4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2011, 2020 by Delphix. All rights reserved.
25 * Copyright (c) 2018, Nexenta Systems, Inc. All rights reserved.
26 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
27 * Copyright 2013 Saso Kiselkov. All rights reserved.
28 * Copyright (c) 2014 Integros [integros.com]
29 * Copyright 2016 Toomas Soome <tsoome@me.com>
30 * Copyright (c) 2016 Actifio, Inc. All rights reserved.
31 * Copyright 2018 Joyent, Inc.
32 * Copyright (c) 2017, 2019, Datto Inc. All rights reserved.
33 * Copyright 2017 Joyent, Inc.
34 * Copyright (c) 2017, Intel Corporation.
38 * SPA: Storage Pool Allocator
40 * This file contains all the routines used when modifying on-disk SPA state.
41 * This includes opening, importing, destroying, exporting a pool, and syncing a
45 #include <sys/zfs_context.h>
46 #include <sys/fm/fs/zfs.h>
47 #include <sys/spa_impl.h>
49 #include <sys/zio_checksum.h>
51 #include <sys/dmu_tx.h>
55 #include <sys/vdev_impl.h>
56 #include <sys/vdev_removal.h>
57 #include <sys/vdev_indirect_mapping.h>
58 #include <sys/vdev_indirect_births.h>
59 #include <sys/vdev_initialize.h>
60 #include <sys/vdev_rebuild.h>
61 #include <sys/vdev_trim.h>
62 #include <sys/vdev_disk.h>
63 #include <sys/vdev_draid.h>
64 #include <sys/metaslab.h>
65 #include <sys/metaslab_impl.h>
67 #include <sys/uberblock_impl.h>
70 #include <sys/bpobj.h>
71 #include <sys/dmu_traverse.h>
72 #include <sys/dmu_objset.h>
73 #include <sys/unique.h>
74 #include <sys/dsl_pool.h>
75 #include <sys/dsl_dataset.h>
76 #include <sys/dsl_dir.h>
77 #include <sys/dsl_prop.h>
78 #include <sys/dsl_synctask.h>
79 #include <sys/fs/zfs.h>
81 #include <sys/callb.h>
82 #include <sys/systeminfo.h>
83 #include <sys/spa_boot.h>
84 #include <sys/zfs_ioctl.h>
85 #include <sys/dsl_scan.h>
86 #include <sys/zfeature.h>
87 #include <sys/dsl_destroy.h>
91 #include <sys/fm/protocol.h>
92 #include <sys/fm/util.h>
93 #include <sys/callb.h>
95 #include <sys/vmsystm.h>
99 #include "zfs_comutil.h"
102 * The interval, in seconds, at which failed configuration cache file writes
105 int zfs_ccw_retry_interval = 300;
107 typedef enum zti_modes {
108 ZTI_MODE_FIXED, /* value is # of threads (min 1) */
109 ZTI_MODE_BATCH, /* cpu-intensive; value is ignored */
110 ZTI_MODE_NULL, /* don't create a taskq */
114 #define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) }
115 #define ZTI_PCT(n) { ZTI_MODE_ONLINE_PERCENT, (n), 1 }
116 #define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 }
117 #define ZTI_NULL { ZTI_MODE_NULL, 0, 0 }
119 #define ZTI_N(n) ZTI_P(n, 1)
120 #define ZTI_ONE ZTI_N(1)
122 typedef struct zio_taskq_info {
123 zti_modes_t zti_mode;
128 static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
129 "iss", "iss_h", "int", "int_h"
133 * This table defines the taskq settings for each ZFS I/O type. When
134 * initializing a pool, we use this table to create an appropriately sized
135 * taskq. Some operations are low volume and therefore have a small, static
136 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE
137 * macros. Other operations process a large amount of data; the ZTI_BATCH
138 * macro causes us to create a taskq oriented for throughput. Some operations
139 * are so high frequency and short-lived that the taskq itself can become a
140 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an
141 * additional degree of parallelism specified by the number of threads per-
142 * taskq and the number of taskqs; when dispatching an event in this case, the
143 * particular taskq is chosen at random.
145 * The different taskq priorities are to handle the different contexts (issue
146 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that
147 * need to be handled with minimum delay.
149 const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
150 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
151 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* NULL */
152 { ZTI_N(8), ZTI_NULL, ZTI_P(12, 8), ZTI_NULL }, /* READ */
153 { ZTI_BATCH, ZTI_N(5), ZTI_P(12, 8), ZTI_N(5) }, /* WRITE */
154 { ZTI_P(12, 8), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FREE */
155 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* CLAIM */
156 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* IOCTL */
157 { ZTI_N(4), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* TRIM */
160 static void spa_sync_version(void *arg, dmu_tx_t *tx);
161 static void spa_sync_props(void *arg, dmu_tx_t *tx);
162 static boolean_t spa_has_active_shared_spare(spa_t *spa);
163 static int spa_load_impl(spa_t *spa, spa_import_type_t type, char **ereport);
164 static void spa_vdev_resilver_done(spa_t *spa);
166 uint_t zio_taskq_batch_pct = 75; /* 1 thread per cpu in pset */
167 boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */
168 uint_t zio_taskq_basedc = 80; /* base duty cycle */
170 boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */
173 * Report any spa_load_verify errors found, but do not fail spa_load.
174 * This is used by zdb to analyze non-idle pools.
176 boolean_t spa_load_verify_dryrun = B_FALSE;
179 * This (illegal) pool name is used when temporarily importing a spa_t in order
180 * to get the vdev stats associated with the imported devices.
182 #define TRYIMPORT_NAME "$import"
185 * For debugging purposes: print out vdev tree during pool import.
187 int spa_load_print_vdev_tree = B_FALSE;
190 * A non-zero value for zfs_max_missing_tvds means that we allow importing
191 * pools with missing top-level vdevs. This is strictly intended for advanced
192 * pool recovery cases since missing data is almost inevitable. Pools with
193 * missing devices can only be imported read-only for safety reasons, and their
194 * fail-mode will be automatically set to "continue".
196 * With 1 missing vdev we should be able to import the pool and mount all
197 * datasets. User data that was not modified after the missing device has been
198 * added should be recoverable. This means that snapshots created prior to the
199 * addition of that device should be completely intact.
201 * With 2 missing vdevs, some datasets may fail to mount since there are
202 * dataset statistics that are stored as regular metadata. Some data might be
203 * recoverable if those vdevs were added recently.
205 * With 3 or more missing vdevs, the pool is severely damaged and MOS entries
206 * may be missing entirely. Chances of data recovery are very low. Note that
207 * there are also risks of performing an inadvertent rewind as we might be
208 * missing all the vdevs with the latest uberblocks.
210 unsigned long zfs_max_missing_tvds = 0;
213 * The parameters below are similar to zfs_max_missing_tvds but are only
214 * intended for a preliminary open of the pool with an untrusted config which
215 * might be incomplete or out-dated.
217 * We are more tolerant for pools opened from a cachefile since we could have
218 * an out-dated cachefile where a device removal was not registered.
219 * We could have set the limit arbitrarily high but in the case where devices
220 * are really missing we would want to return the proper error codes; we chose
221 * SPA_DVAS_PER_BP - 1 so that some copies of the MOS would still be available
222 * and we get a chance to retrieve the trusted config.
224 uint64_t zfs_max_missing_tvds_cachefile = SPA_DVAS_PER_BP - 1;
227 * In the case where config was assembled by scanning device paths (/dev/dsks
228 * by default) we are less tolerant since all the existing devices should have
229 * been detected and we want spa_load to return the right error codes.
231 uint64_t zfs_max_missing_tvds_scan = 0;
234 * Debugging aid that pauses spa_sync() towards the end.
236 boolean_t zfs_pause_spa_sync = B_FALSE;
239 * Variables to indicate the livelist condense zthr func should wait at certain
240 * points for the livelist to be removed - used to test condense/destroy races
242 int zfs_livelist_condense_zthr_pause = 0;
243 int zfs_livelist_condense_sync_pause = 0;
246 * Variables to track whether or not condense cancellation has been
247 * triggered in testing.
249 int zfs_livelist_condense_sync_cancel = 0;
250 int zfs_livelist_condense_zthr_cancel = 0;
253 * Variable to track whether or not extra ALLOC blkptrs were added to a
254 * livelist entry while it was being condensed (caused by the way we track
255 * remapped blkptrs in dbuf_remap_impl)
257 int zfs_livelist_condense_new_alloc = 0;
260 * ==========================================================================
261 * SPA properties routines
262 * ==========================================================================
266 * Add a (source=src, propname=propval) list to an nvlist.
269 spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
270 uint64_t intval, zprop_source_t src)
272 const char *propname = zpool_prop_to_name(prop);
275 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
276 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
279 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
281 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
283 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
284 nvlist_free(propval);
288 * Get property values from the spa configuration.
291 spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
293 vdev_t *rvd = spa->spa_root_vdev;
294 dsl_pool_t *pool = spa->spa_dsl_pool;
295 uint64_t size, alloc, cap, version;
296 const zprop_source_t src = ZPROP_SRC_NONE;
297 spa_config_dirent_t *dp;
298 metaslab_class_t *mc = spa_normal_class(spa);
300 ASSERT(MUTEX_HELD(&spa->spa_props_lock));
303 alloc = metaslab_class_get_alloc(mc);
304 alloc += metaslab_class_get_alloc(spa_special_class(spa));
305 alloc += metaslab_class_get_alloc(spa_dedup_class(spa));
307 size = metaslab_class_get_space(mc);
308 size += metaslab_class_get_space(spa_special_class(spa));
309 size += metaslab_class_get_space(spa_dedup_class(spa));
311 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
312 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
313 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
314 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
316 spa_prop_add_list(*nvp, ZPOOL_PROP_CHECKPOINT, NULL,
317 spa->spa_checkpoint_info.sci_dspace, src);
319 spa_prop_add_list(*nvp, ZPOOL_PROP_FRAGMENTATION, NULL,
320 metaslab_class_fragmentation(mc), src);
321 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL,
322 metaslab_class_expandable_space(mc), src);
323 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
324 (spa_mode(spa) == SPA_MODE_READ), src);
326 cap = (size == 0) ? 0 : (alloc * 100 / size);
327 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
329 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
330 ddt_get_pool_dedup_ratio(spa), src);
332 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
333 rvd->vdev_state, src);
335 version = spa_version(spa);
336 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION)) {
337 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL,
338 version, ZPROP_SRC_DEFAULT);
340 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL,
341 version, ZPROP_SRC_LOCAL);
343 spa_prop_add_list(*nvp, ZPOOL_PROP_LOAD_GUID,
344 NULL, spa_load_guid(spa), src);
349 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
350 * when opening pools before this version freedir will be NULL.
352 if (pool->dp_free_dir != NULL) {
353 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
354 dsl_dir_phys(pool->dp_free_dir)->dd_used_bytes,
357 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
361 if (pool->dp_leak_dir != NULL) {
362 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, NULL,
363 dsl_dir_phys(pool->dp_leak_dir)->dd_used_bytes,
366 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED,
371 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
373 if (spa->spa_comment != NULL) {
374 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
378 if (spa->spa_root != NULL)
379 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
382 if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) {
383 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL,
384 MIN(zfs_max_recordsize, SPA_MAXBLOCKSIZE), ZPROP_SRC_NONE);
386 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL,
387 SPA_OLD_MAXBLOCKSIZE, ZPROP_SRC_NONE);
390 if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_DNODE)) {
391 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXDNODESIZE, NULL,
392 DNODE_MAX_SIZE, ZPROP_SRC_NONE);
394 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXDNODESIZE, NULL,
395 DNODE_MIN_SIZE, ZPROP_SRC_NONE);
398 if ((dp = list_head(&spa->spa_config_list)) != NULL) {
399 if (dp->scd_path == NULL) {
400 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
401 "none", 0, ZPROP_SRC_LOCAL);
402 } else if (strcmp(dp->scd_path, spa_config_path) != 0) {
403 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
404 dp->scd_path, 0, ZPROP_SRC_LOCAL);
410 * Get zpool property values.
413 spa_prop_get(spa_t *spa, nvlist_t **nvp)
415 objset_t *mos = spa->spa_meta_objset;
421 err = nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP);
425 dp = spa_get_dsl(spa);
426 dsl_pool_config_enter(dp, FTAG);
427 mutex_enter(&spa->spa_props_lock);
430 * Get properties from the spa config.
432 spa_prop_get_config(spa, nvp);
434 /* If no pool property object, no more prop to get. */
435 if (mos == NULL || spa->spa_pool_props_object == 0)
439 * Get properties from the MOS pool property object.
441 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
442 (err = zap_cursor_retrieve(&zc, &za)) == 0;
443 zap_cursor_advance(&zc)) {
446 zprop_source_t src = ZPROP_SRC_DEFAULT;
449 if ((prop = zpool_name_to_prop(za.za_name)) == ZPOOL_PROP_INVAL)
452 switch (za.za_integer_length) {
454 /* integer property */
455 if (za.za_first_integer !=
456 zpool_prop_default_numeric(prop))
457 src = ZPROP_SRC_LOCAL;
459 if (prop == ZPOOL_PROP_BOOTFS) {
460 dsl_dataset_t *ds = NULL;
462 err = dsl_dataset_hold_obj(dp,
463 za.za_first_integer, FTAG, &ds);
467 strval = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN,
469 dsl_dataset_name(ds, strval);
470 dsl_dataset_rele(ds, FTAG);
473 intval = za.za_first_integer;
476 spa_prop_add_list(*nvp, prop, strval, intval, src);
479 kmem_free(strval, ZFS_MAX_DATASET_NAME_LEN);
484 /* string property */
485 strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
486 err = zap_lookup(mos, spa->spa_pool_props_object,
487 za.za_name, 1, za.za_num_integers, strval);
489 kmem_free(strval, za.za_num_integers);
492 spa_prop_add_list(*nvp, prop, strval, 0, src);
493 kmem_free(strval, za.za_num_integers);
500 zap_cursor_fini(&zc);
502 mutex_exit(&spa->spa_props_lock);
503 dsl_pool_config_exit(dp, FTAG);
504 if (err && err != ENOENT) {
514 * Validate the given pool properties nvlist and modify the list
515 * for the property values to be set.
518 spa_prop_validate(spa_t *spa, nvlist_t *props)
521 int error = 0, reset_bootfs = 0;
523 boolean_t has_feature = B_FALSE;
526 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
528 char *strval, *slash, *check, *fname;
529 const char *propname = nvpair_name(elem);
530 zpool_prop_t prop = zpool_name_to_prop(propname);
533 case ZPOOL_PROP_INVAL:
534 if (!zpool_prop_feature(propname)) {
535 error = SET_ERROR(EINVAL);
540 * Sanitize the input.
542 if (nvpair_type(elem) != DATA_TYPE_UINT64) {
543 error = SET_ERROR(EINVAL);
547 if (nvpair_value_uint64(elem, &intval) != 0) {
548 error = SET_ERROR(EINVAL);
553 error = SET_ERROR(EINVAL);
557 fname = strchr(propname, '@') + 1;
558 if (zfeature_lookup_name(fname, NULL) != 0) {
559 error = SET_ERROR(EINVAL);
563 has_feature = B_TRUE;
566 case ZPOOL_PROP_VERSION:
567 error = nvpair_value_uint64(elem, &intval);
569 (intval < spa_version(spa) ||
570 intval > SPA_VERSION_BEFORE_FEATURES ||
572 error = SET_ERROR(EINVAL);
575 case ZPOOL_PROP_DELEGATION:
576 case ZPOOL_PROP_AUTOREPLACE:
577 case ZPOOL_PROP_LISTSNAPS:
578 case ZPOOL_PROP_AUTOEXPAND:
579 case ZPOOL_PROP_AUTOTRIM:
580 error = nvpair_value_uint64(elem, &intval);
581 if (!error && intval > 1)
582 error = SET_ERROR(EINVAL);
585 case ZPOOL_PROP_MULTIHOST:
586 error = nvpair_value_uint64(elem, &intval);
587 if (!error && intval > 1)
588 error = SET_ERROR(EINVAL);
591 uint32_t hostid = zone_get_hostid(NULL);
593 spa->spa_hostid = hostid;
595 error = SET_ERROR(ENOTSUP);
600 case ZPOOL_PROP_BOOTFS:
602 * If the pool version is less than SPA_VERSION_BOOTFS,
603 * or the pool is still being created (version == 0),
604 * the bootfs property cannot be set.
606 if (spa_version(spa) < SPA_VERSION_BOOTFS) {
607 error = SET_ERROR(ENOTSUP);
612 * Make sure the vdev config is bootable
614 if (!vdev_is_bootable(spa->spa_root_vdev)) {
615 error = SET_ERROR(ENOTSUP);
621 error = nvpair_value_string(elem, &strval);
626 if (strval == NULL || strval[0] == '\0') {
627 objnum = zpool_prop_default_numeric(
632 error = dmu_objset_hold(strval, FTAG, &os);
637 if (dmu_objset_type(os) != DMU_OST_ZFS) {
638 error = SET_ERROR(ENOTSUP);
640 objnum = dmu_objset_id(os);
642 dmu_objset_rele(os, FTAG);
646 case ZPOOL_PROP_FAILUREMODE:
647 error = nvpair_value_uint64(elem, &intval);
648 if (!error && intval > ZIO_FAILURE_MODE_PANIC)
649 error = SET_ERROR(EINVAL);
652 * This is a special case which only occurs when
653 * the pool has completely failed. This allows
654 * the user to change the in-core failmode property
655 * without syncing it out to disk (I/Os might
656 * currently be blocked). We do this by returning
657 * EIO to the caller (spa_prop_set) to trick it
658 * into thinking we encountered a property validation
661 if (!error && spa_suspended(spa)) {
662 spa->spa_failmode = intval;
663 error = SET_ERROR(EIO);
667 case ZPOOL_PROP_CACHEFILE:
668 if ((error = nvpair_value_string(elem, &strval)) != 0)
671 if (strval[0] == '\0')
674 if (strcmp(strval, "none") == 0)
677 if (strval[0] != '/') {
678 error = SET_ERROR(EINVAL);
682 slash = strrchr(strval, '/');
683 ASSERT(slash != NULL);
685 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
686 strcmp(slash, "/..") == 0)
687 error = SET_ERROR(EINVAL);
690 case ZPOOL_PROP_COMMENT:
691 if ((error = nvpair_value_string(elem, &strval)) != 0)
693 for (check = strval; *check != '\0'; check++) {
694 if (!isprint(*check)) {
695 error = SET_ERROR(EINVAL);
699 if (strlen(strval) > ZPROP_MAX_COMMENT)
700 error = SET_ERROR(E2BIG);
711 (void) nvlist_remove_all(props,
712 zpool_prop_to_name(ZPOOL_PROP_DEDUPDITTO));
714 if (!error && reset_bootfs) {
715 error = nvlist_remove(props,
716 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
719 error = nvlist_add_uint64(props,
720 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
728 spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
731 spa_config_dirent_t *dp;
733 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
737 dp = kmem_alloc(sizeof (spa_config_dirent_t),
740 if (cachefile[0] == '\0')
741 dp->scd_path = spa_strdup(spa_config_path);
742 else if (strcmp(cachefile, "none") == 0)
745 dp->scd_path = spa_strdup(cachefile);
747 list_insert_head(&spa->spa_config_list, dp);
749 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
753 spa_prop_set(spa_t *spa, nvlist_t *nvp)
756 nvpair_t *elem = NULL;
757 boolean_t need_sync = B_FALSE;
759 if ((error = spa_prop_validate(spa, nvp)) != 0)
762 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
763 zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem));
765 if (prop == ZPOOL_PROP_CACHEFILE ||
766 prop == ZPOOL_PROP_ALTROOT ||
767 prop == ZPOOL_PROP_READONLY)
770 if (prop == ZPOOL_PROP_VERSION || prop == ZPOOL_PROP_INVAL) {
773 if (prop == ZPOOL_PROP_VERSION) {
774 VERIFY(nvpair_value_uint64(elem, &ver) == 0);
776 ASSERT(zpool_prop_feature(nvpair_name(elem)));
777 ver = SPA_VERSION_FEATURES;
781 /* Save time if the version is already set. */
782 if (ver == spa_version(spa))
786 * In addition to the pool directory object, we might
787 * create the pool properties object, the features for
788 * read object, the features for write object, or the
789 * feature descriptions object.
791 error = dsl_sync_task(spa->spa_name, NULL,
792 spa_sync_version, &ver,
793 6, ZFS_SPACE_CHECK_RESERVED);
804 return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props,
805 nvp, 6, ZFS_SPACE_CHECK_RESERVED));
812 * If the bootfs property value is dsobj, clear it.
815 spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
817 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
818 VERIFY(zap_remove(spa->spa_meta_objset,
819 spa->spa_pool_props_object,
820 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
827 spa_change_guid_check(void *arg, dmu_tx_t *tx)
829 uint64_t *newguid __maybe_unused = arg;
830 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
831 vdev_t *rvd = spa->spa_root_vdev;
834 if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) {
835 int error = (spa_has_checkpoint(spa)) ?
836 ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT;
837 return (SET_ERROR(error));
840 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
841 vdev_state = rvd->vdev_state;
842 spa_config_exit(spa, SCL_STATE, FTAG);
844 if (vdev_state != VDEV_STATE_HEALTHY)
845 return (SET_ERROR(ENXIO));
847 ASSERT3U(spa_guid(spa), !=, *newguid);
853 spa_change_guid_sync(void *arg, dmu_tx_t *tx)
855 uint64_t *newguid = arg;
856 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
858 vdev_t *rvd = spa->spa_root_vdev;
860 oldguid = spa_guid(spa);
862 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
863 rvd->vdev_guid = *newguid;
864 rvd->vdev_guid_sum += (*newguid - oldguid);
865 vdev_config_dirty(rvd);
866 spa_config_exit(spa, SCL_STATE, FTAG);
868 spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu",
869 (u_longlong_t)oldguid, (u_longlong_t)*newguid);
873 * Change the GUID for the pool. This is done so that we can later
874 * re-import a pool built from a clone of our own vdevs. We will modify
875 * the root vdev's guid, our own pool guid, and then mark all of our
876 * vdevs dirty. Note that we must make sure that all our vdevs are
877 * online when we do this, or else any vdevs that weren't present
878 * would be orphaned from our pool. We are also going to issue a
879 * sysevent to update any watchers.
882 spa_change_guid(spa_t *spa)
887 mutex_enter(&spa->spa_vdev_top_lock);
888 mutex_enter(&spa_namespace_lock);
889 guid = spa_generate_guid(NULL);
891 error = dsl_sync_task(spa->spa_name, spa_change_guid_check,
892 spa_change_guid_sync, &guid, 5, ZFS_SPACE_CHECK_RESERVED);
895 spa_write_cachefile(spa, B_FALSE, B_TRUE);
896 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_REGUID);
899 mutex_exit(&spa_namespace_lock);
900 mutex_exit(&spa->spa_vdev_top_lock);
906 * ==========================================================================
907 * SPA state manipulation (open/create/destroy/import/export)
908 * ==========================================================================
912 spa_error_entry_compare(const void *a, const void *b)
914 const spa_error_entry_t *sa = (const spa_error_entry_t *)a;
915 const spa_error_entry_t *sb = (const spa_error_entry_t *)b;
918 ret = memcmp(&sa->se_bookmark, &sb->se_bookmark,
919 sizeof (zbookmark_phys_t));
921 return (TREE_ISIGN(ret));
925 * Utility function which retrieves copies of the current logs and
926 * re-initializes them in the process.
929 spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
931 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
933 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
934 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
936 avl_create(&spa->spa_errlist_scrub,
937 spa_error_entry_compare, sizeof (spa_error_entry_t),
938 offsetof(spa_error_entry_t, se_avl));
939 avl_create(&spa->spa_errlist_last,
940 spa_error_entry_compare, sizeof (spa_error_entry_t),
941 offsetof(spa_error_entry_t, se_avl));
945 spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
947 const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
948 enum zti_modes mode = ztip->zti_mode;
949 uint_t value = ztip->zti_value;
950 uint_t count = ztip->zti_count;
951 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
953 boolean_t batch = B_FALSE;
955 if (mode == ZTI_MODE_NULL) {
957 tqs->stqs_taskq = NULL;
961 ASSERT3U(count, >, 0);
963 tqs->stqs_count = count;
964 tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP);
968 ASSERT3U(value, >=, 1);
969 value = MAX(value, 1);
970 flags |= TASKQ_DYNAMIC;
975 flags |= TASKQ_THREADS_CPU_PCT;
976 value = MIN(zio_taskq_batch_pct, 100);
980 panic("unrecognized mode for %s_%s taskq (%u:%u) in "
982 zio_type_name[t], zio_taskq_types[q], mode, value);
986 for (uint_t i = 0; i < count; i++) {
990 (void) snprintf(name, sizeof (name), "%s_%s",
991 zio_type_name[t], zio_taskq_types[q]);
993 if (zio_taskq_sysdc && spa->spa_proc != &p0) {
995 flags |= TASKQ_DC_BATCH;
997 tq = taskq_create_sysdc(name, value, 50, INT_MAX,
998 spa->spa_proc, zio_taskq_basedc, flags);
1000 pri_t pri = maxclsyspri;
1002 * The write issue taskq can be extremely CPU
1003 * intensive. Run it at slightly less important
1004 * priority than the other taskqs.
1006 * Under Linux and FreeBSD this means incrementing
1007 * the priority value as opposed to platforms like
1008 * illumos where it should be decremented.
1010 * On FreeBSD, if priorities divided by four (RQ_PPQ)
1011 * are equal then a difference between them is
1014 if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE) {
1015 #if defined(__linux__)
1017 #elif defined(__FreeBSD__)
1023 tq = taskq_create_proc(name, value, pri, 50,
1024 INT_MAX, spa->spa_proc, flags);
1027 tqs->stqs_taskq[i] = tq;
1032 spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
1034 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
1036 if (tqs->stqs_taskq == NULL) {
1037 ASSERT3U(tqs->stqs_count, ==, 0);
1041 for (uint_t i = 0; i < tqs->stqs_count; i++) {
1042 ASSERT3P(tqs->stqs_taskq[i], !=, NULL);
1043 taskq_destroy(tqs->stqs_taskq[i]);
1046 kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *));
1047 tqs->stqs_taskq = NULL;
1051 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
1052 * Note that a type may have multiple discrete taskqs to avoid lock contention
1053 * on the taskq itself. In that case we choose which taskq at random by using
1054 * the low bits of gethrtime().
1057 spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q,
1058 task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent)
1060 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
1063 ASSERT3P(tqs->stqs_taskq, !=, NULL);
1064 ASSERT3U(tqs->stqs_count, !=, 0);
1066 if (tqs->stqs_count == 1) {
1067 tq = tqs->stqs_taskq[0];
1069 tq = tqs->stqs_taskq[((uint64_t)gethrtime()) % tqs->stqs_count];
1072 taskq_dispatch_ent(tq, func, arg, flags, ent);
1076 * Same as spa_taskq_dispatch_ent() but block on the task until completion.
1079 spa_taskq_dispatch_sync(spa_t *spa, zio_type_t t, zio_taskq_type_t q,
1080 task_func_t *func, void *arg, uint_t flags)
1082 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
1086 ASSERT3P(tqs->stqs_taskq, !=, NULL);
1087 ASSERT3U(tqs->stqs_count, !=, 0);
1089 if (tqs->stqs_count == 1) {
1090 tq = tqs->stqs_taskq[0];
1092 tq = tqs->stqs_taskq[((uint64_t)gethrtime()) % tqs->stqs_count];
1095 id = taskq_dispatch(tq, func, arg, flags);
1097 taskq_wait_id(tq, id);
1101 spa_create_zio_taskqs(spa_t *spa)
1103 for (int t = 0; t < ZIO_TYPES; t++) {
1104 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1105 spa_taskqs_init(spa, t, q);
1111 * Disabled until spa_thread() can be adapted for Linux.
1113 #undef HAVE_SPA_THREAD
1115 #if defined(_KERNEL) && defined(HAVE_SPA_THREAD)
1117 spa_thread(void *arg)
1119 psetid_t zio_taskq_psrset_bind = PS_NONE;
1120 callb_cpr_t cprinfo;
1123 user_t *pu = PTOU(curproc);
1125 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
1128 ASSERT(curproc != &p0);
1129 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
1130 "zpool-%s", spa->spa_name);
1131 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
1133 /* bind this thread to the requested psrset */
1134 if (zio_taskq_psrset_bind != PS_NONE) {
1136 mutex_enter(&cpu_lock);
1137 mutex_enter(&pidlock);
1138 mutex_enter(&curproc->p_lock);
1140 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
1141 0, NULL, NULL) == 0) {
1142 curthread->t_bind_pset = zio_taskq_psrset_bind;
1145 "Couldn't bind process for zfs pool \"%s\" to "
1146 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
1149 mutex_exit(&curproc->p_lock);
1150 mutex_exit(&pidlock);
1151 mutex_exit(&cpu_lock);
1155 if (zio_taskq_sysdc) {
1156 sysdc_thread_enter(curthread, 100, 0);
1159 spa->spa_proc = curproc;
1160 spa->spa_did = curthread->t_did;
1162 spa_create_zio_taskqs(spa);
1164 mutex_enter(&spa->spa_proc_lock);
1165 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
1167 spa->spa_proc_state = SPA_PROC_ACTIVE;
1168 cv_broadcast(&spa->spa_proc_cv);
1170 CALLB_CPR_SAFE_BEGIN(&cprinfo);
1171 while (spa->spa_proc_state == SPA_PROC_ACTIVE)
1172 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1173 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
1175 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
1176 spa->spa_proc_state = SPA_PROC_GONE;
1177 spa->spa_proc = &p0;
1178 cv_broadcast(&spa->spa_proc_cv);
1179 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */
1181 mutex_enter(&curproc->p_lock);
1187 * Activate an uninitialized pool.
1190 spa_activate(spa_t *spa, spa_mode_t mode)
1192 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
1194 spa->spa_state = POOL_STATE_ACTIVE;
1195 spa->spa_mode = mode;
1197 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
1198 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
1199 spa->spa_special_class = metaslab_class_create(spa, zfs_metaslab_ops);
1200 spa->spa_dedup_class = metaslab_class_create(spa, zfs_metaslab_ops);
1202 /* Try to create a covering process */
1203 mutex_enter(&spa->spa_proc_lock);
1204 ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
1205 ASSERT(spa->spa_proc == &p0);
1208 #ifdef HAVE_SPA_THREAD
1209 /* Only create a process if we're going to be around a while. */
1210 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
1211 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
1213 spa->spa_proc_state = SPA_PROC_CREATED;
1214 while (spa->spa_proc_state == SPA_PROC_CREATED) {
1215 cv_wait(&spa->spa_proc_cv,
1216 &spa->spa_proc_lock);
1218 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1219 ASSERT(spa->spa_proc != &p0);
1220 ASSERT(spa->spa_did != 0);
1224 "Couldn't create process for zfs pool \"%s\"\n",
1229 #endif /* HAVE_SPA_THREAD */
1230 mutex_exit(&spa->spa_proc_lock);
1232 /* If we didn't create a process, we need to create our taskqs. */
1233 if (spa->spa_proc == &p0) {
1234 spa_create_zio_taskqs(spa);
1237 for (size_t i = 0; i < TXG_SIZE; i++) {
1238 spa->spa_txg_zio[i] = zio_root(spa, NULL, NULL,
1242 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
1243 offsetof(vdev_t, vdev_config_dirty_node));
1244 list_create(&spa->spa_evicting_os_list, sizeof (objset_t),
1245 offsetof(objset_t, os_evicting_node));
1246 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
1247 offsetof(vdev_t, vdev_state_dirty_node));
1249 txg_list_create(&spa->spa_vdev_txg_list, spa,
1250 offsetof(struct vdev, vdev_txg_node));
1252 avl_create(&spa->spa_errlist_scrub,
1253 spa_error_entry_compare, sizeof (spa_error_entry_t),
1254 offsetof(spa_error_entry_t, se_avl));
1255 avl_create(&spa->spa_errlist_last,
1256 spa_error_entry_compare, sizeof (spa_error_entry_t),
1257 offsetof(spa_error_entry_t, se_avl));
1259 spa_keystore_init(&spa->spa_keystore);
1262 * This taskq is used to perform zvol-minor-related tasks
1263 * asynchronously. This has several advantages, including easy
1264 * resolution of various deadlocks.
1266 * The taskq must be single threaded to ensure tasks are always
1267 * processed in the order in which they were dispatched.
1269 * A taskq per pool allows one to keep the pools independent.
1270 * This way if one pool is suspended, it will not impact another.
1272 * The preferred location to dispatch a zvol minor task is a sync
1273 * task. In this context, there is easy access to the spa_t and minimal
1274 * error handling is required because the sync task must succeed.
1276 spa->spa_zvol_taskq = taskq_create("z_zvol", 1, defclsyspri,
1280 * Taskq dedicated to prefetcher threads: this is used to prevent the
1281 * pool traverse code from monopolizing the global (and limited)
1282 * system_taskq by inappropriately scheduling long running tasks on it.
1284 spa->spa_prefetch_taskq = taskq_create("z_prefetch", boot_ncpus,
1285 defclsyspri, 1, INT_MAX, TASKQ_DYNAMIC);
1288 * The taskq to upgrade datasets in this pool. Currently used by
1289 * feature SPA_FEATURE_USEROBJ_ACCOUNTING/SPA_FEATURE_PROJECT_QUOTA.
1291 spa->spa_upgrade_taskq = taskq_create("z_upgrade", boot_ncpus,
1292 defclsyspri, 1, INT_MAX, TASKQ_DYNAMIC);
1296 * Opposite of spa_activate().
1299 spa_deactivate(spa_t *spa)
1301 ASSERT(spa->spa_sync_on == B_FALSE);
1302 ASSERT(spa->spa_dsl_pool == NULL);
1303 ASSERT(spa->spa_root_vdev == NULL);
1304 ASSERT(spa->spa_async_zio_root == NULL);
1305 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
1307 spa_evicting_os_wait(spa);
1309 if (spa->spa_zvol_taskq) {
1310 taskq_destroy(spa->spa_zvol_taskq);
1311 spa->spa_zvol_taskq = NULL;
1314 if (spa->spa_prefetch_taskq) {
1315 taskq_destroy(spa->spa_prefetch_taskq);
1316 spa->spa_prefetch_taskq = NULL;
1319 if (spa->spa_upgrade_taskq) {
1320 taskq_destroy(spa->spa_upgrade_taskq);
1321 spa->spa_upgrade_taskq = NULL;
1324 txg_list_destroy(&spa->spa_vdev_txg_list);
1326 list_destroy(&spa->spa_config_dirty_list);
1327 list_destroy(&spa->spa_evicting_os_list);
1328 list_destroy(&spa->spa_state_dirty_list);
1330 taskq_cancel_id(system_delay_taskq, spa->spa_deadman_tqid);
1332 for (int t = 0; t < ZIO_TYPES; t++) {
1333 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1334 spa_taskqs_fini(spa, t, q);
1338 for (size_t i = 0; i < TXG_SIZE; i++) {
1339 ASSERT3P(spa->spa_txg_zio[i], !=, NULL);
1340 VERIFY0(zio_wait(spa->spa_txg_zio[i]));
1341 spa->spa_txg_zio[i] = NULL;
1344 metaslab_class_destroy(spa->spa_normal_class);
1345 spa->spa_normal_class = NULL;
1347 metaslab_class_destroy(spa->spa_log_class);
1348 spa->spa_log_class = NULL;
1350 metaslab_class_destroy(spa->spa_special_class);
1351 spa->spa_special_class = NULL;
1353 metaslab_class_destroy(spa->spa_dedup_class);
1354 spa->spa_dedup_class = NULL;
1357 * If this was part of an import or the open otherwise failed, we may
1358 * still have errors left in the queues. Empty them just in case.
