = spa_feature_is_active(spa,
8703 SPA_FEATURE_LZ4_COMPRESS);
8705 if (lz4_en && !lz4_ac)
8706 spa_feature_incr(spa, SPA_FEATURE_LZ4_COMPRESS, tx);
8710 * If we haven't written the salt, do so now. Note that the
8711 * feature may not be activated yet, but that's fine since
8712 * the presence of this ZAP entry is backwards compatible.
8714 if (zap_contains(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
8715 DMU_POOL_CHECKSUM_SALT) == ENOENT) {
8716 VERIFY0(zap_add(spa->spa_meta_objset,
8717 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CHECKSUM_SALT, 1,
8718 sizeof (spa->spa_cksum_salt.zcs_bytes),
8719 spa->spa_cksum_salt.zcs_bytes, tx));
8722 rrw_exit(&dp->dp_config_rwlock, FTAG);
8726 vdev_indirect_state_sync_verify(vdev_t *vd)
8728 vdev_indirect_mapping_t *vim __maybe_unused = vd->vdev_indirect_mapping;
8729 vdev_indirect_births_t *vib __maybe_unused = vd->vdev_indirect_births;
8731 if (vd->vdev_ops == &vdev_indirect_ops) {
8732 ASSERT(vim != NULL);
8733 ASSERT(vib != NULL);
8736 uint64_t obsolete_sm_object = 0;
8737 ASSERT0(vdev_obsolete_sm_object(vd, &obsolete_sm_object));
8738 if (obsolete_sm_object != 0) {
8739 ASSERT(vd->vdev_obsolete_sm != NULL);
8740 ASSERT(vd->vdev_removing ||
8741 vd->vdev_ops == &vdev_indirect_ops);
8742 ASSERT(vdev_indirect_mapping_num_entries(vim) > 0);
8743 ASSERT(vdev_indirect_mapping_bytes_mapped(vim) > 0);
8744 ASSERT3U(obsolete_sm_object, ==,
8745 space_map_object(vd->vdev_obsolete_sm));
8746 ASSERT3U(vdev_indirect_mapping_bytes_mapped(vim), >=,
8747 space_map_allocated(vd->vdev_obsolete_sm));
8749 ASSERT(vd->vdev_obsolete_segments != NULL);
8752 * Since frees / remaps to an indirect vdev can only
8753 * happen in syncing context, the obsolete segments
8754 * tree must be empty when we start syncing.
8756 ASSERT0(range_tree_space(vd->vdev_obsolete_segments));
8760 * Set the top-level vdev's max queue depth. Evaluate each top-level's
8761 * async write queue depth in case it changed. The max queue depth will
8762 * not change in the middle of syncing out this txg.
8765 spa_sync_adjust_vdev_max_queue_depth(spa_t *spa)
8767 ASSERT(spa_writeable(spa));
8769 vdev_t *rvd = spa->spa_root_vdev;
8770 uint32_t max_queue_depth = zfs_vdev_async_write_max_active *
8771 zfs_vdev_queue_depth_pct / 100;
8772 metaslab_class_t *normal = spa_normal_class(spa);
8773 metaslab_class_t *special = spa_special_class(spa);
8774 metaslab_class_t *dedup = spa_dedup_class(spa);
8776 uint64_t slots_per_allocator = 0;
8777 for (int c = 0; c < rvd->vdev_children; c++) {
8778 vdev_t *tvd = rvd->vdev_child[c];
8780 metaslab_group_t *mg = tvd->vdev_mg;
8781 if (mg == NULL || !metaslab_group_initialized(mg))
8784 metaslab_class_t *mc = mg->mg_class;
8785 if (mc != normal && mc != special && mc != dedup)
8789 * It is safe to do a lock-free check here because only async
8790 * allocations look at mg_max_alloc_queue_depth, and async
8791 * allocations all happen from spa_sync().
8793 for (int i = 0; i < mg->mg_allocators; i++) {
8794 ASSERT0(zfs_refcount_count(
8795 &(mg->mg_allocator[i].mga_alloc_queue_depth)));
8797 mg->mg_max_alloc_queue_depth = max_queue_depth;
8799 for (int i = 0; i < mg->mg_allocators; i++) {
8800 mg->mg_allocator[i].mga_cur_max_alloc_queue_depth =
8801 zfs_vdev_def_queue_depth;
8803 slots_per_allocator += zfs_vdev_def_queue_depth;
8806 for (int i = 0; i < spa->spa_alloc_count; i++) {
8807 ASSERT0(zfs_refcount_count(&normal->mc_alloc_slots[i]));
8808 ASSERT0(zfs_refcount_count(&special->mc_alloc_slots[i]));
8809 ASSERT0(zfs_refcount_count(&dedup->mc_alloc_slots[i]));
8810 normal->mc_alloc_max_slots[i] = slots_per_allocator;
8811 special->mc_alloc_max_slots[i] = slots_per_allocator;
8812 dedup->mc_alloc_max_slots[i] = slots_per_allocator;
8814 normal->mc_alloc_throttle_enabled = zio_dva_throttle_enabled;
8815 special->mc_alloc_throttle_enabled = zio_dva_throttle_enabled;
8816 dedup->mc_alloc_throttle_enabled = zio_dva_throttle_enabled;
8820 spa_sync_condense_indirect(spa_t *spa, dmu_tx_t *tx)
8822 ASSERT(spa_writeable(spa));
8824 vdev_t *rvd = spa->spa_root_vdev;
8825 for (int c = 0; c < rvd->vdev_children; c++) {
8826 vdev_t *vd = rvd->vdev_child[c];
8827 vdev_indirect_state_sync_verify(vd);
8829 if (vdev_indirect_should_condense(vd)) {
8830 spa_condense_indirect_start_sync(vd, tx);
8837 spa_sync_iterate_to_convergence(spa_t *spa, dmu_tx_t *tx)
8839 objset_t *mos = spa->spa_meta_objset;
8840 dsl_pool_t *dp = spa->spa_dsl_pool;
8841 uint64_t txg = tx->tx_txg;
8842 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
8845 int pass = ++spa->spa_sync_pass;
8847 spa_sync_config_object(spa, tx);
8848 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
8849 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
8850 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
8851 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
8852 spa_errlog_sync(spa, txg);
8853 dsl_pool_sync(dp, txg);
8855 if (pass < zfs_sync_pass_deferred_free ||
8856 spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP)) {
8858 * If the log space map feature is active we don't
8859 * care about deferred frees and the deferred bpobj
8860 * as the log space map should effectively have the
8861 * same results (i.e. appending only to one object).
8863 spa_sync_frees(spa, free_bpl, tx);
8866 * We can not defer frees in pass 1, because
8867 * we sync the deferred frees later in pass 1.
8869 ASSERT3U(pass, >, 1);
8870 bplist_iterate(free_bpl, bpobj_enqueue_alloc_cb,
8871 &spa->spa_deferred_bpobj, tx);
8875 dsl_scan_sync(dp, tx);
8877 spa_sync_upgrades(spa, tx);
8879 spa_flush_metaslabs(spa, tx);
8882 while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
8887 * Note: We need to check if the MOS is dirty because we could
8888 * have marked the MOS dirty without updating the uberblock
8889 * (e.g. if we have sync tasks but no dirty user data). We need
8890 * to check the uberblock's rootbp because it is updated if we
8891 * have synced out dirty data (though in this case the MOS will
8892 * most likely also be dirty due to second order effects, we
8893 * don't want to rely on that here).
8896 spa->spa_uberblock.ub_rootbp.blk_birth < txg &&
8897 !dmu_objset_is_dirty(mos, txg)) {
8899 * Nothing changed on the first pass, therefore this
8900 * TXG is a no-op. Avoid syncing deferred frees, so
8901 * that we can keep this TXG as a no-op.
8903 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
8904 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
8905 ASSERT(txg_list_empty(&dp->dp_sync_tasks, txg));
8906 ASSERT(txg_list_empty(&dp->dp_early_sync_tasks, txg));
8910 spa_sync_deferred_frees(spa, tx);
8911 } while (dmu_objset_is_dirty(mos, txg));
8915 * Rewrite the vdev configuration (which includes the uberblock) to
8916 * commit the transaction group.
8918 * If there are no dirty vdevs, we sync the uberblock to a few random
8919 * top-level vdevs that are known to be visible in the config cache
8920 * (see spa_vdev_add() for a complete description). If there *are* dirty
8921 * vdevs, sync the uberblock to all vdevs.
8924 spa_sync_rewrite_vdev_config(spa_t *spa, dmu_tx_t *tx)
8926 vdev_t *rvd = spa->spa_root_vdev;
8927 uint64_t txg = tx->tx_txg;
8933 * We hold SCL_STATE to prevent vdev open/close/etc.
8934 * while we're attempting to write the vdev labels.