1360 spa_errlog_drain(spa);
1361 avl_destroy(&spa->spa_errlist_scrub);
1362 avl_destroy(&spa->spa_errlist_last);
1364 spa_keystore_fini(&spa->spa_keystore);
1366 spa->spa_state = POOL_STATE_UNINITIALIZED;
1368 mutex_enter(&spa->spa_proc_lock);
1369 if (spa->spa_proc_state != SPA_PROC_NONE) {
1370 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1371 spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1372 cv_broadcast(&spa->spa_proc_cv);
1373 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1374 ASSERT(spa->spa_proc != &p0);
1375 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1377 ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1378 spa->spa_proc_state = SPA_PROC_NONE;
1380 ASSERT(spa->spa_proc == &p0);
1381 mutex_exit(&spa->spa_proc_lock);
1384 * We want to make sure spa_thread() has actually exited the ZFS
1385 * module, so that the module can't be unloaded out from underneath
1388 if (spa->spa_did != 0) {
1389 thread_join(spa->spa_did);
1395 * Verify a pool configuration, and construct the vdev tree appropriately. This
1396 * will create all the necessary vdevs in the appropriate layout, with each vdev
1397 * in the CLOSED state. This will prep the pool before open/creation/import.
1398 * All vdev validation is done by the vdev_alloc() routine.
1401 spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1402 uint_t id, int atype)
1408 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1411 if ((*vdp)->vdev_ops->vdev_op_leaf)
1414 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1417 if (error == ENOENT)
1423 return (SET_ERROR(EINVAL));
1426 for (int c = 0; c < children; c++) {
1428 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1436 ASSERT(*vdp != NULL);
1442 spa_should_flush_logs_on_unload(spa_t *spa)
1444 if (!spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP))
1447 if (!spa_writeable(spa))
1450 if (!spa->spa_sync_on)
1453 if (spa_state(spa) != POOL_STATE_EXPORTED)
1456 if (zfs_keep_log_spacemaps_at_export)
1463 * Opens a transaction that will set the flag that will instruct
1464 * spa_sync to attempt to flush all the metaslabs for that txg.
1467 spa_unload_log_sm_flush_all(spa_t *spa)
1469 dmu_tx_t *tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
1470 VERIFY0(dmu_tx_assign(tx, TXG_WAIT));
1472 ASSERT3U(spa->spa_log_flushall_txg, ==, 0);
1473 spa->spa_log_flushall_txg = dmu_tx_get_txg(tx);
1476 txg_wait_synced(spa_get_dsl(spa), spa->spa_log_flushall_txg);
1480 spa_unload_log_sm_metadata(spa_t *spa)
1482 void *cookie = NULL;
1484 while ((sls = avl_destroy_nodes(&spa->spa_sm_logs_by_txg,
1485 &cookie)) != NULL) {
1486 VERIFY0(sls->sls_mscount);
1487 kmem_free(sls, sizeof (spa_log_sm_t));
1490 for (log_summary_entry_t *e = list_head(&spa->spa_log_summary);
1491 e != NULL; e = list_head(&spa->spa_log_summary)) {
1492 VERIFY0(e->lse_mscount);
1493 list_remove(&spa->spa_log_summary, e);
1494 kmem_free(e, sizeof (log_summary_entry_t));
1497 spa->spa_unflushed_stats.sus_nblocks = 0;
1498 spa->spa_unflushed_stats.sus_memused = 0;
1499 spa->spa_unflushed_stats.sus_blocklimit = 0;
1503 spa_destroy_aux_threads(spa_t *spa)
1505 if (spa->spa_condense_zthr != NULL) {
1506 zthr_destroy(spa->spa_condense_zthr);
1507 spa->spa_condense_zthr = NULL;
1509 if (spa->spa_checkpoint_discard_zthr != NULL) {
1510 zthr_destroy(spa->spa_checkpoint_discard_zthr);
1511 spa->spa_checkpoint_discard_zthr = NULL;
1513 if (spa->spa_livelist_delete_zthr != NULL) {
1514 zthr_destroy(spa->spa_livelist_delete_zthr);
1515 spa->spa_livelist_delete_zthr = NULL;
1517 if (spa->spa_livelist_condense_zthr != NULL) {
1518 zthr_destroy(spa->spa_livelist_condense_zthr);
1519 spa->spa_livelist_condense_zthr = NULL;
1524 * Opposite of spa_load().
1527 spa_unload(spa_t *spa)
1529 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1530 ASSERT(spa_state(spa) != POOL_STATE_UNINITIALIZED);
1532 spa_import_progress_remove(spa_guid(spa));
1533 spa_load_note(spa, "UNLOADING");
1535 spa_wake_waiters(spa);
1538 * If the log space map feature is enabled and the pool is getting
1539 * exported (but not destroyed), we want to spend some time flushing
1540 * as many metaslabs as we can in an attempt to destroy log space
1541 * maps and save import time.
1543 if (spa_should_flush_logs_on_unload(spa))
1544 spa_unload_log_sm_flush_all(spa);
1549 spa_async_suspend(spa);
1551 if (spa->spa_root_vdev) {
1552 vdev_t *root_vdev = spa->spa_root_vdev;
1553 vdev_initialize_stop_all(root_vdev, VDEV_INITIALIZE_ACTIVE);
1554 vdev_trim_stop_all(root_vdev, VDEV_TRIM_ACTIVE);
1555 vdev_autotrim_stop_all(spa);
1556 vdev_rebuild_stop_all(spa);
1562 if (spa->spa_sync_on) {
1563 txg_sync_stop(spa->spa_dsl_pool);
1564 spa->spa_sync_on = B_FALSE;
1568 * This ensures that there is no async metaslab prefetching
1569 * while we attempt to unload the spa.
1571 if (spa->spa_root_vdev != NULL) {
1572 for (int c = 0; c < spa->spa_root_vdev->vdev_children; c++) {
1573 vdev_t *vc = spa->spa_root_vdev->vdev_child[c];
1574 if (vc->vdev_mg != NULL)
1575 taskq_wait(vc->vdev_mg->mg_taskq);
1579 if (spa->spa_mmp.mmp_thread)
1580 mmp_thread_stop(spa);
1583 * Wait for any outstanding async I/O to complete.
1585 if (spa->spa_async_zio_root != NULL) {
1586 for (int i = 0; i < max_ncpus; i++)
1587 (void) zio_wait(spa->spa_async_zio_root[i]);
1588 kmem_free(spa->spa_async_zio_root, max_ncpus * sizeof (void *));
1589 spa->spa_async_zio_root = NULL;
1592 if (spa->spa_vdev_removal != NULL) {
1593 spa_vdev_removal_destroy(spa->spa_vdev_removal);
1594 spa->spa_vdev_removal = NULL;
1597 spa_destroy_aux_threads(spa);
1599 spa_condense_fini(spa);
1601 bpobj_close(&spa->spa_deferred_bpobj);
1603 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
1608 if (spa->spa_root_vdev)
1609 vdev_free(spa->spa_root_vdev);
1610 ASSERT(spa->spa_root_vdev == NULL);
1613 * Close the dsl pool.
1615 if (spa->spa_dsl_pool) {
1616 dsl_pool_close(spa->spa_dsl_pool);
1617 spa->spa_dsl_pool = NULL;
1618 spa->spa_meta_objset = NULL;
1622 spa_unload_log_sm_metadata(spa);
1625 * Drop and purge level 2 cache
1627 spa_l2cache_drop(spa);
1629 for (int i = 0; i < spa->spa_spares.sav_count; i++)
1630 vdev_free(spa->spa_spares.sav_vdevs[i]);
1631 if (spa->spa_spares.sav_vdevs) {
1632 kmem_free(spa->spa_spares.sav_vdevs,
1633 spa->spa_spares.sav_count * sizeof (void *));
1634 spa->spa_spares.sav_vdevs = NULL;
1636 if (spa->spa_spares.sav_config) {
1637 nvlist_free(spa->spa_spares.sav_config);
1638 spa->spa_spares.sav_config = NULL;
1640 spa->spa_spares.sav_count = 0;
1642 for (int i = 0; i < spa->spa_l2cache.sav_count; i++) {
1643 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1644 vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1646 if (spa->spa_l2cache.sav_vdevs) {
1647 kmem_free(spa->spa_l2cache.sav_vdevs,
1648 spa->spa_l2cache.sav_count * sizeof (void *));
1649 spa->spa_l2cache.sav_vdevs = NULL;
1651 if (spa->spa_l2cache.sav_config) {
1652 nvlist_free(spa->spa_l2cache.sav_config);
1653 spa->spa_l2cache.sav_config = NULL;
1655 spa->spa_l2cache.sav_count = 0;
1657 spa->spa_async_suspended = 0;
1659 spa->spa_indirect_vdevs_loaded = B_FALSE;
1661 if (spa->spa_comment != NULL) {
1662 spa_strfree(spa->spa_comment);
1663 spa->spa_comment = NULL;
1666 spa_config_exit(spa, SCL_ALL, spa);
1670 * Load (or re-load) the current list of vdevs describing the active spares for
1671 * this pool. When this is called, we have some form of basic information in
1672 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1673 * then re-generate a more complete list including status information.
1676 spa_load_spares(spa_t *spa)
1685 * zdb opens both the current state of the pool and the
1686 * checkpointed state (if present), with a different spa_t.
1688 * As spare vdevs are shared among open pools, we skip loading
1689 * them when we load the checkpointed state of the pool.
1691 if (!spa_writeable(spa))
1695 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1698 * First, close and free any existing spare vdevs.
1700 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1701 vd = spa->spa_spares.sav_vdevs[i];
1703 /* Undo the call to spa_activate() below */
1704 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1705 B_FALSE)) != NULL && tvd->vdev_isspare)
1706 spa_spare_remove(tvd);
1711 if (spa->spa_spares.sav_vdevs)
1712 kmem_free(spa->spa_spares.sav_vdevs,
1713 spa->spa_spares.sav_count * sizeof (void *));
1715 if (spa->spa_spares.sav_config == NULL)
1718 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1719 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1721 spa->spa_spares.sav_count = (int)nspares;
1722 spa->spa_spares.sav_vdevs = NULL;
1728 * Construct the array of vdevs, opening them to get status in the
1729 * process. For each spare, there is potentially two different vdev_t
1730 * structures associated with it: one in the list of spares (used only
1731 * for basic validation purposes) and one in the active vdev
1732 * configuration (if it's spared in). During this phase we open and
1733 * validate each vdev on the spare list. If the vdev also exists in the
1734 * active configuration, then we also mark this vdev as an active spare.
1736 spa->spa_spares.sav_vdevs = kmem_zalloc(nspares * sizeof (void *),
1738 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1739 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1740 VDEV_ALLOC_SPARE) == 0);
1743 spa->spa_spares.sav_vdevs[i] = vd;
1745 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1746 B_FALSE)) != NULL) {
1747 if (!tvd->vdev_isspare)
1751 * We only mark the spare active if we were successfully
1752 * able to load the vdev. Otherwise, importing a pool
1753 * with a bad active spare would result in strange
1754 * behavior, because multiple pool would think the spare
1755 * is actively in use.
1757 * There is a vulnerability here to an equally bizarre
1758 * circumstance, where a dead active spare is later
1759 * brought back to life (onlined or otherwise). Given
1760 * the rarity of this scenario, and the extra complexity
1761 * it adds, we ignore the possibility.
1763 if (!vdev_is_dead(tvd))
1764 spa_spare_activate(tvd);
1768 vd->vdev_aux = &spa->spa_spares;
1770 if (vdev_open(vd) != 0)
1773 if (vdev_validate_aux(vd) == 0)
1778 * Recompute the stashed list of spares, with status information
1781 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1782 DATA_TYPE_NVLIST_ARRAY) == 0);
1784 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1786 for (i = 0; i < spa->spa_spares.sav_count; i++)
1787 spares[i] = vdev_config_generate(spa,
1788 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1789 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1790 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1791 for (i = 0; i < spa->spa_spares.sav_count; i++)
1792 nvlist_free(spares[i]);
1793 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1797 * Load (or re-load) the current list of vdevs describing the active l2cache for
1798 * this pool. When this is called, we have some form of basic information in
1799 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1800 * then re-generate a more complete list including status information.
1801 * Devices which are already active have their details maintained, and are
1805 spa_load_l2cache(spa_t *spa)
1807 nvlist_t **l2cache = NULL;
1809 int i, j, oldnvdevs;
1811 vdev_t *vd, **oldvdevs, **newvdevs;
1812 spa_aux_vdev_t *sav = &spa->spa_l2cache;
1816 * zdb opens both the current state of the pool and the
1817 * checkpointed state (if present), with a different spa_t.
1819 * As L2 caches are part of the ARC which is shared among open
1820 * pools, we skip loading them when we load the checkpointed
1821 * state of the pool.
1823 if (!spa_writeable(spa))
1827 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1829 oldvdevs = sav->sav_vdevs;
1830 oldnvdevs = sav->sav_count;
1831 sav->sav_vdevs = NULL;
1834 if (sav->sav_config == NULL) {
1840 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1841 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1842 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1845 * Process new nvlist of vdevs.
1847 for (i = 0; i < nl2cache; i++) {
1848 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1852 for (j = 0; j < oldnvdevs; j++) {
1854 if (vd != NULL && guid == vd->vdev_guid) {
1856 * Retain previous vdev for add/remove ops.
1864 if (newvdevs[i] == NULL) {
1868 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1869 VDEV_ALLOC_L2CACHE) == 0);
1874 * Commit this vdev as an l2cache device,
1875 * even if it fails to open.
1877 spa_l2cache_add(vd);
1882 spa_l2cache_activate(vd);
1884 if (vdev_open(vd) != 0)
1887 (void) vdev_validate_aux(vd);
1889 if (!vdev_is_dead(vd))
1890 l2arc_add_vdev(spa, vd);
1893 * Upon cache device addition to a pool or pool
1894 * creation with a cache device or if the header
1895 * of the device is invalid we issue an async
1896 * TRIM command for the whole device which will
1897 * execute if l2arc_trim_ahead > 0.
1899 spa_async_request(spa, SPA_ASYNC_L2CACHE_TRIM);
1903 sav->sav_vdevs = newvdevs;
1904 sav->sav_count = (int)nl2cache;
1907 * Recompute the stashed list of l2cache devices, with status
1908 * information this time.
1910 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1911 DATA_TYPE_NVLIST_ARRAY) == 0);
1913 if (sav->sav_count > 0)
1914 l2cache = kmem_alloc(sav->sav_count * sizeof (void *),
1916 for (i = 0; i < sav->sav_count; i++)
1917 l2cache[i] = vdev_config_generate(spa,
1918 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1919 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1920 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1924 * Purge vdevs that were dropped
1926 for (i = 0; i < oldnvdevs; i++) {
1931 ASSERT(vd->vdev_isl2cache);
1933 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1934 pool != 0ULL && l2arc_vdev_present(vd))
1935 l2arc_remove_vdev(vd);
1936 vdev_clear_stats(vd);
1942 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1944 for (i = 0; i < sav->sav_count; i++)
1945 nvlist_free(l2cache[i]);
1947 kmem_free(l2cache, sav->sav_count * sizeof (void *));
1951 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1954 char *packed = NULL;
1959 error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db);
1963 nvsize = *(uint64_t *)db->db_data;
1964 dmu_buf_rele(db, FTAG);
1966 packed = vmem_alloc(nvsize, KM_SLEEP);
1967 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1970 error = nvlist_unpack(packed, nvsize, value, 0);
1971 vmem_free(packed, nvsize);
1977 * Concrete top-level vdevs that are not missing and are not logs. At every
1978 * spa_sync we write new uberblocks to at least SPA_SYNC_MIN_VDEVS core tvds.
1981 spa_healthy_core_tvds(spa_t *spa)
1983 vdev_t *rvd = spa->spa_root_vdev;
1986 for (uint64_t i = 0; i < rvd->vdev_children; i++) {
1987 vdev_t *vd = rvd->vdev_child[i];
1990 if (vdev_is_concrete(vd) && !vdev_is_dead(vd))
1998 * Checks to see if the given vdev could not be opened, in which case we post a
1999 * sysevent to notify the autoreplace code that the device has been removed.
2002 spa_check_removed(vdev_t *vd)
2004 for (uint64_t c = 0; c < vd->vdev_children; c++)
2005 spa_check_removed(vd->vdev_child[c]);
2007 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) &&
2008 vdev_is_concrete(vd)) {
2009 zfs_post_autoreplace(vd->vdev_spa, vd);
2010 spa_event_notify(vd->vdev_spa, vd, NULL, ESC_ZFS_VDEV_CHECK);
2015 spa_check_for_missing_logs(spa_t *spa)
2017 vdev_t *rvd = spa->spa_root_vdev;
2020 * If we're doing a normal import, then build up any additional
2021 * diagnostic information about missing log devices.
2022 * We'll pass this up to the user for further processing.
2024 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
2025 nvlist_t **child, *nv;
2028 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t *),
2030 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2032 for (uint64_t c = 0; c < rvd->vdev_children; c++) {
2033 vdev_t *tvd = rvd->vdev_child[c];
2036 * We consider a device as missing only if it failed
2037 * to open (i.e. offline or faulted is not considered
2040 if (tvd->vdev_islog &&
2041 tvd->vdev_state == VDEV_STATE_CANT_OPEN) {
2042 child[idx++] = vdev_config_generate(spa, tvd,
2043 B_FALSE, VDEV_CONFIG_MISSING);
2048 fnvlist_add_nvlist_array(nv,
2049 ZPOOL_CONFIG_CHILDREN, child, idx);
2050 fnvlist_add_nvlist(spa->spa_load_info,
2051 ZPOOL_CONFIG_MISSING_DEVICES, nv);
2053 for (uint64_t i = 0; i < idx; i++)
2054 nvlist_free(child[i]);
2057 kmem_free(child, rvd->vdev_children * sizeof (char **));
2060 spa_load_failed(spa, "some log devices are missing");
2061 vdev_dbgmsg_print_tree(rvd, 2);
2062 return (SET_ERROR(ENXIO));
2065 for (uint64_t c = 0; c < rvd->vdev_children; c++) {
2066 vdev_t *tvd = rvd->vdev_child[c];
2068 if (tvd->vdev_islog &&
2069 tvd->vdev_state == VDEV_STATE_CANT_OPEN) {
2070 spa_set_log_state(spa, SPA_LOG_CLEAR);
2071 spa_load_note(spa, "some log devices are "
2072 "missing, ZIL is dropped.");
2073 vdev_dbgmsg_print_tree(rvd, 2);
2083 * Check for missing log devices
2086 spa_check_logs(spa_t *spa)
2088 boolean_t rv = B_FALSE;
2089 dsl_pool_t *dp = spa_get_dsl(spa);
2091 switch (spa->spa_log_state) {
2094 case SPA_LOG_MISSING:
2095 /* need to recheck in case slog has been restored */
2096 case SPA_LOG_UNKNOWN:
2097 rv = (dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
2098 zil_check_log_chain, NULL, DS_FIND_CHILDREN) != 0);
2100 spa_set_log_state(spa, SPA_LOG_MISSING);
2107 spa_passivate_log(spa_t *spa)
2109 vdev_t *rvd = spa->spa_root_vdev;
2110 boolean_t slog_found = B_FALSE;
2112 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
2114 if (!spa_has_slogs(spa))
2117 for (int c = 0; c < rvd->vdev_children; c++) {
2118 vdev_t *tvd = rvd->vdev_child[c];
2119 metaslab_group_t *mg = tvd->vdev_mg;
2121 if (tvd->vdev_islog) {
2122 metaslab_group_passivate(mg);
2123 slog_found = B_TRUE;
2127 return (slog_found);
2131 spa_activate_log(spa_t *spa)
2133 vdev_t *rvd = spa->spa_root_vdev;
2135 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
2137 for (int c = 0; c < rvd->vdev_children; c++) {
2138 vdev_t *tvd = rvd->vdev_child[c];
2139 metaslab_group_t *mg = tvd->vdev_mg;
2141 if (tvd->vdev_islog)
2142 metaslab_group_activate(mg);
2147 spa_reset_logs(spa_t *spa)
2151 error = dmu_objset_find(spa_name(spa), zil_reset,
2152 NULL, DS_FIND_CHILDREN);
2155 * We successfully offlined the log device, sync out the
2156 * current txg so that the "stubby" block can be removed
2159 txg_wait_synced(spa->spa_dsl_pool, 0);
2165 spa_aux_check_removed(spa_aux_vdev_t *sav)
2167 for (int i = 0; i < sav->sav_count; i++)
2168 spa_check_removed(sav->sav_vdevs[i]);
2172 spa_claim_notify(zio_t *zio)
2174 spa_t *spa = zio->io_spa;
2179 mutex_enter(&spa->spa_props_lock); /* any mutex will do */
2180 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
2181 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
2182 mutex_exit(&spa->spa_props_lock);
2185 typedef struct spa_load_error {
2186 uint64_t sle_meta_count;
2187 uint64_t sle_data_count;
2191 spa_load_verify_done(zio_t *zio)
2193 blkptr_t *bp = zio->io_bp;
2194 spa_load_error_t *sle = zio->io_private;
2195 dmu_object_type_t type = BP_GET_TYPE(bp);
2196 int error = zio->io_error;
2197 spa_t *spa = zio->io_spa;
2199 abd_free(zio->io_abd);
2201 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
2202 type != DMU_OT_INTENT_LOG)
2203 atomic_inc_64(&sle->sle_meta_count);
2205 atomic_inc_64(&sle->sle_data_count);
2208 mutex_enter(&spa->spa_scrub_lock);
2209 spa->spa_load_verify_bytes -= BP_GET_PSIZE(bp);
2210 cv_broadcast(&spa->spa_scrub_io_cv);
2211 mutex_exit(&spa->spa_scrub_lock);
2215 * Maximum number of inflight bytes is the log2 fraction of the arc size.
2216 * By default, we set it to 1/16th of the arc.
2218 int spa_load_verify_shift = 4;
2219 int spa_load_verify_metadata = B_TRUE;
2220 int spa_load_verify_data = B_TRUE;
2224 spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
2225 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
2227 if (zb->zb_level == ZB_DNODE_LEVEL || BP_IS_HOLE(bp) ||
2228 BP_IS_EMBEDDED(bp) || BP_IS_REDACTED(bp))
2231 * Note: normally this routine will not be called if
2232 * spa_load_verify_metadata is not set. However, it may be useful
2233 * to manually set the flag after the traversal has begun.
2235 if (!spa_load_verify_metadata)
2237 if (!BP_IS_METADATA(bp) && !spa_load_verify_data)
2240 uint64_t maxinflight_bytes =
2241 arc_target_bytes() >> spa_load_verify_shift;
2243 size_t size = BP_GET_PSIZE(bp);
2245 mutex_enter(&spa->spa_scrub_lock);
2246 while (spa->spa_load_verify_bytes >= maxinflight_bytes)
2247 cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
2248 spa->spa_load_verify_bytes += size;
2249 mutex_exit(&spa->spa_scrub_lock);
2251 zio_nowait(zio_read(rio, spa, bp, abd_alloc_for_io(size, B_FALSE), size,
2252 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
2253 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
2254 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
2260 verify_dataset_name_len(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg)
2262 if (dsl_dataset_namelen(ds) >= ZFS_MAX_DATASET_NAME_LEN)
2263 return (SET_ERROR(ENAMETOOLONG));
2269 spa_load_verify(spa_t *spa)
2272 spa_load_error_t sle = { 0 };
2273 zpool_load_policy_t policy;
2274 boolean_t verify_ok = B_FALSE;
2277 zpool_get_load_policy(spa->spa_config, &policy);
2279 if (policy.zlp_rewind & ZPOOL_NEVER_REWIND)
2282 dsl_pool_config_enter(spa->spa_dsl_pool, FTAG);
2283 error = dmu_objset_find_dp(spa->spa_dsl_pool,
2284 spa->spa_dsl_pool->dp_root_dir_obj, verify_dataset_name_len, NULL,
2286 dsl_pool_config_exit(spa->spa_dsl_pool, FTAG);
2290 rio = zio_root(spa, NULL, &sle,
2291 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
2293 if (spa_load_verify_metadata) {
2294 if (spa->spa_extreme_rewind) {
2295 spa_load_note(spa, "performing a complete scan of the "
2296 "pool since extreme rewind is on. This may take "
2297 "a very long time.\n (spa_load_verify_data=%u, "
2298 "spa_load_verify_metadata=%u)",
2299 spa_load_verify_data, spa_load_verify_metadata);
2302 error = traverse_pool(spa, spa->spa_verify_min_txg,
2303 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA |
2304 TRAVERSE_NO_DECRYPT, spa_load_verify_cb, rio);
2307 (void) zio_wait(rio);
2308 ASSERT0(spa->spa_load_verify_bytes);
2310 spa->spa_load_meta_errors = sle.sle_meta_count;
2311 spa->spa_load_data_errors = sle.sle_data_count;
2313 if (sle.sle_meta_count != 0 || sle.sle_data_count != 0) {
2314 spa_load_note(spa, "spa_load_verify found %llu metadata errors "
2315 "and %llu data errors", (u_longlong_t)sle.sle_meta_count,
2316 (u_longlong_t)sle.sle_data_count);
2319 if (spa_load_verify_dryrun ||
2320 (!error && sle.sle_meta_count <= policy.zlp_maxmeta &&
2321 sle.sle_data_count <= policy.zlp_maxdata)) {
2325 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
2326 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
2328 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
2329 VERIFY(nvlist_add_uint64(spa->spa_load_info,
2330 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
2331 VERIFY(nvlist_add_int64(spa->spa_load_info,
2332 ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
2333 VERIFY(nvlist_add_uint64(spa->spa_load_info,
2334 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
2336 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
2339 if (spa_load_verify_dryrun)
2343 if (error != ENXIO && error != EIO)
2344 error = SET_ERROR(EIO);
2348 return (verify_ok ? 0 : EIO);
2352 * Find a value in the pool props object.
2355 spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
2357 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
2358 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
2362 * Find a value in the pool directory object.
2365 spa_dir_prop(spa_t *spa, const char *name, uint64_t *val, boolean_t log_enoent)
2367 int error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
2368 name, sizeof (uint64_t), 1, val);
2370 if (error != 0 && (error != ENOENT || log_enoent)) {
2371 spa_load_failed(spa, "couldn't get '%s' value in MOS directory "
2372 "[error=%d]", name, error);
2379 spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
2381 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
2382 return (SET_ERROR(err));
2386 spa_livelist_delete_check(spa_t *spa)
2388 return (spa->spa_livelists_to_delete != 0);
2393 spa_livelist_delete_cb_check(void *arg, zthr_t *z)
2396 return (spa_livelist_delete_check(spa));
2400 delete_blkptr_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
2403 zio_free(spa, tx->tx_txg, bp);
2404 dsl_dir_diduse_space(tx->tx_pool->dp_free_dir, DD_USED_HEAD,
2405 -bp_get_dsize_sync(spa, bp),
2406 -BP_GET_PSIZE(bp), -BP_GET_UCSIZE(bp), tx);
2411 dsl_get_next_livelist_obj(objset_t *os, uint64_t zap_obj, uint64_t *llp)
2416 zap_cursor_init(&zc, os, zap_obj);
2417 err = zap_cursor_retrieve(&zc, &za);
2418 zap_cursor_fini(&zc);
2420 *llp = za.za_first_integer;
2425 * Components of livelist deletion that must be performed in syncing
2426 * context: freeing block pointers and updating the pool-wide data
2427 * structures to indicate how much work is left to do
2429 typedef struct sublist_delete_arg {
2434 } sublist_delete_arg_t;
2437 sublist_delete_sync(void *arg, dmu_tx_t *tx)
2439 sublist_delete_arg_t *sda = arg;
2440 spa_t *spa = sda->spa;
2441 dsl_deadlist_t *ll = sda->ll;
2442 uint64_t key = sda->key;
2443 bplist_t *to_free = sda->to_free;
2445 bplist_iterate(to_free, delete_blkptr_cb, spa, tx);
2446 dsl_deadlist_remove_entry(ll, key, tx);
2449 typedef struct livelist_delete_arg {
2453 } livelist_delete_arg_t;
2456 livelist_delete_sync(void *arg, dmu_tx_t *tx)
2458 livelist_delete_arg_t *lda = arg;
2459 spa_t *spa = lda->spa;
2460 uint64_t ll_obj = lda->ll_obj;
2461 uint64_t zap_obj = lda->zap_obj;
2462 objset_t *mos = spa->spa_meta_objset;
2465 /* free the livelist and decrement the feature count */
2466 VERIFY0(zap_remove_int(mos, zap_obj, ll_obj, tx));
2467 dsl_deadlist_free(mos, ll_obj, tx);
2468 spa_feature_decr(spa, SPA_FEATURE_LIVELIST, tx);
2469 VERIFY0(zap_count(mos, zap_obj, &count));
2471 /* no more livelists to delete */
2472 VERIFY0(zap_remove(mos, DMU_POOL_DIRECTORY_OBJECT,
2473 DMU_POOL_DELETED_CLONES, tx));
2474 VERIFY0(zap_destroy(mos, zap_obj, tx));
2475 spa->spa_livelists_to_delete = 0;
2476 spa_notify_waiters(spa);
2481 * Load in the value for the livelist to be removed and open it. Then,
2482 * load its first sublist and determine which block pointers should actually
2483 * be freed. Then, call a synctask which performs the actual frees and updates
2484 * the pool-wide livelist data.
2488 spa_livelist_delete_cb(void *arg, zthr_t *z)
2491 uint64_t ll_obj = 0, count;
2492 objset_t *mos = spa->spa_meta_objset;
2493 uint64_t zap_obj = spa->spa_livelists_to_delete;
2495 * Determine the next livelist to delete. This function should only
2496 * be called if there is at least one deleted clone.
2498 VERIFY0(dsl_get_next_livelist_obj(mos, zap_obj, &ll_obj));
2499 VERIFY0(zap_count(mos, ll_obj, &count));
2502 dsl_deadlist_entry_t *dle;
2504 ll = kmem_zalloc(sizeof (dsl_deadlist_t), KM_SLEEP);
2505 dsl_deadlist_open(ll, mos, ll_obj);
2506 dle = dsl_deadlist_first(ll);
2507 ASSERT3P(dle, !=, NULL);
2508 bplist_create(&to_free);
2509 int err = dsl_process_sub_livelist(&dle->dle_bpobj, &to_free,
2512 sublist_delete_arg_t sync_arg = {
2515 .key = dle->dle_mintxg,
2518 zfs_dbgmsg("deleting sublist (id %llu) from"
2519 " livelist %llu, %d remaining",
2520 dle->dle_bpobj.bpo_object, ll_obj, count - 1);
2521 VERIFY0(dsl_sync_task(spa_name(spa), NULL,
2522 sublist_delete_sync, &sync_arg, 0,
2523 ZFS_SPACE_CHECK_DESTROY));
2525 VERIFY3U(err, ==, EINTR);
2527 bplist_clear(&to_free);
2528 bplist_destroy(&to_free);
2529 dsl_deadlist_close(ll);
2530 kmem_free(ll, sizeof (dsl_deadlist_t));
2532 livelist_delete_arg_t sync_arg = {
2537 zfs_dbgmsg("deletion of livelist %llu completed", ll_obj);
2538 VERIFY0(dsl_sync_task(spa_name(spa), NULL, livelist_delete_sync,
2539 &sync_arg, 0, ZFS_SPACE_CHECK_DESTROY));
2544 spa_start_livelist_destroy_thread(spa_t *spa)
2546 ASSERT3P(spa->spa_livelist_delete_zthr, ==, NULL);
2547 spa->spa_livelist_delete_zthr =
2548 zthr_create("z_livelist_destroy",
2549 spa_livelist_delete_cb_check, spa_livelist_delete_cb, spa);
2552 typedef struct livelist_new_arg {
2555 } livelist_new_arg_t;
2558 livelist_track_new_cb(void *arg, const blkptr_t *bp, boolean_t bp_freed,
2562 livelist_new_arg_t *lna = arg;
2564 bplist_append(lna->frees, bp);
2566 bplist_append(lna->allocs, bp);
2567 zfs_livelist_condense_new_alloc++;
2572 typedef struct livelist_condense_arg {
2575 uint64_t first_size;
2577 } livelist_condense_arg_t;
2580 spa_livelist_condense_sync(void *arg, dmu_tx_t *tx)
2582 livelist_condense_arg_t *lca = arg;
2583 spa_t *spa = lca->spa;
2585 dsl_dataset_t *ds = spa->spa_to_condense.ds;
2587 /* Have we been cancelled? */
2588 if (spa->spa_to_condense.cancelled) {
2589 zfs_livelist_condense_sync_cancel++;
2593 dsl_deadlist_entry_t *first = spa->spa_to_condense.first;
2594 dsl_deadlist_entry_t *next = spa->spa_to_condense.next;
2595 dsl_deadlist_t *ll = &ds->ds_dir->dd_livelist;
2598 * It's possible that the livelist was changed while the zthr was
2599 * running. Therefore, we need to check for new blkptrs in the two
2600 * entries being condensed and continue to track them in the livelist.
2601 * Because of the way we handle remapped blkptrs (see dbuf_remap_impl),
2602 * it's possible that the newly added blkptrs are FREEs or ALLOCs so
2603 * we need to sort them into two different bplists.