8936 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
8938 if (list_is_empty(&spa->spa_config_dirty_list)) {
8939 vdev_t *svd[SPA_SYNC_MIN_VDEVS] = { NULL };
8941 int children = rvd->vdev_children;
8942 int c0 = spa_get_random(children);
8944 for (int c = 0; c < children; c++) {
8946 rvd->vdev_child[(c0 + c) % children];
8948 /* Stop when revisiting the first vdev */
8949 if (c > 0 && svd[0] == vd)
8952 if (vd->vdev_ms_array == 0 ||
8954 !vdev_is_concrete(vd))
8957 svd[svdcount++] = vd;
8958 if (svdcount == SPA_SYNC_MIN_VDEVS)
8961 error = vdev_config_sync(svd, svdcount, txg);
8963 error = vdev_config_sync(rvd->vdev_child,
8964 rvd->vdev_children, txg);
8968 spa->spa_last_synced_guid = rvd->vdev_guid;
8970 spa_config_exit(spa, SCL_STATE, FTAG);
8974 zio_suspend(spa, NULL, ZIO_SUSPEND_IOERR);
8975 zio_resume_wait(spa);
8980 * Sync the specified transaction group. New blocks may be dirtied as
8981 * part of the process, so we iterate until it converges.
8984 spa_sync(spa_t *spa, uint64_t txg)
8988 VERIFY(spa_writeable(spa));
8991 * Wait for i/os issued in open context that need to complete
8992 * before this txg syncs.
8994 (void) zio_wait(spa->spa_txg_zio[txg & TXG_MASK]);
8995 spa->spa_txg_zio[txg & TXG_MASK] = zio_root(spa, NULL, NULL,
8999 * Lock out configuration changes.
9001 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
9003 spa->spa_syncing_txg = txg;
9004 spa->spa_sync_pass = 0;
9006 for (int i = 0; i < spa->spa_alloc_count; i++) {
9007 mutex_enter(&spa->spa_alloc_locks[i]);
9008 VERIFY0(avl_numnodes(&spa->spa_alloc_trees[i]));
9009 mutex_exit(&spa->spa_alloc_locks[i]);
9013 * If there are any pending vdev state changes, convert them
9014 * into config changes that go out with this transaction group.
9016 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
9017 while (list_head(&spa->spa_state_dirty_list) != NULL) {
9019 * We need the write lock here because, for aux vdevs,
9020 * calling vdev_config_dirty() modifies sav_config.
9021 * This is ugly and will become unnecessary when we
9022 * eliminate the aux vdev wart by integrating all vdevs
9023 * into the root vdev tree.
9025 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
9026 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
9027 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
9028 vdev_state_clean(vd);
9029 vdev_config_dirty(vd);
9031 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
9032 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
9034 spa_config_exit(spa, SCL_STATE, FTAG);
9036 dsl_pool_t *dp = spa->spa_dsl_pool;
9037 dmu_tx_t *tx = dmu_tx_create_assigned(dp, txg);
9039 spa->spa_sync_starttime = gethrtime();
9040 taskq_cancel_id(system_delay_taskq, spa->spa_deadman_tqid);
9041 spa->spa_deadman_tqid = taskq_dispatch_delay(system_delay_taskq,
9042 spa_deadman, spa, TQ_SLEEP, ddi_get_lbolt() +
9043 NSEC_TO_TICK(spa->spa_deadman_synctime));
9046 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
9047 * set spa_deflate if we have no raid-z vdevs.
9049 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
9050 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
9051 vdev_t *rvd = spa->spa_root_vdev;
9054 for (i = 0; i < rvd->vdev_children; i++) {
9055 vd = rvd->vdev_child[i];
9056 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
9059 if (i == rvd->vdev_children) {
9060 spa->spa_deflate = TRUE;
9061 VERIFY0(zap_add(spa->spa_meta_objset,
9062 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
9063 sizeof (uint64_t), 1, &spa->spa_deflate, tx));
9067 spa_sync_adjust_vdev_max_queue_depth(spa);
9069 spa_sync_condense_indirect(spa, tx);
9071 spa_sync_iterate_to_convergence(spa, tx);
9074 if (!list_is_empty(&spa->spa_config_dirty_list)) {
9076 * Make sure that the number of ZAPs for all the vdevs matches
9077 * the number of ZAPs in the per-vdev ZAP list. This only gets
9078 * called if the config is dirty; otherwise there may be
9079 * outstanding AVZ operations that weren't completed in
9080 * spa_sync_config_object.