2605 uint64_t first_obj = first->dle_bpobj.bpo_object;
2606 uint64_t next_obj = next->dle_bpobj.bpo_object;
2607 uint64_t cur_first_size = first->dle_bpobj.bpo_phys->bpo_num_blkptrs;
2608 uint64_t cur_next_size = next->dle_bpobj.bpo_phys->bpo_num_blkptrs;
2610 bplist_create(&new_frees);
2611 livelist_new_arg_t new_bps = {
2612 .allocs = &lca->to_keep,
2613 .frees = &new_frees,
2616 if (cur_first_size > lca->first_size) {
2617 VERIFY0(livelist_bpobj_iterate_from_nofree(&first->dle_bpobj,
2618 livelist_track_new_cb, &new_bps, lca->first_size));
2620 if (cur_next_size > lca->next_size) {
2621 VERIFY0(livelist_bpobj_iterate_from_nofree(&next->dle_bpobj,
2622 livelist_track_new_cb, &new_bps, lca->next_size));
2625 dsl_deadlist_clear_entry(first, ll, tx);
2626 ASSERT(bpobj_is_empty(&first->dle_bpobj));
2627 dsl_deadlist_remove_entry(ll, next->dle_mintxg, tx);
2629 bplist_iterate(&lca->to_keep, dsl_deadlist_insert_alloc_cb, ll, tx);
2630 bplist_iterate(&new_frees, dsl_deadlist_insert_free_cb, ll, tx);
2631 bplist_destroy(&new_frees);
2633 char dsname[ZFS_MAX_DATASET_NAME_LEN];
2634 dsl_dataset_name(ds, dsname);
2635 zfs_dbgmsg("txg %llu condensing livelist of %s (id %llu), bpobj %llu "
2636 "(%llu blkptrs) and bpobj %llu (%llu blkptrs) -> bpobj %llu "
2637 "(%llu blkptrs)", tx->tx_txg, dsname, ds->ds_object, first_obj,
2638 cur_first_size, next_obj, cur_next_size,
2639 first->dle_bpobj.bpo_object,
2640 first->dle_bpobj.bpo_phys->bpo_num_blkptrs);
2642 dmu_buf_rele(ds->ds_dbuf, spa);
2643 spa->spa_to_condense.ds = NULL;
2644 bplist_clear(&lca->to_keep);
2645 bplist_destroy(&lca->to_keep);
2646 kmem_free(lca, sizeof (livelist_condense_arg_t));
2647 spa->spa_to_condense.syncing = B_FALSE;
2651 spa_livelist_condense_cb(void *arg, zthr_t *t)
2653 while (zfs_livelist_condense_zthr_pause &&
2654 !(zthr_has_waiters(t) || zthr_iscancelled(t)))
2658 dsl_deadlist_entry_t *first = spa->spa_to_condense.first;
2659 dsl_deadlist_entry_t *next = spa->spa_to_condense.next;
2660 uint64_t first_size, next_size;
2662 livelist_condense_arg_t *lca =
2663 kmem_alloc(sizeof (livelist_condense_arg_t), KM_SLEEP);
2664 bplist_create(&lca->to_keep);
2667 * Process the livelists (matching FREEs and ALLOCs) in open context
2668 * so we have minimal work in syncing context to condense.
2670 * We save bpobj sizes (first_size and next_size) to use later in
2671 * syncing context to determine if entries were added to these sublists
2672 * while in open context. This is possible because the clone is still
2673 * active and open for normal writes and we want to make sure the new,
2674 * unprocessed blockpointers are inserted into the livelist normally.
2676 * Note that dsl_process_sub_livelist() both stores the size number of
2677 * blockpointers and iterates over them while the bpobj's lock held, so
2678 * the sizes returned to us are consistent which what was actually
2681 int err = dsl_process_sub_livelist(&first->dle_bpobj, &lca->to_keep, t,
2684 err = dsl_process_sub_livelist(&next->dle_bpobj, &lca->to_keep,
2688 while (zfs_livelist_condense_sync_pause &&
2689 !(zthr_has_waiters(t) || zthr_iscancelled(t)))
2692 dmu_tx_t *tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
2693 dmu_tx_mark_netfree(tx);
2694 dmu_tx_hold_space(tx, 1);
2695 err = dmu_tx_assign(tx, TXG_NOWAIT | TXG_NOTHROTTLE);
2698 * Prevent the condense zthr restarting before
2699 * the synctask completes.
2701 spa->spa_to_condense.syncing = B_TRUE;
2703 lca->first_size = first_size;
2704 lca->next_size = next_size;
2705 dsl_sync_task_nowait(spa_get_dsl(spa),
2706 spa_livelist_condense_sync, lca, tx);
2712 * Condensing can not continue: either it was externally stopped or
2713 * we were unable to assign to a tx because the pool has run out of
2714 * space. In the second case, we'll just end up trying to condense
2715 * again in a later txg.
2718 bplist_clear(&lca->to_keep);
2719 bplist_destroy(&lca->to_keep);
2720 kmem_free(lca, sizeof (livelist_condense_arg_t));
2721 dmu_buf_rele(spa->spa_to_condense.ds->ds_dbuf, spa);
2722 spa->spa_to_condense.ds = NULL;
2724 zfs_livelist_condense_zthr_cancel++;
2729 * Check that there is something to condense but that a condense is not
2730 * already in progress and that condensing has not been cancelled.
2733 spa_livelist_condense_cb_check(void *arg, zthr_t *z)
2736 if ((spa->spa_to_condense.ds != NULL) &&
2737 (spa->spa_to_condense.syncing == B_FALSE) &&
2738 (spa->spa_to_condense.cancelled == B_FALSE)) {
2745 spa_start_livelist_condensing_thread(spa_t *spa)
2747 spa->spa_to_condense.ds = NULL;
2748 spa->spa_to_condense.first = NULL;
2749 spa->spa_to_condense.next = NULL;
2750 spa->spa_to_condense.syncing = B_FALSE;
2751 spa->spa_to_condense.cancelled = B_FALSE;
2753 ASSERT3P(spa->spa_livelist_condense_zthr, ==, NULL);
2754 spa->spa_livelist_condense_zthr =
2755 zthr_create("z_livelist_condense",
2756 spa_livelist_condense_cb_check,
2757 spa_livelist_condense_cb, spa);
2761 spa_spawn_aux_threads(spa_t *spa)
2763 ASSERT(spa_writeable(spa));
2765 ASSERT(MUTEX_HELD(&spa_namespace_lock));
2767 spa_start_indirect_condensing_thread(spa);
2768 spa_start_livelist_destroy_thread(spa);
2769 spa_start_livelist_condensing_thread(spa);
2771 ASSERT3P(spa->spa_checkpoint_discard_zthr, ==, NULL);
2772 spa->spa_checkpoint_discard_zthr =
2773 zthr_create("z_checkpoint_discard",
2774 spa_checkpoint_discard_thread_check,
2775 spa_checkpoint_discard_thread, spa);
2779 * Fix up config after a partly-completed split. This is done with the
2780 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
2781 * pool have that entry in their config, but only the splitting one contains
2782 * a list of all the guids of the vdevs that are being split off.
2784 * This function determines what to do with that list: either rejoin
2785 * all the disks to the pool, or complete the splitting process. To attempt
2786 * the rejoin, each disk that is offlined is marked online again, and
2787 * we do a reopen() call. If the vdev label for every disk that was
2788 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
2789 * then we call vdev_split() on each disk, and complete the split.
2791 * Otherwise we leave the config alone, with all the vdevs in place in
2792 * the original pool.
2795 spa_try_repair(spa_t *spa, nvlist_t *config)
2802 boolean_t attempt_reopen;
2804 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
2807 /* check that the config is complete */
2808 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
2809 &glist, &gcount) != 0)
2812 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
2814 /* attempt to online all the vdevs & validate */
2815 attempt_reopen = B_TRUE;
2816 for (i = 0; i < gcount; i++) {
2817 if (glist[i] == 0) /* vdev is hole */
2820 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
2821 if (vd[i] == NULL) {
2823 * Don't bother attempting to reopen the disks;
2824 * just do the split.
2826 attempt_reopen = B_FALSE;
2828 /* attempt to re-online it */
2829 vd[i]->vdev_offline = B_FALSE;
2833 if (attempt_reopen) {
2834 vdev_reopen(spa->spa_root_vdev);
2836 /* check each device to see what state it's in */
2837 for (extracted = 0, i = 0; i < gcount; i++) {
2838 if (vd[i] != NULL &&
2839 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
2846 * If every disk has been moved to the new pool, or if we never
2847 * even attempted to look at them, then we split them off for
2850 if (!attempt_reopen || gcount == extracted) {
2851 for (i = 0; i < gcount; i++)
2854 vdev_reopen(spa->spa_root_vdev);
2857 kmem_free(vd, gcount * sizeof (vdev_t *));
2861 spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type)
2863 char *ereport = FM_EREPORT_ZFS_POOL;
2866 spa->spa_load_state = state;
2867 (void) spa_import_progress_set_state(spa_guid(spa),
2868 spa_load_state(spa));
2870 gethrestime(&spa->spa_loaded_ts);
2871 error = spa_load_impl(spa, type, &ereport);
2874 * Don't count references from objsets that are already closed
2875 * and are making their way through the eviction process.
2877 spa_evicting_os_wait(spa);
2878 spa->spa_minref = zfs_refcount_count(&spa->spa_refcount);
2880 if (error != EEXIST) {
2881 spa->spa_loaded_ts.tv_sec = 0;
2882 spa->spa_loaded_ts.tv_nsec = 0;
2884 if (error != EBADF) {
2885 (void) zfs_ereport_post(ereport, spa,
2886 NULL, NULL, NULL, 0);
2889 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2892 (void) spa_import_progress_set_state(spa_guid(spa),
2893 spa_load_state(spa));
2900 * Count the number of per-vdev ZAPs associated with all of the vdevs in the
2901 * vdev tree rooted in the given vd, and ensure that each ZAP is present in the
2902 * spa's per-vdev ZAP list.
2905 vdev_count_verify_zaps(vdev_t *vd)
2907 spa_t *spa = vd->vdev_spa;
2910 if (vd->vdev_top_zap != 0) {
2912 ASSERT0(zap_lookup_int(spa->spa_meta_objset,
2913 spa->spa_all_vdev_zaps, vd->vdev_top_zap));
2915 if (vd->vdev_leaf_zap != 0) {
2917 ASSERT0(zap_lookup_int(spa->spa_meta_objset,
2918 spa->spa_all_vdev_zaps, vd->vdev_leaf_zap));
2921 for (uint64_t i = 0; i < vd->vdev_children; i++) {
2922 total += vdev_count_verify_zaps(vd->vdev_child[i]);
2930 * Determine whether the activity check is required.
2933 spa_activity_check_required(spa_t *spa, uberblock_t *ub, nvlist_t *label,
2937 uint64_t hostid = 0;
2938 uint64_t tryconfig_txg = 0;
2939 uint64_t tryconfig_timestamp = 0;
2940 uint16_t tryconfig_mmp_seq = 0;
2943 if (nvlist_exists(config, ZPOOL_CONFIG_LOAD_INFO)) {
2944 nvinfo = fnvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO);
2945 (void) nvlist_lookup_uint64(nvinfo, ZPOOL_CONFIG_MMP_TXG,
2947 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
2948 &tryconfig_timestamp);
2949 (void) nvlist_lookup_uint16(nvinfo, ZPOOL_CONFIG_MMP_SEQ,
2950 &tryconfig_mmp_seq);
2953 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE, &state);
2956 * Disable the MMP activity check - This is used by zdb which
2957 * is intended to be used on potentially active pools.
2959 if (spa->spa_import_flags & ZFS_IMPORT_SKIP_MMP)
2963 * Skip the activity check when the MMP feature is disabled.
2965 if (ub->ub_mmp_magic == MMP_MAGIC && ub->ub_mmp_delay == 0)
2969 * If the tryconfig_ values are nonzero, they are the results of an
2970 * earlier tryimport. If they all match the uberblock we just found,
2971 * then the pool has not changed and we return false so we do not test
2974 if (tryconfig_txg && tryconfig_txg == ub->ub_txg &&
2975 tryconfig_timestamp && tryconfig_timestamp == ub->ub_timestamp &&
2976 tryconfig_mmp_seq && tryconfig_mmp_seq ==
2977 (MMP_SEQ_VALID(ub) ? MMP_SEQ(ub) : 0))
2981 * Allow the activity check to be skipped when importing the pool
2982 * on the same host which last imported it. Since the hostid from
2983 * configuration may be stale use the one read from the label.
2985 if (nvlist_exists(label, ZPOOL_CONFIG_HOSTID))
2986 hostid = fnvlist_lookup_uint64(label, ZPOOL_CONFIG_HOSTID);
2988 if (hostid == spa_get_hostid(spa))
2992 * Skip the activity test when the pool was cleanly exported.
2994 if (state != POOL_STATE_ACTIVE)
3001 * Nanoseconds the activity check must watch for changes on-disk.
3004 spa_activity_check_duration(spa_t *spa, uberblock_t *ub)
3006 uint64_t import_intervals = MAX(zfs_multihost_import_intervals, 1);
3007 uint64_t multihost_interval = MSEC2NSEC(
3008 MMP_INTERVAL_OK(zfs_multihost_interval));
3009 uint64_t import_delay = MAX(NANOSEC, import_intervals *
3010 multihost_interval);
3013 * Local tunables determine a minimum duration except for the case
3014 * where we know when the remote host will suspend the pool if MMP
3015 * writes do not land.
3017 * See Big Theory comment at the top of mmp.c for the reasoning behind
3018 * these cases and times.
3021 ASSERT(MMP_IMPORT_SAFETY_FACTOR >= 100);
3023 if (MMP_INTERVAL_VALID(ub) && MMP_FAIL_INT_VALID(ub) &&
3024 MMP_FAIL_INT(ub) > 0) {
3026 /* MMP on remote host will suspend pool after failed writes */
3027 import_delay = MMP_FAIL_INT(ub) * MSEC2NSEC(MMP_INTERVAL(ub)) *
3028 MMP_IMPORT_SAFETY_FACTOR / 100;
3030 zfs_dbgmsg("fail_intvals>0 import_delay=%llu ub_mmp "
3031 "mmp_fails=%llu ub_mmp mmp_interval=%llu "
3032 "import_intervals=%u", import_delay, MMP_FAIL_INT(ub),
3033 MMP_INTERVAL(ub), import_intervals);
3035 } else if (MMP_INTERVAL_VALID(ub) && MMP_FAIL_INT_VALID(ub) &&
3036 MMP_FAIL_INT(ub) == 0) {
3038 /* MMP on remote host will never suspend pool */
3039 import_delay = MAX(import_delay, (MSEC2NSEC(MMP_INTERVAL(ub)) +
3040 ub->ub_mmp_delay) * import_intervals);
3042 zfs_dbgmsg("fail_intvals=0 import_delay=%llu ub_mmp "
3043 "mmp_interval=%llu ub_mmp_delay=%llu "
3044 "import_intervals=%u", import_delay, MMP_INTERVAL(ub),
3045 ub->ub_mmp_delay, import_intervals);
3047 } else if (MMP_VALID(ub)) {
3049 * zfs-0.7 compatibility case
3052 import_delay = MAX(import_delay, (multihost_interval +
3053 ub->ub_mmp_delay) * import_intervals);
3055 zfs_dbgmsg("import_delay=%llu ub_mmp_delay=%llu "
3056 "import_intervals=%u leaves=%u", import_delay,
3057 ub->ub_mmp_delay, import_intervals,
3058 vdev_count_leaves(spa));
3060 /* Using local tunings is the only reasonable option */
3061 zfs_dbgmsg("pool last imported on non-MMP aware "
3062 "host using import_delay=%llu multihost_interval=%llu "
3063 "import_intervals=%u", import_delay, multihost_interval,
3067 return (import_delay);
3071 * Perform the import activity check. If the user canceled the import or
3072 * we detected activity then fail.
3075 spa_activity_check(spa_t *spa, uberblock_t *ub, nvlist_t *config)
3077 uint64_t txg = ub->ub_txg;
3078 uint64_t timestamp = ub->ub_timestamp;
3079 uint64_t mmp_config = ub->ub_mmp_config;
3080 uint16_t mmp_seq = MMP_SEQ_VALID(ub) ? MMP_SEQ(ub) : 0;
3081 uint64_t import_delay;
3082 hrtime_t import_expire;
3083 nvlist_t *mmp_label = NULL;
3084 vdev_t *rvd = spa->spa_root_vdev;
3089 cv_init(&cv, NULL, CV_DEFAULT, NULL);
3090 mutex_init(&mtx, NULL, MUTEX_DEFAULT, NULL);
3094 * If ZPOOL_CONFIG_MMP_TXG is present an activity check was performed
3095 * during the earlier tryimport. If the txg recorded there is 0 then
3096 * the pool is known to be active on another host.
3098 * Otherwise, the pool might be in use on another host. Check for
3099 * changes in the uberblocks on disk if necessary.
3101 if (nvlist_exists(config, ZPOOL_CONFIG_LOAD_INFO)) {
3102 nvlist_t *nvinfo = fnvlist_lookup_nvlist(config,
3103 ZPOOL_CONFIG_LOAD_INFO);
3105 if (nvlist_exists(nvinfo, ZPOOL_CONFIG_MMP_TXG) &&
3106 fnvlist_lookup_uint64(nvinfo, ZPOOL_CONFIG_MMP_TXG) == 0) {
3107 vdev_uberblock_load(rvd, ub, &mmp_label);
3108 error = SET_ERROR(EREMOTEIO);
3113 import_delay = spa_activity_check_duration(spa, ub);
3115 /* Add a small random factor in case of simultaneous imports (0-25%) */
3116 import_delay += import_delay * spa_get_random(250) / 1000;
3118 import_expire = gethrtime() + import_delay;
3120 while (gethrtime() < import_expire) {
3121 (void) spa_import_progress_set_mmp_check(spa_guid(spa),
3122 NSEC2SEC(import_expire - gethrtime()));
3124 vdev_uberblock_load(rvd, ub, &mmp_label);
3126 if (txg != ub->ub_txg || timestamp != ub->ub_timestamp ||
3127 mmp_seq != (MMP_SEQ_VALID(ub) ? MMP_SEQ(ub) : 0)) {
3128 zfs_dbgmsg("multihost activity detected "
3129 "txg %llu ub_txg %llu "
3130 "timestamp %llu ub_timestamp %llu "
3131 "mmp_config %#llx ub_mmp_config %#llx",
3132 txg, ub->ub_txg, timestamp, ub->ub_timestamp,
3133 mmp_config, ub->ub_mmp_config);
3135 error = SET_ERROR(EREMOTEIO);
3140 nvlist_free(mmp_label);
3144 error = cv_timedwait_sig(&cv, &mtx, ddi_get_lbolt() + hz);
3146 error = SET_ERROR(EINTR);
3154 mutex_destroy(&mtx);
3158 * If the pool is determined to be active store the status in the
3159 * spa->spa_load_info nvlist. If the remote hostname or hostid are
3160 * available from configuration read from disk store them as well.
3161 * This allows 'zpool import' to generate a more useful message.
3163 * ZPOOL_CONFIG_MMP_STATE - observed pool status (mandatory)
3164 * ZPOOL_CONFIG_MMP_HOSTNAME - hostname from the active pool
3165 * ZPOOL_CONFIG_MMP_HOSTID - hostid from the active pool
3167 if (error == EREMOTEIO) {
3168 char *hostname = "<unknown>";
3169 uint64_t hostid = 0;
3172 if (nvlist_exists(mmp_label, ZPOOL_CONFIG_HOSTNAME)) {
3173 hostname = fnvlist_lookup_string(mmp_label,
3174 ZPOOL_CONFIG_HOSTNAME);
3175 fnvlist_add_string(spa->spa_load_info,
3176 ZPOOL_CONFIG_MMP_HOSTNAME, hostname);
3179 if (nvlist_exists(mmp_label, ZPOOL_CONFIG_HOSTID)) {
3180 hostid = fnvlist_lookup_uint64(mmp_label,
3181 ZPOOL_CONFIG_HOSTID);
3182 fnvlist_add_uint64(spa->spa_load_info,
3183 ZPOOL_CONFIG_MMP_HOSTID, hostid);
3187 fnvlist_add_uint64(spa->spa_load_info,
3188 ZPOOL_CONFIG_MMP_STATE, MMP_STATE_ACTIVE);
3189 fnvlist_add_uint64(spa->spa_load_info,
3190 ZPOOL_CONFIG_MMP_TXG, 0);
3192 error = spa_vdev_err(rvd, VDEV_AUX_ACTIVE, EREMOTEIO);
3196 nvlist_free(mmp_label);
3202 spa_verify_host(spa_t *spa, nvlist_t *mos_config)
3206 uint64_t myhostid = 0;
3208 if (!spa_is_root(spa) && nvlist_lookup_uint64(mos_config,
3209 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
3210 hostname = fnvlist_lookup_string(mos_config,
3211 ZPOOL_CONFIG_HOSTNAME);
3213 myhostid = zone_get_hostid(NULL);
3215 if (hostid != 0 && myhostid != 0 && hostid != myhostid) {
3216 cmn_err(CE_WARN, "pool '%s' could not be "
3217 "loaded as it was last accessed by "
3218 "another system (host: %s hostid: 0x%llx). "
3219 "See: https://openzfs.github.io/openzfs-docs/msg/"
3221 spa_name(spa), hostname, (u_longlong_t)hostid);
3222 spa_load_failed(spa, "hostid verification failed: pool "
3223 "last accessed by host: %s (hostid: 0x%llx)",
3224 hostname, (u_longlong_t)hostid);
3225 return (SET_ERROR(EBADF));
3233 spa_ld_parse_config(spa_t *spa, spa_import_type_t type)
3236 nvlist_t *nvtree, *nvl, *config = spa->spa_config;
3243 * Versioning wasn't explicitly added to the label until later, so if
3244 * it's not present treat it as the initial version.
3246 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
3247 &spa->spa_ubsync.ub_version) != 0)
3248 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
3250 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) {
3251 spa_load_failed(spa, "invalid config provided: '%s' missing",
3252 ZPOOL_CONFIG_POOL_GUID);
3253 return (SET_ERROR(EINVAL));
3257 * If we are doing an import, ensure that the pool is not already
3258 * imported by checking if its pool guid already exists in the
3261 * The only case that we allow an already imported pool to be
3262 * imported again, is when the pool is checkpointed and we want to
3263 * look at its checkpointed state from userland tools like zdb.
3266 if ((spa->spa_load_state == SPA_LOAD_IMPORT ||
3267 spa->spa_load_state == SPA_LOAD_TRYIMPORT) &&
3268 spa_guid_exists(pool_guid, 0)) {
3270 if ((spa->spa_load_state == SPA_LOAD_IMPORT ||
3271 spa->spa_load_state == SPA_LOAD_TRYIMPORT) &&
3272 spa_guid_exists(pool_guid, 0) &&
3273 !spa_importing_readonly_checkpoint(spa)) {
3275 spa_load_failed(spa, "a pool with guid %llu is already open",
3276 (u_longlong_t)pool_guid);
3277 return (SET_ERROR(EEXIST));
3280 spa->spa_config_guid = pool_guid;
3282 nvlist_free(spa->spa_load_info);
3283 spa->spa_load_info = fnvlist_alloc();
3285 ASSERT(spa->spa_comment == NULL);
3286 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
3287 spa->spa_comment = spa_strdup(comment);
3289 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
3290 &spa->spa_config_txg);
3292 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) == 0)
3293 spa->spa_config_splitting = fnvlist_dup(nvl);
3295 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvtree)) {
3296 spa_load_failed(spa, "invalid config provided: '%s' missing",
3297 ZPOOL_CONFIG_VDEV_TREE);
3298 return (SET_ERROR(EINVAL));
3302 * Create "The Godfather" zio to hold all async IOs
3304 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
3306 for (int i = 0; i < max_ncpus; i++) {
3307 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
3308 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
3309 ZIO_FLAG_GODFATHER);
3313 * Parse the configuration into a vdev tree. We explicitly set the
3314 * value that will be returned by spa_version() since parsing the
3315 * configuration requires knowing the version number.
3317 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3318 parse = (type == SPA_IMPORT_EXISTING ?
3319 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
3320 error = spa_config_parse(spa, &rvd, nvtree, NULL, 0, parse);
3321 spa_config_exit(spa, SCL_ALL, FTAG);
3324 spa_load_failed(spa, "unable to parse config [error=%d]",
3329 ASSERT(spa->spa_root_vdev == rvd);
3330 ASSERT3U(spa->spa_min_ashift, >=, SPA_MINBLOCKSHIFT);
3331 ASSERT3U(spa->spa_max_ashift, <=, SPA_MAXBLOCKSHIFT);
3333 if (type != SPA_IMPORT_ASSEMBLE) {
3334 ASSERT(spa_guid(spa) == pool_guid);
3341 * Recursively open all vdevs in the vdev tree. This function is called twice:
3342 * first with the untrusted config, then with the trusted config.
3345 spa_ld_open_vdevs(spa_t *spa)
3350 * spa_missing_tvds_allowed defines how many top-level vdevs can be
3351 * missing/unopenable for the root vdev to be still considered openable.
3353 if (spa->spa_trust_config) {
3354 spa->spa_missing_tvds_allowed = zfs_max_missing_tvds;
3355 } else if (spa->spa_config_source == SPA_CONFIG_SRC_CACHEFILE) {
3356 spa->spa_missing_tvds_allowed = zfs_max_missing_tvds_cachefile;
3357 } else if (spa->spa_config_source == SPA_CONFIG_SRC_SCAN) {
3358 spa->spa_missing_tvds_allowed = zfs_max_missing_tvds_scan;
3360 spa->spa_missing_tvds_allowed = 0;
3363 spa->spa_missing_tvds_allowed =
3364 MAX(zfs_max_missing_tvds, spa->spa_missing_tvds_allowed);
3366 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3367 error = vdev_open(spa->spa_root_vdev);
3368 spa_config_exit(spa, SCL_ALL, FTAG);
3370 if (spa->spa_missing_tvds != 0) {
3371 spa_load_note(spa, "vdev tree has %lld missing top-level "
3372 "vdevs.", (u_longlong_t)spa->spa_missing_tvds);
3373 if (spa->spa_trust_config && (spa->spa_mode & SPA_MODE_WRITE)) {
3375 * Although theoretically we could allow users to open
3376 * incomplete pools in RW mode, we'd need to add a lot
3377 * of extra logic (e.g. adjust pool space to account
3378 * for missing vdevs).
3379 * This limitation also prevents users from accidentally
3380 * opening the pool in RW mode during data recovery and
3381 * damaging it further.
3383 spa_load_note(spa, "pools with missing top-level "
3384 "vdevs can only be opened in read-only mode.");
3385 error = SET_ERROR(ENXIO);
3387 spa_load_note(spa, "current settings allow for maximum "
3388 "%lld missing top-level vdevs at this stage.",
3389 (u_longlong_t)spa->spa_missing_tvds_allowed);
3393 spa_load_failed(spa, "unable to open vdev tree [error=%d]",
3396 if (spa->spa_missing_tvds != 0 || error != 0)
3397 vdev_dbgmsg_print_tree(spa->spa_root_vdev, 2);
3403 * We need to validate the vdev labels against the configuration that
3404 * we have in hand. This function is called twice: first with an untrusted
3405 * config, then with a trusted config. The validation is more strict when the
3406 * config is trusted.
3409 spa_ld_validate_vdevs(spa_t *spa)
3412 vdev_t *rvd = spa->spa_root_vdev;
3414 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3415 error = vdev_validate(rvd);
3416 spa_config_exit(spa, SCL_ALL, FTAG);
3419 spa_load_failed(spa, "vdev_validate failed [error=%d]", error);
3423 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) {
3424 spa_load_failed(spa, "cannot open vdev tree after invalidating "
3426 vdev_dbgmsg_print_tree(rvd, 2);
3427 return (SET_ERROR(ENXIO));
3434 spa_ld_select_uberblock_done(spa_t *spa, uberblock_t *ub)
3436 spa->spa_state = POOL_STATE_ACTIVE;
3437 spa->spa_ubsync = spa->spa_uberblock;
3438 spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
3439 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
3440 spa->spa_first_txg = spa->spa_last_ubsync_txg ?
3441 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
3442 spa->spa_claim_max_txg = spa->spa_first_txg;
3443 spa->spa_prev_software_version = ub->ub_software_version;
3447 spa_ld_select_uberblock(spa_t *spa, spa_import_type_t type)
3449 vdev_t *rvd = spa->spa_root_vdev;
3451 uberblock_t *ub = &spa->spa_uberblock;
3452 boolean_t activity_check = B_FALSE;
3455 * If we are opening the checkpointed state of the pool by
3456 * rewinding to it, at this point we will have written the
3457 * checkpointed uberblock to the vdev labels, so searching
3458 * the labels will find the right uberblock. However, if
3459 * we are opening the checkpointed state read-only, we have
3460 * not modified the labels. Therefore, we must ignore the
3461 * labels and continue using the spa_uberblock that was set
3462 * by spa_ld_checkpoint_rewind.
3464 * Note that it would be fine to ignore the labels when
3465 * rewinding (opening writeable) as well. However, if we
3466 * crash just after writing the labels, we will end up
3467 * searching the labels. Doing so in the common case means
3468 * that this code path gets exercised normally, rather than
3469 * just in the edge case.
3471 if (ub->ub_checkpoint_txg != 0 &&
3472 spa_importing_readonly_checkpoint(spa)) {
3473 spa_ld_select_uberblock_done(spa, ub);
3478 * Find the best uberblock.
3480 vdev_uberblock_load(rvd, ub, &label);
3483 * If we weren't able to find a single valid uberblock, return failure.
3485 if (ub->ub_txg == 0) {
3487 spa_load_failed(spa, "no valid uberblock found");
3488 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
3491 if (spa->spa_load_max_txg != UINT64_MAX) {
3492 (void) spa_import_progress_set_max_txg(spa_guid(spa),
3493 (u_longlong_t)spa->spa_load_max_txg);
3495 spa_load_note(spa, "using uberblock with txg=%llu",
3496 (u_longlong_t)ub->ub_txg);
3500 * For pools which have the multihost property on determine if the
3501 * pool is truly inactive and can be safely imported. Prevent
3502 * hosts which don't have a hostid set from importing the pool.
3504 activity_check = spa_activity_check_required(spa, ub, label,
3506 if (activity_check) {
3507 if (ub->ub_mmp_magic == MMP_MAGIC && ub->ub_mmp_delay &&
3508 spa_get_hostid(spa) == 0) {
3510 fnvlist_add_uint64(spa->spa_load_info,
3511 ZPOOL_CONFIG_MMP_STATE, MMP_STATE_NO_HOSTID);
3512 return (spa_vdev_err(rvd, VDEV_AUX_ACTIVE, EREMOTEIO));
3515 int error = spa_activity_check(spa, ub, spa->spa_config);
3521 fnvlist_add_uint64(spa->spa_load_info,
3522 ZPOOL_CONFIG_MMP_STATE, MMP_STATE_INACTIVE);
3523 fnvlist_add_uint64(spa->spa_load_info,
3524 ZPOOL_CONFIG_MMP_TXG, ub->ub_txg);
3525 fnvlist_add_uint16(spa->spa_load_info,
3526 ZPOOL_CONFIG_MMP_SEQ,
3527 (MMP_SEQ_VALID(ub) ? MMP_SEQ(ub) : 0));
3531 * If the pool has an unsupported version we can't open it.
3533 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
3535 spa_load_failed(spa, "version %llu is not supported",
3536 (u_longlong_t)ub->ub_version);
3537 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
3540 if (ub->ub_version >= SPA_VERSION_FEATURES) {
3544 * If we weren't able to find what's necessary for reading the
3545 * MOS in the label, return failure.
3547 if (label == NULL) {
3548 spa_load_failed(spa, "label config unavailable");
3549 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
3553 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_FEATURES_FOR_READ,
3556 spa_load_failed(spa, "invalid label: '%s' missing",
3557 ZPOOL_CONFIG_FEATURES_FOR_READ);
3558 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
3563 * Update our in-core representation with the definitive values
3566 nvlist_free(spa->spa_label_features);
3567 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
3573 * Look through entries in the label nvlist's features_for_read. If
3574 * there is a feature listed there which we don't understand then we
3575 * cannot open a pool.
3577 if (ub->ub_version >= SPA_VERSION_FEATURES) {
3578 nvlist_t *unsup_feat;
3580 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
3583 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
3585 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
3586 if (!zfeature_is_supported(nvpair_name(nvp))) {
3587 VERIFY(nvlist_add_string(unsup_feat,
3588 nvpair_name(nvp), "") == 0);
3592 if (!nvlist_empty(unsup_feat)) {
3593 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
3594 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
3595 nvlist_free(unsup_feat);
3596 spa_load_failed(spa, "some features are unsupported");
3597 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
3601 nvlist_free(unsup_feat);
3604 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
3605 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3606 spa_try_repair(spa, spa->spa_config);
3607 spa_config_exit(spa, SCL_ALL, FTAG);
3608 nvlist_free(spa->spa_config_splitting);
3609 spa->spa_config_splitting = NULL;
3613 * Initialize internal SPA structures.
3615 spa_ld_select_uberblock_done(spa, ub);
3621 spa_ld_open_rootbp(spa_t *spa)
3624 vdev_t *rvd = spa->spa_root_vdev;
3626 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
3628 spa_load_failed(spa, "unable to open rootbp in dsl_pool_init "
3629 "[error=%d]", error);
3630 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3632 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
3638 spa_ld_trusted_config(spa_t *spa, spa_import_type_t type,
3639 boolean_t reloading)
3641 vdev_t *mrvd, *rvd = spa->spa_root_vdev;
3642 nvlist_t *nv, *mos_config, *policy;
3643 int error = 0, copy_error;
3644 uint64_t healthy_tvds, healthy_tvds_mos;
3645 uint64_t mos_config_txg;
3647 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object, B_TRUE)
3649 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3652 * If we're assembling a pool from a split, the config provided is
3653 * already trusted so there is nothing to do.
3655 if (type == SPA_IMPORT_ASSEMBLE)
3658 healthy_tvds = spa_healthy_core_tvds(spa);
3660 if (load_nvlist(spa, spa->spa_config_object, &mos_config)
3662 spa_load_failed(spa, "unable to retrieve MOS config");
3663 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3667 * If we are doing an open, pool owner wasn't verified yet, thus do
3668 * the verification here.
3670 if (spa->spa_load_state == SPA_LOAD_OPEN) {
3671 error = spa_verify_host(spa, mos_config);
3673 nvlist_free(mos_config);
3678 nv = fnvlist_lookup_nvlist(mos_config, ZPOOL_CONFIG_VDEV_TREE);
3680 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3683 * Build a new vdev tree from the trusted config
3685 error = spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD);
3687 nvlist_free(mos_config);
3688 spa_config_exit(spa, SCL_ALL, FTAG);
3689 spa_load_failed(spa, "spa_config_parse failed [error=%d]",
3691 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, error));
3695 * Vdev paths in the MOS may be obsolete. If the untrusted config was
3696 * obtained by scanning /dev/dsk, then it will have the right vdev
3697 * paths. We update the trusted MOS config with this information.
3698 * We first try to copy the paths with vdev_copy_path_strict, which
3699 * succeeds only when both configs have exactly the same vdev tree.
3700 * If that fails, we fall back to a more flexible method that has a
3701 * best effort policy.
3703 copy_error = vdev_copy_path_strict(rvd, mrvd);
3704 if (copy_error != 0 || spa_load_print_vdev_tree) {
3705 spa_load_note(spa, "provided vdev tree:");
3706 vdev_dbgmsg_print_tree(rvd, 2);
3707 spa_load_note(spa, "MOS vdev tree:");
3708 vdev_dbgmsg_print_tree(mrvd, 2);
3710 if (copy_error != 0) {
3711 spa_load_note(spa, "vdev_copy_path_strict failed, falling "
3712 "back to vdev_copy_path_relaxed");
3713 vdev_copy_path_relaxed(rvd, mrvd);
3718 spa->spa_root_vdev = mrvd;
3720 spa_config_exit(spa, SCL_ALL, FTAG);
3723 * We will use spa_config if we decide to reload the spa or if spa_load
3724 * fails and we rewind. We must thus regenerate the config using the
3725 * MOS information with the updated paths. ZPOOL_LOAD_POLICY is used to
3726 * pass settings on how to load the pool and is not stored in the MOS.
3727 * We copy it over to our new, trusted config.