9082 uint64_t all_vdev_zap_entry_count;
9083 ASSERT0(zap_count(spa->spa_meta_objset,
9084 spa->spa_all_vdev_zaps, &all_vdev_zap_entry_count));
9085 ASSERT3U(vdev_count_verify_zaps(spa->spa_root_vdev), ==,
9086 all_vdev_zap_entry_count);
9090 if (spa->spa_vdev_removal != NULL) {
9091 ASSERT0(spa->spa_vdev_removal->svr_bytes_done[txg & TXG_MASK]);
9094 spa_sync_rewrite_vdev_config(spa, tx);
9097 taskq_cancel_id(system_delay_taskq, spa->spa_deadman_tqid);
9098 spa->spa_deadman_tqid = 0;
9101 * Clear the dirty config list.
9103 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
9104 vdev_config_clean(vd);
9107 * Now that the new config has synced transactionally,
9108 * let it become visible to the config cache.
9110 if (spa->spa_config_syncing != NULL) {
9111 spa_config_set(spa, spa->spa_config_syncing);
9112 spa->spa_config_txg = txg;
9113 spa->spa_config_syncing = NULL;
9116 dsl_pool_sync_done(dp, txg);
9118 for (int i = 0; i < spa->spa_alloc_count; i++) {
9119 mutex_enter(&spa->spa_alloc_locks[i]);
9120 VERIFY0(avl_numnodes(&spa->spa_alloc_trees[i]));
9121 mutex_exit(&spa->spa_alloc_locks[i]);
9125 * Update usable space statistics.
9127 while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
9129 vdev_sync_done(vd, txg);
9131 metaslab_class_evict_old(spa->spa_normal_class, txg);
9132 metaslab_class_evict_old(spa->spa_log_class, txg);
9134 spa_sync_close_syncing_log_sm(spa);
9136 spa_update_dspace(spa);
9139 * It had better be the case that we didn't dirty anything
9140 * since vdev_config_sync().
9142 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
9143 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
9144 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
9146 while (zfs_pause_spa_sync)
9149 spa->spa_sync_pass = 0;
9152 * Update the last synced uberblock here. We want to do this at
9153 * the end of spa_sync() so that consumers of spa_last_synced_txg()
9154 * will be guaranteed that all the processing associated with
9155 * that txg has been completed.
9157 spa->spa_ubsync = spa->spa_uberblock;
9158 spa_config_exit(spa, SCL_CONFIG, FTAG);
9160 spa_handle_ignored_writes(spa);
9163 * If any async tasks have been requested, kick them off.
9165 spa_async_dispatch(spa);
9169 * Sync all pools. We don't want to hold the namespace lock across these
9170 * operations, so we take a reference on the spa_t and drop the lock during the
9174 spa_sync_allpools(void)
9177 mutex_enter(&spa_namespace_lock);
9178 while ((spa = spa_next(spa)) != NULL) {
9179 if (spa_state(spa) != POOL_STATE_ACTIVE ||
9180 !spa_writeable(spa) || spa_suspended(spa))
9182 spa_open_ref(spa, FTAG);
9183 mutex_exit(&spa_namespace_lock);
9184 txg_wait_synced(spa_get_dsl(spa), 0);
9185 mutex_enter(&spa_namespace_lock);
9186 spa_close(spa, FTAG);
9188 mutex_exit(&spa_namespace_lock);
9192 * ==========================================================================
9193 * Miscellaneous routines
9194 * ==========================================================================
9198 * Remove all pools in the system.
9206 * Remove all cached state. All pools should be closed now,
9207 * so every spa in the AVL tree should be unreferenced.
9209 mutex_enter(&spa_namespace_lock);
9210 while ((spa = spa_next(NULL)) != NULL) {
9212 * Stop async tasks. The async thread may need to detach
9213 * a device that's been replaced, which requires grabbing
9214 * spa_namespace_lock, so we must drop it here.
9216 spa_open_ref(spa, FTAG);
9217 mutex_exit(&spa_namespace_lock);
9218 spa_async_suspend(spa);
9219 mutex_enter(&spa_namespace_lock);
9220 spa_close(spa, FTAG);
9222 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
9224 spa_deactivate(spa);
9228 mutex_exit(&spa_namespace_lock);
9232 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
9237 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
9241 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
9242 vd = spa->spa_l2cache.sav_vdevs[i];
9243 if (vd->vdev_guid == guid)
9247 for (i = 0; i < spa->spa_spares.sav_count; i++) {
9248 vd = spa->spa_spares.sav_vdevs[i];
9249 if (vd->vdev_guid == guid)
9258 spa_upgrade(spa_t *spa, uint64_t version)
9260 ASSERT(spa_writeable(spa));
9262 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
9265 * This should only be called for a non-faulted pool, and since a
9266 * future version would result in an unopenable pool, this shouldn't be
9269 ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version));
9270 ASSERT3U(version, >=, spa->spa_uberblock.ub_version);
9272 spa->spa_uberblock.ub_version = version;
9273 vdev_config_dirty(spa->spa_root_vdev);
9275 spa_config_exit(spa, SCL_ALL, FTAG);
9277 txg_wait_synced(spa_get_dsl(spa), 0);
9281 spa_has_spare(spa_t *spa, uint64_t guid)
9285 spa_aux_vdev_t *sav = &spa->spa_spares;
9287 for (i = 0; i < sav->sav_count; i++)
9288 if (sav->sav_vdevs[i]->vdev_guid == guid)
9291 for (i = 0; i < sav->sav_npending; i++) {
9292 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
9293 &spareguid) == 0 && spareguid == guid)
9301 * Check if a pool has an active shared spare device.
9302 * Note: reference count of an active spare is 2, as a spare and as a replace
9305 spa_has_active_shared_spare(spa_t *spa)
9309 spa_aux_vdev_t *sav = &spa->spa_spares;
9311 for (i = 0; i < sav->sav_count; i++) {
9312 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
9313 &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
9322 spa_total_metaslabs(spa_t *spa)
9324 vdev_t *rvd = spa->spa_root_vdev;
9327 for (uint64_t c = 0; c < rvd->vdev_children; c++) {
9328 vdev_t *vd = rvd->vdev_child[c];
9329 if (!vdev_is_concrete(vd))
9331 m += vd->vdev_ms_count;
9337 * Notify any waiting threads that some activity has switched from being in-
9338 * progress to not-in-progress so that the thread can wake up and determine
9339 * whether it is finished waiting.
9342 spa_notify_waiters(spa_t *spa)
9345 * Acquiring spa_activities_lock here prevents the cv_broadcast from
9346 * happening between the waiting thread's check and cv_wait.
9348 mutex_enter(&spa->spa_activities_lock);
9349 cv_broadcast(&spa->spa_activities_cv);
9350 mutex_exit(&spa->spa_activities_lock);
9354 * Notify any waiting threads that the pool is exporting, and then block until
9355 * they are finished using the spa_t.
9358 spa_wake_waiters(spa_t *spa)
9360 mutex_enter(&spa->spa_activities_lock);
9361 spa->spa_waiters_cancel = B_TRUE;
9362 cv_broadcast(&spa->spa_activities_cv);
9363 while (spa->spa_waiters != 0)
9364 cv_wait(&spa->spa_waiters_cv, &spa->spa_activities_lock);
9365 spa->spa_waiters_cancel = B_FALSE;
9366 mutex_exit(&spa->spa_activities_lock);
9369 /* Whether the vdev or any of its descendants are being initialized/trimmed. */
9371 spa_vdev_activity_in_progress_impl(vdev_t *vd, zpool_wait_activity_t activity)
9373 spa_t *spa = vd->vdev_spa;
9375 ASSERT(spa_config_held(spa, SCL_CONFIG | SCL_STATE, RW_READER));
9376 ASSERT(MUTEX_HELD(&spa->spa_activities_lock));
9377 ASSERT(activity == ZPOOL_WAIT_INITIALIZE ||
9378 activity == ZPOOL_WAIT_TRIM);
9380 kmutex_t *lock = activity == ZPOOL_WAIT_INITIALIZE ?
9381 &vd->vdev_initialize_lock : &vd->vdev_trim_lock;
9383 mutex_exit(&spa->spa_activities_lock);
9385 mutex_enter(&spa->spa_activities_lock);
9387 boolean_t in_progress = (activity == ZPOOL_WAIT_INITIALIZE) ?
9388 (vd->vdev_initialize_state == VDEV_INITIALIZE_ACTIVE) :
9389 (vd->vdev_trim_state == VDEV_TRIM_ACTIVE);
9395 for (int i = 0; i < vd->vdev_children; i++) {
9396 if (spa_vdev_activity_in_progress_impl(vd->vdev_child[i],
9405 * If use_guid is true, this checks whether the vdev specified by guid is
9406 * being initialized/trimmed. Otherwise, it checks whether any vdev in the pool
9407 * is being initialized/trimmed. The caller must hold the config lock and
9408 * spa_activities_lock.