3729 mos_config_txg = fnvlist_lookup_uint64(mos_config,
3730 ZPOOL_CONFIG_POOL_TXG);
3731 nvlist_free(mos_config);
3732 mos_config = spa_config_generate(spa, NULL, mos_config_txg, B_FALSE);
3733 if (nvlist_lookup_nvlist(spa->spa_config, ZPOOL_LOAD_POLICY,
3735 fnvlist_add_nvlist(mos_config, ZPOOL_LOAD_POLICY, policy);
3736 spa_config_set(spa, mos_config);
3737 spa->spa_config_source = SPA_CONFIG_SRC_MOS;
3740 * Now that we got the config from the MOS, we should be more strict
3741 * in checking blkptrs and can make assumptions about the consistency
3742 * of the vdev tree. spa_trust_config must be set to true before opening
3743 * vdevs in order for them to be writeable.
3745 spa->spa_trust_config = B_TRUE;
3748 * Open and validate the new vdev tree
3750 error = spa_ld_open_vdevs(spa);
3754 error = spa_ld_validate_vdevs(spa);
3758 if (copy_error != 0 || spa_load_print_vdev_tree) {
3759 spa_load_note(spa, "final vdev tree:");
3760 vdev_dbgmsg_print_tree(rvd, 2);
3763 if (spa->spa_load_state != SPA_LOAD_TRYIMPORT &&
3764 !spa->spa_extreme_rewind && zfs_max_missing_tvds == 0) {
3766 * Sanity check to make sure that we are indeed loading the
3767 * latest uberblock. If we missed SPA_SYNC_MIN_VDEVS tvds
3768 * in the config provided and they happened to be the only ones
3769 * to have the latest uberblock, we could involuntarily perform
3770 * an extreme rewind.
3772 healthy_tvds_mos = spa_healthy_core_tvds(spa);
3773 if (healthy_tvds_mos - healthy_tvds >=
3774 SPA_SYNC_MIN_VDEVS) {
3775 spa_load_note(spa, "config provided misses too many "
3776 "top-level vdevs compared to MOS (%lld vs %lld). ",
3777 (u_longlong_t)healthy_tvds,
3778 (u_longlong_t)healthy_tvds_mos);
3779 spa_load_note(spa, "vdev tree:");
3780 vdev_dbgmsg_print_tree(rvd, 2);
3782 spa_load_failed(spa, "config was already "
3783 "provided from MOS. Aborting.");
3784 return (spa_vdev_err(rvd,
3785 VDEV_AUX_CORRUPT_DATA, EIO));
3787 spa_load_note(spa, "spa must be reloaded using MOS "
3789 return (SET_ERROR(EAGAIN));
3793 error = spa_check_for_missing_logs(spa);
3795 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
3797 if (rvd->vdev_guid_sum != spa->spa_uberblock.ub_guid_sum) {
3798 spa_load_failed(spa, "uberblock guid sum doesn't match MOS "
3799 "guid sum (%llu != %llu)",
3800 (u_longlong_t)spa->spa_uberblock.ub_guid_sum,
3801 (u_longlong_t)rvd->vdev_guid_sum);
3802 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
3810 spa_ld_open_indirect_vdev_metadata(spa_t *spa)
3813 vdev_t *rvd = spa->spa_root_vdev;
3816 * Everything that we read before spa_remove_init() must be stored
3817 * on concreted vdevs. Therefore we do this as early as possible.
3819 error = spa_remove_init(spa);
3821 spa_load_failed(spa, "spa_remove_init failed [error=%d]",
3823 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3827 * Retrieve information needed to condense indirect vdev mappings.
3829 error = spa_condense_init(spa);
3831 spa_load_failed(spa, "spa_condense_init failed [error=%d]",
3833 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, error));
3840 spa_ld_check_features(spa_t *spa, boolean_t *missing_feat_writep)
3843 vdev_t *rvd = spa->spa_root_vdev;
3845 if (spa_version(spa) >= SPA_VERSION_FEATURES) {
3846 boolean_t missing_feat_read = B_FALSE;
3847 nvlist_t *unsup_feat, *enabled_feat;
3849 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
3850 &spa->spa_feat_for_read_obj, B_TRUE) != 0) {
3851 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3854 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
3855 &spa->spa_feat_for_write_obj, B_TRUE) != 0) {
3856 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3859 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
3860 &spa->spa_feat_desc_obj, B_TRUE) != 0) {
3861 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3864 enabled_feat = fnvlist_alloc();
3865 unsup_feat = fnvlist_alloc();
3867 if (!spa_features_check(spa, B_FALSE,
3868 unsup_feat, enabled_feat))
3869 missing_feat_read = B_TRUE;
3871 if (spa_writeable(spa) ||
3872 spa->spa_load_state == SPA_LOAD_TRYIMPORT) {
3873 if (!spa_features_check(spa, B_TRUE,
3874 unsup_feat, enabled_feat)) {
3875 *missing_feat_writep = B_TRUE;
3879 fnvlist_add_nvlist(spa->spa_load_info,
3880 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
3882 if (!nvlist_empty(unsup_feat)) {
3883 fnvlist_add_nvlist(spa->spa_load_info,
3884 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
3887 fnvlist_free(enabled_feat);
3888 fnvlist_free(unsup_feat);
3890 if (!missing_feat_read) {
3891 fnvlist_add_boolean(spa->spa_load_info,
3892 ZPOOL_CONFIG_CAN_RDONLY);
3896 * If the state is SPA_LOAD_TRYIMPORT, our objective is
3897 * twofold: to determine whether the pool is available for
3898 * import in read-write mode and (if it is not) whether the
3899 * pool is available for import in read-only mode. If the pool
3900 * is available for import in read-write mode, it is displayed
3901 * as available in userland; if it is not available for import
3902 * in read-only mode, it is displayed as unavailable in
3903 * userland. If the pool is available for import in read-only
3904 * mode but not read-write mode, it is displayed as unavailable
3905 * in userland with a special note that the pool is actually
3906 * available for open in read-only mode.
3908 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
3909 * missing a feature for write, we must first determine whether
3910 * the pool can be opened read-only before returning to
3911 * userland in order to know whether to display the
3912 * abovementioned note.
3914 if (missing_feat_read || (*missing_feat_writep &&
3915 spa_writeable(spa))) {
3916 spa_load_failed(spa, "pool uses unsupported features");
3917 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
3922 * Load refcounts for ZFS features from disk into an in-memory
3923 * cache during SPA initialization.
3925 for (spa_feature_t i = 0; i < SPA_FEATURES; i++) {
3928 error = feature_get_refcount_from_disk(spa,
3929 &spa_feature_table[i], &refcount);
3931 spa->spa_feat_refcount_cache[i] = refcount;
3932 } else if (error == ENOTSUP) {
3933 spa->spa_feat_refcount_cache[i] =
3934 SPA_FEATURE_DISABLED;
3936 spa_load_failed(spa, "error getting refcount "
3937 "for feature %s [error=%d]",
3938 spa_feature_table[i].fi_guid, error);
3939 return (spa_vdev_err(rvd,
3940 VDEV_AUX_CORRUPT_DATA, EIO));
3945 if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) {
3946 if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG,
3947 &spa->spa_feat_enabled_txg_obj, B_TRUE) != 0)
3948 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3952 * Encryption was added before bookmark_v2, even though bookmark_v2
3953 * is now a dependency. If this pool has encryption enabled without
3954 * bookmark_v2, trigger an errata message.
3956 if (spa_feature_is_enabled(spa, SPA_FEATURE_ENCRYPTION) &&
3957 !spa_feature_is_enabled(spa, SPA_FEATURE_BOOKMARK_V2)) {
3958 spa->spa_errata = ZPOOL_ERRATA_ZOL_8308_ENCRYPTION;
3965 spa_ld_load_special_directories(spa_t *spa)
3968 vdev_t *rvd = spa->spa_root_vdev;
3970 spa->spa_is_initializing = B_TRUE;
3971 error = dsl_pool_open(spa->spa_dsl_pool);
3972 spa->spa_is_initializing = B_FALSE;
3974 spa_load_failed(spa, "dsl_pool_open failed [error=%d]", error);
3975 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3982 spa_ld_get_props(spa_t *spa)
3986 vdev_t *rvd = spa->spa_root_vdev;
3988 /* Grab the checksum salt from the MOS. */
3989 error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
3990 DMU_POOL_CHECKSUM_SALT, 1,
3991 sizeof (spa->spa_cksum_salt.zcs_bytes),
3992 spa->spa_cksum_salt.zcs_bytes);
3993 if (error == ENOENT) {
3994 /* Generate a new salt for subsequent use */
3995 (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes,
3996 sizeof (spa->spa_cksum_salt.zcs_bytes));
3997 } else if (error != 0) {
3998 spa_load_failed(spa, "unable to retrieve checksum salt from "
3999 "MOS [error=%d]", error);
4000 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4003 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj, B_TRUE) != 0)
4004 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4005 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
4007 spa_load_failed(spa, "error opening deferred-frees bpobj "
4008 "[error=%d]", error);
4009 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4013 * Load the bit that tells us to use the new accounting function
4014 * (raid-z deflation). If we have an older pool, this will not
4017 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate, B_FALSE);
4018 if (error != 0 && error != ENOENT)
4019 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4021 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
4022 &spa->spa_creation_version, B_FALSE);
4023 if (error != 0 && error != ENOENT)
4024 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4027 * Load the persistent error log. If we have an older pool, this will
4030 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last,
4032 if (error != 0 && error != ENOENT)
4033 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4035 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
4036 &spa->spa_errlog_scrub, B_FALSE);
4037 if (error != 0 && error != ENOENT)
4038 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4041 * Load the livelist deletion field. If a livelist is queued for
4042 * deletion, indicate that in the spa
4044 error = spa_dir_prop(spa, DMU_POOL_DELETED_CLONES,
4045 &spa->spa_livelists_to_delete, B_FALSE);
4046 if (error != 0 && error != ENOENT)
4047 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4050 * Load the history object. If we have an older pool, this
4051 * will not be present.
4053 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history, B_FALSE);
4054 if (error != 0 && error != ENOENT)
4055 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4058 * Load the per-vdev ZAP map. If we have an older pool, this will not
4059 * be present; in this case, defer its creation to a later time to
4060 * avoid dirtying the MOS this early / out of sync context. See
4061 * spa_sync_config_object.
4064 /* The sentinel is only available in the MOS config. */
4065 nvlist_t *mos_config;
4066 if (load_nvlist(spa, spa->spa_config_object, &mos_config) != 0) {
4067 spa_load_failed(spa, "unable to retrieve MOS config");
4068 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4071 error = spa_dir_prop(spa, DMU_POOL_VDEV_ZAP_MAP,
4072 &spa->spa_all_vdev_zaps, B_FALSE);
4074 if (error == ENOENT) {
4075 VERIFY(!nvlist_exists(mos_config,
4076 ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS));
4077 spa->spa_avz_action = AVZ_ACTION_INITIALIZE;
4078 ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev));
4079 } else if (error != 0) {
4080 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4081 } else if (!nvlist_exists(mos_config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)) {
4083 * An older version of ZFS overwrote the sentinel value, so
4084 * we have orphaned per-vdev ZAPs in the MOS. Defer their
4085 * destruction to later; see spa_sync_config_object.
4087 spa->spa_avz_action = AVZ_ACTION_DESTROY;
4089 * We're assuming that no vdevs have had their ZAPs created
4090 * before this. Better be sure of it.
4092 ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev));
4094 nvlist_free(mos_config);
4096 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
4098 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object,
4100 if (error && error != ENOENT)
4101 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4104 uint64_t autoreplace;
4106 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
4107 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
4108 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
4109 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
4110 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
4111 spa_prop_find(spa, ZPOOL_PROP_MULTIHOST, &spa->spa_multihost);
4112 spa_prop_find(spa, ZPOOL_PROP_AUTOTRIM, &spa->spa_autotrim);
4113 spa->spa_autoreplace = (autoreplace != 0);
4117 * If we are importing a pool with missing top-level vdevs,
4118 * we enforce that the pool doesn't panic or get suspended on
4119 * error since the likelihood of missing data is extremely high.
4121 if (spa->spa_missing_tvds > 0 &&
4122 spa->spa_failmode != ZIO_FAILURE_MODE_CONTINUE &&
4123 spa->spa_load_state != SPA_LOAD_TRYIMPORT) {
4124 spa_load_note(spa, "forcing failmode to 'continue' "
4125 "as some top level vdevs are missing");
4126 spa->spa_failmode = ZIO_FAILURE_MODE_CONTINUE;
4133 spa_ld_open_aux_vdevs(spa_t *spa, spa_import_type_t type)
4136 vdev_t *rvd = spa->spa_root_vdev;
4139 * If we're assembling the pool from the split-off vdevs of
4140 * an existing pool, we don't want to attach the spares & cache
4145 * Load any hot spares for this pool.
4147 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object,
4149 if (error != 0 && error != ENOENT)
4150 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4151 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
4152 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
4153 if (load_nvlist(spa, spa->spa_spares.sav_object,
4154 &spa->spa_spares.sav_config) != 0) {
4155 spa_load_failed(spa, "error loading spares nvlist");
4156 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4159 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4160 spa_load_spares(spa);
4161 spa_config_exit(spa, SCL_ALL, FTAG);
4162 } else if (error == 0) {
4163 spa->spa_spares.sav_sync = B_TRUE;
4167 * Load any level 2 ARC devices for this pool.
4169 error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
4170 &spa->spa_l2cache.sav_object, B_FALSE);
4171 if (error != 0 && error != ENOENT)
4172 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4173 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
4174 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
4175 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
4176 &spa->spa_l2cache.sav_config) != 0) {
4177 spa_load_failed(spa, "error loading l2cache nvlist");
4178 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4181 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4182 spa_load_l2cache(spa);
4183 spa_config_exit(spa, SCL_ALL, FTAG);
4184 } else if (error == 0) {
4185 spa->spa_l2cache.sav_sync = B_TRUE;
4192 spa_ld_load_vdev_metadata(spa_t *spa)
4195 vdev_t *rvd = spa->spa_root_vdev;
4198 * If the 'multihost' property is set, then never allow a pool to
4199 * be imported when the system hostid is zero. The exception to
4200 * this rule is zdb which is always allowed to access pools.
4202 if (spa_multihost(spa) && spa_get_hostid(spa) == 0 &&
4203 (spa->spa_import_flags & ZFS_IMPORT_SKIP_MMP) == 0) {
4204 fnvlist_add_uint64(spa->spa_load_info,
4205 ZPOOL_CONFIG_MMP_STATE, MMP_STATE_NO_HOSTID);
4206 return (spa_vdev_err(rvd, VDEV_AUX_ACTIVE, EREMOTEIO));
4210 * If the 'autoreplace' property is set, then post a resource notifying
4211 * the ZFS DE that it should not issue any faults for unopenable
4212 * devices. We also iterate over the vdevs, and post a sysevent for any
4213 * unopenable vdevs so that the normal autoreplace handler can take
4216 if (spa->spa_autoreplace && spa->spa_load_state != SPA_LOAD_TRYIMPORT) {
4217 spa_check_removed(spa->spa_root_vdev);
4219 * For the import case, this is done in spa_import(), because
4220 * at this point we're using the spare definitions from
4221 * the MOS config, not necessarily from the userland config.
4223 if (spa->spa_load_state != SPA_LOAD_IMPORT) {
4224 spa_aux_check_removed(&spa->spa_spares);
4225 spa_aux_check_removed(&spa->spa_l2cache);
4230 * Load the vdev metadata such as metaslabs, DTLs, spacemap object, etc.
4232 error = vdev_load(rvd);
4234 spa_load_failed(spa, "vdev_load failed [error=%d]", error);
4235 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, error));
4238 error = spa_ld_log_spacemaps(spa);
4240 spa_load_failed(spa, "spa_ld_log_sm_data failed [error=%d]",
4242 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, error));
4246 * Propagate the leaf DTLs we just loaded all the way up the vdev tree.
4248 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4249 vdev_dtl_reassess(rvd, 0, 0, B_FALSE, B_FALSE);
4250 spa_config_exit(spa, SCL_ALL, FTAG);
4256 spa_ld_load_dedup_tables(spa_t *spa)
4259 vdev_t *rvd = spa->spa_root_vdev;
4261 error = ddt_load(spa);
4263 spa_load_failed(spa, "ddt_load failed [error=%d]", error);
4264 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4271 spa_ld_verify_logs(spa_t *spa, spa_import_type_t type, char **ereport)
4273 vdev_t *rvd = spa->spa_root_vdev;
4275 if (type != SPA_IMPORT_ASSEMBLE && spa_writeable(spa)) {
4276 boolean_t missing = spa_check_logs(spa);
4278 if (spa->spa_missing_tvds != 0) {
4279 spa_load_note(spa, "spa_check_logs failed "
4280 "so dropping the logs");
4282 *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
4283 spa_load_failed(spa, "spa_check_logs failed");
4284 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG,
4294 spa_ld_verify_pool_data(spa_t *spa)
4297 vdev_t *rvd = spa->spa_root_vdev;
4300 * We've successfully opened the pool, verify that we're ready
4301 * to start pushing transactions.
4303 if (spa->spa_load_state != SPA_LOAD_TRYIMPORT) {
4304 error = spa_load_verify(spa);
4306 spa_load_failed(spa, "spa_load_verify failed "
4307 "[error=%d]", error);
4308 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
4317 spa_ld_claim_log_blocks(spa_t *spa)
4320 dsl_pool_t *dp = spa_get_dsl(spa);
4323 * Claim log blocks that haven't been committed yet.
4324 * This must all happen in a single txg.
4325 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
4326 * invoked from zil_claim_log_block()'s i/o done callback.
4327 * Price of rollback is that we abandon the log.
4329 spa->spa_claiming = B_TRUE;
4331 tx = dmu_tx_create_assigned(dp, spa_first_txg(spa));
4332 (void) dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
4333 zil_claim, tx, DS_FIND_CHILDREN);
4336 spa->spa_claiming = B_FALSE;
4338 spa_set_log_state(spa, SPA_LOG_GOOD);
4342 spa_ld_check_for_config_update(spa_t *spa, uint64_t config_cache_txg,
4343 boolean_t update_config_cache)
4345 vdev_t *rvd = spa->spa_root_vdev;
4346 int need_update = B_FALSE;
4349 * If the config cache is stale, or we have uninitialized
4350 * metaslabs (see spa_vdev_add()), then update the config.
4352 * If this is a verbatim import, trust the current
4353 * in-core spa_config and update the disk labels.
4355 if (update_config_cache || config_cache_txg != spa->spa_config_txg ||
4356 spa->spa_load_state == SPA_LOAD_IMPORT ||
4357 spa->spa_load_state == SPA_LOAD_RECOVER ||
4358 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
4359 need_update = B_TRUE;
4361 for (int c = 0; c < rvd->vdev_children; c++)
4362 if (rvd->vdev_child[c]->vdev_ms_array == 0)
4363 need_update = B_TRUE;
4366 * Update the config cache asynchronously in case we're the
4367 * root pool, in which case the config cache isn't writable yet.
4370 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
4374 spa_ld_prepare_for_reload(spa_t *spa)
4376 spa_mode_t mode = spa->spa_mode;
4377 int async_suspended = spa->spa_async_suspended;
4380 spa_deactivate(spa);
4381 spa_activate(spa, mode);
4384 * We save the value of spa_async_suspended as it gets reset to 0 by
4385 * spa_unload(). We want to restore it back to the original value before
4386 * returning as we might be calling spa_async_resume() later.
4388 spa->spa_async_suspended = async_suspended;
4392 spa_ld_read_checkpoint_txg(spa_t *spa)
4394 uberblock_t checkpoint;
4397 ASSERT0(spa->spa_checkpoint_txg);
4398 ASSERT(MUTEX_HELD(&spa_namespace_lock));
4400 error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
4401 DMU_POOL_ZPOOL_CHECKPOINT, sizeof (uint64_t),
4402 sizeof (uberblock_t) / sizeof (uint64_t), &checkpoint);
4404 if (error == ENOENT)
4410 ASSERT3U(checkpoint.ub_txg, !=, 0);
4411 ASSERT3U(checkpoint.ub_checkpoint_txg, !=, 0);
4412 ASSERT3U(checkpoint.ub_timestamp, !=, 0);
4413 spa->spa_checkpoint_txg = checkpoint.ub_txg;
4414 spa->spa_checkpoint_info.sci_timestamp = checkpoint.ub_timestamp;
4420 spa_ld_mos_init(spa_t *spa, spa_import_type_t type)
4424 ASSERT(MUTEX_HELD(&spa_namespace_lock));
4425 ASSERT(spa->spa_config_source != SPA_CONFIG_SRC_NONE);
4428 * Never trust the config that is provided unless we are assembling
4429 * a pool following a split.
4430 * This means don't trust blkptrs and the vdev tree in general. This
4431 * also effectively puts the spa in read-only mode since
4432 * spa_writeable() checks for spa_trust_config to be true.
4433 * We will later load a trusted config from the MOS.
4435 if (type != SPA_IMPORT_ASSEMBLE)
4436 spa->spa_trust_config = B_FALSE;
4439 * Parse the config provided to create a vdev tree.
4441 error = spa_ld_parse_config(spa, type);
4445 spa_import_progress_add(spa);
4448 * Now that we have the vdev tree, try to open each vdev. This involves
4449 * opening the underlying physical device, retrieving its geometry and
4450 * probing the vdev with a dummy I/O. The state of each vdev will be set
4451 * based on the success of those operations. After this we'll be ready
4452 * to read from the vdevs.
4454 error = spa_ld_open_vdevs(spa);
4459 * Read the label of each vdev and make sure that the GUIDs stored
4460 * there match the GUIDs in the config provided.
4461 * If we're assembling a new pool that's been split off from an
4462 * existing pool, the labels haven't yet been updated so we skip
4463 * validation for now.
4465 if (type != SPA_IMPORT_ASSEMBLE) {
4466 error = spa_ld_validate_vdevs(spa);
4472 * Read all vdev labels to find the best uberblock (i.e. latest,
4473 * unless spa_load_max_txg is set) and store it in spa_uberblock. We
4474 * get the list of features required to read blkptrs in the MOS from
4475 * the vdev label with the best uberblock and verify that our version
4476 * of zfs supports them all.
4478 error = spa_ld_select_uberblock(spa, type);
4483 * Pass that uberblock to the dsl_pool layer which will open the root
4484 * blkptr. This blkptr points to the latest version of the MOS and will
4485 * allow us to read its contents.
4487 error = spa_ld_open_rootbp(spa);
4495 spa_ld_checkpoint_rewind(spa_t *spa)
4497 uberblock_t checkpoint;
4500 ASSERT(MUTEX_HELD(&spa_namespace_lock));
4501 ASSERT(spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT);
4503 error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
4504 DMU_POOL_ZPOOL_CHECKPOINT, sizeof (uint64_t),
4505 sizeof (uberblock_t) / sizeof (uint64_t), &checkpoint);
4508 spa_load_failed(spa, "unable to retrieve checkpointed "
4509 "uberblock from the MOS config [error=%d]", error);
4511 if (error == ENOENT)
4512 error = ZFS_ERR_NO_CHECKPOINT;
4517 ASSERT3U(checkpoint.ub_txg, <, spa->spa_uberblock.ub_txg);
4518 ASSERT3U(checkpoint.ub_txg, ==, checkpoint.ub_checkpoint_txg);
4521 * We need to update the txg and timestamp of the checkpointed
4522 * uberblock to be higher than the latest one. This ensures that
4523 * the checkpointed uberblock is selected if we were to close and
4524 * reopen the pool right after we've written it in the vdev labels.
4525 * (also see block comment in vdev_uberblock_compare)
4527 checkpoint.ub_txg = spa->spa_uberblock.ub_txg + 1;
4528 checkpoint.ub_timestamp = gethrestime_sec();
4531 * Set current uberblock to be the checkpointed uberblock.
4533 spa->spa_uberblock = checkpoint;
4536 * If we are doing a normal rewind, then the pool is open for
4537 * writing and we sync the "updated" checkpointed uberblock to
4538 * disk. Once this is done, we've basically rewound the whole
4539 * pool and there is no way back.
4541 * There are cases when we don't want to attempt and sync the
4542 * checkpointed uberblock to disk because we are opening a
4543 * pool as read-only. Specifically, verifying the checkpointed
4544 * state with zdb, and importing the checkpointed state to get
4545 * a "preview" of its content.
4547 if (spa_writeable(spa)) {
4548 vdev_t *rvd = spa->spa_root_vdev;
4550 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4551 vdev_t *svd[SPA_SYNC_MIN_VDEVS] = { NULL };
4553 int children = rvd->vdev_children;
4554 int c0 = spa_get_random(children);
4556 for (int c = 0; c < children; c++) {
4557 vdev_t *vd = rvd->vdev_child[(c0 + c) % children];
4559 /* Stop when revisiting the first vdev */
4560 if (c > 0 && svd[0] == vd)
4563 if (vd->vdev_ms_array == 0 || vd->vdev_islog ||
4564 !vdev_is_concrete(vd))
4567 svd[svdcount++] = vd;
4568 if (svdcount == SPA_SYNC_MIN_VDEVS)
4571 error = vdev_config_sync(svd, svdcount, spa->spa_first_txg);
4573 spa->spa_last_synced_guid = rvd->vdev_guid;
4574 spa_config_exit(spa, SCL_ALL, FTAG);
4577 spa_load_failed(spa, "failed to write checkpointed "
4578 "uberblock to the vdev labels [error=%d]", error);
4587 spa_ld_mos_with_trusted_config(spa_t *spa, spa_import_type_t type,
4588 boolean_t *update_config_cache)
4593 * Parse the config for pool, open and validate vdevs,
4594 * select an uberblock, and use that uberblock to open
4597 error = spa_ld_mos_init(spa, type);
4602 * Retrieve the trusted config stored in the MOS and use it to create
4603 * a new, exact version of the vdev tree, then reopen all vdevs.
4605 error = spa_ld_trusted_config(spa, type, B_FALSE);
4606 if (error == EAGAIN) {
4607 if (update_config_cache != NULL)
4608 *update_config_cache = B_TRUE;
4611 * Redo the loading process with the trusted config if it is
4612 * too different from the untrusted config.
4614 spa_ld_prepare_for_reload(spa);
4615 spa_load_note(spa, "RELOADING");
4616 error = spa_ld_mos_init(spa, type);
4620 error = spa_ld_trusted_config(spa, type, B_TRUE);
4624 } else if (error != 0) {
4632 * Load an existing storage pool, using the config provided. This config
4633 * describes which vdevs are part of the pool and is later validated against
4634 * partial configs present in each vdev's label and an entire copy of the
4635 * config stored in the MOS.
4638 spa_load_impl(spa_t *spa, spa_import_type_t type, char **ereport)
4641 boolean_t missing_feat_write = B_FALSE;
4642 boolean_t checkpoint_rewind =
4643 (spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT);
4644 boolean_t update_config_cache = B_FALSE;
4646 ASSERT(MUTEX_HELD(&spa_namespace_lock));
4647 ASSERT(spa->spa_config_source != SPA_CONFIG_SRC_NONE);
4649 spa_load_note(spa, "LOADING");
4651 error = spa_ld_mos_with_trusted_config(spa, type, &update_config_cache);
4656 * If we are rewinding to the checkpoint then we need to repeat
4657 * everything we've done so far in this function but this time
4658 * selecting the checkpointed uberblock and using that to open
4661 if (checkpoint_rewind) {
4663 * If we are rewinding to the checkpoint update config cache
4666 update_config_cache = B_TRUE;
4669 * Extract the checkpointed uberblock from the current MOS
4670 * and use this as the pool's uberblock from now on. If the
4671 * pool is imported as writeable we also write the checkpoint
4672 * uberblock to the labels, making the rewind permanent.
4674 error = spa_ld_checkpoint_rewind(spa);
4679 * Redo the loading process again with the
4680 * checkpointed uberblock.
4682 spa_ld_prepare_for_reload(spa);
4683 spa_load_note(spa, "LOADING checkpointed uberblock");
4684 error = spa_ld_mos_with_trusted_config(spa, type, NULL);
4690 * Retrieve the checkpoint txg if the pool has a checkpoint.
4692 error = spa_ld_read_checkpoint_txg(spa);
4697 * Retrieve the mapping of indirect vdevs. Those vdevs were removed
4698 * from the pool and their contents were re-mapped to other vdevs. Note
4699 * that everything that we read before this step must have been
4700 * rewritten on concrete vdevs after the last device removal was
4701 * initiated. Otherwise we could be reading from indirect vdevs before
4702 * we have loaded their mappings.
4704 error = spa_ld_open_indirect_vdev_metadata(spa);
4709 * Retrieve the full list of active features from the MOS and check if
4710 * they are all supported.
4712 error = spa_ld_check_features(spa, &missing_feat_write);
4717 * Load several special directories from the MOS needed by the dsl_pool
4720 error = spa_ld_load_special_directories(spa);
4725 * Retrieve pool properties from the MOS.
4727 error = spa_ld_get_props(spa);
4732 * Retrieve the list of auxiliary devices - cache devices and spares -
4735 error = spa_ld_open_aux_vdevs(spa, type);
4740 * Load the metadata for all vdevs. Also check if unopenable devices
4741 * should be autoreplaced.
4743 error = spa_ld_load_vdev_metadata(spa);
4747 error = spa_ld_load_dedup_tables(spa);
4752 * Verify the logs now to make sure we don't have any unexpected errors
4753 * when we claim log blocks later.
4755 error = spa_ld_verify_logs(spa, type, ereport);
4759 if (missing_feat_write) {
4760 ASSERT(spa->spa_load_state == SPA_LOAD_TRYIMPORT);
4763 * At this point, we know that we can open the pool in
4764 * read-only mode but not read-write mode. We now have enough
4765 * information and can return to userland.
4767 return (spa_vdev_err(spa->spa_root_vdev, VDEV_AUX_UNSUP_FEAT,
4772 * Traverse the last txgs to make sure the pool was left off in a safe
4773 * state. When performing an extreme rewind, we verify the whole pool,
4774 * which can take a very long time.
4776 error = spa_ld_verify_pool_data(spa);
4781 * Calculate the deflated space for the pool. This must be done before
4782 * we write anything to the pool because we'd need to update the space
4783 * accounting using the deflated sizes.
4785 spa_update_dspace(spa);
4788 * We have now retrieved all the information we needed to open the
4789 * pool. If we are importing the pool in read-write mode, a few
4790 * additional steps must be performed to finish the import.
4792 if (spa_writeable(spa) && (spa->spa_load_state == SPA_LOAD_RECOVER ||
4793 spa->spa_load_max_txg == UINT64_MAX)) {
4794 uint64_t config_cache_txg = spa->spa_config_txg;
4796 ASSERT(spa->spa_load_state != SPA_LOAD_TRYIMPORT);
4799 * In case of a checkpoint rewind, log the original txg
4800 * of the checkpointed uberblock.
4802 if (checkpoint_rewind) {
4803 spa_history_log_internal(spa, "checkpoint rewind",
4804 NULL, "rewound state to txg=%llu",
4805 (u_longlong_t)spa->spa_uberblock.ub_checkpoint_txg);
4809 * Traverse the ZIL and claim all blocks.
4811 spa_ld_claim_log_blocks(spa);
4814 * Kick-off the syncing thread.
4816 spa->spa_sync_on = B_TRUE;
4817 txg_sync_start(spa->spa_dsl_pool);
4818 mmp_thread_start(spa);
4821 * Wait for all claims to sync. We sync up to the highest
4822 * claimed log block birth time so that claimed log blocks
4823 * don't appear to be from the future. spa_claim_max_txg
4824 * will have been set for us by ZIL traversal operations
4827 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
4830 * Check if we need to request an update of the config. On the
4831 * next sync, we would update the config stored in vdev labels
4832 * and the cachefile (by default /etc/zfs/zpool.cache).
4834 spa_ld_check_for_config_update(spa, config_cache_txg,
4835 update_config_cache);
4838 * Check if a rebuild was in progress and if so resume it.
4839 * Then check all DTLs to see if anything needs resilvering.
4840 * The resilver will be deferred if a rebuild was started.
4842 if (vdev_rebuild_active(spa->spa_root_vdev)) {
4843 vdev_rebuild_restart(spa);
4844 } else if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
4845 vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
4846 spa_async_request(spa, SPA_ASYNC_RESILVER);
4850 * Log the fact that we booted up (so that we can detect if
4851 * we rebooted in the middle of an operation).
4853 spa_history_log_version(spa, "open", NULL);
4855 spa_restart_removal(spa);
4856 spa_spawn_aux_threads(spa);
4859 * Delete any inconsistent datasets.
4862 * Since we may be issuing deletes for clones here,
4863 * we make sure to do so after we've spawned all the
4864 * auxiliary threads above (from which the livelist
4865 * deletion zthr is part of).
4867 (void) dmu_objset_find(spa_name(spa),
4868 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
4871 * Clean up any stale temporary dataset userrefs.
4873 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
4875 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4876 vdev_initialize_restart(spa->spa_root_vdev);
4877 vdev_trim_restart(spa->spa_root_vdev);
4878 vdev_autotrim_restart(spa);
4879 spa_config_exit(spa, SCL_CONFIG, FTAG);
4882 spa_import_progress_remove(spa_guid(spa));
4883 spa_async_request(spa, SPA_ASYNC_L2CACHE_REBUILD);
4885 spa_load_note(spa, "LOADED");
4891 spa_load_retry(spa_t *spa, spa_load_state_t state)
4893 spa_mode_t mode = spa->spa_mode;
4896 spa_deactivate(spa);
4898 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg - 1;
4900 spa_activate(spa, mode);
4901 spa_async_suspend(spa);
4903 spa_load_note(spa, "spa_load_retry: rewind, max txg: %llu",
4904 (u_longlong_t)spa->spa_load_max_txg);
4906 return (spa_load(spa, state, SPA_IMPORT_EXISTING));
4910 * If spa_load() fails this function will try loading prior txg's. If
4911 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
4912 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
4913 * function will not rewind the pool and will return the same error as
4917 spa_load_best(spa_t *spa, spa_load_state_t state, uint64_t max_request,
4920 nvlist_t *loadinfo = NULL;
4921 nvlist_t *config = NULL;
4922 int load_error, rewind_error;
4923 uint64_t safe_rewind_txg;
4926 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
4927 spa->spa_load_max_txg = spa->spa_load_txg;
4928 spa_set_log_state(spa, SPA_LOG_CLEAR);
4930 spa->spa_load_max_txg = max_request;
4931 if (max_request != UINT64_MAX)
4932 spa->spa_extreme_rewind = B_TRUE;
4935 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING);
4936 if (load_error == 0)
4938 if (load_error == ZFS_ERR_NO_CHECKPOINT) {
4940 * When attempting checkpoint-rewind on a pool with no
4941 * checkpoint, we should not attempt to load uberblocks
4942 * from previous txgs when spa_load fails.
4944 ASSERT(spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT);
4945 spa_import_progress_remove(spa_guid(spa));
4946 return (load_error);
4949 if (spa->spa_root_vdev != NULL)
4950 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4952 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
4953 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
4955 if (rewind_flags & ZPOOL_NEVER_REWIND) {
4956 nvlist_free(config);
4957 spa_import_progress_remove(spa_guid(spa));
4958 return (load_error);
4961 if (state == SPA_LOAD_RECOVER) {
4962 /* Price of rolling back is discarding txgs, including log */
4963 spa_set_log_state(spa, SPA_LOG_CLEAR);
4966 * If we aren't rolling back save the load info from our first
4967 * import attempt so that we can restore it after attempting
4970 loadinfo = spa->spa_load_info;
4971 spa->spa_load_info = fnvlist_alloc();
4974 spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
4975 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
4976 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
4977 TXG_INITIAL : safe_rewind_txg;
4980 * Continue as long as we're finding errors, we're still within
4981 * the acceptable rewind range, and we're still finding uberblocks
4983 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
4984 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
4985 if (spa->spa_load_max_txg < safe_rewind_txg)
4986 spa->spa_extreme_rewind = B_TRUE;
4987 rewind_error = spa_load_retry(spa, state);
4990 spa->spa_extreme_rewind = B_FALSE;
4991 spa->spa_load_max_txg = UINT64_MAX;
4993 if (config && (rewind_error || state != SPA_LOAD_RECOVER))
4994 spa_config_set(spa, config);
4996 nvlist_free(config);
4998 if (state == SPA_LOAD_RECOVER) {
4999 ASSERT3P(loadinfo, ==, NULL);
5000 spa_import_progress_remove(spa_guid(spa));
5001 return (rewind_error);
5003 /* Store the rewind info as part of the initial load info */
5004 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
5005 spa->spa_load_info);
5007 /* Restore the initial load info */
5008 fnvlist_free(spa->spa_load_info);
5009 spa->spa_load_info = loadinfo;
5011 spa_import_progress_remove(spa_guid(spa));
5012 return (load_error);
5019 * The import case is identical to an open except that the configuration is sent
5020 * down from userland, instead of grabbed from the configuration cache. For the
5021 * case of an open, the pool configuration will exist in the
5022 * POOL_STATE_UNINITIALIZED state.
5024 * The stats information (gen/count/ustats) is used to gather vdev statistics at
5025 * the same time open the pool, without having to keep around the spa_t in some
5029 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
5033 spa_load_state_t state = SPA_LOAD_OPEN;
5035 int locked = B_FALSE;
5036 int firstopen = B_FALSE;
5041 * As disgusting as this is, we need to support recursive calls to this
5042 * function because dsl_dir_open() is called during spa_load(), and ends
5043 * up calling spa_open() again. The real fix is to figure out how to
5044 * avoid dsl_dir_open() calling this in the first place.
5046 if (MUTEX_NOT_HELD(&spa_namespace_lock)) {
5047 mutex_enter(&spa_namespace_lock);
5051 if ((spa = spa_lookup(pool)) == NULL) {
5053 mutex_exit(&spa_namespace_lock);
5054 return (SET_ERROR(ENOENT));
5057 if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
5058 zpool_load_policy_t policy;
5062 zpool_get_load_policy(nvpolicy ? nvpolicy : spa->spa_config,
5064 if (policy.zlp_rewind & ZPOOL_DO_REWIND)
5065 state = SPA_LOAD_RECOVER;
5067 spa_activate(spa, spa_mode_global);
5069 if (state != SPA_LOAD_RECOVER)
5070 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
5071 spa->spa_config_source = SPA_CONFIG_SRC_CACHEFILE;
5073 zfs_dbgmsg("spa_open_common: opening %s", pool);
5074 error = spa_load_best(spa, state, policy.zlp_txg,
5077 if (error == EBADF) {
5079 * If vdev_validate() returns failure (indicated by
5080 * EBADF), it indicates that one of the vdevs indicates
5081 * that the pool has been exported or destroyed. If
5082 * this is the case, the config cache is out of sync and
5083 * we should remove the pool from the namespace.
5086 spa_deactivate(spa);
5087 spa_write_cachefile(spa, B_TRUE, B_TRUE);
5090 mutex_exit(&spa_namespace_lock);
5091 return (SET_ERROR(ENOENT));
5096 * We can't open the pool, but we still have useful
5097 * information: the state of each vdev after the
5098 * attempted vdev_open(). Return this to the user.
5100 if (config != NULL && spa->spa_config) {
5101 VERIFY(nvlist_dup(spa->spa_config, config,
5103 VERIFY(nvlist_add_nvlist(*config,
5104 ZPOOL_CONFIG_LOAD_INFO,
5105 spa->spa_load_info) == 0);
5108 spa_deactivate(spa);
5109 spa->spa_last_open_failed = error;
5111 mutex_exit(&spa_namespace_lock);
5117 spa_open_ref(spa, tag);
5120 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
5123 * If we've recovered the pool, pass back any information we
5124 * gathered while doing the load.
5126 if (state == SPA_LOAD_RECOVER) {
5127 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
5128 spa->spa_load_info) == 0);
5132 spa->spa_last_open_failed = 0;
5133 spa->spa_last_ubsync_txg = 0;
5134 spa->spa_load_txg = 0;
5135 mutex_exit(&spa_namespace_lock);
5139 zvol_create_minors_recursive(spa_name(spa));
5147 spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
5150 return (spa_open_common(name, spapp, tag, policy, config));
5154 spa_open(const char *name, spa_t **spapp, void *tag)
5156 return (spa_open_common(name, spapp, tag, NULL, NULL));
5160 * Lookup the given spa_t, incrementing the inject count in the process,
5161 * preventing it from being exported or destroyed.
5164 spa_inject_addref(char *name)
5168 mutex_enter(&spa_namespace_lock);
5169 if ((spa = spa_lookup(name)) == NULL) {
5170 mutex_exit(&spa_namespace_lock);
5173 spa->spa_inject_ref++;
5174 mutex_exit(&spa_namespace_lock);
5180 spa_inject_delref(spa_t *spa)
5182 mutex_enter(&spa_namespace_lock);
5183 spa->spa_inject_ref--;
5184 mutex_exit(&spa_namespace_lock);
5188 * Add spares device information to the nvlist.
5191 spa_add_spares(spa_t *spa, nvlist_t *config)
5201 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
5203 if (spa->spa_spares.sav_count == 0)
5206 VERIFY(nvlist_lookup_nvlist(config,
5207 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
5208 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5209 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
5211 VERIFY(nvlist_add_nvlist_array(nvroot,
5212 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
5213 VERIFY(nvlist_lookup_nvlist_array(nvroot,
5214 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
5217 * Go through and find any spares which have since been
5218 * repurposed as an active spare. If this is the case, update
5219 * their status appropriately.
5221 for (i = 0; i < nspares; i++) {
5222 VERIFY(nvlist_lookup_uint64(spares[i],
5223 ZPOOL_CONFIG_GUID, &guid) == 0);
5224 if (spa_spare_exists(guid, &pool, NULL) &&
5226 VERIFY(nvlist_lookup_uint64_array(
5227 spares[i], ZPOOL_CONFIG_VDEV_STATS,
5228 (uint64_t **)&vs, &vsc) == 0);
5229 vs->vs_state = VDEV_STATE_CANT_OPEN;
5230 vs->vs_aux = VDEV_AUX_SPARED;
5237 * Add l2cache device information to the nvlist, including vdev stats.
5240 spa_add_l2cache(spa_t *spa, nvlist_t *config)
5243 uint_t i, j, nl2cache;
5250 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
5252 if (spa->spa_l2cache.sav_count == 0)
5255 VERIFY(nvlist_lookup_nvlist(config,
5256 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
5257 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5258 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
5259 if (nl2cache != 0) {
5260 VERIFY(nvlist_add_nvlist_array(nvroot,
5261 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
5262 VERIFY(nvlist_lookup_nvlist_array(nvroot,
5263 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
5266 * Update level 2 cache device stats.
5269 for (i = 0; i < nl2cache; i++) {
5270 VERIFY(nvlist_lookup_uint64(l2cache[i],
5271 ZPOOL_CONFIG_GUID, &guid) == 0);
5274 for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
5276 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
5277 vd = spa->spa_l2cache.sav_vdevs[j];
5283 VERIFY(nvlist_lookup_uint64_array(l2cache[i],
5284 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
5286 vdev_get_stats(vd, vs);
5287 vdev_config_generate_stats(vd, l2cache[i]);
5294 spa_feature_stats_from_disk(spa_t *spa, nvlist_t *features)
5299 if (spa->spa_feat_for_read_obj != 0) {
5300 for (zap_cursor_init(&zc, spa->spa_meta_objset,
5301 spa->spa_feat_for_read_obj);
5302 zap_cursor_retrieve(&zc, &za) == 0;
5303 zap_cursor_advance(&zc)) {
5304 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
5305 za.za_num_integers == 1);
5306 VERIFY0(nvlist_add_uint64(features, za.za_name,
5307 za.za_first_integer));
5309 zap_cursor_fini(&zc);
5312 if (spa->spa_feat_for_write_obj != 0) {
5313 for (zap_cursor_init(&zc, spa->spa_meta_objset,
5314 spa->spa_feat_for_write_obj);
5315 zap_cursor_retrieve(&zc, &za) == 0;
5316 zap_cursor_advance(&zc)) {
5317 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
5318 za.za_num_integers == 1);
5319 VERIFY0(nvlist_add_uint64(features, za.za_name,
5320 za.za_first_integer));
5322 zap_cursor_fini(&zc);
5327 spa_feature_stats_from_cache(spa_t *spa, nvlist_t *features)
5331 for (i = 0; i < SPA_FEATURES; i++) {
5332 zfeature_info_t feature = spa_feature_table[i];
5335 if (feature_get_refcount(spa, &feature, &refcount) != 0)
5338 VERIFY0(nvlist_add_uint64(features, feature.fi_guid, refcount));
5343 * Store a list of pool features and their reference counts in the
5346 * The first time this is called on a spa, allocate a new nvlist, fetch
5347 * the pool features and reference counts from disk, then save the list
5348 * in the spa. In subsequent calls on the same spa use the saved nvlist
5349 * and refresh its values from the cached reference counts. This
5350 * ensures we don't block here on I/O on a suspended pool so 'zpool
5351 * clear' can resume the pool.
5354 spa_add_feature_stats(spa_t *spa, nvlist_t *config)
5358 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
5360 mutex_enter(&spa->spa_feat_stats_lock);
5361 features = spa->spa_feat_stats;
5363 if (features != NULL) {
5364 spa_feature_stats_from_cache(spa, features);
5366 VERIFY0(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP));
5367 spa->spa_feat_stats = features;
5368 spa_feature_stats_from_disk(spa, features);
5371 VERIFY0(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
5374 mutex_exit(&spa->spa_feat_stats_lock);
5378 spa_get_stats(const char *name, nvlist_t **config,
5379 char *altroot, size_t buflen)
5385 error = spa_open_common(name, &spa, FTAG, NULL, config);
5389 * This still leaves a window of inconsistency where the spares
5390 * or l2cache devices could change and the config would be
5391 * self-inconsistent.
5393 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5395 if (*config != NULL) {
5396 uint64_t loadtimes[2];
5398 loadtimes[0] = spa->spa_loaded_ts.tv_sec;
5399 loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
5400 VERIFY(nvlist_add_uint64_array(*config,
5401 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
5403 VERIFY(nvlist_add_uint64(*config,
5404 ZPOOL_CONFIG_ERRCOUNT,
5405 spa_get_errlog_size(spa)) == 0);
5407 if (spa_suspended(spa)) {
5408 VERIFY(nvlist_add_uint64(*config,
5409 ZPOOL_CONFIG_SUSPENDED,
5410 spa->spa_failmode) == 0);
5411 VERIFY(nvlist_add_uint64(*config,
5412 ZPOOL_CONFIG_SUSPENDED_REASON,
5413 spa->spa_suspended) == 0);
5416 spa_add_spares(spa, *config);
5417 spa_add_l2cache(spa, *config);
5418 spa_add_feature_stats(spa, *config);
5423 * We want to get the alternate root even for faulted pools, so we cheat
5424 * and call spa_lookup() directly.
5428 mutex_enter(&spa_namespace_lock);
5429 spa = spa_lookup(name);
5431 spa_altroot(spa, altroot, buflen);
5435 mutex_exit(&spa_namespace_lock);
5437 spa_altroot(spa, altroot, buflen);
5442 spa_config_exit(spa, SCL_CONFIG, FTAG);
5443 spa_close(spa, FTAG);
5450 * Validate that the auxiliary device array is well formed. We must have an
5451 * array of nvlists, each which describes a valid leaf vdev. If this is an
5452 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
5453 * specified, as long as they are well-formed.
5456 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
5457 spa_aux_vdev_t *sav, const char *config, uint64_t version,
5458 vdev_labeltype_t label)
5465 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5468 * It's acceptable to have no devs specified.
5470 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
5474 return (SET_ERROR(EINVAL));
5477 * Make sure the pool is formatted with a version that supports this
5480 if (spa_version(spa) < version)
5481 return (SET_ERROR(ENOTSUP));
5484 * Set the pending device list so we correctly handle device in-use
5487 sav->sav_pending = dev;
5488 sav->sav_npending = ndev;
5490 for (i = 0; i < ndev; i++) {
5491 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
5495 if (!vd->vdev_ops->vdev_op_leaf) {
5497 error = SET_ERROR(EINVAL);
5503 if ((error = vdev_open(vd)) == 0 &&
5504 (error = vdev_label_init(vd, crtxg, label)) == 0) {
5505 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
5506 vd->vdev_guid) == 0);
5512 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
5519 sav->sav_pending = NULL;
5520 sav->sav_npending = 0;
5525 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
5529 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5531 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
5532 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
5533 VDEV_LABEL_SPARE)) != 0) {
5537 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
5538 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
5539 VDEV_LABEL_L2CACHE));
5543 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
5548 if (sav->sav_config != NULL) {
5554 * Generate new dev list by concatenating with the
5557 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
5558 &olddevs, &oldndevs) == 0);
5560 newdevs = kmem_alloc(sizeof (void *) *
5561 (ndevs + oldndevs), KM_SLEEP);
5562 for (i = 0; i < oldndevs; i++)
5563 VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
5565 for (i = 0; i < ndevs; i++)
5566 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
5569 VERIFY(nvlist_remove(sav->sav_config, config,
5570 DATA_TYPE_NVLIST_ARRAY) == 0);
5572 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
5573 config, newdevs, ndevs + oldndevs) == 0);
5574 for (i = 0; i < oldndevs + ndevs; i++)
5575 nvlist_free(newdevs[i]);
5576 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
5579 * Generate a new dev list.
5581 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
5583 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
5589 * Stop and drop level 2 ARC devices
5592 spa_l2cache_drop(spa_t *spa)
5596 spa_aux_vdev_t *sav = &spa->spa_l2cache;
5598 for (i = 0; i < sav->sav_count; i++) {
5601 vd = sav->sav_vdevs[i];
5604 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
5605 pool != 0ULL && l2arc_vdev_present(vd))
5606 l2arc_remove_vdev(vd);
5611 * Verify encryption parameters for spa creation. If we are encrypting, we must
5612 * have the encryption feature flag enabled.
5615 spa_create_check_encryption_params(dsl_crypto_params_t *dcp,
5616 boolean_t has_encryption)
5618 if (dcp->cp_crypt != ZIO_CRYPT_OFF &&
5619 dcp->cp_crypt != ZIO_CRYPT_INHERIT &&
5621 return (SET_ERROR(ENOTSUP));
5623 return (dmu_objset_create_crypt_check(NULL, dcp, NULL));
5630 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
5631 nvlist_t *zplprops, dsl_crypto_params_t *dcp)
5634 char *altroot = NULL;
5639 uint64_t txg = TXG_INITIAL;
5640 nvlist_t **spares, **l2cache;
5641 uint_t nspares, nl2cache;
5642 uint64_t version, obj, ndraid = 0;
5643 boolean_t has_features;
5644 boolean_t has_encryption;
5645 boolean_t has_allocclass;
5651 if (props == NULL ||
5652 nvlist_lookup_string(props, "tname", &poolname) != 0)
5653 poolname = (char *)pool;
5656 * If this pool already exists, return failure.
5658 mutex_enter(&spa_namespace_lock);
5659 if (spa_lookup(poolname) != NULL) {
5660 mutex_exit(&spa_namespace_lock);
5661 return (SET_ERROR(EEXIST));
5665 * Allocate a new spa_t structure.
5667 nvl = fnvlist_alloc();
5668 fnvlist_add_string(nvl, ZPOOL_CONFIG_POOL_NAME, pool);
5669 (void) nvlist_lookup_string(props,
5670 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
5671 spa = spa_add(poolname, nvl, altroot);
5673 spa_activate(spa, spa_mode_global);
5675 if (props && (error = spa_prop_validate(spa, props))) {
5676 spa_deactivate(spa);
5678 mutex_exit(&spa_namespace_lock);
5683 * Temporary pool names should never be written to disk.
5685 if (poolname != pool)
5686 spa->spa_import_flags |= ZFS_IMPORT_TEMP_NAME;
5688 has_features = B_FALSE;
5689 has_encryption = B_FALSE;
5690 has_allocclass = B_FALSE;
5691 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
5692 elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
5693 if (zpool_prop_feature(nvpair_name(elem))) {
5694 has_features = B_TRUE;
5696 feat_name = strchr(nvpair_name(elem), '@') + 1;
5697 VERIFY0(zfeature_lookup_name(feat_name, &feat));
5698 if (feat == SPA_FEATURE_ENCRYPTION)
5699 has_encryption = B_TRUE;
5700 if (feat == SPA_FEATURE_ALLOCATION_CLASSES)
5701 has_allocclass = B_TRUE;
5705 /* verify encryption params, if they were provided */
5707 error = spa_create_check_encryption_params(dcp, has_encryption);
5709 spa_deactivate(spa);
5711 mutex_exit(&spa_namespace_lock);
5715 if (!has_allocclass && zfs_special_devs(nvroot, NULL)) {
5716 spa_deactivate(spa);
5718 mutex_exit(&spa_namespace_lock);
5722 if (has_features || nvlist_lookup_uint64(props,
5723 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
5724 version = SPA_VERSION;
5726 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
5728 spa->spa_first_txg = txg;
5729 spa->spa_uberblock.ub_txg = txg - 1;
5730 spa->spa_uberblock.ub_version = version;
5731 spa->spa_ubsync = spa->spa_uberblock;
5732 spa->spa_load_state = SPA_LOAD_CREATE;
5733 spa->spa_removing_phys.sr_state = DSS_NONE;
5734 spa->spa_removing_phys.sr_removing_vdev = -1;
5735 spa->spa_removing_phys.sr_prev_indirect_vdev = -1;
5736 spa->spa_indirect_vdevs_loaded = B_TRUE;
5739 * Create "The Godfather" zio to hold all async IOs
5741 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
5743 for (int i = 0; i < max_ncpus; i++) {
5744 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
5745 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
5746 ZIO_FLAG_GODFATHER);
5750 * Create the root vdev.
5752 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5754 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
5756 ASSERT(error != 0 || rvd != NULL);
5757 ASSERT(error != 0 || spa->spa_root_vdev == rvd);
5759 if (error == 0 && !zfs_allocatable_devs(nvroot))
5760 error = SET_ERROR(EINVAL);
5763 (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
5764 (error = vdev_draid_spare_create(nvroot, rvd, &ndraid, 0)) == 0 &&
5765 (error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) == 0) {
5767 * instantiate the metaslab groups (this will dirty the vdevs)
5768 * we can no longer error exit past this point
5770 for (int c = 0; error == 0 && c < rvd->vdev_children; c++) {
5771 vdev_t *vd = rvd->vdev_child[c];
5773 vdev_metaslab_set_size(vd);
5774 vdev_expand(vd, txg);
5778 spa_config_exit(spa, SCL_ALL, FTAG);
5782 spa_deactivate(spa);
5784 mutex_exit(&spa_namespace_lock);
5789 * Get the list of spares, if specified.
5791 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
5792 &spares, &nspares) == 0) {
5793 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
5795 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
5796 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
5797 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5798 spa_load_spares(spa);
5799 spa_config_exit(spa, SCL_ALL, FTAG);
5800 spa->spa_spares.sav_sync = B_TRUE;
5804 * Get the list of level 2 cache devices, if specified.
5806 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
5807 &l2cache, &nl2cache) == 0) {
5808 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
5809 NV_UNIQUE_NAME, KM_SLEEP) == 0);
5810 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
5811 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
5812 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5813 spa_load_l2cache(spa);
5814 spa_config_exit(spa, SCL_ALL, FTAG);
5815 spa->spa_l2cache.sav_sync = B_TRUE;
5818 spa->spa_is_initializing = B_TRUE;
5819 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, dcp, txg);
5820 spa->spa_is_initializing = B_FALSE;
5823 * Create DDTs (dedup tables).
5827 spa_update_dspace(spa);
5829 tx = dmu_tx_create_assigned(dp, txg);
5832 * Create the pool's history object.
5834 if (version >= SPA_VERSION_ZPOOL_HISTORY && !spa->spa_history)
5835 spa_history_create_obj(spa, tx);
5837 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_CREATE);
5838 spa_history_log_version(spa, "create", tx);
5841 * Create the pool config object.
5843 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
5844 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
5845 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
5847 if (zap_add(spa->spa_meta_objset,
5848 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
5849 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
5850 cmn_err(CE_PANIC, "failed to add pool config");
5853 if (zap_add(spa->spa_meta_objset,
5854 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
5855 sizeof (uint64_t), 1, &version, tx) != 0) {
5856 cmn_err(CE_PANIC, "failed to add pool version");
5859 /* Newly created pools with the right version are always deflated. */
5860 if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
5861 spa->spa_deflate = TRUE;
5862 if (zap_add(spa->spa_meta_objset,
5863 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
5864 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
5865 cmn_err(CE_PANIC, "failed to add deflate");
5870 * Create the deferred-free bpobj. Turn off compression
5871 * because sync-to-convergence takes longer if the blocksize
5874 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
5875 dmu_object_set_compress(spa->spa_meta_objset, obj,
5876 ZIO_COMPRESS_OFF, tx);
5877 if (zap_add(spa->spa_meta_objset,
5878 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
5879 sizeof (uint64_t), 1, &obj, tx) != 0) {
5880 cmn_err(CE_PANIC, "failed to add bpobj");
5882 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
5883 spa->spa_meta_objset, obj));
5886 * Generate some random noise for salted checksums to operate on.
5888 (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes,
5889 sizeof (spa->spa_cksum_salt.zcs_bytes));
5892 * Set pool properties.
5894 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
5895 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
5896 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
5897 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
5898 spa->spa_multihost = zpool_prop_default_numeric(ZPOOL_PROP_MULTIHOST);
5899 spa->spa_autotrim = zpool_prop_default_numeric(ZPOOL_PROP_AUTOTRIM);
5901 if (props != NULL) {
5902 spa_configfile_set(spa, props, B_FALSE);
5903 spa_sync_props(props, tx);
5906 for (int i = 0; i < ndraid; i++)
5907 spa_feature_incr(spa, SPA_FEATURE_DRAID, tx);
5911 spa->spa_sync_on = B_TRUE;
5913 mmp_thread_start(spa);
5914 txg_wait_synced(dp, txg);
5916 spa_spawn_aux_threads(spa);
5918 spa_write_cachefile(spa, B_FALSE, B_TRUE);
5921 * Don't count references from objsets that are already closed
5922 * and are making their way through the eviction process.
5924 spa_evicting_os_wait(spa);
5925 spa->spa_minref = zfs_refcount_count(&spa->spa_refcount);
5926 spa->spa_load_state = SPA_LOAD_NONE;
5928 mutex_exit(&spa_namespace_lock);
5934 * Import a non-root pool into the system.
5937 spa_import(char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
5940 char *altroot = NULL;
5941 spa_load_state_t state = SPA_LOAD_IMPORT;
5942 zpool_load_policy_t policy;
5943 spa_mode_t mode = spa_mode_global;
5944 uint64_t readonly = B_FALSE;
5947 nvlist_t **spares, **l2cache;
5948 uint_t nspares, nl2cache;
5951 * If a pool with this name exists, return failure.
5953 mutex_enter(&spa_namespace_lock);
5954 if (spa_lookup(pool) != NULL) {
5955 mutex_exit(&spa_namespace_lock);
5956 return (SET_ERROR(EEXIST));
5960 * Create and initialize the spa structure.
5962 (void) nvlist_lookup_string(props,
5963 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
5964 (void) nvlist_lookup_uint64(props,
5965 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
5967 mode = SPA_MODE_READ;
5968 spa = spa_add(pool, config, altroot);
5969 spa->spa_import_flags = flags;
5972 * Verbatim import - Take a pool and insert it into the namespace
5973 * as if it had been loaded at boot.
5975 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
5977 spa_configfile_set(spa, props, B_FALSE);
5979 spa_write_cachefile(spa, B_FALSE, B_TRUE);
5980 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_IMPORT);
5981 zfs_dbgmsg("spa_import: verbatim import of %s", pool);
5982 mutex_exit(&spa_namespace_lock);
5986 spa_activate(spa, mode);
5989 * Don't start async tasks until we know everything is healthy.
5991 spa_async_suspend(spa);
5993 zpool_get_load_policy(config, &policy);
5994 if (policy.zlp_rewind & ZPOOL_DO_REWIND)
5995 state = SPA_LOAD_RECOVER;
5997 spa->spa_config_source = SPA_CONFIG_SRC_TRYIMPORT;
5999 if (state != SPA_LOAD_RECOVER) {
6000 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
6001 zfs_dbgmsg("spa_import: importing %s", pool);
6003 zfs_dbgmsg("spa_import: importing %s, max_txg=%lld "
6004 "(RECOVERY MODE)", pool, (longlong_t)policy.zlp_txg);
6006 error = spa_load_best(spa, state, policy.zlp_txg, policy.zlp_rewind);
6009 * Propagate anything learned while loading the pool and pass it
6010 * back to caller (i.e. rewind info, missing devices, etc).
6012 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
6013 spa->spa_load_info) == 0);
6015 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6017 * Toss any existing sparelist, as it doesn't have any validity
6018 * anymore, and conflicts with spa_has_spare().
6020 if (spa->spa_spares.sav_config) {
6021 nvlist_free(spa->spa_spares.sav_config);
6022 spa->spa_spares.sav_config = NULL;
6023 spa_load_spares(spa);
6025 if (spa->spa_l2cache.sav_config) {
6026 nvlist_free(spa->spa_l2cache.sav_config);
6027 spa->spa_l2cache.sav_config = NULL;
6028 spa_load_l2cache(spa);
6031 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
6033 spa_config_exit(spa, SCL_ALL, FTAG);
6036 spa_configfile_set(spa, props, B_FALSE);
6038 if (error != 0 || (props && spa_writeable(spa) &&
6039 (error = spa_prop_set(spa, props)))) {
6041 spa_deactivate(spa);
6043 mutex_exit(&spa_namespace_lock);
6047 spa_async_resume(spa);
6050 * Override any spares and level 2 cache devices as specified by
6051 * the user, as these may have correct device names/devids, etc.
6053 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
6054 &spares, &nspares) == 0) {
6055 if (spa->spa_spares.sav_config)
6056 VERIFY(nvlist_remove(spa->spa_spares.sav_config,
6057 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
6059 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
6060 NV_UNIQUE_NAME, KM_SLEEP) == 0);
6061 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
6062 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
6063 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6064 spa_load_spares(spa);
6065 spa_config_exit(spa, SCL_ALL, FTAG);
6066 spa->spa_spares.sav_sync = B_TRUE;
6068 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
6069 &l2cache, &nl2cache) == 0) {
6070 if (spa->spa_l2cache.sav_config)
6071 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
6072 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
6074 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
6075 NV_UNIQUE_NAME, KM_SLEEP) == 0);
6076 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
6077 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
6078 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6079 spa_load_l2cache(spa);
6080 spa_config_exit(spa, SCL_ALL, FTAG);
6081 spa->spa_l2cache.sav_sync = B_TRUE;
6085 * Check for any removed devices.
6087 if (spa->spa_autoreplace) {
6088 spa_aux_check_removed(&spa->spa_spares);
6089 spa_aux_check_removed(&spa->spa_l2cache);
6092 if (spa_writeable(spa)) {
6094 * Update the config cache to include the newly-imported pool.
6096 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
6100 * It's possible that the pool was expanded while it was exported.
6101 * We kick off an async task to handle this for us.
6103 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
6105 spa_history_log_version(spa, "import", NULL);
6107 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_IMPORT);
6109 mutex_exit(&spa_namespace_lock);
6111 zvol_create_minors_recursive(pool);
6117 spa_tryimport(nvlist_t *tryconfig)
6119 nvlist_t *config = NULL;
6120 char *poolname, *cachefile;
6124 zpool_load_policy_t policy;
6126 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
6129 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
6133 * Create and initialize the spa structure.
6135 mutex_enter(&spa_namespace_lock);
6136 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
6137 spa_activate(spa, SPA_MODE_READ);
6140 * Rewind pool if a max txg was provided.
6142 zpool_get_load_policy(spa->spa_config, &policy);
6143 if (policy.zlp_txg != UINT64_MAX) {
6144 spa->spa_load_max_txg = policy.zlp_txg;
6145 spa->spa_extreme_rewind = B_TRUE;
6146 zfs_dbgmsg("spa_tryimport: importing %s, max_txg=%lld",
6147 poolname, (longlong_t)policy.zlp_txg);
6149 zfs_dbgmsg("spa_tryimport: importing %s", poolname);
6152 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_CACHEFILE, &cachefile)
6154 zfs_dbgmsg("spa_tryimport: using cachefile '%s'", cachefile);
6155 spa->spa_config_source = SPA_CONFIG_SRC_CACHEFILE;
6157 spa->spa_config_source = SPA_CONFIG_SRC_SCAN;
6160 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING);
6163 * If 'tryconfig' was at least parsable, return the current config.
6165 if (spa->spa_root_vdev != NULL) {
6166 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
6167 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
6169 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
6171 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
6172 spa->spa_uberblock.ub_timestamp) == 0);
6173 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
6174 spa->spa_load_info) == 0);
6175 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_ERRATA,
6176 spa->spa_errata) == 0);
6179 * If the bootfs property exists on this pool then we
6180 * copy it out so that external consumers can tell which
6181 * pools are bootable.
6183 if ((!error || error == EEXIST) && spa->spa_bootfs) {
6184 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
6187 * We have to play games with the name since the
6188 * pool was opened as TRYIMPORT_NAME.
6190 if (dsl_dsobj_to_dsname(spa_name(spa),
6191 spa->spa_bootfs, tmpname) == 0) {
6195 dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
6197 cp = strchr(tmpname, '/');
6199 (void) strlcpy(dsname, tmpname,
6202 (void) snprintf(dsname, MAXPATHLEN,
6203 "%s/%s", poolname, ++cp);
6205 VERIFY(nvlist_add_string(config,
6206 ZPOOL_CONFIG_BOOTFS, dsname) == 0);
6207 kmem_free(dsname, MAXPATHLEN);
6209 kmem_free(tmpname, MAXPATHLEN);
6213 * Add the list of hot spares and level 2 cache devices.
6215 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
6216 spa_add_spares(spa, config);
6217 spa_add_l2cache(spa, config);
6218 spa_config_exit(spa, SCL_CONFIG, FTAG);
6222 spa_deactivate(spa);
6224 mutex_exit(&spa_namespace_lock);
6230 * Pool export/destroy
6232 * The act of destroying or exporting a pool is very simple. We make sure there
6233 * is no more pending I/O and any references to the pool are gone. Then, we
6234 * update the pool state and sync all the labels to disk, removing the
6235 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
6236 * we don't sync the labels or remove the configuration cache.
6239 spa_export_common(const char *pool, int new_state, nvlist_t **oldconfig,
6240 boolean_t force, boolean_t hardforce)
6247 if (!(spa_mode_global & SPA_MODE_WRITE))
6248 return (SET_ERROR(EROFS));
6250 mutex_enter(&spa_namespace_lock);
6251 if ((spa = spa_lookup(pool)) == NULL) {
6252 mutex_exit(&spa_namespace_lock);
6253 return (SET_ERROR(ENOENT));
6256 if (spa->spa_is_exporting) {
6257 /* the pool is being exported by another thread */
6258 mutex_exit(&spa_namespace_lock);
6259 return (SET_ERROR(ZFS_ERR_EXPORT_IN_PROGRESS));
6261 spa->spa_is_exporting = B_TRUE;
6264 * Put a hold on the pool, drop the namespace lock, stop async tasks,
6265 * reacquire the namespace lock, and see if we can export.
6267 spa_open_ref(spa, FTAG);
6268 mutex_exit(&spa_namespace_lock);
6269 spa_async_suspend(spa);
6270 if (spa->spa_zvol_taskq) {
6271 zvol_remove_minors(spa, spa_name(spa), B_TRUE);
6272 taskq_wait(spa->spa_zvol_taskq);
6274 mutex_enter(&spa_namespace_lock);
6275 spa_close(spa, FTAG);
6277 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
6280 * The pool will be in core if it's openable, in which case we can
6281 * modify its state. Objsets may be open only because they're dirty,
6282 * so we have to force it to sync before checking spa_refcnt.
6284 if (spa->spa_sync_on) {
6285 txg_wait_synced(spa->spa_dsl_pool, 0);
6286 spa_evicting_os_wait(spa);
6290 * A pool cannot be exported or destroyed if there are active
6291 * references. If we are resetting a pool, allow references by
6292 * fault injection handlers.
6294 if (!spa_refcount_zero(spa) ||
6295 (spa->spa_inject_ref != 0 &&
6296 new_state != POOL_STATE_UNINITIALIZED)) {
6297 spa_async_resume(spa);
6298 spa->spa_is_exporting = B_FALSE;
6299 mutex_exit(&spa_namespace_lock);
6300 return (SET_ERROR(EBUSY));
6303 if (spa->spa_sync_on) {
6305 * A pool cannot be exported if it has an active shared spare.
6306 * This is to prevent other pools stealing the active spare
6307 * from an exported pool. At user's own will, such pool can
6308 * be forcedly exported.
6310 if (!force && new_state == POOL_STATE_EXPORTED &&
6311 spa_has_active_shared_spare(spa)) {
6312 spa_async_resume(spa);
6313 spa->spa_is_exporting = B_FALSE;
6314 mutex_exit(&spa_namespace_lock);
6315 return (SET_ERROR(EXDEV));
6319 * We're about to export or destroy this pool. Make sure
6320 * we stop all initialization and trim activity here before
6321 * we set the spa_final_txg. This will ensure that all
6322 * dirty data resulting from the initialization is
6323 * committed to disk before we unload the pool.
6325 if (spa->spa_root_vdev != NULL) {
6326 vdev_t *rvd = spa->spa_root_vdev;
6327 vdev_initialize_stop_all(rvd, VDEV_INITIALIZE_ACTIVE);
6328 vdev_trim_stop_all(rvd, VDEV_TRIM_ACTIVE);
6329 vdev_autotrim_stop_all(spa);
6330 vdev_rebuild_stop_all(spa);
6334 * We want this to be reflected on every label,
6335 * so mark them all dirty. spa_unload() will do the
6336 * final sync that pushes these changes out.
6338 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
6339 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6340 spa->spa_state = new_state;
6341 spa->spa_final_txg = spa_last_synced_txg(spa) +
6343 vdev_config_dirty(spa->spa_root_vdev);
6344 spa_config_exit(spa, SCL_ALL, FTAG);
6349 if (new_state == POOL_STATE_DESTROYED)
6350 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_DESTROY);
6351 else if (new_state == POOL_STATE_EXPORTED)
6352 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_EXPORT);
6354 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
6356 spa_deactivate(spa);
6359 if (oldconfig && spa->spa_config)
6360 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
6362 if (new_state != POOL_STATE_UNINITIALIZED) {
6364 spa_write_cachefile(spa, B_TRUE, B_TRUE);
6368 * If spa_remove() is not called for this spa_t and
6369 * there is any possibility that it can be reused,
6370 * we make sure to reset the exporting flag.