9411 spa_vdev_activity_in_progress(spa_t *spa, boolean_t use_guid, uint64_t guid,
9412 zpool_wait_activity_t activity, boolean_t *in_progress)
9414 mutex_exit(&spa->spa_activities_lock);
9415 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
9416 mutex_enter(&spa->spa_activities_lock);
9420 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
9421 if (vd == NULL || !vd->vdev_ops->vdev_op_leaf) {
9422 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
9426 vd = spa->spa_root_vdev;
9429 *in_progress = spa_vdev_activity_in_progress_impl(vd, activity);
9431 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
9436 * Locking for waiting threads
9437 * ---------------------------
9439 * Waiting threads need a way to check whether a given activity is in progress,
9440 * and then, if it is, wait for it to complete. Each activity will have some
9441 * in-memory representation of the relevant on-disk state which can be used to
9442 * determine whether or not the activity is in progress. The in-memory state and
9443 * the locking used to protect it will be different for each activity, and may
9444 * not be suitable for use with a cvar (e.g., some state is protected by the
9445 * config lock). To allow waiting threads to wait without any races, another
9446 * lock, spa_activities_lock, is used.
9448 * When the state is checked, both the activity-specific lock (if there is one)
9449 * and spa_activities_lock are held. In some cases, the activity-specific lock
9450 * is acquired explicitly (e.g. the config lock). In others, the locking is
9451 * internal to some check (e.g. bpobj_is_empty). After checking, the waiting
9452 * thread releases the activity-specific lock and, if the activity is in
9453 * progress, then cv_waits using spa_activities_lock.
9455 * The waiting thread is woken when another thread, one completing some
9456 * activity, updates the state of the activity and then calls
9457 * spa_notify_waiters, which will cv_broadcast. This 'completing' thread only
9458 * needs to hold its activity-specific lock when updating the state, and this
9459 * lock can (but doesn't have to) be dropped before calling spa_notify_waiters.
9461 * Because spa_notify_waiters acquires spa_activities_lock before broadcasting,
9462 * and because it is held when the waiting thread checks the state of the
9463 * activity, it can never be the case that the completing thread both updates
9464 * the activity state and cv_broadcasts in between the waiting thread's check
9465 * and cv_wait. Thus, a waiting thread can never miss a wakeup.
9467 * In order to prevent deadlock, when the waiting thread does its check, in some
9468 * cases it will temporarily drop spa_activities_lock in order to acquire the
9469 * activity-specific lock. The order in which spa_activities_lock and the
9470 * activity specific lock are acquired in the waiting thread is determined by
9471 * the order in which they are acquired in the completing thread; if the
9472 * completing thread calls spa_notify_waiters with the activity-specific lock
9473 * held, then the waiting thread must also acquire the activity-specific lock
9478 spa_activity_in_progress(spa_t *spa, zpool_wait_activity_t activity,
9479 boolean_t use_tag, uint64_t tag, boolean_t *in_progress)
9483 ASSERT(MUTEX_HELD(&spa->spa_activities_lock));
9486 case ZPOOL_WAIT_CKPT_DISCARD:
9488 (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT) &&
9489 zap_contains(spa_meta_objset(spa),
9490 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_ZPOOL_CHECKPOINT) ==
9493 case ZPOOL_WAIT_FREE:
9494 *in_progress = ((spa_version(spa) >= SPA_VERSION_DEADLISTS &&
9495 !bpobj_is_empty(&spa->spa_dsl_pool->dp_free_bpobj)) ||
9496 spa_feature_is_active(spa, SPA_FEATURE_ASYNC_DESTROY) ||
9497 spa_livelist_delete_check(spa));
9499 case ZPOOL_WAIT_INITIALIZE:
9500 case ZPOOL_WAIT_TRIM:
9501 error = spa_vdev_activity_in_progress(spa, use_tag, tag,
9502 activity, in_progress);
9504 case ZPOOL_WAIT_REPLACE:
9505 mutex_exit(&spa->spa_activities_lock);
9506 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
9507 mutex_enter(&spa->spa_activities_lock);
9509 *in_progress = vdev_replace_in_progress(spa->spa_root_vdev);
9510 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
9512 case ZPOOL_WAIT_REMOVE:
9513 *in_progress = (spa->spa_removing_phys.sr_state ==
9516 case ZPOOL_WAIT_RESILVER:
9517 if ((*in_progress = vdev_rebuild_active(spa->spa_root_vdev)))
9520 case ZPOOL_WAIT_SCRUB:
9522 boolean_t scanning, paused, is_scrub;
9523 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
9525 is_scrub = (scn->scn_phys.scn_func == POOL_SCAN_SCRUB);
9526 scanning = (scn->scn_phys.scn_state == DSS_SCANNING);
9527 paused = dsl_scan_is_paused_scrub(scn);
9528 *in_progress = (scanning && !paused &&
9529 is_scrub == (activity == ZPOOL_WAIT_SCRUB));
9533 panic("unrecognized value for activity %d", activity);
9540 spa_wait_common(const char *pool, zpool_wait_activity_t activity,
9541 boolean_t use_tag, uint64_t tag, boolean_t *waited)
9544 * The tag is used to distinguish between instances of an activity.