6372 spa->spa_is_exporting = B_FALSE;
6375 mutex_exit(&spa_namespace_lock);
6380 * Destroy a storage pool.
6383 spa_destroy(const char *pool)
6385 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
6390 * Export a storage pool.
6393 spa_export(const char *pool, nvlist_t **oldconfig, boolean_t force,
6394 boolean_t hardforce)
6396 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
6401 * Similar to spa_export(), this unloads the spa_t without actually removing it
6402 * from the namespace in any way.
6405 spa_reset(const char *pool)
6407 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
6412 * ==========================================================================
6413 * Device manipulation
6414 * ==========================================================================
6418 * This is called as a synctask to increment the draid feature flag
6421 spa_draid_feature_incr(void *arg, dmu_tx_t *tx)
6423 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
6424 int draid = (int)(uintptr_t)arg;
6426 for (int c = 0; c < draid; c++)
6427 spa_feature_incr(spa, SPA_FEATURE_DRAID, tx);
6431 * Add a device to a storage pool.
6434 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
6436 uint64_t txg, ndraid = 0;
6438 vdev_t *rvd = spa->spa_root_vdev;
6440 nvlist_t **spares, **l2cache;
6441 uint_t nspares, nl2cache;
6443 ASSERT(spa_writeable(spa));
6445 txg = spa_vdev_enter(spa);
6447 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
6448 VDEV_ALLOC_ADD)) != 0)
6449 return (spa_vdev_exit(spa, NULL, txg, error));
6451 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */
6453 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
6457 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
6461 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
6462 return (spa_vdev_exit(spa, vd, txg, EINVAL));
6464 if (vd->vdev_children != 0 &&
6465 (error = vdev_create(vd, txg, B_FALSE)) != 0) {
6466 return (spa_vdev_exit(spa, vd, txg, error));
6470 * The virtual dRAID spares must be added after vdev tree is created
6471 * and the vdev guids are generated. The guid of their assoicated
6472 * dRAID is stored in the config and used when opening the spare.
6474 if ((error = vdev_draid_spare_create(nvroot, vd, &ndraid,
6475 rvd->vdev_children)) == 0) {
6476 if (ndraid > 0 && nvlist_lookup_nvlist_array(nvroot,
6477 ZPOOL_CONFIG_SPARES, &spares, &nspares) != 0)
6480 return (spa_vdev_exit(spa, vd, txg, error));
6484 * We must validate the spares and l2cache devices after checking the
6485 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
6487 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
6488 return (spa_vdev_exit(spa, vd, txg, error));
6491 * If we are in the middle of a device removal, we can only add
6492 * devices which match the existing devices in the pool.
6493 * If we are in the middle of a removal, or have some indirect
6494 * vdevs, we can not add raidz or dRAID top levels.
6496 if (spa->spa_vdev_removal != NULL ||
6497 spa->spa_removing_phys.sr_prev_indirect_vdev != -1) {
6498 for (int c = 0; c < vd->vdev_children; c++) {
6499 tvd = vd->vdev_child[c];
6500 if (spa->spa_vdev_removal != NULL &&
6501 tvd->vdev_ashift != spa->spa_max_ashift) {
6502 return (spa_vdev_exit(spa, vd, txg, EINVAL));
6504 /* Fail if top level vdev is raidz or a dRAID */
6505 if (vdev_get_nparity(tvd) != 0)
6506 return (spa_vdev_exit(spa, vd, txg, EINVAL));
6509 * Need the top level mirror to be
6510 * a mirror of leaf vdevs only
6512 if (tvd->vdev_ops == &vdev_mirror_ops) {
6513 for (uint64_t cid = 0;
6514 cid < tvd->vdev_children; cid++) {
6515 vdev_t *cvd = tvd->vdev_child[cid];
6516 if (!cvd->vdev_ops->vdev_op_leaf) {
6517 return (spa_vdev_exit(spa, vd,
6525 for (int c = 0; c < vd->vdev_children; c++) {
6526 tvd = vd->vdev_child[c];
6527 vdev_remove_child(vd, tvd);
6528 tvd->vdev_id = rvd->vdev_children;
6529 vdev_add_child(rvd, tvd);
6530 vdev_config_dirty(tvd);
6534 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
6535 ZPOOL_CONFIG_SPARES);
6536 spa_load_spares(spa);
6537 spa->spa_spares.sav_sync = B_TRUE;
6540 if (nl2cache != 0) {
6541 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
6542 ZPOOL_CONFIG_L2CACHE);
6543 spa_load_l2cache(spa);
6544 spa->spa_l2cache.sav_sync = B_TRUE;
6548 * We can't increment a feature while holding spa_vdev so we
6549 * have to do it in a synctask.
6554 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
6555 dsl_sync_task_nowait(spa->spa_dsl_pool, spa_draid_feature_incr,
6556 (void *)(uintptr_t)ndraid, tx);
6561 * We have to be careful when adding new vdevs to an existing pool.
6562 * If other threads start allocating from these vdevs before we
6563 * sync the config cache, and we lose power, then upon reboot we may
6564 * fail to open the pool because there are DVAs that the config cache
6565 * can't translate. Therefore, we first add the vdevs without
6566 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
6567 * and then let spa_config_update() initialize the new metaslabs.
6569 * spa_load() checks for added-but-not-initialized vdevs, so that
6570 * if we lose power at any point in this sequence, the remaining
6571 * steps will be completed the next time we load the pool.
6573 (void) spa_vdev_exit(spa, vd, txg, 0);
6575 mutex_enter(&spa_namespace_lock);
6576 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
6577 spa_event_notify(spa, NULL, NULL, ESC_ZFS_VDEV_ADD);
6578 mutex_exit(&spa_namespace_lock);
6584 * Attach a device to a mirror. The arguments are the path to any device
6585 * in the mirror, and the nvroot for the new device. If the path specifies
6586 * a device that is not mirrored, we automatically insert the mirror vdev.
6588 * If 'replacing' is specified, the new device is intended to replace the
6589 * existing device; in this case the two devices are made into their own
6590 * mirror using the 'replacing' vdev, which is functionally identical to
6591 * the mirror vdev (it actually reuses all the same ops) but has a few
6592 * extra rules: you can't attach to it after it's been created, and upon
6593 * completion of resilvering, the first disk (the one being replaced)
6594 * is automatically detached.
6596 * If 'rebuild' is specified, then sequential reconstruction (a.ka. rebuild)
6597 * should be performed instead of traditional healing reconstruction. From
6598 * an administrators perspective these are both resilver operations.
6601 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing,
6604 uint64_t txg, dtl_max_txg;
6605 vdev_t *rvd = spa->spa_root_vdev;
6606 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
6608 char *oldvdpath, *newvdpath;
6612 ASSERT(spa_writeable(spa));
6614 txg = spa_vdev_enter(spa);
6616 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
6618 ASSERT(MUTEX_HELD(&spa_namespace_lock));
6619 if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) {
6620 error = (spa_has_checkpoint(spa)) ?
6621 ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT;
6622 return (spa_vdev_exit(spa, NULL, txg, error));
6626 if (!spa_feature_is_enabled(spa, SPA_FEATURE_DEVICE_REBUILD))
6627 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
6629 if (dsl_scan_resilvering(spa_get_dsl(spa)))
6630 return (spa_vdev_exit(spa, NULL, txg,
6631 ZFS_ERR_RESILVER_IN_PROGRESS));
6633 if (vdev_rebuild_active(rvd))
6634 return (spa_vdev_exit(spa, NULL, txg,
6635 ZFS_ERR_REBUILD_IN_PROGRESS));
6638 if (spa->spa_vdev_removal != NULL)
6639 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
6642 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
6644 if (!oldvd->vdev_ops->vdev_op_leaf)
6645 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
6647 pvd = oldvd->vdev_parent;
6649 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
6650 VDEV_ALLOC_ATTACH)) != 0)
6651 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
6653 if (newrootvd->vdev_children != 1)
6654 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
6656 newvd = newrootvd->vdev_child[0];
6658 if (!newvd->vdev_ops->vdev_op_leaf)
6659 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
6661 if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
6662 return (spa_vdev_exit(spa, newrootvd, txg, error));
6665 * Spares can't replace logs
6667 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
6668 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
6671 * A dRAID spare can only replace a child of its parent dRAID vdev.
6673 if (newvd->vdev_ops == &vdev_draid_spare_ops &&
6674 oldvd->vdev_top != vdev_draid_spare_get_parent(newvd)) {
6675 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
6680 * For rebuilds, the top vdev must support reconstruction
6681 * using only space maps. This means the only allowable
6682 * vdevs types are the root vdev, a mirror, or dRAID.
6685 if (pvd->vdev_top != NULL)
6686 tvd = pvd->vdev_top;
6688 if (tvd->vdev_ops != &vdev_mirror_ops &&
6689 tvd->vdev_ops != &vdev_root_ops &&
6690 tvd->vdev_ops != &vdev_draid_ops) {
6691 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
6697 * For attach, the only allowable parent is a mirror or the root
6700 if (pvd->vdev_ops != &vdev_mirror_ops &&
6701 pvd->vdev_ops != &vdev_root_ops)
6702 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
6704 pvops = &vdev_mirror_ops;
6707 * Active hot spares can only be replaced by inactive hot
6710 if (pvd->vdev_ops == &vdev_spare_ops &&
6711 oldvd->vdev_isspare &&
6712 !spa_has_spare(spa, newvd->vdev_guid))
6713 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
6716 * If the source is a hot spare, and the parent isn't already a
6717 * spare, then we want to create a new hot spare. Otherwise, we
6718 * want to create a replacing vdev. The user is not allowed to
6719 * attach to a spared vdev child unless the 'isspare' state is
6720 * the same (spare replaces spare, non-spare replaces
6723 if (pvd->vdev_ops == &vdev_replacing_ops &&
6724 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
6725 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
6726 } else if (pvd->vdev_ops == &vdev_spare_ops &&
6727 newvd->vdev_isspare != oldvd->vdev_isspare) {
6728 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
6731 if (newvd->vdev_isspare)
6732 pvops = &vdev_spare_ops;
6734 pvops = &vdev_replacing_ops;
6738 * Make sure the new device is big enough.
6740 if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
6741 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
6744 * The new device cannot have a higher alignment requirement
6745 * than the top-level vdev.
6747 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
6748 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
6751 * If this is an in-place replacement, update oldvd's path and devid
6752 * to make it distinguishable from newvd, and unopenable from now on.
6754 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
6755 spa_strfree(oldvd->vdev_path);
6756 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
6758 (void) snprintf(oldvd->vdev_path, strlen(newvd->vdev_path) + 5,
6759 "%s/%s", newvd->vdev_path, "old");
6760 if (oldvd->vdev_devid != NULL) {
6761 spa_strfree(oldvd->vdev_devid);
6762 oldvd->vdev_devid = NULL;
6767 * If the parent is not a mirror, or if we're replacing, insert the new
6768 * mirror/replacing/spare vdev above oldvd.
6770 if (pvd->vdev_ops != pvops)
6771 pvd = vdev_add_parent(oldvd, pvops);
6773 ASSERT(pvd->vdev_top->vdev_parent == rvd);
6774 ASSERT(pvd->vdev_ops == pvops);
6775 ASSERT(oldvd->vdev_parent == pvd);
6778 * Extract the new device from its root and add it to pvd.
6780 vdev_remove_child(newrootvd, newvd);
6781 newvd->vdev_id = pvd->vdev_children;
6782 newvd->vdev_crtxg = oldvd->vdev_crtxg;
6783 vdev_add_child(pvd, newvd);
6786 * Reevaluate the parent vdev state.
6788 vdev_propagate_state(pvd);
6790 tvd = newvd->vdev_top;
6791 ASSERT(pvd->vdev_top == tvd);
6792 ASSERT(tvd->vdev_parent == rvd);
6794 vdev_config_dirty(tvd);
6797 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
6798 * for any dmu_sync-ed blocks. It will propagate upward when
6799 * spa_vdev_exit() calls vdev_dtl_reassess().
6801 dtl_max_txg = txg + TXG_CONCURRENT_STATES;
6803 vdev_dtl_dirty(newvd, DTL_MISSING,
6804 TXG_INITIAL, dtl_max_txg - TXG_INITIAL);
6806 if (newvd->vdev_isspare) {
6807 spa_spare_activate(newvd);
6808 spa_event_notify(spa, newvd, NULL, ESC_ZFS_VDEV_SPARE);
6811 oldvdpath = spa_strdup(oldvd->vdev_path);
6812 newvdpath = spa_strdup(newvd->vdev_path);
6813 newvd_isspare = newvd->vdev_isspare;
6816 * Mark newvd's DTL dirty in this txg.
6818 vdev_dirty(tvd, VDD_DTL, newvd, txg);
6821 * Schedule the resilver or rebuild to restart in the future. We do
6822 * this to ensure that dmu_sync-ed blocks have been stitched into the
6823 * respective datasets.
6826 newvd->vdev_rebuild_txg = txg;
6830 newvd->vdev_resilver_txg = txg;
6832 if (dsl_scan_resilvering(spa_get_dsl(spa)) &&
6833 spa_feature_is_enabled(spa, SPA_FEATURE_RESILVER_DEFER)) {
6834 vdev_defer_resilver(newvd);
6836 dsl_scan_restart_resilver(spa->spa_dsl_pool,
6841 if (spa->spa_bootfs)
6842 spa_event_notify(spa, newvd, NULL, ESC_ZFS_BOOTFS_VDEV_ATTACH);
6844 spa_event_notify(spa, newvd, NULL, ESC_ZFS_VDEV_ATTACH);
6849 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
6851 spa_history_log_internal(spa, "vdev attach", NULL,
6852 "%s vdev=%s %s vdev=%s",
6853 replacing && newvd_isspare ? "spare in" :
6854 replacing ? "replace" : "attach", newvdpath,
6855 replacing ? "for" : "to", oldvdpath);
6857 spa_strfree(oldvdpath);
6858 spa_strfree(newvdpath);
6864 * Detach a device from a mirror or replacing vdev.
6866 * If 'replace_done' is specified, only detach if the parent
6867 * is a replacing vdev.
6870 spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
6874 vdev_t *rvd __maybe_unused = spa->spa_root_vdev;
6875 vdev_t *vd, *pvd, *cvd, *tvd;
6876 boolean_t unspare = B_FALSE;
6877 uint64_t unspare_guid = 0;
6880 ASSERT(spa_writeable(spa));
6882 txg = spa_vdev_detach_enter(spa, guid);
6884 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
6887 * Besides being called directly from the userland through the
6888 * ioctl interface, spa_vdev_detach() can be potentially called
6889 * at the end of spa_vdev_resilver_done().
6891 * In the regular case, when we have a checkpoint this shouldn't
6892 * happen as we never empty the DTLs of a vdev during the scrub
6893 * [see comment in dsl_scan_done()]. Thus spa_vdev_resilvering_done()
6894 * should never get here when we have a checkpoint.
6896 * That said, even in a case when we checkpoint the pool exactly
6897 * as spa_vdev_resilver_done() calls this function everything
6898 * should be fine as the resilver will return right away.
6900 ASSERT(MUTEX_HELD(&spa_namespace_lock));
6901 if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) {
6902 error = (spa_has_checkpoint(spa)) ?
6903 ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT;
6904 return (spa_vdev_exit(spa, NULL, txg, error));
6908 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
6910 if (!vd->vdev_ops->vdev_op_leaf)
6911 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
6913 pvd = vd->vdev_parent;
6916 * If the parent/child relationship is not as expected, don't do it.
6917 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
6918 * vdev that's replacing B with C. The user's intent in replacing
6919 * is to go from M(A,B) to M(A,C). If the user decides to cancel
6920 * the replace by detaching C, the expected behavior is to end up
6921 * M(A,B). But suppose that right after deciding to detach C,
6922 * the replacement of B completes. We would have M(A,C), and then
6923 * ask to detach C, which would leave us with just A -- not what
6924 * the user wanted. To prevent this, we make sure that the
6925 * parent/child relationship hasn't changed -- in this example,
6926 * that C's parent is still the replacing vdev R.
6928 if (pvd->vdev_guid != pguid && pguid != 0)
6929 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
6932 * Only 'replacing' or 'spare' vdevs can be replaced.
6934 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
6935 pvd->vdev_ops != &vdev_spare_ops)
6936 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
6938 ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
6939 spa_version(spa) >= SPA_VERSION_SPARES);
6942 * Only mirror, replacing, and spare vdevs support detach.
6944 if (pvd->vdev_ops != &vdev_replacing_ops &&
6945 pvd->vdev_ops != &vdev_mirror_ops &&
6946 pvd->vdev_ops != &vdev_spare_ops)
6947 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
6950 * If this device has the only valid copy of some data,
6951 * we cannot safely detach it.
6953 if (vdev_dtl_required(vd))
6954 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
6956 ASSERT(pvd->vdev_children >= 2);
6959 * If we are detaching the second disk from a replacing vdev, then
6960 * check to see if we changed the original vdev's path to have "/old"
6961 * at the end in spa_vdev_attach(). If so, undo that change now.
6963 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
6964 vd->vdev_path != NULL) {
6965 size_t len = strlen(vd->vdev_path);
6967 for (int c = 0; c < pvd->vdev_children; c++) {
6968 cvd = pvd->vdev_child[c];
6970 if (cvd == vd || cvd->vdev_path == NULL)
6973 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
6974 strcmp(cvd->vdev_path + len, "/old") == 0) {
6975 spa_strfree(cvd->vdev_path);
6976 cvd->vdev_path = spa_strdup(vd->vdev_path);
6983 * If we are detaching the original disk from a normal spare, then it
6984 * implies that the spare should become a real disk, and be removed
6985 * from the active spare list for the pool. dRAID spares on the
6986 * other hand are coupled to the pool and thus should never be removed
6987 * from the spares list.
6989 if (pvd->vdev_ops == &vdev_spare_ops && vd->vdev_id == 0) {
6990 vdev_t *last_cvd = pvd->vdev_child[pvd->vdev_children - 1];
6992 if (last_cvd->vdev_isspare &&
6993 last_cvd->vdev_ops != &vdev_draid_spare_ops) {
6999 * Erase the disk labels so the disk can be used for other things.
7000 * This must be done after all other error cases are handled,
7001 * but before we disembowel vd (so we can still do I/O to it).
7002 * But if we can't do it, don't treat the error as fatal --
7003 * it may be that the unwritability of the disk is the reason
7004 * it's being detached!
7006 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
7009 * Remove vd from its parent and compact the parent's children.
7011 vdev_remove_child(pvd, vd);
7012 vdev_compact_children(pvd);
7015 * Remember one of the remaining children so we can get tvd below.
7017 cvd = pvd->vdev_child[pvd->vdev_children - 1];
7020 * If we need to remove the remaining child from the list of hot spares,
7021 * do it now, marking the vdev as no longer a spare in the process.
7022 * We must do this before vdev_remove_parent(), because that can
7023 * change the GUID if it creates a new toplevel GUID. For a similar
7024 * reason, we must remove the spare now, in the same txg as the detach;
7025 * otherwise someone could attach a new sibling, change the GUID, and
7026 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
7029 ASSERT(cvd->vdev_isspare);
7030 spa_spare_remove(cvd);
7031 unspare_guid = cvd->vdev_guid;
7032 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
7033 cvd->vdev_unspare = B_TRUE;
7037 * If the parent mirror/replacing vdev only has one child,
7038 * the parent is no longer needed. Remove it from the tree.
7040 if (pvd->vdev_children == 1) {
7041 if (pvd->vdev_ops == &vdev_spare_ops)
7042 cvd->vdev_unspare = B_FALSE;
7043 vdev_remove_parent(cvd);
7047 * We don't set tvd until now because the parent we just removed
7048 * may have been the previous top-level vdev.
7050 tvd = cvd->vdev_top;
7051 ASSERT(tvd->vdev_parent == rvd);
7054 * Reevaluate the parent vdev state.
7056 vdev_propagate_state(cvd);
7059 * If the 'autoexpand' property is set on the pool then automatically
7060 * try to expand the size of the pool. For example if the device we
7061 * just detached was smaller than the others, it may be possible to
7062 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
7063 * first so that we can obtain the updated sizes of the leaf vdevs.
7065 if (spa->spa_autoexpand) {
7067 vdev_expand(tvd, txg);
7070 vdev_config_dirty(tvd);
7073 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
7074 * vd->vdev_detached is set and free vd's DTL object in syncing context.
7075 * But first make sure we're not on any *other* txg's DTL list, to
7076 * prevent vd from being accessed after it's freed.
7078 vdpath = spa_strdup(vd->vdev_path ? vd->vdev_path : "none");
7079 for (int t = 0; t < TXG_SIZE; t++)
7080 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
7081 vd->vdev_detached = B_TRUE;
7082 vdev_dirty(tvd, VDD_DTL, vd, txg);
7084 spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_REMOVE);
7085 spa_notify_waiters(spa);
7087 /* hang on to the spa before we release the lock */
7088 spa_open_ref(spa, FTAG);
7090 error = spa_vdev_exit(spa, vd, txg, 0);
7092 spa_history_log_internal(spa, "detach", NULL,
7094 spa_strfree(vdpath);
7097 * If this was the removal of the original device in a hot spare vdev,
7098 * then we want to go through and remove the device from the hot spare
7099 * list of every other pool.
7102 spa_t *altspa = NULL;
7104 mutex_enter(&spa_namespace_lock);
7105 while ((altspa = spa_next(altspa)) != NULL) {
7106 if (altspa->spa_state != POOL_STATE_ACTIVE ||
7110 spa_open_ref(altspa, FTAG);
7111 mutex_exit(&spa_namespace_lock);
7112 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
7113 mutex_enter(&spa_namespace_lock);
7114 spa_close(altspa, FTAG);
7116 mutex_exit(&spa_namespace_lock);
7118 /* search the rest of the vdevs for spares to remove */
7119 spa_vdev_resilver_done(spa);
7122 /* all done with the spa; OK to release */
7123 mutex_enter(&spa_namespace_lock);
7124 spa_close(spa, FTAG);
7125 mutex_exit(&spa_namespace_lock);
7131 spa_vdev_initialize_impl(spa_t *spa, uint64_t guid, uint64_t cmd_type,
7134 ASSERT(MUTEX_HELD(&spa_namespace_lock));
7136 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
7138 /* Look up vdev and ensure it's a leaf. */
7139 vdev_t *vd = spa_lookup_by_guid(spa, guid, B_FALSE);
7140 if (vd == NULL || vd->vdev_detached) {
7141 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
7142 return (SET_ERROR(ENODEV));
7143 } else if (!vd->vdev_ops->vdev_op_leaf || !vdev_is_concrete(vd)) {
7144 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
7145 return (SET_ERROR(EINVAL));
7146 } else if (!vdev_writeable(vd)) {
7147 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
7148 return (SET_ERROR(EROFS));
7150 mutex_enter(&vd->vdev_initialize_lock);
7151 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
7154 * When we activate an initialize action we check to see
7155 * if the vdev_initialize_thread is NULL. We do this instead
7156 * of using the vdev_initialize_state since there might be
7157 * a previous initialization process which has completed but
7158 * the thread is not exited.
7160 if (cmd_type == POOL_INITIALIZE_START &&
7161 (vd->vdev_initialize_thread != NULL ||
7162 vd->vdev_top->vdev_removing)) {
7163 mutex_exit(&vd->vdev_initialize_lock);
7164 return (SET_ERROR(EBUSY));
7165 } else if (cmd_type == POOL_INITIALIZE_CANCEL &&
7166 (vd->vdev_initialize_state != VDEV_INITIALIZE_ACTIVE &&
7167 vd->vdev_initialize_state != VDEV_INITIALIZE_SUSPENDED)) {
7168 mutex_exit(&vd->vdev_initialize_lock);
7169 return (SET_ERROR(ESRCH));
7170 } else if (cmd_type == POOL_INITIALIZE_SUSPEND &&
7171 vd->vdev_initialize_state != VDEV_INITIALIZE_ACTIVE) {
7172 mutex_exit(&vd->vdev_initialize_lock);
7173 return (SET_ERROR(ESRCH));
7177 case POOL_INITIALIZE_START:
7178 vdev_initialize(vd);
7180 case POOL_INITIALIZE_CANCEL:
7181 vdev_initialize_stop(vd, VDEV_INITIALIZE_CANCELED, vd_list);
7183 case POOL_INITIALIZE_SUSPEND:
7184 vdev_initialize_stop(vd, VDEV_INITIALIZE_SUSPENDED, vd_list);
7187 panic("invalid cmd_type %llu", (unsigned long long)cmd_type);
7189 mutex_exit(&vd->vdev_initialize_lock);
7195 spa_vdev_initialize(spa_t *spa, nvlist_t *nv, uint64_t cmd_type,
7196 nvlist_t *vdev_errlist)
7198 int total_errors = 0;
7201 list_create(&vd_list, sizeof (vdev_t),
7202 offsetof(vdev_t, vdev_initialize_node));
7205 * We hold the namespace lock through the whole function
7206 * to prevent any changes to the pool while we're starting or
7207 * stopping initialization. The config and state locks are held so that
7208 * we can properly assess the vdev state before we commit to
7209 * the initializing operation.
7211 mutex_enter(&spa_namespace_lock);
7213 for (nvpair_t *pair = nvlist_next_nvpair(nv, NULL);
7214 pair != NULL; pair = nvlist_next_nvpair(nv, pair)) {
7215 uint64_t vdev_guid = fnvpair_value_uint64(pair);
7217 int error = spa_vdev_initialize_impl(spa, vdev_guid, cmd_type,
7220 char guid_as_str[MAXNAMELEN];
7222 (void) snprintf(guid_as_str, sizeof (guid_as_str),
7223 "%llu", (unsigned long long)vdev_guid);
7224 fnvlist_add_int64(vdev_errlist, guid_as_str, error);
7229 /* Wait for all initialize threads to stop. */
7230 vdev_initialize_stop_wait(spa, &vd_list);
7232 /* Sync out the initializing state */
7233 txg_wait_synced(spa->spa_dsl_pool, 0);
7234 mutex_exit(&spa_namespace_lock);
7236 list_destroy(&vd_list);
7238 return (total_errors);
7242 spa_vdev_trim_impl(spa_t *spa, uint64_t guid, uint64_t cmd_type,
7243 uint64_t rate, boolean_t partial, boolean_t secure, list_t *vd_list)
7245 ASSERT(MUTEX_HELD(&spa_namespace_lock));
7247 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
7249 /* Look up vdev and ensure it's a leaf. */
7250 vdev_t *vd = spa_lookup_by_guid(spa, guid, B_FALSE);
7251 if (vd == NULL || vd->vdev_detached) {
7252 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
7253 return (SET_ERROR(ENODEV));
7254 } else if (!vd->vdev_ops->vdev_op_leaf || !vdev_is_concrete(vd)) {
7255 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
7256 return (SET_ERROR(EINVAL));
7257 } else if (!vdev_writeable(vd)) {
7258 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
7259 return (SET_ERROR(EROFS));
7260 } else if (!vd->vdev_has_trim) {
7261 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
7262 return (SET_ERROR(EOPNOTSUPP));
7263 } else if (secure && !vd->vdev_has_securetrim) {
7264 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
7265 return (SET_ERROR(EOPNOTSUPP));
7267 mutex_enter(&vd->vdev_trim_lock);
7268 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
7271 * When we activate a TRIM action we check to see if the
7272 * vdev_trim_thread is NULL. We do this instead of using the
7273 * vdev_trim_state since there might be a previous TRIM process
7274 * which has completed but the thread is not exited.
7276 if (cmd_type == POOL_TRIM_START &&
7277 (vd->vdev_trim_thread != NULL || vd->vdev_top->vdev_removing)) {
7278 mutex_exit(&vd->vdev_trim_lock);
7279 return (SET_ERROR(EBUSY));
7280 } else if (cmd_type == POOL_TRIM_CANCEL &&
7281 (vd->vdev_trim_state != VDEV_TRIM_ACTIVE &&
7282 vd->vdev_trim_state != VDEV_TRIM_SUSPENDED)) {
7283 mutex_exit(&vd->vdev_trim_lock);
7284 return (SET_ERROR(ESRCH));
7285 } else if (cmd_type == POOL_TRIM_SUSPEND &&
7286 vd->vdev_trim_state != VDEV_TRIM_ACTIVE) {
7287 mutex_exit(&vd->vdev_trim_lock);
7288 return (SET_ERROR(ESRCH));
7292 case POOL_TRIM_START:
7293 vdev_trim(vd, rate, partial, secure);
7295 case POOL_TRIM_CANCEL:
7296 vdev_trim_stop(vd, VDEV_TRIM_CANCELED, vd_list);
7298 case POOL_TRIM_SUSPEND:
7299 vdev_trim_stop(vd, VDEV_TRIM_SUSPENDED, vd_list);
7302 panic("invalid cmd_type %llu", (unsigned long long)cmd_type);
7304 mutex_exit(&vd->vdev_trim_lock);
7310 * Initiates a manual TRIM for the requested vdevs. This kicks off individual
7311 * TRIM threads for each child vdev. These threads pass over all of the free
7312 * space in the vdev's metaslabs and issues TRIM commands for that space.
7315 spa_vdev_trim(spa_t *spa, nvlist_t *nv, uint64_t cmd_type, uint64_t rate,
7316 boolean_t partial, boolean_t secure, nvlist_t *vdev_errlist)
7318 int total_errors = 0;
7321 list_create(&vd_list, sizeof (vdev_t),
7322 offsetof(vdev_t, vdev_trim_node));
7325 * We hold the namespace lock through the whole function
7326 * to prevent any changes to the pool while we're starting or
7327 * stopping TRIM. The config and state locks are held so that
7328 * we can properly assess the vdev state before we commit to
7329 * the TRIM operation.
7331 mutex_enter(&spa_namespace_lock);
7333 for (nvpair_t *pair = nvlist_next_nvpair(nv, NULL);
7334 pair != NULL; pair = nvlist_next_nvpair(nv, pair)) {
7335 uint64_t vdev_guid = fnvpair_value_uint64(pair);
7337 int error = spa_vdev_trim_impl(spa, vdev_guid, cmd_type,
7338 rate, partial, secure, &vd_list);
7340 char guid_as_str[MAXNAMELEN];
7342 (void) snprintf(guid_as_str, sizeof (guid_as_str),
7343 "%llu", (unsigned long long)vdev_guid);
7344 fnvlist_add_int64(vdev_errlist, guid_as_str, error);
7349 /* Wait for all TRIM threads to stop. */
7350 vdev_trim_stop_wait(spa, &vd_list);
7352 /* Sync out the TRIM state */
7353 txg_wait_synced(spa->spa_dsl_pool, 0);
7354 mutex_exit(&spa_namespace_lock);
7356 list_destroy(&vd_list);
7358 return (total_errors);
7362 * Split a set of devices from their mirrors, and create a new pool from them.
7365 spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
7366 nvlist_t *props, boolean_t exp)
7369 uint64_t txg, *glist;
7371 uint_t c, children, lastlog;
7372 nvlist_t **child, *nvl, *tmp;
7374 char *altroot = NULL;
7375 vdev_t *rvd, **vml = NULL; /* vdev modify list */
7376 boolean_t activate_slog;
7378 ASSERT(spa_writeable(spa));
7380 txg = spa_vdev_enter(spa);
7382 ASSERT(MUTEX_HELD(&spa_namespace_lock));
7383 if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) {
7384 error = (spa_has_checkpoint(spa)) ?
7385 ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT;
7386 return (spa_vdev_exit(spa, NULL, txg, error));
7389 /* clear the log and flush everything up to now */
7390 activate_slog = spa_passivate_log(spa);
7391 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
7392 error = spa_reset_logs(spa);
7393 txg = spa_vdev_config_enter(spa);
7396 spa_activate_log(spa);
7399 return (spa_vdev_exit(spa, NULL, txg, error));
7401 /* check new spa name before going any further */
7402 if (spa_lookup(newname) != NULL)
7403 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
7406 * scan through all the children to ensure they're all mirrors
7408 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
7409 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
7411 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
7413 /* first, check to ensure we've got the right child count */
7414 rvd = spa->spa_root_vdev;
7416 for (c = 0; c < rvd->vdev_children; c++) {
7417 vdev_t *vd = rvd->vdev_child[c];
7419 /* don't count the holes & logs as children */
7420 if (vd->vdev_islog || (vd->vdev_ops != &vdev_indirect_ops &&
7421 !vdev_is_concrete(vd))) {
7429 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
7430 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
7432 /* next, ensure no spare or cache devices are part of the split */
7433 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
7434 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
7435 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
7437 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
7438 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
7440 /* then, loop over each vdev and validate it */
7441 for (c = 0; c < children; c++) {
7442 uint64_t is_hole = 0;
7444 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
7448 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
7449 spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
7452 error = SET_ERROR(EINVAL);
7457 /* deal with indirect vdevs */
7458 if (spa->spa_root_vdev->vdev_child[c]->vdev_ops ==
7462 /* which disk is going to be split? */
7463 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
7465 error = SET_ERROR(EINVAL);
7469 /* look it up in the spa */
7470 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
7471 if (vml[c] == NULL) {
7472 error = SET_ERROR(ENODEV);
7476 /* make sure there's nothing stopping the split */
7477 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
7478 vml[c]->vdev_islog ||
7479 !vdev_is_concrete(vml[c]) ||
7480 vml[c]->vdev_isspare ||
7481 vml[c]->vdev_isl2cache ||
7482 !vdev_writeable(vml[c]) ||
7483 vml[c]->vdev_children != 0 ||
7484 vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
7485 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
7486 error = SET_ERROR(EINVAL);
7490 if (vdev_dtl_required(vml[c]) ||
7491 vdev_resilver_needed(vml[c], NULL, NULL)) {
7492 error = SET_ERROR(EBUSY);
7496 /* we need certain info from the top level */
7497 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
7498 vml[c]->vdev_top->vdev_ms_array) == 0);
7499 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
7500 vml[c]->vdev_top->vdev_ms_shift) == 0);
7501 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
7502 vml[c]->vdev_top->vdev_asize) == 0);
7503 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
7504 vml[c]->vdev_top->vdev_ashift) == 0);
7506 /* transfer per-vdev ZAPs */
7507 ASSERT3U(vml[c]->vdev_leaf_zap, !=, 0);
7508 VERIFY0(nvlist_add_uint64(child[c],
7509 ZPOOL_CONFIG_VDEV_LEAF_ZAP, vml[c]->vdev_leaf_zap));
7511 ASSERT3U(vml[c]->vdev_top->vdev_top_zap, !=, 0);
7512 VERIFY0(nvlist_add_uint64(child[c],
7513 ZPOOL_CONFIG_VDEV_TOP_ZAP,
7514 vml[c]->vdev_parent->vdev_top_zap));
7518 kmem_free(vml, children * sizeof (vdev_t *));
7519 kmem_free(glist, children * sizeof (uint64_t));
7520 return (spa_vdev_exit(spa, NULL, txg, error));
7523 /* stop writers from using the disks */
7524 for (c = 0; c < children; c++) {
7526 vml[c]->vdev_offline = B_TRUE;
7528 vdev_reopen(spa->spa_root_vdev);
7531 * Temporarily record the splitting vdevs in the spa config. This
7532 * will disappear once the config is regenerated.
7534 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
7535 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
7536 glist, children) == 0);
7537 kmem_free(glist, children * sizeof (uint64_t));
7539 mutex_enter(&spa->spa_props_lock);
7540 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
7542 mutex_exit(&spa->spa_props_lock);
7543 spa->spa_config_splitting = nvl;
7544 vdev_config_dirty(spa->spa_root_vdev);
7546 /* configure and create the new pool */
7547 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
7548 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
7549 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
7550 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
7551 spa_version(spa)) == 0);
7552 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
7553 spa->spa_config_txg) == 0);
7554 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
7555 spa_generate_guid(NULL)) == 0);
7556 VERIFY0(nvlist_add_boolean(config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS));
7557 (void) nvlist_lookup_string(props,
7558 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
7560 /* add the new pool to the namespace */
7561 newspa = spa_add(newname, config, altroot);
7562 newspa->spa_avz_action = AVZ_ACTION_REBUILD;
7563 newspa->spa_config_txg = spa->spa_config_txg;
7564 spa_set_log_state(newspa, SPA_LOG_CLEAR);
7566 /* release the spa config lock, retaining the namespace lock */
7567 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
7569 if (zio_injection_enabled)
7570 zio_handle_panic_injection(spa, FTAG, 1);
7572 spa_activate(newspa, spa_mode_global);
7573 spa_async_suspend(newspa);
7576 * Temporarily stop the initializing and TRIM activity. We set the
7577 * state to ACTIVE so that we know to resume initializing or TRIM
7578 * once the split has completed.
7580 list_t vd_initialize_list;
7581 list_create(&vd_initialize_list, sizeof (vdev_t),
7582 offsetof(vdev_t, vdev_initialize_node));
7584 list_t vd_trim_list;
7585 list_create(&vd_trim_list, sizeof (vdev_t),
7586 offsetof(vdev_t, vdev_trim_node));
7588 for (c = 0; c < children; c++) {
7589 if (vml[c] != NULL && vml[c]->vdev_ops != &vdev_indirect_ops) {
7590 mutex_enter(&vml[c]->vdev_initialize_lock);
7591 vdev_initialize_stop(vml[c],
7592 VDEV_INITIALIZE_ACTIVE, &vd_initialize_list);
7593 mutex_exit(&vml[c]->vdev_initialize_lock);
7595 mutex_enter(&vml[c]->vdev_trim_lock);
7596 vdev_trim_stop(vml[c], VDEV_TRIM_ACTIVE, &vd_trim_list);
7597 mutex_exit(&vml[c]->vdev_trim_lock);
7601 vdev_initialize_stop_wait(spa, &vd_initialize_list);
7602 vdev_trim_stop_wait(spa, &vd_trim_list);
7604 list_destroy(&vd_initialize_list);
7605 list_destroy(&vd_trim_list);
7607 newspa->spa_config_source = SPA_CONFIG_SRC_SPLIT;
7608 newspa->spa_is_splitting = B_TRUE;
7610 /* create the new pool from the disks of the original pool */
7611 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE);
7615 /* if that worked, generate a real config for the new pool */
7616 if (newspa->spa_root_vdev != NULL) {
7617 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
7618 NV_UNIQUE_NAME, KM_SLEEP) == 0);
7619 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
7620 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
7621 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
7626 if (props != NULL) {
7627 spa_configfile_set(newspa, props, B_FALSE);
7628 error = spa_prop_set(newspa, props);
7633 /* flush everything */
7634 txg = spa_vdev_config_enter(newspa);
7635 vdev_config_dirty(newspa->spa_root_vdev);
7636 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
7638 if (zio_injection_enabled)
7639 zio_handle_panic_injection(spa, FTAG, 2);
7641 spa_async_resume(newspa);
7643 /* finally, update the original pool's config */
7644 txg = spa_vdev_config_enter(spa);
7645 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
7646 error = dmu_tx_assign(tx, TXG_WAIT);
7649 for (c = 0; c < children; c++) {
7650 if (vml[c] != NULL && vml[c]->vdev_ops != &vdev_indirect_ops) {
7651 vdev_t *tvd = vml[c]->vdev_top;
7654 * Need to be sure the detachable VDEV is not
7655 * on any *other* txg's DTL list to prevent it
7656 * from being accessed after it's freed.
7658 for (int t = 0; t < TXG_SIZE; t++) {
7659 (void) txg_list_remove_this(
7660 &tvd->vdev_dtl_list, vml[c], t);
7665 spa_history_log_internal(spa, "detach", tx,
7666 "vdev=%s", vml[c]->vdev_path);
7671 spa->spa_avz_action = AVZ_ACTION_REBUILD;
7672 vdev_config_dirty(spa->spa_root_vdev);
7673 spa->spa_config_splitting = NULL;
7677 (void) spa_vdev_exit(spa, NULL, txg, 0);
7679 if (zio_injection_enabled)
7680 zio_handle_panic_injection(spa, FTAG, 3);
7682 /* split is complete; log a history record */
7683 spa_history_log_internal(newspa, "split", NULL,
7684 "from pool %s", spa_name(spa));
7686 newspa->spa_is_splitting = B_FALSE;
7687 kmem_free(vml, children * sizeof (vdev_t *));
7689 /* if we're not going to mount the filesystems in userland, export */
7691 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
7698 spa_deactivate(newspa);
7701 txg = spa_vdev_config_enter(spa);
7703 /* re-online all offlined disks */
7704 for (c = 0; c < children; c++) {
7706 vml[c]->vdev_offline = B_FALSE;
7709 /* restart initializing or trimming disks as necessary */
7710 spa_async_request(spa, SPA_ASYNC_INITIALIZE_RESTART);
7711 spa_async_request(spa, SPA_ASYNC_TRIM_RESTART);
7712 spa_async_request(spa, SPA_ASYNC_AUTOTRIM_RESTART);
7714 vdev_reopen(spa->spa_root_vdev);
7716 nvlist_free(spa->spa_config_splitting);
7717 spa->spa_config_splitting = NULL;
7718 (void) spa_vdev_exit(spa, NULL, txg, error);
7720 kmem_free(vml, children * sizeof (vdev_t *));
7725 * Find any device that's done replacing, or a vdev marked 'unspare' that's
7726 * currently spared, so we can detach it.
7729 spa_vdev_resilver_done_hunt(vdev_t *vd)
7731 vdev_t *newvd, *oldvd;
7733 for (int c = 0; c < vd->vdev_children; c++) {
7734 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
7740 * Check for a completed replacement. We always consider the first
7741 * vdev in the list to be the oldest vdev, and the last one to be
7742 * the newest (see spa_vdev_attach() for how that works). In
7743 * the case where the newest vdev is faulted, we will not automatically
7744 * remove it after a resilver completes. This is OK as it will require
7745 * user intervention to determine which disk the admin wishes to keep.
7747 if (vd->vdev_ops == &vdev_replacing_ops) {
7748 ASSERT(vd->vdev_children > 1);
7750 newvd = vd->vdev_child[vd->vdev_children - 1];
7751 oldvd = vd->vdev_child[0];
7753 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
7754 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
7755 !vdev_dtl_required(oldvd))
7760 * Check for a completed resilver with the 'unspare' flag set.
7761 * Also potentially update faulted state.
7763 if (vd->vdev_ops == &vdev_spare_ops) {
7764 vdev_t *first = vd->vdev_child[0];
7765 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
7767 if (last->vdev_unspare) {
7770 } else if (first->vdev_unspare) {
7777 if (oldvd != NULL &&
7778 vdev_dtl_empty(newvd, DTL_MISSING) &&
7779 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
7780 !vdev_dtl_required(oldvd))
7783 vdev_propagate_state(vd);
7786 * If there are more than two spares attached to a disk,
7787 * and those spares are not required, then we want to
7788 * attempt to free them up now so that they can be used
7789 * by other pools. Once we're back down to a single
7790 * disk+spare, we stop removing them.
7792 if (vd->vdev_children > 2) {
7793 newvd = vd->vdev_child[1];
7795 if (newvd->vdev_isspare && last->vdev_isspare &&
7796 vdev_dtl_empty(last, DTL_MISSING) &&
7797 vdev_dtl_empty(last, DTL_OUTAGE) &&
7798 !vdev_dtl_required(newvd))
7807 spa_vdev_resilver_done(spa_t *spa)
7809 vdev_t *vd, *pvd, *ppvd;
7810 uint64_t guid, sguid, pguid, ppguid;
7812 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
7814 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
7815 pvd = vd->vdev_parent;
7816 ppvd = pvd->vdev_parent;
7817 guid = vd->vdev_guid;
7818 pguid = pvd->vdev_guid;
7819 ppguid = ppvd->vdev_guid;
7822 * If we have just finished replacing a hot spared device, then
7823 * we need to detach the parent's first child (the original hot
7826 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
7827 ppvd->vdev_children == 2) {
7828 ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
7829 sguid = ppvd->vdev_child[1]->vdev_guid;
7831 ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd));
7833 spa_config_exit(spa, SCL_ALL, FTAG);
7834 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
7836 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
7838 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
7841 spa_config_exit(spa, SCL_ALL, FTAG);
7844 * If a detach was not performed above replace waiters will not have
7845 * been notified. In which case we must do so now.
7847 spa_notify_waiters(spa);
7851 * Update the stored path or FRU for this vdev.
7854 spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
7858 boolean_t sync = B_FALSE;
7860 ASSERT(spa_writeable(spa));
7862 spa_vdev_state_enter(spa, SCL_ALL);
7864 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
7865 return (spa_vdev_state_exit(spa, NULL, ENOENT));
7867 if (!vd->vdev_ops->vdev_op_leaf)
7868 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
7871 if (strcmp(value, vd->vdev_path) != 0) {
7872 spa_strfree(vd->vdev_path);
7873 vd->vdev_path = spa_strdup(value);
7877 if (vd->vdev_fru == NULL) {
7878 vd->vdev_fru = spa_strdup(value);
7880 } else if (strcmp(value, vd->vdev_fru) != 0) {
7881 spa_strfree(vd->vdev_fru);
7882 vd->vdev_fru = spa_strdup(value);
7887 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
7891 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
7893 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
7897 spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
7899 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
7903 * ==========================================================================
7905 * ==========================================================================
7908 spa_scrub_pause_resume(spa_t *spa, pool_scrub_cmd_t cmd)
7910 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
7912 if (dsl_scan_resilvering(spa->spa_dsl_pool))
7913 return (SET_ERROR(EBUSY));
7915 return (dsl_scrub_set_pause_resume(spa->spa_dsl_pool, cmd));
7919 spa_scan_stop(spa_t *spa)
7921 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
7922 if (dsl_scan_resilvering(spa->spa_dsl_pool))
7923 return (SET_ERROR(EBUSY));
7924 return (dsl_scan_cancel(spa->spa_dsl_pool));
7928 spa_scan(spa_t *spa, pool_scan_func_t func)
7930 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
7932 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
7933 return (SET_ERROR(ENOTSUP));
7935 if (func == POOL_SCAN_RESILVER &&
7936 !spa_feature_is_enabled(spa, SPA_FEATURE_RESILVER_DEFER))
7937 return (SET_ERROR(ENOTSUP));
7940 * If a resilver was requested, but there is no DTL on a
7941 * writeable leaf device, we have nothing to do.
7943 if (func == POOL_SCAN_RESILVER &&
7944 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
7945 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
7949 return (dsl_scan(spa->spa_dsl_pool, func));
7953 * ==========================================================================
7954 * SPA async task processing
7955 * ==========================================================================
7959 spa_async_remove(spa_t *spa, vdev_t *vd)
7961 if (vd->vdev_remove_wanted) {
7962 vd->vdev_remove_wanted = B_FALSE;
7963 vd->vdev_delayed_close = B_FALSE;
7964 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
7967 * We want to clear the stats, but we don't want to do a full
7968 * vdev_clear() as that will cause us to throw away
7969 * degraded/faulted state as well as attempt to reopen the
7970 * device, all of which is a waste.
7972 vd->vdev_stat.vs_read_errors = 0;
7973 vd->vdev_stat.vs_write_errors = 0;
7974 vd->vdev_stat.vs_checksum_errors = 0;
7976 vdev_state_dirty(vd->vdev_top);
7979 for (int c = 0; c < vd->vdev_children; c++)
7980 spa_async_remove(spa, vd->vdev_child[c]);
7984 spa_async_probe(spa_t *spa, vdev_t *vd)
7986 if (vd->vdev_probe_wanted) {
7987 vd->vdev_probe_wanted = B_FALSE;
7988 vdev_reopen(vd); /* vdev_open() does the actual probe */
7991 for (int c = 0; c < vd->vdev_children; c++)
7992 spa_async_probe(spa, vd->vdev_child[c]);
7996 spa_async_autoexpand(spa_t *spa, vdev_t *vd)
7998 if (!spa->spa_autoexpand)
8001 for (int c = 0; c < vd->vdev_children; c++) {
8002 vdev_t *cvd = vd->vdev_child[c];
8003 spa_async_autoexpand(spa, cvd);
8006 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
8009 spa_event_notify(vd->vdev_spa, vd, NULL, ESC_ZFS_VDEV_AUTOEXPAND);
8013 spa_async_thread(void *arg)
8015 spa_t *spa = (spa_t *)arg;
8016 dsl_pool_t *dp = spa->spa_dsl_pool;
8019 ASSERT(spa->spa_sync_on);
8021 mutex_enter(&spa->spa_async_lock);
8022 tasks = spa->spa_async_tasks;
8023 spa->spa_async_tasks = 0;
8024 mutex_exit(&spa->spa_async_lock);
8027 * See if the config needs to be updated.
8029 if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
8030 uint64_t old_space, new_space;
8032 mutex_enter(&spa_namespace_lock);
8033 old_space = metaslab_class_get_space(spa_normal_class(spa));
8034 old_space += metaslab_class_get_space(spa_special_class(spa));
8035 old_space += metaslab_class_get_space(spa_dedup_class(spa));
8037 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
8039 new_space = metaslab_class_get_space(spa_normal_class(spa));
8040 new_space += metaslab_class_get_space(spa_special_class(spa));
8041 new_space += metaslab_class_get_space(spa_dedup_class(spa));
8042 mutex_exit(&spa_namespace_lock);
8045 * If the pool grew as a result of the config update,
8046 * then log an internal history event.
8048 if (new_space != old_space) {
8049 spa_history_log_internal(spa, "vdev online", NULL,
8050 "pool '%s' size: %llu(+%llu)",
8051 spa_name(spa), (u_longlong_t)new_space,
8052 (u_longlong_t)(new_space - old_space));
8057 * See if any devices need to be marked REMOVED.
8059 if (tasks & SPA_ASYNC_REMOVE) {
8060 spa_vdev_state_enter(spa, SCL_NONE);
8061 spa_async_remove(spa, spa->spa_root_vdev);
8062 for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
8063 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
8064 for (int i = 0; i < spa->spa_spares.sav_count; i++)
8065 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
8066 (void) spa_vdev_state_exit(spa, NULL, 0);
8069 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
8070 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
8071 spa_async_autoexpand(spa, spa->spa_root_vdev);
8072 spa_config_exit(spa, SCL_CONFIG, FTAG);
8076 * See if any devices need to be probed.
8078 if (tasks & SPA_ASYNC_PROBE) {
8079 spa_vdev_state_enter(spa, SCL_NONE);
8080 spa_async_probe(spa, spa->spa_root_vdev);
8081 (void) spa_vdev_state_exit(spa, NULL, 0);
8085 * If any devices are done replacing, detach them.
8087 if (tasks & SPA_ASYNC_RESILVER_DONE ||
8088 tasks & SPA_ASYNC_REBUILD_DONE) {
8089 spa_vdev_resilver_done(spa);
8093 * Kick off a resilver.
8095 if (tasks & SPA_ASYNC_RESILVER &&
8096 !vdev_rebuild_active(spa->spa_root_vdev) &&
8097 (!dsl_scan_resilvering(dp) ||
8098 !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_RESILVER_DEFER)))
8099 dsl_scan_restart_resilver(dp, 0);
8101 if (tasks & SPA_ASYNC_INITIALIZE_RESTART) {
8102 mutex_enter(&spa_namespace_lock);
8103 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
8104 vdev_initialize_restart(spa->spa_root_vdev);
8105 spa_config_exit(spa, SCL_CONFIG, FTAG);
8106 mutex_exit(&spa_namespace_lock);
8109 if (tasks & SPA_ASYNC_TRIM_RESTART) {
8110 mutex_enter(&spa_namespace_lock);
8111 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
8112 vdev_trim_restart(spa->spa_root_vdev);
8113 spa_config_exit(spa, SCL_CONFIG, FTAG);
8114 mutex_exit(&spa_namespace_lock);
8117 if (tasks & SPA_ASYNC_AUTOTRIM_RESTART) {
8118 mutex_enter(&spa_namespace_lock);
8119 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
8120 vdev_autotrim_restart(spa);
8121 spa_config_exit(spa, SCL_CONFIG, FTAG);
8122 mutex_exit(&spa_namespace_lock);
8126 * Kick off L2 cache whole device TRIM.
8128 if (tasks & SPA_ASYNC_L2CACHE_TRIM) {
8129 mutex_enter(&spa_namespace_lock);
8130 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
8131 vdev_trim_l2arc(spa);
8132 spa_config_exit(spa, SCL_CONFIG, FTAG);
8133 mutex_exit(&spa_namespace_lock);
8137 * Kick off L2 cache rebuilding.
8139 if (tasks & SPA_ASYNC_L2CACHE_REBUILD) {
8140 mutex_enter(&spa_namespace_lock);
8141 spa_config_enter(spa, SCL_L2ARC, FTAG, RW_READER);
8142 l2arc_spa_rebuild_start(spa);
8143 spa_config_exit(spa, SCL_L2ARC, FTAG);
8144 mutex_exit(&spa_namespace_lock);
8148 * Let the world know that we're done.
8150 mutex_enter(&spa->spa_async_lock);
8151 spa->spa_async_thread = NULL;
8152 cv_broadcast(&spa->spa_async_cv);
8153 mutex_exit(&spa->spa_async_lock);
8158 spa_async_suspend(spa_t *spa)
8160 mutex_enter(&spa->spa_async_lock);
8161 spa->spa_async_suspended++;
8162 while (spa->spa_async_thread != NULL)
8163 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
8164 mutex_exit(&spa->spa_async_lock);
8166 spa_vdev_remove_suspend(spa);
8168 zthr_t *condense_thread = spa->spa_condense_zthr;
8169 if (condense_thread != NULL)
8170 zthr_cancel(condense_thread);
8172 zthr_t *discard_thread = spa->spa_checkpoint_discard_zthr;
8173 if (discard_thread != NULL)
8174 zthr_cancel(discard_thread);
8176 zthr_t *ll_delete_thread = spa->spa_livelist_delete_zthr;
8177 if (ll_delete_thread != NULL)
8178 zthr_cancel(ll_delete_thread);
8180 zthr_t *ll_condense_thread = spa->spa_livelist_condense_zthr;
8181 if (ll_condense_thread != NULL)
8182 zthr_cancel(ll_condense_thread);
8186 spa_async_resume(spa_t *spa)
8188 mutex_enter(&spa->spa_async_lock);
8189 ASSERT(spa->spa_async_suspended != 0);
8190 spa->spa_async_suspended--;
8191 mutex_exit(&spa->spa_async_lock);
8192 spa_restart_removal(spa);
8194 zthr_t *condense_thread = spa->spa_condense_zthr;
8195 if (condense_thread != NULL)
8196 zthr_resume(condense_thread);
8198 zthr_t *discard_thread = spa->spa_checkpoint_discard_zthr;
8199 if (discard_thread != NULL)
8200 zthr_resume(discard_thread);
8202 zthr_t *ll_delete_thread = spa->spa_livelist_delete_zthr;
8203 if (ll_delete_thread != NULL)
8204 zthr_resume(ll_delete_thread);
8206 zthr_t *ll_condense_thread = spa->spa_livelist_condense_zthr;
8207 if (ll_condense_thread != NULL)
8208 zthr_resume(ll_condense_thread);
8212 spa_async_tasks_pending(spa_t *spa)
8214 uint_t non_config_tasks;
8216 boolean_t config_task_suspended;
8218 non_config_tasks = spa->spa_async_tasks & ~SPA_ASYNC_CONFIG_UPDATE;
8219 config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE;
8220 if (spa->spa_ccw_fail_time == 0) {
8221 config_task_suspended = B_FALSE;
8223 config_task_suspended =
8224 (gethrtime() - spa->spa_ccw_fail_time) <
8225 ((hrtime_t)zfs_ccw_retry_interval * NANOSEC);
8228 return (non_config_tasks || (config_task && !config_task_suspended));
8232 spa_async_dispatch(spa_t *spa)
8234 mutex_enter(&spa->spa_async_lock);
8235 if (spa_async_tasks_pending(spa) &&
8236 !spa->spa_async_suspended &&
8237 spa->spa_async_thread == NULL)
8238 spa->spa_async_thread = thread_create(NULL, 0,
8239 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
8240 mutex_exit(&spa->spa_async_lock);
8244 spa_async_request(spa_t *spa, int task)
8246 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
8247 mutex_enter(&spa->spa_async_lock);
8248 spa->spa_async_tasks |= task;
8249 mutex_exit(&spa->spa_async_lock);
8253 spa_async_tasks(spa_t *spa)
8255 return (spa->spa_async_tasks);
8259 * ==========================================================================
8260 * SPA syncing routines
8261 * ==========================================================================
8266 bpobj_enqueue_cb(void *arg, const blkptr_t *bp, boolean_t bp_freed,
8270 bpobj_enqueue(bpo, bp, bp_freed, tx);
8275 bpobj_enqueue_alloc_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
8277 return (bpobj_enqueue_cb(arg, bp, B_FALSE, tx));
8281 bpobj_enqueue_free_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
8283 return (bpobj_enqueue_cb(arg, bp, B_TRUE, tx));
8287 spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
8291 zio_nowait(zio_free_sync(pio, pio->io_spa, dmu_tx_get_txg(tx), bp,
8297 bpobj_spa_free_sync_cb(void *arg, const blkptr_t *bp, boolean_t bp_freed,
8301 return (spa_free_sync_cb(arg, bp, tx));
8305 * Note: this simple function is not inlined to make it easier to dtrace the
8306 * amount of time spent syncing frees.
8309 spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx)
8311 zio_t *zio = zio_root(spa, NULL, NULL, 0);
8312 bplist_iterate(bpl, spa_free_sync_cb, zio, tx);
8313 VERIFY(zio_wait(zio) == 0);
8317 * Note: this simple function is not inlined to make it easier to dtrace the
8318 * amount of time spent syncing deferred frees.
8321 spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx)
8323 if (spa_sync_pass(spa) != 1)
8328 * If the log space map feature is active, we stop deferring
8329 * frees to the next TXG and therefore running this function
8330 * would be considered a no-op as spa_deferred_bpobj should
8331 * not have any entries.
8333 * That said we run this function anyway (instead of returning
8334 * immediately) for the edge-case scenario where we just
8335 * activated the log space map feature in this TXG but we have
8336 * deferred frees from the previous TXG.
8338 zio_t *zio = zio_root(spa, NULL, NULL, 0);
8339 VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj,
8340 bpobj_spa_free_sync_cb, zio, tx), ==, 0);
8341 VERIFY0(zio_wait(zio));
8345 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
8347 char *packed = NULL;
8352 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
8355 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
8356 * information. This avoids the dmu_buf_will_dirty() path and
8357 * saves us a pre-read to get data we don't actually care about.
8359 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
8360 packed = vmem_alloc(bufsize, KM_SLEEP);
8362 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
8364 bzero(packed + nvsize, bufsize - nvsize);
8366 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
8368 vmem_free(packed, bufsize);
8370 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
8371 dmu_buf_will_dirty(db, tx);
8372 *(uint64_t *)db->db_data = nvsize;
8373 dmu_buf_rele(db, FTAG);
8377 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
8378 const char *config, const char *entry)
8388 * Update the MOS nvlist describing the list of available devices.
8389 * spa_validate_aux() will have already made sure this nvlist is
8390 * valid and the vdevs are labeled appropriately.
8392 if (sav->sav_object == 0) {
8393 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
8394 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
8395 sizeof (uint64_t), tx);
8396 VERIFY(zap_update(spa->spa_meta_objset,
8397 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
8398 &sav->sav_object, tx) == 0);
8401 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
8402 if (sav->sav_count == 0) {
8403 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
8405 list = kmem_alloc(sav->sav_count*sizeof (void *), KM_SLEEP);
8406 for (i = 0; i < sav->sav_count; i++)
8407 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
8408 B_FALSE, VDEV_CONFIG_L2CACHE);
8409 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
8410 sav->sav_count) == 0);
8411 for (i = 0; i < sav->sav_count; i++)
8412 nvlist_free(list[i]);
8413 kmem_free(list, sav->sav_count * sizeof (void *));
8416 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
8417 nvlist_free(nvroot);
8419 sav->sav_sync = B_FALSE;
8423 * Rebuild spa's all-vdev ZAP from the vdev ZAPs indicated in each vdev_t.
8424 * The all-vdev ZAP must be empty.
8427 spa_avz_build(vdev_t *vd, uint64_t avz, dmu_tx_t *tx)
8429 spa_t *spa = vd->vdev_spa;
8431 if (vd->vdev_top_zap != 0) {
8432 VERIFY0(zap_add_int(spa->spa_meta_objset, avz,
8433 vd->vdev_top_zap, tx));
8435 if (vd->vdev_leaf_zap != 0) {
8436 VERIFY0(zap_add_int(spa->spa_meta_objset, avz,
8437 vd->vdev_leaf_zap, tx));
8439 for (uint64_t i = 0; i < vd->vdev_children; i++) {
8440 spa_avz_build(vd->vdev_child[i], avz, tx);
8445 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
8450 * If the pool is being imported from a pre-per-vdev-ZAP version of ZFS,
8451 * its config may not be dirty but we still need to build per-vdev ZAPs.
8452 * Similarly, if the pool is being assembled (e.g. after a split), we
8453 * need to rebuild the AVZ although the config may not be dirty.
8455 if (list_is_empty(&spa->spa_config_dirty_list) &&
8456 spa->spa_avz_action == AVZ_ACTION_NONE)
8459 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
8461 ASSERT(spa->spa_avz_action == AVZ_ACTION_NONE ||
8462 spa->spa_avz_action == AVZ_ACTION_INITIALIZE ||
8463 spa->spa_all_vdev_zaps != 0);
8465 if (spa->spa_avz_action == AVZ_ACTION_REBUILD) {
8466 /* Make and build the new AVZ */
8467 uint64_t new_avz = zap_create(spa->spa_meta_objset,
8468 DMU_OTN_ZAP_METADATA, DMU_OT_NONE, 0, tx);
8469 spa_avz_build(spa->spa_root_vdev, new_avz, tx);
8471 /* Diff old AVZ with new one */
8475 for (zap_cursor_init(&zc, spa->spa_meta_objset,
8476 spa->spa_all_vdev_zaps);
8477 zap_cursor_retrieve(&zc, &za) == 0;
8478 zap_cursor_advance(&zc)) {
8479 uint64_t vdzap = za.za_first_integer;
8480 if (zap_lookup_int(spa->spa_meta_objset, new_avz,
8483 * ZAP is listed in old AVZ but not in new one;
8486 VERIFY0(zap_destroy(spa->spa_meta_objset, vdzap,
8491 zap_cursor_fini(&zc);
8493 /* Destroy the old AVZ */
8494 VERIFY0(zap_destroy(spa->spa_meta_objset,
8495 spa->spa_all_vdev_zaps, tx));
8497 /* Replace the old AVZ in the dir obj with the new one */
8498 VERIFY0(zap_update(spa->spa_meta_objset,
8499 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP,
8500 sizeof (new_avz), 1, &new_avz, tx));
8502 spa->spa_all_vdev_zaps = new_avz;
8503 } else if (spa->spa_avz_action == AVZ_ACTION_DESTROY) {
8507 /* Walk through the AVZ and destroy all listed ZAPs */
8508 for (zap_cursor_init(&zc, spa->spa_meta_objset,
8509 spa->spa_all_vdev_zaps);
8510 zap_cursor_retrieve(&zc, &za) == 0;
8511 zap_cursor_advance(&zc)) {
8512 uint64_t zap = za.za_first_integer;
8513 VERIFY0(zap_destroy(spa->spa_meta_objset, zap, tx));
8516 zap_cursor_fini(&zc);
8518 /* Destroy and unlink the AVZ itself */
8519 VERIFY0(zap_destroy(spa->spa_meta_objset,
8520 spa->spa_all_vdev_zaps, tx));
8521 VERIFY0(zap_remove(spa->spa_meta_objset,
8522 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP, tx));
8523 spa->spa_all_vdev_zaps = 0;
8526 if (spa->spa_all_vdev_zaps == 0) {
8527 spa->spa_all_vdev_zaps = zap_create_link(spa->spa_meta_objset,
8528 DMU_OTN_ZAP_METADATA, DMU_POOL_DIRECTORY_OBJECT,
8529 DMU_POOL_VDEV_ZAP_MAP, tx);
8531 spa->spa_avz_action = AVZ_ACTION_NONE;
8533 /* Create ZAPs for vdevs that don't have them. */
8534 vdev_construct_zaps(spa->spa_root_vdev, tx);
8536 config = spa_config_generate(spa, spa->spa_root_vdev,
8537 dmu_tx_get_txg(tx), B_FALSE);
8540 * If we're upgrading the spa version then make sure that
8541 * the config object gets updated with the correct version.
8543 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
8544 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
8545 spa->spa_uberblock.ub_version);
8547 spa_config_exit(spa, SCL_STATE, FTAG);
8549 nvlist_free(spa->spa_config_syncing);
8550 spa->spa_config_syncing = config;
8552 spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
8556 spa_sync_version(void *arg, dmu_tx_t *tx)
8558 uint64_t *versionp = arg;
8559 uint64_t version = *versionp;
8560 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
8563 * Setting the version is special cased when first creating the pool.
8565 ASSERT(tx->tx_txg != TXG_INITIAL);
8567 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
8568 ASSERT(version >= spa_version(spa));
8570 spa->spa_uberblock.ub_version = version;
8571 vdev_config_dirty(spa->spa_root_vdev);
8572 spa_history_log_internal(spa, "set", tx, "version=%lld",
8573 (longlong_t)version);
8577 * Set zpool properties.
8580 spa_sync_props(void *arg, dmu_tx_t *tx)
8582 nvlist_t *nvp = arg;
8583 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
8584 objset_t *mos = spa->spa_meta_objset;
8585 nvpair_t *elem = NULL;
8587 mutex_enter(&spa->spa_props_lock);
8589 while ((elem = nvlist_next_nvpair(nvp, elem))) {
8591 char *strval, *fname;
8593 const char *propname;
8594 zprop_type_t proptype;
8597 switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
8598 case ZPOOL_PROP_INVAL:
8600 * We checked this earlier in spa_prop_validate().
8602 ASSERT(zpool_prop_feature(nvpair_name(elem)));
8604 fname = strchr(nvpair_name(elem), '@') + 1;
8605 VERIFY0(zfeature_lookup_name(fname, &fid));
8607 spa_feature_enable(spa, fid, tx);
8608 spa_history_log_internal(spa, "set", tx,
8609 "%s=enabled", nvpair_name(elem));
8612 case ZPOOL_PROP_VERSION:
8613 intval = fnvpair_value_uint64(elem);
8615 * The version is synced separately before other
8616 * properties and should be correct by now.
8618 ASSERT3U(spa_version(spa), >=, intval);
8621 case ZPOOL_PROP_ALTROOT:
8623 * 'altroot' is a non-persistent property. It should
8624 * have been set temporarily at creation or import time.
8626 ASSERT(spa->spa_root != NULL);
8629 case ZPOOL_PROP_READONLY:
8630 case ZPOOL_PROP_CACHEFILE:
8632 * 'readonly' and 'cachefile' are also non-persistent
8636 case ZPOOL_PROP_COMMENT:
8637 strval = fnvpair_value_string(elem);
8638 if (spa->spa_comment != NULL)
8639 spa_strfree(spa->spa_comment);
8640 spa->spa_comment = spa_strdup(strval);
8642 * We need to dirty the configuration on all the vdevs
8643 * so that their labels get updated. It's unnecessary
8644 * to do this for pool creation since the vdev's
8645 * configuration has already been dirtied.
8647 if (tx->tx_txg != TXG_INITIAL)
8648 vdev_config_dirty(spa->spa_root_vdev);
8649 spa_history_log_internal(spa, "set", tx,
8650 "%s=%s", nvpair_name(elem), strval);
8654 * Set pool property values in the poolprops mos object.
8656 if (spa->spa_pool_props_object == 0) {
8657 spa->spa_pool_props_object =
8658 zap_create_link(mos, DMU_OT_POOL_PROPS,
8659 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
8663 /* normalize the property name */
8664 propname = zpool_prop_to_name(prop);
8665 proptype = zpool_prop_get_type(prop);
8667 if (nvpair_type(elem) == DATA_TYPE_STRING) {
8668 ASSERT(proptype == PROP_TYPE_STRING);
8669 strval = fnvpair_value_string(elem);
8670 VERIFY0(zap_update(mos,
8671 spa->spa_pool_props_object, propname,
8672 1, strlen(strval) + 1, strval, tx));
8673 spa_history_log_internal(spa, "set", tx,
8674 "%s=%s", nvpair_name(elem), strval);
8675 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
8676 intval = fnvpair_value_uint64(elem);
8678 if (proptype == PROP_TYPE_INDEX) {
8680 VERIFY0(zpool_prop_index_to_string(
8681 prop, intval, &unused));
8683 VERIFY0(zap_update(mos,
8684 spa->spa_pool_props_object, propname,
8685 8, 1, &intval, tx));
8686 spa_history_log_internal(spa, "set", tx,
8687 "%s=%lld", nvpair_name(elem),
8688 (longlong_t)intval);
8690 ASSERT(0); /* not allowed */
8694 case ZPOOL_PROP_DELEGATION:
8695 spa->spa_delegation = intval;
8697 case ZPOOL_PROP_BOOTFS:
8698 spa->spa_bootfs = intval;
8700 case ZPOOL_PROP_FAILUREMODE:
8701 spa->spa_failmode = intval;
8703 case ZPOOL_PROP_AUTOTRIM:
8704 spa->spa_autotrim = intval;
8705 spa_async_request(spa,
8706 SPA_ASYNC_AUTOTRIM_RESTART);
8708 case ZPOOL_PROP_AUTOEXPAND:
8709 spa->spa_autoexpand = intval;
8710 if (tx->tx_txg != TXG_INITIAL)
8711 spa_async_request(spa,
8712 SPA_ASYNC_AUTOEXPAND);
8714 case ZPOOL_PROP_MULTIHOST:
8715 spa->spa_multihost = intval;
8724 mutex_exit(&spa->spa_props_lock);
8728 * Perform one-time upgrade on-disk changes. spa_version() does not
8729 * reflect the new version this txg, so there must be no changes this
8730 * txg to anything that the upgrade code depends on after it executes.