9545 * 'initialize' and 'trim' are the only activities that we use this for.
9546 * The other activities can only have a single instance in progress in a
9547 * pool at one time, making the tag unnecessary.
9549 * There can be multiple devices being replaced at once, but since they
9550 * all finish once resilvering finishes, we don't bother keeping track
9551 * of them individually, we just wait for them all to finish.
9553 if (use_tag && activity != ZPOOL_WAIT_INITIALIZE &&
9554 activity != ZPOOL_WAIT_TRIM)
9557 if (activity < 0 || activity >= ZPOOL_WAIT_NUM_ACTIVITIES)
9561 int error = spa_open(pool, &spa, FTAG);
9566 * Increment the spa's waiter count so that we can call spa_close and
9567 * still ensure that the spa_t doesn't get freed before this thread is
9568 * finished with it when the pool is exported. We want to call spa_close
9569 * before we start waiting because otherwise the additional ref would
9570 * prevent the pool from being exported or destroyed throughout the
9571 * potentially long wait.
9573 mutex_enter(&spa->spa_activities_lock);
9575 spa_close(spa, FTAG);
9579 boolean_t in_progress;
9580 error = spa_activity_in_progress(spa, activity, use_tag, tag,
9583 if (error || !in_progress || spa->spa_waiters_cancel)
9588 if (cv_wait_sig(&spa->spa_activities_cv,
9589 &spa->spa_activities_lock) == 0) {
9596 cv_signal(&spa->spa_waiters_cv);
9597 mutex_exit(&spa->spa_activities_lock);
9603 * Wait for a particular instance of the specified activity to complete, where
9604 * the instance is identified by 'tag'
9607 spa_wait_tag(const char *pool, zpool_wait_activity_t activity, uint64_t tag,
9610 return (spa_wait_common(pool, activity, B_TRUE, tag, waited));
9614 * Wait for all instances of the specified activity complete
9617 spa_wait(const char *pool, zpool_wait_activity_t activity, boolean_t *waited)
9620 return (spa_wait_common(pool, activity, B_FALSE, 0, waited));
9624 spa_event_create(spa_t *spa, vdev_t *vd, nvlist_t *hist_nvl, const char *name)
9626 sysevent_t *ev = NULL;
9630 resource = zfs_event_create(spa, vd, FM_SYSEVENT_CLASS, name, hist_nvl);
9632 ev = kmem_alloc(sizeof (sysevent_t), KM_SLEEP);
9633 ev->resource = resource;
9640 spa_event_post(sysevent_t *ev)
9644 zfs_zevent_post(ev->resource, NULL, zfs_zevent_post_cb);
9645 kmem_free(ev, sizeof (*ev));
9651 * Post a zevent corresponding to the given sysevent. The 'name' must be one
9652 * of the event definitions in sys/sysevent/eventdefs.h. The payload will be
9653 * filled in from the spa and (optionally) the vdev. This doesn't do anything
9654 * in the userland libzpool, as we don't want consumers to misinterpret ztest
9655 * or zdb as real changes.
9658 spa_event_notify(spa_t *spa, vdev_t *vd, nvlist_t *hist_nvl, const char *name)
9660 spa_event_post(spa_event_create(spa, vd, hist_nvl, name));
9663 /* state manipulation functions */
9664 EXPORT_SYMBOL(spa_open);
9665 EXPORT_SYMBOL(spa_open_rewind);
9666 EXPORT_SYMBOL(spa_get_stats);
9667 EXPORT_SYMBOL(spa_create);
9668 EXPORT_SYMBOL(spa_import);
9669 EXPORT_SYMBOL(spa_tryimport);
9670 EXPORT_SYMBOL(spa_destroy);
9671 EXPORT_SYMBOL(spa_export);
9672 EXPORT_SYMBOL(spa_reset);
9673 EXPORT_SYMBOL(spa_async_request);
9674 EXPORT_SYMBOL(spa_async_suspend);
9675 EXPORT_SYMBOL(spa_async_resume);
9676 EXPORT_SYMBOL(spa_inject_addref);
9677 EXPORT_SYMBOL(spa_inject_delref);
9678 EXPORT_SYMBOL(spa_scan_stat_init);
9679 EXPORT_SYMBOL(spa_scan_get_stats);
9681 /* device manipulation */
9682 EXPORT_SYMBOL(spa_vdev_add);
9683 EXPORT_SYMBOL(spa_vdev_attach);
9684 EXPORT_SYMBOL(spa_vdev_detach);
9685 EXPORT_SYMBOL(spa_vdev_setpath);
9686 EXPORT_SYMBOL(spa_vdev_setfru);
9687 EXPORT_SYMBOL(spa_vdev_split_mirror);
9689 /* spare statech is global across all pools) */
9690 EXPORT_SYMBOL(spa_spare_add);
9691 EXPORT_SYMBOL(spa_spare_remove);
9692 EXPORT_SYMBOL(spa_spare_exists);
9693 EXPORT_SYMBOL(spa_spare_activate);
9695 /* L2ARC statech is global across all pools) */
9696 EXPORT_SYMBOL(spa_l2cache_add);
9697 EXPORT_SYMBOL(spa_l2cache_remove);
9698 EXPORT_SYMBOL(spa_l2cache_exists);
9699 EXPORT_SYMBOL(spa_l2cache_activate);
9700 EXPORT_SYMBOL(spa_l2cache_drop);
9703 EXPORT_SYMBOL(spa_scan);
9704 EXPORT_SYMBOL(spa_scan_stop);
9707 EXPORT_SYMBOL(spa_sync); /* only for DMU use */
9708 EXPORT_SYMBOL(spa_sync_allpools);
9711 EXPORT_SYMBOL(spa_prop_set);
9712 EXPORT_SYMBOL(spa_prop_get);
9713 EXPORT_SYMBOL(spa_prop_clear_bootfs);
9715 /* asynchronous event notification */
9716 EXPORT_SYMBOL(spa_event_notify);
9719 ZFS_MODULE_PARAM(zfs_spa, spa_, load_verify_shift, INT, ZMOD_RW,
9720 "log2(fraction of arc that can be used by inflight I/Os when "
9721 "verifying pool during import");
9723 ZFS_MODULE_PARAM(zfs_spa, spa_, load_verify_metadata, INT, ZMOD_RW,
9724 "Set to traverse metadata on pool import");
9726 ZFS_MODULE_PARAM(zfs_spa, spa_, load_verify_data, INT, ZMOD_RW,
9727 "Set to traverse data on pool import");
9729 ZFS_MODULE_PARAM(zfs_spa, spa_, load_print_vdev_tree, INT, ZMOD_RW,
9730 "Print vdev tree to zfs_dbgmsg during pool import");
9732 ZFS_MODULE_PARAM(zfs_zio, zio_, taskq_batch_pct, UINT, ZMOD_RD,
9733 "Percentage of CPUs to run an IO worker thread");
9735 ZFS_MODULE_PARAM(zfs, zfs_, max_missing_tvds, ULONG, ZMOD_RW,
9736 "Allow importing pool with up to this number of missing top-level "
9737 "vdevs (in read-only mode)");
9739 ZFS_MODULE_PARAM(zfs_livelist_condense, zfs_livelist_condense_, zthr_pause, INT, ZMOD_RW,
9740 "Set the livelist condense zthr to pause");
9742 ZFS_MODULE_PARAM(zfs_livelist_condense, zfs_livelist_condense_, sync_pause, INT, ZMOD_RW,
9743 "Set the livelist condense synctask to pause");
9745 ZFS_MODULE_PARAM(zfs_livelist_condense, zfs_livelist_condense_, sync_cancel, INT, ZMOD_RW,
9746 "Whether livelist condensing was canceled in the synctask");
9748 ZFS_MODULE_PARAM(zfs_livelist_condense, zfs_livelist_condense_, zthr_cancel, INT, ZMOD_RW,
9749 "Whether livelist condensing was canceled in the zthr function");
9751 ZFS_MODULE_PARAM(zfs_livelist_condense, zfs_livelist_condense_, new_alloc, INT, ZMOD_RW,
9752 "Whether extra ALLOC blkptrs were added to a livelist entry while it "
9753 "was being condensed");