8731 * Therefore this must be called after dsl_pool_sync() does the sync
8735 spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
8737 if (spa_sync_pass(spa) != 1)
8740 dsl_pool_t *dp = spa->spa_dsl_pool;
8741 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
8743 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
8744 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
8745 dsl_pool_create_origin(dp, tx);
8747 /* Keeping the origin open increases spa_minref */
8748 spa->spa_minref += 3;
8751 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
8752 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
8753 dsl_pool_upgrade_clones(dp, tx);
8756 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
8757 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
8758 dsl_pool_upgrade_dir_clones(dp, tx);
8760 /* Keeping the freedir open increases spa_minref */
8761 spa->spa_minref += 3;
8764 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES &&
8765 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
8766 spa_feature_create_zap_objects(spa, tx);
8770 * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable
8771 * when possibility to use lz4 compression for metadata was added
8772 * Old pools that have this feature enabled must be upgraded to have
8773 * this feature active
8775 if (spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
8776 boolean_t lz4_en = spa_feature_is_enabled(spa,
8777 SPA_FEATURE_LZ4_COMPRESS);
8778 boolean_t lz4_ac = spa_feature_is_active(spa,
8779 SPA_FEATURE_LZ4_COMPRESS);
8781 if (lz4_en && !lz4_ac)
8782 spa_feature_incr(spa, SPA_FEATURE_LZ4_COMPRESS, tx);
8786 * If we haven't written the salt, do so now. Note that the
8787 * feature may not be activated yet, but that's fine since
8788 * the presence of this ZAP entry is backwards compatible.
8790 if (zap_contains(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
8791 DMU_POOL_CHECKSUM_SALT) == ENOENT) {
8792 VERIFY0(zap_add(spa->spa_meta_objset,
8793 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CHECKSUM_SALT, 1,
8794 sizeof (spa->spa_cksum_salt.zcs_bytes),
8795 spa->spa_cksum_salt.zcs_bytes, tx));
8798 rrw_exit(&dp->dp_config_rwlock, FTAG);
8802 vdev_indirect_state_sync_verify(vdev_t *vd)
8804 vdev_indirect_mapping_t *vim __maybe_unused = vd->vdev_indirect_mapping;
8805 vdev_indirect_births_t *vib __maybe_unused = vd->vdev_indirect_births;
8807 if (vd->vdev_ops == &vdev_indirect_ops) {
8808 ASSERT(vim != NULL);
8809 ASSERT(vib != NULL);
8812 uint64_t obsolete_sm_object = 0;
8813 ASSERT0(vdev_obsolete_sm_object(vd, &obsolete_sm_object));
8814 if (obsolete_sm_object != 0) {
8815 ASSERT(vd->vdev_obsolete_sm != NULL);
8816 ASSERT(vd->vdev_removing ||
8817 vd->vdev_ops == &vdev_indirect_ops);
8818 ASSERT(vdev_indirect_mapping_num_entries(vim) > 0);
8819 ASSERT(vdev_indirect_mapping_bytes_mapped(vim) > 0);
8820 ASSERT3U(obsolete_sm_object, ==,
8821 space_map_object(vd->vdev_obsolete_sm));
8822 ASSERT3U(vdev_indirect_mapping_bytes_mapped(vim), >=,
8823 space_map_allocated(vd->vdev_obsolete_sm));
8825 ASSERT(vd->vdev_obsolete_segments != NULL);
8828 * Since frees / remaps to an indirect vdev can only
8829 * happen in syncing context, the obsolete segments
8830 * tree must be empty when we start syncing.
8832 ASSERT0(range_tree_space(vd->vdev_obsolete_segments));
8836 * Set the top-level vdev's max queue depth. Evaluate each top-level's
8837 * async write queue depth in case it changed. The max queue depth will
8838 * not change in the middle of syncing out this txg.
8841 spa_sync_adjust_vdev_max_queue_depth(spa_t *spa)
8843 ASSERT(spa_writeable(spa));
8845 vdev_t *rvd = spa->spa_root_vdev;
8846 uint32_t max_queue_depth = zfs_vdev_async_write_max_active *
8847 zfs_vdev_queue_depth_pct / 100;
8848 metaslab_class_t *normal = spa_normal_class(spa);
8849 metaslab_class_t *special = spa_special_class(spa);
8850 metaslab_class_t *dedup = spa_dedup_class(spa);
8852 uint64_t slots_per_allocator = 0;
8853 for (int c = 0; c < rvd->vdev_children; c++) {
8854 vdev_t *tvd = rvd->vdev_child[c];
8856 metaslab_group_t *mg = tvd->vdev_mg;
8857 if (mg == NULL || !metaslab_group_initialized(mg))
8860 metaslab_class_t *mc = mg->mg_class;
8861 if (mc != normal && mc != special && mc != dedup)
8865 * It is safe to do a lock-free check here because only async
8866 * allocations look at mg_max_alloc_queue_depth, and async
8867 * allocations all happen from spa_sync().
8869 for (int i = 0; i < mg->mg_allocators; i++) {
8870 ASSERT0(zfs_refcount_count(
8871 &(mg->mg_allocator[i].mga_alloc_queue_depth)));
8873 mg->mg_max_alloc_queue_depth = max_queue_depth;
8875 for (int i = 0; i < mg->mg_allocators; i++) {
8876 mg->mg_allocator[i].mga_cur_max_alloc_queue_depth =
8877 zfs_vdev_def_queue_depth;
8879 slots_per_allocator += zfs_vdev_def_queue_depth;
8882 for (int i = 0; i < spa->spa_alloc_count; i++) {
8883 ASSERT0(zfs_refcount_count(&normal->mc_alloc_slots[i]));
8884 ASSERT0(zfs_refcount_count(&special->mc_alloc_slots[i]));
8885 ASSERT0(zfs_refcount_count(&dedup->mc_alloc_slots[i]));
8886 normal->mc_alloc_max_slots[i] = slots_per_allocator;
8887 special->mc_alloc_max_slots[i] = slots_per_allocator;
8888 dedup->mc_alloc_max_slots[i] = slots_per_allocator;
8890 normal->mc_alloc_throttle_enabled = zio_dva_throttle_enabled;
8891 special->mc_alloc_throttle_enabled = zio_dva_throttle_enabled;
8892 dedup->mc_alloc_throttle_enabled = zio_dva_throttle_enabled;
8896 spa_sync_condense_indirect(spa_t *spa, dmu_tx_t *tx)
8898 ASSERT(spa_writeable(spa));
8900 vdev_t *rvd = spa->spa_root_vdev;
8901 for (int c = 0; c < rvd->vdev_children; c++) {
8902 vdev_t *vd = rvd->vdev_child[c];
8903 vdev_indirect_state_sync_verify(vd);
8905 if (vdev_indirect_should_condense(vd)) {
8906 spa_condense_indirect_start_sync(vd, tx);
8913 spa_sync_iterate_to_convergence(spa_t *spa, dmu_tx_t *tx)
8915 objset_t *mos = spa->spa_meta_objset;
8916 dsl_pool_t *dp = spa->spa_dsl_pool;
8917 uint64_t txg = tx->tx_txg;
8918 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
8921 int pass = ++spa->spa_sync_pass;
8923 spa_sync_config_object(spa, tx);
8924 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
8925 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
8926 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
8927 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
8928 spa_errlog_sync(spa, txg);
8929 dsl_pool_sync(dp, txg);
8931 if (pass < zfs_sync_pass_deferred_free ||
8932 spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP)) {
8934 * If the log space map feature is active we don't
8935 * care about deferred frees and the deferred bpobj
8936 * as the log space map should effectively have the
8937 * same results (i.e. appending only to one object).
8939 spa_sync_frees(spa, free_bpl, tx);
8942 * We can not defer frees in pass 1, because
8943 * we sync the deferred frees later in pass 1.
8945 ASSERT3U(pass, >, 1);
8946 bplist_iterate(free_bpl, bpobj_enqueue_alloc_cb,
8947 &spa->spa_deferred_bpobj, tx);
8951 dsl_scan_sync(dp, tx);
8953 spa_sync_upgrades(spa, tx);
8955 spa_flush_metaslabs(spa, tx);
8958 while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
8963 * Note: We need to check if the MOS is dirty because we could
8964 * have marked the MOS dirty without updating the uberblock
8965 * (e.g. if we have sync tasks but no dirty user data). We need
8966 * to check the uberblock's rootbp because it is updated if we
8967 * have synced out dirty data (though in this case the MOS will
8968 * most likely also be dirty due to second order effects, we
8969 * don't want to rely on that here).
8972 spa->spa_uberblock.ub_rootbp.blk_birth < txg &&
8973 !dmu_objset_is_dirty(mos, txg)) {
8975 * Nothing changed on the first pass, therefore this
8976 * TXG is a no-op. Avoid syncing deferred frees, so
8977 * that we can keep this TXG as a no-op.
8979 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
8980 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
8981 ASSERT(txg_list_empty(&dp->dp_sync_tasks, txg));
8982 ASSERT(txg_list_empty(&dp->dp_early_sync_tasks, txg));
8986 spa_sync_deferred_frees(spa, tx);
8987 } while (dmu_objset_is_dirty(mos, txg));
8991 * Rewrite the vdev configuration (which includes the uberblock) to
8992 * commit the transaction group.
8994 * If there are no dirty vdevs, we sync the uberblock to a few random
8995 * top-level vdevs that are known to be visible in the config cache
8996 * (see spa_vdev_add() for a complete description). If there *are* dirty
8997 * vdevs, sync the uberblock to all vdevs.
9000 spa_sync_rewrite_vdev_config(spa_t *spa, dmu_tx_t *tx)
9002 vdev_t *rvd = spa->spa_root_vdev;
9003 uint64_t txg = tx->tx_txg;
9009 * We hold SCL_STATE to prevent vdev open/close/etc.
9010 * while we're attempting to write the vdev labels.
9012 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
9014 if (list_is_empty(&spa->spa_config_dirty_list)) {
9015 vdev_t *svd[SPA_SYNC_MIN_VDEVS] = { NULL };
9017 int children = rvd->vdev_children;
9018 int c0 = spa_get_random(children);
9020 for (int c = 0; c < children; c++) {
9022 rvd->vdev_child[(c0 + c) % children];
9024 /* Stop when revisiting the first vdev */
9025 if (c > 0 && svd[0] == vd)
9028 if (vd->vdev_ms_array == 0 ||
9030 !vdev_is_concrete(vd))
9033 svd[svdcount++] = vd;
9034 if (svdcount == SPA_SYNC_MIN_VDEVS)
9037 error = vdev_config_sync(svd, svdcount, txg);
9039 error = vdev_config_sync(rvd->vdev_child,
9040 rvd->vdev_children, txg);
9044 spa->spa_last_synced_guid = rvd->vdev_guid;
9046 spa_config_exit(spa, SCL_STATE, FTAG);
9050 zio_suspend(spa, NULL, ZIO_SUSPEND_IOERR);
9051 zio_resume_wait(spa);
9056 * Sync the specified transaction group. New blocks may be dirtied as
9057 * part of the process, so we iterate until it converges.
9060 spa_sync(spa_t *spa, uint64_t txg)
9064 VERIFY(spa_writeable(spa));
9067 * Wait for i/os issued in open context that need to complete
9068 * before this txg syncs.
9070 (void) zio_wait(spa->spa_txg_zio[txg & TXG_MASK]);
9071 spa->spa_txg_zio[txg & TXG_MASK] = zio_root(spa, NULL, NULL,
9075 * Lock out configuration changes.
9077 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
9079 spa->spa_syncing_txg = txg;
9080 spa->spa_sync_pass = 0;
9082 for (int i = 0; i < spa->spa_alloc_count; i++) {
9083 mutex_enter(&spa->spa_alloc_locks[i]);
9084 VERIFY0(avl_numnodes(&spa->spa_alloc_trees[i]));
9085 mutex_exit(&spa->spa_alloc_locks[i]);
9089 * If there are any pending vdev state changes, convert them
9090 * into config changes that go out with this transaction group.
9092 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
9093 while (list_head(&spa->spa_state_dirty_list) != NULL) {
9095 * We need the write lock here because, for aux vdevs,
9096 * calling vdev_config_dirty() modifies sav_config.
9097 * This is ugly and will become unnecessary when we
9098 * eliminate the aux vdev wart by integrating all vdevs
9099 * into the root vdev tree.
9101 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
9102 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
9103 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
9104 vdev_state_clean(vd);
9105 vdev_config_dirty(vd);
9107 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
9108 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
9110 spa_config_exit(spa, SCL_STATE, FTAG);
9112 dsl_pool_t *dp = spa->spa_dsl_pool;
9113 dmu_tx_t *tx = dmu_tx_create_assigned(dp, txg);
9115 spa->spa_sync_starttime = gethrtime();
9116 taskq_cancel_id(system_delay_taskq, spa->spa_deadman_tqid);
9117 spa->spa_deadman_tqid = taskq_dispatch_delay(system_delay_taskq,
9118 spa_deadman, spa, TQ_SLEEP, ddi_get_lbolt() +
9119 NSEC_TO_TICK(spa->spa_deadman_synctime));
9122 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
9123 * set spa_deflate if we have no raid-z vdevs.
9125 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
9126 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
9127 vdev_t *rvd = spa->spa_root_vdev;
9130 for (i = 0; i < rvd->vdev_children; i++) {
9131 vd = rvd->vdev_child[i];
9132 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
9135 if (i == rvd->vdev_children) {
9136 spa->spa_deflate = TRUE;
9137 VERIFY0(zap_add(spa->spa_meta_objset,
9138 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
9139 sizeof (uint64_t), 1, &spa->spa_deflate, tx));
9143 spa_sync_adjust_vdev_max_queue_depth(spa);
9145 spa_sync_condense_indirect(spa, tx);
9147 spa_sync_iterate_to_convergence(spa, tx);
9150 if (!list_is_empty(&spa->spa_config_dirty_list)) {
9152 * Make sure that the number of ZAPs for all the vdevs matches
9153 * the number of ZAPs in the per-vdev ZAP list. This only gets
9154 * called if the config is dirty; otherwise there may be
9155 * outstanding AVZ operations that weren't completed in
9156 * spa_sync_config_object.
9158 uint64_t all_vdev_zap_entry_count;
9159 ASSERT0(zap_count(spa->spa_meta_objset,
9160 spa->spa_all_vdev_zaps, &all_vdev_zap_entry_count));
9161 ASSERT3U(vdev_count_verify_zaps(spa->spa_root_vdev), ==,
9162 all_vdev_zap_entry_count);
9166 if (spa->spa_vdev_removal != NULL) {
9167 ASSERT0(spa->spa_vdev_removal->svr_bytes_done[txg & TXG_MASK]);
9170 spa_sync_rewrite_vdev_config(spa, tx);
9173 taskq_cancel_id(system_delay_taskq, spa->spa_deadman_tqid);
9174 spa->spa_deadman_tqid = 0;
9177 * Clear the dirty config list.
9179 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
9180 vdev_config_clean(vd);
9183 * Now that the new config has synced transactionally,
9184 * let it become visible to the config cache.
9186 if (spa->spa_config_syncing != NULL) {
9187 spa_config_set(spa, spa->spa_config_syncing);
9188 spa->spa_config_txg = txg;
9189 spa->spa_config_syncing = NULL;
9192 dsl_pool_sync_done(dp, txg);
9194 for (int i = 0; i < spa->spa_alloc_count; i++) {
9195 mutex_enter(&spa->spa_alloc_locks[i]);
9196 VERIFY0(avl_numnodes(&spa->spa_alloc_trees[i]));
9197 mutex_exit(&spa->spa_alloc_locks[i]);
9201 * Update usable space statistics.
9203 while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
9205 vdev_sync_done(vd, txg);
9207 metaslab_class_evict_old(spa->spa_normal_class, txg);
9208 metaslab_class_evict_old(spa->spa_log_class, txg);
9210 spa_sync_close_syncing_log_sm(spa);
9212 spa_update_dspace(spa);
9215 * It had better be the case that we didn't dirty anything
9216 * since vdev_config_sync().
9218 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
9219 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
9220 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
9222 while (zfs_pause_spa_sync)
9225 spa->spa_sync_pass = 0;
9228 * Update the last synced uberblock here. We want to do this at
9229 * the end of spa_sync() so that consumers of spa_last_synced_txg()
9230 * will be guaranteed that all the processing associated with
9231 * that txg has been completed.
9233 spa->spa_ubsync = spa->spa_uberblock;
9234 spa_config_exit(spa, SCL_CONFIG, FTAG);
9236 spa_handle_ignored_writes(spa);
9239 * If any async tasks have been requested, kick them off.
9241 spa_async_dispatch(spa);
9245 * Sync all pools. We don't want to hold the namespace lock across these
9246 * operations, so we take a reference on the spa_t and drop the lock during the
9250 spa_sync_allpools(void)
9253 mutex_enter(&spa_namespace_lock);
9254 while ((spa = spa_next(spa)) != NULL) {
9255 if (spa_state(spa) != POOL_STATE_ACTIVE ||
9256 !spa_writeable(spa) || spa_suspended(spa))
9258 spa_open_ref(spa, FTAG);
9259 mutex_exit(&spa_namespace_lock);
9260 txg_wait_synced(spa_get_dsl(spa), 0);
9261 mutex_enter(&spa_namespace_lock);
9262 spa_close(spa, FTAG);
9264 mutex_exit(&spa_namespace_lock);
9268 * ==========================================================================
9269 * Miscellaneous routines
9270 * ==========================================================================
9274 * Remove all pools in the system.
9282 * Remove all cached state. All pools should be closed now,
9283 * so every spa in the AVL tree should be unreferenced.
9285 mutex_enter(&spa_namespace_lock);
9286 while ((spa = spa_next(NULL)) != NULL) {
9288 * Stop async tasks. The async thread may need to detach
9289 * a device that's been replaced, which requires grabbing
9290 * spa_namespace_lock, so we must drop it here.
9292 spa_open_ref(spa, FTAG);
9293 mutex_exit(&spa_namespace_lock);
9294 spa_async_suspend(spa);
9295 mutex_enter(&spa_namespace_lock);
9296 spa_close(spa, FTAG);
9298 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
9300 spa_deactivate(spa);
9304 mutex_exit(&spa_namespace_lock);
9308 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
9313 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
9317 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
9318 vd = spa->spa_l2cache.sav_vdevs[i];
9319 if (vd->vdev_guid == guid)
9323 for (i = 0; i < spa->spa_spares.sav_count; i++) {
9324 vd = spa->spa_spares.sav_vdevs[i];
9325 if (vd->vdev_guid == guid)
9334 spa_upgrade(spa_t *spa, uint64_t version)
9336 ASSERT(spa_writeable(spa));
9338 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
9341 * This should only be called for a non-faulted pool, and since a
9342 * future version would result in an unopenable pool, this shouldn't be
9345 ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version));
9346 ASSERT3U(version, >=, spa->spa_uberblock.ub_version);
9348 spa->spa_uberblock.ub_version = version;
9349 vdev_config_dirty(spa->spa_root_vdev);
9351 spa_config_exit(spa, SCL_ALL, FTAG);
9353 txg_wait_synced(spa_get_dsl(spa), 0);
9357 spa_has_spare(spa_t *spa, uint64_t guid)
9361 spa_aux_vdev_t *sav = &spa->spa_spares;
9363 for (i = 0; i < sav->sav_count; i++)
9364 if (sav->sav_vdevs[i]->vdev_guid == guid)
9367 for (i = 0; i < sav->sav_npending; i++) {
9368 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
9369 &spareguid) == 0 && spareguid == guid)
9377 * Check if a pool has an active shared spare device.
9378 * Note: reference count of an active spare is 2, as a spare and as a replace
9381 spa_has_active_shared_spare(spa_t *spa)
9385 spa_aux_vdev_t *sav = &spa->spa_spares;
9387 for (i = 0; i < sav->sav_count; i++) {
9388 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
9389 &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
9398 spa_total_metaslabs(spa_t *spa)
9400 vdev_t *rvd = spa->spa_root_vdev;
9403 for (uint64_t c = 0; c < rvd->vdev_children; c++) {
9404 vdev_t *vd = rvd->vdev_child[c];
9405 if (!vdev_is_concrete(vd))
9407 m += vd->vdev_ms_count;
9413 * Notify any waiting threads that some activity has switched from being in-
9414 * progress to not-in-progress so that the thread can wake up and determine
9415 * whether it is finished waiting.
9418 spa_notify_waiters(spa_t *spa)
9421 * Acquiring spa_activities_lock here prevents the cv_broadcast from
9422 * happening between the waiting thread's check and cv_wait.
9424 mutex_enter(&spa->spa_activities_lock);
9425 cv_broadcast(&spa->spa_activities_cv);
9426 mutex_exit(&spa->spa_activities_lock);
9430 * Notify any waiting threads that the pool is exporting, and then block until
9431 * they are finished using the spa_t.
9434 spa_wake_waiters(spa_t *spa)
9436 mutex_enter(&spa->spa_activities_lock);
9437 spa->spa_waiters_cancel = B_TRUE;
9438 cv_broadcast(&spa->spa_activities_cv);
9439 while (spa->spa_waiters != 0)
9440 cv_wait(&spa->spa_waiters_cv, &spa->spa_activities_lock);
9441 spa->spa_waiters_cancel = B_FALSE;
9442 mutex_exit(&spa->spa_activities_lock);
9445 /* Whether the vdev or any of its descendants are being initialized/trimmed. */
9447 spa_vdev_activity_in_progress_impl(vdev_t *vd, zpool_wait_activity_t activity)
9449 spa_t *spa = vd->vdev_spa;
9451 ASSERT(spa_config_held(spa, SCL_CONFIG | SCL_STATE, RW_READER));
9452 ASSERT(MUTEX_HELD(&spa->spa_activities_lock));
9453 ASSERT(activity == ZPOOL_WAIT_INITIALIZE ||
9454 activity == ZPOOL_WAIT_TRIM);
9456 kmutex_t *lock = activity == ZPOOL_WAIT_INITIALIZE ?
9457 &vd->vdev_initialize_lock : &vd->vdev_trim_lock;
9459 mutex_exit(&spa->spa_activities_lock);
9461 mutex_enter(&spa->spa_activities_lock);
9463 boolean_t in_progress = (activity == ZPOOL_WAIT_INITIALIZE) ?
9464 (vd->vdev_initialize_state == VDEV_INITIALIZE_ACTIVE) :
9465 (vd->vdev_trim_state == VDEV_TRIM_ACTIVE);
9471 for (int i = 0; i < vd->vdev_children; i++) {
9472 if (spa_vdev_activity_in_progress_impl(vd->vdev_child[i],
9481 * If use_guid is true, this checks whether the vdev specified by guid is
9482 * being initialized/trimmed. Otherwise, it checks whether any vdev in the pool
9483 * is being initialized/trimmed. The caller must hold the config lock and
9484 * spa_activities_lock.
9487 spa_vdev_activity_in_progress(spa_t *spa, boolean_t use_guid, uint64_t guid,
9488 zpool_wait_activity_t activity, boolean_t *in_progress)
9490 mutex_exit(&spa->spa_activities_lock);
9491 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
9492 mutex_enter(&spa->spa_activities_lock);
9496 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
9497 if (vd == NULL || !vd->vdev_ops->vdev_op_leaf) {
9498 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
9502 vd = spa->spa_root_vdev;
9505 *in_progress = spa_vdev_activity_in_progress_impl(vd, activity);
9507 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
9512 * Locking for waiting threads
9513 * ---------------------------
9515 * Waiting threads need a way to check whether a given activity is in progress,
9516 * and then, if it is, wait for it to complete. Each activity will have some
9517 * in-memory representation of the relevant on-disk state which can be used to
9518 * determine whether or not the activity is in progress. The in-memory state and
9519 * the locking used to protect it will be different for each activity, and may
9520 * not be suitable for use with a cvar (e.g., some state is protected by the
9521 * config lock). To allow waiting threads to wait without any races, another
9522 * lock, spa_activities_lock, is used.
9524 * When the state is checked, both the activity-specific lock (if there is one)
9525 * and spa_activities_lock are held. In some cases, the activity-specific lock
9526 * is acquired explicitly (e.g. the config lock). In others, the locking is
9527 * internal to some check (e.g. bpobj_is_empty). After checking, the waiting
9528 * thread releases the activity-specific lock and, if the activity is in
9529 * progress, then cv_waits using spa_activities_lock.
9531 * The waiting thread is woken when another thread, one completing some
9532 * activity, updates the state of the activity and then calls
9533 * spa_notify_waiters, which will cv_broadcast. This 'completing' thread only
9534 * needs to hold its activity-specific lock when updating the state, and this
9535 * lock can (but doesn't have to) be dropped before calling spa_notify_waiters.
9537 * Because spa_notify_waiters acquires spa_activities_lock before broadcasting,
9538 * and because it is held when the waiting thread checks the state of the
9539 * activity, it can never be the case that the completing thread both updates
9540 * the activity state and cv_broadcasts in between the waiting thread's check
9541 * and cv_wait. Thus, a waiting thread can never miss a wakeup.
9543 * In order to prevent deadlock, when the waiting thread does its check, in some
9544 * cases it will temporarily drop spa_activities_lock in order to acquire the
9545 * activity-specific lock. The order in which spa_activities_lock and the
9546 * activity specific lock are acquired in the waiting thread is determined by
9547 * the order in which they are acquired in the completing thread; if the
9548 * completing thread calls spa_notify_waiters with the activity-specific lock
9549 * held, then the waiting thread must also acquire the activity-specific lock
9554 spa_activity_in_progress(spa_t *spa, zpool_wait_activity_t activity,
9555 boolean_t use_tag, uint64_t tag, boolean_t *in_progress)
9559 ASSERT(MUTEX_HELD(&spa->spa_activities_lock));
9562 case ZPOOL_WAIT_CKPT_DISCARD:
9564 (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT) &&
9565 zap_contains(spa_meta_objset(spa),
9566 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_ZPOOL_CHECKPOINT) ==
9569 case ZPOOL_WAIT_FREE:
9570 *in_progress = ((spa_version(spa) >= SPA_VERSION_DEADLISTS &&
9571 !bpobj_is_empty(&spa->spa_dsl_pool->dp_free_bpobj)) ||
9572 spa_feature_is_active(spa, SPA_FEATURE_ASYNC_DESTROY) ||
9573 spa_livelist_delete_check(spa));
9575 case ZPOOL_WAIT_INITIALIZE:
9576 case ZPOOL_WAIT_TRIM:
9577 error = spa_vdev_activity_in_progress(spa, use_tag, tag,
9578 activity, in_progress);
9580 case ZPOOL_WAIT_REPLACE:
9581 mutex_exit(&spa->spa_activities_lock);
9582 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
9583 mutex_enter(&spa->spa_activities_lock);
9585 *in_progress = vdev_replace_in_progress(spa->spa_root_vdev);
9586 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
9588 case ZPOOL_WAIT_REMOVE:
9589 *in_progress = (spa->spa_removing_phys.sr_state ==
9592 case ZPOOL_WAIT_RESILVER:
9593 if ((*in_progress = vdev_rebuild_active(spa->spa_root_vdev)))
9596 case ZPOOL_WAIT_SCRUB:
9598 boolean_t scanning, paused, is_scrub;
9599 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
9601 is_scrub = (scn->scn_phys.scn_func == POOL_SCAN_SCRUB);
9602 scanning = (scn->scn_phys.scn_state == DSS_SCANNING);
9603 paused = dsl_scan_is_paused_scrub(scn);
9604 *in_progress = (scanning && !paused &&
9605 is_scrub == (activity == ZPOOL_WAIT_SCRUB));
9609 panic("unrecognized value for activity %d", activity);
9616 spa_wait_common(const char *pool, zpool_wait_activity_t activity,
9617 boolean_t use_tag, uint64_t tag, boolean_t *waited)
9620 * The tag is used to distinguish between instances of an activity.
9621 * 'initialize' and 'trim' are the only activities that we use this for.
9622 * The other activities can only have a single instance in progress in a
9623 * pool at one time, making the tag unnecessary.
9625 * There can be multiple devices being replaced at once, but since they
9626 * all finish once resilvering finishes, we don't bother keeping track
9627 * of them individually, we just wait for them all to finish.
9629 if (use_tag && activity != ZPOOL_WAIT_INITIALIZE &&
9630 activity != ZPOOL_WAIT_TRIM)
9633 if (activity < 0 || activity >= ZPOOL_WAIT_NUM_ACTIVITIES)
9637 int error = spa_open(pool, &spa, FTAG);
9642 * Increment the spa's waiter count so that we can call spa_close and
9643 * still ensure that the spa_t doesn't get freed before this thread is
9644 * finished with it when the pool is exported. We want to call spa_close
9645 * before we start waiting because otherwise the additional ref would
9646 * prevent the pool from being exported or destroyed throughout the
9647 * potentially long wait.
9649 mutex_enter(&spa->spa_activities_lock);
9651 spa_close(spa, FTAG);
9655 boolean_t in_progress;
9656 error = spa_activity_in_progress(spa, activity, use_tag, tag,
9659 if (error || !in_progress || spa->spa_waiters_cancel)
9664 if (cv_wait_sig(&spa->spa_activities_cv,
9665 &spa->spa_activities_lock) == 0) {
9672 cv_signal(&spa->spa_waiters_cv);
9673 mutex_exit(&spa->spa_activities_lock);
9679 * Wait for a particular instance of the specified activity to complete, where
9680 * the instance is identified by 'tag'
9683 spa_wait_tag(const char *pool, zpool_wait_activity_t activity, uint64_t tag,
9686 return (spa_wait_common(pool, activity, B_TRUE, tag, waited));
9690 * Wait for all instances of the specified activity complete
9693 spa_wait(const char *pool, zpool_wait_activity_t activity, boolean_t *waited)
9696 return (spa_wait_common(pool, activity, B_FALSE, 0, waited));
9700 spa_event_create(spa_t *spa, vdev_t *vd, nvlist_t *hist_nvl, const char *name)
9702 sysevent_t *ev = NULL;
9706 resource = zfs_event_create(spa, vd, FM_SYSEVENT_CLASS, name, hist_nvl);
9708 ev = kmem_alloc(sizeof (sysevent_t), KM_SLEEP);
9709 ev->resource = resource;
9716 spa_event_post(sysevent_t *ev)
9720 zfs_zevent_post(ev->resource, NULL, zfs_zevent_post_cb);
9721 kmem_free(ev, sizeof (*ev));
9727 * Post a zevent corresponding to the given sysevent. The 'name' must be one
9728 * of the event definitions in sys/sysevent/eventdefs.h. The payload will be
9729 * filled in from the spa and (optionally) the vdev. This doesn't do anything
9730 * in the userland libzpool, as we don't want consumers to misinterpret ztest
9731 * or zdb as real changes.
9734 spa_event_notify(spa_t *spa, vdev_t *vd, nvlist_t *hist_nvl, const char *name)
9736 spa_event_post(spa_event_create(spa, vd, hist_nvl, name));
9739 /* state manipulation functions */
9740 EXPORT_SYMBOL(spa_open);
9741 EXPORT_SYMBOL(spa_open_rewind);
9742 EXPORT_SYMBOL(spa_get_stats);
9743 EXPORT_SYMBOL(spa_create);
9744 EXPORT_SYMBOL(spa_import);
9745 EXPORT_SYMBOL(spa_tryimport);
9746 EXPORT_SYMBOL(spa_destroy);
9747 EXPORT_SYMBOL(spa_export);
9748 EXPORT_SYMBOL(spa_reset);
9749 EXPORT_SYMBOL(spa_async_request);
9750 EXPORT_SYMBOL(spa_async_suspend);
9751 EXPORT_SYMBOL(spa_async_resume);
9752 EXPORT_SYMBOL(spa_inject_addref);
9753 EXPORT_SYMBOL(spa_inject_delref);
9754 EXPORT_SYMBOL(spa_scan_stat_init);
9755 EXPORT_SYMBOL(spa_scan_get_stats);
9757 /* device manipulation */
9758 EXPORT_SYMBOL(spa_vdev_add);
9759 EXPORT_SYMBOL(spa_vdev_attach);
9760 EXPORT_SYMBOL(spa_vdev_detach);
9761 EXPORT_SYMBOL(spa_vdev_setpath);
9762 EXPORT_SYMBOL(spa_vdev_setfru);
9763 EXPORT_SYMBOL(spa_vdev_split_mirror);
9765 /* spare statech is global across all pools) */
9766 EXPORT_SYMBOL(spa_spare_add);
9767 EXPORT_SYMBOL(spa_spare_remove);
9768 EXPORT_SYMBOL(spa_spare_exists);
9769 EXPORT_SYMBOL(spa_spare_activate);
9771 /* L2ARC statech is global across all pools) */
9772 EXPORT_SYMBOL(spa_l2cache_add);
9773 EXPORT_SYMBOL(spa_l2cache_remove);
9774 EXPORT_SYMBOL(spa_l2cache_exists);
9775 EXPORT_SYMBOL(spa_l2cache_activate);
9776 EXPORT_SYMBOL(spa_l2cache_drop);
9779 EXPORT_SYMBOL(spa_scan);
9780 EXPORT_SYMBOL(spa_scan_stop);
9783 EXPORT_SYMBOL(spa_sync); /* only for DMU use */
9784 EXPORT_SYMBOL(spa_sync_allpools);
9787 EXPORT_SYMBOL(spa_prop_set);
9788 EXPORT_SYMBOL(spa_prop_get);
9789 EXPORT_SYMBOL(spa_prop_clear_bootfs);
9791 /* asynchronous event notification */
9792 EXPORT_SYMBOL(spa_event_notify);
9795 ZFS_MODULE_PARAM(zfs_spa, spa_, load_verify_shift, INT, ZMOD_RW,
9796 "log2(fraction of arc that can be used by inflight I/Os when "
9797 "verifying pool during import");
9799 ZFS_MODULE_PARAM(zfs_spa, spa_, load_verify_metadata, INT, ZMOD_RW,
9800 "Set to traverse metadata on pool import");
9802 ZFS_MODULE_PARAM(zfs_spa, spa_, load_verify_data, INT, ZMOD_RW,
9803 "Set to traverse data on pool import");
9805 ZFS_MODULE_PARAM(zfs_spa, spa_, load_print_vdev_tree, INT, ZMOD_RW,
9806 "Print vdev tree to zfs_dbgmsg during pool import");
9808 ZFS_MODULE_PARAM(zfs_zio, zio_, taskq_batch_pct, UINT, ZMOD_RD,
9809 "Percentage of CPUs to run an IO worker thread");
9811 ZFS_MODULE_PARAM(zfs, zfs_, max_missing_tvds, ULONG, ZMOD_RW,
9812 "Allow importing pool with up to this number of missing top-level "
9813 "vdevs (in read-only mode)");
9815 ZFS_MODULE_PARAM(zfs_livelist_condense, zfs_livelist_condense_, zthr_pause, INT, ZMOD_RW,
9816 "Set the livelist condense zthr to pause");
9818 ZFS_MODULE_PARAM(zfs_livelist_condense, zfs_livelist_condense_, sync_pause, INT, ZMOD_RW,
9819 "Set the livelist condense synctask to pause");
9821 ZFS_MODULE_PARAM(zfs_livelist_condense, zfs_livelist_condense_, sync_cancel, INT, ZMOD_RW,
9822 "Whether livelist condensing was canceled in the synctask");
9824 ZFS_MODULE_PARAM(zfs_livelist_condense, zfs_livelist_condense_, zthr_cancel, INT, ZMOD_RW,
9825 "Whether livelist condensing was canceled in the zthr function");
9827 ZFS_MODULE_PARAM(zfs_livelist_condense, zfs_livelist_condense_, new_alloc, INT, ZMOD_RW,
9828 "Whether extra ALLOC blkptrs were added to a livelist entry while it "
9829 "was being condensed");